romcc: kill gcc warnings and .gitignore generated files
[coreboot.git] / util / romcc / romcc.c
index 561ba2411052d6e7961f955a6ec35992bc96af7c..c7ef22366751613d40174a3170b1bd900d74b84e 100644 (file)
@@ -1,3 +1,12 @@
+#undef VERSION_MAJOR
+#undef VERSION_MINOR
+#undef RELEASE_DATE
+#undef VERSION
+#define VERSION_MAJOR "0"
+#define VERSION_MINOR "72"
+#define RELEASE_DATE "10 February 2010"
+#define VERSION VERSION_MAJOR "." VERSION_MINOR
+
 #include <stdarg.h>
 #include <errno.h>
 #include <stdint.h>
 #include <unistd.h>
 #include <stdio.h>
 #include <string.h>
-#include <ctype.h>
 #include <limits.h>
+#include <locale.h>
+#include <time.h>
 
-#define DEBUG_ERROR_MESSAGES 0
-#define DEBUG_COLOR_GRAPH 0
-#define DEBUG_SCC 0
-#define DEBUG_CONSISTENCY 1
+#define MAX_CWD_SIZE 4096
+#define MAX_ALLOCATION_PASSES 100
+
+/* NOTE: Before you even start thinking to touch anything
+ * in this code, set DEBUG_ROMCC_WARNINGS to 1 to get an
+ * insight on the original author's thoughts. We introduced
+ * this switch as romcc was about the only thing producing
+ * massive warnings in our code..
+ */
+#define DEBUG_ROMCC_WARNINGS 0
 
-#warning "FIXME boundary cases with small types in larger registers"
+#define DEBUG_CONSISTENCY 1
+#define DEBUG_SDP_BLOCKS 0
+#define DEBUG_TRIPLE_COLOR 0
+
+#define DEBUG_DISPLAY_USES 1
+#define DEBUG_DISPLAY_TYPES 1
+#define DEBUG_REPLACE_CLOSURE_TYPE_HIRES 0
+#define DEBUG_DECOMPOSE_PRINT_TUPLES 0
+#define DEBUG_DECOMPOSE_HIRES  0
+#define DEBUG_INITIALIZER 0
+#define DEBUG_UPDATE_CLOSURE_TYPE 0
+#define DEBUG_LOCAL_TRIPLE 0
+#define DEBUG_BASIC_BLOCKS_VERBOSE 0
+#define DEBUG_CPS_RENAME_VARIABLES_HIRES 0
+#define DEBUG_SIMPLIFY_HIRES 0
+#define DEBUG_SHRINKING 0
+#define DEBUG_COALESCE_HITCHES 0
+#define DEBUG_CODE_ELIMINATION 0
+
+#define DEBUG_EXPLICIT_CLOSURES 0
+
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME give clear error messages about unused variables"
+#warning "FIXME properly handle multi dimensional arrays"
+#warning "FIXME handle multiple register sizes"
+#endif
 
 /*  Control flow graph of a loop without goto.
- * 
+ *
  *        AAA
  *   +---/
  *  /
  *   |\ GGG HHH   |   continue;
  *   | \  \   |   |
  *   |  \ III |  /
- *   |   \ | /  / 
- *   |    vvv  /  
- *   +----BBB /   
+ *   |   \ | /  /
+ *   |    vvv  /
+ *   +----BBB /
  *         | /
  *         vv
  *        JJJ
  *
- * 
+ *
  *             AAA
  *     +-----+  |  +----+
  *     |      \ | /     |
  *     |       BBB  +-+ |
  *     |       / \ /  | |
  *     |     CCC JJJ / /
- *     |     / \    / / 
- *     |   DDD EEE / /  
+ *     |     / \    / /
+ *     |   DDD EEE / /
  *     |    |   +-/ /
- *     |   FFF     /    
- *     |   / \    /     
- *     | GGG HHH /      
+ *     |   FFF     /
+ *     |   / \    /
+ *     | GGG HHH /
  *     |  |   +-/
  *     | III
- *     +--+ 
+ *     +--+
+ *
  *
- * 
  * DFlocal(X) = { Y <- Succ(X) | idom(Y) != X }
  * DFup(Z)    = { Y <- DF(Z) | idom(Y) != X }
  *
  *     DDD   EEE     DDD: [ ] ( BBB ) EEE: [ JJJ ] ()
  *      |
  *     FFF           FFF: [ ] ( BBB )
- *     / \         
+ *     / \
  *  GGG   HHH        GGG: [ ] ( BBB ) HHH: [ BBB ] ()
  *   |
  *  III              III: [ BBB ] ()
  *
  * BBB and JJJ are definitely the dominance frontier.
  * Where do I place phi functions and how do I make that decision.
- *   
+ *
  */
-static void die(char *fmt, ...)
+
+struct filelist {
+       const char *filename;
+       struct filelist *next;
+};
+
+struct filelist *include_filelist = NULL;
+
+static void __attribute__((noreturn)) die(char *fmt, ...)
 {
        va_list args;
 
@@ -98,7 +147,6 @@ static void die(char *fmt, ...)
        exit(1);
 }
 
-#define MALLOC_STRONG_DEBUG
 static void *xmalloc(size_t size, const char *name)
 {
        void *buf;
@@ -118,6 +166,17 @@ static void *xcmalloc(size_t size, const char *name)
        return buf;
 }
 
+static void *xrealloc(void *ptr, size_t size, const char *name)
+{
+       void *buf;
+       buf = realloc(ptr, size);
+       if (!buf) {
+               die("Cannot realloc %ld bytes to hold %s: %s\n",
+                       size + 0UL, name, strerror(errno));
+       }
+       return buf;
+}
+
 static void xfree(const void *ptr)
 {
        free((void *)ptr);
@@ -137,55 +196,66 @@ static char *xstrdup(const char *str)
 static void xchdir(const char *path)
 {
        if (chdir(path) != 0) {
-               die("chdir to %s failed: %s\n",
+               die("chdir to `%s' failed: %s\n",
                        path, strerror(errno));
        }
 }
 
 static int exists(const char *dirname, const char *filename)
 {
-       int does_exist = 1;
-       xchdir(dirname);
-       if (access(filename, O_RDONLY) < 0) {
+       char cwd[MAX_CWD_SIZE];
+       int does_exist;
+
+       if (getcwd(cwd, sizeof(cwd)) == 0) {
+               die("cwd buffer to small");
+       }
+
+       does_exist = 1;
+       if (chdir(dirname) != 0) {
+               does_exist = 0;
+       }
+       if (does_exist && (access(filename, O_RDONLY) < 0)) {
                if ((errno != EACCES) && (errno != EROFS)) {
                        does_exist = 0;
                }
        }
+       xchdir(cwd);
        return does_exist;
 }
 
 
 static char *slurp_file(const char *dirname, const char *filename, off_t *r_size)
 {
-       int fd;
+       char cwd[MAX_CWD_SIZE];
        char *buf;
        off_t size, progress;
        ssize_t result;
-       struct stat stats;
-       
+       FILE* file;
+
        if (!filename) {
                *r_size = 0;
                return 0;
        }
+       if (getcwd(cwd, sizeof(cwd)) == 0) {
+               die("cwd buffer to small");
+       }
        xchdir(dirname);
-       fd = open(filename, O_RDONLY);
-       if (fd < 0) {
+       file = fopen(filename, "rb");
+       xchdir(cwd);
+       if (file == NULL) {
                die("Cannot open '%s' : %s\n",
                        filename, strerror(errno));
        }
-       result = fstat(fd, &stats);
-       if (result < 0) {
-               die("Cannot stat: %s: %s\n",
-                       filename, strerror(errno));
-       }
-       size = stats.st_size;
+       fseek(file, 0, SEEK_END);
+       size = ftell(file);
+       fseek(file, 0, SEEK_SET);
        *r_size = size +1;
        buf = xmalloc(size +2, filename);
        buf[size] = '\n'; /* Make certain the file is newline terminated */
        buf[size+1] = '\0'; /* Null terminate the file for good measure */
        progress = 0;
        while(progress < size) {
-               result = read(fd, buf + progress, size - progress);
+               result = fread(buf + progress, 1, size - progress, file);
                if (result < 0) {
                        if ((errno == EINTR) || (errno == EAGAIN))
                                continue;
@@ -194,36 +264,104 @@ static char *slurp_file(const char *dirname, const char *filename, off_t *r_size
                }
                progress += result;
        }
-       result = close(fd);
-       if (result < 0) {
-               die("Close of %s failed: %s\n",
-                       filename, strerror(errno));
-       }
+       fclose(file);
        return buf;
 }
 
-/* Long on the destination platform */
-typedef unsigned long ulong_t;
-typedef long long_t;
+/* Types on the destination platform */
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME this assumes 32bit x86 is the destination"
+#endif
+typedef int8_t   schar_t;
+typedef uint8_t  uchar_t;
+typedef int8_t   char_t;
+typedef int16_t  short_t;
+typedef uint16_t ushort_t;
+typedef int32_t  int_t;
+typedef uint32_t uint_t;
+typedef int32_t  long_t;
+#define ulong_t uint32_t
+
+#define SCHAR_T_MIN (-128)
+#define SCHAR_T_MAX 127
+#define UCHAR_T_MAX 255
+#define CHAR_T_MIN  SCHAR_T_MIN
+#define CHAR_T_MAX  SCHAR_T_MAX
+#define SHRT_T_MIN  (-32768)
+#define SHRT_T_MAX  32767
+#define USHRT_T_MAX 65535
+#define INT_T_MIN   (-LONG_T_MAX - 1)
+#define INT_T_MAX   2147483647
+#define UINT_T_MAX  4294967295U
+#define LONG_T_MIN  (-LONG_T_MAX - 1)
+#define LONG_T_MAX  2147483647
+#define ULONG_T_MAX 4294967295U
+
+#define SIZEOF_I8    8
+#define SIZEOF_I16   16
+#define SIZEOF_I32   32
+#define SIZEOF_I64   64
+
+#define SIZEOF_CHAR    8
+#define SIZEOF_SHORT   16
+#define SIZEOF_INT     32
+#define SIZEOF_LONG    (sizeof(long_t)*SIZEOF_CHAR)
+
+
+#define ALIGNOF_CHAR    8
+#define ALIGNOF_SHORT   16
+#define ALIGNOF_INT     32
+#define ALIGNOF_LONG    (sizeof(long_t)*SIZEOF_CHAR)
+
+#define REG_SIZEOF_REG     32
+#define REG_SIZEOF_CHAR    REG_SIZEOF_REG
+#define REG_SIZEOF_SHORT   REG_SIZEOF_REG
+#define REG_SIZEOF_INT     REG_SIZEOF_REG
+#define REG_SIZEOF_LONG    REG_SIZEOF_REG
+
+#define REG_ALIGNOF_REG     REG_SIZEOF_REG
+#define REG_ALIGNOF_CHAR    REG_SIZEOF_REG
+#define REG_ALIGNOF_SHORT   REG_SIZEOF_REG
+#define REG_ALIGNOF_INT     REG_SIZEOF_REG
+#define REG_ALIGNOF_LONG    REG_SIZEOF_REG
+
+/* Additional definitions for clarity.
+ * I currently assume a long is the largest native
+ * machine word and that a pointer fits into it.
+ */
+#define SIZEOF_WORD     SIZEOF_LONG
+#define SIZEOF_POINTER  SIZEOF_LONG
+#define ALIGNOF_WORD    ALIGNOF_LONG
+#define ALIGNOF_POINTER ALIGNOF_LONG
+#define REG_SIZEOF_POINTER  REG_SIZEOF_LONG
+#define REG_ALIGNOF_POINTER REG_ALIGNOF_LONG
 
 struct file_state {
        struct file_state *prev;
        const char *basename;
        char *dirname;
-       char *buf;
+       const char *buf;
        off_t size;
-       char *pos;
+       const char *pos;
        int line;
-       char *line_start;
+       const char *line_start;
+       int report_line;
+       const char *report_name;
+       const char *report_dir;
+       int macro      : 1;
+       int trigraphs  : 1;
+       int join_lines : 1;
 };
 struct hash_entry;
 struct token {
        int tok;
        struct hash_entry *ident;
+       const char *pos;
        int str_len;
        union {
                ulong_t integer;
                const char *str;
+               int notmacro;
        } val;
 };
 
@@ -256,24 +394,26 @@ struct token {
 /* Operations on general purpose registers.
  */
 
-#define OP_SMUL       0
-#define OP_UMUL       1
-#define OP_SDIV       2
-#define OP_UDIV       3
-#define OP_SMOD       4
-#define OP_UMOD       5
-#define OP_ADD        6
-#define OP_SUB        7
-#define OP_SL         8
-#define OP_USR        9
-#define OP_SSR       10 
-#define OP_AND       11 
-#define OP_XOR       12
-#define OP_OR        13
-#define OP_POS       14 /* Dummy positive operator don't use it */
-#define OP_NEG       15
-#define OP_INVERT    16
-                    
+#define OP_SDIVT      0
+#define OP_UDIVT      1
+#define OP_SMUL       2
+#define OP_UMUL       3
+#define OP_SDIV       4
+#define OP_UDIV       5
+#define OP_SMOD       6
+#define OP_UMOD       7
+#define OP_ADD        8
+#define OP_SUB        9
+#define OP_SL        10
+#define OP_USR       11
+#define OP_SSR       12
+#define OP_AND       13
+#define OP_XOR       14
+#define OP_OR        15
+#define OP_POS       16 /* Dummy positive operator don't use it */
+#define OP_NEG       17
+#define OP_INVERT    18
+
 #define OP_EQ        20
 #define OP_NOTEQ     21
 #define OP_SLESS     22
@@ -284,19 +424,48 @@ struct token {
 #define OP_ULESSEQ   27
 #define OP_SMOREEQ   28
 #define OP_UMOREEQ   29
-                    
+
 #define OP_LFALSE    30  /* Test if the expression is logically false */
 #define OP_LTRUE     31  /* Test if the expression is logcially true */
 
 #define OP_LOAD      32
 #define OP_STORE     33
+/* For OP_STORE ->type holds the type
+ * RHS(0) holds the destination address
+ * RHS(1) holds the value to store.
+ */
+
+#define OP_UEXTRACT  34
+/* OP_UEXTRACT extracts an unsigned bitfield from a pseudo register
+ * RHS(0) holds the psuedo register to extract from
+ * ->type holds the size of the bitfield.
+ * ->u.bitfield.size holds the size of the bitfield.
+ * ->u.bitfield.offset holds the offset to extract from
+ */
+#define OP_SEXTRACT  35
+/* OP_SEXTRACT extracts a signed bitfield from a pseudo register
+ * RHS(0) holds the psuedo register to extract from
+ * ->type holds the size of the bitfield.
+ * ->u.bitfield.size holds the size of the bitfield.
+ * ->u.bitfield.offset holds the offset to extract from
+ */
+#define OP_DEPOSIT   36
+/* OP_DEPOSIT replaces a bitfield with a new value.
+ * RHS(0) holds the value to replace a bitifield in.
+ * RHS(1) holds the replacement value
+ * ->u.bitfield.size holds the size of the bitfield.
+ * ->u.bitfield.offset holds the deposit into
+ */
 
-#define OP_NOOP      34
+#define OP_NOOP      37
 
 #define OP_MIN_CONST 50
-#define OP_MAX_CONST 59
+#define OP_MAX_CONST 58
 #define IS_CONST_OP(X) (((X) >= OP_MIN_CONST) && ((X) <= OP_MAX_CONST))
 #define OP_INTCONST  50
+/* For OP_INTCONST ->type holds the type.
+ * ->u.cval holds the constant value.
+ */
 #define OP_BLOBCONST 51
 /* For OP_BLOBCONST ->type holds the layout and size
  * information.  u.blob holds a pointer to the raw binary
@@ -307,10 +476,16 @@ struct token {
  * MISC(0) holds the reference to the static variable.
  * ->u.cval holds an offset from that value.
  */
+#define OP_UNKNOWNVAL 59
+/* For OP_UNKNOWNAL ->type holds the type.
+ * For some reason we don't know what value this type has.
+ * This allows for variables that have don't have values
+ * assigned yet, or variables whose value we simply do not know.
+ */
 
-#define OP_WRITE     60 
+#define OP_WRITE     60
 /* OP_WRITE moves one pseudo register to another.
- * LHS(0) holds the destination pseudo register, which must be an OP_DECL.
+ * MISC(0) holds the destination pseudo register, which must be an OP_DECL.
  * RHS(0) holds the psuedo to move.
  */
 
@@ -321,14 +496,18 @@ struct token {
  * RHS(0) holds points to the triple to read from.
  */
 #define OP_COPY      62
-/* OP_COPY makes a copy of the psedo register or constant in RHS(0).
+/* OP_COPY makes a copy of the pseudo register or constant in RHS(0).
+ */
+#define OP_CONVERT   63
+/* OP_CONVERT makes a copy of the pseudo register or constant in RHS(0).
+ * And then the type is converted appropriately.
  */
-#define OP_PIECE     63
+#define OP_PIECE     64
 /* OP_PIECE returns one piece of a instruction that returns a structure.
  * MISC(0) is the instruction
  * u.cval is the LHS piece of the instruction to return.
  */
-#define OP_ASM       64
+#define OP_ASM       65
 /* OP_ASM holds a sequence of assembly instructions, the result
  * of a C asm directive.
  * RHS(x) holds input value x to the assembly sequence.
@@ -336,20 +515,27 @@ struct token {
  * u.blob holds the string of assembly instructions.
  */
 
-#define OP_DEREF     65
+#define OP_DEREF     66
 /* OP_DEREF generates an lvalue from a pointer.
  * RHS(0) holds the pointer value.
  * OP_DEREF serves as a place holder to indicate all necessary
  * checks have been done to indicate a value is an lvalue.
  */
-#define OP_DOT       66
+#define OP_DOT       67
 /* OP_DOT references a submember of a structure lvalue.
- * RHS(0) holds the lvalue.
+ * MISC(0) holds the lvalue.
  * ->u.field holds the name of the field we want.
  *
- * Not seen outside of expressions.
+ * Not seen after structures are flattened.
+ */
+#define OP_INDEX     68
+/* OP_INDEX references a submember of a tuple or array lvalue.
+ * MISC(0) holds the lvalue.
+ * ->u.cval holds the index into the lvalue.
+ *
+ * Not seen after structures are flattened.
  */
-#define OP_VAL       67
+#define OP_VAL       69
 /* OP_VAL returns the value of a subexpression of the current expression.
  * Useful for operators that have side effects.
  * RHS(0) holds the expression.
@@ -358,68 +544,97 @@ struct token {
  *
  * Not seen outside of expressions.
  */
-#define OP_LAND      68
-/* OP_LAND performs a C logical and between RHS(0) and RHS(1).
- * Not seen outside of expressions.
- */
-#define OP_LOR       69
-/* OP_LOR performs a C logical or between RHS(0) and RHS(1).
- * Not seen outside of expressions.
- */
-#define OP_COND      70
-/* OP_CODE performas a C ? : operation. 
- * RHS(0) holds the test.
- * RHS(1) holds the expression to evaluate if the test returns true.
- * RHS(2) holds the expression to evaluate if the test returns false.
- * Not seen outside of expressions.
+
+#define OP_TUPLE     70
+/* OP_TUPLE is an array of triples that are either variable
+ * or values for a structure or an array.  It is used as
+ * a place holder when flattening compound types.
+ * The value represented by an OP_TUPLE is held in N registers.
+ * LHS(0..N-1) refer to those registers.
+ * ->use is a list of statements that use the value.
+ *
+ * Although OP_TUPLE always has register sized pieces they are not
+ * used until structures are flattened/decomposed into their register
+ * components.
+ * ???? registers ????
  */
-#define OP_COMMA     71
-/* OP_COMMA performacs a C comma operation.
- * That is RHS(0) is evaluated, then RHS(1)
- * and the value of RHS(1) is returned.
- * Not seen outside of expressions.
+
+#define OP_BITREF    71
+/* OP_BITREF describes a bitfield as an lvalue.
+ * RHS(0) holds the register value.
+ * ->type holds the type of the bitfield.
+ * ->u.bitfield.size holds the size of the bitfield.
+ * ->u.bitfield.offset holds the offset of the bitfield in the register
  */
 
-#define OP_CALL      72
-/* OP_CALL performs a procedure call. 
+
+#define OP_FCALL     72
+/* OP_FCALL performs a procedure call.
  * MISC(0) holds a pointer to the OP_LIST of a function
  * RHS(x) holds argument x of a function
- * 
+ *
  * Currently not seen outside of expressions.
  */
-#define OP_VAL_VEC   74
-/* OP_VAL_VEC is an array of triples that are either variable
- * or values for a structure or an array.
- * RHS(x) holds element x of the vector.
- * triple->type->elements holds the size of the vector.
+#define OP_PROG      73
+/* OP_PROG is an expression that holds a list of statements, or
+ * expressions.  The final expression is the value of the expression.
+ * RHS(0) holds the start of the list.
  */
 
 /* statements */
 #define OP_LIST      80
-/* OP_LIST Holds a list of statements, and a result value.
+/* OP_LIST Holds a list of statements that compose a function, and a result value.
  * RHS(0) holds the list of statements.
- * MISC(0) holds the value of the statements.
+ * A list of all functions is maintained.
  */
 
-#define OP_BRANCH    81 /* branch */
+#define OP_BRANCH    81 /* an unconditional branch */
 /* For branch instructions
  * TARG(0) holds the branch target.
- * RHS(0) if present holds the branch condition.
  * ->next holds where to branch to if the branch is not taken.
- * The branch target can only be a decl...
+ * The branch target can only be a label
+ */
+
+#define OP_CBRANCH   82 /* a conditional branch */
+/* For conditional branch instructions
+ * RHS(0) holds the branch condition.
+ * TARG(0) holds the branch target.
+ * ->next holds where to branch to if the branch is not taken.
+ * The branch target can only be a label
  */
 
-#define OP_LABEL     83
+#define OP_CALL      83 /* an uncontional branch that will return */
+/* For call instructions
+ * MISC(0) holds the OP_RET that returns from the branch
+ * TARG(0) holds the branch target.
+ * ->next holds where to branch to if the branch is not taken.
+ * The branch target can only be a label
+ */
+
+#define OP_RET       84 /* an uncontinonal branch through a variable back to an OP_CALL */
+/* For call instructions
+ * RHS(0) holds the variable with the return address
+ * The branch target can only be a label
+ */
+
+#define OP_LABEL     86
 /* OP_LABEL is a triple that establishes an target for branches.
  * ->use is the list of all branches that use this label.
  */
 
-#define OP_ADECL     84 
-/* OP_DECL is a triple that establishes an lvalue for assignments.
+#define OP_ADECL     87
+/* OP_ADECL is a triple that establishes an lvalue for assignments.
+ * A variable takes N registers to contain.
+ * LHS(0..N-1) refer to an OP_PIECE triple that represents
+ * the Xth register that the variable is stored in.
  * ->use is a list of statements that use the variable.
+ *
+ * Although OP_ADECL always has register sized pieces they are not
+ * used until structures are flattened/decomposed into their register
+ * components.
  */
 
-#define OP_SDECL     85
+#define OP_SDECL     88
 /* OP_SDECL is a triple that establishes a variable of static
  * storage duration.
  * ->use is a list of statements that use the variable.
@@ -427,13 +642,13 @@ struct token {
  */
 
 
-#define OP_PHI       86
-/* OP_PHI is a triple used in SSA form code.  
+#define OP_PHI       89
+/* OP_PHI is a triple used in SSA form code.
  * It is used when multiple code paths merge and a variable needs
  * a single assignment from any of those code paths.
  * The operation is a cross between OP_DECL and OP_WRITE, which
- * is what OP_PHI is geneared from.
- * 
+ * is what OP_PHI is generated from.
+ *
  * RHS(x) points to the value from code path x
  * The number of RHS entries is the number of control paths into the block
  * in which OP_PHI resides.  The elements of the array point to point
@@ -442,6 +657,46 @@ struct token {
  * MISC(0) holds a pointer to the orginal OP_DECL node.
  */
 
+#if 0
+/* continuation helpers
+ */
+#define OP_CPS_BRANCH    90 /* an unconditional branch */
+/* OP_CPS_BRANCH calls a continuation
+ * RHS(x) holds argument x of the function
+ * TARG(0) holds OP_CPS_START target
+ */
+#define OP_CPS_CBRANCH   91  /* a conditional branch */
+/* OP_CPS_CBRANCH conditionally calls one of two continuations
+ * RHS(0) holds the branch condition
+ * RHS(x + 1) holds argument x of the function
+ * TARG(0) holds the OP_CPS_START to jump to when true
+ * ->next holds the OP_CPS_START to jump to when false
+ */
+#define OP_CPS_CALL      92  /* an uncontional branch that will return */
+/* For OP_CPS_CALL instructions
+ * RHS(x) holds argument x of the function
+ * MISC(0) holds the OP_CPS_RET that returns from the branch
+ * TARG(0) holds the branch target.
+ * ->next holds where the OP_CPS_RET will return to.
+ */
+#define OP_CPS_RET       93
+/* OP_CPS_RET conditionally calls one of two continuations
+ * RHS(0) holds the variable with the return function address
+ * RHS(x + 1) holds argument x of the function
+ * The branch target may be any OP_CPS_START
+ */
+#define OP_CPS_END       94
+/* OP_CPS_END is the triple at the end of the program.
+ * For most practical purposes it is a branch.
+ */
+#define OP_CPS_START     95
+/* OP_CPS_START is a triple at the start of a continuation
+ * The arguments variables takes N registers to contain.
+ * LHS(0..N-1) refer to an OP_PIECE triple that represents
+ * the Xth register that the arguments are stored in.
+ */
+#endif
+
 /* Architecture specific instructions */
 #define OP_CMP         100
 #define OP_TEST        101
@@ -484,12 +739,21 @@ struct token {
 struct op_info {
        const char *name;
        unsigned flags;
-#define PURE   1
-#define IMPURE 2
+#define PURE       0x001 /* Triple has no side effects */
+#define IMPURE     0x002 /* Triple has side effects */
 #define PURE_BITS(FLAGS) ((FLAGS) & 0x3)
-#define DEF    4
-#define BLOCK  8 /* Triple stores the current block */
-       unsigned char lhs, rhs, misc, targ;
+#define DEF        0x004 /* Triple is a variable definition */
+#define BLOCK      0x008 /* Triple stores the current block */
+#define STRUCTURAL 0x010 /* Triple does not generate a machine instruction */
+#define BRANCH_BITS(FLAGS) ((FLAGS) & 0xe0 )
+#define UBRANCH    0x020 /* Triple is an unconditional branch instruction */
+#define CBRANCH    0x040 /* Triple is a conditional branch instruction */
+#define RETBRANCH  0x060 /* Triple is a return instruction */
+#define CALLBRANCH 0x080 /* Triple is a call instruction */
+#define ENDBRANCH  0x0a0 /* Triple is an end instruction */
+#define PART       0x100 /* Triple is really part of another triple */
+#define BITFIELD   0x200 /* Triple manipulates a bitfield */
+       signed char lhs, rhs, misc, targ;
 };
 
 #define OP(LHS, RHS, MISC, TARG, FLAGS, NAME) { \
@@ -501,6 +765,8 @@ struct op_info {
        .targ = (TARG), \
         }
 static const struct op_info table_ops[] = {
+[OP_SDIVT      ] = OP( 2,  2, 0, 0, PURE | BLOCK , "sdivt"),
+[OP_UDIVT      ] = OP( 2,  2, 0, 0, PURE | BLOCK , "udivt"),
 [OP_SMUL       ] = OP( 0,  2, 0, 0, PURE | DEF | BLOCK , "smul"),
 [OP_UMUL       ] = OP( 0,  2, 0, 0, PURE | DEF | BLOCK , "umul"),
 [OP_SDIV       ] = OP( 0,  2, 0, 0, PURE | DEF | BLOCK , "sdiv"),
@@ -532,42 +798,61 @@ static const struct op_info table_ops[] = {
 [OP_LFALSE     ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK , "lfalse"),
 [OP_LTRUE      ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK , "ltrue"),
 
-[OP_LOAD       ] = OP( 0,  1, 0, 0, IMPURE | DEF | BLOCK, "load"),
-[OP_STORE      ] = OP( 1,  1, 0, 0, IMPURE | BLOCK , "store"),
+[OP_LOAD       ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "load"),
+[OP_STORE      ] = OP( 0,  2, 0, 0, PURE | BLOCK , "store"),
+
+[OP_UEXTRACT   ] = OP( 0,  1, 0, 0, PURE | DEF | BITFIELD, "uextract"),
+[OP_SEXTRACT   ] = OP( 0,  1, 0, 0, PURE | DEF | BITFIELD, "sextract"),
+[OP_DEPOSIT    ] = OP( 0,  2, 0, 0, PURE | DEF | BITFIELD, "deposit"),
 
-[OP_NOOP       ] = OP( 0,  0, 0, 0, PURE | BLOCK, "noop"),
+[OP_NOOP       ] = OP( 0,  0, 0, 0, PURE | BLOCK | STRUCTURAL, "noop"),
 
 [OP_INTCONST   ] = OP( 0,  0, 0, 0, PURE | DEF, "intconst"),
-[OP_BLOBCONST  ] = OP( 0,  0, 0, 0, PURE, "blobconst"),
+[OP_BLOBCONST  ] = OP( 0,  0, 0, 0, PURE , "blobconst"),
 [OP_ADDRCONST  ] = OP( 0,  0, 1, 0, PURE | DEF, "addrconst"),
+[OP_UNKNOWNVAL ] = OP( 0,  0, 0, 0, PURE | DEF, "unknown"),
 
-[OP_WRITE      ] = OP( 1,  1, 0, 0, PURE | BLOCK, "write"),
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME is it correct for OP_WRITE to be a def?  I currently use it as one..."
+#endif
+[OP_WRITE      ] = OP( 0,  1, 1, 0, PURE | DEF | BLOCK, "write"),
 [OP_READ       ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "read"),
 [OP_COPY       ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "copy"),
-[OP_PIECE      ] = OP( 0,  0, 1, 0, PURE | DEF, "piece"),
-[OP_ASM        ] = OP(-1, -1, 0, 0, IMPURE, "asm"),
-[OP_DEREF      ] = OP( 0,  1, 0, 0, 0 | DEF | BLOCK, "deref"), 
-[OP_DOT        ] = OP( 0,  1, 0, 0, 0 | DEF | BLOCK, "dot"),
+[OP_CONVERT    ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "convert"),
+[OP_PIECE      ] = OP( 0,  0, 1, 0, PURE | DEF | STRUCTURAL | PART, "piece"),
+[OP_ASM        ] = OP(-1, -1, 0, 0, PURE, "asm"),
+[OP_DEREF      ] = OP( 0,  1, 0, 0, 0 | DEF | BLOCK, "deref"),
+[OP_DOT        ] = OP( 0,  0, 1, 0, PURE | DEF | PART, "dot"),
+[OP_INDEX      ] = OP( 0,  0, 1, 0, PURE | DEF | PART, "index"),
 
 [OP_VAL        ] = OP( 0,  1, 1, 0, 0 | DEF | BLOCK, "val"),
-[OP_LAND       ] = OP( 0,  2, 0, 0, 0 | DEF | BLOCK, "land"),
-[OP_LOR        ] = OP( 0,  2, 0, 0, 0 | DEF | BLOCK, "lor"),
-[OP_COND       ] = OP( 0,  3, 0, 0, 0 | DEF | BLOCK, "cond"),
-[OP_COMMA      ] = OP( 0,  2, 0, 0, 0 | DEF | BLOCK, "comma"),
+[OP_TUPLE      ] = OP(-1,  0, 0, 0, 0 | PURE | BLOCK | STRUCTURAL, "tuple"),
+[OP_BITREF     ] = OP( 0,  1, 0, 0, 0 | DEF | PURE | STRUCTURAL | BITFIELD, "bitref"),
 /* Call is special most it can stand in for anything so it depends on context */
-[OP_CALL       ] = OP(-1, -1, 1, 0, 0 | BLOCK, "call"),
-/* The sizes of OP_CALL and OP_VAL_VEC depend upon context */
-[OP_VAL_VEC    ] = OP( 0, -1, 0, 0, 0 | BLOCK, "valvec"),
-
-[OP_LIST       ] = OP( 0,  1, 1, 0, 0 | DEF, "list"),
-/* The number of targets for OP_BRANCH depends on context */
-[OP_BRANCH     ] = OP( 0, -1, 0, 1, PURE | BLOCK, "branch"),
-[OP_LABEL      ] = OP( 0,  0, 0, 0, PURE | BLOCK, "label"),
-[OP_ADECL      ] = OP( 0,  0, 0, 0, PURE | BLOCK, "adecl"),
-[OP_SDECL      ] = OP( 0,  0, 1, 0, PURE | BLOCK, "sdecl"),
+[OP_FCALL      ] = OP( 0, -1, 1, 0, 0 | BLOCK | CALLBRANCH, "fcall"),
+[OP_PROG       ] = OP( 0,  1, 0, 0, 0 | IMPURE | BLOCK | STRUCTURAL, "prog"),
+/* The sizes of OP_FCALL depends upon context */
+
+[OP_LIST       ] = OP( 0,  1, 1, 0, 0 | DEF | STRUCTURAL, "list"),
+[OP_BRANCH     ] = OP( 0,  0, 0, 1, PURE | BLOCK | UBRANCH, "branch"),
+[OP_CBRANCH    ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "cbranch"),
+[OP_CALL       ] = OP( 0,  0, 1, 1, PURE | BLOCK | CALLBRANCH, "call"),
+[OP_RET        ] = OP( 0,  1, 0, 0, PURE | BLOCK | RETBRANCH, "ret"),
+[OP_LABEL      ] = OP( 0,  0, 0, 0, PURE | BLOCK | STRUCTURAL, "label"),
+[OP_ADECL      ] = OP( 0,  0, 0, 0, PURE | BLOCK | STRUCTURAL, "adecl"),
+[OP_SDECL      ] = OP( 0,  0, 1, 0, PURE | BLOCK | STRUCTURAL, "sdecl"),
 /* The number of RHS elements of OP_PHI depend upon context */
 [OP_PHI        ] = OP( 0, -1, 1, 0, PURE | DEF | BLOCK, "phi"),
 
+#if 0
+[OP_CPS_BRANCH ] = OP( 0, -1, 0, 1, PURE | BLOCK | UBRANCH,     "cps_branch"),
+[OP_CPS_CBRANCH] = OP( 0, -1, 0, 1, PURE | BLOCK | CBRANCH,     "cps_cbranch"),
+[OP_CPS_CALL   ] = OP( 0, -1, 1, 1, PURE | BLOCK | CALLBRANCH,  "cps_call"),
+[OP_CPS_RET    ] = OP( 0, -1, 0, 0, PURE | BLOCK | RETBRANCH,   "cps_ret"),
+[OP_CPS_END    ] = OP( 0, -1, 0, 0, IMPURE | BLOCK | ENDBRANCH, "cps_end"),
+[OP_CPS_START  ] = OP( -1, 0, 0, 0, PURE | BLOCK | STRUCTURAL,  "cps_start"),
+#endif
+
 [OP_CMP        ] = OP( 0,  2, 0, 0, PURE | DEF | BLOCK, "cmp"),
 [OP_TEST       ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "test"),
 [OP_SET_EQ     ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "set_eq"),
@@ -580,17 +865,17 @@ static const struct op_info table_ops[] = {
 [OP_SET_ULESSEQ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "set_ulesseq"),
 [OP_SET_SMOREEQ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "set_smoreq"),
 [OP_SET_UMOREEQ] = OP( 0,  1, 0, 0, PURE | DEF | BLOCK, "set_umoreq"),
-[OP_JMP        ] = OP( 0,  0, 0, 1, PURE | BLOCK, "jmp"),
-[OP_JMP_EQ     ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_eq"),
-[OP_JMP_NOTEQ  ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_noteq"),
-[OP_JMP_SLESS  ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_sless"),
-[OP_JMP_ULESS  ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_uless"),
-[OP_JMP_SMORE  ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_smore"),
-[OP_JMP_UMORE  ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_umore"),
-[OP_JMP_SLESSEQ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_slesseq"),
-[OP_JMP_ULESSEQ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_ulesseq"),
-[OP_JMP_SMOREEQ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_smoreq"),
-[OP_JMP_UMOREEQ] = OP( 0,  1, 0, 1, PURE | BLOCK, "jmp_umoreq"),
+[OP_JMP        ] = OP( 0,  0, 0, 1, PURE | BLOCK | UBRANCH, "jmp"),
+[OP_JMP_EQ     ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_eq"),
+[OP_JMP_NOTEQ  ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_noteq"),
+[OP_JMP_SLESS  ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_sless"),
+[OP_JMP_ULESS  ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_uless"),
+[OP_JMP_SMORE  ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_smore"),
+[OP_JMP_UMORE  ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_umore"),
+[OP_JMP_SLESSEQ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_slesseq"),
+[OP_JMP_ULESSEQ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_ulesseq"),
+[OP_JMP_SMOREEQ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_smoreq"),
+[OP_JMP_UMOREEQ] = OP( 0,  1, 0, 1, PURE | BLOCK | CBRANCH, "jmp_umoreq"),
 
 [OP_INB        ] = OP( 0,  1, 0, 0, IMPURE | DEF | BLOCK, "__inb"),
 [OP_INW        ] = OP( 0,  1, 0, 0, IMPURE | DEF | BLOCK, "__inw"),
@@ -607,7 +892,7 @@ static const struct op_info table_ops[] = {
 #undef OP
 #define OP_MAX      (sizeof(table_ops)/sizeof(table_ops[0]))
 
-static const char *tops(int index) 
+static const char *tops(int index)
 {
        static const char unknown[] = "unknown op";
        if (index < 0) {
@@ -627,53 +912,62 @@ struct triple_set {
        struct triple *member;
 };
 
-#define MAX_LHS  15
-#define MAX_RHS  15
-#define MAX_MISC 15
-#define MAX_TARG 15
+#define MAX_LHS  63
+#define MAX_RHS  127
+#define MAX_MISC 3
+#define MAX_TARG 1
 
+struct occurance {
+       int count;
+       const char *filename;
+       const char *function;
+       int line;
+       int col;
+       struct occurance *parent;
+};
+struct bitfield {
+       ulong_t size : 8;
+       ulong_t offset : 24;
+};
 struct triple {
        struct triple *next, *prev;
        struct triple_set *use;
        struct type *type;
-       unsigned char op;
-       unsigned char template_id;
-       unsigned short sizes;
-#define TRIPLE_LHS(SIZES)  (((SIZES) >>  0) & 0x0f)
-#define TRIPLE_RHS(SIZES)  (((SIZES) >>  4) & 0x0f)
-#define TRIPLE_MISC(SIZES) (((SIZES) >>  8) & 0x0f)
-#define TRIPLE_TARG(SIZES) (((SIZES) >> 12) & 0x0f)
-#define TRIPLE_SIZE(SIZES) \
-       ((((SIZES) >> 0) & 0x0f) + \
-       (((SIZES) >>  4) & 0x0f) + \
-       (((SIZES) >>  8) & 0x0f) + \
-       (((SIZES) >> 12) & 0x0f))
-#define TRIPLE_SIZES(LHS, RHS, MISC, TARG) \
-       ((((LHS) & 0x0f) <<  0) | \
-       (((RHS) & 0x0f)  <<  4) | \
-       (((MISC) & 0x0f) <<  8) | \
-       (((TARG) & 0x0f) << 12))
-#define TRIPLE_LHS_OFF(SIZES)  (0)
-#define TRIPLE_RHS_OFF(SIZES)  (TRIPLE_LHS_OFF(SIZES) + TRIPLE_LHS(SIZES))
-#define TRIPLE_MISC_OFF(SIZES) (TRIPLE_RHS_OFF(SIZES) + TRIPLE_RHS(SIZES))
-#define TRIPLE_TARG_OFF(SIZES) (TRIPLE_MISC_OFF(SIZES) + TRIPLE_MISC(SIZES))
-#define LHS(PTR,INDEX) ((PTR)->param[TRIPLE_LHS_OFF((PTR)->sizes) + (INDEX)])
-#define RHS(PTR,INDEX) ((PTR)->param[TRIPLE_RHS_OFF((PTR)->sizes) + (INDEX)])
-#define TARG(PTR,INDEX) ((PTR)->param[TRIPLE_TARG_OFF((PTR)->sizes) + (INDEX)])
-#define MISC(PTR,INDEX) ((PTR)->param[TRIPLE_MISC_OFF((PTR)->sizes) + (INDEX)])
+       unsigned int op : 8;
+       unsigned int template_id : 7;
+       unsigned int lhs  : 6;
+       unsigned int rhs  : 7;
+       unsigned int misc : 2;
+       unsigned int targ : 1;
+#define TRIPLE_SIZE(TRIPLE) \
+       ((TRIPLE)->lhs + (TRIPLE)->rhs + (TRIPLE)->misc + (TRIPLE)->targ)
+#define TRIPLE_LHS_OFF(PTR)  (0)
+#define TRIPLE_RHS_OFF(PTR)  (TRIPLE_LHS_OFF(PTR) + (PTR)->lhs)
+#define TRIPLE_MISC_OFF(PTR) (TRIPLE_RHS_OFF(PTR) + (PTR)->rhs)
+#define TRIPLE_TARG_OFF(PTR) (TRIPLE_MISC_OFF(PTR) + (PTR)->misc)
+#define LHS(PTR,INDEX) ((PTR)->param[TRIPLE_LHS_OFF(PTR) + (INDEX)])
+#define RHS(PTR,INDEX) ((PTR)->param[TRIPLE_RHS_OFF(PTR) + (INDEX)])
+#define TARG(PTR,INDEX) ((PTR)->param[TRIPLE_TARG_OFF(PTR) + (INDEX)])
+#define MISC(PTR,INDEX) ((PTR)->param[TRIPLE_MISC_OFF(PTR) + (INDEX)])
        unsigned id; /* A scratch value and finally the register */
 #define TRIPLE_FLAG_FLATTENED   (1 << 31)
 #define TRIPLE_FLAG_PRE_SPLIT   (1 << 30)
 #define TRIPLE_FLAG_POST_SPLIT  (1 << 29)
-       const char *filename;
-       int line;
-       int col;
+#define TRIPLE_FLAG_VOLATILE    (1 << 28)
+#define TRIPLE_FLAG_INLINE      (1 << 27) /* ???? */
+#define TRIPLE_FLAG_LOCAL      (1 << 26)
+
+#define TRIPLE_FLAG_COPY TRIPLE_FLAG_VOLATILE
+       struct occurance *occurance;
        union {
                ulong_t cval;
+               struct bitfield bitfield;
                struct block  *block;
                void *blob;
                struct hash_entry *field;
                struct asm_info *ainfo;
+               struct triple *func;
+               struct symbol *symbol;
        } u;
        struct triple *param[2];
 };
@@ -697,8 +991,9 @@ struct block_set {
 };
 struct block {
        struct block *work_next;
-       struct block *left, *right;
        struct triple *first, *last;
+       int edge_count;
+       struct block_set *edges;
        int users;
        struct block_set *use;
        struct block_set *idominates;
@@ -708,7 +1003,7 @@ struct block {
        struct block_set *ipdomfrontier;
        struct block *ipdom;
        int vertex;
-       
+
 };
 
 struct symbol {
@@ -719,10 +1014,16 @@ struct symbol {
        int scope_depth;
 };
 
+struct macro_arg {
+       struct macro_arg *next;
+       struct hash_entry *ident;
+};
 struct macro {
        struct hash_entry *ident;
-       char *buf;
+       const char *buf;
        int buf_len;
+       struct macro_arg *args;
+       int argc;
 };
 
 struct hash_entry {
@@ -732,66 +1033,115 @@ struct hash_entry {
        int tok;
        struct macro *sym_define;
        struct symbol *sym_label;
-       struct symbol *sym_struct;
+       struct symbol *sym_tag;
        struct symbol *sym_ident;
 };
 
 #define HASH_TABLE_SIZE 2048
 
-struct compile_state {
+struct compiler_state {
        const char *label_prefix;
        const char *ofilename;
+       unsigned long flags;
+       unsigned long debug;
+       unsigned long max_allocation_passes;
+
+       size_t include_path_count;
+       const char **include_paths;
+
+       size_t define_count;
+       const char **defines;
+
+       size_t undef_count;
+       const char **undefs;
+};
+struct arch_state {
+       unsigned long features;
+};
+struct basic_blocks {
+       struct triple *func;
+       struct triple *first;
+       struct block *first_block, *last_block;
+       int last_vertex;
+};
+#define MAX_PP_IF_DEPTH 63
+struct compile_state {
+       struct compiler_state *compiler;
+       struct arch_state *arch;
        FILE *output;
-       struct triple *vars;
+       FILE *errout;
+       FILE *dbgout;
        struct file_state *file;
-       struct token token[4];
+       struct occurance *last_occurance;
+       const char *function;
+       int    token_base;
+       struct token token[6];
        struct hash_entry *hash_table[HASH_TABLE_SIZE];
+       struct hash_entry *i_switch;
+       struct hash_entry *i_case;
        struct hash_entry *i_continue;
        struct hash_entry *i_break;
+       struct hash_entry *i_default;
+       struct hash_entry *i_return;
+       struct hash_entry *i_noreturn;
+       /* Additional hash entries for predefined macros */
+       struct hash_entry *i_defined;
+       struct hash_entry *i___VA_ARGS__;
+       struct hash_entry *i___FILE__;
+       struct hash_entry *i___LINE__;
+       /* Additional hash entries for predefined identifiers */
+       struct hash_entry *i___func__;
+       /* Additional hash entries for attributes */
+       struct hash_entry *i_noinline;
+       struct hash_entry *i_always_inline;
        int scope_depth;
-       int if_depth, if_value;
-       int macro_line;
+       unsigned char if_bytes[(MAX_PP_IF_DEPTH + CHAR_BIT -1)/CHAR_BIT];
+       int if_depth;
+       int eat_depth, eat_targ;
        struct file_state *macro_file;
+       struct triple *functions;
        struct triple *main_function;
-       struct block *first_block, *last_block;
-       int last_vertex;
-       int cpu;
-       int debug;
-       int optimize;
+       struct triple *first;
+       struct triple *global_pool;
+       struct basic_blocks bb;
+       int functions_joined;
 };
 
 /* visibility global/local */
 /* static/auto duration */
 /* typedef, register, inline */
 #define STOR_SHIFT         0
-#define STOR_MASK     0x000f
+#define STOR_MASK     0x001f
 /* Visibility */
 #define STOR_GLOBAL   0x0001
 /* Duration */
 #define STOR_PERM     0x0002
+/* Definition locality */
+#define STOR_NONLOCAL 0x0004  /* The definition is not in this translation unit */
 /* Storage specifiers */
 #define STOR_AUTO     0x0000
 #define STOR_STATIC   0x0002
-#define STOR_EXTERN   0x0003
-#define STOR_REGISTER 0x0004
-#define STOR_TYPEDEF  0x0008
-#define STOR_INLINE   0x000c
-
-#define QUAL_SHIFT         4
-#define QUAL_MASK     0x0070
+#define STOR_LOCAL    0x0003
+#define STOR_EXTERN   0x0007
+#define STOR_INLINE   0x0008
+#define STOR_REGISTER 0x0010
+#define STOR_TYPEDEF  0x0018
+
+#define QUAL_SHIFT         5
+#define QUAL_MASK     0x00e0
 #define QUAL_NONE     0x0000
-#define QUAL_CONST    0x0010
-#define QUAL_VOLATILE 0x0020
-#define QUAL_RESTRICT 0x0040
+#define QUAL_CONST    0x0020
+#define QUAL_VOLATILE 0x0040
+#define QUAL_RESTRICT 0x0080
 
 #define TYPE_SHIFT         8
 #define TYPE_MASK     0x1f00
-#define TYPE_INTEGER(TYPE)    (((TYPE) >= TYPE_CHAR) && ((TYPE) <= TYPE_ULLONG))
-#define TYPE_ARITHMETIC(TYPE) (((TYPE) >= TYPE_CHAR) && ((TYPE) <= TYPE_LDOUBLE))
+#define TYPE_INTEGER(TYPE)    ((((TYPE) >= TYPE_CHAR) && ((TYPE) <= TYPE_ULLONG)) || ((TYPE) == TYPE_ENUM) || ((TYPE) == TYPE_BITFIELD))
+#define TYPE_ARITHMETIC(TYPE) ((((TYPE) >= TYPE_CHAR) && ((TYPE) <= TYPE_LDOUBLE)) || ((TYPE) == TYPE_ENUM) || ((TYPE) == TYPE_BITFIELD))
 #define TYPE_UNSIGNED(TYPE)   ((TYPE) & 0x0100)
 #define TYPE_SIGNED(TYPE)     (!TYPE_UNSIGNED(TYPE))
-#define TYPE_MKUNSIGNED(TYPE) ((TYPE) | 0x0100)
-#define TYPE_RANK(TYPE)       ((TYPE) & ~0x0100)
+#define TYPE_MKUNSIGNED(TYPE) (((TYPE) & ~0xF000) | 0x0100)
+#define TYPE_RANK(TYPE)       ((TYPE) & ~0xF1FF)
 #define TYPE_PTR(TYPE)        (((TYPE) & TYPE_MASK) == TYPE_POINTER)
 #define TYPE_DEFAULT  0x0000
 #define TYPE_VOID     0x0100
@@ -808,16 +1158,36 @@ struct compile_state {
 #define TYPE_FLOAT    0x0c00
 #define TYPE_DOUBLE   0x0d00
 #define TYPE_LDOUBLE  0x0e00 /* long double */
+
+/* Note: TYPE_ENUM is chosen very carefully so TYPE_RANK works */
+#define TYPE_ENUM     0x1600
+#define TYPE_LIST     0x1700
+/* TYPE_LIST is a basic building block when defining enumerations
+ * type->field_ident holds the name of this enumeration entry.
+ * type->right holds the entry in the list.
+ */
+
 #define TYPE_STRUCT   0x1000
-#define TYPE_ENUM     0x1100
-#define TYPE_POINTER  0x1200 
+/* For TYPE_STRUCT
+ * type->left holds the link list of TYPE_PRODUCT entries that
+ * make up the structure.
+ * type->elements hold the length of the linked list
+ */
+#define TYPE_UNION    0x1100
+/* For TYPE_UNION
+ * type->left holds the link list of TYPE_OVERLAP entries that
+ * make up the union.
+ * type->elements hold the length of the linked list
+ */
+#define TYPE_POINTER  0x1200
 /* For TYPE_POINTER:
  * type->left holds the type pointed to.
  */
-#define TYPE_FUNCTION 0x1300 
+#define TYPE_FUNCTION 0x1300
 /* For TYPE_FUNCTION:
  * type->left holds the return type.
- * type->right holds the...
+ * type->right holds the type of the arguments
+ * type->elements holds the count of the arguments
  */
 #define TYPE_PRODUCT  0x1400
 /* TYPE_PRODUCT is a basic building block when defining structures
@@ -829,13 +1199,45 @@ struct compile_state {
  * type->left and type->right holds to types that overlap
  * each other in memory.
  */
-#define TYPE_ARRAY    0x1600
+#define TYPE_ARRAY    0x1800
 /* TYPE_ARRAY is a basic building block when definitng arrays.
  * type->left holds the type we are an array of.
- * type-> holds the number of elements.
+ * type->elements holds the number of elements.
+ */
+#define TYPE_TUPLE    0x1900
+/* TYPE_TUPLE is a basic building block when defining
+ * positionally reference type conglomerations. (i.e. closures)
+ * In essence it is a wrapper for TYPE_PRODUCT, like TYPE_STRUCT
+ * except it has no field names.
+ * type->left holds the liked list of TYPE_PRODUCT entries that
+ * make up the closure type.
+ * type->elements hold the number of elements in the closure.
+ */
+#define TYPE_JOIN     0x1a00
+/* TYPE_JOIN is a basic building block when defining
+ * positionally reference type conglomerations. (i.e. closures)
+ * In essence it is a wrapper for TYPE_OVERLAP, like TYPE_UNION
+ * except it has no field names.
+ * type->left holds the liked list of TYPE_OVERLAP entries that
+ * make up the closure type.
+ * type->elements hold the number of elements in the closure.
+ */
+#define TYPE_BITFIELD 0x1b00
+/* TYPE_BITFIED is the type of a bitfield.
+ * type->left holds the type basic type TYPE_BITFIELD is derived from.
+ * type->elements holds the number of bits in the bitfield.
+ */
+#define TYPE_UNKNOWN  0x1c00
+/* TYPE_UNKNOWN is the type of an unknown value.
+ * Used on unknown consts and other places where I don't know the type.
  */
 
-#define ELEMENT_COUNT_UNSPECIFIED (~0UL)
+#define ATTRIB_SHIFT                 16
+#define ATTRIB_MASK          0xffff0000
+#define ATTRIB_NOINLINE      0x00010000
+#define ATTRIB_ALWAYS_INLINE 0x00020000
+
+#define ELEMENT_COUNT_UNSPECIFIED ULONG_T_MAX
 
 struct type {
        unsigned int type;
@@ -845,29 +1247,54 @@ struct type {
        struct hash_entry *type_ident;
 };
 
-#define MAX_REGISTERS      75
+#define TEMPLATE_BITS      7
+#define MAX_TEMPLATES      (1<<TEMPLATE_BITS)
 #define MAX_REG_EQUIVS     16
-#define REGISTER_BITS      28
+#define MAX_REGC           14
+#define MAX_REGISTERS      75
+#define REGISTER_BITS      7
 #define MAX_VIRT_REGISTERS (1<<REGISTER_BITS)
-#define TEMPLATE_BITS      6
-#define MAX_TEMPLATES      (1<<TEMPLATE_BITS)
-#define MAX_REGC           12
-#define REG_UNSET          0
-#define REG_UNNEEDED       1
+#define REG_ERROR          0
+#define REG_UNSET          1
+#define REG_UNNEEDED       2
 #define REG_VIRT0          (MAX_REGISTERS + 0)
 #define REG_VIRT1          (MAX_REGISTERS + 1)
 #define REG_VIRT2          (MAX_REGISTERS + 2)
 #define REG_VIRT3          (MAX_REGISTERS + 3)
 #define REG_VIRT4          (MAX_REGISTERS + 4)
 #define REG_VIRT5          (MAX_REGISTERS + 5)
+#define REG_VIRT6          (MAX_REGISTERS + 6)
+#define REG_VIRT7          (MAX_REGISTERS + 7)
+#define REG_VIRT8          (MAX_REGISTERS + 8)
+#define REG_VIRT9          (MAX_REGISTERS + 9)
+
+#if (MAX_REGISTERS + 9) > MAX_VIRT_REGISTERS
+#error "MAX_VIRT_REGISTERS to small"
+#endif
+#if (MAX_REGC + REGISTER_BITS) >= 26
+#error "Too many id bits used"
+#endif
 
 /* Provision for 8 register classes */
+#define REG_SHIFT  0
+#define REGC_SHIFT REGISTER_BITS
+#define REGC_MASK (((1 << MAX_REGC) - 1) << REGISTER_BITS)
 #define REG_MASK (MAX_VIRT_REGISTERS -1)
 #define ID_REG(ID)              ((ID) & REG_MASK)
 #define SET_REG(ID, REG)        ((ID) = (((ID) & ~REG_MASK) | ((REG) & REG_MASK)))
+#define ID_REGCM(ID)           (((ID) & REGC_MASK) >> REGC_SHIFT)
+#define SET_REGCM(ID, REGCM)   ((ID) = (((ID) & ~REGC_MASK) | (((REGCM) << REGC_SHIFT) & REGC_MASK)))
+#define SET_INFO(ID, INFO)     ((ID) = (((ID) & ~(REG_MASK | REGC_MASK)) | \
+               (((INFO).reg) & REG_MASK) | ((((INFO).regcm) << REGC_SHIFT) & REGC_MASK)))
 
+#define ARCH_INPUT_REGS 4
+#define ARCH_OUTPUT_REGS 4
+
+static const struct reg_info arch_input_regs[ARCH_INPUT_REGS];
+static const struct reg_info arch_output_regs[ARCH_OUTPUT_REGS];
 static unsigned arch_reg_regcm(struct compile_state *state, int reg);
 static unsigned arch_regcm_normalize(struct compile_state *state, unsigned regcm);
+static unsigned arch_regcm_reg_normalize(struct compile_state *state, unsigned regcm);
 static void arch_reg_equivs(
        struct compile_state *state, unsigned *equiv, int reg);
 static int arch_select_free_register(
@@ -880,49 +1307,442 @@ static struct reg_info arch_reg_constraint(
        struct compile_state *state, struct type *type, const char *constraint);
 static struct reg_info arch_reg_clobber(
        struct compile_state *state, const char *clobber);
-static struct reg_info arch_reg_lhs(struct compile_state *state, 
+static struct reg_info arch_reg_lhs(struct compile_state *state,
        struct triple *ins, int index);
-static struct reg_info arch_reg_rhs(struct compile_state *state, 
+static struct reg_info arch_reg_rhs(struct compile_state *state,
        struct triple *ins, int index);
+static int arch_reg_size(int reg);
 static struct triple *transform_to_arch_instruction(
        struct compile_state *state, struct triple *ins);
+static struct triple *flatten(
+       struct compile_state *state, struct triple *first, struct triple *ptr);
+static void print_dominators(struct compile_state *state,
+       FILE *fp, struct basic_blocks *bb);
+static void print_dominance_frontiers(struct compile_state *state,
+       FILE *fp, struct basic_blocks *bb);
+
+
+
+#define DEBUG_ABORT_ON_ERROR    0x00000001
+#define DEBUG_BASIC_BLOCKS      0x00000002
+#define DEBUG_FDOMINATORS       0x00000004
+#define DEBUG_RDOMINATORS       0x00000008
+#define DEBUG_TRIPLES           0x00000010
+#define DEBUG_INTERFERENCE      0x00000020
+#define DEBUG_SCC_TRANSFORM     0x00000040
+#define DEBUG_SCC_TRANSFORM2    0x00000080
+#define DEBUG_REBUILD_SSA_FORM  0x00000100
+#define DEBUG_INLINE            0x00000200
+#define DEBUG_RANGE_CONFLICTS   0x00000400
+#define DEBUG_RANGE_CONFLICTS2  0x00000800
+#define DEBUG_COLOR_GRAPH       0x00001000
+#define DEBUG_COLOR_GRAPH2      0x00002000
+#define DEBUG_COALESCING        0x00004000
+#define DEBUG_COALESCING2       0x00008000
+#define DEBUG_VERIFICATION     0x00010000
+#define DEBUG_CALLS            0x00020000
+#define DEBUG_CALLS2           0x00040000
+#define DEBUG_TOKENS            0x80000000
+
+#define DEBUG_DEFAULT ( \
+       DEBUG_ABORT_ON_ERROR | \
+       DEBUG_BASIC_BLOCKS | \
+       DEBUG_FDOMINATORS | \
+       DEBUG_RDOMINATORS | \
+       DEBUG_TRIPLES | \
+       0 )
+
+#define DEBUG_ALL ( \
+       DEBUG_ABORT_ON_ERROR   | \
+       DEBUG_BASIC_BLOCKS     | \
+       DEBUG_FDOMINATORS      | \
+       DEBUG_RDOMINATORS      | \
+       DEBUG_TRIPLES          | \
+       DEBUG_INTERFERENCE     | \
+       DEBUG_SCC_TRANSFORM    | \
+       DEBUG_SCC_TRANSFORM2   | \
+       DEBUG_REBUILD_SSA_FORM | \
+       DEBUG_INLINE           | \
+       DEBUG_RANGE_CONFLICTS  | \
+       DEBUG_RANGE_CONFLICTS2 | \
+       DEBUG_COLOR_GRAPH      | \
+       DEBUG_COLOR_GRAPH2     | \
+       DEBUG_COALESCING       | \
+       DEBUG_COALESCING2      | \
+       DEBUG_VERIFICATION     | \
+       DEBUG_CALLS            | \
+       DEBUG_CALLS2           | \
+       DEBUG_TOKENS           | \
+       0 )
+
+#define COMPILER_INLINE_MASK               0x00000007
+#define COMPILER_INLINE_ALWAYS             0x00000000
+#define COMPILER_INLINE_NEVER              0x00000001
+#define COMPILER_INLINE_DEFAULTON          0x00000002
+#define COMPILER_INLINE_DEFAULTOFF         0x00000003
+#define COMPILER_INLINE_NOPENALTY          0x00000004
+#define COMPILER_ELIMINATE_INEFECTUAL_CODE 0x00000008
+#define COMPILER_SIMPLIFY                  0x00000010
+#define COMPILER_SCC_TRANSFORM             0x00000020
+#define COMPILER_SIMPLIFY_OP               0x00000040
+#define COMPILER_SIMPLIFY_PHI              0x00000080
+#define COMPILER_SIMPLIFY_LABEL            0x00000100
+#define COMPILER_SIMPLIFY_BRANCH           0x00000200
+#define COMPILER_SIMPLIFY_COPY             0x00000400
+#define COMPILER_SIMPLIFY_ARITH            0x00000800
+#define COMPILER_SIMPLIFY_SHIFT            0x00001000
+#define COMPILER_SIMPLIFY_BITWISE          0x00002000
+#define COMPILER_SIMPLIFY_LOGICAL          0x00004000
+#define COMPILER_SIMPLIFY_BITFIELD         0x00008000
+
+#define COMPILER_TRIGRAPHS                 0x40000000
+#define COMPILER_PP_ONLY                   0x80000000
+
+#define COMPILER_DEFAULT_FLAGS ( \
+       COMPILER_TRIGRAPHS | \
+       COMPILER_ELIMINATE_INEFECTUAL_CODE | \
+       COMPILER_INLINE_DEFAULTON | \
+       COMPILER_SIMPLIFY_OP | \
+       COMPILER_SIMPLIFY_PHI | \
+       COMPILER_SIMPLIFY_LABEL | \
+       COMPILER_SIMPLIFY_BRANCH | \
+       COMPILER_SIMPLIFY_COPY | \
+       COMPILER_SIMPLIFY_ARITH | \
+       COMPILER_SIMPLIFY_SHIFT | \
+       COMPILER_SIMPLIFY_BITWISE | \
+       COMPILER_SIMPLIFY_LOGICAL | \
+       COMPILER_SIMPLIFY_BITFIELD | \
+       0 )
+
+#define GLOBAL_SCOPE_DEPTH   1
+#define FUNCTION_SCOPE_DEPTH (GLOBAL_SCOPE_DEPTH + 1)
+
+static void compile_file(struct compile_state *old_state, const char *filename, int local);
 
 
 
-#define DEBUG_ABORT_ON_ERROR    0x0001
-#define DEBUG_INTERMEDIATE_CODE 0x0002
-#define DEBUG_CONTROL_FLOW      0x0004
-#define DEBUG_BASIC_BLOCKS      0x0008
-#define DEBUG_FDOMINATORS       0x0010
-#define DEBUG_RDOMINATORS       0x0020
-#define DEBUG_TRIPLES           0x0040
-#define DEBUG_INTERFERENCE      0x0080
-#define DEBUG_ARCH_CODE         0x0100
-#define DEBUG_CODE_ELIMINATION  0x0200
-#define DEBUG_INSERTED_COPIES   0x0400
+static void init_compiler_state(struct compiler_state *compiler)
+{
+       memset(compiler, 0, sizeof(*compiler));
+       compiler->label_prefix = "";
+       compiler->ofilename = "auto.inc";
+       compiler->flags = COMPILER_DEFAULT_FLAGS;
+       compiler->debug = 0;
+       compiler->max_allocation_passes = MAX_ALLOCATION_PASSES;
+       compiler->include_path_count = 1;
+       compiler->include_paths      = xcmalloc(sizeof(char *), "include_paths");
+       compiler->define_count       = 1;
+       compiler->defines            = xcmalloc(sizeof(char *), "defines");
+       compiler->undef_count        = 1;
+       compiler->undefs             = xcmalloc(sizeof(char *), "undefs");
+}
 
-#define GLOBAL_SCOPE_DEPTH 1
+struct compiler_flag {
+       const char *name;
+       unsigned long flag;
+};
 
-static void compile_file(struct compile_state *old_state, const char *filename, int local);
+struct compiler_arg {
+       const char *name;
+       unsigned long mask;
+       struct compiler_flag flags[16];
+};
+
+static int set_flag(
+       const struct compiler_flag *ptr, unsigned long *flags,
+       int act, const char *flag)
+{
+       int result = -1;
+       for(; ptr->name; ptr++) {
+               if (strcmp(ptr->name, flag) == 0) {
+                       break;
+               }
+       }
+       if (ptr->name) {
+               result = 0;
+               *flags &= ~(ptr->flag);
+               if (act) {
+                       *flags |= ptr->flag;
+               }
+       }
+       return result;
+}
+
+static int set_arg(
+       const struct compiler_arg *ptr, unsigned long *flags, const char *arg)
+{
+       const char *val;
+       int result = -1;
+       int len;
+       val = strchr(arg, '=');
+       if (val) {
+               len = val - arg;
+               val++;
+               for(; ptr->name; ptr++) {
+                       if (strncmp(ptr->name, arg, len) == 0) {
+                               break;
+                       }
+               }
+               if (ptr->name) {
+                       *flags &= ~ptr->mask;
+                       result = set_flag(&ptr->flags[0], flags, 1, val);
+               }
+       }
+       return result;
+}
+
+
+static void flag_usage(FILE *fp, const struct compiler_flag *ptr,
+       const char *prefix, const char *invert_prefix)
+{
+       for(;ptr->name; ptr++) {
+               fprintf(fp, "%s%s\n", prefix, ptr->name);
+               if (invert_prefix) {
+                       fprintf(fp, "%s%s\n", invert_prefix, ptr->name);
+               }
+       }
+}
+
+static void arg_usage(FILE *fp, const struct compiler_arg *ptr,
+       const char *prefix)
+{
+       for(;ptr->name; ptr++) {
+               const struct compiler_flag *flag;
+               for(flag = &ptr->flags[0]; flag->name; flag++) {
+                       fprintf(fp, "%s%s=%s\n",
+                               prefix, ptr->name, flag->name);
+               }
+       }
+}
+
+static int append_string(size_t *max, const char ***vec, const char *str,
+       const char *name)
+{
+       size_t count;
+       count = ++(*max);
+       *vec = xrealloc(*vec, sizeof(char *)*count, "name");
+       (*vec)[count -1] = 0;
+       (*vec)[count -2] = str;
+       return 0;
+}
+
+static void arg_error(char *fmt, ...);
+static const char *identifier(const char *str, const char *end);
+
+static int append_include_path(struct compiler_state *compiler, const char *str)
+{
+       int result;
+       if (!exists(str, ".")) {
+               arg_error("Nonexistent include path: `%s'\n",
+                       str);
+       }
+       result = append_string(&compiler->include_path_count,
+               &compiler->include_paths, str, "include_paths");
+       return result;
+}
+
+static int append_define(struct compiler_state *compiler, const char *str)
+{
+       const char *end, *rest;
+       int result;
+
+       end = strchr(str, '=');
+       if (!end) {
+               end = str + strlen(str);
+       }
+       rest = identifier(str, end);
+       if (rest != end) {
+               int len = end - str - 1;
+               arg_error("Invalid name cannot define macro: `%*.*s'\n",
+                       len, len, str);
+       }
+       result = append_string(&compiler->define_count,
+               &compiler->defines, str, "defines");
+       return result;
+}
+
+static int append_undef(struct compiler_state *compiler, const char *str)
+{
+       const char *end, *rest;
+       int result;
+
+       end = str + strlen(str);
+       rest = identifier(str, end);
+       if (rest != end) {
+               int len = end - str - 1;
+               arg_error("Invalid name cannot undefine macro: `%*.*s'\n",
+                       len, len, str);
+       }
+       result = append_string(&compiler->undef_count,
+               &compiler->undefs, str, "undefs");
+       return result;
+}
+
+static const struct compiler_flag romcc_flags[] = {
+       { "trigraphs",                 COMPILER_TRIGRAPHS },
+       { "pp-only",                   COMPILER_PP_ONLY },
+       { "eliminate-inefectual-code", COMPILER_ELIMINATE_INEFECTUAL_CODE },
+       { "simplify",                  COMPILER_SIMPLIFY },
+       { "scc-transform",             COMPILER_SCC_TRANSFORM },
+       { "simplify-op",               COMPILER_SIMPLIFY_OP },
+       { "simplify-phi",              COMPILER_SIMPLIFY_PHI },
+       { "simplify-label",            COMPILER_SIMPLIFY_LABEL },
+       { "simplify-branch",           COMPILER_SIMPLIFY_BRANCH },
+       { "simplify-copy",             COMPILER_SIMPLIFY_COPY },
+       { "simplify-arith",            COMPILER_SIMPLIFY_ARITH },
+       { "simplify-shift",            COMPILER_SIMPLIFY_SHIFT },
+       { "simplify-bitwise",          COMPILER_SIMPLIFY_BITWISE },
+       { "simplify-logical",          COMPILER_SIMPLIFY_LOGICAL },
+       { "simplify-bitfield",         COMPILER_SIMPLIFY_BITFIELD },
+       { 0, 0 },
+};
+static const struct compiler_arg romcc_args[] = {
+       { "inline-policy",             COMPILER_INLINE_MASK,
+               {
+                       { "always",      COMPILER_INLINE_ALWAYS, },
+                       { "never",       COMPILER_INLINE_NEVER, },
+                       { "defaulton",   COMPILER_INLINE_DEFAULTON, },
+                       { "defaultoff",  COMPILER_INLINE_DEFAULTOFF, },
+                       { "nopenalty",   COMPILER_INLINE_NOPENALTY, },
+                       { 0, 0 },
+               },
+       },
+       { 0, 0 },
+};
+static const struct compiler_flag romcc_opt_flags[] = {
+       { "-O",  COMPILER_SIMPLIFY },
+       { "-O2", COMPILER_SIMPLIFY | COMPILER_SCC_TRANSFORM },
+       { "-E",  COMPILER_PP_ONLY },
+       { 0, 0, },
+};
+static const struct compiler_flag romcc_debug_flags[] = {
+       { "all",                   DEBUG_ALL },
+       { "abort-on-error",        DEBUG_ABORT_ON_ERROR },
+       { "basic-blocks",          DEBUG_BASIC_BLOCKS },
+       { "fdominators",           DEBUG_FDOMINATORS },
+       { "rdominators",           DEBUG_RDOMINATORS },
+       { "triples",               DEBUG_TRIPLES },
+       { "interference",          DEBUG_INTERFERENCE },
+       { "scc-transform",         DEBUG_SCC_TRANSFORM },
+       { "scc-transform2",        DEBUG_SCC_TRANSFORM2 },
+       { "rebuild-ssa-form",      DEBUG_REBUILD_SSA_FORM },
+       { "inline",                DEBUG_INLINE },
+       { "live-range-conflicts",  DEBUG_RANGE_CONFLICTS },
+       { "live-range-conflicts2", DEBUG_RANGE_CONFLICTS2 },
+       { "color-graph",           DEBUG_COLOR_GRAPH },
+       { "color-graph2",          DEBUG_COLOR_GRAPH2 },
+       { "coalescing",            DEBUG_COALESCING },
+       { "coalescing2",           DEBUG_COALESCING2 },
+       { "verification",          DEBUG_VERIFICATION },
+       { "calls",                 DEBUG_CALLS },
+       { "calls2",                DEBUG_CALLS2 },
+       { "tokens",                DEBUG_TOKENS },
+       { 0, 0 },
+};
+
+static int compiler_encode_flag(
+       struct compiler_state *compiler, const char *flag)
+{
+       int act;
+       int result;
+
+       act = 1;
+       result = -1;
+       if (strncmp(flag, "no-", 3) == 0) {
+               flag += 3;
+               act = 0;
+       }
+       if (strncmp(flag, "-O", 2) == 0) {
+               result = set_flag(romcc_opt_flags, &compiler->flags, act, flag);
+       }
+       else if (strncmp(flag, "-E", 2) == 0) {
+               result = set_flag(romcc_opt_flags, &compiler->flags, act, flag);
+       }
+       else if (strncmp(flag, "-I", 2) == 0) {
+               result = append_include_path(compiler, flag + 2);
+       }
+       else if (strncmp(flag, "-D", 2) == 0) {
+               result = append_define(compiler, flag + 2);
+       }
+       else if (strncmp(flag, "-U", 2) == 0) {
+               result = append_undef(compiler, flag + 2);
+       }
+       else if (act && strncmp(flag, "label-prefix=", 13) == 0) {
+               result = 0;
+               compiler->label_prefix = flag + 13;
+       }
+       else if (act && strncmp(flag, "max-allocation-passes=", 22) == 0) {
+               unsigned long max_passes;
+               char *end;
+               max_passes = strtoul(flag + 22, &end, 10);
+               if (end[0] == '\0') {
+                       result = 0;
+                       compiler->max_allocation_passes = max_passes;
+               }
+       }
+       else if (act && strcmp(flag, "debug") == 0) {
+               result = 0;
+               compiler->debug |= DEBUG_DEFAULT;
+       }
+       else if (strncmp(flag, "debug-", 6) == 0) {
+               flag += 6;
+               result = set_flag(romcc_debug_flags, &compiler->debug, act, flag);
+       }
+       else {
+               result = set_flag(romcc_flags, &compiler->flags, act, flag);
+               if (result < 0) {
+                       result = set_arg(romcc_args, &compiler->flags, flag);
+               }
+       }
+       return result;
+}
+
+static void compiler_usage(FILE *fp)
+{
+       flag_usage(fp, romcc_opt_flags, "", 0);
+       flag_usage(fp, romcc_flags, "-f", "-fno-");
+       arg_usage(fp,  romcc_args, "-f");
+       flag_usage(fp, romcc_debug_flags, "-fdebug-", "-fno-debug-");
+       fprintf(fp, "-flabel-prefix=<prefix for assembly language labels>\n");
+       fprintf(fp, "--label-prefix=<prefix for assembly language labels>\n");
+       fprintf(fp, "-I<include path>\n");
+       fprintf(fp, "-D<macro>[=defn]\n");
+       fprintf(fp, "-U<macro>\n");
+}
 
 static void do_cleanup(struct compile_state *state)
 {
        if (state->output) {
                fclose(state->output);
-               unlink(state->ofilename);
+               unlink(state->compiler->ofilename);
+               state->output = 0;
+       }
+       if (state->dbgout) {
+               fflush(state->dbgout);
+       }
+       if (state->errout) {
+               fflush(state->errout);
+       }
+}
+
+static struct compile_state *exit_state;
+static void exit_cleanup(void)
+{
+       if (exit_state) {
+               do_cleanup(exit_state);
        }
 }
 
 static int get_col(struct file_state *file)
 {
        int col;
-       char *ptr, *end;
+       const char *ptr, *end;
        ptr = file->line_start;
        end = file->pos;
        for(col = 0; ptr < end; ptr++) {
                if (*ptr != '\t') {
                        col++;
-               } 
+               }
                else {
                        col = (col & ~7) + 8;
                }
@@ -933,90 +1753,97 @@ static int get_col(struct file_state *file)
 static void loc(FILE *fp, struct compile_state *state, struct triple *triple)
 {
        int col;
-       if (triple) {
-               fprintf(fp, "%s:%d.%d: ", 
-                       triple->filename, triple->line, triple->col);
+       if (triple && triple->occurance) {
+               struct occurance *spot;
+               for(spot = triple->occurance; spot; spot = spot->parent) {
+                       fprintf(fp, "%s:%d.%d: ",
+                               spot->filename, spot->line, spot->col);
+               }
                return;
        }
        if (!state->file) {
                return;
        }
        col = get_col(state->file);
-       fprintf(fp, "%s:%d.%d: ", 
-               state->file->basename, state->file->line, col);
+       fprintf(fp, "%s:%d.%d: ",
+               state->file->report_name, state->file->report_line, col);
 }
 
-static void __internal_error(struct compile_state *state, struct triple *ptr, 
-       char *fmt, ...)
+static void __attribute__ ((noreturn)) internal_error(struct compile_state *state, struct triple *ptr,
+       const char *fmt, ...)
 {
+       FILE *fp = state->errout;
        va_list args;
        va_start(args, fmt);
-       loc(stderr, state, ptr);
+       loc(fp, state, ptr);
+       fputc('\n', fp);
        if (ptr) {
-               fprintf(stderr, "%p %s ", ptr, tops(ptr->op));
+               fprintf(fp, "%p %-10s ", ptr, tops(ptr->op));
        }
-       fprintf(stderr, "Internal compiler error: ");
-       vfprintf(stderr, fmt, args);
-       fprintf(stderr, "\n");
+       fprintf(fp, "Internal compiler error: ");
+       vfprintf(fp, fmt, args);
+       fprintf(fp, "\n");
        va_end(args);
        do_cleanup(state);
        abort();
 }
 
 
-static void __internal_warning(struct compile_state *state, struct triple *ptr, 
-       char *fmt, ...)
+static void internal_warning(struct compile_state *state, struct triple *ptr,
+       const char *fmt, ...)
 {
+       FILE *fp = state->errout;
        va_list args;
        va_start(args, fmt);
-       loc(stderr, state, ptr);
-       fprintf(stderr, "Internal compiler warning: ");
-       vfprintf(stderr, fmt, args);
-       fprintf(stderr, "\n");
+       loc(fp, state, ptr);
+       if (ptr) {
+               fprintf(fp, "%p %-10s ", ptr, tops(ptr->op));
+       }
+       fprintf(fp, "Internal compiler warning: ");
+       vfprintf(fp, fmt, args);
+       fprintf(fp, "\n");
        va_end(args);
 }
 
 
 
-static void __error(struct compile_state *state, struct triple *ptr, 
-       char *fmt, ...)
+static void __attribute__ ((noreturn)) error(struct compile_state *state, struct triple *ptr,
+       const char *fmt, ...)
 {
+       FILE *fp = state->errout;
        va_list args;
        va_start(args, fmt);
-       loc(stderr, state, ptr);
-       vfprintf(stderr, fmt, args);
+       loc(fp, state, ptr);
+       fputc('\n', fp);
+       if (ptr && (state->compiler->debug & DEBUG_ABORT_ON_ERROR)) {
+               fprintf(fp, "%p %-10s ", ptr, tops(ptr->op));
+       }
+       vfprintf(fp, fmt, args);
        va_end(args);
-       fprintf(stderr, "\n");
+       fprintf(fp, "\n");
        do_cleanup(state);
-       if (state->debug & DEBUG_ABORT_ON_ERROR) {
+       if (state->compiler->debug & DEBUG_ABORT_ON_ERROR) {
                abort();
        }
        exit(1);
 }
 
-static void __warning(struct compile_state *state, struct triple *ptr, 
-       char *fmt, ...)
+static void warning(struct compile_state *state, struct triple *ptr,
+       const char *fmt, ...)
 {
+       FILE *fp = state->errout;
        va_list args;
        va_start(args, fmt);
-       loc(stderr, state, ptr);
-       fprintf(stderr, "warning: "); 
-       vfprintf(stderr, fmt, args);
-       fprintf(stderr, "\n");
+       loc(fp, state, ptr);
+       fprintf(fp, "warning: ");
+       if (ptr && (state->compiler->debug & DEBUG_ABORT_ON_ERROR)) {
+               fprintf(fp, "%p %-10s ", ptr, tops(ptr->op));
+       }
+       vfprintf(fp, fmt, args);
+       fprintf(fp, "\n");
        va_end(args);
 }
 
-#if DEBUG_ERROR_MESSAGES 
-#  define internal_error fprintf(stderr,  "@ %s.%s:%d \t", __FILE__, __func__, __LINE__),__internal_error
-#  define internal_warning fprintf(stderr,  "@ %s.%s:%d \t", __FILE__, __func__, __LINE__),__internal_warning
-#  define error fprintf(stderr, "@ %s.%s:%d \t", __FILE__, __func__, __LINE__),__error
-#  define warning fprintf(stderr, "@ %s.%s:%d \t", __FILE__, __func__, __LINE__),__warning
-#else
-#  define internal_error __internal_error
-#  define internal_warning __internal_warning
-#  define error __error
-#  define warning __warning
-#endif
 #define FINISHME() warning(state, 0, "FINISHME @ %s.%s:%d", __FILE__, __func__, __LINE__)
 
 static void valid_op(struct compile_state *state, int op)
@@ -1035,67 +1862,33 @@ static void valid_ins(struct compile_state *state, struct triple *ptr)
        valid_op(state, ptr->op);
 }
 
-static void process_trigraphs(struct compile_state *state)
+#if DEBUG_ROMCC_WARNING
+static void valid_param_count(struct compile_state *state, struct triple *ins)
 {
-       char *src, *dest, *end;
-       struct file_state *file;
-       file = state->file;
-       src = dest = file->buf;
-       end = file->buf + file->size;
-       while((end - src) >= 3) {
-               if ((src[0] == '?') && (src[1] == '?')) {
-                       int c = -1;
-                       switch(src[2]) {
-                       case '=': c = '#'; break;
-                       case '/': c = '\\'; break;
-                       case '\'': c = '^'; break;
-                       case '(': c = '['; break;
-                       case ')': c = ']'; break;
-                       case '!': c = '!'; break;
-                       case '<': c = '{'; break;
-                       case '>': c = '}'; break;
-                       case '-': c = '~'; break;
-                       }
-                       if (c != -1) {
-                               *dest++ = c;
-                               src += 3;
-                       }
-                       else {
-                               *dest++ = *src++;
-                       }
-               }
-               else {
-                       *dest++ = *src++;
-               }
+       int lhs, rhs, misc, targ;
+       valid_ins(state, ins);
+       lhs  = table_ops[ins->op].lhs;
+       rhs  = table_ops[ins->op].rhs;
+       misc = table_ops[ins->op].misc;
+       targ = table_ops[ins->op].targ;
+
+       if ((lhs >= 0) && (ins->lhs != lhs)) {
+               internal_error(state, ins, "Bad lhs count");
        }
-       while(src != end) {
-               *dest++ = *src++;
+       if ((rhs >= 0) && (ins->rhs != rhs)) {
+               internal_error(state, ins, "Bad rhs count");
        }
-       file->size = dest - file->buf;
-}
-
-static void splice_lines(struct compile_state *state)
-{
-       char *src, *dest, *end;
-       struct file_state *file;
-       file = state->file;
-       src = dest = file->buf;
-       end = file->buf + file->size;
-       while((end - src) >= 2) {
-               if ((src[0] == '\\') && (src[1] == '\n')) {
-                       src += 2;
-               }
-               else {
-                       *dest++ = *src++;
-               }
+       if ((misc >= 0) && (ins->misc != misc)) {
+               internal_error(state, ins, "Bad misc count");
        }
-       while(src != end) {
-               *dest++ = *src++;
+       if ((targ >= 0) && (ins->targ != targ)) {
+               internal_error(state, ins, "Bad targ count");
        }
-       file->size = dest - file->buf;
 }
+#endif
 
 static struct type void_type;
+static struct type unknown_type;
 static void use_triple(struct triple *used, struct triple *user)
 {
        struct triple_set **ptr, *new;
@@ -1110,7 +1903,7 @@ static void use_triple(struct triple *used, struct triple *user)
                }
                ptr = &(*ptr)->next;
        }
-       /* Append new to the head of the list, 
+       /* Append new to the head of the list,
         * copy_func and rename_block_variables
         * depends on this.
         */
@@ -1139,134 +1932,221 @@ static void unuse_triple(struct triple *used, struct triple *unuser)
        }
 }
 
-static void push_triple(struct triple *used, struct triple *user)
+static void put_occurance(struct occurance *occurance)
 {
-       struct triple_set *new;
-       if (!used)
-               return;
-       if (!user)
-               return;
-       /* Append new to the head of the list,
-        * it's the only sensible behavoir for a stack.
-        */
-       new = xcmalloc(sizeof(*new), "triple_set");
-       new->member = user;
-       new->next   = used->use;
-       used->use   = new;
+       if (occurance) {
+               occurance->count -= 1;
+               if (occurance->count <= 0) {
+                       if (occurance->parent) {
+                               put_occurance(occurance->parent);
+                       }
+                       xfree(occurance);
+               }
+       }
 }
 
-static void pop_triple(struct triple *used, struct triple *unuser)
+static void get_occurance(struct occurance *occurance)
 {
-       struct triple_set *use, **ptr;
-       ptr = &used->use;
-       while(*ptr) {
-               use = *ptr;
-               if (use->member == unuser) {
-                       *ptr = use->next;
-                       xfree(use);
-                       /* Only free one occurance from the stack */
-                       return;
-               }
-               else {
-                       ptr = &use->next;
-               }
+       if (occurance) {
+               occurance->count += 1;
        }
 }
 
 
-/* The zero triple is used as a place holder when we are removing pointers
- * from a triple.  Having allows certain sanity checks to pass even
- * when the original triple that was pointed to is gone.
- */
-static struct triple zero_triple = {
-       .next     = &zero_triple,
-       .prev     = &zero_triple,
-       .use      = 0,
-       .op       = OP_INTCONST,
-       .sizes    = TRIPLE_SIZES(0, 0, 0, 0),
-       .id       = -1, /* An invalid id */
-       .u = { .cval   = 0, },
+static struct occurance *new_occurance(struct compile_state *state)
+{
+       struct occurance *result, *last;
+       const char *filename;
+       const char *function;
+       int line, col;
+
+       function = "";
+       filename = 0;
+       line = 0;
+       col  = 0;
+       if (state->file) {
+               filename = state->file->report_name;
+               line     = state->file->report_line;
+               col      = get_col(state->file);
+       }
+       if (state->function) {
+               function = state->function;
+       }
+       last = state->last_occurance;
+       if (last &&
+               (last->col == col) &&
+               (last->line == line) &&
+               (last->function == function) &&
+               ((last->filename == filename) ||
+                       (strcmp(last->filename, filename) == 0)))
+       {
+               get_occurance(last);
+               return last;
+       }
+       if (last) {
+               state->last_occurance = 0;
+               put_occurance(last);
+       }
+       result = xmalloc(sizeof(*result), "occurance");
+       result->count    = 2;
+       result->filename = filename;
+       result->function = function;
+       result->line     = line;
+       result->col      = col;
+       result->parent   = 0;
+       state->last_occurance = result;
+       return result;
+}
+
+static struct occurance *inline_occurance(struct compile_state *state,
+       struct occurance *base, struct occurance *top)
+{
+       struct occurance *result, *last;
+       if (top->parent) {
+               internal_error(state, 0, "inlining an already inlined function?");
+       }
+       /* If I have a null base treat it that way */
+       if ((base->parent == 0) &&
+               (base->col == 0) &&
+               (base->line == 0) &&
+               (base->function[0] == '\0') &&
+               (base->filename[0] == '\0')) {
+               base = 0;
+       }
+       /* See if I can reuse the last occurance I had */
+       last = state->last_occurance;
+       if (last &&
+               (last->parent   == base) &&
+               (last->col      == top->col) &&
+               (last->line     == top->line) &&
+               (last->function == top->function) &&
+               (last->filename == top->filename)) {
+               get_occurance(last);
+               return last;
+       }
+       /* I can't reuse the last occurance so free it */
+       if (last) {
+               state->last_occurance = 0;
+               put_occurance(last);
+       }
+       /* Generate a new occurance structure */
+       get_occurance(base);
+       result = xmalloc(sizeof(*result), "occurance");
+       result->count    = 2;
+       result->filename = top->filename;
+       result->function = top->function;
+       result->line     = top->line;
+       result->col      = top->col;
+       result->parent   = base;
+       state->last_occurance = result;
+       return result;
+}
+
+static struct occurance dummy_occurance = {
+       .count    = 2,
        .filename = __FILE__,
+       .function = "",
        .line     = __LINE__,
        .col      = 0,
-       .param { [0] = 0, [1] = 0, },
+       .parent   = 0,
+};
+
+/* The undef triple is used as a place holder when we are removing pointers
+ * from a triple.  Having allows certain sanity checks to pass even
+ * when the original triple that was pointed to is gone.
+ */
+static struct triple unknown_triple = {
+       .next      = &unknown_triple,
+       .prev      = &unknown_triple,
+       .use       = 0,
+       .op        = OP_UNKNOWNVAL,
+       .lhs       = 0,
+       .rhs       = 0,
+       .misc      = 0,
+       .targ      = 0,
+       .type      = &unknown_type,
+       .id        = -1, /* An invalid id */
+       .u = { .cval = 0, },
+       .occurance = &dummy_occurance,
+       .param = { [0] = 0, [1] = 0, },
 };
 
 
-static unsigned short triple_sizes(struct compile_state *state,
-       int op, struct type *type, int lhs_wanted, int rhs_wanted)
+static size_t registers_of(struct compile_state *state, struct type *type);
+
+static struct triple *alloc_triple(struct compile_state *state,
+       int op, struct type *type, int lhs_wanted, int rhs_wanted,
+       struct occurance *occurance)
 {
+       size_t size, extra_count, min_count;
        int lhs, rhs, misc, targ;
+       struct triple *ret, dummy;
+       dummy.op = op;
+       dummy.occurance = occurance;
        valid_op(state, op);
        lhs = table_ops[op].lhs;
        rhs = table_ops[op].rhs;
        misc = table_ops[op].misc;
        targ = table_ops[op].targ;
-       
-       
-       if (op == OP_CALL) {
-               struct type *param;
-               rhs = 0;
-               param = type->right;
-               while((param->type & TYPE_MASK) == TYPE_PRODUCT) {
-                       rhs++;
-                       param = param->right;
-               }
-               if ((param->type & TYPE_MASK) != TYPE_VOID) {
-                       rhs++;
-               }
-               lhs = 0;
-               if ((type->left->type & TYPE_MASK) == TYPE_STRUCT) {
-                       lhs = type->left->elements;
-               }
-       }
-       else if (op == OP_VAL_VEC) {
-               rhs = type->elements;
-       }
-       else if ((op == OP_BRANCH) || (op == OP_PHI)) {
+
+       switch(op) {
+       case OP_FCALL:
                rhs = rhs_wanted;
-       }
-       else if (op == OP_ASM) {
+               break;
+       case OP_PHI:
+               rhs = rhs_wanted;
+               break;
+       case OP_ADECL:
+               lhs = registers_of(state, type);
+               break;
+       case OP_TUPLE:
+               lhs = registers_of(state, type);
+               break;
+       case OP_ASM:
                rhs = rhs_wanted;
                lhs = lhs_wanted;
+               break;
        }
        if ((rhs < 0) || (rhs > MAX_RHS)) {
-               internal_error(state, 0, "bad rhs");
+               internal_error(state, &dummy, "bad rhs count %d", rhs);
        }
        if ((lhs < 0) || (lhs > MAX_LHS)) {
-               internal_error(state, 0, "bad lhs");
+               internal_error(state, &dummy, "bad lhs count %d", lhs);
        }
        if ((misc < 0) || (misc > MAX_MISC)) {
-               internal_error(state, 0, "bad misc");
+               internal_error(state, &dummy, "bad misc count %d", misc);
        }
        if ((targ < 0) || (targ > MAX_TARG)) {
-               internal_error(state, 0, "bad targs");
+               internal_error(state, &dummy, "bad targs count %d", targ);
        }
-       return TRIPLE_SIZES(lhs, rhs, misc, targ);
-}
-
-static struct triple *alloc_triple(struct compile_state *state, 
-       int op, struct type *type, int lhs, int rhs,
-       const char *filename, int line, int col)
-{
-       size_t size, sizes, extra_count, min_count;
-       struct triple *ret;
-       sizes = triple_sizes(state, op, type, lhs, rhs);
 
        min_count = sizeof(ret->param)/sizeof(ret->param[0]);
-       extra_count = TRIPLE_SIZE(sizes);
+       extra_count = lhs + rhs + misc + targ;
        extra_count = (extra_count < min_count)? 0 : extra_count - min_count;
 
        size = sizeof(*ret) + sizeof(ret->param[0]) * extra_count;
        ret = xcmalloc(size, "tripple");
-       ret->op       = op;
-       ret->sizes    = sizes;
-       ret->type     = type;
-       ret->next     = ret;
-       ret->prev     = ret;
-       ret->filename = filename;
-       ret->line     = line;
-       ret->col      = col;
+       ret->op        = op;
+       ret->lhs       = lhs;
+       ret->rhs       = rhs;
+       ret->misc      = misc;
+       ret->targ      = targ;
+       ret->type      = type;
+       ret->next      = ret;
+       ret->prev      = ret;
+       ret->occurance = occurance;
+       /* A simple sanity check */
+       if ((ret->op != op) ||
+               (ret->lhs != lhs) ||
+               (ret->rhs != rhs) ||
+               (ret->misc != misc) ||
+               (ret->targ != targ) ||
+               (ret->type != type) ||
+               (ret->next != ret) ||
+               (ret->prev != ret) ||
+               (ret->occurance != occurance)) {
+               internal_error(state, ret, "huh?");
+       }
        return ret;
 }
 
@@ -1274,43 +2154,44 @@ struct triple *dup_triple(struct compile_state *state, struct triple *src)
 {
        struct triple *dup;
        int src_lhs, src_rhs, src_size;
-       src_lhs = TRIPLE_LHS(src->sizes);
-       src_rhs = TRIPLE_RHS(src->sizes);
-       src_size = TRIPLE_SIZE(src->sizes);
+       src_lhs = src->lhs;
+       src_rhs = src->rhs;
+       src_size = TRIPLE_SIZE(src);
+       get_occurance(src->occurance);
        dup = alloc_triple(state, src->op, src->type, src_lhs, src_rhs,
-               src->filename, src->line, src->col);
+               src->occurance);
        memcpy(dup, src, sizeof(*src));
        memcpy(dup->param, src->param, src_size * sizeof(src->param[0]));
        return dup;
 }
 
-static struct triple *new_triple(struct compile_state *state, 
+static struct triple *copy_triple(struct compile_state *state, struct triple *src)
+{
+       struct triple *copy;
+       copy = dup_triple(state, src);
+       copy->use = 0;
+       copy->next = copy->prev = copy;
+       return copy;
+}
+
+static struct triple *new_triple(struct compile_state *state,
        int op, struct type *type, int lhs, int rhs)
 {
        struct triple *ret;
-       const char *filename;
-       int line, col;
-       filename = 0;
-       line = 0;
-       col  = 0;
-       if (state->file) {
-               filename = state->file->basename;
-               line     = state->file->line;
-               col      = get_col(state->file);
-       }
-       ret = alloc_triple(state, op, type, lhs, rhs,
-               filename, line, col);
+       struct occurance *occurance;
+       occurance = new_occurance(state);
+       ret = alloc_triple(state, op, type, lhs, rhs, occurance);
        return ret;
 }
 
-static struct triple *build_triple(struct compile_state *state, 
+static struct triple *build_triple(struct compile_state *state,
        int op, struct type *type, struct triple *left, struct triple *right,
-       const char *filename, int line, int col)
+       struct occurance *occurance)
 {
        struct triple *ret;
        size_t count;
-       ret = alloc_triple(state, op, type, -1, -1, filename, line, col);
-       count = TRIPLE_SIZE(ret->sizes);
+       ret = alloc_triple(state, op, type, -1, -1, occurance);
+       count = TRIPLE_SIZE(ret);
        if (count > 0) {
                ret->param[0] = left;
        }
@@ -1320,13 +2201,13 @@ static struct triple *build_triple(struct compile_state *state,
        return ret;
 }
 
-static struct triple *triple(struct compile_state *state, 
+static struct triple *triple(struct compile_state *state,
        int op, struct type *type, struct triple *left, struct triple *right)
 {
        struct triple *ret;
        size_t count;
        ret = new_triple(state, op, type, -1, -1);
-       count = TRIPLE_SIZE(ret->sizes);
+       count = TRIPLE_SIZE(ret);
        if (count >= 1) {
                ret->param[0] = left;
        }
@@ -1336,24 +2217,27 @@ static struct triple *triple(struct compile_state *state,
        return ret;
 }
 
-static struct triple *branch(struct compile_state *state, 
+static struct triple *branch(struct compile_state *state,
        struct triple *targ, struct triple *test)
 {
        struct triple *ret;
-       ret = new_triple(state, OP_BRANCH, &void_type, -1, test?1:0);
        if (test) {
+               ret = new_triple(state, OP_CBRANCH, &void_type, -1, 1);
                RHS(ret, 0) = test;
+       } else {
+               ret = new_triple(state, OP_BRANCH, &void_type, -1, 0);
        }
        TARG(ret, 0) = targ;
        /* record the branch target was used */
        if (!targ || (targ->op != OP_LABEL)) {
                internal_error(state, 0, "branch not to label");
-               use_triple(targ, ret);
        }
        return ret;
 }
 
-
+static int triple_is_label(struct compile_state *state, struct triple *ins);
+static int triple_is_call(struct compile_state *state, struct triple *ins);
+static int triple_is_cbranch(struct compile_state *state, struct triple *ins);
 static void insert_triple(struct compile_state *state,
        struct triple *first, struct triple *ptr)
 {
@@ -1365,8 +2249,9 @@ static void insert_triple(struct compile_state *state,
                ptr->prev       = first->prev;
                ptr->prev->next = ptr;
                ptr->next->prev = ptr;
-               if ((ptr->prev->op == OP_BRANCH) && 
-                       TRIPLE_RHS(ptr->prev->sizes)) {
+
+               if (triple_is_cbranch(state, ptr->prev) ||
+                       triple_is_call(state, ptr->prev)) {
                        unuse_triple(first, ptr->prev);
                        use_triple(ptr, ptr->prev);
                }
@@ -1375,7 +2260,7 @@ static void insert_triple(struct compile_state *state,
 
 static int triple_stores_block(struct compile_state *state, struct triple *ins)
 {
-       /* This function is used to determine if u.block 
+       /* This function is used to determine if u.block
         * is utilized to store the current block number.
         */
        int stores_block;
@@ -1384,37 +2269,54 @@ static int triple_stores_block(struct compile_state *state, struct triple *ins)
        return stores_block;
 }
 
-static struct block *block_of_triple(struct compile_state *state, 
+static int triple_is_branch(struct compile_state *state, struct triple *ins);
+static struct block *block_of_triple(struct compile_state *state,
        struct triple *ins)
 {
        struct triple *first;
-       first = RHS(state->main_function, 0);
-       while(ins != first && !triple_stores_block(state, ins)) {
+       if (!ins || ins == &unknown_triple) {
+               return 0;
+       }
+       first = state->first;
+       while(ins != first && !triple_is_branch(state, ins->prev) &&
+               !triple_stores_block(state, ins))
+       {
                if (ins == ins->prev) {
-                       internal_error(state, 0, "ins == ins->prev?");
+                       internal_error(state, ins, "ins == ins->prev?");
                }
                ins = ins->prev;
        }
-       if (!triple_stores_block(state, ins)) {
-               internal_error(state, ins, "Cannot find block");
-       }
-       return ins->u.block;
+       return triple_stores_block(state, ins)? ins->u.block: 0;
 }
 
+static void generate_lhs_pieces(struct compile_state *state, struct triple *ins);
 static struct triple *pre_triple(struct compile_state *state,
        struct triple *base,
        int op, struct type *type, struct triple *left, struct triple *right)
 {
        struct block *block;
        struct triple *ret;
+       int i;
+       /* If I am an OP_PIECE jump to the real instruction */
+       if (base->op == OP_PIECE) {
+               base = MISC(base, 0);
+       }
        block = block_of_triple(state, base);
-       ret = build_triple(state, op, type, left, right, 
-               base->filename, base->line, base->col);
+       get_occurance(base->occurance);
+       ret = build_triple(state, op, type, left, right, base->occurance);
+       generate_lhs_pieces(state, ret);
        if (triple_stores_block(state, ret)) {
                ret->u.block = block;
        }
        insert_triple(state, base, ret);
-       if (block->first == base) {
+       for(i = 0; i < ret->lhs; i++) {
+               struct triple *piece;
+               piece = LHS(ret, i);
+               insert_triple(state, base, piece);
+               use_triple(ret, piece);
+               use_triple(piece, ret);
+       }
+       if (block && (block->first == base)) {
                block->first = ret;
        }
        return ret;
@@ -1425,20 +2327,82 @@ static struct triple *post_triple(struct compile_state *state,
        int op, struct type *type, struct triple *left, struct triple *right)
 {
        struct block *block;
-       struct triple *ret;
+       struct triple *ret, *next;
+       int zlhs, i;
+       /* If I am an OP_PIECE jump to the real instruction */
+       if (base->op == OP_PIECE) {
+               base = MISC(base, 0);
+       }
+       /* If I have a left hand side skip over it */
+       zlhs = base->lhs;
+       if (zlhs) {
+               base = LHS(base, zlhs - 1);
+       }
+
        block = block_of_triple(state, base);
-       ret = build_triple(state, op, type, left, right, 
-               base->filename, base->line, base->col);
+       get_occurance(base->occurance);
+       ret = build_triple(state, op, type, left, right, base->occurance);
+       generate_lhs_pieces(state, ret);
        if (triple_stores_block(state, ret)) {
                ret->u.block = block;
        }
-       insert_triple(state, base->next, ret);
-       if (block->last == base) {
+       next = base->next;
+       insert_triple(state, next, ret);
+       zlhs = ret->lhs;
+       for(i = 0; i < zlhs; i++) {
+               struct triple *piece;
+               piece = LHS(ret, i);
+               insert_triple(state, next, piece);
+               use_triple(ret, piece);
+               use_triple(piece, ret);
+       }
+       if (block && (block->last == base)) {
                block->last = ret;
+               if (zlhs) {
+                       block->last = LHS(ret, zlhs - 1);
+               }
        }
        return ret;
 }
 
+static struct type *reg_type(
+       struct compile_state *state, struct type *type, int reg);
+
+static void generate_lhs_piece(
+       struct compile_state *state, struct triple *ins, int index)
+{
+       struct type *piece_type;
+       struct triple *piece;
+       get_occurance(ins->occurance);
+       piece_type = reg_type(state, ins->type, index * REG_SIZEOF_REG);
+
+       if ((piece_type->type & TYPE_MASK) == TYPE_BITFIELD) {
+               piece_type = piece_type->left;
+       }
+#if 0
+{
+       static void name_of(FILE *fp, struct type *type);
+       FILE * fp = state->errout;
+       fprintf(fp, "piece_type(%d): ", index);
+       name_of(fp, piece_type);
+       fprintf(fp, "\n");
+}
+#endif
+       piece = alloc_triple(state, OP_PIECE, piece_type, -1, -1, ins->occurance);
+       piece->u.cval  = index;
+       LHS(ins, piece->u.cval) = piece;
+       MISC(piece, 0) = ins;
+}
+
+static void generate_lhs_pieces(struct compile_state *state, struct triple *ins)
+{
+       int i, zlhs;
+       zlhs = ins->lhs;
+       for(i = 0; i < zlhs; i++) {
+               generate_lhs_piece(state, ins, i);
+       }
+}
+
 static struct triple *label(struct compile_state *state)
 {
        /* Labels don't get a type */
@@ -1447,122 +2411,327 @@ static struct triple *label(struct compile_state *state)
        return result;
 }
 
+static struct triple *mkprog(struct compile_state *state, ...)
+{
+       struct triple *prog, *head, *arg;
+       va_list args;
+       int i;
+
+       head = label(state);
+       prog = new_triple(state, OP_PROG, &void_type, -1, -1);
+       RHS(prog, 0) = head;
+       va_start(args, state);
+       i = 0;
+       while((arg = va_arg(args, struct triple *)) != 0) {
+               if (++i >= 100) {
+                       internal_error(state, 0, "too many arguments to mkprog");
+               }
+               flatten(state, head, arg);
+       }
+       va_end(args);
+       prog->type = head->prev->type;
+       return prog;
+}
+static void name_of(FILE *fp, struct type *type);
 static void display_triple(FILE *fp, struct triple *ins)
 {
-       if (ins->op == OP_INTCONST) {
-               fprintf(fp, "(%p) %3d %-2d %-10s <0x%08lx>          @ %s:%d.%d\n",
-                       ins, ID_REG(ins->id), ins->template_id, tops(ins->op), 
-                       ins->u.cval,
-                       ins->filename, ins->line, ins->col);
+       struct occurance *ptr;
+       const char *reg;
+       char pre, post, vol;
+       pre = post = vol = ' ';
+       if (ins) {
+               if (ins->id & TRIPLE_FLAG_PRE_SPLIT) {
+                       pre = '^';
+               }
+               if (ins->id & TRIPLE_FLAG_POST_SPLIT) {
+                       post = ',';
+               }
+               if (ins->id & TRIPLE_FLAG_VOLATILE) {
+                       vol = 'v';
+               }
+               reg = arch_reg_str(ID_REG(ins->id));
+       }
+       if (ins == 0) {
+               fprintf(fp, "(%p) <nothing> ", ins);
+       }
+       else if (ins->op == OP_INTCONST) {
+               fprintf(fp, "(%p) %c%c%c %-7s %-2d %-10s <0x%08lx>         ",
+                       ins, pre, post, vol, reg, ins->template_id, tops(ins->op),
+                       (unsigned long)(ins->u.cval));
        }
        else if (ins->op == OP_ADDRCONST) {
-               fprintf(fp, "(%p) %3d %-2d %-10s %-10p <0x%08lx> @ %s:%d.%d\n",
-                       ins, ID_REG(ins->id), ins->template_id, tops(ins->op), 
-                       MISC(ins, 0), ins->u.cval,
-                       ins->filename, ins->line, ins->col);
+               fprintf(fp, "(%p) %c%c%c %-7s %-2d %-10s %-10p <0x%08lx>",
+                       ins, pre, post, vol, reg, ins->template_id, tops(ins->op),
+                       MISC(ins, 0), (unsigned long)(ins->u.cval));
+       }
+       else if (ins->op == OP_INDEX) {
+               fprintf(fp, "(%p) %c%c%c %-7s %-2d %-10s %-10p <0x%08lx>",
+                       ins, pre, post, vol, reg, ins->template_id, tops(ins->op),
+                       RHS(ins, 0), (unsigned long)(ins->u.cval));
+       }
+       else if (ins->op == OP_PIECE) {
+               fprintf(fp, "(%p) %c%c%c %-7s %-2d %-10s %-10p <0x%08lx>",
+                       ins, pre, post, vol, reg, ins->template_id, tops(ins->op),
+                       MISC(ins, 0), (unsigned long)(ins->u.cval));
        }
        else {
                int i, count;
-               fprintf(fp, "(%p) %3d %-2d %-10s", 
-                       ins, ID_REG(ins->id), ins->template_id, tops(ins->op));
-               count = TRIPLE_SIZE(ins->sizes);
+               fprintf(fp, "(%p) %c%c%c %-7s %-2d %-10s",
+                       ins, pre, post, vol, reg, ins->template_id, tops(ins->op));
+               if (table_ops[ins->op].flags & BITFIELD) {
+                       fprintf(fp, " <%2d-%2d:%2d>",
+                               ins->u.bitfield.offset,
+                               ins->u.bitfield.offset + ins->u.bitfield.size,
+                               ins->u.bitfield.size);
+               }
+               count = TRIPLE_SIZE(ins);
                for(i = 0; i < count; i++) {
                        fprintf(fp, " %-10p", ins->param[i]);
                }
                for(; i < 2; i++) {
-                       printf("           ");
+                       fprintf(fp, "           ");
+               }
+       }
+       if (ins) {
+               struct triple_set *user;
+#if DEBUG_DISPLAY_TYPES
+               fprintf(fp, " <");
+               name_of(fp, ins->type);
+               fprintf(fp, "> ");
+#endif
+#if DEBUG_DISPLAY_USES
+               fprintf(fp, " [");
+               for(user = ins->use; user; user = user->next) {
+                       fprintf(fp, " %-10p", user->member);
+               }
+               fprintf(fp, " ]");
+#endif
+               fprintf(fp, " @");
+               for(ptr = ins->occurance; ptr; ptr = ptr->parent) {
+                       fprintf(fp, " %s,%s:%d.%d",
+                               ptr->function,
+                               ptr->filename,
+                               ptr->line,
+                               ptr->col);
+               }
+               if (ins->op == OP_ASM) {
+                       fprintf(fp, "\n\t%s", ins->u.ainfo->str);
                }
-               fprintf(fp, " @ %s:%d.%d\n", 
-                       ins->filename, ins->line, ins->col);
        }
+       fprintf(fp, "\n");
        fflush(fp);
 }
 
-static int triple_is_pure(struct compile_state *state, struct triple *ins)
+static int equiv_types(struct type *left, struct type *right);
+static void display_triple_changes(
+       FILE *fp, const struct triple *new, const struct triple *orig)
+{
+
+       int new_count, orig_count;
+       new_count = TRIPLE_SIZE(new);
+       orig_count = TRIPLE_SIZE(orig);
+       if ((new->op != orig->op) ||
+               (new_count != orig_count) ||
+               (memcmp(orig->param, new->param,
+                       orig_count * sizeof(orig->param[0])) != 0) ||
+               (memcmp(&orig->u, &new->u, sizeof(orig->u)) != 0))
+       {
+               struct occurance *ptr;
+               int i, min_count, indent;
+               fprintf(fp, "(%p %p)", new, orig);
+               if (orig->op == new->op) {
+                       fprintf(fp, " %-11s", tops(orig->op));
+               } else {
+                       fprintf(fp, " [%-10s %-10s]",
+                               tops(new->op), tops(orig->op));
+               }
+               min_count = new_count;
+               if (min_count > orig_count) {
+                       min_count = orig_count;
+               }
+               for(indent = i = 0; i < min_count; i++) {
+                       if (orig->param[i] == new->param[i]) {
+                               fprintf(fp, " %-11p",
+                                       orig->param[i]);
+                               indent += 12;
+                       } else {
+                               fprintf(fp, " [%-10p %-10p]",
+                                       new->param[i],
+                                       orig->param[i]);
+                               indent += 24;
+                       }
+               }
+               for(; i < orig_count; i++) {
+                       fprintf(fp, " [%-9p]", orig->param[i]);
+                       indent += 12;
+               }
+               for(; i < new_count; i++) {
+                       fprintf(fp, " [%-9p]", new->param[i]);
+                       indent += 12;
+               }
+               if ((new->op == OP_INTCONST)||
+                       (new->op == OP_ADDRCONST)) {
+                       fprintf(fp, " <0x%08lx>",
+                               (unsigned long)(new->u.cval));
+                       indent += 13;
+               }
+               for(;indent < 36; indent++) {
+                       putc(' ', fp);
+               }
+
+#if DEBUG_DISPLAY_TYPES
+               fprintf(fp, " <");
+               name_of(fp, new->type);
+               if (!equiv_types(new->type, orig->type)) {
+                       fprintf(fp, " -- ");
+                       name_of(fp, orig->type);
+               }
+               fprintf(fp, "> ");
+#endif
+
+               fprintf(fp, " @");
+               for(ptr = orig->occurance; ptr; ptr = ptr->parent) {
+                       fprintf(fp, " %s,%s:%d.%d",
+                               ptr->function,
+                               ptr->filename,
+                               ptr->line,
+                               ptr->col);
+
+               }
+               fprintf(fp, "\n");
+               fflush(fp);
+       }
+}
+
+static int triple_is_pure(struct compile_state *state, struct triple *ins, unsigned id)
 {
        /* Does the triple have no side effects.
-        * I.e. Rexecuting the triple with the same arguments 
+        * I.e. Rexecuting the triple with the same arguments
         * gives the same value.
         */
        unsigned pure;
        valid_ins(state, ins);
        pure = PURE_BITS(table_ops[ins->op].flags);
        if ((pure != PURE) && (pure != IMPURE)) {
-               internal_error(state, 0, "Purity of %s not known\n",
+               internal_error(state, 0, "Purity of %s not known",
                        tops(ins->op));
        }
-       return pure == PURE;
+       return (pure == PURE) && !(id & TRIPLE_FLAG_VOLATILE);
 }
 
-static int triple_is_branch(struct compile_state *state, struct triple *ins)
+static int triple_is_branch_type(struct compile_state *state,
+       struct triple *ins, unsigned type)
 {
-       /* This function is used to determine which triples need
-        * a register.
-        */
-       int is_branch;
+       /* Is this one of the passed branch types? */
        valid_ins(state, ins);
-       is_branch = (table_ops[ins->op].targ != 0);
-       return is_branch;
+       return (BRANCH_BITS(table_ops[ins->op].flags) == type);
 }
 
-static int triple_is_def(struct compile_state *state, struct triple *ins)
+static int triple_is_branch(struct compile_state *state, struct triple *ins)
 {
-       /* This function is used to determine which triples need
-        * a register.
-        */
-       int is_def;
+       /* Is this triple a branch instruction? */
        valid_ins(state, ins);
-       is_def = (table_ops[ins->op].flags & DEF) == DEF;
-       return is_def;
+       return (BRANCH_BITS(table_ops[ins->op].flags) != 0);
 }
 
-static struct triple **triple_iter(struct compile_state *state,
-       size_t count, struct triple **vector,
-       struct triple *ins, struct triple **last)
+static int triple_is_cbranch(struct compile_state *state, struct triple *ins)
 {
-       struct triple **ret;
-       ret = 0;
-       if (count) {
-               if (!last) {
-                       ret = vector;
-               }
-               else if ((last >= vector) && (last < (vector + count - 1))) {
-                       ret = last + 1;
-               }
-       }
-       return ret;
-       
+       /* Is this triple a conditional branch instruction? */
+       return triple_is_branch_type(state, ins, CBRANCH);
 }
 
-static struct triple **triple_lhs(struct compile_state *state,
-       struct triple *ins, struct triple **last)
+static int triple_is_ubranch(struct compile_state *state, struct triple *ins)
 {
-       return triple_iter(state, TRIPLE_LHS(ins->sizes), &LHS(ins,0), 
-               ins, last);
+       /* Is this triple a unconditional branch instruction? */
+       unsigned type;
+       valid_ins(state, ins);
+       type = BRANCH_BITS(table_ops[ins->op].flags);
+       return (type != 0) && (type != CBRANCH);
 }
 
-static struct triple **triple_rhs(struct compile_state *state,
-       struct triple *ins, struct triple **last)
+static int triple_is_call(struct compile_state *state, struct triple *ins)
 {
-       return triple_iter(state, TRIPLE_RHS(ins->sizes), &RHS(ins,0), 
-               ins, last);
+       /* Is this triple a call instruction? */
+       return triple_is_branch_type(state, ins, CALLBRANCH);
 }
 
-static struct triple **triple_misc(struct compile_state *state,
-       struct triple *ins, struct triple **last)
+static int triple_is_ret(struct compile_state *state, struct triple *ins)
 {
-       return triple_iter(state, TRIPLE_MISC(ins->sizes), &MISC(ins,0), 
-               ins, last);
+       /* Is this triple a return instruction? */
+       return triple_is_branch_type(state, ins, RETBRANCH);
 }
 
-static struct triple **triple_targ(struct compile_state *state,
-       struct triple *ins, struct triple **last)
+#if DEBUG_ROMCC_WARNING
+static int triple_is_simple_ubranch(struct compile_state *state, struct triple *ins)
 {
-       size_t count;
-       struct triple **ret, **vector;
+       /* Is this triple an unconditional branch and not a call or a
+        * return? */
+       return triple_is_branch_type(state, ins, UBRANCH);
+}
+#endif
+
+static int triple_is_end(struct compile_state *state, struct triple *ins)
+{
+       return triple_is_branch_type(state, ins, ENDBRANCH);
+}
+
+static int triple_is_label(struct compile_state *state, struct triple *ins)
+{
+       valid_ins(state, ins);
+       return (ins->op == OP_LABEL);
+}
+
+static struct triple *triple_to_block_start(
+       struct compile_state *state, struct triple *start)
+{
+       while(!triple_is_branch(state, start->prev) &&
+               (!triple_is_label(state, start) || !start->use)) {
+               start = start->prev;
+       }
+       return start;
+}
+
+static int triple_is_def(struct compile_state *state, struct triple *ins)
+{
+       /* This function is used to determine which triples need
+        * a register.
+        */
+       int is_def;
+       valid_ins(state, ins);
+       is_def = (table_ops[ins->op].flags & DEF) == DEF;
+       if (ins->lhs >= 1) {
+               is_def = 0;
+       }
+       return is_def;
+}
+
+static int triple_is_structural(struct compile_state *state, struct triple *ins)
+{
+       int is_structural;
+       valid_ins(state, ins);
+       is_structural = (table_ops[ins->op].flags & STRUCTURAL) == STRUCTURAL;
+       return is_structural;
+}
+
+static int triple_is_part(struct compile_state *state, struct triple *ins)
+{
+       int is_part;
+       valid_ins(state, ins);
+       is_part = (table_ops[ins->op].flags & PART) == PART;
+       return is_part;
+}
+
+static int triple_is_auto_var(struct compile_state *state, struct triple *ins)
+{
+       return (ins->op == OP_PIECE) && (MISC(ins, 0)->op == OP_ADECL);
+}
+
+static struct triple **triple_iter(struct compile_state *state,
+       size_t count, struct triple **vector,
+       struct triple *ins, struct triple **last)
+{
+       struct triple **ret;
        ret = 0;
-       count = TRIPLE_TARG(ins->sizes);
-       vector = &TARG(ins, 0);
        if (count) {
                if (!last) {
                        ret = vector;
@@ -1570,20 +2739,189 @@ static struct triple **triple_targ(struct compile_state *state,
                else if ((last >= vector) && (last < (vector + count - 1))) {
                        ret = last + 1;
                }
-               else if ((last == (vector + count - 1)) && 
-                       TRIPLE_RHS(ins->sizes)) {
+       }
+       return ret;
+
+}
+
+static struct triple **triple_lhs(struct compile_state *state,
+       struct triple *ins, struct triple **last)
+{
+       return triple_iter(state, ins->lhs, &LHS(ins,0),
+               ins, last);
+}
+
+static struct triple **triple_rhs(struct compile_state *state,
+       struct triple *ins, struct triple **last)
+{
+       return triple_iter(state, ins->rhs, &RHS(ins,0),
+               ins, last);
+}
+
+static struct triple **triple_misc(struct compile_state *state,
+       struct triple *ins, struct triple **last)
+{
+       return triple_iter(state, ins->misc, &MISC(ins,0),
+               ins, last);
+}
+
+static struct triple **do_triple_targ(struct compile_state *state,
+       struct triple *ins, struct triple **last, int call_edges, int next_edges)
+{
+       size_t count;
+       struct triple **ret, **vector;
+       int next_is_targ;
+       ret = 0;
+       count = ins->targ;
+       next_is_targ = 0;
+       if (triple_is_cbranch(state, ins)) {
+               next_is_targ = 1;
+       }
+       if (!call_edges && triple_is_call(state, ins)) {
+               count = 0;
+       }
+       if (next_edges && triple_is_call(state, ins)) {
+               next_is_targ = 1;
+       }
+       vector = &TARG(ins, 0);
+       if (!ret && next_is_targ) {
+               if (!last) {
                        ret = &ins->next;
+               } else if (last == &ins->next) {
+                       last = 0;
+               }
+       }
+       if (!ret && count) {
+               if (!last) {
+                       ret = vector;
+               }
+               else if ((last >= vector) && (last < (vector + count - 1))) {
+                       ret = last + 1;
+               }
+               else if (last == vector + count - 1) {
+                       last = 0;
+               }
+       }
+       if (!ret && triple_is_ret(state, ins) && call_edges) {
+               struct triple_set *use;
+               for(use = ins->use; use; use = use->next) {
+                       if (!triple_is_call(state, use->member)) {
+                               continue;
+                       }
+                       if (!last) {
+                               ret = &use->member->next;
+                               break;
+                       }
+                       else if (last == &use->member->next) {
+                               last = 0;
+                       }
                }
        }
        return ret;
 }
 
+static struct triple **triple_targ(struct compile_state *state,
+       struct triple *ins, struct triple **last)
+{
+       return do_triple_targ(state, ins, last, 1, 1);
+}
+
+static struct triple **triple_edge_targ(struct compile_state *state,
+       struct triple *ins, struct triple **last)
+{
+       return do_triple_targ(state, ins, last,
+               state->functions_joined, !state->functions_joined);
+}
+
+static struct triple *after_lhs(struct compile_state *state, struct triple *ins)
+{
+       struct triple *next;
+       int lhs, i;
+       lhs = ins->lhs;
+       next = ins->next;
+       for(i = 0; i < lhs; i++) {
+               struct triple *piece;
+               piece = LHS(ins, i);
+               if (next != piece) {
+                       internal_error(state, ins, "malformed lhs on %s",
+                               tops(ins->op));
+               }
+               if (next->op != OP_PIECE) {
+                       internal_error(state, ins, "bad lhs op %s at %d on %s",
+                               tops(next->op), i, tops(ins->op));
+               }
+               if (next->u.cval != i) {
+                       internal_error(state, ins, "bad u.cval of %d %d expected",
+                               next->u.cval, i);
+               }
+               next = next->next;
+       }
+       return next;
+}
+
+/* Function piece accessor functions */
+static struct triple *do_farg(struct compile_state *state,
+       struct triple *func, unsigned index)
+{
+       struct type *ftype;
+       struct triple *first, *arg;
+       unsigned i;
+
+       ftype = func->type;
+       if((index < 0) || (index >= (ftype->elements + 2))) {
+               internal_error(state, func, "bad argument index: %d", index);
+       }
+       first = RHS(func, 0);
+       arg = first->next;
+       for(i = 0; i < index; i++, arg = after_lhs(state, arg)) {
+               /* do nothing */
+       }
+       if (arg->op != OP_ADECL) {
+               internal_error(state, 0, "arg not adecl?");
+       }
+       return arg;
+}
+static struct triple *fresult(struct compile_state *state, struct triple *func)
+{
+       return do_farg(state, func, 0);
+}
+static struct triple *fretaddr(struct compile_state *state, struct triple *func)
+{
+       return do_farg(state, func, 1);
+}
+static struct triple *farg(struct compile_state *state,
+       struct triple *func, unsigned index)
+{
+       return do_farg(state, func, index + 2);
+}
+
+
+static void display_func(struct compile_state *state, FILE *fp, struct triple *func)
+{
+       struct triple *first, *ins;
+       fprintf(fp, "display_func %s\n", func->type->type_ident->name);
+       first = ins = RHS(func, 0);
+       do {
+               if (triple_is_label(state, ins) && ins->use) {
+                       fprintf(fp, "%p:\n", ins);
+               }
+               display_triple(fp, ins);
+
+               if (triple_is_branch(state, ins)) {
+                       fprintf(fp, "\n");
+               }
+               if (ins->next->prev != ins) {
+                       internal_error(state, ins->next, "bad prev");
+               }
+               ins = ins->next;
+       } while(ins != first);
+}
 
 static void verify_use(struct compile_state *state,
        struct triple *user, struct triple *used)
 {
        int size, i;
-       size = TRIPLE_SIZE(user->sizes);
+       size = TRIPLE_SIZE(user);
        for(i = 0; i < size; i++) {
                if (user->param[i] == used) {
                        break;
@@ -1600,13 +2938,17 @@ static void verify_use(struct compile_state *state,
        }
 }
 
-static int find_rhs_use(struct compile_state *state, 
+static int find_rhs_use(struct compile_state *state,
        struct triple *user, struct triple *used)
 {
        struct triple **param;
        int size, i;
        verify_use(state, user, used);
-       size = TRIPLE_RHS(user->sizes);
+
+#if DEBUG_ROMCC_WARNINGS
+#warning "AUDIT ME ->rhs"
+#endif
+       size = user->rhs;
        param = &RHS(user, 0);
        for(i = 0; i < size; i++) {
                if (param[i] == used) {
@@ -1620,12 +2962,13 @@ static void free_triple(struct compile_state *state, struct triple *ptr)
 {
        size_t size;
        size = sizeof(*ptr) - sizeof(ptr->param) +
-               (sizeof(ptr->param[0])*TRIPLE_SIZE(ptr->sizes));
+               (sizeof(ptr->param[0])*TRIPLE_SIZE(ptr));
        ptr->prev->next = ptr->next;
        ptr->next->prev = ptr->prev;
        if (ptr->use) {
                internal_error(state, ptr, "ptr->use != 0");
        }
+       put_occurance(ptr->occurance);
        memset(ptr, -1, size);
        xfree(ptr);
 }
@@ -1634,6 +2977,24 @@ static void release_triple(struct compile_state *state, struct triple *ptr)
 {
        struct triple_set *set, *next;
        struct triple **expr;
+       struct block *block;
+       if (ptr == &unknown_triple) {
+               return;
+       }
+       valid_ins(state, ptr);
+       /* Make certain the we are not the first or last element of a block */
+       block = block_of_triple(state, ptr);
+       if (block) {
+               if ((block->last == ptr) && (block->first == ptr)) {
+                       block->last = block->first = 0;
+               }
+               else if (block->last == ptr) {
+                       block->last = ptr->prev;
+               }
+               else if (block->first == ptr) {
+                       block->first = ptr->next;
+               }
+       }
        /* Remove ptr from use chains where it is the user */
        expr = triple_rhs(state, ptr, 0);
        for(; expr; expr = triple_rhs(state, ptr, expr)) {
@@ -1655,35 +3016,36 @@ static void release_triple(struct compile_state *state, struct triple *ptr)
        }
        expr = triple_targ(state, ptr, 0);
        for(; expr; expr = triple_targ(state, ptr, expr)) {
-               if (*expr) {
+               if (*expr){
                        unuse_triple(*expr, ptr);
                }
        }
        /* Reomve ptr from use chains where it is used */
        for(set = ptr->use; set; set = next) {
                next = set->next;
+               valid_ins(state, set->member);
                expr = triple_rhs(state, set->member, 0);
                for(; expr; expr = triple_rhs(state, set->member, expr)) {
                        if (*expr == ptr) {
-                               *expr = &zero_triple;
+                               *expr = &unknown_triple;
                        }
                }
                expr = triple_lhs(state, set->member, 0);
                for(; expr; expr = triple_lhs(state, set->member, expr)) {
                        if (*expr == ptr) {
-                               *expr = &zero_triple;
+                               *expr = &unknown_triple;
                        }
                }
                expr = triple_misc(state, set->member, 0);
                for(; expr; expr = triple_misc(state, set->member, expr)) {
                        if (*expr == ptr) {
-                               *expr = &zero_triple;
+                               *expr = &unknown_triple;
                        }
                }
                expr = triple_targ(state, set->member, 0);
                for(; expr; expr = triple_targ(state, set->member, expr)) {
                        if (*expr == ptr) {
-                               *expr = &zero_triple;
+                               *expr = &unknown_triple;
                        }
                }
                unuse_triple(ptr, set->member);
@@ -1691,127 +3053,134 @@ static void release_triple(struct compile_state *state, struct triple *ptr)
        free_triple(state, ptr);
 }
 
-static void print_triple(struct compile_state *state, struct triple *ptr);
-
-#define TOK_UNKNOWN     0
-#define TOK_SPACE       1
-#define TOK_SEMI        2
-#define TOK_LBRACE      3
-#define TOK_RBRACE      4
-#define TOK_COMMA       5
-#define TOK_EQ          6
-#define TOK_COLON       7
-#define TOK_LBRACKET    8
-#define TOK_RBRACKET    9
-#define TOK_LPAREN      10
-#define TOK_RPAREN      11
-#define TOK_STAR        12
-#define TOK_DOTS        13
-#define TOK_MORE        14
-#define TOK_LESS        15
-#define TOK_TIMESEQ     16
-#define TOK_DIVEQ       17
-#define TOK_MODEQ       18
-#define TOK_PLUSEQ      19
-#define TOK_MINUSEQ     20
-#define TOK_SLEQ        21
-#define TOK_SREQ        22
-#define TOK_ANDEQ       23
-#define TOK_XOREQ       24
-#define TOK_OREQ        25
-#define TOK_EQEQ        26
-#define TOK_NOTEQ       27
-#define TOK_QUEST       28
-#define TOK_LOGOR       29
-#define TOK_LOGAND      30
-#define TOK_OR          31
-#define TOK_AND         32
-#define TOK_XOR         33
-#define TOK_LESSEQ      34
-#define TOK_MOREEQ      35
-#define TOK_SL          36
-#define TOK_SR          37
-#define TOK_PLUS        38
-#define TOK_MINUS       39
-#define TOK_DIV         40
-#define TOK_MOD         41
-#define TOK_PLUSPLUS    42
-#define TOK_MINUSMINUS  43
-#define TOK_BANG        44
-#define TOK_ARROW       45
-#define TOK_DOT         46
-#define TOK_TILDE       47
-#define TOK_LIT_STRING  48
-#define TOK_LIT_CHAR    49
-#define TOK_LIT_INT     50
-#define TOK_LIT_FLOAT   51
-#define TOK_MACRO       52
-#define TOK_CONCATENATE 53
-
-#define TOK_IDENT       54
-#define TOK_STRUCT_NAME 55
-#define TOK_ENUM_CONST  56
-#define TOK_TYPE_NAME   57
-
-#define TOK_AUTO        58
-#define TOK_BREAK       59
-#define TOK_CASE        60
-#define TOK_CHAR        61
-#define TOK_CONST       62
-#define TOK_CONTINUE    63
-#define TOK_DEFAULT     64
-#define TOK_DO          65
-#define TOK_DOUBLE      66
-#define TOK_ELSE        67
-#define TOK_ENUM        68
-#define TOK_EXTERN      69
-#define TOK_FLOAT       70
-#define TOK_FOR         71
-#define TOK_GOTO        72
-#define TOK_IF          73
-#define TOK_INLINE      74
-#define TOK_INT         75
-#define TOK_LONG        76
-#define TOK_REGISTER    77
-#define TOK_RESTRICT    78
-#define TOK_RETURN      79
-#define TOK_SHORT       80
-#define TOK_SIGNED      81
-#define TOK_SIZEOF      82
-#define TOK_STATIC      83
-#define TOK_STRUCT      84
-#define TOK_SWITCH      85
-#define TOK_TYPEDEF     86
-#define TOK_UNION       87
-#define TOK_UNSIGNED    88
-#define TOK_VOID        89
-#define TOK_VOLATILE    90
-#define TOK_WHILE       91
-#define TOK_ASM         92
-#define TOK_ATTRIBUTE   93
-#define TOK_ALIGNOF     94
+static void print_triples(struct compile_state *state);
+static void print_blocks(struct compile_state *state, const char *func, FILE *fp);
+
+#define TOK_UNKNOWN       0
+#define TOK_SPACE         1
+#define TOK_SEMI          2
+#define TOK_LBRACE        3
+#define TOK_RBRACE        4
+#define TOK_COMMA         5
+#define TOK_EQ            6
+#define TOK_COLON         7
+#define TOK_LBRACKET      8
+#define TOK_RBRACKET      9
+#define TOK_LPAREN        10
+#define TOK_RPAREN        11
+#define TOK_STAR          12
+#define TOK_DOTS          13
+#define TOK_MORE          14
+#define TOK_LESS          15
+#define TOK_TIMESEQ       16
+#define TOK_DIVEQ         17
+#define TOK_MODEQ         18
+#define TOK_PLUSEQ        19
+#define TOK_MINUSEQ       20
+#define TOK_SLEQ          21
+#define TOK_SREQ          22
+#define TOK_ANDEQ         23
+#define TOK_XOREQ         24
+#define TOK_OREQ          25
+#define TOK_EQEQ          26
+#define TOK_NOTEQ         27
+#define TOK_QUEST         28
+#define TOK_LOGOR         29
+#define TOK_LOGAND        30
+#define TOK_OR            31
+#define TOK_AND           32
+#define TOK_XOR           33
+#define TOK_LESSEQ        34
+#define TOK_MOREEQ        35
+#define TOK_SL            36
+#define TOK_SR            37
+#define TOK_PLUS          38
+#define TOK_MINUS         39
+#define TOK_DIV           40
+#define TOK_MOD           41
+#define TOK_PLUSPLUS      42
+#define TOK_MINUSMINUS    43
+#define TOK_BANG          44
+#define TOK_ARROW         45
+#define TOK_DOT           46
+#define TOK_TILDE         47
+#define TOK_LIT_STRING    48
+#define TOK_LIT_CHAR      49
+#define TOK_LIT_INT       50
+#define TOK_LIT_FLOAT     51
+#define TOK_MACRO         52
+#define TOK_CONCATENATE   53
+
+#define TOK_IDENT         54
+#define TOK_STRUCT_NAME   55
+#define TOK_ENUM_CONST    56
+#define TOK_TYPE_NAME     57
+
+#define TOK_AUTO          58
+#define TOK_BREAK         59
+#define TOK_CASE          60
+#define TOK_CHAR          61
+#define TOK_CONST         62
+#define TOK_CONTINUE      63
+#define TOK_DEFAULT       64
+#define TOK_DO            65
+#define TOK_DOUBLE        66
+#define TOK_ELSE          67
+#define TOK_ENUM          68
+#define TOK_EXTERN        69
+#define TOK_FLOAT         70
+#define TOK_FOR           71
+#define TOK_GOTO          72
+#define TOK_IF            73
+#define TOK_INLINE        74
+#define TOK_INT           75
+#define TOK_LONG          76
+#define TOK_REGISTER      77
+#define TOK_RESTRICT      78
+#define TOK_RETURN        79
+#define TOK_SHORT         80
+#define TOK_SIGNED        81
+#define TOK_SIZEOF        82
+#define TOK_STATIC        83
+#define TOK_STRUCT        84
+#define TOK_SWITCH        85
+#define TOK_TYPEDEF       86
+#define TOK_UNION         87
+#define TOK_UNSIGNED      88
+#define TOK_VOID          89
+#define TOK_VOLATILE      90
+#define TOK_WHILE         91
+#define TOK_ASM           92
+#define TOK_ATTRIBUTE     93
+#define TOK_ALIGNOF       94
 #define TOK_FIRST_KEYWORD TOK_AUTO
 #define TOK_LAST_KEYWORD  TOK_ALIGNOF
 
-#define TOK_DEFINE      100
-#define TOK_UNDEF       101
-#define TOK_INCLUDE     102
-#define TOK_LINE        103
-#define TOK_ERROR       104
-#define TOK_WARNING     105
-#define TOK_PRAGMA      106
-#define TOK_IFDEF       107
-#define TOK_IFNDEF      108
-#define TOK_ELIF        109
-#define TOK_ENDIF       110
-
-#define TOK_FIRST_MACRO TOK_DEFINE
-#define TOK_LAST_MACRO  TOK_ENDIF
-         
-#define TOK_EOF         111
+#define TOK_MDEFINE       100
+#define TOK_MDEFINED      101
+#define TOK_MUNDEF        102
+#define TOK_MINCLUDE      103
+#define TOK_MLINE         104
+#define TOK_MERROR        105
+#define TOK_MWARNING      106
+#define TOK_MPRAGMA       107
+#define TOK_MIFDEF        108
+#define TOK_MIFNDEF       109
+#define TOK_MELIF         110
+#define TOK_MENDIF        111
+
+#define TOK_FIRST_MACRO   TOK_MDEFINE
+#define TOK_LAST_MACRO    TOK_MENDIF
+
+#define TOK_MIF           112
+#define TOK_MELSE         113
+#define TOK_MIDENT        114
+
+#define TOK_EOL                  115
+#define TOK_EOF           116
 
 static const char *tokens[] = {
-[TOK_UNKNOWN     ] = "unknown",
+[TOK_UNKNOWN     ] = ":unknown:",
 [TOK_SPACE       ] = ":space:",
 [TOK_SEMI        ] = ";",
 [TOK_LBRACE      ] = "{",
@@ -1906,18 +3275,23 @@ static const char *tokens[] = {
 [TOK_ATTRIBUTE   ] = "__attribute__",
 [TOK_ALIGNOF     ] = "__alignof__",
 
-[TOK_DEFINE      ] = "define",
-[TOK_UNDEF       ] = "undef",
-[TOK_INCLUDE     ] = "include",
-[TOK_LINE        ] = "line",
-[TOK_ERROR       ] = "error",
-[TOK_WARNING     ] = "warning",
-[TOK_PRAGMA      ] = "pragma",
-[TOK_IFDEF       ] = "ifdef",
-[TOK_IFNDEF      ] = "ifndef",
-[TOK_ELIF        ] = "elif",
-[TOK_ENDIF       ] = "endif",
-
+[TOK_MDEFINE     ] = "#define",
+[TOK_MDEFINED    ] = "#defined",
+[TOK_MUNDEF      ] = "#undef",
+[TOK_MINCLUDE    ] = "#include",
+[TOK_MLINE       ] = "#line",
+[TOK_MERROR      ] = "#error",
+[TOK_MWARNING    ] = "#warning",
+[TOK_MPRAGMA     ] = "#pragma",
+[TOK_MIFDEF      ] = "#ifdef",
+[TOK_MIFNDEF     ] = "#ifndef",
+[TOK_MELIF       ] = "#elif",
+[TOK_MENDIF      ] = "#endif",
+
+[TOK_MIF         ] = "#if",
+[TOK_MELSE       ] = "#else",
+[TOK_MIDENT      ] = "#:ident:",
+[TOK_EOL         ] = "EOL",
 [TOK_EOF         ] = "EOF",
 };
 
@@ -1941,7 +3315,7 @@ static struct hash_entry *lookup(
        unsigned int index;
        index = hash(name, name_len);
        entry = state->hash_table[index];
-       while(entry && 
+       while(entry &&
                ((entry->name_len != name_len) ||
                        (memcmp(entry->name, name, name_len) != 0))) {
                entry = entry->next;
@@ -1971,7 +3345,7 @@ static void ident_to_keyword(struct compile_state *state, struct token *tk)
        entry = tk->ident;
        if (entry && ((entry->tok == TOK_TYPE_NAME) ||
                (entry->tok == TOK_ENUM_CONST) ||
-               ((entry->tok >= TOK_FIRST_KEYWORD) && 
+               ((entry->tok >= TOK_FIRST_KEYWORD) &&
                        (entry->tok <= TOK_LAST_KEYWORD)))) {
                tk->tok = entry->tok;
        }
@@ -1981,11 +3355,20 @@ static void ident_to_macro(struct compile_state *state, struct token *tk)
 {
        struct hash_entry *entry;
        entry = tk->ident;
-       if (entry && 
-               (entry->tok >= TOK_FIRST_MACRO) &&
-               (entry->tok <= TOK_LAST_MACRO)) {
+       if (!entry)
+               return;
+       if ((entry->tok >= TOK_FIRST_MACRO) && (entry->tok <= TOK_LAST_MACRO)) {
                tk->tok = entry->tok;
        }
+       else if (entry->tok == TOK_IF) {
+               tk->tok = TOK_MIF;
+       }
+       else if (entry->tok == TOK_ELSE) {
+               tk->tok = TOK_MELSE;
+       }
+       else {
+               tk->tok = TOK_MIDENT;
+       }
 }
 
 static void hash_keyword(
@@ -1999,29 +3382,52 @@ static void hash_keyword(
        entry->tok  = tok;
 }
 
-static void symbol(
+static void romcc_symbol(
        struct compile_state *state, struct hash_entry *ident,
-       struct symbol **chain, struct triple *def, struct type *type)
+       struct symbol **chain, struct triple *def, struct type *type, int depth)
 {
        struct symbol *sym;
-       if (*chain && ((*chain)->scope_depth == state->scope_depth)) {
+       if (*chain && ((*chain)->scope_depth >= depth)) {
                error(state, 0, "%s already defined", ident->name);
        }
        sym = xcmalloc(sizeof(*sym), "symbol");
        sym->ident = ident;
        sym->def   = def;
        sym->type  = type;
-       sym->scope_depth = state->scope_depth;
+       sym->scope_depth = depth;
        sym->next = *chain;
        *chain    = sym;
 }
 
+static void symbol(
+       struct compile_state *state, struct hash_entry *ident,
+       struct symbol **chain, struct triple *def, struct type *type)
+{
+       romcc_symbol(state, ident, chain, def, type, state->scope_depth);
+}
+
+static void var_symbol(struct compile_state *state,
+       struct hash_entry *ident, struct triple *def)
+{
+       if ((def->type->type & TYPE_MASK) == TYPE_PRODUCT) {
+               internal_error(state, 0, "bad var type");
+       }
+       symbol(state, ident, &ident->sym_ident, def, def->type);
+}
+
+static void label_symbol(struct compile_state *state,
+       struct hash_entry *ident, struct triple *label, int depth)
+{
+       romcc_symbol(state, ident, &ident->sym_label, label, &void_type, depth);
+}
+
 static void start_scope(struct compile_state *state)
 {
        state->scope_depth++;
 }
 
-static void end_scope_syms(struct symbol **chain, int depth)
+static void end_scope_syms(struct compile_state *state,
+       struct symbol **chain, int depth)
 {
        struct symbol *sym, *next;
        sym = *chain;
@@ -2038,16 +3444,16 @@ static void end_scope(struct compile_state *state)
        int i;
        int depth;
        /* Walk through the hash table and remove all symbols
-        * in the current scope. 
+        * in the current scope.
         */
        depth = state->scope_depth;
        for(i = 0; i < HASH_TABLE_SIZE; i++) {
                struct hash_entry *entry;
                entry = state->hash_table[i];
                while(entry) {
-                       end_scope_syms(&entry->sym_label,  depth);
-                       end_scope_syms(&entry->sym_struct, depth);
-                       end_scope_syms(&entry->sym_ident,  depth);
+                       end_scope_syms(state, &entry->sym_label, depth);
+                       end_scope_syms(state, &entry->sym_tag,   depth);
+                       end_scope_syms(state, &entry->sym_ident, depth);
                        entry = entry->next;
                }
        }
@@ -2099,17 +3505,168 @@ static void register_keywords(struct compile_state *state)
 
 static void register_macro_keywords(struct compile_state *state)
 {
-       hash_keyword(state, "define",        TOK_DEFINE);
-       hash_keyword(state, "undef",         TOK_UNDEF);
-       hash_keyword(state, "include",       TOK_INCLUDE);
-       hash_keyword(state, "line",          TOK_LINE);
-       hash_keyword(state, "error",         TOK_ERROR);
-       hash_keyword(state, "warning",       TOK_WARNING);
-       hash_keyword(state, "pragma",        TOK_PRAGMA);
-       hash_keyword(state, "ifdef",         TOK_IFDEF);
-       hash_keyword(state, "ifndef",        TOK_IFNDEF);
-       hash_keyword(state, "elif",          TOK_ELIF);
-       hash_keyword(state, "endif",         TOK_ENDIF);
+       hash_keyword(state, "define",        TOK_MDEFINE);
+       hash_keyword(state, "defined",       TOK_MDEFINED);
+       hash_keyword(state, "undef",         TOK_MUNDEF);
+       hash_keyword(state, "include",       TOK_MINCLUDE);
+       hash_keyword(state, "line",          TOK_MLINE);
+       hash_keyword(state, "error",         TOK_MERROR);
+       hash_keyword(state, "warning",       TOK_MWARNING);
+       hash_keyword(state, "pragma",        TOK_MPRAGMA);
+       hash_keyword(state, "ifdef",         TOK_MIFDEF);
+       hash_keyword(state, "ifndef",        TOK_MIFNDEF);
+       hash_keyword(state, "elif",          TOK_MELIF);
+       hash_keyword(state, "endif",         TOK_MENDIF);
+}
+
+
+static void undef_macro(struct compile_state *state, struct hash_entry *ident)
+{
+       if (ident->sym_define != 0) {
+               struct macro *macro;
+               struct macro_arg *arg, *anext;
+               macro = ident->sym_define;
+               ident->sym_define = 0;
+
+               /* Free the macro arguments... */
+               anext = macro->args;
+               while(anext) {
+                       arg = anext;
+                       anext = arg->next;
+                       xfree(arg);
+               }
+
+               /* Free the macro buffer */
+               xfree(macro->buf);
+
+               /* Now free the macro itself */
+               xfree(macro);
+       }
+}
+
+static void do_define_macro(struct compile_state *state,
+       struct hash_entry *ident, const char *body,
+       int argc, struct macro_arg *args)
+{
+       struct macro *macro;
+       struct macro_arg *arg;
+       size_t body_len;
+
+       /* Find the length of the body */
+       body_len = strlen(body);
+       macro = ident->sym_define;
+       if (macro != 0) {
+               int identical_bodies, identical_args;
+               struct macro_arg *oarg;
+               /* Explicitly allow identical redfinitions of the same macro */
+               identical_bodies =
+                       (macro->buf_len == body_len) &&
+                       (memcmp(macro->buf, body, body_len) == 0);
+               identical_args = macro->argc == argc;
+               oarg = macro->args;
+               arg = args;
+               while(identical_args && arg) {
+                       identical_args = oarg->ident == arg->ident;
+                       arg = arg->next;
+                       oarg = oarg->next;
+               }
+               if (identical_bodies && identical_args) {
+                       xfree(body);
+                       return;
+               }
+               error(state, 0, "macro %s already defined\n", ident->name);
+       }
+#if 0
+       fprintf(state->errout, "#define %s: `%*.*s'\n",
+               ident->name, body_len, body_len, body);
+#endif
+       macro = xmalloc(sizeof(*macro), "macro");
+       macro->ident   = ident;
+       macro->buf     = body;
+       macro->buf_len = body_len;
+       macro->args    = args;
+       macro->argc    = argc;
+
+       ident->sym_define = macro;
+}
+
+static void define_macro(
+       struct compile_state *state,
+       struct hash_entry *ident,
+       const char *body, int body_len,
+       int argc, struct macro_arg *args)
+{
+       char *buf;
+       buf = xmalloc(body_len + 1, "macro buf");
+       memcpy(buf, body, body_len);
+       buf[body_len] = '\0';
+       do_define_macro(state, ident, buf, argc, args);
+}
+
+static void register_builtin_macro(struct compile_state *state,
+       const char *name, const char *value)
+{
+       struct hash_entry *ident;
+
+       if (value[0] == '(') {
+               internal_error(state, 0, "Builtin macros with arguments not supported");
+       }
+       ident = lookup(state, name, strlen(name));
+       define_macro(state, ident, value, strlen(value), -1, 0);
+}
+
+static void register_builtin_macros(struct compile_state *state)
+{
+       char buf[30];
+       char scratch[30];
+       time_t now;
+       struct tm *tm;
+       now = time(NULL);
+       tm = localtime(&now);
+
+       register_builtin_macro(state, "__ROMCC__", VERSION_MAJOR);
+       register_builtin_macro(state, "__ROMCC_MINOR__", VERSION_MINOR);
+       register_builtin_macro(state, "__FILE__", "\"This should be the filename\"");
+       register_builtin_macro(state, "__LINE__", "54321");
+
+       strftime(scratch, sizeof(scratch), "%b %e %Y", tm);
+       sprintf(buf, "\"%s\"", scratch);
+       register_builtin_macro(state, "__DATE__", buf);
+
+       strftime(scratch, sizeof(scratch), "%H:%M:%S", tm);
+       sprintf(buf, "\"%s\"", scratch);
+       register_builtin_macro(state, "__TIME__", buf);
+
+       /* I can't be a conforming implementation of C :( */
+       register_builtin_macro(state, "__STDC__", "0");
+       /* In particular I don't conform to C99 */
+       register_builtin_macro(state, "__STDC_VERSION__", "199901L");
+
+}
+
+static void process_cmdline_macros(struct compile_state *state)
+{
+       const char **macro, *name;
+       struct hash_entry *ident;
+       for(macro = state->compiler->defines; (name = *macro); macro++) {
+               const char *body;
+               size_t name_len;
+
+               name_len = strlen(name);
+               body = strchr(name, '=');
+               if (!body) {
+                       body = "\0";
+               } else {
+                       name_len = body - name;
+                       body++;
+               }
+               ident = lookup(state, name, name_len);
+               define_macro(state, ident, body, strlen(body), -1, 0);
+       }
+       for(macro = state->compiler->undefs; (name = *macro); macro++) {
+               ident = lookup(state, name, strlen(name));
+               undef_macro(state, ident);
+       }
 }
 
 static int spacep(int c)
@@ -2121,7 +3678,6 @@ static int spacep(int c)
        case '\f':
        case '\v':
        case '\r':
-       case '\n':
                ret = 1;
                break;
        }
@@ -2132,19 +3688,27 @@ static int digitp(int c)
 {
        int ret = 0;
        switch(c) {
-       case '0': case '1': case '2': case '3': case '4': 
+       case '0': case '1': case '2': case '3': case '4':
        case '5': case '6': case '7': case '8': case '9':
                ret = 1;
                break;
        }
        return ret;
 }
+static int digval(int c)
+{
+       int val = -1;
+       if ((c >= '0') && (c <= '9')) {
+               val = c - '0';
+       }
+       return val;
+}
 
 static int hexdigitp(int c)
 {
        int ret = 0;
        switch(c) {
-       case '0': case '1': case '2': case '3': case '4': 
+       case '0': case '1': case '2': case '3': case '4':
        case '5': case '6': case '7': case '8': case '9':
        case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
        case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
@@ -2153,7 +3717,7 @@ static int hexdigitp(int c)
        }
        return ret;
 }
-static int hexdigval(int c) 
+static int hexdigval(int c)
 {
        int val = -1;
        if ((c >= '0') && (c <= '9')) {
@@ -2172,7 +3736,7 @@ static int octdigitp(int c)
 {
        int ret = 0;
        switch(c) {
-       case '0': case '1': case '2': case '3': 
+       case '0': case '1': case '2': case '3':
        case '4': case '5': case '6': case '7':
                ret = 1;
                break;
@@ -2211,6 +3775,20 @@ static int letterp(int c)
        return ret;
 }
 
+static const char *identifier(const char *str, const char *end)
+{
+       if (letterp(*str)) {
+               for(; str < end; str++) {
+                       int c;
+                       c = *str;
+                       if (!letterp(c) && !digitp(c)) {
+                               break;
+                       }
+               }
+       }
+       return str;
+}
+
 static int char_value(struct compile_state *state,
        const signed char **strp, const signed char *end)
 {
@@ -2230,8 +3808,8 @@ static int char_value(struct compile_state *state,
                case '\\': c = '\\'; str++; break;
                case '?':  c = '?';  str++; break;
                case '\'': c = '\''; str++; break;
-               case '"':  c = '"';  break;
-               case 'x': 
+               case '"':  c = '"';  str++; break;
+               case 'x':
                        c = 0;
                        str++;
                        while((str < end) && hexdigitp(*str)) {
@@ -2240,7 +3818,7 @@ static int char_value(struct compile_state *state,
                                str++;
                        }
                        break;
-               case '0': case '1': case '2': case '3': 
+               case '0': case '1': case '2': case '3':
                case '4': case '5': case '6': case '7':
                        c = 0;
                        while((str < end) && octdigitp(*str)) {
@@ -2258,152 +3836,263 @@ static int char_value(struct compile_state *state,
        return c;
 }
 
-static char *after_digits(char *ptr, char *end)
+static const char *next_char(struct file_state *file, const char *pos, int index)
 {
-       while((ptr < end) && digitp(*ptr)) {
-               ptr++;
+       const char *end = file->buf + file->size;
+       while(pos < end) {
+               /* Lookup the character */
+               int size = 1;
+               int c = *pos;
+               /* Is this a trigraph? */
+               if (file->trigraphs &&
+                       (c == '?') && ((end - pos) >= 3) && (pos[1] == '?'))
+               {
+                       switch(pos[2]) {
+                       case '=': c = '#'; break;
+                       case '/': c = '\\'; break;
+                       case '\'': c = '^'; break;
+                       case '(': c = '['; break;
+                       case ')': c = ']'; break;
+                       case '!': c = '!'; break;
+                       case '<': c = '{'; break;
+                       case '>': c = '}'; break;
+                       case '-': c = '~'; break;
+                       }
+                       if (c != '?') {
+                               size = 3;
+                       }
+               }
+               /* Is this an escaped newline? */
+               if (file->join_lines &&
+                       (c == '\\') && (pos + size < end) && ((pos[1] == '\n') || ((pos[1] == '\r') && (pos[2] == '\n'))))
+               {
+                       int cr_offset = ((pos[1] == '\r') && (pos[2] == '\n'))?1:0;
+                       /* At the start of a line just eat it */
+                       if (pos == file->pos) {
+                               file->line++;
+                               file->report_line++;
+                               file->line_start = pos + size + 1 + cr_offset;
+                       }
+                       pos += size + 1 + cr_offset;
+               }
+               /* Do I need to ga any farther? */
+               else if (index == 0) {
+                       break;
+               }
+               /* Process a normal character */
+               else {
+                       pos += size;
+                       index -= 1;
+               }
        }
-       return ptr;
+       return pos;
 }
 
-static char *after_octdigits(char *ptr, char *end)
+static int get_char(struct file_state *file, const char *pos)
 {
-       while((ptr < end) && octdigitp(*ptr)) {
-               ptr++;
-       }
+       const char *end = file->buf + file->size;
+       int c;
+       c = -1;
+       pos = next_char(file, pos, 0);
+       if (pos < end) {
+               /* Lookup the character */
+               c = *pos;
+               /* If it is a trigraph get the trigraph value */
+               if (file->trigraphs &&
+                       (c == '?') && ((end - pos) >= 3) && (pos[1] == '?'))
+               {
+                       switch(pos[2]) {
+                       case '=': c = '#'; break;
+                       case '/': c = '\\'; break;
+                       case '\'': c = '^'; break;
+                       case '(': c = '['; break;
+                       case ')': c = ']'; break;
+                       case '!': c = '!'; break;
+                       case '<': c = '{'; break;
+                       case '>': c = '}'; break;
+                       case '-': c = '~'; break;
+                       }
+               }
+       }
+       return c;
+}
+
+static void eat_chars(struct file_state *file, const char *targ)
+{
+       const char *pos = file->pos;
+       while(pos < targ) {
+               /* Do we have a newline? */
+               if (pos[0] == '\n') {
+                       file->line++;
+                       file->report_line++;
+                       file->line_start = pos + 1;
+               }
+               pos++;
+       }
+       file->pos = pos;
+}
+
+
+static size_t char_strlen(struct file_state *file, const char *src, const char *end)
+{
+       size_t len;
+       len = 0;
+       while(src < end) {
+               src = next_char(file, src, 1);
+               len++;
+       }
+       return len;
+}
+
+static void char_strcpy(char *dest,
+       struct file_state *file, const char *src, const char *end)
+{
+       while(src < end) {
+               int c;
+               c = get_char(file, src);
+               src = next_char(file, src, 1);
+               *dest++ = c;
+       }
+}
+
+static char *char_strdup(struct file_state *file,
+       const char *start, const char *end, const char *id)
+{
+       char *str;
+       size_t str_len;
+       str_len = char_strlen(file, start, end);
+       str = xcmalloc(str_len + 1, id);
+       char_strcpy(str, file, start, end);
+       str[str_len] = '\0';
+       return str;
+}
+
+static const char *after_digits(struct file_state *file, const char *ptr)
+{
+       while(digitp(get_char(file, ptr))) {
+               ptr = next_char(file, ptr, 1);
+       }
        return ptr;
 }
 
-static char *after_hexdigits(char *ptr, char *end)
+static const char *after_octdigits(struct file_state *file, const char *ptr)
 {
-       while((ptr < end) && hexdigitp(*ptr)) {
-               ptr++;
+       while(octdigitp(get_char(file, ptr))) {
+               ptr = next_char(file, ptr, 1);
+       }
+       return ptr;
+}
+
+static const char *after_hexdigits(struct file_state *file, const char *ptr)
+{
+       while(hexdigitp(get_char(file, ptr))) {
+               ptr = next_char(file, ptr, 1);
+       }
+       return ptr;
+}
+
+static const char *after_alnums(struct file_state *file, const char *ptr)
+{
+       int c;
+       c = get_char(file, ptr);
+       while(letterp(c) || digitp(c)) {
+               ptr = next_char(file, ptr, 1);
+               c = get_char(file, ptr);
        }
        return ptr;
 }
 
-static void save_string(struct compile_state *state, 
-       struct token *tk, char *start, char *end, const char *id)
+static void save_string(struct file_state *file,
+       struct token *tk, const char *start, const char *end, const char *id)
 {
        char *str;
-       int str_len;
+
        /* Create a private copy of the string */
-       str_len = end - start + 1;
-       str = xmalloc(str_len + 1, id);
-       memcpy(str, start, str_len);
-       str[str_len] = '\0';
+       str = char_strdup(file, start, end, id);
 
        /* Store the copy in the token */
        tk->val.str = str;
-       tk->str_len = str_len;
+       tk->str_len = strlen(str);
 }
-static void next_token(struct compile_state *state, int index)
+
+static void raw_next_token(struct compile_state *state,
+       struct file_state *file, struct token *tk)
 {
-       struct file_state *file;
-       struct token *tk;
-       char *token;
+       const char *token;
        int c, c1, c2, c3;
-       char *tokp, *end;
+       const char *tokp;
+       int eat;
        int tok;
-next_token:
-       file = state->file;
-       tk = &state->token[index];
+
        tk->str_len = 0;
        tk->ident = 0;
-       token = tokp = file->pos;
-       end = file->buf + file->size;
+       token = tokp = next_char(file, file->pos, 0);
        tok = TOK_UNKNOWN;
-       c = -1;
-       if (tokp < end) {
-               c = *tokp;
-       }
-       c1 = -1;
-       if ((tokp + 1) < end) {
-               c1 = tokp[1];
-       }
-       c2 = -1;
-       if ((tokp + 2) < end) {
-               c2 = tokp[2];
-       }
-       c3 = -1;
-       if ((tokp + 3) < end) {
-               c3 = tokp[3];
-       }
-       if (tokp >= end) {
+       c  = get_char(file, tokp);
+       tokp = next_char(file, tokp, 1);
+       eat = 0;
+       c1 = get_char(file, tokp);
+       c2 = get_char(file, next_char(file, tokp, 1));
+       c3 = get_char(file, next_char(file, tokp, 2));
+
+       /* The end of the file */
+       if (c == -1) {
                tok = TOK_EOF;
-               tokp = end;
        }
        /* Whitespace */
        else if (spacep(c)) {
                tok = TOK_SPACE;
-               while ((tokp < end) && spacep(c)) {
-                       if (c == '\n') {
-                               file->line++;
-                               file->line_start = tokp + 1;
-                       }
-                       c = *(++tokp);
-               }
-               if (!spacep(c)) {
-                       tokp--;
+               while (spacep(get_char(file, tokp))) {
+                       tokp = next_char(file, tokp, 1);
                }
        }
        /* EOL Comments */
        else if ((c == '/') && (c1 == '/')) {
                tok = TOK_SPACE;
-               for(tokp += 2; tokp < end; tokp++) {
-                       c = *tokp;
+               tokp = next_char(file, tokp, 1);
+               while((c = get_char(file, tokp)) != -1) {
+                       /* Advance to the next character only after we verify
+                        * the current character is not a newline.
+                        * EOL is special to the preprocessor so we don't
+                        * want to loose any.
+                        */
                        if (c == '\n') {
-                               file->line++;
-                               file->line_start = tokp +1;
                                break;
                        }
+                       tokp = next_char(file, tokp, 1);
                }
        }
        /* Comments */
        else if ((c == '/') && (c1 == '*')) {
-               int line;
-               char *line_start;
-               line = file->line;
-               line_start = file->line_start;
-               for(tokp += 2; (end - tokp) >= 2; tokp++) {
-                       c = *tokp;
-                       if (c == '\n') {
-                               line++;
-                               line_start = tokp +1;
-                       }
-                       else if ((c == '*') && (tokp[1] == '/')) {
+               tokp = next_char(file, tokp, 2);
+               c = c2;
+               while((c1 = get_char(file, tokp)) != -1) {
+                       tokp = next_char(file, tokp, 1);
+                       if ((c == '*') && (c1 == '/')) {
                                tok = TOK_SPACE;
-                               tokp += 1;
                                break;
                        }
+                       c = c1;
                }
                if (tok == TOK_UNKNOWN) {
                        error(state, 0, "unterminated comment");
                }
-               file->line = line;
-               file->line_start = line_start;
        }
        /* string constants */
-       else if ((c == '"') ||
-               ((c == 'L') && (c1 == '"'))) {
-               int line;
-               char *line_start;
-               int wchar;
-               line = file->line;
-               line_start = file->line_start;
-               wchar = 0;
+       else if ((c == '"') || ((c == 'L') && (c1 == '"'))) {
+               int multiline;
+
+               multiline = 0;
                if (c == 'L') {
-                       wchar = 1;
-                       tokp++;
+                       tokp = next_char(file, tokp, 1);
                }
-               for(tokp += 1; tokp < end; tokp++) {
-                       c = *tokp;
+               while((c = get_char(file, tokp)) != -1) {
+                       tokp = next_char(file, tokp, 1);
                        if (c == '\n') {
-                               line++;
-                               line_start = tokp + 1;
+                               multiline = 1;
                        }
-                       else if ((c == '\\') && (tokp +1 < end)) {
-                               tokp++;
+                       else if (c == '\\') {
+                               tokp = next_char(file, tokp, 1);
                        }
                        else if (c == '"') {
                                tok = TOK_LIT_STRING;
@@ -2413,36 +4102,28 @@ next_token:
                if (tok == TOK_UNKNOWN) {
                        error(state, 0, "unterminated string constant");
                }
-               if (line != file->line) {
+               if (multiline) {
                        warning(state, 0, "multiline string constant");
                }
-               file->line = line;
-               file->line_start = line_start;
 
                /* Save the string value */
-               save_string(state, tk, token, tokp, "literal string");
+               save_string(file, tk, token, tokp, "literal string");
        }
        /* character constants */
-       else if ((c == '\'') ||
-               ((c == 'L') && (c1 == '\''))) {
-               int line;
-               char *line_start;
-               int wchar;
-               line = file->line;
-               line_start = file->line_start;
-               wchar = 0;
+       else if ((c == '\'') || ((c == 'L') && (c1 == '\''))) {
+               int multiline;
+
+               multiline = 0;
                if (c == 'L') {
-                       wchar = 1;
-                       tokp++;
+                       tokp = next_char(file, tokp, 1);
                }
-               for(tokp += 1; tokp < end; tokp++) {
-                       c = *tokp;
+               while((c = get_char(file, tokp)) != -1) {
+                       tokp = next_char(file, tokp, 1);
                        if (c == '\n') {
-                               line++;
-                               line_start = tokp + 1;
+                               multiline = 1;
                        }
-                       else if ((c == '\\') && (tokp +1 < end)) {
-                               tokp++;
+                       else if (c == '\\') {
+                               tokp = next_char(file, tokp, 1);
                        }
                        else if (c == '\'') {
                                tok = TOK_LIT_CHAR;
@@ -2452,21 +4133,19 @@ next_token:
                if (tok == TOK_UNKNOWN) {
                        error(state, 0, "unterminated character constant");
                }
-               if (line != file->line) {
+               if (multiline) {
                        warning(state, 0, "multiline character constant");
                }
-               file->line = line;
-               file->line_start = line_start;
 
                /* Save the character value */
-               save_string(state, tk, token, tokp, "literal character");
+               save_string(file, tk, token, tokp, "literal character");
        }
-       /* integer and floating constants 
+       /* integer and floating constants
         * Integer Constants
         * {digits}
         * 0[Xx]{hexdigits}
         * 0{octdigit}+
-        * 
+        *
          * Floating constants
         * {digits}.{digits}[Ee][+-]?{digits}
         * {digits}.{digits}
@@ -2474,121 +4153,140 @@ next_token:
         * .{digits}[Ee][+-]?{digits}
         * .{digits}
         */
-       
        else if (digitp(c) || ((c == '.') && (digitp(c1)))) {
-               char *next, *new;
+               const char *next;
                int is_float;
+               int cn;
                is_float = 0;
                if (c != '.') {
-                       next = after_digits(tokp, end);
+                       next = after_digits(file, tokp);
                }
                else {
-                       next = tokp;
-               }
-               if (next[0] == '.') {
-                       new = after_digits(next, end);
-                       is_float = (new != next);
-                       next = new;
-               }
-               if ((next[0] == 'e') || (next[0] == 'E')) {
-                       if (((next + 1) < end) && 
-                               ((next[1] == '+') || (next[1] == '-'))) {
-                               next++;
+                       next = token;
+               }
+               cn = get_char(file, next);
+               if (cn == '.') {
+                       next = next_char(file, next, 1);
+                       next = after_digits(file, next);
+                       is_float = 1;
+               }
+               cn = get_char(file, next);
+               if ((cn == 'e') || (cn == 'E')) {
+                       const char *new;
+                       next = next_char(file, next, 1);
+                       cn = get_char(file, next);
+                       if ((cn == '+') || (cn == '-')) {
+                               next = next_char(file, next, 1);
                        }
-                       new = after_digits(next, end);
-                       is_float = (new != next);
+                       new = after_digits(file, next);
+                       is_float |= (new != next);
                        next = new;
                }
                if (is_float) {
                        tok = TOK_LIT_FLOAT;
-                       if ((next < end) && (
-                               (next[0] == 'f') ||
-                               (next[0] == 'F') ||
-                               (next[0] == 'l') ||
-                               (next[0] == 'L'))
-                               ) {
-                               next++;
+                       cn = get_char(file, next);
+                       if ((cn  == 'f') || (cn == 'F') || (cn == 'l') || (cn == 'L')) {
+                               next = next_char(file, next, 1);
                        }
                }
                if (!is_float && digitp(c)) {
                        tok = TOK_LIT_INT;
                        if ((c == '0') && ((c1 == 'x') || (c1 == 'X'))) {
-                               next = after_hexdigits(tokp + 2, end);
+                               next = next_char(file, tokp, 1);
+                               next = after_hexdigits(file, next);
                        }
                        else if (c == '0') {
-                               next = after_octdigits(tokp, end);
+                               next = after_octdigits(file, tokp);
                        }
                        else {
-                               next = after_digits(tokp, end);
+                               next = after_digits(file, tokp);
                        }
                        /* crazy integer suffixes */
-                       if ((next < end) && 
-                               ((next[0] == 'u') || (next[0] == 'U'))) { 
-                               next++;
-                               if ((next < end) &&
-                                       ((next[0] == 'l') || (next[0] == 'L'))) {
-                                       next++;
+                       cn = get_char(file, next);
+                       if ((cn == 'u') || (cn == 'U')) {
+                               next = next_char(file, next, 1);
+                               cn = get_char(file, next);
+                               if ((cn == 'l') || (cn == 'L')) {
+                                       next = next_char(file, next, 1);
+                                       cn = get_char(file, next);
+                               }
+                               if ((cn == 'l') || (cn == 'L')) {
+                                       next = next_char(file, next, 1);
                                }
                        }
-                       else if ((next < end) &&
-                               ((next[0] == 'l') || (next[0] == 'L'))) {
-                               next++;
-                               if ((next < end) && 
-                                       ((next[0] == 'u') || (next[0] == 'U'))) { 
-                                       next++;
+                       else if ((cn == 'l') || (cn == 'L')) {
+                               next = next_char(file, next, 1);
+                               cn = get_char(file, next);
+                               if ((cn == 'l') || (cn == 'L')) {
+                                       next = next_char(file, next, 1);
+                                       cn = get_char(file, next);
+                               }
+                               if ((cn == 'u') || (cn == 'U')) {
+                                       next = next_char(file, next, 1);
                                }
                        }
                }
-               tokp = next - 1;
+               tokp = next;
 
                /* Save the integer/floating point value */
-               save_string(state, tk, token, tokp, "literal number");
+               save_string(file, tk, token, tokp, "literal number");
        }
        /* identifiers */
        else if (letterp(c)) {
                tok = TOK_IDENT;
-               for(tokp += 1; tokp < end; tokp++) {
-                       c = *tokp;
-                       if (!letterp(c) && !digitp(c)) {
-                               break;
-                       }
+
+               /* Find and save the identifier string */
+               tokp = after_alnums(file, tokp);
+               save_string(file, tk, token, tokp, "identifier");
+
+               /* Look up to see which identifier it is */
+               tk->ident = lookup(state, tk->val.str, tk->str_len);
+
+               /* Free the identifier string */
+               tk->str_len = 0;
+               xfree(tk->val.str);
+
+               /* See if this identifier can be macro expanded */
+               tk->val.notmacro = 0;
+               c = get_char(file, tokp);
+               if (c == '$') {
+                       tokp = next_char(file, tokp, 1);
+                       tk->val.notmacro = 1;
                }
-               tokp -= 1;
-               tk->ident = lookup(state, token, tokp +1 - token);
        }
        /* C99 alternate macro characters */
-       else if ((c == '%') && (c1 == ':') && (c2 == '%') && (c3 == ':')) { 
-               tokp += 3; 
-               tok = TOK_CONCATENATE; 
-       }
-       else if ((c == '.') && (c1 == '.') && (c2 == '.')) { tokp += 2; tok = TOK_DOTS; }
-       else if ((c == '<') && (c1 == '<') && (c2 == '=')) { tokp += 2; tok = TOK_SLEQ; }
-       else if ((c == '>') && (c1 == '>') && (c2 == '=')) { tokp += 2; tok = TOK_SREQ; }
-       else if ((c == '*') && (c1 == '=')) { tokp += 1; tok = TOK_TIMESEQ; }
-       else if ((c == '/') && (c1 == '=')) { tokp += 1; tok = TOK_DIVEQ; }
-       else if ((c == '%') && (c1 == '=')) { tokp += 1; tok = TOK_MODEQ; }
-       else if ((c == '+') && (c1 == '=')) { tokp += 1; tok = TOK_PLUSEQ; }
-       else if ((c == '-') && (c1 == '=')) { tokp += 1; tok = TOK_MINUSEQ; }
-       else if ((c == '&') && (c1 == '=')) { tokp += 1; tok = TOK_ANDEQ; }
-       else if ((c == '^') && (c1 == '=')) { tokp += 1; tok = TOK_XOREQ; }
-       else if ((c == '|') && (c1 == '=')) { tokp += 1; tok = TOK_OREQ; }
-       else if ((c == '=') && (c1 == '=')) { tokp += 1; tok = TOK_EQEQ; }
-       else if ((c == '!') && (c1 == '=')) { tokp += 1; tok = TOK_NOTEQ; }
-       else if ((c == '|') && (c1 == '|')) { tokp += 1; tok = TOK_LOGOR; }
-       else if ((c == '&') && (c1 == '&')) { tokp += 1; tok = TOK_LOGAND; }
-       else if ((c == '<') && (c1 == '=')) { tokp += 1; tok = TOK_LESSEQ; }
-       else if ((c == '>') && (c1 == '=')) { tokp += 1; tok = TOK_MOREEQ; }
-       else if ((c == '<') && (c1 == '<')) { tokp += 1; tok = TOK_SL; }
-       else if ((c == '>') && (c1 == '>')) { tokp += 1; tok = TOK_SR; }
-       else if ((c == '+') && (c1 == '+')) { tokp += 1; tok = TOK_PLUSPLUS; }
-       else if ((c == '-') && (c1 == '-')) { tokp += 1; tok = TOK_MINUSMINUS; }
-       else if ((c == '-') && (c1 == '>')) { tokp += 1; tok = TOK_ARROW; }
-       else if ((c == '<') && (c1 == ':')) { tokp += 1; tok = TOK_LBRACKET; }
-       else if ((c == ':') && (c1 == '>')) { tokp += 1; tok = TOK_RBRACKET; }
-       else if ((c == '<') && (c1 == '%')) { tokp += 1; tok = TOK_LBRACE; }
-       else if ((c == '%') && (c1 == '>')) { tokp += 1; tok = TOK_RBRACE; }
-       else if ((c == '%') && (c1 == ':')) { tokp += 1; tok = TOK_MACRO; }
-       else if ((c == '#') && (c1 == '#')) { tokp += 1; tok = TOK_CONCATENATE; }
+       else if ((c == '%') && (c1 == ':') && (c2 == '%') && (c3 == ':')) {
+               eat += 3;
+               tok = TOK_CONCATENATE;
+       }
+       else if ((c == '.') && (c1 == '.') && (c2 == '.')) { eat += 2; tok = TOK_DOTS; }
+       else if ((c == '<') && (c1 == '<') && (c2 == '=')) { eat += 2; tok = TOK_SLEQ; }
+       else if ((c == '>') && (c1 == '>') && (c2 == '=')) { eat += 2; tok = TOK_SREQ; }
+       else if ((c == '*') && (c1 == '=')) { eat += 1; tok = TOK_TIMESEQ; }
+       else if ((c == '/') && (c1 == '=')) { eat += 1; tok = TOK_DIVEQ; }
+       else if ((c == '%') && (c1 == '=')) { eat += 1; tok = TOK_MODEQ; }
+       else if ((c == '+') && (c1 == '=')) { eat += 1; tok = TOK_PLUSEQ; }
+       else if ((c == '-') && (c1 == '=')) { eat += 1; tok = TOK_MINUSEQ; }
+       else if ((c == '&') && (c1 == '=')) { eat += 1; tok = TOK_ANDEQ; }
+       else if ((c == '^') && (c1 == '=')) { eat += 1; tok = TOK_XOREQ; }
+       else if ((c == '|') && (c1 == '=')) { eat += 1; tok = TOK_OREQ; }
+       else if ((c == '=') && (c1 == '=')) { eat += 1; tok = TOK_EQEQ; }
+       else if ((c == '!') && (c1 == '=')) { eat += 1; tok = TOK_NOTEQ; }
+       else if ((c == '|') && (c1 == '|')) { eat += 1; tok = TOK_LOGOR; }
+       else if ((c == '&') && (c1 == '&')) { eat += 1; tok = TOK_LOGAND; }
+       else if ((c == '<') && (c1 == '=')) { eat += 1; tok = TOK_LESSEQ; }
+       else if ((c == '>') && (c1 == '=')) { eat += 1; tok = TOK_MOREEQ; }
+       else if ((c == '<') && (c1 == '<')) { eat += 1; tok = TOK_SL; }
+       else if ((c == '>') && (c1 == '>')) { eat += 1; tok = TOK_SR; }
+       else if ((c == '+') && (c1 == '+')) { eat += 1; tok = TOK_PLUSPLUS; }
+       else if ((c == '-') && (c1 == '-')) { eat += 1; tok = TOK_MINUSMINUS; }
+       else if ((c == '-') && (c1 == '>')) { eat += 1; tok = TOK_ARROW; }
+       else if ((c == '<') && (c1 == ':')) { eat += 1; tok = TOK_LBRACKET; }
+       else if ((c == ':') && (c1 == '>')) { eat += 1; tok = TOK_RBRACKET; }
+       else if ((c == '<') && (c1 == '%')) { eat += 1; tok = TOK_LBRACE; }
+       else if ((c == '%') && (c1 == '>')) { eat += 1; tok = TOK_RBRACE; }
+       else if ((c == '%') && (c1 == ':')) { eat += 1; tok = TOK_MACRO; }
+       else if ((c == '#') && (c1 == '#')) { eat += 1; tok = TOK_CONCATENATE; }
        else if (c == ';') { tok = TOK_SEMI; }
        else if (c == '{') { tok = TOK_LBRACE; }
        else if (c == '}') { tok = TOK_RBRACE; }
@@ -2614,756 +4312,755 @@ next_token:
        else if (c == '.') { tok = TOK_DOT; }
        else if (c == '~') { tok = TOK_TILDE; }
        else if (c == '#') { tok = TOK_MACRO; }
-       if (tok == TOK_MACRO) {
-               /* Only match preprocessor directives at the start of a line */
-               char *ptr;
-               for(ptr = file->line_start; spacep(*ptr); ptr++)
-                       ;
-               if (ptr != tokp) {
-                       tok = TOK_UNKNOWN;
-               }
-       }
-       if (tok == TOK_UNKNOWN) {
-               error(state, 0, "unknown token");
-       }
+       else if (c == '\n') { tok = TOK_EOL; }
 
-       file->pos = tokp + 1;
+       tokp = next_char(file, tokp, eat);
+       eat_chars(file, tokp);
        tk->tok = tok;
-       if (tok == TOK_IDENT) {
-               ident_to_keyword(state, tk);
-       }
-       /* Don't return space tokens. */
-       if (tok == TOK_SPACE) {
-               goto next_token;
-       }
+       tk->pos = token;
 }
 
-static void compile_macro(struct compile_state *state, struct token *tk)
+static void check_tok(struct compile_state *state, struct token *tk, int tok)
 {
-       struct file_state *file;
-       struct hash_entry *ident;
-       ident = tk->ident;
-       file = xmalloc(sizeof(*file), "file_state");
-       file->basename = xstrdup(tk->ident->name);
-       file->dirname = xstrdup("");
-       file->size = ident->sym_define->buf_len;
-       file->buf = xmalloc(file->size +2,  file->basename);
-       memcpy(file->buf, ident->sym_define->buf, file->size);
-       file->buf[file->size] = '\n';
-       file->buf[file->size + 1] = '\0';
-       file->pos = file->buf;
-       file->line_start = file->pos;
-       file->line = 1;
-       file->prev = state->file;
-       state->file = file;
+       if (tk->tok != tok) {
+               const char *name1, *name2;
+               name1 = tokens[tk->tok];
+               name2 = "";
+               if ((tk->tok == TOK_IDENT) || (tk->tok == TOK_MIDENT)) {
+                       name2 = tk->ident->name;
+               }
+               error(state, 0, "\tfound %s %s expected %s",
+                       name1, name2, tokens[tok]);
+       }
 }
 
-
-static int mpeek(struct compile_state *state, int index)
+struct macro_arg_value {
+       struct hash_entry *ident;
+       char *value;
+       size_t len;
+};
+static struct macro_arg_value *read_macro_args(
+       struct compile_state *state, struct macro *macro,
+       struct file_state *file, struct token *tk)
 {
-       struct token *tk;
-       int rescan;
-       tk = &state->token[index + 1];
-       if (tk->tok == -1) {
-               next_token(state, index + 1);
+       struct macro_arg_value *argv;
+       struct macro_arg *arg;
+       int paren_depth;
+       int i;
+
+       if (macro->argc == 0) {
+               do {
+                       raw_next_token(state, file, tk);
+               } while(tk->tok == TOK_SPACE);
+               return NULL;
        }
-       do {
-               rescan = 0;
-               if ((tk->tok == TOK_EOF) && 
-                       (state->file != state->macro_file) &&
-                       (state->file->prev)) {
-                       struct file_state *file = state->file;
-                       state->file = file->prev;
-                       /* file->basename is used keep it */
-                       xfree(file->dirname);
-                       xfree(file->buf);
-                       xfree(file);
-                       next_token(state, index + 1);
-                       rescan = 1;
+       argv = xcmalloc(sizeof(*argv) * macro->argc, "macro args");
+       for(i = 0, arg = macro->args; arg; arg = arg->next, i++) {
+               argv[i].value = 0;
+               argv[i].len   = 0;
+               argv[i].ident = arg->ident;
+       }
+       paren_depth = 0;
+       i = 0;
+
+       for(;;) {
+               const char *start;
+               size_t len;
+               start = file->pos;
+               raw_next_token(state, file, tk);
+
+               if (!paren_depth && (tk->tok == TOK_COMMA) &&
+                       (argv[i].ident != state->i___VA_ARGS__))
+               {
+                       i++;
+                       if (i >= macro->argc) {
+                               error(state, 0, "too many args to %s\n",
+                                       macro->ident->name);
+                       }
+                       continue;
                }
-               else if (tk->ident && tk->ident->sym_define) {
-                       compile_macro(state, tk);
-                       next_token(state, index + 1);
-                       rescan = 1;
+
+               if (tk->tok == TOK_LPAREN) {
+                       paren_depth++;
                }
-       } while(rescan);
-       /* Don't show the token on the next line */
-       if (state->macro_line < state->macro_file->line) {
-               return TOK_EOF;
+
+               if (tk->tok == TOK_RPAREN) {
+                       if (paren_depth == 0) {
+                               break;
+                       }
+                       paren_depth--;
+               }
+               if (tk->tok == TOK_EOF) {
+                       error(state, 0, "End of file encountered while parsing macro arguments");
+               }
+
+               len = char_strlen(file, start, file->pos);
+               argv[i].value = xrealloc(
+                       argv[i].value, argv[i].len + len, "macro args");
+               char_strcpy((char *)argv[i].value + argv[i].len, file, start, file->pos);
+               argv[i].len += len;
        }
-       return state->token[index +1].tok;
+       if (i != macro->argc -1) {
+               error(state, 0, "missing %s arg %d\n",
+                       macro->ident->name, i +2);
+       }
+       return argv;
 }
 
-static void meat(struct compile_state *state, int index, int tok)
+
+static void free_macro_args(struct macro *macro, struct macro_arg_value *argv)
 {
-       int next_tok;
        int i;
-       next_tok = mpeek(state, index);
-       if (next_tok != tok) {
-               const char *name1, *name2;
-               name1 = tokens[next_tok];
-               name2 = "";
-               if (next_tok == TOK_IDENT) {
-                       name2 = state->token[index + 1].ident->name;
-               }
-               error(state, 0, "found %s %s expected %s", 
-                       name1, name2, tokens[tok]);
-       }
-       /* Free the old token value */
-       if (state->token[index].str_len) {
-               memset((void *)(state->token[index].val.str), -1, 
-                       state->token[index].str_len);
-               xfree(state->token[index].val.str);
-       }
-       for(i = index; i < sizeof(state->token)/sizeof(state->token[0]) - 1; i++) {
-               state->token[i] = state->token[i + 1];
+       for(i = 0; i < macro->argc; i++) {
+               xfree(argv[i].value);
        }
-       memset(&state->token[i], 0, sizeof(state->token[i]));
-       state->token[i].tok = -1;
+       xfree(argv);
 }
 
-static long_t mcexpr(struct compile_state *state, int index);
+struct macro_buf {
+       char *str;
+       size_t len, pos;
+};
 
-static long_t mprimary_expr(struct compile_state *state, int index)
+static void grow_macro_buf(struct compile_state *state,
+       const char *id, struct macro_buf *buf,
+       size_t grow)
 {
-       long_t val;
-       int tok;
-       tok = mpeek(state, index);
-       while(state->token[index + 1].ident && 
-               state->token[index + 1].ident->sym_define) {
-               meat(state, index, tok);
-               compile_macro(state, &state->token[index]);
-               tok = mpeek(state, index);
-       }
-       switch(tok) {
-       case TOK_LPAREN:
-               meat(state, index, TOK_LPAREN);
-               val = mcexpr(state, index);
-               meat(state, index, TOK_RPAREN);
-               break;
-       case TOK_LIT_INT:
-       {
-               char *end;
-               meat(state, index, TOK_LIT_INT);
-               errno = 0;
-               val = strtol(state->token[index].val.str, &end, 0);
-               if (((val == LONG_MIN) || (val == LONG_MAX)) &&
-                       (errno == ERANGE)) {
-                       error(state, 0, "Integer constant to large");
-               }
-               break;
-       }
-       default:
-               meat(state, index, TOK_LIT_INT);
-               val = 0;
+       if ((buf->pos + grow) >= buf->len) {
+               buf->str = xrealloc(buf->str, buf->len + grow, id);
+               buf->len += grow;
        }
-       return val;
 }
-static long_t munary_expr(struct compile_state *state, int index)
+
+static void append_macro_text(struct compile_state *state,
+       const char *id, struct macro_buf *buf,
+       const char *fstart, size_t flen)
 {
-       long_t val;
-       switch(mpeek(state, index)) {
-       case TOK_PLUS:
-               meat(state, index, TOK_PLUS);
-               val = munary_expr(state, index);
-               val = + val;
-               break;
-       case TOK_MINUS:
-               meat(state, index, TOK_MINUS);
-               val = munary_expr(state, index);
-               val = - val;
-               break;
-       case TOK_TILDE:
-               meat(state, index, TOK_BANG);
-               val = munary_expr(state, index);
-               val = ~ val;
-               break;
-       case TOK_BANG:
-               meat(state, index, TOK_BANG);
-               val = munary_expr(state, index);
-               val = ! val;
-               break;
-       default:
-               val = mprimary_expr(state, index);
-               break;
-       }
-       return val;
-       
+       grow_macro_buf(state, id, buf, flen);
+       memcpy(buf->str + buf->pos, fstart, flen);
+#if 0
+       fprintf(state->errout, "append: `%*.*s' `%*.*s'\n",
+               buf->pos, buf->pos, buf->str,
+               flen, flen, buf->str + buf->pos);
+#endif
+       buf->pos += flen;
 }
-static long_t mmul_expr(struct compile_state *state, int index)
+
+
+static void append_macro_chars(struct compile_state *state,
+       const char *id, struct macro_buf *buf,
+       struct file_state *file, const char *start, const char *end)
 {
-       long_t val;
-       int done;
-       val = munary_expr(state, index);
-       do {
-               long_t right;
-               done = 0;
-               switch(mpeek(state, index)) {
-               case TOK_STAR:
-                       meat(state, index, TOK_STAR);
-                       right = munary_expr(state, index);
-                       val = val * right;
-                       break;
-               case TOK_DIV:
-                       meat(state, index, TOK_DIV);
-                       right = munary_expr(state, index);
-                       val = val / right;
-                       break;
-               case TOK_MOD:
-                       meat(state, index, TOK_MOD);
-                       right = munary_expr(state, index);
-                       val = val % right;
-                       break;
-               default:
-                       done = 1;
-                       break;
+       size_t flen;
+       flen = char_strlen(file, start, end);
+       grow_macro_buf(state, id, buf, flen);
+       char_strcpy(buf->str + buf->pos, file, start, end);
+#if 0
+       fprintf(state->errout, "append: `%*.*s' `%*.*s'\n",
+               buf->pos, buf->pos, buf->str,
+               flen, flen, buf->str + buf->pos);
+#endif
+       buf->pos += flen;
+}
+
+static int compile_macro(struct compile_state *state,
+       struct file_state **filep, struct token *tk);
+
+static void macro_expand_args(struct compile_state *state,
+       struct macro *macro, struct macro_arg_value *argv, struct token *tk)
+{
+       int i;
+
+       for(i = 0; i < macro->argc; i++) {
+               struct file_state fmacro, *file;
+               struct macro_buf buf;
+
+               fmacro.prev        = 0;
+               fmacro.basename    = argv[i].ident->name;
+               fmacro.dirname     = "";
+               fmacro.buf         = (char *)argv[i].value;
+               fmacro.size        = argv[i].len;
+               fmacro.pos         = fmacro.buf;
+               fmacro.line        = 1;
+               fmacro.line_start  = fmacro.buf;
+               fmacro.report_line = 1;
+               fmacro.report_name = fmacro.basename;
+               fmacro.report_dir  = fmacro.dirname;
+               fmacro.macro       = 1;
+               fmacro.trigraphs   = 0;
+               fmacro.join_lines  = 0;
+
+               buf.len = argv[i].len;
+               buf.str = xmalloc(buf.len, argv[i].ident->name);
+               buf.pos = 0;
+
+               file = &fmacro;
+               for(;;) {
+                       raw_next_token(state, file, tk);
+
+                       /* If we have recursed into another macro body
+                        * get out of it.
+                        */
+                       if (tk->tok == TOK_EOF) {
+                               struct file_state *old;
+                               old = file;
+                               file = file->prev;
+                               if (!file) {
+                                       break;
+                               }
+                               /* old->basename is used keep it */
+                               xfree(old->dirname);
+                               xfree(old->buf);
+                               xfree(old);
+                               continue;
+                       }
+                       else if (tk->ident && tk->ident->sym_define) {
+                               if (compile_macro(state, &file, tk)) {
+                                       continue;
+                               }
+                       }
+
+                       append_macro_chars(state, macro->ident->name, &buf,
+                               file, tk->pos, file->pos);
                }
-       } while(!done);
 
-       return val;
+               xfree(argv[i].value);
+               argv[i].value = buf.str;
+               argv[i].len   = buf.pos;
+       }
+       return;
 }
 
-static long_t madd_expr(struct compile_state *state, int index)
+static void expand_macro(struct compile_state *state,
+       struct macro *macro, struct macro_buf *buf,
+       struct macro_arg_value *argv, struct token *tk)
 {
-       long_t val;
-       int done;
-       val = mmul_expr(state, index);
-       do {
-               long_t right;
-               done = 0;
-               switch(mpeek(state, index)) {
-               case TOK_PLUS:
-                       meat(state, index, TOK_PLUS);
-                       right = mmul_expr(state, index);
-                       val = val + right;
+       struct file_state fmacro;
+       const char space[] = " ";
+       const char *fstart;
+       size_t flen;
+       int i, j;
+
+       /* Place the macro body in a dummy file */
+       fmacro.prev        = 0;
+       fmacro.basename    = macro->ident->name;
+       fmacro.dirname     = "";
+       fmacro.buf         = macro->buf;
+       fmacro.size        = macro->buf_len;
+       fmacro.pos         = fmacro.buf;
+       fmacro.line        = 1;
+       fmacro.line_start  = fmacro.buf;
+       fmacro.report_line = 1;
+       fmacro.report_name = fmacro.basename;
+       fmacro.report_dir  = fmacro.dirname;
+       fmacro.macro       = 1;
+       fmacro.trigraphs   = 0;
+       fmacro.join_lines  = 0;
+
+       /* Allocate a buffer to hold the macro expansion */
+       buf->len = macro->buf_len + 3;
+       buf->str = xmalloc(buf->len, macro->ident->name);
+       buf->pos = 0;
+
+       fstart = fmacro.pos;
+       raw_next_token(state, &fmacro, tk);
+       while(tk->tok != TOK_EOF) {
+               flen = fmacro.pos - fstart;
+               switch(tk->tok) {
+               case TOK_IDENT:
+                       for(i = 0; i < macro->argc; i++) {
+                               if (argv[i].ident == tk->ident) {
+                                       break;
+                               }
+                       }
+                       if (i >= macro->argc) {
+                               break;
+                       }
+                       /* Substitute macro parameter */
+                       fstart = argv[i].value;
+                       flen   = argv[i].len;
                        break;
-               case TOK_MINUS:
-                       meat(state, index, TOK_MINUS);
-                       right = mmul_expr(state, index);
-                       val = val - right;
+               case TOK_MACRO:
+                       if (macro->argc < 0) {
+                               break;
+                       }
+                       do {
+                               raw_next_token(state, &fmacro, tk);
+                       } while(tk->tok == TOK_SPACE);
+                       check_tok(state, tk, TOK_IDENT);
+                       for(i = 0; i < macro->argc; i++) {
+                               if (argv[i].ident == tk->ident) {
+                                       break;
+                               }
+                       }
+                       if (i >= macro->argc) {
+                               error(state, 0, "parameter `%s' not found",
+                                       tk->ident->name);
+                       }
+                       /* Stringize token */
+                       append_macro_text(state, macro->ident->name, buf, "\"", 1);
+                       for(j = 0; j < argv[i].len; j++) {
+                               char *str = argv[i].value + j;
+                               size_t len = 1;
+                               if (*str == '\\') {
+                                       str = "\\";
+                                       len = 2;
+                               }
+                               else if (*str == '"') {
+                                       str = "\\\"";
+                                       len = 2;
+                               }
+                               append_macro_text(state, macro->ident->name, buf, str, len);
+                       }
+                       append_macro_text(state, macro->ident->name, buf, "\"", 1);
+                       fstart = 0;
+                       flen   = 0;
+                       break;
+               case TOK_CONCATENATE:
+                       /* Concatenate tokens */
+                       /* Delete the previous whitespace token */
+                       if (buf->str[buf->pos - 1] == ' ') {
+                               buf->pos -= 1;
+                       }
+                       /* Skip the next sequence of whitspace tokens */
+                       do {
+                               fstart = fmacro.pos;
+                               raw_next_token(state, &fmacro, tk);
+                       } while(tk->tok == TOK_SPACE);
+                       /* Restart at the top of the loop.
+                        * I need to process the non white space token.
+                        */
+                       continue;
+                       break;
+               case TOK_SPACE:
+                       /* Collapse multiple spaces into one */
+                       if (buf->str[buf->pos - 1] != ' ') {
+                               fstart = space;
+                               flen   = 1;
+                       } else {
+                               fstart = 0;
+                               flen   = 0;
+                       }
                        break;
                default:
-                       done = 1;
                        break;
                }
-       } while(!done);
 
-       return val;
+               append_macro_text(state, macro->ident->name, buf, fstart, flen);
+
+               fstart = fmacro.pos;
+               raw_next_token(state, &fmacro, tk);
+       }
 }
 
-static long_t mshift_expr(struct compile_state *state, int index)
+static void tag_macro_name(struct compile_state *state,
+       struct macro *macro, struct macro_buf *buf,
+       struct token *tk)
 {
-       long_t val;
-       int done;
-       val = madd_expr(state, index);
-       do {
-               long_t right;
-               done = 0;
-               switch(mpeek(state, index)) {
-               case TOK_SL:
-                       meat(state, index, TOK_SL);
-                       right = madd_expr(state, index);
-                       val = val << right;
-                       break;
-               case TOK_SR:
-                       meat(state, index, TOK_SR);
-                       right = madd_expr(state, index);
-                       val = val >> right;
-                       break;
-               default:
-                       done = 1;
-                       break;
+       /* Guard all instances of the macro name in the replacement
+        * text from further macro expansion.
+        */
+       struct file_state fmacro;
+       const char *fstart;
+       size_t flen;
+
+       /* Put the old macro expansion buffer in a file */
+       fmacro.prev        = 0;
+       fmacro.basename    = macro->ident->name;
+       fmacro.dirname     = "";
+       fmacro.buf         = buf->str;
+       fmacro.size        = buf->pos;
+       fmacro.pos         = fmacro.buf;
+       fmacro.line        = 1;
+       fmacro.line_start  = fmacro.buf;
+       fmacro.report_line = 1;
+       fmacro.report_name = fmacro.basename;
+       fmacro.report_dir  = fmacro.dirname;
+       fmacro.macro       = 1;
+       fmacro.trigraphs   = 0;
+       fmacro.join_lines  = 0;
+
+       /* Allocate a new macro expansion buffer */
+       buf->len = macro->buf_len + 3;
+       buf->str = xmalloc(buf->len, macro->ident->name);
+       buf->pos = 0;
+
+       fstart = fmacro.pos;
+       raw_next_token(state, &fmacro, tk);
+       while(tk->tok != TOK_EOF) {
+               flen = fmacro.pos - fstart;
+               if ((tk->tok == TOK_IDENT) &&
+                       (tk->ident == macro->ident) &&
+                       (tk->val.notmacro == 0))
+               {
+                       append_macro_text(state, macro->ident->name, buf, fstart, flen);
+                       fstart = "$";
+                       flen   = 1;
                }
-       } while(!done);
 
-       return val;
-}
+               append_macro_text(state, macro->ident->name, buf, fstart, flen);
 
-static long_t mrel_expr(struct compile_state *state, int index)
-{
-       long_t val;
-       int done;
-       val = mshift_expr(state, index);
-       do {
-               long_t right;
-               done = 0;
-               switch(mpeek(state, index)) {
-               case TOK_LESS:
-                       meat(state, index, TOK_LESS);
-                       right = mshift_expr(state, index);
-                       val = val < right;
-                       break;
-               case TOK_MORE:
-                       meat(state, index, TOK_MORE);
-                       right = mshift_expr(state, index);
-                       val = val > right;
-                       break;
-               case TOK_LESSEQ:
-                       meat(state, index, TOK_LESSEQ);
-                       right = mshift_expr(state, index);
-                       val = val <= right;
-                       break;
-               case TOK_MOREEQ:
-                       meat(state, index, TOK_MOREEQ);
-                       right = mshift_expr(state, index);
-                       val = val >= right;
-                       break;
-               default:
-                       done = 1;
-                       break;
-               }
-       } while(!done);
-       return val;
+               fstart = fmacro.pos;
+               raw_next_token(state, &fmacro, tk);
+       }
+       xfree(fmacro.buf);
 }
 
-static long_t meq_expr(struct compile_state *state, int index)
+static int compile_macro(struct compile_state *state,
+       struct file_state **filep, struct token *tk)
 {
-       long_t val;
-       int done;
-       val = mrel_expr(state, index);
-       do {
-               long_t right;
-               done = 0;
-               switch(mpeek(state, index)) {
-               case TOK_EQEQ:
-                       meat(state, index, TOK_EQEQ);
-                       right = mrel_expr(state, index);
-                       val = val == right;
-                       break;
-               case TOK_NOTEQ:
-                       meat(state, index, TOK_NOTEQ);
-                       right = mrel_expr(state, index);
-                       val = val != right;
-                       break;
-               default:
-                       done = 1;
-                       break;
-               }
-       } while(!done);
-       return val;
-}
+       struct file_state *file;
+       struct hash_entry *ident;
+       struct macro *macro;
+       struct macro_arg_value *argv;
+       struct macro_buf buf;
 
-static long_t mand_expr(struct compile_state *state, int index)
-{
-       long_t val;
-       val = meq_expr(state, index);
-       if (mpeek(state, index) == TOK_AND) {
-               long_t right;
-               meat(state, index, TOK_AND);
-               right = meq_expr(state, index);
-               val = val & right;
+#if 0
+       fprintf(state->errout, "macro: %s\n", tk->ident->name);
+#endif
+       ident = tk->ident;
+       macro = ident->sym_define;
+
+       /* If this token comes from a macro expansion ignore it */
+       if (tk->val.notmacro) {
+               return 0;
+       }
+       /* If I am a function like macro and the identifier is not followed
+        * by a left parenthesis, do nothing.
+        */
+       if ((macro->argc >= 0) && (get_char(*filep, (*filep)->pos) != '(')) {
+               return 0;
        }
-       return val;
-}
 
-static long_t mxor_expr(struct compile_state *state, int index)
-{
-       long_t val;
-       val = mand_expr(state, index);
-       if (mpeek(state, index) == TOK_XOR) {
-               long_t right;
-               meat(state, index, TOK_XOR);
-               right = mand_expr(state, index);
-               val = val ^ right;
+       /* Read in the macro arguments */
+       argv = 0;
+       if (macro->argc >= 0) {
+               raw_next_token(state, *filep, tk);
+               check_tok(state, tk, TOK_LPAREN);
+
+               argv = read_macro_args(state, macro, *filep, tk);
+
+               check_tok(state, tk, TOK_RPAREN);
        }
-       return val;
-}
+       /* Macro expand the macro arguments */
+       macro_expand_args(state, macro, argv, tk);
 
-static long_t mor_expr(struct compile_state *state, int index)
-{
-       long_t val;
-       val = mxor_expr(state, index);
-       if (mpeek(state, index) == TOK_OR) {
-               long_t right;
-               meat(state, index, TOK_OR);
-               right = mxor_expr(state, index);
-               val = val | right;
+       buf.str = 0;
+       buf.len = 0;
+       buf.pos = 0;
+       if (ident == state->i___FILE__) {
+               buf.len = strlen(state->file->basename) + 1 + 2 + 3;
+               buf.str = xmalloc(buf.len, ident->name);
+               sprintf(buf.str, "\"%s\"", state->file->basename);
+               buf.pos = strlen(buf.str);
        }
-       return val;
+       else if (ident == state->i___LINE__) {
+               buf.len = 30;
+               buf.str = xmalloc(buf.len, ident->name);
+               sprintf(buf.str, "%d", state->file->line);
+               buf.pos = strlen(buf.str);
+       }
+       else {
+               expand_macro(state, macro, &buf, argv, tk);
+       }
+       /* Tag the macro name with a $ so it will no longer
+        * be regonized as a canidate for macro expansion.
+        */
+       tag_macro_name(state, macro, &buf, tk);
+
+#if 0
+       fprintf(state->errout, "%s: %d -> `%*.*s'\n",
+               ident->name, buf.pos, buf.pos, (int)(buf.pos), buf.str);
+#endif
+
+       free_macro_args(macro, argv);
+
+       file = xmalloc(sizeof(*file), "file_state");
+       file->prev        = *filep;
+       file->basename    = xstrdup(ident->name);
+       file->dirname     = xstrdup("");
+       file->buf         = buf.str;
+       file->size        = buf.pos;
+       file->pos         = file->buf;
+       file->line        = 1;
+       file->line_start  = file->pos;
+       file->report_line = 1;
+       file->report_name = file->basename;
+       file->report_dir  = file->dirname;
+       file->macro       = 1;
+       file->trigraphs   = 0;
+       file->join_lines  = 0;
+       *filep = file;
+       return 1;
 }
 
-static long_t mland_expr(struct compile_state *state, int index)
+static void eat_tokens(struct compile_state *state, int targ_tok)
 {
-       long_t val;
-       val = mor_expr(state, index);
-       if (mpeek(state, index) == TOK_LOGAND) {
-               long_t right;
-               meat(state, index, TOK_LOGAND);
-               right = mor_expr(state, index);
-               val = val && right;
+       if (state->eat_depth > 0) {
+               internal_error(state, 0, "Already eating...");
        }
-       return val;
+       state->eat_depth = state->if_depth;
+       state->eat_targ = targ_tok;
 }
-static long_t mlor_expr(struct compile_state *state, int index)
+static int if_eat(struct compile_state *state)
 {
-       long_t val;
-       val = mland_expr(state, index);
-       if (mpeek(state, index) == TOK_LOGOR) {
-               long_t right;
-               meat(state, index, TOK_LOGOR);
-               right = mland_expr(state, index);
-               val = val || right;
-       }
-       return val;
+       return state->eat_depth > 0;
 }
-
-static long_t mcexpr(struct compile_state *state, int index)
+static int if_value(struct compile_state *state)
 {
-       return mlor_expr(state, index);
+       int index, offset;
+       index = state->if_depth / CHAR_BIT;
+       offset = state->if_depth % CHAR_BIT;
+       return !!(state->if_bytes[index] & (1 << (offset)));
 }
-static void preprocess(struct compile_state *state, int index)
+static void set_if_value(struct compile_state *state, int value)
 {
-       /* Doing much more with the preprocessor would require
-        * a parser and a major restructuring.
-        * Postpone that for later.
-        */
-       struct file_state *file;
-       struct token *tk;
-       int line;
-       int tok;
-       
-       file = state->file;
-       tk = &state->token[index];
-       state->macro_line = line = file->line;
-       state->macro_file = file;
-
-       next_token(state, index);
-       ident_to_macro(state, tk);
-       if (tk->tok == TOK_IDENT) {
-               error(state, 0, "undefined preprocessing directive `%s'",
-                       tk->ident->name);
-       }
-       switch(tk->tok) {
-       case TOK_UNDEF:
-       case TOK_LINE:
-       case TOK_PRAGMA:
-               if (state->if_value < 0) {
-                       break;
-               }
-               warning(state, 0, "Ignoring preprocessor directive: %s", 
-                       tk->ident->name);
-               break;
-       case TOK_ELIF:
-               error(state, 0, "#elif not supported");
-#warning "FIXME multiple #elif and #else in an #if do not work properly"
-               if (state->if_depth == 0) {
-                       error(state, 0, "#elif without #if");
-               }
-               /* If the #if was taken the #elif just disables the following code */
-               if (state->if_value >= 0) {
-                       state->if_value = - state->if_value;
-               }
-               /* If the previous #if was not taken see if the #elif enables the 
-                * trailing code.
-                */
-               else if ((state->if_value < 0) && 
-                       (state->if_depth == - state->if_value))
-               {
-                       if (mcexpr(state, index) != 0) {
-                               state->if_value = state->if_depth;
-                       }
-                       else {
-                               state->if_value = - state->if_depth;
-                       }
-               }
-               break;
-       case TOK_IF:
-               state->if_depth++;
-               if (state->if_value < 0) {
-                       break;
-               }
-               if (mcexpr(state, index) != 0) {
-                       state->if_value = state->if_depth;
-               }
-               else {
-                       state->if_value = - state->if_depth;
-               }
-               break;
-       case TOK_IFNDEF:
-               state->if_depth++;
-               if (state->if_value < 0) {
-                       break;
-               }
-               next_token(state, index);
-               if ((line != file->line) || (tk->tok != TOK_IDENT)) {
-                       error(state, 0, "Invalid macro name");
-               }
-               if (tk->ident->sym_define == 0) {
-                       state->if_value = state->if_depth;
-               } 
-               else {
-                       state->if_value = - state->if_depth;
-               }
-               break;
-       case TOK_IFDEF:
-               state->if_depth++;
-               if (state->if_value < 0) {
-                       break;
-               }
-               next_token(state, index);
-               if ((line != file->line) || (tk->tok != TOK_IDENT)) {
-                       error(state, 0, "Invalid macro name");
-               }
-               if (tk->ident->sym_define != 0) {
-                       state->if_value = state->if_depth;
-               }
-               else {
-                       state->if_value = - state->if_depth;
-               }
-               break;
-       case TOK_ELSE:
-               if (state->if_depth == 0) {
-                       error(state, 0, "#else without #if");
-               }
-               if ((state->if_value >= 0) ||
-                       ((state->if_value < 0) && 
-                               (state->if_depth == -state->if_value)))
-               {
-                       state->if_value = - state->if_value;
-               }
-               break;
-       case TOK_ENDIF:
-               if (state->if_depth == 0) {
-                       error(state, 0, "#endif without #if");
-               }
-               if ((state->if_value >= 0) ||
-                       ((state->if_value < 0) &&
-                               (state->if_depth == -state->if_value))) 
-               {
-                       state->if_value = state->if_depth - 1;
-               }
-               state->if_depth--;
-               break;
-       case TOK_DEFINE:
-       {
-               struct hash_entry *ident;
-               struct macro *macro;
-               char *ptr;
-               
-               if (state->if_value < 0) /* quit early when #if'd out */
-                       break;
-
-               meat(state, index, TOK_IDENT);
-               ident = tk->ident;
-               
-
-               if (*file->pos == '(') {
-#warning "FIXME macros with arguments not supported"
-                       error(state, 0, "Macros with arguments not supported");
-               }
-
-               /* Find the end of the line to get an estimate of
-                * the macro's length.
-                */
-               for(ptr = file->pos; *ptr != '\n'; ptr++)  
-                       ;
-
-               if (ident->sym_define != 0) {
-                       error(state, 0, "macro %s already defined\n", ident->name);
-               }
-               macro = xmalloc(sizeof(*macro), "macro");
-               macro->ident = ident;
-               macro->buf_len = ptr - file->pos +1;
-               macro->buf = xmalloc(macro->buf_len +2, "macro buf");
-
-               memcpy(macro->buf, file->pos, macro->buf_len);
-               macro->buf[macro->buf_len] = '\n';
-               macro->buf[macro->buf_len +1] = '\0';
+       int index, offset;
+       index = state->if_depth / CHAR_BIT;
+       offset = state->if_depth % CHAR_BIT;
 
-               ident->sym_define = macro;
-               break;
+       state->if_bytes[index] &= ~(1 << offset);
+       if (value) {
+               state->if_bytes[index] |= (1 << offset);
        }
-       case TOK_ERROR:
-       {
-               char *end;
-               int len;
-               /* Find the end of the line */
-               for(end = file->pos; *end != '\n'; end++)
-                       ;
-               len = (end - file->pos);
-               if (state->if_value >= 0) {
-                       error(state, 0, "%*.*s", len, len, file->pos);
-               }
-               file->pos = end;
-               break;
+}
+static void in_if(struct compile_state *state, const char *name)
+{
+       if (state->if_depth <= 0) {
+               error(state, 0, "%s without #if", name);
        }
-       case TOK_WARNING:
-       {
-               char *end;
-               int len;
-               /* Find the end of the line */
-               for(end = file->pos; *end != '\n'; end++)
-                       ;
-               len = (end - file->pos);
-               if (state->if_value >= 0) {
-                       warning(state, 0, "%*.*s", len, len, file->pos);
-               }
-               file->pos = end;
-               break;
+}
+static void enter_if(struct compile_state *state)
+{
+       state->if_depth += 1;
+       if (state->if_depth > MAX_PP_IF_DEPTH) {
+               error(state, 0, "#if depth too great");
        }
-       case TOK_INCLUDE:
-       {
-               char *name;
-               char *ptr;
-               int local;
-               local = 0;
-               name = 0;
-               next_token(state, index);
-               if (tk->tok == TOK_LIT_STRING) {
-                       const char *token;
-                       int name_len;
-                       name = xmalloc(tk->str_len, "include");
-                       token = tk->val.str +1;
-                       name_len = tk->str_len -2;
-                       if (*token == '"') {
-                               token++;
-                               name_len--;
-                       }
-                       memcpy(name, token, name_len);
-                       name[name_len] = '\0';
-                       local = 1;
-               }
-               else if (tk->tok == TOK_LESS) {
-                       char *start, *end;
-                       start = file->pos;
-                       for(end = start; *end != '\n'; end++) {
-                               if (*end == '>') {
-                                       break;
-                               }
-                       }
-                       if (*end == '\n') {
-                               error(state, 0, "Unterminated included directive");
-                       }
-                       name = xmalloc(end - start + 1, "include");
-                       memcpy(name, start, end - start);
-                       name[end - start] = '\0';
-                       file->pos = end +1;
-                       local = 0;
-               }
-               else {
-                       error(state, 0, "Invalid include directive");
-               }
-               /* Error if there are any characters after the include */
-               for(ptr = file->pos; *ptr != '\n'; ptr++) {
-                       if (!isspace(*ptr)) {
-                               error(state, 0, "garbage after include directive");
-                       }
-               }
-               if (state->if_value >= 0) {
-                       compile_file(state, name, local);
-               }
-               xfree(name);
-               next_token(state, index);
-               return;
+}
+static void reenter_if(struct compile_state *state, const char *name)
+{
+       in_if(state, name);
+       if ((state->eat_depth == state->if_depth) &&
+               (state->eat_targ == TOK_MELSE)) {
+               state->eat_depth = 0;
+               state->eat_targ = 0;
        }
-       default:
-               /* Ignore # without a following ident */
-               if (tk->tok == TOK_IDENT) {
-                       error(state, 0, "Invalid preprocessor directive: %s", 
-                               tk->ident->name);
-               }
-               break;
+}
+static void enter_else(struct compile_state *state, const char *name)
+{
+       in_if(state, name);
+       if ((state->eat_depth == state->if_depth) &&
+               (state->eat_targ == TOK_MELSE)) {
+               state->eat_depth = 0;
+               state->eat_targ = 0;
        }
-       /* Consume the rest of the macro line */
-       do {
-               tok = mpeek(state, index);
-               meat(state, index, tok);
-       } while(tok != TOK_EOF);
-       return;
+}
+static void exit_if(struct compile_state *state, const char *name)
+{
+       in_if(state, name);
+       if (state->eat_depth == state->if_depth) {
+               state->eat_depth = 0;
+               state->eat_targ = 0;
+       }
+       state->if_depth -= 1;
 }
 
-static void token(struct compile_state *state, int index)
+static void raw_token(struct compile_state *state, struct token *tk)
 {
        struct file_state *file;
-       struct token *tk;
        int rescan;
 
-       tk = &state->token[index];
-       next_token(state, index);
+       file = state->file;
+       raw_next_token(state, file, tk);
        do {
                rescan = 0;
                file = state->file;
-               if (tk->tok == TOK_EOF && file->prev) {
+               /* Exit out of an include directive or macro call */
+               if ((tk->tok == TOK_EOF) &&
+                       (file != state->macro_file) && file->prev)
+               {
                        state->file = file->prev;
                        /* file->basename is used keep it */
                        xfree(file->dirname);
                        xfree(file->buf);
                        xfree(file);
-                       next_token(state, index);
+                       file = 0;
+                       raw_next_token(state, state->file, tk);
                        rescan = 1;
                }
-               else if (tk->tok == TOK_MACRO) {
-                       preprocess(state, index);
+       } while(rescan);
+}
+
+static void pp_token(struct compile_state *state, struct token *tk)
+{
+       int rescan;
+
+       raw_token(state, tk);
+       do {
+               rescan = 0;
+               if (tk->tok == TOK_SPACE) {
+                       raw_token(state, tk);
                        rescan = 1;
                }
+               else if (tk->tok == TOK_IDENT) {
+                       if (state->token_base == 0) {
+                               ident_to_keyword(state, tk);
+                       } else {
+                               ident_to_macro(state, tk);
+                       }
+               }
+       } while(rescan);
+}
+
+static void preprocess(struct compile_state *state, struct token *tk);
+
+static void token(struct compile_state *state, struct token *tk)
+{
+       int rescan;
+       pp_token(state, tk);
+       do {
+               rescan = 0;
+               /* Process a macro directive */
+               if (tk->tok == TOK_MACRO) {
+                       /* Only match preprocessor directives at the start of a line */
+                       const char *ptr;
+                       ptr = state->file->line_start;
+                       while((ptr < tk->pos)
+                               && spacep(get_char(state->file, ptr)))
+                       {
+                               ptr = next_char(state->file, ptr, 1);
+                       }
+                       if (ptr == tk->pos) {
+                               preprocess(state, tk);
+                               rescan = 1;
+                       }
+               }
+               /* Expand a macro call */
                else if (tk->ident && tk->ident->sym_define) {
-                       compile_macro(state, tk);
-                       next_token(state, index);
+                       rescan = compile_macro(state, &state->file, tk);
+                       if (rescan) {
+                               pp_token(state, tk);
+                       }
+               }
+               /* Eat tokens disabled by the preprocessor
+                * (Unless we are parsing a preprocessor directive
+                */
+               else if (if_eat(state) && (state->token_base == 0)) {
+                       pp_token(state, tk);
                        rescan = 1;
                }
-               else if (state->if_value < 0) {
-                       next_token(state, index);
+               /* Make certain EOL only shows up in preprocessor directives */
+               else if ((tk->tok == TOK_EOL) && (state->token_base == 0)) {
+                       pp_token(state, tk);
                        rescan = 1;
                }
+               /* Error on unknown tokens */
+               else if (tk->tok == TOK_UNKNOWN) {
+                       error(state, 0, "unknown token");
+               }
        } while(rescan);
 }
 
-static int peek(struct compile_state *state)
-{
-       if (state->token[1].tok == -1) {
-               token(state, 1);
-       }
-       return state->token[1].tok;
-}
 
-static int peek2(struct compile_state *state)
+static inline struct token *get_token(struct compile_state *state, int offset)
 {
-       if (state->token[1].tok == -1) {
-               token(state, 1);
-       }
-       if (state->token[2].tok == -1) {
-               token(state, 2);
+       int index;
+       index = state->token_base + offset;
+       if (index >= sizeof(state->token)/sizeof(state->token[0])) {
+               internal_error(state, 0, "token array to small");
        }
-       return state->token[2].tok;
+       return &state->token[index];
 }
 
-static void eat(struct compile_state *state, int tok)
+static struct token *do_eat_token(struct compile_state *state, int tok)
 {
-       int next_tok;
+       struct token *tk;
        int i;
-       next_tok = peek(state);
-       if (next_tok != tok) {
-               const char *name1, *name2;
-               name1 = tokens[next_tok];
-               name2 = "";
-               if (next_tok == TOK_IDENT) {
-                       name2 = state->token[1].ident->name;
-               }
-               error(state, 0, "\tfound %s %s expected %s",
-                       name1, name2 ,tokens[tok]);
-       }
+       check_tok(state, get_token(state, 1), tok);
+
        /* Free the old token value */
-       if (state->token[0].str_len) {
-               xfree((void *)(state->token[0].val.str));
+       tk = get_token(state, 0);
+       if (tk->str_len) {
+               memset((void *)tk->val.str, -1, tk->str_len);
+               xfree(tk->val.str);
        }
-       for(i = 0; i < sizeof(state->token)/sizeof(state->token[0]) - 1; i++) {
+       /* Overwrite the old token with newer tokens */
+       for(i = state->token_base; i < sizeof(state->token)/sizeof(state->token[0]) - 1; i++) {
                state->token[i] = state->token[i + 1];
        }
+       /* Clear the last token */
        memset(&state->token[i], 0, sizeof(state->token[i]));
        state->token[i].tok = -1;
-}
 
-#warning "FIXME do not hardcode the include paths"
-static char *include_paths[] = {
-       "/home/eric/projects/linuxbios/checkin/solo/freebios2/src/include",
-       "/home/eric/projects/linuxbios/checkin/solo/freebios2/src/arch/i386/include",
-       "/home/eric/projects/linuxbios/checkin/solo/freebios2/src",
-       0
-};
+       /* Return the token */
+       return tk;
+}
 
-static void compile_file(struct compile_state *state, const char *filename, int local)
+static int raw_peek(struct compile_state *state)
 {
-       char cwd[4096];
+       struct token *tk1;
+       tk1 = get_token(state, 1);
+       if (tk1->tok == -1) {
+               raw_token(state, tk1);
+       }
+       return tk1->tok;
+}
+
+static struct token *raw_eat(struct compile_state *state, int tok)
+{
+       raw_peek(state);
+       return do_eat_token(state, tok);
+}
+
+static int pp_peek(struct compile_state *state)
+{
+       struct token *tk1;
+       tk1 = get_token(state, 1);
+       if (tk1->tok == -1) {
+               pp_token(state, tk1);
+       }
+       return tk1->tok;
+}
+
+static struct token *pp_eat(struct compile_state *state, int tok)
+{
+       pp_peek(state);
+       return do_eat_token(state, tok);
+}
+
+static int peek(struct compile_state *state)
+{
+       struct token *tk1;
+       tk1 = get_token(state, 1);
+       if (tk1->tok == -1) {
+               token(state, tk1);
+       }
+       return tk1->tok;
+}
+
+static int peek2(struct compile_state *state)
+{
+       struct token *tk1, *tk2;
+       tk1 = get_token(state, 1);
+       tk2 = get_token(state, 2);
+       if (tk1->tok == -1) {
+               token(state, tk1);
+       }
+       if (tk2->tok == -1) {
+               token(state, tk2);
+       }
+       return tk2->tok;
+}
+
+static struct token *eat(struct compile_state *state, int tok)
+{
+       peek(state);
+       return do_eat_token(state, tok);
+}
+
+static void compile_file(struct compile_state *state, const char *filename, int local)
+{
+       char cwd[MAX_CWD_SIZE];
        const char *subdir, *base;
        int subdir_len;
        struct file_state *file;
@@ -3387,16 +5084,15 @@ static void compile_file(struct compile_state *state, const char *filename, int
        if (getcwd(cwd, sizeof(cwd)) == 0) {
                die("cwd buffer to small");
        }
-       
-       if (subdir[0] == '/') {
+       if ((subdir[0] == '/') || ((subdir[1] == ':') && ((subdir[2] == '/') || (subdir[2] == '\\')))) {
                file->dirname = xmalloc(subdir_len + 1, "dirname");
                memcpy(file->dirname, subdir, subdir_len);
                file->dirname[subdir_len] = '\0';
        }
        else {
-               char *dir;
+               const char *dir;
                int dirlen;
-               char **path;
+               const char **path;
                /* Find the appropriate directory... */
                dir = 0;
                if (!state->file && exists(cwd, filename)) {
@@ -3405,13 +5101,13 @@ static void compile_file(struct compile_state *state, const char *filename, int
                if (local && state->file && exists(state->file->dirname, filename)) {
                        dir = state->file->dirname;
                }
-               for(path = include_paths; !dir && *path; path++) {
+               for(path = state->compiler->include_paths; !dir && *path; path++) {
                        if (exists(*path, filename)) {
                                dir = *path;
                        }
                }
                if (!dir) {
-                       error(state, 0, "Cannot find `%s'\n", filename);
+                       error(state, 0, "Cannot open `%s'\n", filename);
                }
                dirlen = strlen(dir);
                file->dirname = xmalloc(dirlen + 1 + subdir_len + 1, "dirname");
@@ -3421,17 +5117,432 @@ static void compile_file(struct compile_state *state, const char *filename, int
                file->dirname[dirlen + 1 + subdir_len] = '\0';
        }
        file->buf = slurp_file(file->dirname, file->basename, &file->size);
-       xchdir(cwd);
 
        file->pos = file->buf;
        file->line_start = file->pos;
        file->line = 1;
 
+       file->report_line = 1;
+       file->report_name = file->basename;
+       file->report_dir  = file->dirname;
+       file->macro       = 0;
+       file->trigraphs   = (state->compiler->flags & COMPILER_TRIGRAPHS)? 1: 0;
+       file->join_lines  = 1;
+
        file->prev = state->file;
        state->file = file;
-       
-       process_trigraphs(state);
-       splice_lines(state);
+}
+
+static struct triple *constant_expr(struct compile_state *state);
+static void integral(struct compile_state *state, struct triple *def);
+
+static int mcexpr(struct compile_state *state)
+{
+       struct triple *cvalue;
+       cvalue = constant_expr(state);
+       integral(state, cvalue);
+       if (cvalue->op != OP_INTCONST) {
+               error(state, 0, "integer constant expected");
+       }
+       return cvalue->u.cval != 0;
+}
+
+static void preprocess(struct compile_state *state, struct token *current_token)
+{
+       /* Doing much more with the preprocessor would require
+        * a parser and a major restructuring.
+        * Postpone that for later.
+        */
+       int old_token_base;
+       int tok;
+
+       state->macro_file = state->file;
+
+       old_token_base = state->token_base;
+       state->token_base = current_token - state->token;
+
+       tok = pp_peek(state);
+       switch(tok) {
+       case TOK_LIT_INT:
+       {
+               struct token *tk;
+               int override_line;
+               tk = pp_eat(state, TOK_LIT_INT);
+               override_line = strtoul(tk->val.str, 0, 10);
+               /* I have a preprocessor  line marker parse it */
+               if (pp_peek(state) == TOK_LIT_STRING) {
+                       const char *token, *base;
+                       char *name, *dir;
+                       int name_len, dir_len;
+                       tk = pp_eat(state, TOK_LIT_STRING);
+                       name = xmalloc(tk->str_len, "report_name");
+                       token = tk->val.str + 1;
+                       base = strrchr(token, '/');
+                       name_len = tk->str_len -2;
+                       if (base != 0) {
+                               dir_len = base - token;
+                               base++;
+                               name_len -= base - token;
+                       } else {
+                               dir_len = 0;
+                               base = token;
+                       }
+                       memcpy(name, base, name_len);
+                       name[name_len] = '\0';
+                       dir = xmalloc(dir_len + 1, "report_dir");
+                       memcpy(dir, token, dir_len);
+                       dir[dir_len] = '\0';
+                       state->file->report_line = override_line - 1;
+                       state->file->report_name = name;
+                       state->file->report_dir = dir;
+                       state->file->macro      = 0;
+               }
+               break;
+       }
+       case TOK_MLINE:
+       {
+               struct token *tk;
+               pp_eat(state, TOK_MLINE);
+               tk = eat(state, TOK_LIT_INT);
+               state->file->report_line = strtoul(tk->val.str, 0, 10) -1;
+               if (pp_peek(state) == TOK_LIT_STRING) {
+                       const char *token, *base;
+                       char *name, *dir;
+                       int name_len, dir_len;
+                       tk = pp_eat(state, TOK_LIT_STRING);
+                       name = xmalloc(tk->str_len, "report_name");
+                       token = tk->val.str + 1;
+                       base = strrchr(token, '/');
+                       name_len = tk->str_len - 2;
+                       if (base != 0) {
+                               dir_len = base - token;
+                               base++;
+                               name_len -= base - token;
+                       } else {
+                               dir_len = 0;
+                               base = token;
+                       }
+                       memcpy(name, base, name_len);
+                       name[name_len] = '\0';
+                       dir = xmalloc(dir_len + 1, "report_dir");
+                       memcpy(dir, token, dir_len);
+                       dir[dir_len] = '\0';
+                       state->file->report_name = name;
+                       state->file->report_dir = dir;
+                       state->file->macro      = 0;
+               }
+               break;
+       }
+       case TOK_MUNDEF:
+       {
+               struct hash_entry *ident;
+               pp_eat(state, TOK_MUNDEF);
+               if (if_eat(state))  /* quit early when #if'd out */
+                       break;
+
+               ident = pp_eat(state, TOK_MIDENT)->ident;
+
+               undef_macro(state, ident);
+               break;
+       }
+       case TOK_MPRAGMA:
+               pp_eat(state, TOK_MPRAGMA);
+               if (if_eat(state))  /* quit early when #if'd out */
+                       break;
+               warning(state, 0, "Ignoring pragma");
+               break;
+       case TOK_MELIF:
+               pp_eat(state, TOK_MELIF);
+               reenter_if(state, "#elif");
+               if (if_eat(state))   /* quit early when #if'd out */
+                       break;
+               /* If the #if was taken the #elif just disables the following code */
+               if (if_value(state)) {
+                       eat_tokens(state, TOK_MENDIF);
+               }
+               /* If the previous #if was not taken see if the #elif enables the
+                * trailing code.
+                */
+               else {
+                       set_if_value(state, mcexpr(state));
+                       if (!if_value(state)) {
+                               eat_tokens(state, TOK_MELSE);
+                       }
+               }
+               break;
+       case TOK_MIF:
+               pp_eat(state, TOK_MIF);
+               enter_if(state);
+               if (if_eat(state))  /* quit early when #if'd out */
+                       break;
+               set_if_value(state, mcexpr(state));
+               if (!if_value(state)) {
+                       eat_tokens(state, TOK_MELSE);
+               }
+               break;
+       case TOK_MIFNDEF:
+       {
+               struct hash_entry *ident;
+
+               pp_eat(state, TOK_MIFNDEF);
+               enter_if(state);
+               if (if_eat(state))  /* quit early when #if'd out */
+                       break;
+               ident = pp_eat(state, TOK_MIDENT)->ident;
+               set_if_value(state, ident->sym_define == 0);
+               if (!if_value(state)) {
+                       eat_tokens(state, TOK_MELSE);
+               }
+               break;
+       }
+       case TOK_MIFDEF:
+       {
+               struct hash_entry *ident;
+               pp_eat(state, TOK_MIFDEF);
+               enter_if(state);
+               if (if_eat(state))  /* quit early when #if'd out */
+                       break;
+               ident = pp_eat(state, TOK_MIDENT)->ident;
+               set_if_value(state, ident->sym_define != 0);
+               if (!if_value(state)) {
+                       eat_tokens(state, TOK_MELSE);
+               }
+               break;
+       }
+       case TOK_MELSE:
+               pp_eat(state, TOK_MELSE);
+               enter_else(state, "#else");
+               if (!if_eat(state) && if_value(state)) {
+                       eat_tokens(state, TOK_MENDIF);
+               }
+               break;
+       case TOK_MENDIF:
+               pp_eat(state, TOK_MENDIF);
+               exit_if(state, "#endif");
+               break;
+       case TOK_MDEFINE:
+       {
+               struct hash_entry *ident;
+               struct macro_arg *args, **larg;
+               const char *mstart, *mend;
+               int argc;
+
+               pp_eat(state, TOK_MDEFINE);
+               if (if_eat(state))  /* quit early when #if'd out */
+                       break;
+               ident = pp_eat(state, TOK_MIDENT)->ident;
+               argc = -1;
+               args = 0;
+               larg = &args;
+
+               /* Parse macro parameters */
+               if (raw_peek(state) == TOK_LPAREN) {
+                       raw_eat(state, TOK_LPAREN);
+                       argc += 1;
+
+                       for(;;) {
+                               struct macro_arg *narg, *arg;
+                               struct hash_entry *aident;
+                               int tok;
+
+                               tok = pp_peek(state);
+                               if (!args && (tok == TOK_RPAREN)) {
+                                       break;
+                               }
+                               else if (tok == TOK_DOTS) {
+                                       pp_eat(state, TOK_DOTS);
+                                       aident = state->i___VA_ARGS__;
+                               }
+                               else {
+                                       aident = pp_eat(state, TOK_MIDENT)->ident;
+                               }
+
+                               narg = xcmalloc(sizeof(*arg), "macro arg");
+                               narg->ident = aident;
+
+                               /* Verify I don't have a duplicate identifier */
+                               for(arg = args; arg; arg = arg->next) {
+                                       if (arg->ident == narg->ident) {
+                                               error(state, 0, "Duplicate macro arg `%s'",
+                                                       narg->ident->name);
+                                       }
+                               }
+                               /* Add the new argument to the end of the list */
+                               *larg = narg;
+                               larg = &narg->next;
+                               argc += 1;
+
+                               if ((aident == state->i___VA_ARGS__) ||
+                                       (pp_peek(state) != TOK_COMMA)) {
+                                       break;
+                               }
+                               pp_eat(state, TOK_COMMA);
+                       }
+                       pp_eat(state, TOK_RPAREN);
+               }
+               /* Remove leading whitespace */
+               while(raw_peek(state) == TOK_SPACE) {
+                       raw_eat(state, TOK_SPACE);
+               }
+
+               /* Remember the start of the macro body */
+               tok = raw_peek(state);
+               mend = mstart = get_token(state, 1)->pos;
+
+               /* Find the end of the macro */
+               for(tok = raw_peek(state); tok != TOK_EOL; tok = raw_peek(state)) {
+                       raw_eat(state, tok);
+                       /* Remember the end of the last non space token */
+                       raw_peek(state);
+                       if (tok != TOK_SPACE) {
+                               mend = get_token(state, 1)->pos;
+                       }
+               }
+
+               /* Now that I have found the body defined the token */
+               do_define_macro(state, ident,
+                       char_strdup(state->file, mstart, mend, "macro buf"),
+                       argc, args);
+               break;
+       }
+       case TOK_MERROR:
+       {
+               const char *start, *end;
+               int len;
+
+               pp_eat(state, TOK_MERROR);
+               /* Find the start of the line */
+               raw_peek(state);
+               start = get_token(state, 1)->pos;
+
+               /* Find the end of the line */
+               while((tok = raw_peek(state)) != TOK_EOL) {
+                       raw_eat(state, tok);
+               }
+               end = get_token(state, 1)->pos;
+               len = end - start;
+               if (!if_eat(state)) {
+                       error(state, 0, "%*.*s", len, len, start);
+               }
+               break;
+       }
+       case TOK_MWARNING:
+       {
+               const char *start, *end;
+               int len;
+
+               pp_eat(state, TOK_MWARNING);
+
+               /* Find the start of the line */
+               raw_peek(state);
+               start = get_token(state, 1)->pos;
+
+               /* Find the end of the line */
+               while((tok = raw_peek(state)) != TOK_EOL) {
+                       raw_eat(state, tok);
+               }
+               end = get_token(state, 1)->pos;
+               len = end - start;
+               if (!if_eat(state)) {
+                       warning(state, 0, "%*.*s", len, len, start);
+               }
+               break;
+       }
+       case TOK_MINCLUDE:
+       {
+               char *name;
+               int local;
+               local = 0;
+               name = 0;
+
+               pp_eat(state, TOK_MINCLUDE);
+               if (if_eat(state)) {
+                       /* Find the end of the line */
+                       while((tok = raw_peek(state)) != TOK_EOL) {
+                               raw_eat(state, tok);
+                       }
+                       break;
+               }
+               tok = peek(state);
+               if (tok == TOK_LIT_STRING) {
+                       struct token *tk;
+                       const char *token;
+                       int name_len;
+                       tk = eat(state, TOK_LIT_STRING);
+                       name = xmalloc(tk->str_len, "include");
+                       token = tk->val.str +1;
+                       name_len = tk->str_len -2;
+                       if (*token == '"') {
+                               token++;
+                               name_len--;
+                       }
+                       memcpy(name, token, name_len);
+                       name[name_len] = '\0';
+                       local = 1;
+               }
+               else if (tok == TOK_LESS) {
+                       struct macro_buf buf;
+                       eat(state, TOK_LESS);
+
+                       buf.len = 40;
+                       buf.str = xmalloc(buf.len, "include");
+                       buf.pos = 0;
+
+                       tok = peek(state);
+                       while((tok != TOK_MORE) &&
+                               (tok != TOK_EOL) && (tok != TOK_EOF))
+                       {
+                               struct token *tk;
+                               tk = eat(state, tok);
+                               append_macro_chars(state, "include", &buf,
+                                       state->file, tk->pos, state->file->pos);
+                               tok = peek(state);
+                       }
+                       append_macro_text(state, "include", &buf, "\0", 1);
+                       if (peek(state) != TOK_MORE) {
+                               error(state, 0, "Unterminated include directive");
+                       }
+                       eat(state, TOK_MORE);
+                       local = 0;
+                       name = buf.str;
+               }
+               else {
+                       error(state, 0, "Invalid include directive");
+               }
+               /* Error if there are any tokens after the include */
+               if (pp_peek(state) != TOK_EOL) {
+                       error(state, 0, "garbage after include directive");
+               }
+               if (!if_eat(state)) {
+                       compile_file(state, name, local);
+               }
+               xfree(name);
+               break;
+       }
+       case TOK_EOL:
+               /* Ignore # without a follwing ident */
+               break;
+       default:
+       {
+               const char *name1, *name2;
+               name1 = tokens[tok];
+               name2 = "";
+               if (tok == TOK_MIDENT) {
+                       name2 = get_token(state, 1)->ident->name;
+               }
+               error(state, 0, "Invalid preprocessor directive: %s %s",
+                       name1, name2);
+               break;
+       }
+       }
+       /* Consume the rest of the macro line */
+       do {
+               tok = pp_peek(state);
+               pp_eat(state, tok);
+       } while((tok != TOK_EOF) && (tok != TOK_EOL));
+       state->token_base = old_token_base;
+       state->macro_file = NULL;
+       return;
 }
 
 /* Type helper functions */
@@ -3446,6 +5557,7 @@ static struct type *new_type(
        result->right = right;
        result->field_ident = 0;
        result->type_ident = 0;
+       result->elements = 0;
        return result;
 }
 
@@ -3459,20 +5571,85 @@ static struct type *clone_type(unsigned int specifiers, struct type *old)
        return result;
 }
 
-#define SIZEOF_SHORT 2
-#define SIZEOF_INT   4
-#define SIZEOF_LONG  (sizeof(long_t))
+static struct type *dup_type(struct compile_state *state, struct type *orig)
+{
+       struct type *new;
+       new = xcmalloc(sizeof(*new), "type");
+       new->type = orig->type;
+       new->field_ident = orig->field_ident;
+       new->type_ident  = orig->type_ident;
+       new->elements    = orig->elements;
+       if (orig->left) {
+               new->left = dup_type(state, orig->left);
+       }
+       if (orig->right) {
+               new->right = dup_type(state, orig->right);
+       }
+       return new;
+}
+
+
+static struct type *invalid_type(struct compile_state *state, struct type *type)
+{
+       struct type *invalid, *member;
+       invalid = 0;
+       if (!type) {
+               internal_error(state, 0, "type missing?");
+       }
+       switch(type->type & TYPE_MASK) {
+       case TYPE_VOID:
+       case TYPE_CHAR:         case TYPE_UCHAR:
+       case TYPE_SHORT:        case TYPE_USHORT:
+       case TYPE_INT:          case TYPE_UINT:
+       case TYPE_LONG:         case TYPE_ULONG:
+       case TYPE_LLONG:        case TYPE_ULLONG:
+       case TYPE_POINTER:
+       case TYPE_ENUM:
+               break;
+       case TYPE_BITFIELD:
+               invalid = invalid_type(state, type->left);
+               break;
+       case TYPE_ARRAY:
+               invalid = invalid_type(state, type->left);
+               break;
+       case TYPE_STRUCT:
+       case TYPE_TUPLE:
+               member = type->left;
+               while(member && (invalid == 0) &&
+                       ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       invalid = invalid_type(state, member->left);
+                       member = member->right;
+               }
+               if (!invalid) {
+                       invalid = invalid_type(state, member);
+               }
+               break;
+       case TYPE_UNION:
+       case TYPE_JOIN:
+               member = type->left;
+               while(member && (invalid == 0) &&
+                       ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       invalid = invalid_type(state, member->left);
+                       member = member->right;
+               }
+               if (!invalid) {
+                       invalid = invalid_type(state, member);
+               }
+               break;
+       default:
+               invalid = type;
+               break;
+       }
+       return invalid;
 
-#define ALIGNOF_SHORT 2
-#define ALIGNOF_INT   4
-#define ALIGNOF_LONG  (sizeof(long_t))
+}
 
 #define MASK_UCHAR(X)    ((X) & ((ulong_t)0xff))
-#define MASK_USHORT(X)   ((X) & (((ulong_t)1 << (SIZEOF_SHORT*8)) - 1))
+#define MASK_USHORT(X)   ((X) & (((ulong_t)1 << (SIZEOF_SHORT)) - 1))
 static inline ulong_t mask_uint(ulong_t x)
 {
        if (SIZEOF_INT < SIZEOF_LONG) {
-               ulong_t mask = (((ulong_t)1) << ((ulong_t)(SIZEOF_INT*8))) -1;
+               ulong_t mask = (1ULL << ((ulong_t)(SIZEOF_INT))) -1;
                x &= mask;
        }
        return x;
@@ -3480,38 +5657,43 @@ static inline ulong_t mask_uint(ulong_t x)
 #define MASK_UINT(X)      (mask_uint(X))
 #define MASK_ULONG(X)    (X)
 
-static struct type void_type   = { .type  = TYPE_VOID };
-static struct type char_type   = { .type  = TYPE_CHAR };
-static struct type uchar_type  = { .type  = TYPE_UCHAR };
-static struct type short_type  = { .type  = TYPE_SHORT };
-static struct type ushort_type = { .type  = TYPE_USHORT };
-static struct type int_type    = { .type  = TYPE_INT };
-static struct type uint_type   = { .type  = TYPE_UINT };
-static struct type long_type   = { .type  = TYPE_LONG };
-static struct type ulong_type  = { .type  = TYPE_ULONG };
+static struct type void_type    = { .type  = TYPE_VOID };
+static struct type char_type    = { .type  = TYPE_CHAR };
+static struct type uchar_type   = { .type  = TYPE_UCHAR };
+#if DEBUG_ROMCC_WARNING
+static struct type short_type   = { .type  = TYPE_SHORT };
+#endif
+static struct type ushort_type  = { .type  = TYPE_USHORT };
+static struct type int_type     = { .type  = TYPE_INT };
+static struct type uint_type    = { .type  = TYPE_UINT };
+static struct type long_type    = { .type  = TYPE_LONG };
+static struct type ulong_type   = { .type  = TYPE_ULONG };
+static struct type unknown_type = { .type  = TYPE_UNKNOWN };
+
+static struct type void_ptr_type  = {
+       .type = TYPE_POINTER,
+       .left = &void_type,
+};
+
+#if DEBUG_ROMCC_WARNING
+static struct type void_func_type = {
+       .type  = TYPE_FUNCTION,
+       .left  = &void_type,
+       .right = &void_type,
+};
+#endif
+
+static size_t bits_to_bytes(size_t size)
+{
+       return (size + SIZEOF_CHAR - 1)/SIZEOF_CHAR;
+}
 
 static struct triple *variable(struct compile_state *state, struct type *type)
 {
        struct triple *result;
        if ((type->type & STOR_MASK) != STOR_PERM) {
-               if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
-                       result = triple(state, OP_ADECL, type, 0, 0);
-               } else {
-                       struct type *field;
-                       struct triple **vector;
-                       ulong_t index;
-                       result = new_triple(state, OP_VAL_VEC, type, -1, -1);
-                       vector = &result->param[0];
-
-                       field = type->left;
-                       index = 0;
-                       while((field->type & TYPE_MASK) == TYPE_PRODUCT) {
-                               vector[index] = variable(state, field->left);
-                               field = field->right;
-                               index++;
-                       }
-                       vector[index] = variable(state, field);
-               }
+               result = triple(state, OP_ADECL, type, 0, 0);
+               generate_lhs_pieces(state, result);
        }
        else {
                result = triple(state, OP_SDECL, type, 0, 0);
@@ -3528,6 +5710,9 @@ static void stor_of(FILE *fp, struct type *type)
        case STOR_STATIC:
                fprintf(fp, "static ");
                break;
+       case STOR_LOCAL:
+               fprintf(fp, "local ");
+               break;
        case STOR_EXTERN:
                fprintf(fp, "extern ");
                break;
@@ -3537,9 +5722,18 @@ static void stor_of(FILE *fp, struct type *type)
        case STOR_TYPEDEF:
                fprintf(fp, "typedef ");
                break;
-       case STOR_INLINE:
+       case STOR_INLINE | STOR_LOCAL:
                fprintf(fp, "inline ");
                break;
+       case STOR_INLINE | STOR_STATIC:
+               fprintf(fp, "static inline");
+               break;
+       case STOR_INLINE | STOR_EXTERN:
+               fprintf(fp, "extern inline");
+               break;
+       default:
+               fprintf(fp, "stor:%x", type->type & STOR_MASK);
+               break;
        }
 }
 static void qual_of(FILE *fp, struct type *type)
@@ -3557,8 +5751,12 @@ static void qual_of(FILE *fp, struct type *type)
 
 static void name_of(FILE *fp, struct type *type)
 {
-       stor_of(fp, type);
-       switch(type->type & TYPE_MASK) {
+       unsigned int base_type;
+       base_type = type->type & TYPE_MASK;
+       if ((base_type != TYPE_PRODUCT) && (base_type != TYPE_OVERLAP)) {
+               stor_of(fp, type);
+       }
+       switch(base_type) {
        case TYPE_VOID:
                fprintf(fp, "void");
                qual_of(fp, type);
@@ -3601,34 +5799,70 @@ static void name_of(FILE *fp, struct type *type)
                qual_of(fp, type);
                break;
        case TYPE_PRODUCT:
-       case TYPE_OVERLAP:
                name_of(fp, type->left);
                fprintf(fp, ", ");
                name_of(fp, type->right);
                break;
+       case TYPE_OVERLAP:
+               name_of(fp, type->left);
+               fprintf(fp, ",| ");
+               name_of(fp, type->right);
+               break;
        case TYPE_ENUM:
-               fprintf(fp, "enum %s", type->type_ident->name);
+               fprintf(fp, "enum %s",
+                       (type->type_ident)? type->type_ident->name : "");
                qual_of(fp, type);
                break;
        case TYPE_STRUCT:
-               fprintf(fp, "struct %s", type->type_ident->name);
+               fprintf(fp, "struct %s { ",
+                       (type->type_ident)? type->type_ident->name : "");
+               name_of(fp, type->left);
+               fprintf(fp, " } ");
+               qual_of(fp, type);
+               break;
+       case TYPE_UNION:
+               fprintf(fp, "union %s { ",
+                       (type->type_ident)? type->type_ident->name : "");
+               name_of(fp, type->left);
+               fprintf(fp, " } ");
                qual_of(fp, type);
                break;
        case TYPE_FUNCTION:
-       {
                name_of(fp, type->left);
                fprintf(fp, " (*)(");
                name_of(fp, type->right);
                fprintf(fp, ")");
                break;
-       }
        case TYPE_ARRAY:
                name_of(fp, type->left);
-               fprintf(fp, " [%ld]", type->elements);
+               fprintf(fp, " [%ld]", (long)(type->elements));
                break;
-       default:
-               fprintf(fp, "????: %x", type->type & TYPE_MASK);
+       case TYPE_TUPLE:
+               fprintf(fp, "tuple { ");
+               name_of(fp, type->left);
+               fprintf(fp, " } ");
+               qual_of(fp, type);
                break;
+       case TYPE_JOIN:
+               fprintf(fp, "join { ");
+               name_of(fp, type->left);
+               fprintf(fp, " } ");
+               qual_of(fp, type);
+               break;
+       case TYPE_BITFIELD:
+               name_of(fp, type->left);
+               fprintf(fp, " : %d ", type->elements);
+               qual_of(fp, type);
+               break;
+       case TYPE_UNKNOWN:
+               fprintf(fp, "unknown_t");
+               break;
+       default:
+               fprintf(fp, "????: %x", base_type);
+               break;
+       }
+       if (type->field_ident && type->field_ident->name) {
+               fprintf(fp, " .%s", type->field_ident->name);
        }
 }
 
@@ -3640,9 +5874,12 @@ static size_t align_of(struct compile_state *state, struct type *type)
        case TYPE_VOID:
                align = 1;
                break;
+       case TYPE_BITFIELD:
+               align = 1;
+               break;
        case TYPE_CHAR:
        case TYPE_UCHAR:
-               align = 1;
+               align = ALIGNOF_CHAR;
                break;
        case TYPE_SHORT:
        case TYPE_USHORT:
@@ -3655,9 +5892,11 @@ static size_t align_of(struct compile_state *state, struct type *type)
                break;
        case TYPE_LONG:
        case TYPE_ULONG:
-       case TYPE_POINTER:
                align = ALIGNOF_LONG;
                break;
+       case TYPE_POINTER:
+               align = ALIGNOF_POINTER;
+               break;
        case TYPE_PRODUCT:
        case TYPE_OVERLAP:
        {
@@ -3671,6 +5910,9 @@ static size_t align_of(struct compile_state *state, struct type *type)
                align = align_of(state, type->left);
                break;
        case TYPE_STRUCT:
+       case TYPE_TUPLE:
+       case TYPE_UNION:
+       case TYPE_JOIN:
                align = align_of(state, type->left);
                break;
        default:
@@ -3680,6 +5922,111 @@ static size_t align_of(struct compile_state *state, struct type *type)
        return align;
 }
 
+static size_t reg_align_of(struct compile_state *state, struct type *type)
+{
+       size_t align;
+       align = 0;
+       switch(type->type & TYPE_MASK) {
+       case TYPE_VOID:
+               align = 1;
+               break;
+       case TYPE_BITFIELD:
+               align = 1;
+               break;
+       case TYPE_CHAR:
+       case TYPE_UCHAR:
+               align = REG_ALIGNOF_CHAR;
+               break;
+       case TYPE_SHORT:
+       case TYPE_USHORT:
+               align = REG_ALIGNOF_SHORT;
+               break;
+       case TYPE_INT:
+       case TYPE_UINT:
+       case TYPE_ENUM:
+               align = REG_ALIGNOF_INT;
+               break;
+       case TYPE_LONG:
+       case TYPE_ULONG:
+               align = REG_ALIGNOF_LONG;
+               break;
+       case TYPE_POINTER:
+               align = REG_ALIGNOF_POINTER;
+               break;
+       case TYPE_PRODUCT:
+       case TYPE_OVERLAP:
+       {
+               size_t left_align, right_align;
+               left_align  = reg_align_of(state, type->left);
+               right_align = reg_align_of(state, type->right);
+               align = (left_align >= right_align) ? left_align : right_align;
+               break;
+       }
+       case TYPE_ARRAY:
+               align = reg_align_of(state, type->left);
+               break;
+       case TYPE_STRUCT:
+       case TYPE_UNION:
+       case TYPE_TUPLE:
+       case TYPE_JOIN:
+               align = reg_align_of(state, type->left);
+               break;
+       default:
+               error(state, 0, "alignof not yet defined for type\n");
+               break;
+       }
+       return align;
+}
+
+static size_t align_of_in_bytes(struct compile_state *state, struct type *type)
+{
+       return bits_to_bytes(align_of(state, type));
+}
+static size_t size_of(struct compile_state *state, struct type *type);
+static size_t reg_size_of(struct compile_state *state, struct type *type);
+
+static size_t needed_padding(struct compile_state *state,
+       struct type *type, size_t offset)
+{
+        size_t padding, align;
+       align = align_of(state, type);
+       /* Align to the next machine word if the bitfield does completely
+        * fit into the current word.
+        */
+       if ((type->type & TYPE_MASK) == TYPE_BITFIELD) {
+               size_t size;
+               size = size_of(state, type);
+               if ((offset + type->elements)/size != offset/size) {
+                       align = size;
+               }
+       }
+       padding = 0;
+       if (offset % align) {
+               padding = align - (offset % align);
+       }
+       return padding;
+}
+
+static size_t reg_needed_padding(struct compile_state *state,
+       struct type *type, size_t offset)
+{
+        size_t padding, align;
+       align = reg_align_of(state, type);
+       /* Align to the next register word if the bitfield does completely
+        * fit into the current register.
+        */
+       if (((type->type & TYPE_MASK) == TYPE_BITFIELD) &&
+               (((offset + type->elements)/REG_SIZEOF_REG) != (offset/REG_SIZEOF_REG)))
+       {
+               align = REG_SIZEOF_REG;
+       }
+       padding = 0;
+       if (offset % align) {
+               padding = align - (offset % align);
+       }
+       return padding;
+}
+
 static size_t size_of(struct compile_state *state, struct type *type)
 {
        size_t size;
@@ -3688,9 +6035,12 @@ static size_t size_of(struct compile_state *state, struct type *type)
        case TYPE_VOID:
                size = 0;
                break;
+       case TYPE_BITFIELD:
+               size = type->elements;
+               break;
        case TYPE_CHAR:
        case TYPE_UCHAR:
-               size = 1;
+               size = SIZEOF_CHAR;
                break;
        case TYPE_SHORT:
        case TYPE_USHORT:
@@ -3703,22 +6053,22 @@ static size_t size_of(struct compile_state *state, struct type *type)
                break;
        case TYPE_LONG:
        case TYPE_ULONG:
-       case TYPE_POINTER:
                size = SIZEOF_LONG;
                break;
+       case TYPE_POINTER:
+               size = SIZEOF_POINTER;
+               break;
        case TYPE_PRODUCT:
        {
-               size_t align, pad;
-               size = size_of(state, type->left);
-               while((type->right->type & TYPE_MASK) == TYPE_PRODUCT) {
-                       type = type->right;
-                       align = align_of(state, type->left);
-                       pad = align - (size % align);
+               size_t pad;
+               size = 0;
+               while((type->type & TYPE_MASK) == TYPE_PRODUCT) {
+                       pad = needed_padding(state, type->left, size);
                        size = size + pad + size_of(state, type->left);
+                       type = type->right;
                }
-               align = align_of(state, type->right);
-               pad = align - (size % align);
-               size = size + pad + sizeof(type->right);
+               pad = needed_padding(state, type, size);
+               size = size + pad + size_of(state, type);
                break;
        }
        case TYPE_OVERLAP:
@@ -3737,317 +6087,1051 @@ static size_t size_of(struct compile_state *state, struct type *type)
                }
                break;
        case TYPE_STRUCT:
+       case TYPE_TUPLE:
+       {
+               size_t pad;
+               size = size_of(state, type->left);
+               /* Pad structures so their size is a multiples of their alignment */
+               pad = needed_padding(state, type, size);
+               size = size + pad;
+               break;
+       }
+       case TYPE_UNION:
+       case TYPE_JOIN:
+       {
+               size_t pad;
                size = size_of(state, type->left);
+               /* Pad unions so their size is a multiple of their alignment */
+               pad = needed_padding(state, type, size);
+               size = size + pad;
                break;
+       }
        default:
-               error(state, 0, "sizeof not yet defined for type\n");
+               internal_error(state, 0, "sizeof not yet defined for type");
                break;
        }
        return size;
 }
 
-static size_t field_offset(struct compile_state *state, 
-       struct type *type, struct hash_entry *field)
+static size_t reg_size_of(struct compile_state *state, struct type *type)
 {
-       size_t size, align, pad;
-       if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
-               internal_error(state, 0, "field_offset only works on structures");
-       }
+       size_t size;
        size = 0;
-       type = type->left;
-       while((type->type & TYPE_MASK) == TYPE_PRODUCT) {
-               if (type->left->field_ident == field) {
-                       type = type->left;
+       switch(type->type & TYPE_MASK) {
+       case TYPE_VOID:
+               size = 0;
+               break;
+       case TYPE_BITFIELD:
+               size = type->elements;
+               break;
+       case TYPE_CHAR:
+       case TYPE_UCHAR:
+               size = REG_SIZEOF_CHAR;
+               break;
+       case TYPE_SHORT:
+       case TYPE_USHORT:
+               size = REG_SIZEOF_SHORT;
+               break;
+       case TYPE_INT:
+       case TYPE_UINT:
+       case TYPE_ENUM:
+               size = REG_SIZEOF_INT;
+               break;
+       case TYPE_LONG:
+       case TYPE_ULONG:
+               size = REG_SIZEOF_LONG;
+               break;
+       case TYPE_POINTER:
+               size = REG_SIZEOF_POINTER;
+               break;
+       case TYPE_PRODUCT:
+       {
+               size_t pad;
+               size = 0;
+               while((type->type & TYPE_MASK) == TYPE_PRODUCT) {
+                       pad = reg_needed_padding(state, type->left, size);
+                       size = size + pad + reg_size_of(state, type->left);
+                       type = type->right;
                }
-               size += size_of(state, type->left);
-               type = type->right;
-               align = align_of(state, type->left);
-               pad = align - (size % align);
-               size += pad;
+               pad = reg_needed_padding(state, type, size);
+               size = size + pad + reg_size_of(state, type);
+               break;
        }
-       if (type->field_ident != field) {
-               internal_error(state, 0, "field_offset: member %s not present",
-                       field->name);
+       case TYPE_OVERLAP:
+       {
+               size_t size_left, size_right;
+               size_left  = reg_size_of(state, type->left);
+               size_right = reg_size_of(state, type->right);
+               size = (size_left >= size_right)? size_left : size_right;
+               break;
+       }
+       case TYPE_ARRAY:
+               if (type->elements == ELEMENT_COUNT_UNSPECIFIED) {
+                       internal_error(state, 0, "Invalid array type");
+               } else {
+                       size = reg_size_of(state, type->left) * type->elements;
+               }
+               break;
+       case TYPE_STRUCT:
+       case TYPE_TUPLE:
+       {
+               size_t pad;
+               size = reg_size_of(state, type->left);
+               /* Pad structures so their size is a multiples of their alignment */
+               pad = reg_needed_padding(state, type, size);
+               size = size + pad;
+               break;
+       }
+       case TYPE_UNION:
+       case TYPE_JOIN:
+       {
+               size_t pad;
+               size = reg_size_of(state, type->left);
+               /* Pad unions so their size is a multiple of their alignment */
+               pad = reg_needed_padding(state, type, size);
+               size = size + pad;
+               break;
+       }
+       default:
+               internal_error(state, 0, "sizeof not yet defined for type");
+               break;
        }
        return size;
 }
 
-static struct type *field_type(struct compile_state *state, 
+static size_t registers_of(struct compile_state *state, struct type *type)
+{
+       size_t registers;
+       registers = reg_size_of(state, type);
+       registers += REG_SIZEOF_REG - 1;
+       registers /= REG_SIZEOF_REG;
+       return registers;
+}
+
+static size_t size_of_in_bytes(struct compile_state *state, struct type *type)
+{
+       return bits_to_bytes(size_of(state, type));
+}
+
+static size_t field_offset(struct compile_state *state,
        struct type *type, struct hash_entry *field)
 {
-       if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
-               internal_error(state, 0, "field_type only works on structures");
+       struct type *member;
+       size_t size;
+
+       size = 0;
+       member = 0;
+       if ((type->type & TYPE_MASK) == TYPE_STRUCT) {
+               member = type->left;
+               while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       size += needed_padding(state, member->left, size);
+                       if (member->left->field_ident == field) {
+                               member = member->left;
+                               break;
+                       }
+                       size += size_of(state, member->left);
+                       member = member->right;
+               }
+               size += needed_padding(state, member, size);
        }
-       type = type->left;
-       while((type->type & TYPE_MASK) == TYPE_PRODUCT) {
-               if (type->left->field_ident == field) {
-                       type = type->left;
-                       break;
+       else if ((type->type & TYPE_MASK) == TYPE_UNION) {
+               member = type->left;
+               while(member && ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       if (member->left->field_ident == field) {
+                               member = member->left;
+                               break;
+                       }
+                       member = member->right;
                }
-               type = type->right;
        }
-       if (type->field_ident != field) {
-               internal_error(state, 0, "field_type: member %s not present", 
-                       field->name);
+       else {
+               internal_error(state, 0, "field_offset only works on structures and unions");
        }
-       return type;
+
+       if (!member || (member->field_ident != field)) {
+               error(state, 0, "member %s not present", field->name);
+       }
+       return size;
 }
 
-static struct triple *struct_field(struct compile_state *state,
-       struct triple *decl, struct hash_entry *field)
+static size_t field_reg_offset(struct compile_state *state,
+       struct type *type, struct hash_entry *field)
 {
-       struct triple **vector;
-       struct type *type;
-       ulong_t index;
-       type = decl->type;
-       if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
-               return decl;
-       }
-       if (decl->op != OP_VAL_VEC) {
-               internal_error(state, 0, "Invalid struct variable");
-       }
-       if (!field) {
-               internal_error(state, 0, "Missing structure field");
+       struct type *member;
+       size_t size;
+
+       size = 0;
+       member = 0;
+       if ((type->type & TYPE_MASK) == TYPE_STRUCT) {
+               member = type->left;
+               while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       size += reg_needed_padding(state, member->left, size);
+                       if (member->left->field_ident == field) {
+                               member = member->left;
+                               break;
+                       }
+                       size += reg_size_of(state, member->left);
+                       member = member->right;
+               }
        }
-       type = type->left;
-       vector = &RHS(decl, 0);
-       index = 0;
-       while((type->type & TYPE_MASK) == TYPE_PRODUCT) {
-               if (type->left->field_ident == field) {
-                       type = type->left;
-                       break;
+       else if ((type->type & TYPE_MASK) == TYPE_UNION) {
+               member = type->left;
+               while(member && ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       if (member->left->field_ident == field) {
+                               member = member->left;
+                               break;
+                       }
+                       member = member->right;
                }
-               index += 1;
-               type = type->right;
        }
-       if (type->field_ident != field) {
-               internal_error(state, 0, "field %s not found?", field->name);
+       else {
+               internal_error(state, 0, "field_reg_offset only works on structures and unions");
        }
-       return vector[index];
-}
 
-static void arrays_complete(struct compile_state *state, struct type *type)
-{
-       if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
-               if (type->elements == ELEMENT_COUNT_UNSPECIFIED) {
-                       error(state, 0, "array size not specified");
-               }
-               arrays_complete(state, type->left);
+       size += reg_needed_padding(state, member, size);
+       if (!member || (member->field_ident != field)) {
+               error(state, 0, "member %s not present", field->name);
        }
+       return size;
 }
 
-static unsigned int do_integral_promotion(unsigned int type)
+static struct type *field_type(struct compile_state *state,
+       struct type *type, struct hash_entry *field)
 {
-       type &= TYPE_MASK;
-       if (TYPE_INTEGER(type) && 
-               TYPE_RANK(type) < TYPE_RANK(TYPE_INT)) {
-               type = TYPE_INT;
-       }
-       return type;
-}
+       struct type *member;
 
-static unsigned int do_arithmetic_conversion(
-       unsigned int left, unsigned int right)
-{
-       left &= TYPE_MASK;
-       right &= TYPE_MASK;
-       if ((left == TYPE_LDOUBLE) || (right == TYPE_LDOUBLE)) {
-               return TYPE_LDOUBLE;
-       }
-       else if ((left == TYPE_DOUBLE) || (right == TYPE_DOUBLE)) {
-               return TYPE_DOUBLE;
+       member = 0;
+       if ((type->type & TYPE_MASK) == TYPE_STRUCT) {
+               member = type->left;
+               while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       if (member->left->field_ident == field) {
+                               member = member->left;
+                               break;
+                       }
+                       member = member->right;
+               }
        }
-       else if ((left == TYPE_FLOAT) || (right == TYPE_FLOAT)) {
-               return TYPE_FLOAT;
+       else if ((type->type & TYPE_MASK) == TYPE_UNION) {
+               member = type->left;
+               while(member && ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       if (member->left->field_ident == field) {
+                               member = member->left;
+                               break;
+                       }
+                       member = member->right;
+               }
        }
-       left = do_integral_promotion(left);
-       right = do_integral_promotion(right);
-       /* If both operands have the same size done */
-       if (left == right) {
-               return left;
+       else {
+               internal_error(state, 0, "field_type only works on structures and unions");
        }
-       /* If both operands have the same signedness pick the larger */
-       else if (!!TYPE_UNSIGNED(left) == !!TYPE_UNSIGNED(right)) {
-               return (TYPE_RANK(left) >= TYPE_RANK(right)) ? left : right;
+
+       if (!member || (member->field_ident != field)) {
+               error(state, 0, "member %s not present", field->name);
        }
-       /* If the signed type can hold everything use it */
-       else if (TYPE_SIGNED(left) && (TYPE_RANK(left) > TYPE_RANK(right))) {
-               return left;
+       return member;
+}
+
+static size_t index_offset(struct compile_state *state,
+       struct type *type, ulong_t index)
+{
+       struct type *member;
+       size_t size;
+       size = 0;
+       if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
+               size = size_of(state, type->left) * index;
        }
-       else if (TYPE_SIGNED(right) && (TYPE_RANK(right) > TYPE_RANK(left))) {
-               return right;
+       else if ((type->type & TYPE_MASK) == TYPE_TUPLE) {
+               ulong_t i;
+               member = type->left;
+               i = 0;
+               while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       size += needed_padding(state, member->left, size);
+                       if (i == index) {
+                               member = member->left;
+                               break;
+                       }
+                       size += size_of(state, member->left);
+                       i++;
+                       member = member->right;
+               }
+               size += needed_padding(state, member, size);
+               if (i != index) {
+                       internal_error(state, 0, "Missing member index: %u", index);
+               }
        }
-       /* Convert to the unsigned type with the same rank as the signed type */
-       else if (TYPE_SIGNED(left)) {
-               return TYPE_MKUNSIGNED(left);
+       else if ((type->type & TYPE_MASK) == TYPE_JOIN) {
+               ulong_t i;
+               size = 0;
+               member = type->left;
+               i = 0;
+               while(member && ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       if (i == index) {
+                               member = member->left;
+                               break;
+                       }
+                       i++;
+                       member = member->right;
+               }
+               if (i != index) {
+                       internal_error(state, 0, "Missing member index: %u", index);
+               }
        }
        else {
-               return TYPE_MKUNSIGNED(right);
+               internal_error(state, 0,
+                       "request for index %u in something not an array, tuple or join",
+                       index);
        }
+       return size;
 }
 
-/* see if two types are the same except for qualifiers */
-static int equiv_types(struct type *left, struct type *right)
+static size_t index_reg_offset(struct compile_state *state,
+       struct type *type, ulong_t index)
 {
-       unsigned int type;
-       /* Error if the basic types do not match */
-       if ((left->type & TYPE_MASK) != (right->type & TYPE_MASK)) {
-               return 0;
+       struct type *member;
+       size_t size;
+       size = 0;
+       if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
+               size = reg_size_of(state, type->left) * index;
        }
-       type = left->type & TYPE_MASK;
-       /* if the basic types match and it is an arithmetic type we are done */
-       if (TYPE_ARITHMETIC(type)) {
-               return 1;
+       else if ((type->type & TYPE_MASK) == TYPE_TUPLE) {
+               ulong_t i;
+               member = type->left;
+               i = 0;
+               while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       size += reg_needed_padding(state, member->left, size);
+                       if (i == index) {
+                               member = member->left;
+                               break;
+                       }
+                       size += reg_size_of(state, member->left);
+                       i++;
+                       member = member->right;
+               }
+               size += reg_needed_padding(state, member, size);
+               if (i != index) {
+                       internal_error(state, 0, "Missing member index: %u", index);
+               }
+
        }
-       /* If it is a pointer type recurse and keep testing */
-       if (type == TYPE_POINTER) {
-               return equiv_types(left->left, right->left);
+       else if ((type->type & TYPE_MASK) == TYPE_JOIN) {
+               ulong_t i;
+               size = 0;
+               member = type->left;
+               i = 0;
+               while(member && ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       if (i == index) {
+                               member = member->left;
+                               break;
+                       }
+                       i++;
+                       member = member->right;
+               }
+               if (i != index) {
+                       internal_error(state, 0, "Missing member index: %u", index);
+               }
        }
-       else if (type == TYPE_ARRAY) {
-               return (left->elements == right->elements) &&
-                       equiv_types(left->left, right->left);
+       else {
+               internal_error(state, 0,
+                       "request for index %u in something not an array, tuple or join",
+                       index);
        }
-       /* test for struct/union equality */
-       else if (type == TYPE_STRUCT) {
-               return left->type_ident == right->type_ident;
+       return size;
+}
+
+static struct type *index_type(struct compile_state *state,
+       struct type *type, ulong_t index)
+{
+       struct type *member;
+       if (index >= type->elements) {
+               internal_error(state, 0, "Invalid element %u requested", index);
        }
-       /* Test for equivalent functions */
-       else if (type == TYPE_FUNCTION) {
-               return equiv_types(left->left, right->left) &&
-                       equiv_types(left->right, right->right);
+       if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
+               member = type->left;
        }
-       /* We only see TYPE_PRODUCT as part of function equivalence matching */
-       else if (type == TYPE_PRODUCT) {
-               return equiv_types(left->left, right->left) &&
-                       equiv_types(left->right, right->right);
+       else if ((type->type & TYPE_MASK) == TYPE_TUPLE) {
+               ulong_t i;
+               member = type->left;
+               i = 0;
+               while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       if (i == index) {
+                               member = member->left;
+                               break;
+                       }
+                       i++;
+                       member = member->right;
+               }
+               if (i != index) {
+                       internal_error(state, 0, "Missing member index: %u", index);
+               }
+       }
+       else if ((type->type & TYPE_MASK) == TYPE_JOIN) {
+               ulong_t i;
+               member = type->left;
+               i = 0;
+               while(member && ((member->type & TYPE_MASK) == TYPE_OVERLAP)) {
+                       if (i == index) {
+                               member = member->left;
+                               break;
+                       }
+                       i++;
+                       member = member->right;
+               }
+               if (i != index) {
+                       internal_error(state, 0, "Missing member index: %u", index);
+               }
        }
-       /* We should see TYPE_OVERLAP */
        else {
-               return 0;
+               member = 0;
+               internal_error(state, 0,
+                       "request for index %u in something not an array, tuple or join",
+                       index);
        }
+       return member;
 }
 
-static int equiv_ptrs(struct type *left, struct type *right)
+static struct type *unpack_type(struct compile_state *state, struct type *type)
 {
-       if (((left->type & TYPE_MASK) != TYPE_POINTER) ||
-               ((right->type & TYPE_MASK) != TYPE_POINTER)) {
-               return 0;
+       /* If I have a single register compound type not a bit-field
+        * find the real type.
+        */
+       struct type *start_type;
+       size_t size;
+       /* Get out early if I need multiple registers for this type */
+       size = reg_size_of(state, type);
+       if (size > REG_SIZEOF_REG) {
+               return type;
        }
-       return equiv_types(left->left, right->left);
+       /* Get out early if I don't need any registers for this type */
+       if (size == 0) {
+               return &void_type;
+       }
+       /* Loop until I have no more layers I can remove */
+       do {
+               start_type = type;
+               switch(type->type & TYPE_MASK) {
+               case TYPE_ARRAY:
+                       /* If I have a single element the unpacked type
+                        * is that element.
+                        */
+                       if (type->elements == 1) {
+                               type = type->left;
+                       }
+                       break;
+               case TYPE_STRUCT:
+               case TYPE_TUPLE:
+                       /* If I have a single element the unpacked type
+                        * is that element.
+                        */
+                       if (type->elements == 1) {
+                               type = type->left;
+                       }
+                       /* If I have multiple elements the unpacked
+                        * type is the non-void element.
+                        */
+                       else {
+                               struct type *next, *member;
+                               struct type *sub_type;
+                               sub_type = 0;
+                               next = type->left;
+                               while(next) {
+                                       member = next;
+                                       next = 0;
+                                       if ((member->type & TYPE_MASK) == TYPE_PRODUCT) {
+                                               next = member->right;
+                                               member = member->left;
+                                       }
+                                       if (reg_size_of(state, member) > 0) {
+                                               if (sub_type) {
+                                                       internal_error(state, 0, "true compound type in a register");
+                                               }
+                                               sub_type = member;
+                                       }
+                               }
+                               if (sub_type) {
+                                       type = sub_type;
+                               }
+                       }
+                       break;
+
+               case TYPE_UNION:
+               case TYPE_JOIN:
+                       /* If I have a single element the unpacked type
+                        * is that element.
+                        */
+                       if (type->elements == 1) {
+                               type = type->left;
+                       }
+                       /* I can't in general unpack union types */
+                       break;
+               default:
+                       /* If I'm not a compound type I can't unpack it */
+                       break;
+               }
+       } while(start_type != type);
+       switch(type->type & TYPE_MASK) {
+       case TYPE_STRUCT:
+       case TYPE_ARRAY:
+       case TYPE_TUPLE:
+               internal_error(state, 0, "irredicible type?");
+               break;
+       }
+       return type;
 }
 
-static struct type *compatible_types(struct type *left, struct type *right)
+static int equiv_types(struct type *left, struct type *right);
+static int is_compound_type(struct type *type);
+
+static struct type *reg_type(
+       struct compile_state *state, struct type *type, int reg_offset)
 {
-       struct type *result;
-       unsigned int type, qual_type;
-       /* Error if the basic types do not match */
-       if ((left->type & TYPE_MASK) != (right->type & TYPE_MASK)) {
-               return 0;
+       struct type *member;
+       size_t size;
+#if 1
+       struct type *invalid;
+       invalid = invalid_type(state, type);
+       if (invalid) {
+               fprintf(state->errout, "type: ");
+               name_of(state->errout, type);
+               fprintf(state->errout, "\n");
+               fprintf(state->errout, "invalid: ");
+               name_of(state->errout, invalid);
+               fprintf(state->errout, "\n");
+               internal_error(state, 0, "bad input type?");
        }
-       type = left->type & TYPE_MASK;
-       qual_type = (left->type & ~STOR_MASK) | (right->type & ~STOR_MASK);
-       result = 0;
-       /* if the basic types match and it is an arithmetic type we are done */
-       if (TYPE_ARITHMETIC(type)) {
-               result = new_type(qual_type, 0, 0);
+#endif
+
+       size = reg_size_of(state, type);
+       if (reg_offset > size) {
+               member = 0;
+               fprintf(state->errout, "type: ");
+               name_of(state->errout, type);
+               fprintf(state->errout, "\n");
+               internal_error(state, 0, "offset outside of type");
        }
-       /* If it is a pointer type recurse and keep testing */
-       else if (type == TYPE_POINTER) {
-               result = compatible_types(left->left, right->left);
-               if (result) {
-                       result = new_type(qual_type, result, 0);
+       else {
+               switch(type->type & TYPE_MASK) {
+                       /* Don't do anything with the basic types */
+               case TYPE_VOID:
+               case TYPE_CHAR:         case TYPE_UCHAR:
+               case TYPE_SHORT:        case TYPE_USHORT:
+               case TYPE_INT:          case TYPE_UINT:
+               case TYPE_LONG:         case TYPE_ULONG:
+               case TYPE_LLONG:        case TYPE_ULLONG:
+               case TYPE_FLOAT:        case TYPE_DOUBLE:
+               case TYPE_LDOUBLE:
+               case TYPE_POINTER:
+               case TYPE_ENUM:
+               case TYPE_BITFIELD:
+                       member = type;
+                       break;
+               case TYPE_ARRAY:
+                       member = type->left;
+                       size = reg_size_of(state, member);
+                       if (size > REG_SIZEOF_REG) {
+                               member = reg_type(state, member, reg_offset % size);
+                       }
+                       break;
+               case TYPE_STRUCT:
+               case TYPE_TUPLE:
+               {
+                       size_t offset;
+                       offset = 0;
+                       member = type->left;
+                       while(member && ((member->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                               size = reg_size_of(state, member->left);
+                               offset += reg_needed_padding(state, member->left, offset);
+                               if ((offset + size) > reg_offset) {
+                                       member = member->left;
+                                       break;
+                               }
+                               offset += size;
+                               member = member->right;
+                       }
+                       offset += reg_needed_padding(state, member, offset);
+                       member = reg_type(state, member, reg_offset - offset);
+                       break;
                }
-       }
-       /* test for struct/union equality */
-       else if (type == TYPE_STRUCT) {
-               if (left->type_ident == right->type_ident) {
-                       result = left;
+               case TYPE_UNION:
+               case TYPE_JOIN:
+               {
+                       struct type *join, **jnext, *mnext;
+                       join = new_type(TYPE_JOIN, 0, 0);
+                       jnext = &join->left;
+                       mnext = type->left;
+                       while(mnext) {
+                               size_t size;
+                               member = mnext;
+                               mnext = 0;
+                               if ((member->type & TYPE_MASK) == TYPE_OVERLAP) {
+                                       mnext = member->right;
+                                       member = member->left;
+                               }
+                               size = reg_size_of(state, member);
+                               if (size > reg_offset) {
+                                       struct type *part, *hunt;
+                                       part = reg_type(state, member, reg_offset);
+                                       /* See if this type is already in the union */
+                                       hunt = join->left;
+                                       while(hunt) {
+                                               struct type *test = hunt;
+                                               hunt = 0;
+                                               if ((test->type & TYPE_MASK) == TYPE_OVERLAP) {
+                                                       hunt = test->right;
+                                                       test = test->left;
+                                               }
+                                               if (equiv_types(part, test)) {
+                                                       goto next;
+                                               }
+                                       }
+                                       /* Nope add it */
+                                       if (!*jnext) {
+                                               *jnext = part;
+                                       } else {
+                                               *jnext = new_type(TYPE_OVERLAP, *jnext, part);
+                                               jnext = &(*jnext)->right;
+                                       }
+                                       join->elements++;
+                               }
+                       next:
+                               ;
+                       }
+                       if (join->elements == 0) {
+                               internal_error(state, 0, "No elements?");
+                       }
+                       member = join;
+                       break;
                }
-       }
-       /* Test for equivalent functions */
-       else if (type == TYPE_FUNCTION) {
-               struct type *lf, *rf;
-               lf = compatible_types(left->left, right->left);
-               rf = compatible_types(left->right, right->right);
-               if (lf && rf) {
-                       result = new_type(qual_type, lf, rf);
+               default:
+                       member = 0;
+                       fprintf(state->errout, "type: ");
+                       name_of(state->errout, type);
+                       fprintf(state->errout, "\n");
+                       internal_error(state, 0, "reg_type not yet defined for type");
+
                }
        }
-       /* We only see TYPE_PRODUCT as part of function equivalence matching */
-       else if (type == TYPE_PRODUCT) {
-               struct type *lf, *rf;
-               lf = compatible_types(left->left, right->left);
-               rf = compatible_types(left->right, right->right);
-               if (lf && rf) {
-                       result = new_type(qual_type, lf, rf);
-               }
+       /* If I have a single register compound type not a bit-field
+        * find the real type.
+        */
+       member = unpack_type(state, member);
+               ;
+       size  = reg_size_of(state, member);
+       if (size > REG_SIZEOF_REG) {
+               internal_error(state, 0, "Cannot find type of single register");
        }
-       else {
-               /* Nothing else is compatible */
+#if 1
+       invalid = invalid_type(state, member);
+       if (invalid) {
+               fprintf(state->errout, "type: ");
+               name_of(state->errout, member);
+               fprintf(state->errout, "\n");
+               fprintf(state->errout, "invalid: ");
+               name_of(state->errout, invalid);
+               fprintf(state->errout, "\n");
+               internal_error(state, 0, "returning bad type?");
        }
-       return result;
+#endif
+       return member;
 }
 
-static struct type *compatible_ptrs(struct type *left, struct type *right)
+static struct type *next_field(struct compile_state *state,
+       struct type *type, struct type *prev_member)
 {
-       struct type *result;
-       if (((left->type & TYPE_MASK) != TYPE_POINTER) ||
-               ((right->type & TYPE_MASK) != TYPE_POINTER)) {
-               return 0;
+       struct type *member;
+       if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
+               internal_error(state, 0, "next_field only works on structures");
        }
-       result = compatible_types(left->left, right->left);
-       if (result) {
-               unsigned int qual_type;
-               qual_type = (left->type & ~STOR_MASK) | (right->type & ~STOR_MASK);
-               result = new_type(qual_type, result, 0);
+       member = type->left;
+       while((member->type & TYPE_MASK) == TYPE_PRODUCT) {
+               if (!prev_member) {
+                       member = member->left;
+                       break;
+               }
+               if (member->left == prev_member) {
+                       prev_member = 0;
+               }
+               member = member->right;
        }
-       return result;
-       
+       if (member == prev_member) {
+               prev_member = 0;
+       }
+       if (prev_member) {
+               internal_error(state, 0, "prev_member %s not present",
+                       prev_member->field_ident->name);
+       }
+       return member;
 }
-static struct triple *integral_promotion(
-       struct compile_state *state, struct triple *def)
+
+typedef void (*walk_type_fields_cb_t)(struct compile_state *state, struct type *type,
+       size_t ret_offset, size_t mem_offset, void *arg);
+
+static void walk_type_fields(struct compile_state *state,
+       struct type *type, size_t reg_offset, size_t mem_offset,
+       walk_type_fields_cb_t cb, void *arg);
+
+static void walk_struct_fields(struct compile_state *state,
+       struct type *type, size_t reg_offset, size_t mem_offset,
+       walk_type_fields_cb_t cb, void *arg)
 {
-       struct type *type;
-       type = def->type;
-       /* As all operations are carried out in registers
-        * the values are converted on load I just convert
-        * logical type of the operand.
-        */
-       if (TYPE_INTEGER(type->type)) {
-               unsigned int int_type;
-               int_type = type->type & ~TYPE_MASK;
-               int_type |= do_integral_promotion(type->type);
-               if (int_type != type->type) {
-                       def->type = new_type(int_type, 0, 0);
+       struct type *tptr;
+       ulong_t i;
+       if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
+               internal_error(state, 0, "walk_struct_fields only works on structures");
+       }
+       tptr = type->left;
+       for(i = 0; i < type->elements; i++) {
+               struct type *mtype;
+               mtype = tptr;
+               if ((mtype->type & TYPE_MASK) == TYPE_PRODUCT) {
+                       mtype = mtype->left;
                }
+               walk_type_fields(state, mtype,
+                       reg_offset +
+                       field_reg_offset(state, type, mtype->field_ident),
+                       mem_offset +
+                       field_offset(state, type, mtype->field_ident),
+                       cb, arg);
+               tptr = tptr->right;
        }
-       return def;
-}
 
+}
 
-static void arithmetic(struct compile_state *state, struct triple *def)
+static void walk_type_fields(struct compile_state *state,
+       struct type *type, size_t reg_offset, size_t mem_offset,
+       walk_type_fields_cb_t cb, void *arg)
 {
-       if (!TYPE_ARITHMETIC(def->type->type)) {
-               error(state, 0, "arithmetic type expexted");
+       switch(type->type & TYPE_MASK) {
+       case TYPE_STRUCT:
+               walk_struct_fields(state, type, reg_offset, mem_offset, cb, arg);
+               break;
+       case TYPE_CHAR:
+       case TYPE_UCHAR:
+       case TYPE_SHORT:
+       case TYPE_USHORT:
+       case TYPE_INT:
+       case TYPE_UINT:
+       case TYPE_LONG:
+       case TYPE_ULONG:
+               cb(state, type, reg_offset, mem_offset, arg);
+               break;
+       case TYPE_VOID:
+               break;
+       default:
+               internal_error(state, 0, "walk_type_fields not yet implemented for type");
        }
 }
 
-static void ptr_arithmetic(struct compile_state *state, struct triple *def)
+static void arrays_complete(struct compile_state *state, struct type *type)
 {
-       if (!TYPE_PTR(def->type->type) && !TYPE_ARITHMETIC(def->type->type)) {
-               error(state, def, "pointer or arithmetic type expected");
+       if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
+               if (type->elements == ELEMENT_COUNT_UNSPECIFIED) {
+                       error(state, 0, "array size not specified");
+               }
+               arrays_complete(state, type->left);
        }
 }
 
-static int is_integral(struct triple *ins)
+static unsigned int get_basic_type(struct type *type)
 {
-       return TYPE_INTEGER(ins->type->type);
+       unsigned int basic;
+       basic = type->type & TYPE_MASK;
+       /* Convert enums to ints */
+       if (basic == TYPE_ENUM) {
+               basic = TYPE_INT;
+       }
+       /* Convert bitfields to standard types */
+       else if (basic == TYPE_BITFIELD) {
+               if (type->elements <= SIZEOF_CHAR) {
+                       basic = TYPE_CHAR;
+               }
+               else if (type->elements <= SIZEOF_SHORT) {
+                       basic = TYPE_SHORT;
+               }
+               else if (type->elements <= SIZEOF_INT) {
+                       basic = TYPE_INT;
+               }
+               else if (type->elements <= SIZEOF_LONG) {
+                       basic = TYPE_LONG;
+               }
+               if (!TYPE_SIGNED(type->left->type)) {
+                       basic += 1;
+               }
+       }
+       return basic;
 }
 
-static void integral(struct compile_state *state, struct triple *def)
+static unsigned int do_integral_promotion(unsigned int type)
 {
-       if (!is_integral(def)) {
-               error(state, 0, "integral type expected");
+       if (TYPE_INTEGER(type) && (TYPE_RANK(type) < TYPE_RANK(TYPE_INT))) {
+               type = TYPE_INT;
+       }
+       return type;
+}
+
+static unsigned int do_arithmetic_conversion(
+       unsigned int left, unsigned int right)
+{
+       if ((left == TYPE_LDOUBLE) || (right == TYPE_LDOUBLE)) {
+               return TYPE_LDOUBLE;
+       }
+       else if ((left == TYPE_DOUBLE) || (right == TYPE_DOUBLE)) {
+               return TYPE_DOUBLE;
+       }
+       else if ((left == TYPE_FLOAT) || (right == TYPE_FLOAT)) {
+               return TYPE_FLOAT;
+       }
+       left = do_integral_promotion(left);
+       right = do_integral_promotion(right);
+       /* If both operands have the same size done */
+       if (left == right) {
+               return left;
+       }
+       /* If both operands have the same signedness pick the larger */
+       else if (!!TYPE_UNSIGNED(left) == !!TYPE_UNSIGNED(right)) {
+               return (TYPE_RANK(left) >= TYPE_RANK(right)) ? left : right;
+       }
+       /* If the signed type can hold everything use it */
+       else if (TYPE_SIGNED(left) && (TYPE_RANK(left) > TYPE_RANK(right))) {
+               return left;
+       }
+       else if (TYPE_SIGNED(right) && (TYPE_RANK(right) > TYPE_RANK(left))) {
+               return right;
+       }
+       /* Convert to the unsigned type with the same rank as the signed type */
+       else if (TYPE_SIGNED(left)) {
+               return TYPE_MKUNSIGNED(left);
+       }
+       else {
+               return TYPE_MKUNSIGNED(right);
+       }
+}
+
+/* see if two types are the same except for qualifiers */
+static int equiv_types(struct type *left, struct type *right)
+{
+       unsigned int type;
+       /* Error if the basic types do not match */
+       if ((left->type & TYPE_MASK) != (right->type & TYPE_MASK)) {
+               return 0;
+       }
+       type = left->type & TYPE_MASK;
+       /* If the basic types match and it is a void type we are done */
+       if (type == TYPE_VOID) {
+               return 1;
+       }
+       /* For bitfields we need to compare the sizes */
+       else if (type == TYPE_BITFIELD) {
+               return (left->elements == right->elements) &&
+                       (TYPE_SIGNED(left->left->type) == TYPE_SIGNED(right->left->type));
+       }
+       /* if the basic types match and it is an arithmetic type we are done */
+       else if (TYPE_ARITHMETIC(type)) {
+               return 1;
+       }
+       /* If it is a pointer type recurse and keep testing */
+       else if (type == TYPE_POINTER) {
+               return equiv_types(left->left, right->left);
+       }
+       else if (type == TYPE_ARRAY) {
+               return (left->elements == right->elements) &&
+                       equiv_types(left->left, right->left);
+       }
+       /* test for struct equality */
+       else if (type == TYPE_STRUCT) {
+               return left->type_ident == right->type_ident;
+       }
+       /* test for union equality */
+       else if (type == TYPE_UNION) {
+               return left->type_ident == right->type_ident;
+       }
+       /* Test for equivalent functions */
+       else if (type == TYPE_FUNCTION) {
+               return equiv_types(left->left, right->left) &&
+                       equiv_types(left->right, right->right);
+       }
+       /* We only see TYPE_PRODUCT as part of function equivalence matching */
+       /* We also see TYPE_PRODUCT as part of of tuple equivalence matchin */
+       else if (type == TYPE_PRODUCT) {
+               return equiv_types(left->left, right->left) &&
+                       equiv_types(left->right, right->right);
+       }
+       /* We should see TYPE_OVERLAP when comparing joins */
+       else if (type == TYPE_OVERLAP) {
+               return equiv_types(left->left, right->left) &&
+                       equiv_types(left->right, right->right);
+       }
+       /* Test for equivalence of tuples */
+       else if (type == TYPE_TUPLE) {
+               return (left->elements == right->elements) &&
+                       equiv_types(left->left, right->left);
+       }
+       /* Test for equivalence of joins */
+       else if (type == TYPE_JOIN) {
+               return (left->elements == right->elements) &&
+                       equiv_types(left->left, right->left);
+       }
+       else {
+               return 0;
+       }
+}
+
+static int equiv_ptrs(struct type *left, struct type *right)
+{
+       if (((left->type & TYPE_MASK) != TYPE_POINTER) ||
+               ((right->type & TYPE_MASK) != TYPE_POINTER)) {
+               return 0;
+       }
+       return equiv_types(left->left, right->left);
+}
+
+static struct type *compatible_types(struct type *left, struct type *right)
+{
+       struct type *result;
+       unsigned int type, qual_type;
+       /* Error if the basic types do not match */
+       if ((left->type & TYPE_MASK) != (right->type & TYPE_MASK)) {
+               return 0;
+       }
+       type = left->type & TYPE_MASK;
+       qual_type = (left->type & ~STOR_MASK) | (right->type & ~STOR_MASK);
+       result = 0;
+       /* if the basic types match and it is an arithmetic type we are done */
+       if (TYPE_ARITHMETIC(type)) {
+               result = new_type(qual_type, 0, 0);
+       }
+       /* If it is a pointer type recurse and keep testing */
+       else if (type == TYPE_POINTER) {
+               result = compatible_types(left->left, right->left);
+               if (result) {
+                       result = new_type(qual_type, result, 0);
+               }
+       }
+       /* test for struct equality */
+       else if (type == TYPE_STRUCT) {
+               if (left->type_ident == right->type_ident) {
+                       result = left;
+               }
+       }
+       /* test for union equality */
+       else if (type == TYPE_UNION) {
+               if (left->type_ident == right->type_ident) {
+                       result = left;
+               }
+       }
+       /* Test for equivalent functions */
+       else if (type == TYPE_FUNCTION) {
+               struct type *lf, *rf;
+               lf = compatible_types(left->left, right->left);
+               rf = compatible_types(left->right, right->right);
+               if (lf && rf) {
+                       result = new_type(qual_type, lf, rf);
+               }
+       }
+       /* We only see TYPE_PRODUCT as part of function equivalence matching */
+       else if (type == TYPE_PRODUCT) {
+               struct type *lf, *rf;
+               lf = compatible_types(left->left, right->left);
+               rf = compatible_types(left->right, right->right);
+               if (lf && rf) {
+                       result = new_type(qual_type, lf, rf);
+               }
+       }
+       else {
+               /* Nothing else is compatible */
+       }
+       return result;
+}
+
+/* See if left is a equivalent to right or right is a union member of left */
+static int is_subset_type(struct type *left, struct type *right)
+{
+       if (equiv_types(left, right)) {
+               return 1;
+       }
+       if ((left->type & TYPE_MASK) == TYPE_JOIN) {
+               struct type *member, *mnext;
+               mnext = left->left;
+               while(mnext) {
+                       member = mnext;
+                       mnext = 0;
+                       if ((member->type & TYPE_MASK) == TYPE_OVERLAP) {
+                               mnext = member->right;
+                               member = member->left;
+                       }
+                       if (is_subset_type( member, right)) {
+                               return 1;
+                       }
+               }
+       }
+       return 0;
+}
+
+static struct type *compatible_ptrs(struct type *left, struct type *right)
+{
+       struct type *result;
+       if (((left->type & TYPE_MASK) != TYPE_POINTER) ||
+               ((right->type & TYPE_MASK) != TYPE_POINTER)) {
+               return 0;
+       }
+       result = compatible_types(left->left, right->left);
+       if (result) {
+               unsigned int qual_type;
+               qual_type = (left->type & ~STOR_MASK) | (right->type & ~STOR_MASK);
+               result = new_type(qual_type, result, 0);
+       }
+       return result;
+
+}
+static struct triple *integral_promotion(
+       struct compile_state *state, struct triple *def)
+{
+       struct type *type;
+       type = def->type;
+       /* As all operations are carried out in registers
+        * the values are converted on load I just convert
+        * logical type of the operand.
+        */
+       if (TYPE_INTEGER(type->type)) {
+               unsigned int int_type;
+               int_type = type->type & ~TYPE_MASK;
+               int_type |= do_integral_promotion(get_basic_type(type));
+               if (int_type != type->type) {
+                       if (def->op != OP_LOAD) {
+                               def->type = new_type(int_type, 0, 0);
+                       }
+                       else {
+                               def = triple(state, OP_CONVERT,
+                                       new_type(int_type, 0, 0), def, 0);
+                       }
+               }
+       }
+       return def;
+}
+
+
+static void arithmetic(struct compile_state *state, struct triple *def)
+{
+       if (!TYPE_ARITHMETIC(def->type->type)) {
+               error(state, 0, "arithmetic type expexted");
+       }
+}
+
+static void ptr_arithmetic(struct compile_state *state, struct triple *def)
+{
+       if (!TYPE_PTR(def->type->type) && !TYPE_ARITHMETIC(def->type->type)) {
+               error(state, def, "pointer or arithmetic type expected");
+       }
+}
+
+static int is_integral(struct triple *ins)
+{
+       return TYPE_INTEGER(ins->type->type);
+}
+
+static void integral(struct compile_state *state, struct triple *def)
+{
+       if (!is_integral(def)) {
+               error(state, 0, "integral type expected");
        }
 }
 
@@ -4062,8 +7146,28 @@ static void bool(struct compile_state *state, struct triple *def)
 
 static int is_signed(struct type *type)
 {
+       if ((type->type & TYPE_MASK) == TYPE_BITFIELD) {
+               type = type->left;
+       }
        return !!TYPE_SIGNED(type->type);
 }
+static int is_compound_type(struct type *type)
+{
+       int is_compound;
+       switch((type->type & TYPE_MASK)) {
+       case TYPE_ARRAY:
+       case TYPE_STRUCT:
+       case TYPE_TUPLE:
+       case TYPE_UNION:
+       case TYPE_JOIN:
+               is_compound = 1;
+               break;
+       default:
+               is_compound = 0;
+               break;
+       }
+       return is_compound;
+}
 
 /* Is this value located in a register otherwise it must be in memory */
 static int is_in_reg(struct compile_state *state, struct triple *def)
@@ -4075,77 +7179,53 @@ static int is_in_reg(struct compile_state *state, struct triple *def)
        else if ((def->op == OP_SDECL) || (def->op == OP_DEREF)) {
                in_reg = 0;
        }
-       else if (def->op == OP_VAL_VEC) {
-               in_reg = is_in_reg(state, RHS(def, 0));
-       }
-       else if (def->op == OP_DOT) {
-               in_reg = is_in_reg(state, RHS(def, 0));
+       else if (triple_is_part(state, def)) {
+               in_reg = is_in_reg(state, MISC(def, 0));
        }
        else {
-               internal_error(state, 0, "unknown expr storage location");
+               internal_error(state, def, "unknown expr storage location");
                in_reg = -1;
        }
        return in_reg;
 }
 
-/* Is this a stable variable location otherwise it must be a temporary */
-static int is_stable(struct compile_state *state, struct triple *def)
+/* Is this an auto or static variable location? Something that can
+ * be assigned to.  Otherwise it must must be a pure value, a temporary.
+ */
+static int is_lvalue(struct compile_state *state, struct triple *def)
 {
        int ret;
        ret = 0;
        if (!def) {
                return 0;
        }
-       if ((def->op == OP_ADECL) || 
-               (def->op == OP_SDECL) || 
+       if ((def->op == OP_ADECL) ||
+               (def->op == OP_SDECL) ||
                (def->op == OP_DEREF) ||
-               (def->op == OP_BLOBCONST)) {
+               (def->op == OP_BLOBCONST) ||
+               (def->op == OP_LIST)) {
                ret = 1;
        }
-       else if (def->op == OP_DOT) {
-               ret = is_stable(state, RHS(def, 0));
-       }
-       else if (def->op == OP_VAL_VEC) {
-               struct triple **vector;
-               ulong_t i;
-               ret = 1;
-               vector = &RHS(def, 0);
-               for(i = 0; i < def->type->elements; i++) {
-                       if (!is_stable(state, vector[i])) {
-                               ret = 0;
-                               break;
-                       }
-               }
+       else if (triple_is_part(state, def)) {
+               ret = is_lvalue(state, MISC(def, 0));
        }
        return ret;
 }
 
-static int is_lvalue(struct compile_state *state, struct triple *def)
+static void clvalue(struct compile_state *state, struct triple *def)
 {
-       int ret;
-       ret = 1;
        if (!def) {
-               return 0;
+               internal_error(state, def, "nothing where lvalue expected?");
        }
-       if (!is_stable(state, def)) {
-               return 0;
+       if (!is_lvalue(state, def)) {
+               error(state, def, "lvalue expected");
        }
+}
+static void lvalue(struct compile_state *state, struct triple *def)
+{
+       clvalue(state, def);
        if (def->type->type & QUAL_CONST) {
-               ret = 0;
-       }
-       else if (def->op == OP_DOT) {
-               ret = is_lvalue(state, RHS(def, 0));
-       }
-       return ret;
-}
-
-static void lvalue(struct compile_state *state, struct triple *def)
-{
-       if (!def) {
-               internal_error(state, def, "nothing where lvalue expected?");
-       }
-       if (!is_lvalue(state, def)) { 
-               error(state, def, "lvalue expected");
+               error(state, def, "modifable lvalue expected");
        }
 }
 
@@ -4171,7 +7251,7 @@ static struct triple *int_const(
        case TYPE_LONG:  case TYPE_ULONG:
                break;
        default:
-               internal_error(state, 0, "constant for unkown type");
+               internal_error(state, 0, "constant for unknown type");
        }
        result = triple(state, OP_INTCONST, type, 0, 0);
        result->u.cval = value;
@@ -4179,39 +7259,66 @@ static struct triple *int_const(
 }
 
 
-static struct triple *do_mk_addr_expr(struct compile_state *state, 
+static struct triple *read_expr(struct compile_state *state, struct triple *def);
+
+static struct triple *do_mk_addr_expr(struct compile_state *state,
        struct triple *expr, struct type *type, ulong_t offset)
 {
        struct triple *result;
-       lvalue(state, expr);
+       struct type *ptr_type;
+       clvalue(state, expr);
+
+       ptr_type = new_type(TYPE_POINTER | (type->type & QUAL_MASK), type, 0);
+
 
        result = 0;
        if (expr->op == OP_ADECL) {
                error(state, expr, "address of auto variables not supported");
        }
        else if (expr->op == OP_SDECL) {
-               result = triple(state, OP_ADDRCONST, type, 0, 0);
+               result = triple(state, OP_ADDRCONST, ptr_type, 0, 0);
                MISC(result, 0) = expr;
                result->u.cval = offset;
        }
        else if (expr->op == OP_DEREF) {
-               result = triple(state, OP_ADD, type,
+               result = triple(state, OP_ADD, ptr_type,
                        RHS(expr, 0),
                        int_const(state, &ulong_type, offset));
        }
+       else if (expr->op == OP_BLOBCONST) {
+               FINISHME();
+               internal_error(state, expr, "not yet implemented");
+       }
+       else if (expr->op == OP_LIST) {
+               error(state, 0, "Function addresses not supported");
+       }
+       else if (triple_is_part(state, expr)) {
+               struct triple *part;
+               part = expr;
+               expr = MISC(expr, 0);
+               if (part->op == OP_DOT) {
+                       offset += bits_to_bytes(
+                               field_offset(state, expr->type, part->u.field));
+               }
+               else if (part->op == OP_INDEX) {
+                       offset += bits_to_bytes(
+                               index_offset(state, expr->type, part->u.cval));
+               }
+               else {
+                       internal_error(state, part, "unhandled part type");
+               }
+               result = do_mk_addr_expr(state, expr, type, offset);
+       }
+       if (!result) {
+               internal_error(state, expr, "cannot take address of expression");
+       }
        return result;
 }
 
 static struct triple *mk_addr_expr(
        struct compile_state *state, struct triple *expr, ulong_t offset)
 {
-       struct type *type;
-       
-       type = new_type(
-               TYPE_POINTER | (expr->type->type & QUAL_MASK),
-               expr->type, 0);
-
-       return do_mk_addr_expr(state, expr, type, offset);
+       return do_mk_addr_expr(state, expr, expr->type, offset);
 }
 
 static struct triple *mk_deref_expr(
@@ -4220,84 +7327,126 @@ static struct triple *mk_deref_expr(
        struct type *base_type;
        pointer(state, expr);
        base_type = expr->type->left;
-       if (!TYPE_PTR(base_type->type) && !TYPE_ARITHMETIC(base_type->type)) {
-               error(state, 0, 
-                       "Only pointer and arithmetic values can be dereferenced");
-       }
        return triple(state, OP_DEREF, base_type, expr, 0);
 }
 
+/* lvalue conversions always apply except when certain operators
+ * are applied.  So I apply apply it when I know no more
+ * operators will be applied.
+ */
+static struct triple *lvalue_conversion(struct compile_state *state, struct triple *def)
+{
+       /* Tranform an array to a pointer to the first element */
+       if ((def->type->type & TYPE_MASK) == TYPE_ARRAY) {
+               struct type *type;
+               type = new_type(
+                       TYPE_POINTER | (def->type->type & QUAL_MASK),
+                       def->type->left, 0);
+               if ((def->op == OP_SDECL) || IS_CONST_OP(def->op)) {
+                       struct triple *addrconst;
+                       if ((def->op != OP_SDECL) && (def->op != OP_BLOBCONST)) {
+                               internal_error(state, def, "bad array constant");
+                       }
+                       addrconst = triple(state, OP_ADDRCONST, type, 0, 0);
+                       MISC(addrconst, 0) = def;
+                       def = addrconst;
+               }
+               else {
+                       def = triple(state, OP_CONVERT, type, def, 0);
+               }
+       }
+       /* Transform a function to a pointer to it */
+       else if ((def->type->type & TYPE_MASK) == TYPE_FUNCTION) {
+               def = mk_addr_expr(state, def, 0);
+       }
+       return def;
+}
+
 static struct triple *deref_field(
        struct compile_state *state, struct triple *expr, struct hash_entry *field)
 {
        struct triple *result;
        struct type *type, *member;
+       ulong_t offset;
        if (!field) {
                internal_error(state, 0, "No field passed to deref_field");
        }
        result = 0;
        type = expr->type;
-       if ((type->type & TYPE_MASK) != TYPE_STRUCT) {
+       if (((type->type & TYPE_MASK) != TYPE_STRUCT) &&
+               ((type->type & TYPE_MASK) != TYPE_UNION)) {
                error(state, 0, "request for member %s in something not a struct or union",
                        field->name);
        }
-       member = type->left;
-       while((member->type & TYPE_MASK) == TYPE_PRODUCT) {
-               if (member->left->field_ident == field) {
-                       member = member->left;
-                       break;
-               }
-               member = member->right;
-       }
-       if (member->field_ident != field) {
-               error(state, 0, "%s is not a member", field->name);
-       }
+       member = field_type(state, type, field);
        if ((type->type & STOR_MASK) == STOR_PERM) {
                /* Do the pointer arithmetic to get a deref the field */
-               ulong_t offset;
-               offset = field_offset(state, type, field);
+               offset = bits_to_bytes(field_offset(state, type, field));
                result = do_mk_addr_expr(state, expr, member, offset);
                result = mk_deref_expr(state, result);
        }
        else {
                /* Find the variable for the field I want. */
-               result = triple(state, OP_DOT, 
-                       field_type(state, type, field), expr, 0);
+               result = triple(state, OP_DOT, member, expr, 0);
                result->u.field = field;
        }
        return result;
 }
 
+static struct triple *deref_index(
+       struct compile_state *state, struct triple *expr, size_t index)
+{
+       struct triple *result;
+       struct type *type, *member;
+       ulong_t offset;
+
+       result = 0;
+       type = expr->type;
+       member = index_type(state, type, index);
+
+       if ((type->type & STOR_MASK) == STOR_PERM) {
+               offset = bits_to_bytes(index_offset(state, type, index));
+               result = do_mk_addr_expr(state, expr, member, offset);
+               result = mk_deref_expr(state, result);
+       }
+       else {
+               result = triple(state, OP_INDEX, member, expr, 0);
+               result->u.cval = index;
+       }
+       return result;
+}
+
 static struct triple *read_expr(struct compile_state *state, struct triple *def)
 {
        int op;
        if  (!def) {
                return 0;
        }
-       if (!is_stable(state, def)) {
+#if DEBUG_ROMCC_WARNINGS
+#warning "CHECK_ME is this the only place I need to do lvalue conversions?"
+#endif
+       /* Transform lvalues into something we can read */
+       def = lvalue_conversion(state, def);
+       if (!is_lvalue(state, def)) {
                return def;
        }
-       /* Tranform an array to a pointer to the first element */
-#warning "CHECK_ME is this the right place to transform arrays to pointers?"
-       if ((def->type->type & TYPE_MASK) == TYPE_ARRAY) {
-               struct type *type;
-               struct triple *result;
-               type = new_type(
-                       TYPE_POINTER | (def->type->type & QUAL_MASK),
-                       def->type->left, 0);
-               result = triple(state, OP_ADDRCONST, type, 0, 0);
-               MISC(result, 0) = def;
-               return result;
-       }
        if (is_in_reg(state, def)) {
                op = OP_READ;
        } else {
+               if (def->op == OP_SDECL) {
+                       def = mk_addr_expr(state, def, 0);
+                       def = mk_deref_expr(state, def);
+               }
                op = OP_LOAD;
        }
-       return triple(state, op, def->type, def, 0);
+       def = triple(state, op, def->type, def, 0);
+       if (def->type->type & QUAL_VOLATILE) {
+               def->id |= TRIPLE_FLAG_VOLATILE;
+       }
+       return def;
 }
 
-static void write_compatible(struct compile_state *state,
+int is_write_compatible(struct compile_state *state,
        struct type *dest, struct type *rval)
 {
        int compatible = 0;
@@ -4317,21 +7466,43 @@ static void write_compatible(struct compile_state *state,
                compatible = 1;
        }
        /* test for struct/union equality  */
-       else if (((dest->type & TYPE_MASK) == TYPE_STRUCT) &&
-               ((rval->type & TYPE_MASK) == TYPE_STRUCT) &&
-               (dest->type_ident == rval->type_ident)) {
+       else if (equiv_types(dest, rval)) {
                compatible = 1;
        }
-       if (!compatible) {
+       return compatible;
+}
+
+static void write_compatible(struct compile_state *state,
+       struct type *dest, struct type *rval)
+{
+       if (!is_write_compatible(state, dest, rval)) {
+               FILE *fp = state->errout;
+               fprintf(fp, "dest: ");
+               name_of(fp, dest);
+               fprintf(fp,"\nrval: ");
+               name_of(fp, rval);
+               fprintf(fp, "\n");
                error(state, 0, "Incompatible types in assignment");
        }
 }
 
+static int is_init_compatible(struct compile_state *state,
+       struct type *dest, struct type *rval)
+{
+       int compatible = 0;
+       if (is_write_compatible(state, dest, rval)) {
+               compatible = 1;
+       }
+       else if (equiv_types(dest, rval)) {
+               compatible = 1;
+       }
+       return compatible;
+}
+
 static struct triple *write_expr(
        struct compile_state *state, struct triple *dest, struct triple *rval)
 {
        struct triple *def;
-       int op;
 
        def = 0;
        if (!rval) {
@@ -4344,17 +7515,30 @@ static struct triple *write_expr(
        if (!is_lvalue(state, dest)) {
                internal_error(state, 0, "writing to a non lvalue?");
        }
+       if (dest->type->type & QUAL_CONST) {
+               internal_error(state, 0, "modifable lvalue expexted");
+       }
 
        write_compatible(state, dest->type, rval->type);
+       if (!equiv_types(dest->type, rval->type)) {
+               rval = triple(state, OP_CONVERT, dest->type, rval, 0);
+       }
 
        /* Now figure out which assignment operator to use */
-       op = -1;
        if (is_in_reg(state, dest)) {
-               op = OP_WRITE;
+               def = triple(state, OP_WRITE, dest->type, rval, dest);
+               if (MISC(def, 0) != dest) {
+                       internal_error(state, def, "huh?");
+               }
+               if (RHS(def, 0) != rval) {
+                       internal_error(state, def, "huh?");
+               }
        } else {
-               op = OP_STORE;
+               def = triple(state, OP_STORE, dest->type, dest, rval);
+       }
+       if (def->type->type & QUAL_VOLATILE) {
+               def->id |= TRIPLE_FLAG_VOLATILE;
        }
-       def = triple(state, op, dest->type, dest, rval);
        return def;
 }
 
@@ -4399,8 +7583,9 @@ struct type *arithmetic_result(
        arithmetic(state, right);
        type = new_type(
                do_arithmetic_conversion(
-                       left->type->type, 
-                       right->type->type), 0, 0);
+                       get_basic_type(left->type),
+                       get_basic_type(right->type)),
+               0, 0);
        return type;
 }
 
@@ -4411,7 +7596,7 @@ struct type *ptr_arithmetic_result(
        /* Sanity checks to ensure I am working with the proper types */
        ptr_arithmetic(state, left);
        arithmetic(state, right);
-       if (TYPE_ARITHMETIC(left->type->type) && 
+       if (TYPE_ARITHMETIC(left->type->type) &&
                TYPE_ARITHMETIC(right->type->type)) {
                type = arithmetic_result(state, left, right);
        }
@@ -4425,10 +7610,9 @@ struct type *ptr_arithmetic_result(
        return type;
 }
 
-
 /* boolean helper function */
 
-static struct triple *ltrue_expr(struct compile_state *state, 
+static struct triple *ltrue_expr(struct compile_state *state,
        struct triple *expr)
 {
        switch(expr->op) {
@@ -4444,17 +7628,77 @@ static struct triple *ltrue_expr(struct compile_state *state,
        return expr;
 }
 
-static struct triple *lfalse_expr(struct compile_state *state, 
+static struct triple *lfalse_expr(struct compile_state *state,
        struct triple *expr)
 {
        return triple(state, OP_LFALSE, &int_type, expr, 0);
 }
 
-static struct triple *cond_expr(
-       struct compile_state *state, 
+static struct triple *mkland_expr(
+       struct compile_state *state,
+       struct triple *left, struct triple *right)
+{
+       struct triple *def, *val, *var, *jmp, *mid, *end;
+       struct triple *lstore, *rstore;
+
+       /* Generate some intermediate triples */
+       end = label(state);
+       var = variable(state, &int_type);
+
+       /* Store the left hand side value */
+       lstore = write_expr(state, var, left);
+
+       /* Jump if the value is false */
+       jmp =  branch(state, end,
+               lfalse_expr(state, read_expr(state, var)));
+       mid = label(state);
+
+       /* Store the right hand side value */
+       rstore = write_expr(state, var, right);
+
+       /* An expression for the computed value */
+       val = read_expr(state, var);
+
+       /* Generate the prog for a logical and */
+       def = mkprog(state, var, lstore, jmp, mid, rstore, end, val, 0UL);
+
+       return def;
+}
+
+static struct triple *mklor_expr(
+       struct compile_state *state,
+       struct triple *left, struct triple *right)
+{
+       struct triple *def, *val, *var, *jmp, *mid, *end;
+
+       /* Generate some intermediate triples */
+       end = label(state);
+       var = variable(state, &int_type);
+
+       /* Store the left hand side value */
+       left = write_expr(state, var, left);
+
+       /* Jump if the value is true */
+       jmp = branch(state, end, read_expr(state, var));
+       mid = label(state);
+
+       /* Store the right hand side value */
+       right = write_expr(state, var, right);
+
+       /* An expression for the computed value*/
+       val = read_expr(state, var);
+
+       /* Generate the prog for a logical or */
+       def = mkprog(state, var, left, jmp, mid, right, end, val, 0UL);
+
+       return def;
+}
+
+static struct triple *mkcond_expr(
+       struct compile_state *state,
        struct triple *test, struct triple *left, struct triple *right)
 {
-       struct triple *def;
+       struct triple *def, *val, *var, *jmp1, *jmp2, *top, *mid, *end;
        struct type *result_type;
        unsigned int left_type, right_type;
        bool(state, test);
@@ -4489,18 +7733,39 @@ static struct triple *cond_expr(
        if (!result_type) {
                error(state, 0, "Incompatible types in conditional expression");
        }
-       /* Cleanup and invert the test */
-       test = lfalse_expr(state, read_expr(state, test));
-       def = new_triple(state, OP_COND, result_type, 0, 3);
-       def->param[0] = test;
-       def->param[1] = left;
-       def->param[2] = right;
+       /* Generate some intermediate triples */
+       mid = label(state);
+       end = label(state);
+       var = variable(state, result_type);
+
+       /* Branch if the test is false */
+       jmp1 = branch(state, mid, lfalse_expr(state, read_expr(state, test)));
+       top = label(state);
+
+       /* Store the left hand side value */
+       left = write_expr(state, var, left);
+
+       /* Branch to the end */
+       jmp2 = branch(state, end, 0);
+
+       /* Store the right hand side value */
+       right = write_expr(state, var, right);
+
+       /* An expression for the computed value */
+       val = read_expr(state, var);
+
+       /* Generate the prog for a conditional expression */
+       def = mkprog(state, var, jmp1, top, left, jmp2, mid, right, end, val, 0UL);
+
        return def;
 }
 
 
 static int expr_depth(struct compile_state *state, struct triple *ins)
 {
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME move optimal ordering of subexpressions into the optimizer"
+#endif
        int count;
        count = 0;
        if (!ins || (ins->id & TRIPLE_FLAG_FLATTENED)) {
@@ -4512,13 +7777,7 @@ static int expr_depth(struct compile_state *state, struct triple *ins)
        else if (ins->op == OP_VAL) {
                count = expr_depth(state, RHS(ins, 0)) - 1;
        }
-       else if (ins->op == OP_COMMA) {
-               int ldepth, rdepth;
-               ldepth = expr_depth(state, RHS(ins, 0));
-               rdepth = expr_depth(state, RHS(ins, 1));
-               count = (ldepth >= rdepth)? ldepth : rdepth;
-       }
-       else if (ins->op == OP_CALL) {
+       else if (ins->op == OP_FCALL) {
                /* Don't figure the depth of a call just guess it is huge */
                count = 1000;
        }
@@ -4538,21 +7797,19 @@ static int expr_depth(struct compile_state *state, struct triple *ins)
        return count + 1;
 }
 
-static struct triple *flatten(
-       struct compile_state *state, struct triple *first, struct triple *ptr);
-
 static struct triple *flatten_generic(
-       struct compile_state *state, struct triple *first, struct triple *ptr)
+       struct compile_state *state, struct triple *first, struct triple *ptr,
+       int ignored)
 {
        struct rhs_vector {
                int depth;
                struct triple **ins;
        } vector[MAX_RHS];
        int i, rhs, lhs;
-       /* Only operations with just a rhs should come here */
-       rhs = TRIPLE_RHS(ptr->sizes);
-       lhs = TRIPLE_LHS(ptr->sizes);
-       if (TRIPLE_SIZE(ptr->sizes) != lhs + rhs) {
+       /* Only operations with just a rhs and a lhs should come here */
+       rhs = ptr->rhs;
+       lhs = ptr->lhs;
+       if (TRIPLE_SIZE(ptr) != lhs + rhs + ignored) {
                internal_error(state, ptr, "unexpected args for: %d %s",
                        ptr->op, tops(ptr->op));
        }
@@ -4581,305 +7838,137 @@ static struct triple *flatten_generic(
                *vector[i].ins = flatten(state, first, *vector[i].ins);
                use_triple(*vector[i].ins, ptr);
        }
-       
-       /* Now flatten the lhs elements */
-       for(i = 0; i < lhs; i++) {
-               struct triple **ins = &LHS(ptr, i);
-               *ins = flatten(state, first, *ins);
-               use_triple(*ins, ptr);
+       if (lhs) {
+               insert_triple(state, first, ptr);
+               ptr->id |= TRIPLE_FLAG_FLATTENED;
+               ptr->id &= ~TRIPLE_FLAG_LOCAL;
+
+               /* Now flatten the lhs elements */
+               for(i = 0; i < lhs; i++) {
+                       struct triple **ins = &LHS(ptr, i);
+                       *ins = flatten(state, first, *ins);
+                       use_triple(*ins, ptr);
+               }
        }
        return ptr;
 }
 
-static struct triple *flatten_land(
+static struct triple *flatten_prog(
        struct compile_state *state, struct triple *first, struct triple *ptr)
 {
-       struct triple *left, *right;
-       struct triple *val, *test, *jmp, *label1, *end;
-
-       /* Find the triples */
-       left = RHS(ptr, 0);
-       right = RHS(ptr, 1);
+       struct triple *head, *body, *val;
+       head = RHS(ptr, 0);
+       RHS(ptr, 0) = 0;
+       val  = head->prev;
+       body = head->next;
+       release_triple(state, head);
+       release_triple(state, ptr);
+       val->next        = first;
+       body->prev       = first->prev;
+       body->prev->next = body;
+       val->next->prev  = val;
 
-       /* Generate the needed triples */
-       end = label(state);
+       if (triple_is_cbranch(state, body->prev) ||
+               triple_is_call(state, body->prev)) {
+               unuse_triple(first, body->prev);
+               use_triple(body, body->prev);
+       }
 
-       /* Thread the triples together */
-       val          = flatten(state, first, variable(state, ptr->type));
-       left         = flatten(state, first, write_expr(state, val, left));
-       test         = flatten(state, first, 
-               lfalse_expr(state, read_expr(state, val)));
-       jmp          = flatten(state, first, branch(state, end, test));
-       label1       = flatten(state, first, label(state));
-       right        = flatten(state, first, write_expr(state, val, right));
-       TARG(jmp, 0) = flatten(state, first, end); 
-       
-       /* Now give the caller something to chew on */
-       return read_expr(state, val);
-}
-
-static struct triple *flatten_lor(
-       struct compile_state *state, struct triple *first, struct triple *ptr)
-{
-       struct triple *left, *right;
-       struct triple *val, *jmp, *label1, *end;
+       if (!(val->id & TRIPLE_FLAG_FLATTENED)) {
+               internal_error(state, val, "val not flattened?");
+       }
 
-       /* Find the triples */
-       left = RHS(ptr, 0);
-       right = RHS(ptr, 1);
+       return val;
+}
 
-       /* Generate the needed triples */
-       end = label(state);
 
-       /* Thread the triples together */
-       val          = flatten(state, first, variable(state, ptr->type));
-       left         = flatten(state, first, write_expr(state, val, left));
-       jmp          = flatten(state, first, branch(state, end, left));
-       label1       = flatten(state, first, label(state));
-       right        = flatten(state, first, write_expr(state, val, right));
-       TARG(jmp, 0) = flatten(state, first, end);
-       
-       
-       /* Now give the caller something to chew on */
-       return read_expr(state, val);
-}
-
-static struct triple *flatten_cond(
+static struct triple *flatten_part(
        struct compile_state *state, struct triple *first, struct triple *ptr)
 {
-       struct triple *test, *left, *right;
-       struct triple *val, *mv1, *jmp1, *label1, *mv2, *middle, *jmp2, *end;
-
-       /* Find the triples */
-       test = RHS(ptr, 0);
-       left = RHS(ptr, 1);
-       right = RHS(ptr, 2);
-
-       /* Generate the needed triples */
-       end = label(state);
-       middle = label(state);
+       if (!triple_is_part(state, ptr)) {
+               internal_error(state, ptr,  "not a part");
+       }
+       if (ptr->rhs || ptr->lhs || ptr->targ || (ptr->misc != 1)) {
+               internal_error(state, ptr, "unexpected args for: %d %s",
+                       ptr->op, tops(ptr->op));
+       }
+       MISC(ptr, 0) = flatten(state, first, MISC(ptr, 0));
+       use_triple(MISC(ptr, 0), ptr);
+       return flatten_generic(state, first, ptr, 1);
+}
 
-       /* Thread the triples together */
-       val           = flatten(state, first, variable(state, ptr->type));
-       test          = flatten(state, first, test);
-       jmp1          = flatten(state, first, branch(state, middle, test));
-       label1        = flatten(state, first, label(state));
-       left          = flatten(state, first, left);
-       mv1           = flatten(state, first, write_expr(state, val, left));
-       jmp2          = flatten(state, first, branch(state, end, 0));
-       TARG(jmp1, 0) = flatten(state, first, middle);
-       right         = flatten(state, first, right);
-       mv2           = flatten(state, first, write_expr(state, val, right));
-       TARG(jmp2, 0) = flatten(state, first, end);
-       
-       /* Now give the caller something to chew on */
-       return read_expr(state, val);
-}
-
-struct triple *copy_func(struct compile_state *state, struct triple *ofunc)
+static struct triple *flatten(
+       struct compile_state *state, struct triple *first, struct triple *ptr)
 {
-       struct triple *nfunc;
-       struct triple *nfirst, *ofirst;
-       struct triple *new, *old;
-
-#if 0
-       fprintf(stdout, "\n");
-       loc(stdout, state, 0);
-       fprintf(stdout, "\n__________ copy_func _________\n");
-       print_triple(state, ofunc);
-       fprintf(stdout, "__________ copy_func _________ done\n\n");
-#endif
-
-       /* Make a new copy of the old function */
-       nfunc = triple(state, OP_LIST, ofunc->type, 0, 0);
-       nfirst = 0;
-       ofirst = old = RHS(ofunc, 0);
+       struct triple *orig_ptr;
+       if (!ptr)
+               return 0;
        do {
-               struct triple *new;
-               int old_lhs, old_rhs;
-               old_lhs = TRIPLE_LHS(old->sizes);
-               old_rhs = TRIPLE_RHS(old->sizes);
-               new = alloc_triple(state, old->op, old->type, old_lhs, old_rhs,
-                       old->filename, old->line, old->col);
-               if (!triple_stores_block(state, new)) {
-                       memcpy(&new->u, &old->u, sizeof(new->u));
-               }
-               if (!nfirst) {
-                       RHS(nfunc, 0) = nfirst = new;
-               }
-               else {
-                       insert_triple(state, nfirst, new);
-               }
-               new->id |= TRIPLE_FLAG_FLATTENED;
-               
-               /* During the copy remember new as user of old */
-               use_triple(old, new);
-
-               /* Populate the return type if present */
-               if (old == MISC(ofunc, 0)) {
-                       MISC(nfunc, 0) = new;
+               orig_ptr = ptr;
+               /* Only flatten triples once */
+               if (ptr->id & TRIPLE_FLAG_FLATTENED) {
+                       return ptr;
                }
-               old = old->next;
-       } while(old != ofirst);
-
-       /* Make a second pass to fix up any unresolved references */
-       old = ofirst;
-       new = nfirst;
-       do {
-               struct triple **oexpr, **nexpr;
-               int count, i;
-               /* Lookup where the copy is, to join pointers */
-               count = TRIPLE_SIZE(old->sizes);
-               for(i = 0; i < count; i++) {
-                       oexpr = &old->param[i];
-                       nexpr = &new->param[i];
-                       if (!*nexpr && *oexpr && (*oexpr)->use) {
-                               *nexpr = (*oexpr)->use->member;
-                               if (*nexpr == old) {
-                                       internal_error(state, 0, "new == old?");
-                               }
-                               use_triple(*nexpr, new);
-                       }
-                       if (!*nexpr && *oexpr) {
-                               internal_error(state, 0, "Could not copy %d\n", i);
+               switch(ptr->op) {
+               case OP_VAL:
+                       RHS(ptr, 0) = flatten(state, first, RHS(ptr, 0));
+                       return MISC(ptr, 0);
+                       break;
+               case OP_PROG:
+                       ptr = flatten_prog(state, first, ptr);
+                       break;
+               case OP_FCALL:
+                       ptr = flatten_generic(state, first, ptr, 1);
+                       insert_triple(state, first, ptr);
+                       ptr->id |= TRIPLE_FLAG_FLATTENED;
+                       ptr->id &= ~TRIPLE_FLAG_LOCAL;
+                       if (ptr->next != ptr) {
+                               use_triple(ptr->next, ptr);
                        }
-               }
-               old = old->next;
-               new = new->next;
-       } while((old != ofirst) && (new != nfirst));
-       
-       /* Make a third pass to cleanup the extra useses */
-       old = ofirst;
-       new = nfirst;
-       do {
-               unuse_triple(old, new);
-               old = old->next;
-               new = new->next;
-       } while ((old != ofirst) && (new != nfirst));
-       return nfunc;
-}
-
-static struct triple *flatten_call(
-       struct compile_state *state, struct triple *first, struct triple *ptr)
-{
-       /* Inline the function call */
-       struct type *ptype;
-       struct triple *ofunc, *nfunc, *nfirst, *param, *result;
-       struct triple *end, *nend;
-       int pvals, i;
-
-       /* Find the triples */
-       ofunc = MISC(ptr, 0);
-       if (ofunc->op != OP_LIST) {
-               internal_error(state, 0, "improper function");
-       }
-       nfunc = copy_func(state, ofunc);
-       nfirst = RHS(nfunc, 0)->next;
-       /* Prepend the parameter reading into the new function list */
-       ptype = nfunc->type->right;
-       param = RHS(nfunc, 0)->next;
-       pvals = TRIPLE_RHS(ptr->sizes);
-       for(i = 0; i < pvals; i++) {
-               struct type *atype;
-               struct triple *arg;
-               atype = ptype;
-               if ((ptype->type & TYPE_MASK) == TYPE_PRODUCT) {
-                       atype = ptype->left;
-               }
-               while((param->type->type & TYPE_MASK) != (atype->type & TYPE_MASK)) {
-                       param = param->next;
-               }
-               arg = RHS(ptr, i);
-               flatten(state, nfirst, write_expr(state, param, arg));
-               ptype = ptype->right;
-               param = param->next;
-       }
-       result = 0;
-       if ((nfunc->type->left->type & TYPE_MASK) != TYPE_VOID) {
-               result = read_expr(state, MISC(nfunc,0));
-       }
-#if 0
-       fprintf(stdout, "\n");
-       loc(stdout, state, 0);
-       fprintf(stdout, "\n__________ flatten_call _________\n");
-       print_triple(state, nfunc);
-       fprintf(stdout, "__________ flatten_call _________ done\n\n");
-#endif
-
-       /* Get rid of the extra triples */
-       nfirst = RHS(nfunc, 0)->next;
-       free_triple(state, RHS(nfunc, 0));
-       RHS(nfunc, 0) = 0;
-       free_triple(state, nfunc);
-
-       /* Append the new function list onto the return list */
-       end = first->prev;
-       nend = nfirst->prev;
-       end->next    = nfirst;
-       nfirst->prev = end;
-       nend->next   = first;
-       first->prev  = nend;
-
-       return result;
-}
-
-static struct triple *flatten(
-       struct compile_state *state, struct triple *first, struct triple *ptr)
-{
-       struct triple *orig_ptr;
-       if (!ptr)
-               return 0;
-       do {
-               orig_ptr = ptr;
-               /* Only flatten triples once */
-               if (ptr->id & TRIPLE_FLAG_FLATTENED) {
-                       return ptr;
-               }
-               switch(ptr->op) {
-               case OP_WRITE:
-               case OP_STORE:
-                       RHS(ptr, 0) = flatten(state, first, RHS(ptr, 0));
-                       LHS(ptr, 0) = flatten(state, first, LHS(ptr, 0));
-                       use_triple(LHS(ptr, 0), ptr);
-                       use_triple(RHS(ptr, 0), ptr);
-                       break;
-               case OP_COMMA:
-                       RHS(ptr, 0) = flatten(state, first, RHS(ptr, 0));
-                       ptr = RHS(ptr, 1);
                        break;
-               case OP_VAL:
+               case OP_READ:
+               case OP_LOAD:
                        RHS(ptr, 0) = flatten(state, first, RHS(ptr, 0));
-                       return MISC(ptr, 0);
-                       break;
-               case OP_LAND:
-                       ptr = flatten_land(state, first, ptr);
-                       break;
-               case OP_LOR:
-                       ptr = flatten_lor(state, first, ptr);
+                       use_triple(RHS(ptr, 0), ptr);
                        break;
-               case OP_COND:
-                       ptr = flatten_cond(state, first, ptr);
+               case OP_WRITE:
+                       ptr = flatten_generic(state, first, ptr, 1);
+                       MISC(ptr, 0) = flatten(state, first, MISC(ptr, 0));
+                       use_triple(MISC(ptr, 0), ptr);
                        break;
-               case OP_CALL:
-                       ptr = flatten_call(state, first, ptr);
+               case OP_BRANCH:
+                       use_triple(TARG(ptr, 0), ptr);
                        break;
-               case OP_READ:
-               case OP_LOAD:
+               case OP_CBRANCH:
                        RHS(ptr, 0) = flatten(state, first, RHS(ptr, 0));
                        use_triple(RHS(ptr, 0), ptr);
+                       use_triple(TARG(ptr, 0), ptr);
+                       insert_triple(state, first, ptr);
+                       ptr->id |= TRIPLE_FLAG_FLATTENED;
+                       ptr->id &= ~TRIPLE_FLAG_LOCAL;
+                       if (ptr->next != ptr) {
+                               use_triple(ptr->next, ptr);
+                       }
                        break;
-               case OP_BRANCH:
+               case OP_CALL:
+                       MISC(ptr, 0) = flatten(state, first, MISC(ptr, 0));
+                       use_triple(MISC(ptr, 0), ptr);
                        use_triple(TARG(ptr, 0), ptr);
-                       if (TRIPLE_RHS(ptr->sizes)) {
-                               use_triple(RHS(ptr, 0), ptr);
-                               if (ptr->next != ptr) {
-                                       use_triple(ptr->next, ptr);
-                               }
+                       insert_triple(state, first, ptr);
+                       ptr->id |= TRIPLE_FLAG_FLATTENED;
+                       ptr->id &= ~TRIPLE_FLAG_LOCAL;
+                       if (ptr->next != ptr) {
+                               use_triple(ptr->next, ptr);
                        }
                        break;
+               case OP_RET:
+                       RHS(ptr, 0) = flatten(state, first, RHS(ptr, 0));
+                       use_triple(RHS(ptr, 0), ptr);
+                       break;
                case OP_BLOBCONST:
-                       insert_triple(state, first, ptr);
+                       insert_triple(state, state->global_pool, ptr);
                        ptr->id |= TRIPLE_FLAG_FLATTENED;
+                       ptr->id &= ~TRIPLE_FLAG_LOCAL;
                        ptr = triple(state, OP_SDECL, ptr->type, ptr, 0);
                        use_triple(MISC(ptr, 0), ptr);
                        break;
@@ -4890,32 +7979,66 @@ static struct triple *flatten(
                        free_triple(state, orig_ptr);
                        break;
                case OP_DOT:
-               {
-                       struct triple *base;
-                       base = RHS(ptr, 0);
-                       base = flatten(state, first, base);
-                       if (base->op == OP_VAL_VEC) {
-                               ptr = struct_field(state, base, ptr->u.field);
+                       if (RHS(ptr, 0)->op == OP_DEREF) {
+                               struct triple *base, *left;
+                               ulong_t offset;
+                               base = MISC(ptr, 0);
+                               offset = bits_to_bytes(field_offset(state, base->type, ptr->u.field));
+                               left = RHS(base, 0);
+                               ptr = triple(state, OP_ADD, left->type,
+                                       read_expr(state, left),
+                                       int_const(state, &ulong_type, offset));
+                               free_triple(state, base);
+                       }
+                       else {
+                               ptr = flatten_part(state, first, ptr);
+                       }
+                       break;
+               case OP_INDEX:
+                       if (RHS(ptr, 0)->op == OP_DEREF) {
+                               struct triple *base, *left;
+                               ulong_t offset;
+                               base = MISC(ptr, 0);
+                               offset = bits_to_bytes(index_offset(state, base->type, ptr->u.cval));
+                               left = RHS(base, 0);
+                               ptr = triple(state, OP_ADD, left->type,
+                                       read_expr(state, left),
+                                       int_const(state, &long_type, offset));
+                               free_triple(state, base);
+                       }
+                       else {
+                               ptr = flatten_part(state, first, ptr);
                        }
                        break;
-               }
-               case OP_ADDRCONST:
-               case OP_SDECL:
                case OP_PIECE:
+                       ptr = flatten_part(state, first, ptr);
+                       use_triple(ptr, MISC(ptr, 0));
+                       break;
+               case OP_ADDRCONST:
                        MISC(ptr, 0) = flatten(state, first, MISC(ptr, 0));
                        use_triple(MISC(ptr, 0), ptr);
                        break;
+               case OP_SDECL:
+                       first = state->global_pool;
+                       MISC(ptr, 0) = flatten(state, first, MISC(ptr, 0));
+                       use_triple(MISC(ptr, 0), ptr);
+                       insert_triple(state, first, ptr);
+                       ptr->id |= TRIPLE_FLAG_FLATTENED;
+                       ptr->id &= ~TRIPLE_FLAG_LOCAL;
+                       return ptr;
                case OP_ADECL:
+                       ptr = flatten_generic(state, first, ptr, 0);
                        break;
                default:
                        /* Flatten the easy cases we don't override */
-                       ptr = flatten_generic(state, first, ptr);
+                       ptr = flatten_generic(state, first, ptr, 0);
                        break;
                }
        } while(ptr && (ptr != orig_ptr));
-       if (ptr) {
+       if (ptr && !(ptr->id & TRIPLE_FLAG_FLATTENED)) {
                insert_triple(state, first, ptr);
                ptr->id |= TRIPLE_FLAG_FLATTENED;
+               ptr->id &= ~TRIPLE_FLAG_LOCAL;
        }
        return ptr;
 }
@@ -4971,21 +8094,71 @@ static int replace_lhs_use(struct compile_state *state,
        return found;
 }
 
+static int replace_misc_use(struct compile_state *state,
+       struct triple *orig, struct triple *new, struct triple *use)
+{
+       struct triple **expr;
+       int found;
+       found = 0;
+       expr = triple_misc(state, use, 0);
+       for(;expr; expr = triple_misc(state, use, expr)) {
+               if (*expr == orig) {
+                       *expr = new;
+                       found = 1;
+               }
+       }
+       if (found) {
+               unuse_triple(orig, use);
+               use_triple(new, use);
+       }
+       return found;
+}
+
+static int replace_targ_use(struct compile_state *state,
+       struct triple *orig, struct triple *new, struct triple *use)
+{
+       struct triple **expr;
+       int found;
+       found = 0;
+       expr = triple_targ(state, use, 0);
+       for(;expr; expr = triple_targ(state, use, expr)) {
+               if (*expr == orig) {
+                       *expr = new;
+                       found = 1;
+               }
+       }
+       if (found) {
+               unuse_triple(orig, use);
+               use_triple(new, use);
+       }
+       return found;
+}
+
+static void replace_use(struct compile_state *state,
+       struct triple *orig, struct triple *new, struct triple *use)
+{
+       int found;
+       found = 0;
+       found |= replace_rhs_use(state, orig, new, use);
+       found |= replace_lhs_use(state, orig, new, use);
+       found |= replace_misc_use(state, orig, new, use);
+       found |= replace_targ_use(state, orig, new, use);
+       if (!found) {
+               internal_error(state, use, "use without use");
+       }
+}
+
 static void propogate_use(struct compile_state *state,
        struct triple *orig, struct triple *new)
 {
        struct triple_set *user, *next;
        for(user = orig->use; user; user = next) {
-               struct triple *use;
-               int found;
+               /* Careful replace_use modifies the use chain and
+                * removes use.  So we must get a copy of the next
+                * entry early.
+                */
                next = user->next;
-               use = user->member;
-               found = 0;
-               found |= replace_rhs_use(state, orig, new, use);
-               found |= replace_lhs_use(state, orig, new, use);
-               if (!found) {
-                       internal_error(state, use, "use without use");
-               }
+               replace_use(state, orig, new, user->member);
        }
        if (orig->use) {
                internal_error(state, orig, "used after propogate_use");
@@ -4997,6 +8170,15 @@ static void propogate_use(struct compile_state *state,
  * ===========================
  */
 
+static struct triple *mk_cast_expr(
+       struct compile_state *state, struct type *type, struct triple *expr)
+{
+       struct triple *def;
+       def = read_expr(state, expr);
+       def = triple(state, OP_CONVERT, type, def, 0);
+       return def;
+}
+
 static struct triple *mk_add_expr(
        struct compile_state *state, struct triple *left, struct triple *right)
 {
@@ -5012,12 +8194,21 @@ static struct triple *mk_add_expr(
        right = read_expr(state, right);
        result_type = ptr_arithmetic_result(state, left, right);
        if (is_pointer(left)) {
-               right = triple(state, 
-                       is_signed(right->type)? OP_SMUL : OP_UMUL, 
-                       &ulong_type, 
-                       right, 
-                       int_const(state, &ulong_type, 
-                               size_of(state, left->type->left)));
+               struct type *ptr_math;
+               int op;
+               if (is_signed(right->type)) {
+                       ptr_math = &long_type;
+                       op = OP_SMUL;
+               } else {
+                       ptr_math = &ulong_type;
+                       op = OP_UMUL;
+               }
+               if (!equiv_types(right->type, ptr_math)) {
+                       right = mk_cast_expr(state, ptr_math, right);
+               }
+               right = triple(state, op, ptr_math, right,
+                       int_const(state, ptr_math,
+                               size_of_in_bytes(state, left->type->left)));
        }
        return triple(state, OP_ADD, result_type, left, right);
 }
@@ -5030,12 +8221,21 @@ static struct triple *mk_sub_expr(
        left  = read_expr(state, left);
        right = read_expr(state, right);
        if (is_pointer(left)) {
-               right = triple(state, 
-                       is_signed(right->type)? OP_SMUL : OP_UMUL, 
-                       &ulong_type, 
-                       right, 
-                       int_const(state, &ulong_type, 
-                               size_of(state, left->type->left)));
+               struct type *ptr_math;
+               int op;
+               if (is_signed(right->type)) {
+                       ptr_math = &long_type;
+                       op = OP_SMUL;
+               } else {
+                       ptr_math = &ulong_type;
+                       op = OP_UMUL;
+               }
+               if (!equiv_types(right->type, ptr_math)) {
+                       right = mk_cast_expr(state, ptr_math, right);
+               }
+               right = triple(state, op, ptr_math, right,
+                       int_const(state, ptr_math,
+                               size_of_in_bytes(state, left->type->left)));
        }
        return triple(state, OP_SUB, result_type, left, right);
 }
@@ -5080,7 +8280,7 @@ static struct triple *mk_post_dec_expr(
        struct triple *val;
        lvalue(state, def);
        val = read_expr(state, def);
-       return triple(state, OP_VAL, def->type, 
+       return triple(state, OP_VAL, def->type,
                write_expr(state, def,
                        mk_sub_expr(state, val, int_const(state, &int_type, 1)))
                , val);
@@ -5097,6 +8297,7 @@ static struct triple *mk_subscript_expr(
        return mk_deref_expr(state, mk_add_expr(state, left, right));
 }
 
+
 /*
  * Compile time evaluation
  * ===========================
@@ -5106,11 +8307,24 @@ static int is_const(struct triple *ins)
        return IS_CONST_OP(ins->op);
 }
 
-static int constants_equal(struct compile_state *state, 
+static int is_simple_const(struct triple *ins)
+{
+       /* Is this a constant that u.cval has the value.
+        * Or equivalently is this a constant that read_const
+        * works on.
+        * So far only OP_INTCONST qualifies.
+        */
+       return (ins->op == OP_INTCONST);
+}
+
+static int constants_equal(struct compile_state *state,
        struct triple *left, struct triple *right)
 {
        int equal;
-       if (!is_const(left) || !is_const(right)) {
+       if ((left->op == OP_UNKNOWNVAL) || (right->op == OP_UNKNOWNVAL)) {
+               equal = 0;
+       }
+       else if (!is_const(left) || !is_const(right)) {
                equal = 0;
        }
        else if (left->op != right->op) {
@@ -5129,13 +8343,14 @@ static int constants_equal(struct compile_state *state,
                        break;
                case OP_BLOBCONST:
                {
-                       size_t lsize, rsize;
+                       size_t lsize, rsize, bytes;
                        lsize = size_of(state, left->type);
                        rsize = size_of(state, right->type);
                        if (lsize != rsize) {
                                break;
                        }
-                       if (memcmp(left->u.blob, right->u.blob, lsize) == 0) {
+                       bytes = bits_to_bytes(lsize);
+                       if (memcmp(left->u.blob, right->u.blob, bytes) == 0) {
                                equal = 1;
                        }
                        break;
@@ -5156,13 +8371,32 @@ static int constants_equal(struct compile_state *state,
 
 static int is_zero(struct triple *ins)
 {
-       return is_const(ins) && (ins->u.cval == 0);
+       return is_simple_const(ins) && (ins->u.cval == 0);
 }
 
 static int is_one(struct triple *ins)
 {
-       return is_const(ins) && (ins->u.cval == 1);
+       return is_simple_const(ins) && (ins->u.cval == 1);
+}
+
+#if DEBUG_ROMCC_WARNING
+static long_t bit_count(ulong_t value)
+{
+       int count;
+       int i;
+       count = 0;
+       for(i = (sizeof(ulong_t)*8) -1; i >= 0; i--) {
+               ulong_t mask;
+               mask = 1;
+               mask <<= i;
+               if (value & mask) {
+                       count++;
+               }
+       }
+       return count;
+
 }
+#endif
 
 static long_t bsr(ulong_t value)
 {
@@ -5192,14 +8426,14 @@ static long_t bsf(ulong_t value)
        return -1;
 }
 
-static long_t log2(ulong_t value)
+static long_t ilog2(ulong_t value)
 {
        return bsr(value);
 }
 
 static long_t tlog2(struct triple *ins)
 {
-       return log2(ins->u.cval);
+       return ilog2(ins->u.cval);
 }
 
 static int is_pow2(struct triple *ins)
@@ -5210,7 +8444,7 @@ static int is_pow2(struct triple *ins)
                return 0;
        }
        value = ins->u.cval;
-       log = log2(value);
+       log = ilog2(value);
        if (log == -1) {
                return 0;
        }
@@ -5220,35 +8454,145 @@ static int is_pow2(struct triple *ins)
 }
 
 static ulong_t read_const(struct compile_state *state,
-       struct triple *ins, struct triple **expr)
+       struct triple *ins, struct triple *rhs)
 {
-       struct triple *rhs;
-       rhs = *expr;
        switch(rhs->type->type &TYPE_MASK) {
-       case TYPE_CHAR:   
+       case TYPE_CHAR:
        case TYPE_SHORT:
        case TYPE_INT:
        case TYPE_LONG:
-       case TYPE_UCHAR:   
-       case TYPE_USHORT:  
+       case TYPE_UCHAR:
+       case TYPE_USHORT:
        case TYPE_UINT:
        case TYPE_ULONG:
        case TYPE_POINTER:
+       case TYPE_BITFIELD:
                break;
        default:
-               internal_error(state, rhs, "bad type to read_const\n");
+               fprintf(state->errout, "type: ");
+               name_of(state->errout, rhs->type);
+               fprintf(state->errout, "\n");
+               internal_warning(state, rhs, "bad type to read_const");
                break;
        }
+       if (!is_simple_const(rhs)) {
+               internal_error(state, rhs, "bad op to read_const");
+       }
        return rhs->u.cval;
 }
 
-static long_t read_sconst(struct triple *ins, struct triple **expr)
+static long_t read_sconst(struct compile_state *state,
+       struct triple *ins, struct triple *rhs)
 {
-       struct triple *rhs;
-       rhs = *expr;
        return (long_t)(rhs->u.cval);
 }
 
+int const_ltrue(struct compile_state *state, struct triple *ins, struct triple *rhs)
+{
+       if (!is_const(rhs)) {
+               internal_error(state, 0, "non const passed to const_true");
+       }
+       return !is_zero(rhs);
+}
+
+int const_eq(struct compile_state *state, struct triple *ins,
+       struct triple *left, struct triple *right)
+{
+       int result;
+       if (!is_const(left) || !is_const(right)) {
+               internal_warning(state, ins, "non const passed to const_eq");
+               result = -1;
+       }
+       else if (left == right) {
+               result = 1;
+       }
+       else if (is_simple_const(left) && is_simple_const(right)) {
+               ulong_t lval, rval;
+               lval = read_const(state, ins, left);
+               rval = read_const(state, ins, right);
+               result = (lval == rval);
+       }
+       else if ((left->op == OP_ADDRCONST) &&
+               (right->op == OP_ADDRCONST)) {
+               result = (MISC(left, 0) == MISC(right, 0)) &&
+                       (left->u.cval == right->u.cval);
+       }
+       else {
+               internal_warning(state, ins, "incomparable constants passed to const_eq");
+               result = -1;
+       }
+       return result;
+
+}
+
+int const_ucmp(struct compile_state *state, struct triple *ins,
+       struct triple *left, struct triple *right)
+{
+       int result;
+       if (!is_const(left) || !is_const(right)) {
+               internal_warning(state, ins, "non const past to const_ucmp");
+               result = -2;
+       }
+       else if (left == right) {
+               result = 0;
+       }
+       else if (is_simple_const(left) && is_simple_const(right)) {
+               ulong_t lval, rval;
+               lval = read_const(state, ins, left);
+               rval = read_const(state, ins, right);
+               result = 0;
+               if (lval > rval) {
+                       result = 1;
+               } else if (rval > lval) {
+                       result = -1;
+               }
+       }
+       else if ((left->op == OP_ADDRCONST) &&
+               (right->op == OP_ADDRCONST) &&
+               (MISC(left, 0) == MISC(right, 0))) {
+               result = 0;
+               if (left->u.cval > right->u.cval) {
+                       result = 1;
+               } else if (left->u.cval < right->u.cval) {
+                       result = -1;
+               }
+       }
+       else {
+               internal_warning(state, ins, "incomparable constants passed to const_ucmp");
+               result = -2;
+       }
+       return result;
+}
+
+int const_scmp(struct compile_state *state, struct triple *ins,
+       struct triple *left, struct triple *right)
+{
+       int result;
+       if (!is_const(left) || !is_const(right)) {
+               internal_warning(state, ins, "non const past to ucmp_const");
+               result = -2;
+       }
+       else if (left == right) {
+               result = 0;
+       }
+       else if (is_simple_const(left) && is_simple_const(right)) {
+               long_t lval, rval;
+               lval = read_sconst(state, ins, left);
+               rval = read_sconst(state, ins, right);
+               result = 0;
+               if (lval > rval) {
+                       result = 1;
+               } else if (rval > lval) {
+                       result = -1;
+               }
+       }
+       else {
+               internal_warning(state, ins, "incomparable constants passed to const_scmp");
+               result = -2;
+       }
+       return result;
+}
+
 static void unuse_rhs(struct compile_state *state, struct triple *ins)
 {
        struct triple **expr;
@@ -5271,16 +8615,51 @@ static void unuse_lhs(struct compile_state *state, struct triple *ins)
        }
 }
 
-static void check_lhs(struct compile_state *state, struct triple *ins)
+#if DEBUG_ROMCC_WARNING
+static void unuse_misc(struct compile_state *state, struct triple *ins)
 {
        struct triple **expr;
-       expr = triple_lhs(state, ins, 0);
-       for(;expr;expr = triple_lhs(state, ins, expr)) {
-               internal_error(state, ins, "unexpected lhs");
+       expr = triple_misc(state, ins, 0);
+       for(;expr;expr = triple_misc(state, ins, expr)) {
+               unuse_triple(*expr, ins);
+               *expr = 0;
        }
-       
 }
-static void check_targ(struct compile_state *state, struct triple *ins)
+
+static void unuse_targ(struct compile_state *state, struct triple *ins)
+{
+       int i;
+       struct triple **slot;
+       slot = &TARG(ins, 0);
+       for(i = 0; i < ins->targ; i++) {
+               unuse_triple(slot[i], ins);
+               slot[i] = 0;
+       }
+}
+
+static void check_lhs(struct compile_state *state, struct triple *ins)
+{
+       struct triple **expr;
+       expr = triple_lhs(state, ins, 0);
+       for(;expr;expr = triple_lhs(state, ins, expr)) {
+               internal_error(state, ins, "unexpected lhs");
+       }
+
+}
+#endif
+
+static void check_misc(struct compile_state *state, struct triple *ins)
+{
+       struct triple **expr;
+       expr = triple_misc(state, ins, 0);
+       for(;expr;expr = triple_misc(state, ins, expr)) {
+               if (*expr) {
+                       internal_error(state, ins, "unexpected misc");
+               }
+       }
+}
+
+static void check_targ(struct compile_state *state, struct triple *ins)
 {
        struct triple **expr;
        expr = triple_targ(state, ins, 0);
@@ -5296,156 +8675,604 @@ static void wipe_ins(struct compile_state *state, struct triple *ins)
         * in all instructions to hold all others.
         */
        check_targ(state, ins);
+       check_misc(state, ins);
+       unuse_rhs(state, ins);
+       unuse_lhs(state, ins);
+       ins->lhs  = 0;
+       ins->rhs  = 0;
+       ins->misc = 0;
+       ins->targ = 0;
+}
+
+#if DEBUG_ROMCC_WARNING
+static void wipe_branch(struct compile_state *state, struct triple *ins)
+{
+       /* Becareful which instructions you replace the wiped
+        * instruction with, as there are not enough slots
+        * in all instructions to hold all others.
+        */
        unuse_rhs(state, ins);
        unuse_lhs(state, ins);
+       unuse_misc(state, ins);
+       unuse_targ(state, ins);
+       ins->lhs  = 0;
+       ins->rhs  = 0;
+       ins->misc = 0;
+       ins->targ = 0;
 }
+#endif
 
-static void mkcopy(struct compile_state *state, 
+static void mkcopy(struct compile_state *state,
        struct triple *ins, struct triple *rhs)
 {
+       struct block *block;
+       if (!equiv_types(ins->type, rhs->type)) {
+               FILE *fp = state->errout;
+               fprintf(fp, "src type: ");
+               name_of(fp, rhs->type);
+               fprintf(fp, "\ndst type: ");
+               name_of(fp, ins->type);
+               fprintf(fp, "\n");
+               internal_error(state, ins, "mkcopy type mismatch");
+       }
+       block = block_of_triple(state, ins);
        wipe_ins(state, ins);
        ins->op = OP_COPY;
-       ins->sizes = TRIPLE_SIZES(0, 1, 0, 0);
+       ins->rhs  = 1;
+       ins->u.block = block;
        RHS(ins, 0) = rhs;
        use_triple(RHS(ins, 0), ins);
 }
 
-static void mkconst(struct compile_state *state, 
+static void mkconst(struct compile_state *state,
        struct triple *ins, ulong_t value)
 {
        if (!is_integral(ins) && !is_pointer(ins)) {
-               internal_error(state, ins, "unknown type to make constant\n");
+               fprintf(state->errout, "type: ");
+               name_of(state->errout, ins->type);
+               fprintf(state->errout, "\n");
+               internal_error(state, ins, "unknown type to make constant value: %ld",
+                       value);
        }
        wipe_ins(state, ins);
        ins->op = OP_INTCONST;
-       ins->sizes = TRIPLE_SIZES(0, 0, 0, 0);
        ins->u.cval = value;
 }
 
 static void mkaddr_const(struct compile_state *state,
        struct triple *ins, struct triple *sdecl, ulong_t value)
 {
+       if ((sdecl->op != OP_SDECL) && (sdecl->op != OP_LABEL)) {
+               internal_error(state, ins, "bad base for addrconst");
+       }
        wipe_ins(state, ins);
        ins->op = OP_ADDRCONST;
-       ins->sizes = TRIPLE_SIZES(0, 0, 1, 0);
+       ins->misc = 1;
        MISC(ins, 0) = sdecl;
        ins->u.cval = value;
        use_triple(sdecl, ins);
 }
 
-/* Transform multicomponent variables into simple register variables */
-static void flatten_structures(struct compile_state *state)
+#if DEBUG_DECOMPOSE_PRINT_TUPLES
+static void print_tuple(struct compile_state *state,
+       struct triple *ins, struct triple *tuple)
 {
-       struct triple *ins, *first;
-       first = RHS(state->main_function, 0);
+       FILE *fp = state->dbgout;
+       fprintf(fp, "%5s %p tuple: %p ", tops(ins->op), ins, tuple);
+       name_of(fp, tuple->type);
+       if (tuple->lhs > 0) {
+               fprintf(fp, " lhs: ");
+               name_of(fp, LHS(tuple, 0)->type);
+       }
+       fprintf(fp, "\n");
+
+}
+#endif
+
+static struct triple *decompose_with_tuple(struct compile_state *state,
+       struct triple *ins, struct triple *tuple)
+{
+       struct triple *next;
+       next = ins->next;
+       flatten(state, next, tuple);
+#if DEBUG_DECOMPOSE_PRINT_TUPLES
+       print_tuple(state, ins, tuple);
+#endif
+
+       if (!is_compound_type(tuple->type) && (tuple->lhs > 0)) {
+               struct triple *tmp;
+               if (tuple->lhs != 1) {
+                       internal_error(state, tuple, "plain type in multiple registers?");
+               }
+               tmp = LHS(tuple, 0);
+               release_triple(state, tuple);
+               tuple = tmp;
+       }
+
+       propogate_use(state, ins, tuple);
+       release_triple(state, ins);
+
+       return next;
+}
+
+static struct triple *decompose_unknownval(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple;
+       ulong_t i;
+
+#if DEBUG_DECOMPOSE_HIRES
+       FILE *fp = state->dbgout;
+       fprintf(fp, "unknown type: ");
+       name_of(fp, ins->type);
+       fprintf(fp, "\n");
+#endif
+
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, ins->type, -1, -1,
+               ins->occurance);
+
+       for(i = 0; i < tuple->lhs; i++) {
+               struct type *piece_type;
+               struct triple *unknown;
+
+               piece_type = reg_type(state, ins->type, i * REG_SIZEOF_REG);
+               get_occurance(tuple->occurance);
+               unknown = alloc_triple(state, OP_UNKNOWNVAL, piece_type, 0, 0,
+                       tuple->occurance);
+               LHS(tuple, i) = unknown;
+       }
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+
+static struct triple *decompose_read(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple, *lval;
+       ulong_t i;
+
+       lval = RHS(ins, 0);
+
+       if (lval->op == OP_PIECE) {
+               return ins->next;
+       }
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, lval->type, -1, -1,
+               ins->occurance);
+
+       if ((tuple->lhs != lval->lhs) &&
+               (!triple_is_def(state, lval) || (tuple->lhs != 1)))
+       {
+               internal_error(state, ins, "lhs size inconsistency?");
+       }
+       for(i = 0; i < tuple->lhs; i++) {
+               struct triple *piece, *read, *bitref;
+               if ((i != 0) || !triple_is_def(state, lval)) {
+                       piece = LHS(lval, i);
+               } else {
+                       piece = lval;
+               }
+
+               /* See if the piece is really a bitref */
+               bitref = 0;
+               if (piece->op == OP_BITREF) {
+                       bitref = piece;
+                       piece = RHS(bitref, 0);
+               }
+
+               get_occurance(tuple->occurance);
+               read = alloc_triple(state, OP_READ, piece->type, -1, -1,
+                       tuple->occurance);
+               RHS(read, 0) = piece;
+
+               if (bitref) {
+                       struct triple *extract;
+                       int op;
+                       if (is_signed(bitref->type->left)) {
+                               op = OP_SEXTRACT;
+                       } else {
+                               op = OP_UEXTRACT;
+                       }
+                       get_occurance(tuple->occurance);
+                       extract = alloc_triple(state, op, bitref->type, -1, -1,
+                               tuple->occurance);
+                       RHS(extract, 0) = read;
+                       extract->u.bitfield.size   = bitref->u.bitfield.size;
+                       extract->u.bitfield.offset = bitref->u.bitfield.offset;
+
+                       read = extract;
+               }
+
+               LHS(tuple, i) = read;
+       }
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+static struct triple *decompose_write(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple, *lval, *val;
+       ulong_t i;
+
+       lval = MISC(ins, 0);
+       val = RHS(ins, 0);
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, ins->type, -1, -1,
+               ins->occurance);
+
+       if ((tuple->lhs != lval->lhs) &&
+               (!triple_is_def(state, lval) || tuple->lhs != 1))
+       {
+               internal_error(state, ins, "lhs size inconsistency?");
+       }
+       for(i = 0; i < tuple->lhs; i++) {
+               struct triple *piece, *write, *pval, *bitref;
+               if ((i != 0) || !triple_is_def(state, lval)) {
+                       piece = LHS(lval, i);
+               } else {
+                       piece = lval;
+               }
+               if ((i == 0) && (tuple->lhs == 1) && (val->lhs == 0)) {
+                       pval = val;
+               }
+               else {
+                       if (i > val->lhs) {
+                               internal_error(state, ins, "lhs size inconsistency?");
+                       }
+                       pval = LHS(val, i);
+               }
+
+               /* See if the piece is really a bitref */
+               bitref = 0;
+               if (piece->op == OP_BITREF) {
+                       struct triple *read, *deposit;
+                       bitref = piece;
+                       piece = RHS(bitref, 0);
+
+                       /* Read the destination register */
+                       get_occurance(tuple->occurance);
+                       read = alloc_triple(state, OP_READ, piece->type, -1, -1,
+                               tuple->occurance);
+                       RHS(read, 0) = piece;
+
+                       /* Deposit the new bitfield value */
+                       get_occurance(tuple->occurance);
+                       deposit = alloc_triple(state, OP_DEPOSIT, piece->type, -1, -1,
+                               tuple->occurance);
+                       RHS(deposit, 0) = read;
+                       RHS(deposit, 1) = pval;
+                       deposit->u.bitfield.size   = bitref->u.bitfield.size;
+                       deposit->u.bitfield.offset = bitref->u.bitfield.offset;
+
+                       /* Now write the newly generated value */
+                       pval = deposit;
+               }
+
+               get_occurance(tuple->occurance);
+               write = alloc_triple(state, OP_WRITE, piece->type, -1, -1,
+                       tuple->occurance);
+               MISC(write, 0) = piece;
+               RHS(write, 0) = pval;
+               LHS(tuple, i) = write;
+       }
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+struct decompose_load_info {
+       struct occurance *occurance;
+       struct triple *lval;
+       struct triple *tuple;
+};
+static void decompose_load_cb(struct compile_state *state,
+       struct type *type, size_t reg_offset, size_t mem_offset, void *arg)
+{
+       struct decompose_load_info *info = arg;
+       struct triple *load;
+
+       if (reg_offset > info->tuple->lhs) {
+               internal_error(state, info->tuple, "lhs to small?");
+       }
+       get_occurance(info->occurance);
+       load = alloc_triple(state, OP_LOAD, type, -1, -1, info->occurance);
+       RHS(load, 0) = mk_addr_expr(state, info->lval, mem_offset);
+       LHS(info->tuple, reg_offset/REG_SIZEOF_REG) = load;
+}
+
+static struct triple *decompose_load(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple;
+       struct decompose_load_info info;
+
+       if (!is_compound_type(ins->type)) {
+               return ins->next;
+       }
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, ins->type, -1, -1,
+               ins->occurance);
+
+       info.occurance = ins->occurance;
+       info.lval      = RHS(ins, 0);
+       info.tuple     = tuple;
+       walk_type_fields(state, ins->type, 0, 0, decompose_load_cb, &info);
+
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+
+struct decompose_store_info {
+       struct occurance *occurance;
+       struct triple *lval;
+       struct triple *val;
+       struct triple *tuple;
+};
+static void decompose_store_cb(struct compile_state *state,
+       struct type *type, size_t reg_offset, size_t mem_offset, void *arg)
+{
+       struct decompose_store_info *info = arg;
+       struct triple *store;
+
+       if (reg_offset > info->tuple->lhs) {
+               internal_error(state, info->tuple, "lhs to small?");
+       }
+       get_occurance(info->occurance);
+       store = alloc_triple(state, OP_STORE, type, -1, -1, info->occurance);
+       RHS(store, 0) = mk_addr_expr(state, info->lval, mem_offset);
+       RHS(store, 1) = LHS(info->val, reg_offset);
+       LHS(info->tuple, reg_offset/REG_SIZEOF_REG) = store;
+}
+
+static struct triple *decompose_store(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple;
+       struct decompose_store_info info;
+
+       if (!is_compound_type(ins->type)) {
+               return ins->next;
+       }
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, ins->type, -1, -1,
+               ins->occurance);
+
+       info.occurance = ins->occurance;
+       info.lval      = RHS(ins, 0);
+       info.val       = RHS(ins, 1);
+       info.tuple     = tuple;
+       walk_type_fields(state, ins->type, 0, 0, decompose_store_cb, &info);
+
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+static struct triple *decompose_dot(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple, *lval;
+       struct type *type;
+       size_t reg_offset;
+       int i, idx;
+
+       lval = MISC(ins, 0);
+       reg_offset = field_reg_offset(state, lval->type, ins->u.field);
+       idx  = reg_offset/REG_SIZEOF_REG;
+       type = field_type(state, lval->type, ins->u.field);
+#if DEBUG_DECOMPOSE_HIRES
+       {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "field type: ");
+               name_of(fp, type);
+               fprintf(fp, "\n");
+       }
+#endif
+
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, type, -1, -1,
+               ins->occurance);
+
+       if (((ins->type->type & TYPE_MASK) == TYPE_BITFIELD) &&
+               (tuple->lhs != 1))
+       {
+               internal_error(state, ins, "multi register bitfield?");
+       }
+
+       for(i = 0; i < tuple->lhs; i++, idx++) {
+               struct triple *piece;
+               if (!triple_is_def(state, lval)) {
+                       if (idx > lval->lhs) {
+                               internal_error(state, ins, "inconsistent lhs count");
+                       }
+                       piece = LHS(lval, idx);
+               } else {
+                       if (idx != 0) {
+                               internal_error(state, ins, "bad reg_offset into def");
+                       }
+                       if (i != 0) {
+                               internal_error(state, ins, "bad reg count from def");
+                       }
+                       piece = lval;
+               }
+
+               /* Remember the offset of the bitfield */
+               if ((type->type & TYPE_MASK) == TYPE_BITFIELD) {
+                       get_occurance(ins->occurance);
+                       piece = build_triple(state, OP_BITREF, type, piece, 0,
+                               ins->occurance);
+                       piece->u.bitfield.size   = size_of(state, type);
+                       piece->u.bitfield.offset = reg_offset % REG_SIZEOF_REG;
+               }
+               else if ((reg_offset % REG_SIZEOF_REG) != 0) {
+                       internal_error(state, ins,
+                               "request for a nonbitfield sub register?");
+               }
+
+               LHS(tuple, i) = piece;
+       }
+
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+static struct triple *decompose_index(struct compile_state *state,
+       struct triple *ins)
+{
+       struct triple *tuple, *lval;
+       struct type *type;
+       int i, idx;
+
+       lval = MISC(ins, 0);
+       idx = index_reg_offset(state, lval->type, ins->u.cval)/REG_SIZEOF_REG;
+       type = index_type(state, lval->type, ins->u.cval);
+#if DEBUG_DECOMPOSE_HIRES
+{
+       FILE *fp = state->dbgout;
+       fprintf(fp, "index type: ");
+       name_of(fp, type);
+       fprintf(fp, "\n");
+}
+#endif
+
+       get_occurance(ins->occurance);
+       tuple = alloc_triple(state, OP_TUPLE, type, -1, -1,
+               ins->occurance);
+
+       for(i = 0; i < tuple->lhs; i++, idx++) {
+               struct triple *piece;
+               if (!triple_is_def(state, lval)) {
+                       if (idx > lval->lhs) {
+                               internal_error(state, ins, "inconsistent lhs count");
+                       }
+                       piece = LHS(lval, idx);
+               } else {
+                       if (idx != 0) {
+                               internal_error(state, ins, "bad reg_offset into def");
+                       }
+                       if (i != 0) {
+                               internal_error(state, ins, "bad reg count from def");
+                       }
+                       piece = lval;
+               }
+               LHS(tuple, i) = piece;
+       }
+
+       return decompose_with_tuple(state, ins, tuple);
+}
+
+static void decompose_compound_types(struct compile_state *state)
+{
+       struct triple *ins, *next, *first;
+#if DEBUG_DECOMPOSE_HIRES
+       FILE *fp;
+       fp = state->dbgout;
+#endif
+       first = state->first;
        ins = first;
-       /* Pass one expand structure values into valvecs.
+
+       /* Pass one expand compound values into pseudo registers.
+        */
+       next = first;
+       do {
+               ins = next;
+               next = ins->next;
+               switch(ins->op) {
+               case OP_UNKNOWNVAL:
+                       next = decompose_unknownval(state, ins);
+                       break;
+
+               case OP_READ:
+                       next = decompose_read(state, ins);
+                       break;
+
+               case OP_WRITE:
+                       next = decompose_write(state, ins);
+                       break;
+
+
+               /* Be very careful with the load/store logic. These
+                * operations must convert from the in register layout
+                * to the in memory layout, which is nontrivial.
+                */
+               case OP_LOAD:
+                       next = decompose_load(state, ins);
+                       break;
+               case OP_STORE:
+                       next = decompose_store(state, ins);
+                       break;
+
+               case OP_DOT:
+                       next = decompose_dot(state, ins);
+                       break;
+               case OP_INDEX:
+                       next = decompose_index(state, ins);
+                       break;
+
+               }
+#if DEBUG_DECOMPOSE_HIRES
+               fprintf(fp, "decompose next: %p \n", next);
+               fflush(fp);
+               fprintf(fp, "next->op: %d %s\n",
+                       next->op, tops(next->op));
+               /* High resolution debugging mode */
+               print_triples(state);
+#endif
+       } while (next != first);
+
+       /* Pass two remove the tuples.
         */
        ins = first;
        do {
-               struct triple *next;
                next = ins->next;
-               if ((ins->type->type & TYPE_MASK) == TYPE_STRUCT) {
-                       if (ins->op == OP_VAL_VEC) {
-                               /* Do nothing */
-                       }
-                       else if ((ins->op == OP_LOAD) || (ins->op == OP_READ)) {
-                               struct triple *def, **vector;
-                               struct type *tptr;
-                               int op;
-                               ulong_t i;
-
-                               op = ins->op;
-                               def = RHS(ins, 0);
-                               next = alloc_triple(state, OP_VAL_VEC, ins->type, -1, -1,
-                                       ins->filename, ins->line, ins->col);
-
-                               vector = &RHS(next, 0);
-                               tptr = next->type->left;
-                               for(i = 0; i < next->type->elements; i++) {
-                                       struct triple *sfield;
-                                       struct type *mtype;
-                                       mtype = tptr;
-                                       if ((mtype->type & TYPE_MASK) == TYPE_PRODUCT) {
-                                               mtype = mtype->left;
-                                       }
-                                       sfield = deref_field(state, def, mtype->field_ident);
-                                       
-                                       vector[i] = triple(
-                                               state, op, mtype, sfield, 0);
-                                       vector[i]->filename = next->filename;
-                                       vector[i]->line = next->line;
-                                       vector[i]->col = next->col;
-                                       tptr = tptr->right;
-                               }
-                               propogate_use(state, ins, next);
-                               flatten(state, ins, next);
-                               free_triple(state, ins);
-                       }
-                       else if ((ins->op == OP_STORE) || (ins->op == OP_WRITE)) {
-                               struct triple *src, *dst, **vector;
-                               struct type *tptr;
-                               int op;
-                               ulong_t i;
-
-                               op = ins->op;
-                               src = RHS(ins, 0);
-                               dst = LHS(ins, 0);
-                               next = alloc_triple(state, OP_VAL_VEC, ins->type, -1, -1,
-                                       ins->filename, ins->line, ins->col);
-                               
-                               vector = &RHS(next, 0);
-                               tptr = next->type->left;
-                               for(i = 0; i < ins->type->elements; i++) {
-                                       struct triple *dfield, *sfield;
-                                       struct type *mtype;
-                                       mtype = tptr;
-                                       if ((mtype->type & TYPE_MASK) == TYPE_PRODUCT) {
-                                               mtype = mtype->left;
-                                       }
-                                       sfield = deref_field(state, src, mtype->field_ident);
-                                       dfield = deref_field(state, dst, mtype->field_ident);
-                                       vector[i] = triple(
-                                               state, op, mtype, dfield, sfield);
-                                       vector[i]->filename = next->filename;
-                                       vector[i]->line = next->line;
-                                       vector[i]->col = next->col;
-                                       tptr = tptr->right;
-                               }
-                               propogate_use(state, ins, next);
-                               flatten(state, ins, next);
-                               free_triple(state, ins);
+               if (ins->op == OP_TUPLE) {
+                       if (ins->use) {
+                               internal_error(state, ins, "tuple used");
+                       }
+                       else {
+                               release_triple(state, ins);
                        }
                }
                ins = next;
        } while(ins != first);
-       /* Pass two flatten the valvecs.
-        */
        ins = first;
        do {
-               struct triple *next;
                next = ins->next;
-               if (ins->op == OP_VAL_VEC) {
-                       release_triple(state, ins);
-               } 
+               if (ins->op == OP_BITREF) {
+                       if (ins->use) {
+                               internal_error(state, ins, "bitref used");
+                       }
+                       else {
+                               release_triple(state, ins);
+                       }
+               }
                ins = next;
        } while(ins != first);
+
        /* Pass three verify the state and set ->id to 0.
         */
-       ins = first;
+       next = first;
        do {
+               ins = next;
+               next = ins->next;
                ins->id &= ~TRIPLE_FLAG_FLATTENED;
-               if ((ins->type->type & TYPE_MASK) == TYPE_STRUCT) {
-                       internal_error(state, 0, "STRUCT_TYPE remains?");
+               if (triple_stores_block(state, ins)) {
+                       ins->u.block = 0;
+               }
+               if (triple_is_def(state, ins)) {
+                       if (reg_size_of(state, ins->type) > REG_SIZEOF_REG) {
+                               internal_error(state, ins, "multi register value remains?");
+                       }
                }
                if (ins->op == OP_DOT) {
-                       internal_error(state, 0, "OP_DOT remains?");
+                       internal_error(state, ins, "OP_DOT remains?");
                }
-               if (ins->op == OP_VAL_VEC) {
-                       internal_error(state, 0, "OP_VAL_VEC remains?");
+               if (ins->op == OP_INDEX) {
+                       internal_error(state, ins, "OP_INDEX remains?");
                }
-               ins = ins->next;
-       } while(ins != first);
+               if (ins->op == OP_BITREF) {
+                       internal_error(state, ins, "OP_BITREF remains?");
+               }
+               if (ins->op == OP_TUPLE) {
+                       internal_error(state, ins, "OP_TUPLE remains?");
+               }
+       } while(next != first);
 }
 
 /* For those operations that cannot be simplified */
@@ -5464,8 +9291,8 @@ static void simplify_smul(struct compile_state *state, struct triple *ins)
        }
        if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
                long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
+               left  = read_sconst(state, ins, RHS(ins, 0));
+               right = read_sconst(state, ins, RHS(ins, 1));
                mkconst(state, ins, left * right);
        }
        else if (is_zero(RHS(ins, 1))) {
@@ -5478,7 +9305,7 @@ static void simplify_smul(struct compile_state *state, struct triple *ins)
                struct triple *val;
                val = int_const(state, ins->type, tlog2(RHS(ins, 1)));
                ins->op = OP_SL;
-               insert_triple(state, ins, val);
+               insert_triple(state, state->global_pool, val);
                unuse_triple(RHS(ins, 1), ins);
                use_triple(val, ins);
                RHS(ins, 1) = val;
@@ -5493,10 +9320,10 @@ static void simplify_umul(struct compile_state *state, struct triple *ins)
                RHS(ins, 0) = RHS(ins, 1);
                RHS(ins, 1) = tmp;
        }
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins, left * right);
        }
        else if (is_zero(RHS(ins, 1))) {
@@ -5509,7 +9336,7 @@ static void simplify_umul(struct compile_state *state, struct triple *ins)
                struct triple *val;
                val = int_const(state, ins->type, tlog2(RHS(ins, 1)));
                ins->op = OP_SL;
-               insert_triple(state, ins, val);
+               insert_triple(state, state->global_pool, val);
                unuse_triple(RHS(ins, 1), ins);
                use_triple(val, ins);
                RHS(ins, 1) = val;
@@ -5520,8 +9347,8 @@ static void simplify_sdiv(struct compile_state *state, struct triple *ins)
 {
        if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
                long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
+               left  = read_sconst(state, ins, RHS(ins, 0));
+               right = read_sconst(state, ins, RHS(ins, 1));
                mkconst(state, ins, left / right);
        }
        else if (is_zero(RHS(ins, 0))) {
@@ -5537,7 +9364,7 @@ static void simplify_sdiv(struct compile_state *state, struct triple *ins)
                struct triple *val;
                val = int_const(state, ins->type, tlog2(RHS(ins, 1)));
                ins->op = OP_SSR;
-               insert_triple(state, ins, val);
+               insert_triple(state, state->global_pool, val);
                unuse_triple(RHS(ins, 1), ins);
                use_triple(val, ins);
                RHS(ins, 1) = val;
@@ -5546,10 +9373,10 @@ static void simplify_sdiv(struct compile_state *state, struct triple *ins)
 
 static void simplify_udiv(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins, left / right);
        }
        else if (is_zero(RHS(ins, 0))) {
@@ -5565,7 +9392,7 @@ static void simplify_udiv(struct compile_state *state, struct triple *ins)
                struct triple *val;
                val = int_const(state, ins->type, tlog2(RHS(ins, 1)));
                ins->op = OP_USR;
-               insert_triple(state, ins, val);
+               insert_triple(state, state->global_pool, val);
                unuse_triple(RHS(ins, 1), ins);
                use_triple(val, ins);
                RHS(ins, 1) = val;
@@ -5574,10 +9401,10 @@ static void simplify_udiv(struct compile_state *state, struct triple *ins)
 
 static void simplify_smod(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                long_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins, left % right);
        }
        else if (is_zero(RHS(ins, 0))) {
@@ -5593,18 +9420,19 @@ static void simplify_smod(struct compile_state *state, struct triple *ins)
                struct triple *val;
                val = int_const(state, ins->type, RHS(ins, 1)->u.cval - 1);
                ins->op = OP_AND;
-               insert_triple(state, ins, val);
+               insert_triple(state, state->global_pool, val);
                unuse_triple(RHS(ins, 1), ins);
                use_triple(val, ins);
                RHS(ins, 1) = val;
        }
 }
+
 static void simplify_umod(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins, left % right);
        }
        else if (is_zero(RHS(ins, 0))) {
@@ -5620,7 +9448,7 @@ static void simplify_umod(struct compile_state *state, struct triple *ins)
                struct triple *val;
                val = int_const(state, ins->type, RHS(ins, 1)->u.cval - 1);
                ins->op = OP_AND;
-               insert_triple(state, ins, val);
+               insert_triple(state, state->global_pool, val);
                unuse_triple(RHS(ins, 1), ins);
                use_triple(val, ins);
                RHS(ins, 1) = val;
@@ -5636,14 +9464,14 @@ static void simplify_add(struct compile_state *state, struct triple *ins)
                RHS(ins, 0) = RHS(ins, 1);
                RHS(ins, 1) = tmp;
        }
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               if (!is_pointer(RHS(ins, 0))) {
+       if (is_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
+               if (RHS(ins, 0)->op == OP_INTCONST) {
                        ulong_t left, right;
-                       left  = read_const(state, ins, &RHS(ins, 0));
-                       right = read_const(state, ins, &RHS(ins, 1));
+                       left  = read_const(state, ins, RHS(ins, 0));
+                       right = read_const(state, ins, RHS(ins, 1));
                        mkconst(state, ins, left + right);
                }
-               else /* op == OP_ADDRCONST */ {
+               else if (RHS(ins, 0)->op == OP_ADDRCONST) {
                        struct triple *sdecl;
                        ulong_t left, right;
                        sdecl = MISC(RHS(ins, 0), 0);
@@ -5651,6 +9479,9 @@ static void simplify_add(struct compile_state *state, struct triple *ins)
                        right = RHS(ins, 1)->u.cval;
                        mkaddr_const(state, ins, sdecl, left + right);
                }
+               else {
+                       internal_warning(state, ins, "Optimize me!");
+               }
        }
        else if (is_const(RHS(ins, 0)) && !is_const(RHS(ins, 1))) {
                struct triple *tmp;
@@ -5662,14 +9493,14 @@ static void simplify_add(struct compile_state *state, struct triple *ins)
 
 static void simplify_sub(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               if (!is_pointer(RHS(ins, 0))) {
+       if (is_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
+               if (RHS(ins, 0)->op == OP_INTCONST) {
                        ulong_t left, right;
-                       left  = read_const(state, ins, &RHS(ins, 0));
-                       right = read_const(state, ins, &RHS(ins, 1));
+                       left  = read_const(state, ins, RHS(ins, 0));
+                       right = read_const(state, ins, RHS(ins, 1));
                        mkconst(state, ins, left - right);
                }
-               else /* op == OP_ADDRCONST */ {
+               else if (RHS(ins, 0)->op == OP_ADDRCONST) {
                        struct triple *sdecl;
                        ulong_t left, right;
                        sdecl = MISC(RHS(ins, 0), 0);
@@ -5677,86 +9508,108 @@ static void simplify_sub(struct compile_state *state, struct triple *ins)
                        right = RHS(ins, 1)->u.cval;
                        mkaddr_const(state, ins, sdecl, left - right);
                }
-       }
+               else {
+                       internal_warning(state, ins, "Optimize me!");
+               }
+       }
 }
 
 static void simplify_sl(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 1))) {
                ulong_t right;
-               right = read_const(state, ins, &RHS(ins, 1));
-               if (right >= (size_of(state, ins->type)*8)) {
+               right = read_const(state, ins, RHS(ins, 1));
+               if (right >= (size_of(state, ins->type))) {
                        warning(state, ins, "left shift count >= width of type");
                }
        }
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins,  left << right);
        }
 }
 
 static void simplify_usr(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 1))) {
                ulong_t right;
-               right = read_const(state, ins, &RHS(ins, 1));
-               if (right >= (size_of(state, ins->type)*8)) {
+               right = read_const(state, ins, RHS(ins, 1));
+               if (right >= (size_of(state, ins->type))) {
                        warning(state, ins, "right shift count >= width of type");
                }
        }
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins, left >> right);
        }
 }
 
 static void simplify_ssr(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 1))) {
                ulong_t right;
-               right = read_const(state, ins, &RHS(ins, 1));
-               if (right >= (size_of(state, ins->type)*8)) {
+               right = read_const(state, ins, RHS(ins, 1));
+               if (right >= (size_of(state, ins->type))) {
                        warning(state, ins, "right shift count >= width of type");
                }
        }
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
+               left  = read_sconst(state, ins, RHS(ins, 0));
+               right = read_sconst(state, ins, RHS(ins, 1));
                mkconst(state, ins, left >> right);
        }
 }
 
 static void simplify_and(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left & right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_simple_const(left) && is_simple_const(right)) {
+               ulong_t lval, rval;
+               lval = read_const(state, ins, left);
+               rval = read_const(state, ins, right);
+               mkconst(state, ins, lval & rval);
+       }
+       else if (is_zero(right) || is_zero(left)) {
+               mkconst(state, ins, 0);
        }
 }
 
 static void simplify_or(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left | right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_simple_const(left) && is_simple_const(right)) {
+               ulong_t lval, rval;
+               lval = read_const(state, ins, left);
+               rval = read_const(state, ins, right);
+               mkconst(state, ins, lval | rval);
+       }
+#if 0 /* I need to handle type mismatches here... */
+       else if (is_zero(right)) {
+               mkcopy(state, ins, left);
        }
+       else if (is_zero(left)) {
+               mkcopy(state, ins, right);
+       }
+#endif
 }
 
 static void simplify_xor(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
                ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
+               left  = read_const(state, ins, RHS(ins, 0));
+               right = read_const(state, ins, RHS(ins, 1));
                mkconst(state, ins, left ^ right);
        }
 }
@@ -5773,9 +9626,9 @@ static void simplify_pos(struct compile_state *state, struct triple *ins)
 
 static void simplify_neg(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
+       if (is_simple_const(RHS(ins, 0))) {
                ulong_t left;
-               left = read_const(state, ins, &RHS(ins, 0));
+               left = read_const(state, ins, RHS(ins, 0));
                mkconst(state, ins, -left);
        }
        else if (RHS(ins, 0)->op == OP_NEG) {
@@ -5785,93 +9638,123 @@ static void simplify_neg(struct compile_state *state, struct triple *ins)
 
 static void simplify_invert(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
+       if (is_simple_const(RHS(ins, 0))) {
                ulong_t left;
-               left = read_const(state, ins, &RHS(ins, 0));
+               left = read_const(state, ins, RHS(ins, 0));
                mkconst(state, ins, ~left);
        }
 }
 
 static void simplify_eq(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left == right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_eq(state, ins, left, right);
+               if (val >= 0) {
+                       mkconst(state, ins, val == 1);
+               }
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 1);
        }
 }
 
 static void simplify_noteq(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left != right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_eq(state, ins, left, right);
+               if (val >= 0) {
+                       mkconst(state, ins, val != 1);
+               }
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       if (left == right) {
                mkconst(state, ins, 0);
        }
 }
 
 static void simplify_sless(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
-               mkconst(state, ins, left < right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_scmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val < 0);
+               }
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 0);
        }
 }
 
 static void simplify_uless(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left < right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_ucmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val < 0);
+               }
        }
-       else if (is_zero(RHS(ins, 0))) {
-               mkconst(state, ins, 1);
+       else if (is_zero(right)) {
+               mkconst(state, ins, 0);
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 0);
        }
 }
 
 static void simplify_smore(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
-               mkconst(state, ins, left > right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_scmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val > 0);
+               }
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 0);
        }
 }
 
 static void simplify_umore(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left > right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_ucmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val > 0);
+               }
        }
-       else if (is_zero(RHS(ins, 1))) {
-               mkconst(state, ins, 1);
+       else if (is_zero(left)) {
+               mkconst(state, ins, 0);
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 0);
        }
 }
@@ -5879,132 +9762,284 @@ static void simplify_umore(struct compile_state *state, struct triple *ins)
 
 static void simplify_slesseq(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
-               mkconst(state, ins, left <= right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_scmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val <= 0);
+               }
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 1);
        }
 }
 
 static void simplify_ulesseq(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left <= right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_ucmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val <= 0);
+               }
        }
-       else if (is_zero(RHS(ins, 0))) {
+       else if (is_zero(left)) {
                mkconst(state, ins, 1);
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 1);
        }
 }
 
 static void simplify_smoreeq(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 0))) {
-               long_t left, right;
-               left  = read_sconst(ins, &RHS(ins, 0));
-               right = read_sconst(ins, &RHS(ins, 1));
-               mkconst(state, ins, left >= right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_scmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val >= 0);
+               }
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 1);
        }
 }
 
 static void simplify_umoreeq(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0)) && is_const(RHS(ins, 1))) {
-               ulong_t left, right;
-               left  = read_const(state, ins, &RHS(ins, 0));
-               right = read_const(state, ins, &RHS(ins, 1));
-               mkconst(state, ins, left >= right);
+       struct triple *left, *right;
+       left = RHS(ins, 0);
+       right = RHS(ins, 1);
+
+       if (is_const(left) && is_const(right)) {
+               int val;
+               val = const_ucmp(state, ins, left, right);
+               if ((val >= -1) && (val <= 1)) {
+                       mkconst(state, ins, val >= 0);
+               }
        }
-       else if (is_zero(RHS(ins, 1))) {
+       else if (is_zero(right)) {
                mkconst(state, ins, 1);
        }
-       else if (RHS(ins, 0) == RHS(ins, 1)) {
+       else if (left == right) {
                mkconst(state, ins, 1);
        }
 }
 
 static void simplify_lfalse(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
-               ulong_t left;
-               left = read_const(state, ins, &RHS(ins, 0));
-               mkconst(state, ins, left == 0);
+       struct triple *rhs;
+       rhs = RHS(ins, 0);
+
+       if (is_const(rhs)) {
+               mkconst(state, ins, !const_ltrue(state, ins, rhs));
        }
        /* Otherwise if I am the only user... */
-       else if ((RHS(ins, 0)->use->member == ins) && (RHS(ins, 0)->use->next == 0)) {
+       else if ((rhs->use) &&
+               (rhs->use->member == ins) && (rhs->use->next == 0)) {
                int need_copy = 1;
                /* Invert a boolean operation */
-               switch(RHS(ins, 0)->op) {
-               case OP_LTRUE:   RHS(ins, 0)->op = OP_LFALSE;  break;
-               case OP_LFALSE:  RHS(ins, 0)->op = OP_LTRUE;   break;
-               case OP_EQ:      RHS(ins, 0)->op = OP_NOTEQ;   break;
-               case OP_NOTEQ:   RHS(ins, 0)->op = OP_EQ;      break;
-               case OP_SLESS:   RHS(ins, 0)->op = OP_SMOREEQ; break;
-               case OP_ULESS:   RHS(ins, 0)->op = OP_UMOREEQ; break;
-               case OP_SMORE:   RHS(ins, 0)->op = OP_SLESSEQ; break;
-               case OP_UMORE:   RHS(ins, 0)->op = OP_ULESSEQ; break;
-               case OP_SLESSEQ: RHS(ins, 0)->op = OP_SMORE;   break;
-               case OP_ULESSEQ: RHS(ins, 0)->op = OP_UMORE;   break;
-               case OP_SMOREEQ: RHS(ins, 0)->op = OP_SLESS;   break;
-               case OP_UMOREEQ: RHS(ins, 0)->op = OP_ULESS;   break;
+               switch(rhs->op) {
+               case OP_LTRUE:   rhs->op = OP_LFALSE;  break;
+               case OP_LFALSE:  rhs->op = OP_LTRUE;   break;
+               case OP_EQ:      rhs->op = OP_NOTEQ;   break;
+               case OP_NOTEQ:   rhs->op = OP_EQ;      break;
+               case OP_SLESS:   rhs->op = OP_SMOREEQ; break;
+               case OP_ULESS:   rhs->op = OP_UMOREEQ; break;
+               case OP_SMORE:   rhs->op = OP_SLESSEQ; break;
+               case OP_UMORE:   rhs->op = OP_ULESSEQ; break;
+               case OP_SLESSEQ: rhs->op = OP_SMORE;   break;
+               case OP_ULESSEQ: rhs->op = OP_UMORE;   break;
+               case OP_SMOREEQ: rhs->op = OP_SLESS;   break;
+               case OP_UMOREEQ: rhs->op = OP_ULESS;   break;
                default:
                        need_copy = 0;
                        break;
                }
                if (need_copy) {
-                       mkcopy(state, ins, RHS(ins, 0));
+                       mkcopy(state, ins, rhs);
                }
        }
 }
 
 static void simplify_ltrue (struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
-               ulong_t left;
-               left = read_const(state, ins, &RHS(ins, 0));
-               mkconst(state, ins, left != 0);
+       struct triple *rhs;
+       rhs = RHS(ins, 0);
+
+       if (is_const(rhs)) {
+               mkconst(state, ins, const_ltrue(state, ins, rhs));
        }
-       else switch(RHS(ins, 0)->op) {
+       else switch(rhs->op) {
        case OP_LTRUE:   case OP_LFALSE:  case OP_EQ:      case OP_NOTEQ:
        case OP_SLESS:   case OP_ULESS:   case OP_SMORE:   case OP_UMORE:
        case OP_SLESSEQ: case OP_ULESSEQ: case OP_SMOREEQ: case OP_UMOREEQ:
-               mkcopy(state, ins, RHS(ins, 0));
+               mkcopy(state, ins, rhs);
+       }
+
+}
+
+static void simplify_load(struct compile_state *state, struct triple *ins)
+{
+       struct triple *addr, *sdecl, *blob;
+
+       /* If I am doing a load with a constant pointer from a constant
+        * table get the value.
+        */
+       addr = RHS(ins, 0);
+       if ((addr->op == OP_ADDRCONST) && (sdecl = MISC(addr, 0)) &&
+               (sdecl->op == OP_SDECL) && (blob = MISC(sdecl, 0)) &&
+               (blob->op == OP_BLOBCONST)) {
+               unsigned char buffer[SIZEOF_WORD];
+               size_t reg_size, mem_size;
+               const char *src, *end;
+               ulong_t val;
+               reg_size = reg_size_of(state, ins->type);
+               if (reg_size > REG_SIZEOF_REG) {
+                       internal_error(state, ins, "load size greater than register");
+               }
+               mem_size = size_of(state, ins->type);
+               end = blob->u.blob;
+               end += bits_to_bytes(size_of(state, sdecl->type));
+               src = blob->u.blob;
+               src += addr->u.cval;
+
+               if (src > end) {
+                       error(state, ins, "Load address out of bounds");
+               }
+
+               memset(buffer, 0, sizeof(buffer));
+               memcpy(buffer, src, bits_to_bytes(mem_size));
+
+               switch(mem_size) {
+               case SIZEOF_I8:  val = *((uint8_t *) buffer); break;
+               case SIZEOF_I16: val = *((uint16_t *)buffer); break;
+               case SIZEOF_I32: val = *((uint32_t *)buffer); break;
+               case SIZEOF_I64: val = *((uint64_t *)buffer); break;
+               default:
+                       internal_error(state, ins, "mem_size: %d not handled",
+                               mem_size);
+                       val = 0;
+                       break;
+               }
+               mkconst(state, ins, val);
+       }
+}
+
+static void simplify_uextract(struct compile_state *state, struct triple *ins)
+{
+       if (is_simple_const(RHS(ins, 0))) {
+               ulong_t val;
+               ulong_t mask;
+               val = read_const(state, ins, RHS(ins, 0));
+               mask = 1;
+               mask <<= ins->u.bitfield.size;
+               mask -= 1;
+               val >>= ins->u.bitfield.offset;
+               val &= mask;
+               mkconst(state, ins, val);
+       }
+}
+
+static void simplify_sextract(struct compile_state *state, struct triple *ins)
+{
+       if (is_simple_const(RHS(ins, 0))) {
+               ulong_t val;
+               ulong_t mask;
+               long_t sval;
+               val = read_const(state, ins, RHS(ins, 0));
+               mask = 1;
+               mask <<= ins->u.bitfield.size;
+               mask -= 1;
+               val >>= ins->u.bitfield.offset;
+               val &= mask;
+               val <<= (SIZEOF_LONG - ins->u.bitfield.size);
+               sval = val;
+               sval >>= (SIZEOF_LONG - ins->u.bitfield.size);
+               mkconst(state, ins, sval);
        }
+}
 
+static void simplify_deposit(struct compile_state *state, struct triple *ins)
+{
+       if (is_simple_const(RHS(ins, 0)) && is_simple_const(RHS(ins, 1))) {
+               ulong_t targ, val;
+               ulong_t mask;
+               targ = read_const(state, ins, RHS(ins, 0));
+               val  = read_const(state, ins, RHS(ins, 1));
+               mask = 1;
+               mask <<= ins->u.bitfield.size;
+               mask -= 1;
+               mask <<= ins->u.bitfield.offset;
+               targ &= ~mask;
+               val <<= ins->u.bitfield.offset;
+               val &= mask;
+               targ |= val;
+               mkconst(state, ins, targ);
+       }
 }
 
 static void simplify_copy(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
-               switch(RHS(ins, 0)->op) {
+       struct triple *right;
+       right = RHS(ins, 0);
+       if (is_subset_type(ins->type, right->type)) {
+               ins->type = right->type;
+       }
+       if (equiv_types(ins->type, right->type)) {
+               ins->op = OP_COPY;/* I don't need to convert if the types match */
+       } else {
+               if (ins->op == OP_COPY) {
+                       internal_error(state, ins, "type mismatch on copy");
+               }
+       }
+       if (is_const(right) && (right->op == OP_ADDRCONST) && is_pointer(ins)) {
+               struct triple *sdecl;
+               ulong_t offset;
+               sdecl  = MISC(right, 0);
+               offset = right->u.cval;
+               mkaddr_const(state, ins, sdecl, offset);
+       }
+       else if (is_const(right) && is_write_compatible(state, ins->type, right->type)) {
+               switch(right->op) {
                case OP_INTCONST:
                {
                        ulong_t left;
-                       left = read_const(state, ins, &RHS(ins, 0));
+                       left = read_const(state, ins, right);
+                       /* Ensure I have not overflowed the destination. */
+                       if (size_of(state, right->type) > size_of(state, ins->type)) {
+                               ulong_t mask;
+                               mask = 1;
+                               mask <<= size_of(state, ins->type);
+                               mask -= 1;
+                               left &= mask;
+                       }
+                       /* Ensure I am properly sign extended */
+                       if (size_of(state, right->type) < size_of(state, ins->type) &&
+                               is_signed(right->type)) {
+                               long_t val;
+                               int shift;
+                               shift = SIZEOF_LONG - size_of(state, right->type);
+                               val = left;
+                               val <<= shift;
+                               val >>= shift;
+                               left = val;
+                       }
                        mkconst(state, ins, left);
                        break;
                }
-               case OP_ADDRCONST:
-               {
-                       struct triple *sdecl;
-                       ulong_t offset;
-                       sdecl  = MISC(RHS(ins, 0), 0);
-                       offset = RHS(ins, 0)->u.cval;
-                       mkaddr_const(state, ins, sdecl, offset);
-                       break;
-               }
                default:
                        internal_error(state, ins, "uknown constant");
                        break;
@@ -6012,29 +10047,103 @@ static void simplify_copy(struct compile_state *state, struct triple *ins)
        }
 }
 
+static int phi_present(struct block *block)
+{
+       struct triple *ptr;
+       if (!block) {
+               return 0;
+       }
+       ptr = block->first;
+       do {
+               if (ptr->op == OP_PHI) {
+                       return 1;
+               }
+               ptr = ptr->next;
+       } while(ptr != block->last);
+       return 0;
+}
+
+static int phi_dependency(struct block *block)
+{
+       /* A block has a phi dependency if a phi function
+        * depends on that block to exist, and makes a block
+        * that is otherwise useless unsafe to remove.
+        */
+       if (block) {
+               struct block_set *edge;
+               for(edge = block->edges; edge; edge = edge->next) {
+                       if (phi_present(edge->member)) {
+                               return 1;
+                       }
+               }
+       }
+       return 0;
+}
+
+static struct triple *branch_target(struct compile_state *state, struct triple *ins)
+{
+       struct triple *targ;
+       targ = TARG(ins, 0);
+       /* During scc_transform temporary triples are allocated that
+        * loop back onto themselves. If I see one don't advance the
+        * target.
+        */
+       while(triple_is_structural(state, targ) &&
+               (targ->next != targ) && (targ->next != state->first)) {
+               targ = targ->next;
+       }
+       return targ;
+}
+
+
 static void simplify_branch(struct compile_state *state, struct triple *ins)
 {
-       struct block *block;
-       if (ins->op != OP_BRANCH) {
+       int simplified, loops;
+       if ((ins->op != OP_BRANCH) && (ins->op != OP_CBRANCH)) {
                internal_error(state, ins, "not branch");
        }
        if (ins->use != 0) {
                internal_error(state, ins, "branch use");
        }
-#warning "FIXME implement simplify branch."
-       /* The challenge here with simplify branch is that I need to 
+       /* The challenge here with simplify branch is that I need to
         * make modifications to the control flow graph as well
-        * as to the branch instruction itself.
+        * as to the branch instruction itself.  That is handled
+        * by rebuilding the basic blocks after simplify all is called.
+        */
+
+       /* If we have a branch to an unconditional branch update
+        * our target.  But watch out for dependencies from phi
+        * functions.
+        * Also only do this a limited number of times so
+        * we don't get into an infinite loop.
+        */
+       loops = 0;
+       do {
+               struct triple *targ;
+               simplified = 0;
+               targ = branch_target(state, ins);
+               if ((targ != ins) && (targ->op == OP_BRANCH) &&
+                       !phi_dependency(targ->u.block))
+               {
+                       unuse_triple(TARG(ins, 0), ins);
+                       TARG(ins, 0) = TARG(targ, 0);
+                       use_triple(TARG(ins, 0), ins);
+                       simplified = 1;
+               }
+       } while(simplified && (++loops < 20));
+
+       /* If we have a conditional branch with a constant condition
+        * make it an unconditional branch.
         */
-       block = ins->u.block;
-       
-       if (TRIPLE_RHS(ins->sizes) && is_const(RHS(ins, 0))) {
+       if ((ins->op == OP_CBRANCH) && is_simple_const(RHS(ins, 0))) {
                struct triple *targ;
                ulong_t value;
-               value = read_const(state, ins, &RHS(ins, 0));
+               value = read_const(state, ins, RHS(ins, 0));
                unuse_triple(RHS(ins, 0), ins);
                targ = TARG(ins, 0);
-               ins->sizes = TRIPLE_SIZES(0, 0, 0, 1);
+               ins->rhs  = 0;
+               ins->targ = 1;
+               ins->op = OP_BRANCH;
                if (value) {
                        unuse_triple(ins->next, ins);
                        TARG(ins, 0) = targ;
@@ -6043,206 +10152,251 @@ static void simplify_branch(struct compile_state *state, struct triple *ins)
                        unuse_triple(targ, ins);
                        TARG(ins, 0) = ins->next;
                }
-#warning "FIXME handle the case of making a branch unconditional"
        }
+
+       /* If we have a branch to the next instruction,
+        * make it a noop.
+        */
        if (TARG(ins, 0) == ins->next) {
-               unuse_triple(ins->next, ins);
-               if (TRIPLE_RHS(ins->sizes)) {
+               unuse_triple(TARG(ins, 0), ins);
+               if (ins->op == OP_CBRANCH) {
                        unuse_triple(RHS(ins, 0), ins);
                        unuse_triple(ins->next, ins);
                }
-               ins->sizes = TRIPLE_SIZES(0, 0, 0, 0);
+               ins->lhs = 0;
+               ins->rhs = 0;
+               ins->misc = 0;
+               ins->targ = 0;
                ins->op = OP_NOOP;
                if (ins->use) {
                        internal_error(state, ins, "noop use != 0");
                }
-#warning "FIXME handle the case of killing a branch"
+       }
+}
+
+static void simplify_label(struct compile_state *state, struct triple *ins)
+{
+       /* Ignore volatile labels */
+       if (!triple_is_pure(state, ins, ins->id)) {
+               return;
+       }
+       if (ins->use == 0) {
+               ins->op = OP_NOOP;
+       }
+       else if (ins->prev->op == OP_LABEL) {
+               /* In general it is not safe to merge one label that
+                * imediately follows another.  The problem is that the empty
+                * looking block may have phi functions that depend on it.
+                */
+               if (!phi_dependency(ins->prev->u.block)) {
+                       struct triple_set *user, *next;
+                       ins->op = OP_NOOP;
+                       for(user = ins->use; user; user = next) {
+                               struct triple *use, **expr;
+                               next = user->next;
+                               use = user->member;
+                               expr = triple_targ(state, use, 0);
+                               for(;expr; expr = triple_targ(state, use, expr)) {
+                                       if (*expr == ins) {
+                                               *expr = ins->prev;
+                                               unuse_triple(ins, use);
+                                               use_triple(ins->prev, use);
+                                       }
+
+                               }
+                       }
+                       if (ins->use) {
+                               internal_error(state, ins, "noop use != 0");
+                       }
+               }
        }
 }
 
 static void simplify_phi(struct compile_state *state, struct triple *ins)
 {
-       struct triple **expr;
-       ulong_t value;
-       expr = triple_rhs(state, ins, 0);
-       if (!*expr || !is_const(*expr)) {
+       struct triple **slot;
+       struct triple *value;
+       int zrhs, i;
+       ulong_t cvalue;
+       slot = &RHS(ins, 0);
+       zrhs = ins->rhs;
+       if (zrhs == 0) {
                return;
        }
-       value = read_const(state, ins, expr);
-       for(;expr;expr = triple_rhs(state, ins, expr)) {
-               if (!*expr || !is_const(*expr)) {
-                       return;
+       /* See if all of the rhs members of a phi have the same value */
+       if (slot[0] && is_simple_const(slot[0])) {
+               cvalue = read_const(state, ins, slot[0]);
+               for(i = 1; i < zrhs; i++) {
+                       if (    !slot[i] ||
+                               !is_simple_const(slot[i]) ||
+                               !equiv_types(slot[0]->type, slot[i]->type) ||
+                               (cvalue != read_const(state, ins, slot[i]))) {
+                               break;
+                       }
                }
-               if (value != read_const(state, ins, expr)) {
+               if (i == zrhs) {
+                       mkconst(state, ins, cvalue);
                        return;
                }
        }
-       mkconst(state, ins, value);
+
+       /* See if all of rhs members of a phi are the same */
+       value = slot[0];
+       for(i = 1; i < zrhs; i++) {
+               if (slot[i] != value) {
+                       break;
+               }
+       }
+       if (i == zrhs) {
+               /* If the phi has a single value just copy it */
+               if (!is_subset_type(ins->type, value->type)) {
+                       internal_error(state, ins, "bad input type to phi");
+               }
+               /* Make the types match */
+               if (!equiv_types(ins->type, value->type)) {
+                       ins->type = value->type;
+               }
+               /* Now make the actual copy */
+               mkcopy(state, ins, value);
+               return;
+       }
 }
 
 
 static void simplify_bsf(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
+       if (is_simple_const(RHS(ins, 0))) {
                ulong_t left;
-               left = read_const(state, ins, &RHS(ins, 0));
+               left = read_const(state, ins, RHS(ins, 0));
                mkconst(state, ins, bsf(left));
        }
 }
 
 static void simplify_bsr(struct compile_state *state, struct triple *ins)
 {
-       if (is_const(RHS(ins, 0))) {
+       if (is_simple_const(RHS(ins, 0))) {
                ulong_t left;
-               left = read_const(state, ins, &RHS(ins, 0));
+               left = read_const(state, ins, RHS(ins, 0));
                mkconst(state, ins, bsr(left));
        }
 }
 
 
 typedef void (*simplify_t)(struct compile_state *state, struct triple *ins);
-static const simplify_t table_simplify[] = {
-#if 0
-#define simplify_smul     simplify_noop
-#define simplify_umul    simplify_noop
-#define simplify_sdiv    simplify_noop
-#define simplify_udiv    simplify_noop
-#define simplify_smod    simplify_noop
-#define simplify_umod    simplify_noop
-#endif
-#if 0
-#define simplify_add     simplify_noop
-#define simplify_sub     simplify_noop
-#endif
-#if 0
-#define simplify_sl      simplify_noop
-#define simplify_usr     simplify_noop
-#define simplify_ssr     simplify_noop
-#endif
-#if 0
-#define simplify_and     simplify_noop
-#define simplify_xor     simplify_noop
-#define simplify_or      simplify_noop
-#endif
-#if 0
-#define simplify_pos     simplify_noop
-#define simplify_neg     simplify_noop
-#define simplify_invert          simplify_noop
-#endif
-
-#if 0
-#define simplify_eq      simplify_noop
-#define simplify_noteq   simplify_noop
-#endif
-#if 0
-#define simplify_sless   simplify_noop
-#define simplify_uless   simplify_noop
-#define simplify_smore   simplify_noop
-#define simplify_umore   simplify_noop
-#endif
-#if 0
-#define simplify_slesseq  simplify_noop
-#define simplify_ulesseq  simplify_noop
-#define simplify_smoreeq  simplify_noop
-#define simplify_umoreeq  simplify_noop
-#endif
-#if 0
-#define simplify_lfalse          simplify_noop
-#endif
-#if 0
-#define simplify_ltrue   simplify_noop
-#endif
-
-#if 0
-#define simplify_copy    simplify_noop
-#endif
-
-#if 0
-#define simplify_branch          simplify_noop
-#endif
+static const struct simplify_table {
+       simplify_t func;
+       unsigned long flag;
+} table_simplify[] = {
+#define simplify_sdivt    simplify_noop
+#define simplify_udivt    simplify_noop
+#define simplify_piece    simplify_noop
+
+[OP_SDIVT      ] = { simplify_sdivt,    COMPILER_SIMPLIFY_ARITH },
+[OP_UDIVT      ] = { simplify_udivt,   COMPILER_SIMPLIFY_ARITH },
+[OP_SMUL       ] = { simplify_smul,    COMPILER_SIMPLIFY_ARITH },
+[OP_UMUL       ] = { simplify_umul,    COMPILER_SIMPLIFY_ARITH },
+[OP_SDIV       ] = { simplify_sdiv,    COMPILER_SIMPLIFY_ARITH },
+[OP_UDIV       ] = { simplify_udiv,    COMPILER_SIMPLIFY_ARITH },
+[OP_SMOD       ] = { simplify_smod,    COMPILER_SIMPLIFY_ARITH },
+[OP_UMOD       ] = { simplify_umod,    COMPILER_SIMPLIFY_ARITH },
+[OP_ADD        ] = { simplify_add,     COMPILER_SIMPLIFY_ARITH },
+[OP_SUB        ] = { simplify_sub,     COMPILER_SIMPLIFY_ARITH },
+[OP_SL         ] = { simplify_sl,      COMPILER_SIMPLIFY_SHIFT },
+[OP_USR        ] = { simplify_usr,     COMPILER_SIMPLIFY_SHIFT },
+[OP_SSR        ] = { simplify_ssr,     COMPILER_SIMPLIFY_SHIFT },
+[OP_AND        ] = { simplify_and,     COMPILER_SIMPLIFY_BITWISE },
+[OP_XOR        ] = { simplify_xor,     COMPILER_SIMPLIFY_BITWISE },
+[OP_OR         ] = { simplify_or,      COMPILER_SIMPLIFY_BITWISE },
+[OP_POS        ] = { simplify_pos,     COMPILER_SIMPLIFY_ARITH },
+[OP_NEG        ] = { simplify_neg,     COMPILER_SIMPLIFY_ARITH },
+[OP_INVERT     ] = { simplify_invert,  COMPILER_SIMPLIFY_BITWISE },
+
+[OP_EQ         ] = { simplify_eq,      COMPILER_SIMPLIFY_LOGICAL },
+[OP_NOTEQ      ] = { simplify_noteq,   COMPILER_SIMPLIFY_LOGICAL },
+[OP_SLESS      ] = { simplify_sless,   COMPILER_SIMPLIFY_LOGICAL },
+[OP_ULESS      ] = { simplify_uless,   COMPILER_SIMPLIFY_LOGICAL },
+[OP_SMORE      ] = { simplify_smore,   COMPILER_SIMPLIFY_LOGICAL },
+[OP_UMORE      ] = { simplify_umore,   COMPILER_SIMPLIFY_LOGICAL },
+[OP_SLESSEQ    ] = { simplify_slesseq,         COMPILER_SIMPLIFY_LOGICAL },
+[OP_ULESSEQ    ] = { simplify_ulesseq, COMPILER_SIMPLIFY_LOGICAL },
+[OP_SMOREEQ    ] = { simplify_smoreeq, COMPILER_SIMPLIFY_LOGICAL },
+[OP_UMOREEQ    ] = { simplify_umoreeq, COMPILER_SIMPLIFY_LOGICAL },
+[OP_LFALSE     ] = { simplify_lfalse,  COMPILER_SIMPLIFY_LOGICAL },
+[OP_LTRUE      ] = { simplify_ltrue,   COMPILER_SIMPLIFY_LOGICAL },
+
+[OP_LOAD       ] = { simplify_load,    COMPILER_SIMPLIFY_OP },
+[OP_STORE      ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+
+[OP_UEXTRACT   ] = { simplify_uextract, COMPILER_SIMPLIFY_BITFIELD },
+[OP_SEXTRACT   ] = { simplify_sextract, COMPILER_SIMPLIFY_BITFIELD },
+[OP_DEPOSIT    ] = { simplify_deposit,  COMPILER_SIMPLIFY_BITFIELD },
+
+[OP_NOOP       ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+
+[OP_INTCONST   ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_BLOBCONST  ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_ADDRCONST  ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_UNKNOWNVAL ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+
+[OP_WRITE      ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_READ       ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_COPY       ] = { simplify_copy,    COMPILER_SIMPLIFY_COPY },
+[OP_CONVERT    ] = { simplify_copy,    COMPILER_SIMPLIFY_COPY },
+[OP_PIECE      ] = { simplify_piece,   COMPILER_SIMPLIFY_OP },
+[OP_ASM        ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+
+[OP_DOT        ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_INDEX      ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+
+[OP_LIST       ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_BRANCH     ] = { simplify_branch,  COMPILER_SIMPLIFY_BRANCH },
+[OP_CBRANCH    ] = { simplify_branch,  COMPILER_SIMPLIFY_BRANCH },
+[OP_CALL       ] = { simplify_noop,    COMPILER_SIMPLIFY_BRANCH },
+[OP_RET        ] = { simplify_noop,    COMPILER_SIMPLIFY_BRANCH },
+[OP_LABEL      ] = { simplify_label,   COMPILER_SIMPLIFY_LABEL },
+[OP_ADECL      ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_SDECL      ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_PHI        ] = { simplify_phi,     COMPILER_SIMPLIFY_PHI },
+
+[OP_INB        ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_INW        ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_INL        ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_OUTB       ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_OUTW       ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_OUTL       ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_BSF        ] = { simplify_bsf,     COMPILER_SIMPLIFY_OP },
+[OP_BSR        ] = { simplify_bsr,     COMPILER_SIMPLIFY_OP },
+[OP_RDMSR      ] = { simplify_noop,    COMPILER_SIMPLIFY_OP },
+[OP_WRMSR      ] = { simplify_noop,     COMPILER_SIMPLIFY_OP },
+[OP_HLT        ] = { simplify_noop,     COMPILER_SIMPLIFY_OP },
+};
 
-#if 0
-#define simplify_phi     simplify_noop
+static inline void debug_simplify(struct compile_state *state,
+       simplify_t do_simplify, struct triple *ins)
+{
+#if DEBUG_SIMPLIFY_HIRES
+               if (state->functions_joined && (do_simplify != simplify_noop)) {
+                       /* High resolution debugging mode */
+                       fprintf(state->dbgout, "simplifing: ");
+                       display_triple(state->dbgout, ins);
+               }
 #endif
-
-#if 0
-#define simplify_bsf     simplify_noop
-#define simplify_bsr      simplify_noop
+               do_simplify(state, ins);
+#if DEBUG_SIMPLIFY_HIRES
+               if (state->functions_joined && (do_simplify != simplify_noop)) {
+                       /* High resolution debugging mode */
+                       fprintf(state->dbgout, "simplified: ");
+                       display_triple(state->dbgout, ins);
+               }
 #endif
-
-[OP_SMUL       ] = simplify_smul,
-[OP_UMUL       ] = simplify_umul,
-[OP_SDIV       ] = simplify_sdiv,
-[OP_UDIV       ] = simplify_udiv,
-[OP_SMOD       ] = simplify_smod,
-[OP_UMOD       ] = simplify_umod,
-[OP_ADD        ] = simplify_add,
-[OP_SUB        ] = simplify_sub,
-[OP_SL         ] = simplify_sl,
-[OP_USR        ] = simplify_usr,
-[OP_SSR        ] = simplify_ssr,
-[OP_AND        ] = simplify_and,
-[OP_XOR        ] = simplify_xor,
-[OP_OR         ] = simplify_or,
-[OP_POS        ] = simplify_pos,
-[OP_NEG        ] = simplify_neg,
-[OP_INVERT     ] = simplify_invert,
-
-[OP_EQ         ] = simplify_eq,
-[OP_NOTEQ      ] = simplify_noteq,
-[OP_SLESS      ] = simplify_sless,
-[OP_ULESS      ] = simplify_uless,
-[OP_SMORE      ] = simplify_smore,
-[OP_UMORE      ] = simplify_umore,
-[OP_SLESSEQ    ] = simplify_slesseq,
-[OP_ULESSEQ    ] = simplify_ulesseq,
-[OP_SMOREEQ    ] = simplify_smoreeq,
-[OP_UMOREEQ    ] = simplify_umoreeq,
-[OP_LFALSE     ] = simplify_lfalse,
-[OP_LTRUE      ] = simplify_ltrue,
-
-[OP_LOAD       ] = simplify_noop,
-[OP_STORE      ] = simplify_noop,
-
-[OP_NOOP       ] = simplify_noop,
-
-[OP_INTCONST   ] = simplify_noop,
-[OP_BLOBCONST  ] = simplify_noop,
-[OP_ADDRCONST  ] = simplify_noop,
-
-[OP_WRITE      ] = simplify_noop,
-[OP_READ       ] = simplify_noop,
-[OP_COPY       ] = simplify_copy,
-[OP_PIECE      ] = simplify_noop,
-[OP_ASM        ] = simplify_noop,
-
-[OP_DOT        ] = simplify_noop,
-[OP_VAL_VEC    ] = simplify_noop,
-
-[OP_LIST       ] = simplify_noop,
-[OP_BRANCH     ] = simplify_branch,
-[OP_LABEL      ] = simplify_noop,
-[OP_ADECL      ] = simplify_noop,
-[OP_SDECL      ] = simplify_noop,
-[OP_PHI        ] = simplify_phi,
-
-[OP_INB        ] = simplify_noop,
-[OP_INW        ] = simplify_noop,
-[OP_INL        ] = simplify_noop,
-[OP_OUTB       ] = simplify_noop,
-[OP_OUTW       ] = simplify_noop,
-[OP_OUTL       ] = simplify_noop,
-[OP_BSF        ] = simplify_bsf,
-[OP_BSR        ] = simplify_bsr,
-[OP_RDMSR      ] = simplify_noop,
-[OP_WRMSR      ] = simplify_noop,                    
-[OP_HLT        ] = simplify_noop,
-};
-
+}
 static void simplify(struct compile_state *state, struct triple *ins)
 {
        int op;
        simplify_t do_simplify;
+       if (ins == &unknown_triple) {
+               internal_error(state, ins, "simplifying the unknown triple?");
+       }
        do {
                op = ins->op;
                do_simplify = 0;
@@ -6250,26 +10404,47 @@ static void simplify(struct compile_state *state, struct triple *ins)
                        do_simplify = 0;
                }
                else {
-                       do_simplify = table_simplify[op];
+                       do_simplify = table_simplify[op].func;
+               }
+               if (do_simplify &&
+                       !(state->compiler->flags & table_simplify[op].flag)) {
+                       do_simplify = simplify_noop;
                }
+               if (do_simplify && (ins->id & TRIPLE_FLAG_VOLATILE)) {
+                       do_simplify = simplify_noop;
+               }
+
                if (!do_simplify) {
-                       internal_error(state, ins, "cannot simplify op: %d %s\n",
+                       internal_error(state, ins, "cannot simplify op: %d %s",
                                op, tops(op));
                        return;
                }
-               do_simplify(state, ins);
+               debug_simplify(state, do_simplify, ins);
        } while(ins->op != op);
 }
 
+static void rebuild_ssa_form(struct compile_state *state);
+
 static void simplify_all(struct compile_state *state)
 {
        struct triple *ins, *first;
-       first = RHS(state->main_function, 0);
+       if (!(state->compiler->flags & COMPILER_SIMPLIFY)) {
+               return;
+       }
+       first = state->first;
+       ins = first->prev;
+       do {
+               simplify(state, ins);
+               ins = ins->prev;
+       } while(ins != first->prev);
        ins = first;
        do {
                simplify(state, ins);
                ins = ins->next;
-       } while(ins != first);
+       }while(ins != first);
+       rebuild_ssa_form(state);
+
+       print_blocks(state, __func__, state->dbgout);
 }
 
 /*
@@ -6280,8 +10455,8 @@ static void simplify_all(struct compile_state *state)
 static void register_builtin_function(struct compile_state *state,
        const char *name, int op, struct type *rtype, ...)
 {
-       struct type *ftype, *atype, *param, **next;
-       struct triple *def, *arg, *result, *work, *last, *first;
+       struct type *ftype, *atype, *ctype, *crtype, *param, **next;
+       struct triple *def, *result, *work, *first, *retvar, *ret;
        struct hash_entry *ident;
        struct file_state file;
        int parameters;
@@ -6291,10 +10466,13 @@ static void register_builtin_function(struct compile_state *state,
 
        /* Dummy file state to get debug handling right */
        memset(&file, 0, sizeof(file));
-       file.basename = name;
+       file.basename = "<built-in>";
        file.line = 1;
+       file.report_line = 1;
+       file.report_name = file.basename;
        file.prev = state->file;
        state->file = &file;
+       state->function = name;
 
        /* Find the Parameter count */
        valid_op(state, op);
@@ -6304,7 +10482,8 @@ static void register_builtin_function(struct compile_state *state,
        }
 
        /* Find the function type */
-       ftype = new_type(TYPE_FUNCTION, rtype, 0);
+       ftype = new_type(TYPE_FUNCTION | STOR_INLINE | STOR_STATIC, rtype, 0);
+       ftype->elements = parameters;
        next = &ftype->right;
        va_start(args, rtype);
        for(i = 0; i < parameters; i++) {
@@ -6321,10 +10500,21 @@ static void register_builtin_function(struct compile_state *state,
        }
        va_end(args);
 
+       /* Get the initial closure type */
+       ctype = new_type(TYPE_JOIN, &void_type, 0);
+       ctype->elements = 1;
+
+       /* Get the return type */
+       crtype = new_type(TYPE_TUPLE, new_type(TYPE_PRODUCT, ctype, rtype), 0);
+       crtype->elements = 2;
+
        /* Generate the needed triples */
        def = triple(state, OP_LIST, ftype, 0, 0);
        first = label(state);
        RHS(def, 0) = first;
+       result = flatten(state, first, variable(state, crtype));
+       retvar = flatten(state, first, variable(state, &void_ptr_type));
+       ret = triple(state, OP_RET, &void_type, read_expr(state, retvar), 0);
 
        /* Now string them together */
        param = ftype->right;
@@ -6334,62 +10524,42 @@ static void register_builtin_function(struct compile_state *state,
                } else {
                        atype = param;
                }
-               arg = flatten(state, first, variable(state, atype));
+               flatten(state, first, variable(state, atype));
                param = param->right;
        }
-       result = 0;
-       if ((rtype->type & TYPE_MASK) != TYPE_VOID) {
-               result = flatten(state, first, variable(state, rtype));
-       }
-       MISC(def, 0) = result;
        work = new_triple(state, op, rtype, -1, parameters);
-       for(i = 0, arg = first->next; i < parameters; i++, arg = arg->next) {
-               RHS(work, i) = read_expr(state, arg);
-       }
-       if (result && ((rtype->type & TYPE_MASK) == TYPE_STRUCT)) {
-               struct triple *val;
-               /* Populate the LHS with the target registers */
-               work = flatten(state, first, work);
-               work->type = &void_type;
-               param = rtype->left;
-               if (rtype->elements != TRIPLE_LHS(work->sizes)) {
-                       internal_error(state, 0, "Invalid result type");
-               }
-               val = new_triple(state, OP_VAL_VEC, rtype, -1, -1);
-               for(i = 0; i < rtype->elements; i++) {
-                       struct triple *piece;
-                       atype = param;
-                       if ((param->type & TYPE_MASK) == TYPE_PRODUCT) {
-                               atype = param->left;
-                       }
-                       if (!TYPE_ARITHMETIC(atype->type) &&
-                               !TYPE_PTR(atype->type)) {
-                               internal_error(state, 0, "Invalid lhs type");
-                       }
-                       piece = triple(state, OP_PIECE, atype, work, 0);
-                       piece->u.cval = i;
-                       LHS(work, i) = piece;
-                       RHS(val, i) = piece;
-               }
-               work = val;
+       generate_lhs_pieces(state, work);
+       for(i = 0; i < parameters; i++) {
+               RHS(work, i) = read_expr(state, farg(state, def, i));
        }
-       if (result) {
-               work = write_expr(state, result, work);
+       if ((rtype->type & TYPE_MASK) != TYPE_VOID) {
+               work = write_expr(state, deref_index(state, result, 1), work);
        }
        work = flatten(state, first, work);
-       last = flatten(state, first, label(state));
+       flatten(state, first, label(state));
+       ret  = flatten(state, first, ret);
        name_len = strlen(name);
        ident = lookup(state, name, name_len);
+       ftype->type_ident = ident;
        symbol(state, ident, &ident->sym_ident, def, ftype);
-       
+
        state->file = file.prev;
-#if 0
-       fprintf(stdout, "\n");
-       loc(stdout, state, 0);
-       fprintf(stdout, "\n__________ builtin_function _________\n");
-       print_triple(state, def);
-       fprintf(stdout, "__________ builtin_function _________ done\n\n");
-#endif
+       state->function = 0;
+       state->main_function = 0;
+
+       if (!state->functions) {
+               state->functions = def;
+       } else {
+               insert_triple(state, state->functions, def);
+       }
+       if (state->compiler->debug & DEBUG_INLINE) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "\n");
+               loc(fp, state, 0);
+               fprintf(fp, "\n__________ %s _________\n", __FUNCTION__);
+               display_func(state, fp, def);
+               fprintf(fp, "__________ %s _________ done\n\n", __FUNCTION__);
+       }
 }
 
 static struct type *partial_struct(struct compile_state *state,
@@ -6419,7 +10589,7 @@ static struct type *register_builtin_type(struct compile_state *state,
 
        name_len = strlen(name);
        ident = lookup(state, name, name_len);
-       
+
        if ((type->type & TYPE_MASK) == TYPE_PRODUCT) {
                ulong_t elements = 0;
                struct type *field;
@@ -6430,7 +10600,7 @@ static struct type *register_builtin_type(struct compile_state *state,
                        field = field->right;
                }
                elements++;
-               symbol(state, ident, &ident->sym_struct, 0, type);
+               symbol(state, ident, &ident->sym_tag, 0, type);
                type->type_ident = ident;
                type->elements = elements;
        }
@@ -6442,25 +10612,49 @@ static struct type *register_builtin_type(struct compile_state *state,
 
 static void register_builtins(struct compile_state *state)
 {
+       struct type *div_type, *ldiv_type;
+       struct type *udiv_type, *uldiv_type;
        struct type *msr_type;
 
-       register_builtin_function(state, "__builtin_inb", OP_INB, &uchar_type, 
+       div_type = register_builtin_type(state, "__builtin_div_t",
+               partial_struct(state, "quot", &int_type,
+               partial_struct(state, "rem",  &int_type, 0)));
+       ldiv_type = register_builtin_type(state, "__builtin_ldiv_t",
+               partial_struct(state, "quot", &long_type,
+               partial_struct(state, "rem",  &long_type, 0)));
+       udiv_type = register_builtin_type(state, "__builtin_udiv_t",
+               partial_struct(state, "quot", &uint_type,
+               partial_struct(state, "rem",  &uint_type, 0)));
+       uldiv_type = register_builtin_type(state, "__builtin_uldiv_t",
+               partial_struct(state, "quot", &ulong_type,
+               partial_struct(state, "rem",  &ulong_type, 0)));
+
+       register_builtin_function(state, "__builtin_div",   OP_SDIVT, div_type,
+               &int_type, &int_type);
+       register_builtin_function(state, "__builtin_ldiv",  OP_SDIVT, ldiv_type,
+               &long_type, &long_type);
+       register_builtin_function(state, "__builtin_udiv",  OP_UDIVT, udiv_type,
+               &uint_type, &uint_type);
+       register_builtin_function(state, "__builtin_uldiv", OP_UDIVT, uldiv_type,
+               &ulong_type, &ulong_type);
+
+       register_builtin_function(state, "__builtin_inb", OP_INB, &uchar_type,
                &ushort_type);
        register_builtin_function(state, "__builtin_inw", OP_INW, &ushort_type,
                &ushort_type);
-       register_builtin_function(state, "__builtin_inl", OP_INL, &uint_type,   
+       register_builtin_function(state, "__builtin_inl", OP_INL, &uint_type,
                &ushort_type);
 
-       register_builtin_function(state, "__builtin_outb", OP_OUTB, &void_type, 
+       register_builtin_function(state, "__builtin_outb", OP_OUTB, &void_type,
                &uchar_type, &ushort_type);
-       register_builtin_function(state, "__builtin_outw", OP_OUTW, &void_type, 
+       register_builtin_function(state, "__builtin_outw", OP_OUTW, &void_type,
                &ushort_type, &ushort_type);
-       register_builtin_function(state, "__builtin_outl", OP_OUTL, &void_type, 
+       register_builtin_function(state, "__builtin_outl", OP_OUTL, &void_type,
                &uint_type, &ushort_type);
-       
-       register_builtin_function(state, "__builtin_bsf", OP_BSF, &int_type, 
+
+       register_builtin_function(state, "__builtin_bsf", OP_BSF, &int_type,
                &int_type);
-       register_builtin_function(state, "__builtin_bsr", OP_BSR, &int_type, 
+       register_builtin_function(state, "__builtin_bsr", OP_BSR, &int_type,
                &int_type);
 
        msr_type = register_builtin_type(state, "__builtin_msr_t",
@@ -6471,23 +10665,25 @@ static void register_builtins(struct compile_state *state)
                &ulong_type);
        register_builtin_function(state, "__builtin_wrmsr", OP_WRMSR, &void_type,
                &ulong_type, &ulong_type, &ulong_type);
-       
-       register_builtin_function(state, "__builtin_hlt", OP_HLT, &void_type, 
+
+       register_builtin_function(state, "__builtin_hlt", OP_HLT, &void_type,
                &void_type);
 }
 
 static struct type *declarator(
-       struct compile_state *state, struct type *type, 
+       struct compile_state *state, struct type *type,
        struct hash_entry **ident, int need_ident);
 static void decl(struct compile_state *state, struct triple *first);
 static struct type *specifier_qualifier_list(struct compile_state *state);
+#if DEBUG_ROMCC_WARNING
 static int isdecl_specifier(int tok);
+#endif
 static struct type *decl_specifiers(struct compile_state *state);
 static int istype(int tok);
 static struct triple *expr(struct compile_state *state);
 static struct triple *assignment_expr(struct compile_state *state);
 static struct type *type_name(struct compile_state *state);
-static void statement(struct compile_state *state, struct triple *fist);
+static void statement(struct compile_state *state, struct triple *first);
 
 static struct triple *call_expr(
        struct compile_state *state, struct triple *func)
@@ -6505,10 +10701,17 @@ static struct triple *call_expr(
        eat(state, TOK_LPAREN);
        /* Find the return type without any specifiers */
        type = clone_type(0, func->type->left);
-       def = new_triple(state, OP_CALL, func->type, -1, -1);
-       def->type = type;
-
-       pvals = TRIPLE_RHS(def->sizes);
+       /* Count the number of rhs entries for OP_FCALL */
+       param = func->type->right;
+       pvals = 0;
+       while((param->type & TYPE_MASK) == TYPE_PRODUCT) {
+               pvals++;
+               param = param->right;
+       }
+       if ((param->type & TYPE_MASK) != TYPE_VOID) {
+               pvals++;
+       }
+       def = new_triple(state, OP_FCALL, type, -1, pvals);
        MISC(def, 0) = func;
 
        param = func->type->right;
@@ -6539,9 +10742,8 @@ static struct triple *character_constant(struct compile_state *state)
        const signed char *str, *end;
        int c;
        int str_len;
-       eat(state, TOK_LIT_CHAR);
-       tk = &state->token[0];
-       str = tk->val.str + 1;
+       tk = eat(state, TOK_LIT_CHAR);
+       str = (signed char *)tk->val.str + 1;
        str_len = tk->str_len - 2;
        if (str_len <= 0) {
                error(state, 0, "empty character constant");
@@ -6569,13 +10771,15 @@ static struct triple *string_constant(struct compile_state *state)
        type->elements = 0;
        /* The while loop handles string concatenation */
        do {
-               eat(state, TOK_LIT_STRING);
-               tk = &state->token[0];
-               str = tk->val.str + 1;
+               tk = eat(state, TOK_LIT_STRING);
+               str = (signed char *)tk->val.str + 1;
                str_len = tk->str_len - 2;
                if (str_len < 0) {
                        error(state, 0, "negative string constant length");
                }
+               /* ignore empty string tokens */
+               if ('"' == *str && 0 == str[1])
+                       continue;
                end = str + str_len;
                ptr = buf;
                buf = xmalloc(type->elements + str_len + 1, "string_constant");
@@ -6590,6 +10794,7 @@ static struct triple *string_constant(struct compile_state *state)
        type->elements += 1;
        def = triple(state, OP_BLOBCONST, type, 0, 0);
        def->u.blob = buf;
+
        return def;
 }
 
@@ -6603,12 +10808,11 @@ static struct triple *integer_constant(struct compile_state *state)
        int u, l, decimal;
        struct type *type;
 
-       eat(state, TOK_LIT_INT);
-       tk = &state->token[0];
+       tk = eat(state, TOK_LIT_INT);
        errno = 0;
        decimal = (tk->val.str[0] != '0');
        val = strtoul(tk->val.str, &end, 0);
-       if ((val == ULONG_MAX) && (errno == ERANGE)) {
+       if ((val > ULONG_T_MAX) || ((val == ULONG_MAX) && (errno == ERANGE))) {
                error(state, 0, "Integer constant to large");
        }
        u = l = 0;
@@ -6632,25 +10836,25 @@ static struct triple *integer_constant(struct compile_state *state)
        }
        else if (l) {
                type = &long_type;
-               if (!decimal && (val > LONG_MAX)) {
+               if (!decimal && (val > LONG_T_MAX)) {
                        type = &ulong_type;
                }
        }
        else if (u) {
                type = &uint_type;
-               if (val > UINT_MAX) {
+               if (val > UINT_T_MAX) {
                        type = &ulong_type;
                }
        }
        else {
                type = &int_type;
-               if (!decimal && (val > INT_MAX) && (val <= UINT_MAX)) {
+               if (!decimal && (val > INT_T_MAX) && (val <= UINT_T_MAX)) {
                        type = &uint_type;
                }
-               else if (!decimal && (val > LONG_MAX)) {
+               else if (!decimal && (val > LONG_T_MAX)) {
                        type = &ulong_type;
                }
-               else if (val > INT_MAX) {
+               else if (val > INT_T_MAX) {
                        type = &long_type;
                }
        }
@@ -6670,10 +10874,8 @@ static struct triple *primary_expr(struct compile_state *state)
                /* Here ident is either:
                 * a varable name
                 * a function name
-                * an enumeration constant.
                 */
-               eat(state, TOK_IDENT);
-               ident = state->token[0].ident;
+               ident = eat(state, TOK_IDENT)->ident;
                if (!ident->sym_ident) {
                        error(state, 0, "%s undeclared", ident->name);
                }
@@ -6681,11 +10883,25 @@ static struct triple *primary_expr(struct compile_state *state)
                break;
        }
        case TOK_ENUM_CONST:
+       {
+               struct hash_entry *ident;
                /* Here ident is an enumeration constant */
-               eat(state, TOK_ENUM_CONST);
-               def = 0;
-               FINISHME();
+               ident = eat(state, TOK_ENUM_CONST)->ident;
+               if (!ident->sym_ident) {
+                       error(state, 0, "%s undeclared", ident->name);
+               }
+               def = ident->sym_ident->def;
+               break;
+       }
+       case TOK_MIDENT:
+       {
+               struct hash_entry *ident;
+               ident = eat(state, TOK_MIDENT)->ident;
+               warning(state, 0, "Replacing undefined macro: %s with 0",
+                       ident->name);
+               def = int_const(state, &int_type, 0);
                break;
+       }
        case TOK_LPAREN:
                eat(state, TOK_LPAREN);
                def = expr(state);
@@ -6736,8 +10952,7 @@ static struct triple *postfix_expr(struct compile_state *state)
                {
                        struct hash_entry *field;
                        eat(state, TOK_DOT);
-                       eat(state, TOK_IDENT);
-                       field = state->token[0].ident;
+                       field = eat(state, TOK_IDENT)->ident;
                        def = deref_field(state, def, field);
                        break;
                }
@@ -6745,8 +10960,7 @@ static struct triple *postfix_expr(struct compile_state *state)
                {
                        struct hash_entry *field;
                        eat(state, TOK_ARROW);
-                       eat(state, TOK_IDENT);
-                       field = state->token[0].ident;
+                       field = eat(state, TOK_IDENT)->ident;
                        def = mk_deref_expr(state, read_expr(state, def));
                        def = deref_field(state, def, field);
                        break;
@@ -6834,7 +11048,7 @@ static struct triple *unary_expr(struct compile_state *state)
                        type = expr->type;
                        release_expr(state, expr);
                }
-               def = int_const(state, &ulong_type, size_of(state, type));
+               def = int_const(state, &ulong_type, size_of_in_bytes(state, type));
                break;
        }
        case TOK_ALIGNOF:
@@ -6855,7 +11069,25 @@ static struct triple *unary_expr(struct compile_state *state)
                        type = expr->type;
                        release_expr(state, expr);
                }
-               def = int_const(state, &ulong_type, align_of(state, type));
+               def = int_const(state, &ulong_type, align_of_in_bytes(state, type));
+               break;
+       }
+       case TOK_MDEFINED:
+       {
+               /* We only come here if we are called from the preprocessor */
+               struct hash_entry *ident;
+               int parens;
+               eat(state, TOK_MDEFINED);
+               parens = 0;
+               if (pp_peek(state) == TOK_LPAREN) {
+                       pp_eat(state, TOK_LPAREN);
+                       parens = 1;
+               }
+               ident = pp_eat(state, TOK_MIDENT)->ident;
+               if (parens) {
+                       eat(state, TOK_RPAREN);
+               }
+               def = int_const(state, &int_type, ident->sym_define != 0);
                break;
        }
        default:
@@ -6876,8 +11108,7 @@ static struct triple *cast_expr(struct compile_state *state)
                eat(state, TOK_LPAREN);
                type = type_name(state);
                eat(state, TOK_RPAREN);
-               def = read_expr(state, cast_expr(state));
-               def = triple(state, OP_COPY, type, def, 0);
+               def = mk_cast_expr(state, type, cast_expr(state));
        }
        else {
                def = unary_expr(state);
@@ -6895,7 +11126,8 @@ static struct triple *mult_expr(struct compile_state *state)
                struct type *result_type;
                int tok, op, sign;
                done = 0;
-               switch(tok = (peek(state))) {
+               tok = peek(state);
+               switch(tok) {
                case TOK_STAR:
                case TOK_DIV:
                case TOK_MOD:
@@ -6970,8 +11202,8 @@ static struct triple *shift_expr(struct compile_state *state)
                        right = read_expr(state, add_expr(state));
                        integral(state, right);
                        right = integral_promotion(state, right);
-                       
-                       op = (tok == TOK_SL)? OP_SL : 
+
+                       op = (tok == TOK_SL)? OP_SL :
                                is_signed(left->type)? OP_SSR: OP_USR;
 
                        def = triple(state, op, left->type, left, right);
@@ -6986,7 +11218,9 @@ static struct triple *shift_expr(struct compile_state *state)
 
 static struct triple *relational_expr(struct compile_state *state)
 {
+#if DEBUG_ROMCC_WARNINGS
 #warning "Extend relational exprs to work on more than arithmetic types"
+#endif
        struct triple *def;
        int done;
        def = shift_expr(state);
@@ -7029,7 +11263,9 @@ static struct triple *relational_expr(struct compile_state *state)
 
 static struct triple *equality_expr(struct compile_state *state)
 {
+#if DEBUG_ROMCC_WARNINGS
 #warning "Extend equality exprs to work on more than arithmetic types"
+#endif
        struct triple *def;
        int done;
        def = relational_expr(state);
@@ -7122,7 +11358,7 @@ static struct triple *land_expr(struct compile_state *state)
                right = read_expr(state, or_expr(state));
                bool(state, right);
 
-               def = triple(state, OP_LAND, &int_type,
+               def = mkland_expr(state,
                        ltrue_expr(state, left),
                        ltrue_expr(state, right));
        }
@@ -7140,8 +11376,8 @@ static struct triple *lor_expr(struct compile_state *state)
                eat(state, TOK_LOGOR);
                right = read_expr(state, land_expr(state));
                bool(state, right);
-               
-               def = triple(state, OP_LOR, &int_type,
+
+               def = mklor_expr(state,
                        ltrue_expr(state, left),
                        ltrue_expr(state, right));
        }
@@ -7161,34 +11397,146 @@ static struct triple *conditional_expr(struct compile_state *state)
                eat(state, TOK_COLON);
                right = read_expr(state, conditional_expr(state));
 
-               def = cond_expr(state, test, left, right);
+               def = mkcond_expr(state, test, left, right);
        }
        return def;
 }
 
+struct cv_triple {
+       struct triple *val;
+       int id;
+};
+
+static void set_cv(struct compile_state *state, struct cv_triple *cv,
+       struct triple *dest, struct triple *val)
+{
+       if (cv[dest->id].val) {
+               free_triple(state, cv[dest->id].val);
+       }
+       cv[dest->id].val = val;
+}
+static struct triple *get_cv(struct compile_state *state, struct cv_triple *cv,
+       struct triple *src)
+{
+       return cv[src->id].val;
+}
+
 static struct triple *eval_const_expr(
        struct compile_state *state, struct triple *expr)
 {
        struct triple *def;
-       struct triple *head, *ptr;
-       head = label(state); /* dummy initial triple */
-       flatten(state, head, expr);
-       for(ptr = head->next; ptr != head; ptr = ptr->next) {
-               simplify(state, ptr);
-       }
-       /* Remove the constant value the tail of the list */
-       def = head->prev;
-       def->prev->next = def->next;
-       def->next->prev = def->prev;
-       def->next = def->prev = def;
-       if (!is_const(def)) {
-               internal_error(state, 0, "Not a constant expression");
+       if (is_const(expr)) {
+               def = expr;
        }
-       /* Free the intermediate expressions */
-       while(head->next != head) {
-               release_triple(state, head->next);
+       else {
+               /* If we don't start out as a constant simplify into one */
+               struct triple *head, *ptr;
+               struct cv_triple *cv;
+               int i, count;
+               head = label(state); /* dummy initial triple */
+               flatten(state, head, expr);
+               count = 1;
+               for(ptr = head->next; ptr != head; ptr = ptr->next) {
+                       count++;
+               }
+               cv = xcmalloc(sizeof(struct cv_triple)*count, "const value vector");
+               i = 1;
+               for(ptr = head->next; ptr != head; ptr = ptr->next) {
+                       cv[i].val = 0;
+                       cv[i].id  = ptr->id;
+                       ptr->id   = i;
+                       i++;
+               }
+               ptr = head->next;
+               do {
+                       valid_ins(state, ptr);
+                       if ((ptr->op == OP_PHI) || (ptr->op == OP_LIST)) {
+                               internal_error(state, ptr,
+                                       "unexpected %s in constant expression",
+                                       tops(ptr->op));
+                       }
+                       else if (ptr->op == OP_LIST) {
+                       }
+                       else if (triple_is_structural(state, ptr)) {
+                               ptr = ptr->next;
+                       }
+                       else if (triple_is_ubranch(state, ptr)) {
+                               ptr = TARG(ptr, 0);
+                       }
+                       else if (triple_is_cbranch(state, ptr)) {
+                               struct triple *cond_val;
+                               cond_val = get_cv(state, cv, RHS(ptr, 0));
+                               if (!cond_val || !is_const(cond_val) ||
+                                       (cond_val->op != OP_INTCONST))
+                               {
+                                       internal_error(state, ptr, "bad branch condition");
+                               }
+                               if (cond_val->u.cval == 0) {
+                                       ptr = ptr->next;
+                               } else {
+                                       ptr = TARG(ptr, 0);
+                               }
+                       }
+                       else if (triple_is_branch(state, ptr)) {
+                               error(state, ptr, "bad branch type in constant expression");
+                       }
+                       else if (ptr->op == OP_WRITE) {
+                               struct triple *val;
+                               val = get_cv(state, cv, RHS(ptr, 0));
+
+                               set_cv(state, cv, MISC(ptr, 0),
+                                       copy_triple(state, val));
+                               set_cv(state, cv, ptr,
+                                       copy_triple(state, val));
+                               ptr = ptr->next;
+                       }
+                       else if (ptr->op == OP_READ) {
+                               set_cv(state, cv, ptr,
+                                       copy_triple(state,
+                                               get_cv(state, cv, RHS(ptr, 0))));
+                               ptr = ptr->next;
+                       }
+                       else if (triple_is_pure(state, ptr, cv[ptr->id].id)) {
+                               struct triple *val, **rhs;
+                               val = copy_triple(state, ptr);
+                               rhs = triple_rhs(state, val, 0);
+                               for(; rhs; rhs = triple_rhs(state, val, rhs)) {
+                                       if (!*rhs) {
+                                               internal_error(state, ptr, "Missing rhs");
+                                       }
+                                       *rhs = get_cv(state, cv, *rhs);
+                               }
+                               simplify(state, val);
+                               set_cv(state, cv, ptr, val);
+                               ptr = ptr->next;
+                       }
+                       else {
+                               error(state, ptr, "impure operation in constant expression");
+                       }
+
+               } while(ptr != head);
+
+               /* Get the result value */
+               def = get_cv(state, cv, head->prev);
+               cv[head->prev->id].val = 0;
+
+               /* Free the temporary values */
+               for(i = 0; i < count; i++) {
+                       if (cv[i].val) {
+                               free_triple(state, cv[i].val);
+                               cv[i].val = 0;
+                       }
+               }
+               xfree(cv);
+               /* Free the intermediate expressions */
+               while(head->next != head) {
+                       release_triple(state, head->next);
+               }
+               free_triple(state, head);
+       }
+       if (!is_const(def)) {
+               error(state, expr, "Not a constant expression");
        }
-       free_triple(state, head);
        return def;
 }
 
@@ -7211,7 +11559,7 @@ static struct triple *assignment_expr(struct compile_state *state)
         * a larger set of statements than standard C.   As long
         * as the subset of the grammar that is standard C behaves
         * correctly this should cause no problems.
-        * 
+        *
         * For the extra token strings accepted by the grammar
         * none of them should produce a valid lvalue, so they
         * should not produce functioning programs.
@@ -7224,7 +11572,7 @@ static struct triple *assignment_expr(struct compile_state *state)
        case TOK_EQ:
                lvalue(state, left);
                eat(state, TOK_EQ);
-               def = write_expr(state, left, 
+               def = write_expr(state, left,
                        read_expr(state, assignment_expr(state)));
                break;
        case TOK_TIMESEQ:
@@ -7244,7 +11592,7 @@ static struct triple *assignment_expr(struct compile_state *state)
                case TOK_MODEQ:   op = sign? OP_SMOD : OP_UMOD; break;
                }
                def = write_expr(state, left,
-                       triple(state, op, left->type, 
+                       triple(state, op, left->type,
                                read_expr(state, left), right));
                break;
        case TOK_PLUSEQ:
@@ -7280,7 +11628,7 @@ static struct triple *assignment_expr(struct compile_state *state)
                case TOK_OREQ:  op = OP_OR; break;
                }
                def = write_expr(state, left,
-                       triple(state, op, left->type, 
+                       triple(state, op, left->type,
                                read_expr(state, left), right));
                break;
        }
@@ -7292,11 +11640,8 @@ static struct triple *expr(struct compile_state *state)
        struct triple *def;
        def = assignment_expr(state);
        while(peek(state) == TOK_COMMA) {
-               struct triple *left, *right;
-               left = def;
                eat(state, TOK_COMMA);
-               right = assignment_expr(state);
-               def = triple(state, OP_COMMA, right->type, left, right);
+               def = mkprog(state, def, assignment_expr(state), 0UL);
        }
        return def;
 }
@@ -7304,7 +11649,11 @@ static struct triple *expr(struct compile_state *state)
 static void expr_statement(struct compile_state *state, struct triple *first)
 {
        if (peek(state) != TOK_SEMI) {
-               flatten(state, first, expr(state));
+               /* lvalue conversions always apply except when certian operators
+                * are applied.  I apply the lvalue conversions here
+                * as I know no more operators will be applied.
+                */
+               flatten(state, first, lvalue_conversion(state, expr(state)));
        }
        eat(state, TOK_SEMI);
 }
@@ -7356,7 +11705,7 @@ static void for_statement(struct compile_state *state, struct triple *first)
        head = test = tail = jmp1 = jmp2 = 0;
        if (peek(state) != TOK_SEMI) {
                head = expr(state);
-       } 
+       }
        eat(state, TOK_SEMI);
        if (peek(state) != TOK_SEMI) {
                test = expr(state);
@@ -7477,7 +11826,9 @@ static void return_statement(struct compile_state *state, struct triple *first)
        int last;
        eat(state, TOK_RETURN);
 
+#if DEBUG_ROMCC_WARNINGS
 #warning "FIXME implement a more general excess branch elimination"
+#endif
        val = 0;
        /* If we have a return value do some more work */
        if (peek(state) != TOK_SEMI) {
@@ -7486,13 +11837,14 @@ static void return_statement(struct compile_state *state, struct triple *first)
        eat(state, TOK_SEMI);
 
        /* See if this last statement in a function */
-       last = ((peek(state) == TOK_RBRACE) && 
+       last = ((peek(state) == TOK_RBRACE) &&
                (state->scope_depth == GLOBAL_SCOPE_DEPTH +2));
 
        /* Find the return variable */
-       var = MISC(state->main_function, 0);
+       var = fresult(state, state->main_function);
+
        /* Find the return destination */
-       dest = RHS(state->main_function, 0)->prev;
+       dest = state->i_return->sym_ident->def;
        mv = jmp = 0;
        /* If needed generate a jump instruction */
        if (!last) {
@@ -7500,7 +11852,7 @@ static void return_statement(struct compile_state *state, struct triple *first)
        }
        /* If needed generate an assignment instruction */
        if (val) {
-               mv = write_expr(state, var, val);
+               mv = write_expr(state, deref_index(state, var, 1), val);
        }
        /* Now put the code together */
        if (mv) {
@@ -7538,55 +11890,160 @@ static void continue_statement(struct compile_state *state, struct triple *first
 
 static void goto_statement(struct compile_state *state, struct triple *first)
 {
-       FINISHME();
+       struct hash_entry *ident;
        eat(state, TOK_GOTO);
-       eat(state, TOK_IDENT);
+       ident = eat(state, TOK_IDENT)->ident;
+       if (!ident->sym_label) {
+               /* If this is a forward branch allocate the label now,
+                * it will be flattend in the appropriate location later.
+                */
+               struct triple *ins;
+               ins = label(state);
+               label_symbol(state, ident, ins, FUNCTION_SCOPE_DEPTH);
+       }
        eat(state, TOK_SEMI);
-       error(state, 0, "goto is not implemeted");
-       FINISHME();
+
+       flatten(state, first, branch(state, ident->sym_label->def, 0));
 }
 
 static void labeled_statement(struct compile_state *state, struct triple *first)
 {
-       FINISHME();
-       eat(state, TOK_IDENT);
+       struct triple *ins;
+       struct hash_entry *ident;
+
+       ident = eat(state, TOK_IDENT)->ident;
+       if (ident->sym_label && ident->sym_label->def) {
+               ins = ident->sym_label->def;
+               put_occurance(ins->occurance);
+               ins->occurance = new_occurance(state);
+       }
+       else {
+               ins = label(state);
+               label_symbol(state, ident, ins, FUNCTION_SCOPE_DEPTH);
+       }
+       if (ins->id & TRIPLE_FLAG_FLATTENED) {
+               error(state, 0, "label %s already defined", ident->name);
+       }
+       flatten(state, first, ins);
+
        eat(state, TOK_COLON);
        statement(state, first);
-       error(state, 0, "labeled statements are not implemented");
-       FINISHME();
 }
 
 static void switch_statement(struct compile_state *state, struct triple *first)
 {
-       FINISHME();
+       struct triple *value, *top, *end, *dbranch;
+       struct hash_entry *ident;
+
+       /* See if we have a valid switch statement */
        eat(state, TOK_SWITCH);
        eat(state, TOK_LPAREN);
-       expr(state);
+       value = expr(state);
+       integral(state, value);
+       value = read_expr(state, value);
        eat(state, TOK_RPAREN);
+       /* Generate the needed pieces */
+       top = label(state);
+       end = label(state);
+       dbranch = branch(state, end, 0);
+       /* Remember where case branches and break goes */
+       start_scope(state);
+       ident = state->i_switch;
+       symbol(state, ident, &ident->sym_ident, value, value->type);
+       ident = state->i_case;
+       symbol(state, ident, &ident->sym_ident, top, top->type);
+       ident = state->i_break;
+       symbol(state, ident, &ident->sym_ident, end, end->type);
+       ident = state->i_default;
+       symbol(state, ident, &ident->sym_ident, dbranch, dbranch->type);
+       /* Thread them together */
+       flatten(state, first, value);
+       flatten(state, first, top);
+       flatten(state, first, dbranch);
        statement(state, first);
-       error(state, 0, "switch statements are not implemented");
-       FINISHME();
+       flatten(state, first, end);
+       /* Cleanup the switch scope */
+       end_scope(state);
 }
 
 static void case_statement(struct compile_state *state, struct triple *first)
 {
-       FINISHME();
+       struct triple *cvalue, *dest, *test, *jmp;
+       struct triple *ptr, *value, *top, *dbranch;
+
+       /* See if w have a valid case statement */
        eat(state, TOK_CASE);
-       constant_expr(state);
+       cvalue = constant_expr(state);
+       integral(state, cvalue);
+       if (cvalue->op != OP_INTCONST) {
+               error(state, 0, "integer constant expected");
+       }
        eat(state, TOK_COLON);
+       if (!state->i_case->sym_ident) {
+               error(state, 0, "case statement not within a switch");
+       }
+
+       /* Lookup the interesting pieces */
+       top = state->i_case->sym_ident->def;
+       value = state->i_switch->sym_ident->def;
+       dbranch = state->i_default->sym_ident->def;
+
+       /* See if this case label has already been used */
+       for(ptr = top; ptr != dbranch; ptr = ptr->next) {
+               if (ptr->op != OP_EQ) {
+                       continue;
+               }
+               if (RHS(ptr, 1)->u.cval == cvalue->u.cval) {
+                       error(state, 0, "duplicate case %d statement",
+                               cvalue->u.cval);
+               }
+       }
+       /* Generate the needed pieces */
+       dest = label(state);
+       test = triple(state, OP_EQ, &int_type, value, cvalue);
+       jmp = branch(state, dest, test);
+       /* Thread the pieces together */
+       flatten(state, dbranch, test);
+       flatten(state, dbranch, jmp);
+       flatten(state, dbranch, label(state));
+       flatten(state, first, dest);
        statement(state, first);
-       error(state, 0, "case statements are not implemented");
-       FINISHME();
 }
 
 static void default_statement(struct compile_state *state, struct triple *first)
 {
-       FINISHME();
+       struct triple *dest;
+       struct triple *dbranch, *end;
+
+       /* See if we have a valid default statement */
        eat(state, TOK_DEFAULT);
        eat(state, TOK_COLON);
+
+       if (!state->i_case->sym_ident) {
+               error(state, 0, "default statement not within a switch");
+       }
+
+       /* Lookup the interesting pieces */
+       dbranch = state->i_default->sym_ident->def;
+       end = state->i_break->sym_ident->def;
+
+       /* See if a default statement has already happened */
+       if (TARG(dbranch, 0) != end) {
+               error(state, 0, "duplicate default statement");
+       }
+
+       /* Generate the needed pieces */
+       dest = label(state);
+
+       /* Blame the branch on the default statement */
+       put_occurance(dbranch->occurance);
+       dbranch->occurance = new_occurance(state);
+
+       /* Thread the pieces together */
+       TARG(dbranch, 0) = dest;
+       use_triple(dest, dbranch);
+       flatten(state, first, dest);
        statement(state, first);
-       error(state, 0, "default statements are not implemented");
-       FINISHME();
 }
 
 static void asm_statement(struct compile_state *state, struct triple *first)
@@ -7598,7 +12055,9 @@ static void asm_statement(struct compile_state *state, struct triple *first)
        } out_param[MAX_LHS], in_param[MAX_RHS], clob_param[MAX_LHS];
        struct triple *def, *asm_str;
        int out, in, clobbers, more, colons, i;
+       int flags;
 
+       flags = 0;
        eat(state, TOK_ASM);
        /* For now ignore the qualifiers */
        switch(peek(state)) {
@@ -7607,6 +12066,7 @@ static void asm_statement(struct compile_state *state, struct triple *first)
                break;
        case TOK_VOLATILE:
                eat(state, TOK_VOLATILE);
+               flags |= TRIPLE_FLAG_VOLATILE;
                break;
        }
        eat(state, TOK_LPAREN);
@@ -7622,11 +12082,17 @@ static void asm_statement(struct compile_state *state, struct triple *first)
                while(more) {
                        struct triple *var;
                        struct triple *constraint;
+                       char *str;
                        more = 0;
                        if (out > MAX_LHS) {
                                error(state, 0, "Maximum output count exceeded.");
                        }
                        constraint = string_constant(state);
+                       str = constraint->u.blob;
+                       if (str[0] != '=') {
+                               error(state, 0, "Output constraint does not start with =");
+                       }
+                       constraint->u.blob = str + 1;
                        eat(state, TOK_LPAREN);
                        var = conditional_expr(state);
                        eat(state, TOK_RPAREN);
@@ -7649,11 +12115,20 @@ static void asm_statement(struct compile_state *state, struct triple *first)
                while(more) {
                        struct triple *val;
                        struct triple *constraint;
+                       char *str;
                        more = 0;
                        if (in > MAX_RHS) {
                                error(state, 0, "Maximum input count exceeded.");
                        }
                        constraint = string_constant(state);
+                       str = constraint->u.blob;
+                       if (digitp(str[0] && str[1] == '\0')) {
+                               int val;
+                               val = digval(str[0]);
+                               if ((val < 0) || (val >= out)) {
+                                       error(state, 0, "Invalid input constraint %d", val);
+                               }
+                       }
                        eat(state, TOK_LPAREN);
                        val = conditional_expr(state);
                        eat(state, TOK_RPAREN);
@@ -7680,7 +12155,6 @@ static void asm_statement(struct compile_state *state, struct triple *first)
                                error(state, 0, "Maximum clobber limit exceeded.");
                        }
                        clobber = string_constant(state);
-                       eat(state, TOK_RPAREN);
 
                        clob_param[clobbers].constraint = clobber;
                        if (peek(state) == TOK_COMMA) {
@@ -7700,41 +12174,86 @@ static void asm_statement(struct compile_state *state, struct triple *first)
 
        def = new_triple(state, OP_ASM, &void_type, clobbers + out, in);
        def->u.ainfo = info;
-       for(i = 0; i < in; i++) {
-               struct triple *constraint;
-               constraint = in_param[i].constraint;
-               info->tmpl.rhs[i] = arch_reg_constraint(state, 
-                       in_param[i].expr->type, constraint->u.blob);
+       def->id |= flags;
 
-               RHS(def, i) = read_expr(state,in_param[i].expr);
-               free_triple(state, constraint);
-       }
-       flatten(state, first, def);
+       /* Find the register constraints */
        for(i = 0; i < out; i++) {
-               struct triple *piece;
                struct triple *constraint;
                constraint = out_param[i].constraint;
                info->tmpl.lhs[i] = arch_reg_constraint(state,
                        out_param[i].expr->type, constraint->u.blob);
-
-               piece = triple(state, OP_PIECE, out_param[i].expr->type, def, 0);
-               piece->u.cval = i;
-               LHS(def, i) = piece;
-               flatten(state, first,
-                       write_expr(state, out_param[i].expr, piece));
                free_triple(state, constraint);
        }
        for(; i - out < clobbers; i++) {
-               struct triple *piece;
                struct triple *constraint;
                constraint = clob_param[i - out].constraint;
                info->tmpl.lhs[i] = arch_reg_clobber(state, constraint->u.blob);
+               free_triple(state, constraint);
+       }
+       for(i = 0; i < in; i++) {
+               struct triple *constraint;
+               const char *str;
+               constraint = in_param[i].constraint;
+               str = constraint->u.blob;
+               if (digitp(str[0]) && str[1] == '\0') {
+                       struct reg_info cinfo;
+                       int val;
+                       val = digval(str[0]);
+                       cinfo.reg = info->tmpl.lhs[val].reg;
+                       cinfo.regcm = arch_type_to_regcm(state, in_param[i].expr->type);
+                       cinfo.regcm &= info->tmpl.lhs[val].regcm;
+                       if (cinfo.reg == REG_UNSET) {
+                               cinfo.reg = REG_VIRT0 + val;
+                       }
+                       if (cinfo.regcm == 0) {
+                               error(state, 0, "No registers for %d", val);
+                       }
+                       info->tmpl.lhs[val] = cinfo;
+                       info->tmpl.rhs[i]   = cinfo;
+
+               } else {
+                       info->tmpl.rhs[i] = arch_reg_constraint(state,
+                               in_param[i].expr->type, str);
+               }
+               free_triple(state, constraint);
+       }
 
-               piece = triple(state, OP_PIECE, &void_type, def, 0);
+       /* Now build the helper expressions */
+       for(i = 0; i < in; i++) {
+               RHS(def, i) = read_expr(state, in_param[i].expr);
+       }
+       flatten(state, first, def);
+       for(i = 0; i < (out + clobbers); i++) {
+               struct type *type;
+               struct triple *piece;
+               if (i < out) {
+                       type = out_param[i].expr->type;
+               } else {
+                       size_t size = arch_reg_size(info->tmpl.lhs[i].reg);
+                       if (size >= SIZEOF_LONG) {
+                               type = &ulong_type;
+                       }
+                       else if (size >= SIZEOF_INT) {
+                               type = &uint_type;
+                       }
+                       else if (size >= SIZEOF_SHORT) {
+                               type = &ushort_type;
+                       }
+                       else {
+                               type = &uchar_type;
+                       }
+               }
+               piece = triple(state, OP_PIECE, type, def, 0);
                piece->u.cval = i;
                LHS(def, i) = piece;
                flatten(state, first, piece);
-               free_triple(state, constraint);
+       }
+       /* And write the helpers to their destinations */
+       for(i = 0; i < out; i++) {
+               struct triple *piece;
+               piece = LHS(def, i);
+               flatten(state, first,
+                       write_expr(state, out_param[i].expr, piece));
        }
 }
 
@@ -7790,7 +12309,7 @@ static void statement(struct compile_state *state, struct triple *first)
                compound_statement(state, first);
        }
        else if (tok == TOK_IF) {
-               if_statement(state, first); 
+               if_statement(state, first);
        }
        else if (tok == TOK_FOR) {
                for_statement(state, first);
@@ -7820,7 +12339,7 @@ static void statement(struct compile_state *state, struct triple *first)
                asm_statement(state, first);
        }
        else if ((tok == TOK_IDENT) && (peek2(state) == TOK_COLON)) {
-               labeled_statement(state, first); 
+               labeled_statement(state, first);
        }
        else if (tok == TOK_CASE) {
                case_statement(state, first);
@@ -7842,7 +12361,7 @@ static struct type *param_decl(struct compile_state *state)
        struct type *type;
        struct hash_entry *ident;
        /* Cheat so the declarator will know we are not global */
-       start_scope(state); 
+       start_scope(state);
        ident = 0;
        type = decl_specifiers(state);
        type = declarator(state, type, &ident, 0);
@@ -7854,8 +12373,9 @@ static struct type *param_decl(struct compile_state *state)
 static struct type *param_type_list(struct compile_state *state, struct type *type)
 {
        struct type *ftype, **next;
-       ftype = new_type(TYPE_FUNCTION, type, param_decl(state));
+       ftype = new_type(TYPE_FUNCTION | (type->type & STOR_MASK), type, param_decl(state));
        next = &ftype->right;
+       ftype->elements = 1;
        while(peek(state) == TOK_COMMA) {
                eat(state, TOK_COMMA);
                if (peek(state) == TOK_DOTS) {
@@ -7865,12 +12385,12 @@ static struct type *param_type_list(struct compile_state *state, struct type *ty
                else {
                        *next = new_type(TYPE_PRODUCT, *next, param_decl(state));
                        next = &((*next)->right);
+                       ftype->elements++;
                }
        }
        return ftype;
 }
 
-
 static struct type *type_name(struct compile_state *state)
 {
        struct type *type;
@@ -7881,25 +12401,26 @@ static struct type *type_name(struct compile_state *state)
 }
 
 static struct type *direct_declarator(
-       struct compile_state *state, struct type *type, 
-       struct hash_entry **ident, int need_ident)
+       struct compile_state *state, struct type *type,
+       struct hash_entry **pident, int need_ident)
 {
+       struct hash_entry *ident;
        struct type *outer;
        int op;
        outer = 0;
        arrays_complete(state, type);
        switch(peek(state)) {
        case TOK_IDENT:
-               eat(state, TOK_IDENT);
+               ident = eat(state, TOK_IDENT)->ident;
                if (!ident) {
                        error(state, 0, "Unexpected identifier found");
                }
                /* The name of what we are declaring */
-               *ident = state->token[0].ident;
+               *pident = ident;
                break;
        case TOK_LPAREN:
                eat(state, TOK_LPAREN);
-               outer = declarator(state, type, ident, need_ident);
+               outer = declarator(state, type, pident, need_ident);
                eat(state, TOK_RPAREN);
                break;
        default:
@@ -7958,28 +12479,26 @@ static struct type *direct_declarator(
 }
 
 static struct type *declarator(
-       struct compile_state *state, struct type *type, 
-       struct hash_entry **ident, int need_ident)
+       struct compile_state *state, struct type *type,
+       struct hash_entry **pident, int need_ident)
 {
        while(peek(state) == TOK_STAR) {
                eat(state, TOK_STAR);
                type = new_type(TYPE_POINTER | (type->type & STOR_MASK), type, 0);
        }
-       type = direct_declarator(state, type, ident, need_ident);
+       type = direct_declarator(state, type, pident, need_ident);
        return type;
 }
 
-
 static struct type *typedef_name(
        struct compile_state *state, unsigned int specifiers)
 {
        struct hash_entry *ident;
        struct type *type;
-       eat(state, TOK_TYPE_NAME);
-       ident = state->token[0].ident;
+       ident = eat(state, TOK_TYPE_NAME)->ident;
        type = ident->sym_ident->type;
        specifiers |= type->type & QUAL_MASK;
-       if ((specifiers & (STOR_MASK | QUAL_MASK)) != 
+       if ((specifiers & (STOR_MASK | QUAL_MASK)) !=
                (type->type & (STOR_MASK | QUAL_MASK))) {
                type = clone_type(specifiers, type);
        }
@@ -7987,59 +12506,101 @@ static struct type *typedef_name(
 }
 
 static struct type *enum_specifier(
-       struct compile_state *state, unsigned int specifiers)
+       struct compile_state *state, unsigned int spec)
 {
+       struct hash_entry *ident;
+       ulong_t base;
        int tok;
-       struct type *type;
-       type = 0;
-       FINISHME();
+       struct type *enum_type;
+       enum_type = 0;
+       ident = 0;
        eat(state, TOK_ENUM);
        tok = peek(state);
-       if (tok == TOK_IDENT) {
-               eat(state, TOK_IDENT);
+       if ((tok == TOK_IDENT) || (tok == TOK_ENUM_CONST) || (tok == TOK_TYPE_NAME)) {
+               ident = eat(state, tok)->ident;
        }
-       if ((tok != TOK_IDENT) || (peek(state) == TOK_LBRACE)) {
+       base = 0;
+       if (!ident || (peek(state) == TOK_LBRACE)) {
+               struct type **next;
                eat(state, TOK_LBRACE);
+               enum_type = new_type(TYPE_ENUM | spec, 0, 0);
+               enum_type->type_ident = ident;
+               next = &enum_type->right;
                do {
-                       eat(state, TOK_IDENT);
+                       struct hash_entry *eident;
+                       struct triple *value;
+                       struct type *entry;
+                       eident = eat(state, TOK_IDENT)->ident;
+                       if (eident->sym_ident) {
+                               error(state, 0, "%s already declared",
+                                       eident->name);
+                       }
+                       eident->tok = TOK_ENUM_CONST;
                        if (peek(state) == TOK_EQ) {
+                               struct triple *val;
                                eat(state, TOK_EQ);
-                               constant_expr(state);
+                               val = constant_expr(state);
+                               integral(state, val);
+                               base = val->u.cval;
                        }
+                       value = int_const(state, &int_type, base);
+                       symbol(state, eident, &eident->sym_ident, value, &int_type);
+                       entry = new_type(TYPE_LIST, 0, 0);
+                       entry->field_ident = eident;
+                       *next = entry;
+                       next = &entry->right;
+                       base += 1;
                        if (peek(state) == TOK_COMMA) {
                                eat(state, TOK_COMMA);
                        }
                } while(peek(state) != TOK_RBRACE);
                eat(state, TOK_RBRACE);
+               if (ident) {
+                       symbol(state, ident, &ident->sym_tag, 0, enum_type);
+               }
        }
-       FINISHME();
-       return type;
+       if (ident && ident->sym_tag &&
+               ident->sym_tag->type &&
+               ((ident->sym_tag->type->type & TYPE_MASK) == TYPE_ENUM)) {
+               enum_type = clone_type(spec, ident->sym_tag->type);
+       }
+       else if (ident && !enum_type) {
+               error(state, 0, "enum %s undeclared", ident->name);
+       }
+       return enum_type;
 }
 
-#if 0
 static struct type *struct_declarator(
        struct compile_state *state, struct type *type, struct hash_entry **ident)
 {
-       int tok;
-#warning "struct_declarator is complicated because of bitfields, kill them?"
-       tok = peek(state);
-       if (tok != TOK_COLON) {
+       if (peek(state) != TOK_COLON) {
                type = declarator(state, type, ident, 1);
        }
-       if ((tok == TOK_COLON) || (peek(state) == TOK_COLON)) {
+       if (peek(state) == TOK_COLON) {
+               struct triple *value;
                eat(state, TOK_COLON);
-               constant_expr(state);
+               value = constant_expr(state);
+               if (value->op != OP_INTCONST) {
+                       error(state, 0, "Invalid constant expression");
+               }
+               if (value->u.cval > size_of(state, type)) {
+                       error(state, 0, "bitfield larger than base type");
+               }
+               if (!TYPE_INTEGER(type->type) || ((type->type & TYPE_MASK) == TYPE_BITFIELD)) {
+                       error(state, 0, "bitfield base not an integer type");
+               }
+               type = new_type(TYPE_BITFIELD, type, 0);
+               type->elements = value->u.cval;
        }
-       FINISHME();
        return type;
 }
-#endif
 
 static struct type *struct_or_union_specifier(
-       struct compile_state *state, unsigned int specifiers)
+       struct compile_state *state, unsigned int spec)
 {
        struct type *struct_type;
        struct hash_entry *ident;
+       unsigned int type_main;
        unsigned int type_join;
        int tok;
        struct_type = 0;
@@ -8047,38 +12608,39 @@ static struct type *struct_or_union_specifier(
        switch(peek(state)) {
        case TOK_STRUCT:
                eat(state, TOK_STRUCT);
+               type_main = TYPE_STRUCT;
                type_join = TYPE_PRODUCT;
                break;
        case TOK_UNION:
                eat(state, TOK_UNION);
+               type_main = TYPE_UNION;
                type_join = TYPE_OVERLAP;
-               error(state, 0, "unions not yet supported\n");
                break;
        default:
                eat(state, TOK_STRUCT);
+               type_main = TYPE_STRUCT;
                type_join = TYPE_PRODUCT;
                break;
        }
        tok = peek(state);
-       if ((tok == TOK_IDENT) || (tok == TOK_TYPE_NAME)) {
-               eat(state, tok);
-               ident = state->token[0].ident;
+       if ((tok == TOK_IDENT) || (tok == TOK_ENUM_CONST) || (tok == TOK_TYPE_NAME)) {
+               ident = eat(state, tok)->ident;
        }
        if (!ident || (peek(state) == TOK_LBRACE)) {
                ulong_t elements;
+               struct type **next;
                elements = 0;
                eat(state, TOK_LBRACE);
+               next = &struct_type;
                do {
                        struct type *base_type;
-                       struct type **next;
                        int done;
                        base_type = specifier_qualifier_list(state);
-                       next = &struct_type;
                        do {
                                struct type *type;
                                struct hash_entry *fident;
                                done = 1;
-                               type = declarator(state, base_type, &fident, 1);
+                               type = struct_declarator(state, base_type, &fident);
                                elements++;
                                if (peek(state) == TOK_COMMA) {
                                        done = 0;
@@ -8096,16 +12658,22 @@ static struct type *struct_or_union_specifier(
                        eat(state, TOK_SEMI);
                } while(peek(state) != TOK_RBRACE);
                eat(state, TOK_RBRACE);
-               struct_type = new_type(TYPE_STRUCT, struct_type, 0);
+               struct_type = new_type(type_main | spec, struct_type, 0);
                struct_type->type_ident = ident;
                struct_type->elements = elements;
-               symbol(state, ident, &ident->sym_struct, 0, struct_type);
+               if (ident) {
+                       symbol(state, ident, &ident->sym_tag, 0, struct_type);
+               }
        }
-       if (ident && ident->sym_struct) {
-               struct_type = ident->sym_struct->type;
+       if (ident && ident->sym_tag &&
+               ident->sym_tag->type &&
+               ((ident->sym_tag->type->type & TYPE_MASK) == type_main)) {
+               struct_type = clone_type(spec, ident->sym_tag->type);
        }
-       else if (ident && !ident->sym_struct) {
-               error(state, 0, "struct %s undeclared", ident->name);
+       else if (ident && !struct_type) {
+               error(state, 0, "%s %s undeclared",
+                       (type_main == TYPE_STRUCT)?"struct" : "union",
+                       ident->name);
        }
        return struct_type;
 }
@@ -8136,7 +12704,7 @@ static unsigned int storage_class_specifier_opt(struct compile_state *state)
                break;
        default:
                if (state->scope_depth <= GLOBAL_SCOPE_DEPTH) {
-                       specifiers = STOR_STATIC;
+                       specifiers = STOR_LOCAL;
                }
                else {
                        specifiers = STOR_AUTO;
@@ -8158,29 +12726,94 @@ static unsigned int function_specifier_opt(struct compile_state *state)
        return specifiers;
 }
 
-static unsigned int type_qualifiers(struct compile_state *state)
+static unsigned int attrib(struct compile_state *state, unsigned int attributes)
 {
-       unsigned int specifiers;
-       int done;
-       done = 0;
-       specifiers = QUAL_NONE;
-       do {
-               switch(peek(state)) {
-               case TOK_CONST:
-                       eat(state, TOK_CONST);
-                       specifiers = QUAL_CONST;
-                       break;
-               case TOK_VOLATILE:
-                       eat(state, TOK_VOLATILE);
-                       specifiers = QUAL_VOLATILE;
-                       break;
-               case TOK_RESTRICT:
-                       eat(state, TOK_RESTRICT);
-                       specifiers = QUAL_RESTRICT;
-                       break;
-               default:
-                       done = 1;
-                       break;
+       int tok = peek(state);
+       switch(tok) {
+       case TOK_COMMA:
+       case TOK_LPAREN:
+               /* The empty attribute ignore it */
+               break;
+       case TOK_IDENT:
+       case TOK_ENUM_CONST:
+       case TOK_TYPE_NAME:
+       {
+               struct hash_entry *ident;
+               ident = eat(state, TOK_IDENT)->ident;
+
+               if (ident == state->i_noinline) {
+                       if (attributes & ATTRIB_ALWAYS_INLINE) {
+                               error(state, 0, "both always_inline and noinline attribtes");
+                       }
+                       attributes |= ATTRIB_NOINLINE;
+               }
+               else if (ident == state->i_always_inline) {
+                       if (attributes & ATTRIB_NOINLINE) {
+                               error(state, 0, "both noinline and always_inline attribtes");
+                       }
+                       attributes |= ATTRIB_ALWAYS_INLINE;
+               }
+               else if (ident == state->i_noreturn) {
+                       // attribute((noreturn)) does nothing (yet?)
+               }
+               else {
+                       error(state, 0, "Unknown attribute:%s", ident->name);
+               }
+               break;
+       }
+       default:
+               error(state, 0, "Unexpected token: %s\n", tokens[tok]);
+               break;
+       }
+       return attributes;
+}
+
+static unsigned int attribute_list(struct compile_state *state, unsigned type)
+{
+       type = attrib(state, type);
+       while(peek(state) == TOK_COMMA) {
+               eat(state, TOK_COMMA);
+               type = attrib(state, type);
+       }
+       return type;
+}
+
+static unsigned int attributes_opt(struct compile_state *state, unsigned type)
+{
+       if (peek(state) == TOK_ATTRIBUTE) {
+               eat(state, TOK_ATTRIBUTE);
+               eat(state, TOK_LPAREN);
+               eat(state, TOK_LPAREN);
+               type = attribute_list(state, type);
+               eat(state, TOK_RPAREN);
+               eat(state, TOK_RPAREN);
+       }
+       return type;
+}
+
+static unsigned int type_qualifiers(struct compile_state *state)
+{
+       unsigned int specifiers;
+       int done;
+       done = 0;
+       specifiers = QUAL_NONE;
+       do {
+               switch(peek(state)) {
+               case TOK_CONST:
+                       eat(state, TOK_CONST);
+                       specifiers |= QUAL_CONST;
+                       break;
+               case TOK_VOLATILE:
+                       eat(state, TOK_VOLATILE);
+                       specifiers |= QUAL_VOLATILE;
+                       break;
+               case TOK_RESTRICT:
+                       eat(state, TOK_RESTRICT);
+                       specifiers |= QUAL_RESTRICT;
+                       break;
+               default:
+                       done = 1;
+                       break;
                }
        } while(!done);
        return specifiers;
@@ -8190,8 +12823,9 @@ static struct type *type_specifier(
        struct compile_state *state, unsigned int spec)
 {
        struct type *type;
+       int tok;
        type = 0;
-       switch(peek(state)) {
+       switch((tok = peek(state))) {
        case TOK_VOID:
                eat(state, TOK_VOID);
                type = new_type(TYPE_VOID | spec, 0, 0);
@@ -8325,8 +12959,8 @@ static struct type *type_specifier(
                type = typedef_name(state, spec);
                break;
        default:
-               error(state, 0, "bad type specifier %s", 
-                       tokens[peek(state)]);
+               error(state, 0, "bad type specifier %s",
+                       tokens[tok]);
                break;
        }
        return type;
@@ -8372,6 +13006,7 @@ static struct type *specifier_qualifier_list(struct compile_state *state)
        return type;
 }
 
+#if DEBUG_ROMCC_WARNING
 static int isdecl_specifier(int tok)
 {
        switch(tok) {
@@ -8409,6 +13044,7 @@ static int isdecl_specifier(int tok)
                return 0;
        }
 }
+#endif
 
 static struct type *decl_specifiers(struct compile_state *state)
 {
@@ -8426,6 +13062,9 @@ static struct type *decl_specifiers(struct compile_state *state)
        /* function-specifier */
        specifiers |= function_specifier_opt(state);
 
+       /* attributes */
+       specifiers |= attributes_opt(state, 0);
+
        /* type qualifier */
        specifiers |= type_qualifiers(state);
 
@@ -8434,113 +13073,178 @@ static struct type *decl_specifiers(struct compile_state *state)
        return type;
 }
 
-static unsigned designator(struct compile_state *state)
+struct field_info {
+       struct type *type;
+       size_t offset;
+};
+
+static struct field_info designator(struct compile_state *state, struct type *type)
 {
        int tok;
-       unsigned index;
-       index = -1U;
+       struct field_info info;
+       info.offset = ~0U;
+       info.type = 0;
        do {
                switch(peek(state)) {
                case TOK_LBRACKET:
                {
                        struct triple *value;
+                       if ((type->type & TYPE_MASK) != TYPE_ARRAY) {
+                               error(state, 0, "Array designator not in array initializer");
+                       }
                        eat(state, TOK_LBRACKET);
                        value = constant_expr(state);
                        eat(state, TOK_RBRACKET);
-                       index = value->u.cval;
+
+                       info.type = type->left;
+                       info.offset = value->u.cval * size_of(state, info.type);
                        break;
                }
                case TOK_DOT:
+               {
+                       struct hash_entry *field;
+                       if (((type->type & TYPE_MASK) != TYPE_STRUCT) &&
+                               ((type->type & TYPE_MASK) != TYPE_UNION))
+                       {
+                               error(state, 0, "Struct designator not in struct initializer");
+                       }
                        eat(state, TOK_DOT);
-                       eat(state, TOK_IDENT);
-                       error(state, 0, "Struct Designators not currently supported");
+                       field = eat(state, TOK_IDENT)->ident;
+                       info.offset = field_offset(state, type, field);
+                       info.type   = field_type(state, type, field);
                        break;
+               }
                default:
                        error(state, 0, "Invalid designator");
                }
                tok = peek(state);
        } while((tok == TOK_LBRACKET) || (tok == TOK_DOT));
        eat(state, TOK_EQ);
-       return index;
+       return info;
 }
 
 static struct triple *initializer(
        struct compile_state *state, struct type *type)
 {
        struct triple *result;
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME more consistent initializer handling (where should eval_const_expr go?"
+#endif
        if (peek(state) != TOK_LBRACE) {
                result = assignment_expr(state);
+               if (((type->type & TYPE_MASK) == TYPE_ARRAY) &&
+                       (type->elements == ELEMENT_COUNT_UNSPECIFIED) &&
+                       ((result->type->type & TYPE_MASK) == TYPE_ARRAY) &&
+                       (result->type->elements != ELEMENT_COUNT_UNSPECIFIED) &&
+                       (equiv_types(type->left, result->type->left))) {
+                       type->elements = result->type->elements;
+               }
+               if (is_lvalue(state, result) &&
+                       ((result->type->type & TYPE_MASK) == TYPE_ARRAY) &&
+                       (type->type & TYPE_MASK) != TYPE_ARRAY)
+               {
+                       result = lvalue_conversion(state, result);
+               }
+               if (!is_init_compatible(state, type, result->type)) {
+                       error(state, 0, "Incompatible types in initializer");
+               }
+               if (!equiv_types(type, result->type)) {
+                       result = mk_cast_expr(state, type, result);
+               }
        }
        else {
                int comma;
-               unsigned index, max_index;
+               size_t max_offset;
+               struct field_info info;
                void *buf;
-               max_index = index = 0;
-               if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
-                       max_index = type->elements;
-                       if (type->elements == ELEMENT_COUNT_UNSPECIFIED) {
-                               type->elements = 0;
-                       }
+               if (((type->type & TYPE_MASK) != TYPE_ARRAY) &&
+                       ((type->type & TYPE_MASK) != TYPE_STRUCT)) {
+                       internal_error(state, 0, "unknown initializer type");
+               }
+               info.offset = 0;
+               info.type = type->left;
+               if ((type->type & TYPE_MASK) == TYPE_STRUCT) {
+                       info.type = next_field(state, type, 0);
+               }
+               if (type->elements == ELEMENT_COUNT_UNSPECIFIED) {
+                       max_offset = 0;
                } else {
-                       error(state, 0, "Struct initializers not currently supported");
+                       max_offset = size_of(state, type);
                }
-               buf = xcmalloc(size_of(state, type), "initializer");
+               buf = xcmalloc(bits_to_bytes(max_offset), "initializer");
                eat(state, TOK_LBRACE);
                do {
                        struct triple *value;
                        struct type *value_type;
                        size_t value_size;
+                       void *dest;
                        int tok;
                        comma = 0;
                        tok = peek(state);
                        if ((tok == TOK_LBRACKET) || (tok == TOK_DOT)) {
-                               index = designator(state);
+                               info = designator(state, type);
                        }
-                       if ((max_index != ELEMENT_COUNT_UNSPECIFIED) &&
-                               (index > max_index)) {
+                       if ((type->elements != ELEMENT_COUNT_UNSPECIFIED) &&
+                               (info.offset >= max_offset)) {
                                error(state, 0, "element beyond bounds");
                        }
-                       value_type = 0;
-                       if ((type->type & TYPE_MASK) == TYPE_ARRAY) {
-                               value_type = type->left;
-                       }
+                       value_type = info.type;
                        value = eval_const_expr(state, initializer(state, value_type));
                        value_size = size_of(state, value_type);
                        if (((type->type & TYPE_MASK) == TYPE_ARRAY) &&
-                               (max_index == ELEMENT_COUNT_UNSPECIFIED) &&
-                               (type->elements <= index)) {
+                               (type->elements == ELEMENT_COUNT_UNSPECIFIED) &&
+                               (max_offset <= info.offset)) {
                                void *old_buf;
                                size_t old_size;
                                old_buf = buf;
-                               old_size = size_of(state, type);
-                               type->elements = index + 1;
-                               buf = xmalloc(size_of(state, type), "initializer");
-                               memcpy(buf, old_buf, old_size);
+                               old_size = max_offset;
+                               max_offset = info.offset + value_size;
+                               buf = xmalloc(bits_to_bytes(max_offset), "initializer");
+                               memcpy(buf, old_buf, bits_to_bytes(old_size));
                                xfree(old_buf);
                        }
+                       dest = ((char *)buf) + bits_to_bytes(info.offset);
+#if DEBUG_INITIALIZER
+                       fprintf(state->errout, "dest = buf + %d max_offset: %d value_size: %d op: %d\n",
+                               dest - buf,
+                               bits_to_bytes(max_offset),
+                               bits_to_bytes(value_size),
+                               value->op);
+#endif
                        if (value->op == OP_BLOBCONST) {
-                               memcpy((char *)buf + index * value_size, value->u.blob, value_size);
+                               memcpy(dest, value->u.blob, bits_to_bytes(value_size));
                        }
-                       else if ((value->op == OP_INTCONST) && (value_size == 1)) {
-                               *(((uint8_t *)buf) + index) = value->u.cval & 0xff;
+                       else if ((value->op == OP_INTCONST) && (value_size == SIZEOF_I8)) {
+#if DEBUG_INITIALIZER
+                               fprintf(state->errout, "byte: %02x\n", value->u.cval & 0xff);
+#endif
+                               *((uint8_t *)dest) = value->u.cval & 0xff;
                        }
-                       else if ((value->op == OP_INTCONST) && (value_size == 2)) {
-                               *(((uint16_t *)buf) + index) = value->u.cval & 0xffff;
+                       else if ((value->op == OP_INTCONST) && (value_size == SIZEOF_I16)) {
+                               *((uint16_t *)dest) = value->u.cval & 0xffff;
                        }
-                       else if ((value->op == OP_INTCONST) && (value_size == 4)) {
-                               *(((uint32_t *)buf) + index) = value->u.cval & 0xffffffff;
+                       else if ((value->op == OP_INTCONST) && (value_size == SIZEOF_I32)) {
+                               *((uint32_t *)dest) = value->u.cval & 0xffffffff;
                        }
                        else {
-                               fprintf(stderr, "%d %d\n",
-                                       value->op, value_size);
                                internal_error(state, 0, "unhandled constant initializer");
                        }
+                       free_triple(state, value);
                        if (peek(state) == TOK_COMMA) {
                                eat(state, TOK_COMMA);
                                comma = 1;
                        }
-                       index += 1;
+                       info.offset += value_size;
+                       if ((type->type & TYPE_MASK) == TYPE_STRUCT) {
+                               info.type = next_field(state, type, info.type);
+                               info.offset = field_offset(state, type,
+                                       info.type->field_ident);
+                       }
                } while(comma && (peek(state) != TOK_RBRACE));
+               if ((type->elements == ELEMENT_COUNT_UNSPECIFIED) &&
+                       ((type->type & TYPE_MASK) == TYPE_ARRAY)) {
+                       type->elements = max_offset / size_of(state, type->left);
+               }
                eat(state, TOK_RBRACE);
                result = triple(state, OP_BLOBCONST, type, 0, 0);
                result->u.blob = buf;
@@ -8548,12 +13252,62 @@ static struct triple *initializer(
        return result;
 }
 
+static void resolve_branches(struct compile_state *state, struct triple *first)
+{
+       /* Make a second pass and finish anything outstanding
+        * with respect to branches.  The only outstanding item
+        * is to see if there are goto to labels that have not
+        * been defined and to error about them.
+        */
+       int i;
+       struct triple *ins;
+       /* Also error on branches that do not use their targets */
+       ins = first;
+       do {
+               if (!triple_is_ret(state, ins)) {
+                       struct triple **expr ;
+                       struct triple_set *set;
+                       expr = triple_targ(state, ins, 0);
+                       for(; expr; expr = triple_targ(state, ins, expr)) {
+                               struct triple *targ;
+                               targ = *expr;
+                               for(set = targ?targ->use:0; set; set = set->next) {
+                                       if (set->member == ins) {
+                                               break;
+                                       }
+                               }
+                               if (!set) {
+                                       internal_error(state, ins, "targ not used");
+                               }
+                       }
+               }
+               ins = ins->next;
+       } while(ins != first);
+       /* See if there are goto to labels that have not been defined */
+       for(i = 0; i < HASH_TABLE_SIZE; i++) {
+               struct hash_entry *entry;
+               for(entry = state->hash_table[i]; entry; entry = entry->next) {
+                       struct triple *ins;
+                       if (!entry->sym_label) {
+                               continue;
+                       }
+                       ins = entry->sym_label->def;
+                       if (!(ins->id & TRIPLE_FLAG_FLATTENED)) {
+                               error(state, ins, "label `%s' used but not defined",
+                                       entry->name);
+                       }
+               }
+       }
+}
+
 static struct triple *function_definition(
        struct compile_state *state, struct type *type)
 {
-       struct triple *def, *tmp, *first, *end;
+       struct triple *def, *tmp, *first, *end, *retvar, *ret;
+       struct triple *fname;
+       struct type *fname_type;
        struct hash_entry *ident;
-       struct type *param;
+       struct type *param, *crtype, *ctype;
        int i;
        if ((type->type &TYPE_MASK) != TYPE_FUNCTION) {
                error(state, 0, "Invalid function header");
@@ -8578,7 +13332,7 @@ static struct triple *function_definition(
        if (((param->type & TYPE_MASK) != TYPE_VOID) && !param->field_ident) {
                error(state, 0, "No identifier for paramter %d\n", i);
        }
-       
+
        /* Get a list of statements for this function. */
        def = triple(state, OP_LIST, type, 0, 0);
 
@@ -8592,6 +13346,27 @@ static struct triple *function_definition(
        /* Put a label at the very end of a function */
        end = label(state);
        flatten(state, first, end);
+       /* Remember where return goes */
+       ident = state->i_return;
+       symbol(state, ident, &ident->sym_ident, end, end->type);
+
+       /* Get the initial closure type */
+       ctype = new_type(TYPE_JOIN, &void_type, 0);
+       ctype->elements = 1;
+
+       /* Add a variable for the return value */
+       crtype = new_type(TYPE_TUPLE,
+               /* Remove all type qualifiers from the return type */
+               new_type(TYPE_PRODUCT, ctype, clone_type(0, type->left)), 0);
+       crtype->elements = 2;
+       flatten(state, end, variable(state, crtype));
+
+       /* Allocate a variable for the return address */
+       retvar = flatten(state, end, variable(state, &void_ptr_type));
+
+       /* Add in the return instruction */
+       ret = triple(state, OP_RET, &void_type, read_expr(state, retvar), 0);
+       ret = flatten(state, first, ret);
 
        /* Walk through the parameters and create symbol table entries
         * for them.
@@ -8600,7 +13375,7 @@ static struct triple *function_definition(
        while((param->type & TYPE_MASK) == TYPE_PRODUCT) {
                ident = param->left->field_ident;
                tmp = variable(state, param->left);
-               symbol(state, ident, &ident->sym_ident, tmp, tmp->type);
+               var_symbol(state, ident, tmp);
                flatten(state, end, tmp);
                param = param->right;
        }
@@ -8611,159 +13386,1512 @@ static struct triple *function_definition(
                symbol(state, ident, &ident->sym_ident, tmp, tmp->type);
                flatten(state, end, tmp);
        }
-       /* Add a variable for the return value */
-       MISC(def, 0) = 0;
-       if ((type->left->type & TYPE_MASK) != TYPE_VOID) {
-               /* Remove all type qualifiers from the return type */
-               tmp = variable(state, clone_type(0, type->left));
-               flatten(state, end, tmp);
-               /* Remember where the return value is */
-               MISC(def, 0) = tmp;
-       }
+
+       /* Add the declaration static const char __func__ [] = "func-name"  */
+       fname_type = new_type(TYPE_ARRAY,
+               clone_type(QUAL_CONST | STOR_STATIC, &char_type), 0);
+       fname_type->type |= QUAL_CONST | STOR_STATIC;
+       fname_type->elements = strlen(state->function) + 1;
+
+       fname = triple(state, OP_BLOBCONST, fname_type, 0, 0);
+       fname->u.blob = (void *)state->function;
+       fname = flatten(state, end, fname);
+
+       ident = state->i___func__;
+       symbol(state, ident, &ident->sym_ident, fname, fname_type);
 
        /* Remember which function I am compiling.
         * Also assume the last defined function is the main function.
         */
        state->main_function = def;
 
-       /* Now get the actual function definition */
-       compound_statement(state, end);
+       /* Now get the actual function definition */
+       compound_statement(state, end);
+
+       /* Finish anything unfinished with branches */
+       resolve_branches(state, first);
+
+       /* Remove the parameter scope */
+       end_scope(state);
+
+
+       /* Remember I have defined a function */
+       if (!state->functions) {
+               state->functions = def;
+       } else {
+               insert_triple(state, state->functions, def);
+       }
+       if (state->compiler->debug & DEBUG_INLINE) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "\n");
+               loc(fp, state, 0);
+               fprintf(fp, "\n__________ %s _________\n", __FUNCTION__);
+               display_func(state, fp, def);
+               fprintf(fp, "__________ %s _________ done\n\n", __FUNCTION__);
+       }
+
+       return def;
+}
+
+static struct triple *do_decl(struct compile_state *state,
+       struct type *type, struct hash_entry *ident)
+{
+       struct triple *def;
+       def = 0;
+       /* Clean up the storage types used */
+       switch (type->type & STOR_MASK) {
+       case STOR_AUTO:
+       case STOR_STATIC:
+               /* These are the good types I am aiming for */
+               break;
+       case STOR_REGISTER:
+               type->type &= ~STOR_MASK;
+               type->type |= STOR_AUTO;
+               break;
+       case STOR_LOCAL:
+       case STOR_EXTERN:
+               type->type &= ~STOR_MASK;
+               type->type |= STOR_STATIC;
+               break;
+       case STOR_TYPEDEF:
+               if (!ident) {
+                       error(state, 0, "typedef without name");
+               }
+               symbol(state, ident, &ident->sym_ident, 0, type);
+               ident->tok = TOK_TYPE_NAME;
+               return 0;
+               break;
+       default:
+               internal_error(state, 0, "Undefined storage class");
+       }
+       if ((type->type & TYPE_MASK) == TYPE_FUNCTION) {
+               error(state, 0, "Function prototypes not supported");
+       }
+       if (ident &&
+               ((type->type & TYPE_MASK) == TYPE_ARRAY) &&
+               ((type->type & STOR_MASK) != STOR_STATIC))
+               error(state, 0, "non static arrays not supported");
+       if (ident &&
+               ((type->type & STOR_MASK) == STOR_STATIC) &&
+               ((type->type & QUAL_CONST) == 0)) {
+               error(state, 0, "non const static variables not supported");
+       }
+       if (ident) {
+               def = variable(state, type);
+               var_symbol(state, ident, def);
+       }
+       return def;
+}
+
+static void decl(struct compile_state *state, struct triple *first)
+{
+       struct type *base_type, *type;
+       struct hash_entry *ident;
+       struct triple *def;
+       int global;
+       global = (state->scope_depth <= GLOBAL_SCOPE_DEPTH);
+       base_type = decl_specifiers(state);
+       ident = 0;
+       type = declarator(state, base_type, &ident, 0);
+       type->type = attributes_opt(state, type->type);
+       if (global && ident && (peek(state) == TOK_LBRACE)) {
+               /* function */
+               type->type_ident = ident;
+               state->function = ident->name;
+               def = function_definition(state, type);
+               symbol(state, ident, &ident->sym_ident, def, type);
+               state->function = 0;
+       }
+       else {
+               int done;
+               flatten(state, first, do_decl(state, type, ident));
+               /* type or variable definition */
+               do {
+                       done = 1;
+                       if (peek(state) == TOK_EQ) {
+                               if (!ident) {
+                                       error(state, 0, "cannot assign to a type");
+                               }
+                               eat(state, TOK_EQ);
+                               flatten(state, first,
+                                       init_expr(state,
+                                               ident->sym_ident->def,
+                                               initializer(state, type)));
+                       }
+                       arrays_complete(state, type);
+                       if (peek(state) == TOK_COMMA) {
+                               eat(state, TOK_COMMA);
+                               ident = 0;
+                               type = declarator(state, base_type, &ident, 0);
+                               flatten(state, first, do_decl(state, type, ident));
+                               done = 0;
+                       }
+               } while(!done);
+               eat(state, TOK_SEMI);
+       }
+}
+
+static void decls(struct compile_state *state)
+{
+       struct triple *list;
+       int tok;
+       list = label(state);
+       while(1) {
+               tok = peek(state);
+               if (tok == TOK_EOF) {
+                       return;
+               }
+               if (tok == TOK_SPACE) {
+                       eat(state, TOK_SPACE);
+               }
+               decl(state, list);
+               if (list->next != list) {
+                       error(state, 0, "global variables not supported");
+               }
+       }
+}
+
+/*
+ * Function inlining
+ */
+struct triple_reg_set {
+       struct triple_reg_set *next;
+       struct triple *member;
+       struct triple *new;
+};
+struct reg_block {
+       struct block *block;
+       struct triple_reg_set *in;
+       struct triple_reg_set *out;
+       int vertex;
+};
+static void setup_basic_blocks(struct compile_state *, struct basic_blocks *bb);
+static void analyze_basic_blocks(struct compile_state *state, struct basic_blocks *bb);
+static void free_basic_blocks(struct compile_state *, struct basic_blocks *bb);
+static int tdominates(struct compile_state *state, struct triple *dom, struct triple *sub);
+static void walk_blocks(struct compile_state *state, struct basic_blocks *bb,
+       void (*cb)(struct compile_state *state, struct block *block, void *arg),
+       void *arg);
+static void print_block(
+       struct compile_state *state, struct block *block, void *arg);
+static int do_triple_set(struct triple_reg_set **head,
+       struct triple *member, struct triple *new_member);
+static void do_triple_unset(struct triple_reg_set **head, struct triple *member);
+static struct reg_block *compute_variable_lifetimes(
+       struct compile_state *state, struct basic_blocks *bb);
+static void free_variable_lifetimes(struct compile_state *state,
+       struct basic_blocks *bb, struct reg_block *blocks);
+#if DEBUG_EXPLICIT_CLOSURES
+static void print_live_variables(struct compile_state *state,
+       struct basic_blocks *bb, struct reg_block *rb, FILE *fp);
+#endif
+
+
+static struct triple *call(struct compile_state *state,
+       struct triple *retvar, struct triple *ret_addr,
+       struct triple *targ, struct triple *ret)
+{
+       struct triple *call;
+
+       if (!retvar || !is_lvalue(state, retvar)) {
+               internal_error(state, 0, "writing to a non lvalue?");
+       }
+       write_compatible(state, retvar->type, &void_ptr_type);
+
+       call = new_triple(state, OP_CALL, &void_type, 1, 0);
+       TARG(call, 0) = targ;
+       MISC(call, 0) = ret;
+       if (!targ || (targ->op != OP_LABEL)) {
+               internal_error(state, 0, "call not to a label");
+       }
+       if (!ret || (ret->op != OP_RET)) {
+               internal_error(state, 0, "call not matched with return");
+       }
+       return call;
+}
+
+static void walk_functions(struct compile_state *state,
+       void (*cb)(struct compile_state *state, struct triple *func, void *arg),
+       void *arg)
+{
+       struct triple *func, *first;
+       func = first = state->functions;
+       do {
+               cb(state, func, arg);
+               func = func->next;
+       } while(func != first);
+}
+
+static void reverse_walk_functions(struct compile_state *state,
+       void (*cb)(struct compile_state *state, struct triple *func, void *arg),
+       void *arg)
+{
+       struct triple *func, *first;
+       func = first = state->functions;
+       do {
+               func = func->prev;
+               cb(state, func, arg);
+       } while(func != first);
+}
+
+
+static void mark_live(struct compile_state *state, struct triple *func, void *arg)
+{
+       struct triple *ptr, *first;
+       if (func->u.cval == 0) {
+               return;
+       }
+       ptr = first = RHS(func, 0);
+       do {
+               if (ptr->op == OP_FCALL) {
+                       struct triple *called_func;
+                       called_func = MISC(ptr, 0);
+                       /* Mark the called function as used */
+                       if (!(func->id & TRIPLE_FLAG_FLATTENED)) {
+                               called_func->u.cval++;
+                       }
+                       /* Remove the called function from the list */
+                       called_func->prev->next = called_func->next;
+                       called_func->next->prev = called_func->prev;
+
+                       /* Place the called function before me on the list */
+                       called_func->next       = func;
+                       called_func->prev       = func->prev;
+                       called_func->prev->next = called_func;
+                       called_func->next->prev = called_func;
+               }
+               ptr = ptr->next;
+       } while(ptr != first);
+       func->id |= TRIPLE_FLAG_FLATTENED;
+}
+
+static void mark_live_functions(struct compile_state *state)
+{
+       /* Ensure state->main_function is the last function in
+        * the list of functions.
+        */
+       if ((state->main_function->next != state->functions) ||
+               (state->functions->prev != state->main_function)) {
+               internal_error(state, 0,
+                       "state->main_function is not at the end of the function list ");
+       }
+       state->main_function->u.cval = 1;
+       reverse_walk_functions(state, mark_live, 0);
+}
+
+static int local_triple(struct compile_state *state,
+       struct triple *func, struct triple *ins)
+{
+       int local = (ins->id & TRIPLE_FLAG_LOCAL);
+#if 0
+       if (!local) {
+               FILE *fp = state->errout;
+               fprintf(fp, "global: ");
+               display_triple(fp, ins);
+       }
+#endif
+       return local;
+}
+
+struct triple *copy_func(struct compile_state *state, struct triple *ofunc,
+       struct occurance *base_occurance)
+{
+       struct triple *nfunc;
+       struct triple *nfirst, *ofirst;
+       struct triple *new, *old;
+
+       if (state->compiler->debug & DEBUG_INLINE) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "\n");
+               loc(fp, state, 0);
+               fprintf(fp, "\n__________ %s _________\n", __FUNCTION__);
+               display_func(state, fp, ofunc);
+               fprintf(fp, "__________ %s _________ done\n\n", __FUNCTION__);
+       }
+
+       /* Make a new copy of the old function */
+       nfunc = triple(state, OP_LIST, ofunc->type, 0, 0);
+       nfirst = 0;
+       ofirst = old = RHS(ofunc, 0);
+       do {
+               struct triple *new;
+               struct occurance *occurance;
+               int old_lhs, old_rhs;
+               old_lhs = old->lhs;
+               old_rhs = old->rhs;
+               occurance = inline_occurance(state, base_occurance, old->occurance);
+               if (ofunc->u.cval && (old->op == OP_FCALL)) {
+                       MISC(old, 0)->u.cval += 1;
+               }
+               new = alloc_triple(state, old->op, old->type, old_lhs, old_rhs,
+                       occurance);
+               if (!triple_stores_block(state, new)) {
+                       memcpy(&new->u, &old->u, sizeof(new->u));
+               }
+               if (!nfirst) {
+                       RHS(nfunc, 0) = nfirst = new;
+               }
+               else {
+                       insert_triple(state, nfirst, new);
+               }
+               new->id |= TRIPLE_FLAG_FLATTENED;
+               new->id |= old->id & TRIPLE_FLAG_COPY;
+
+               /* During the copy remember new as user of old */
+               use_triple(old, new);
+
+               /* Remember which instructions are local */
+               old->id |= TRIPLE_FLAG_LOCAL;
+               old = old->next;
+       } while(old != ofirst);
+
+       /* Make a second pass to fix up any unresolved references */
+       old = ofirst;
+       new = nfirst;
+       do {
+               struct triple **oexpr, **nexpr;
+               int count, i;
+               /* Lookup where the copy is, to join pointers */
+               count = TRIPLE_SIZE(old);
+               for(i = 0; i < count; i++) {
+                       oexpr = &old->param[i];
+                       nexpr = &new->param[i];
+                       if (*oexpr && !*nexpr) {
+                               if (!local_triple(state, ofunc, *oexpr)) {
+                                       *nexpr = *oexpr;
+                               }
+                               else if ((*oexpr)->use) {
+                                       *nexpr = (*oexpr)->use->member;
+                               }
+                               if (*nexpr == old) {
+                                       internal_error(state, 0, "new == old?");
+                               }
+                               use_triple(*nexpr, new);
+                       }
+                       if (!*nexpr && *oexpr) {
+                               internal_error(state, 0, "Could not copy %d", i);
+                       }
+               }
+               old = old->next;
+               new = new->next;
+       } while((old != ofirst) && (new != nfirst));
+
+       /* Make a third pass to cleanup the extra useses */
+       old = ofirst;
+       new = nfirst;
+       do {
+               unuse_triple(old, new);
+               /* Forget which instructions are local */
+               old->id &= ~TRIPLE_FLAG_LOCAL;
+               old = old->next;
+               new = new->next;
+       } while ((old != ofirst) && (new != nfirst));
+       return nfunc;
+}
+
+static void expand_inline_call(
+       struct compile_state *state, struct triple *me, struct triple *fcall)
+{
+       /* Inline the function call */
+       struct type *ptype;
+       struct triple *ofunc, *nfunc, *nfirst, *result, *retvar, *ins;
+       struct triple *end, *nend;
+       int pvals, i;
+
+       /* Find the triples */
+       ofunc = MISC(fcall, 0);
+       if (ofunc->op != OP_LIST) {
+               internal_error(state, 0, "improper function");
+       }
+       nfunc = copy_func(state, ofunc, fcall->occurance);
+       /* Prepend the parameter reading into the new function list */
+       ptype = nfunc->type->right;
+       pvals = fcall->rhs;
+       for(i = 0; i < pvals; i++) {
+               struct type *atype;
+               struct triple *arg, *param;
+               atype = ptype;
+               if ((ptype->type & TYPE_MASK) == TYPE_PRODUCT) {
+                       atype = ptype->left;
+               }
+               param = farg(state, nfunc, i);
+               if ((param->type->type & TYPE_MASK) != (atype->type & TYPE_MASK)) {
+                       internal_error(state, fcall, "param %d type mismatch", i);
+               }
+               arg = RHS(fcall, i);
+               flatten(state, fcall, write_expr(state, param, arg));
+               ptype = ptype->right;
+       }
+       result = 0;
+       if ((nfunc->type->left->type & TYPE_MASK) != TYPE_VOID) {
+               result = read_expr(state,
+                       deref_index(state, fresult(state, nfunc), 1));
+       }
+       if (state->compiler->debug & DEBUG_INLINE) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "\n");
+               loc(fp, state, 0);
+               fprintf(fp, "\n__________ %s _________\n", __FUNCTION__);
+               display_func(state, fp, nfunc);
+               fprintf(fp, "__________ %s _________ done\n\n", __FUNCTION__);
+       }
+
+       /*
+        * Get rid of the extra triples
+        */
+       /* Remove the read of the return address */
+       ins = RHS(nfunc, 0)->prev->prev;
+       if ((ins->op != OP_READ) || (RHS(ins, 0) != fretaddr(state, nfunc))) {
+               internal_error(state, ins, "Not return addres read?");
+       }
+       release_triple(state, ins);
+       /* Remove the return instruction */
+       ins = RHS(nfunc, 0)->prev;
+       if (ins->op != OP_RET) {
+               internal_error(state, ins, "Not return?");
+       }
+       release_triple(state, ins);
+       /* Remove the retaddres variable */
+       retvar = fretaddr(state, nfunc);
+       if ((retvar->lhs != 1) ||
+               (retvar->op != OP_ADECL) ||
+               (retvar->next->op != OP_PIECE) ||
+               (MISC(retvar->next, 0) != retvar)) {
+               internal_error(state, retvar, "Not the return address?");
+       }
+       release_triple(state, retvar->next);
+       release_triple(state, retvar);
+
+       /* Remove the label at the start of the function */
+       ins = RHS(nfunc, 0);
+       if (ins->op != OP_LABEL) {
+               internal_error(state, ins, "Not label?");
+       }
+       nfirst = ins->next;
+       free_triple(state, ins);
+       /* Release the new function header */
+       RHS(nfunc, 0) = 0;
+       free_triple(state, nfunc);
+
+       /* Append the new function list onto the return list */
+       end = fcall->prev;
+       nend = nfirst->prev;
+       end->next    = nfirst;
+       nfirst->prev = end;
+       nend->next   = fcall;
+       fcall->prev  = nend;
+
+       /* Now the result reading code */
+       if (result) {
+               result = flatten(state, fcall, result);
+               propogate_use(state, fcall, result);
+       }
+
+       /* Release the original fcall instruction */
+       release_triple(state, fcall);
+
+       return;
+}
+
+/*
+ *
+ * Type of the result variable.
+ *
+ *                                     result
+ *                                        |
+ *                             +----------+------------+
+ *                             |                       |
+ *                     union of closures         result_type
+ *                             |
+ *          +------------------+---------------+
+ *          |                                  |
+ *       closure1                    ...   closuerN
+ *          |                                  |
+ *  +----+--+-+--------+-----+       +----+----+---+-----+
+ *  |    |    |        |     |       |    |        |     |
+ * var1 var2 var3 ... varN result   var1 var2 ... varN result
+ *                           |
+ *                  +--------+---------+
+ *                  |                  |
+ *          union of closures     result_type
+ *                  |
+ *            +-----+-------------------+
+ *            |                         |
+ *         closure1            ...  closureN
+ *            |                         |
+ *  +-----+---+----+----+      +----+---+----+-----+
+ *  |     |        |    |      |    |        |     |
+ * var1 var2 ... varN result  var1 var2 ... varN result
+ */
+
+static int add_closure_type(struct compile_state *state,
+       struct triple *func, struct type *closure_type)
+{
+       struct type *type, *ctype, **next;
+       struct triple *var, *new_var;
+       int i;
+
+#if 0
+       FILE *fp = state->errout;
+       fprintf(fp, "original_type: ");
+       name_of(fp, fresult(state, func)->type);
+       fprintf(fp, "\n");
+#endif
+       /* find the original type */
+       var = fresult(state, func);
+       type = var->type;
+       if (type->elements != 2) {
+               internal_error(state, var, "bad return type");
+       }
+
+       /* Find the complete closure type and update it */
+       ctype = type->left->left;
+       next = &ctype->left;
+       while(((*next)->type & TYPE_MASK) == TYPE_OVERLAP) {
+               next = &(*next)->right;
+       }
+       *next = new_type(TYPE_OVERLAP, *next, dup_type(state, closure_type));
+       ctype->elements += 1;
+
+#if 0
+       fprintf(fp, "new_type: ");
+       name_of(fp, type);
+       fprintf(fp, "\n");
+       fprintf(fp, "ctype: %p %d bits: %d ",
+               ctype, ctype->elements, reg_size_of(state, ctype));
+       name_of(fp, ctype);
+       fprintf(fp, "\n");
+#endif
+
+       /* Regenerate the variable with the new type definition */
+       new_var = pre_triple(state, var, OP_ADECL, type, 0, 0);
+       new_var->id |= TRIPLE_FLAG_FLATTENED;
+       for(i = 0; i < new_var->lhs; i++) {
+               LHS(new_var, i)->id |= TRIPLE_FLAG_FLATTENED;
+       }
+
+       /* Point everyone at the new variable */
+       propogate_use(state, var, new_var);
+
+       /* Release the original variable */
+       for(i = 0; i < var->lhs; i++) {
+               release_triple(state, LHS(var, i));
+       }
+       release_triple(state, var);
+
+       /* Return the index of the added closure type */
+       return ctype->elements - 1;
+}
+
+static struct triple *closure_expr(struct compile_state *state,
+       struct triple *func, int closure_idx, int var_idx)
+{
+       return deref_index(state,
+               deref_index(state,
+                       deref_index(state, fresult(state, func), 0),
+                       closure_idx),
+               var_idx);
+}
+
+
+static void insert_triple_set(
+       struct triple_reg_set **head, struct triple *member)
+{
+       struct triple_reg_set *new;
+       new = xcmalloc(sizeof(*new), "triple_set");
+       new->member = member;
+       new->new    = 0;
+       new->next   = *head;
+       *head       = new;
+}
+
+static int ordered_triple_set(
+       struct triple_reg_set **head, struct triple *member)
+{
+       struct triple_reg_set **ptr;
+       if (!member)
+               return 0;
+       ptr = head;
+       while(*ptr) {
+               if (member == (*ptr)->member) {
+                       return 0;
+               }
+               /* keep the list ordered */
+               if (member->id < (*ptr)->member->id) {
+                       break;
+               }
+               ptr = &(*ptr)->next;
+       }
+       insert_triple_set(ptr, member);
+       return 1;
+}
+
+
+static void free_closure_variables(struct compile_state *state,
+       struct triple_reg_set **enclose)
+{
+       struct triple_reg_set *entry, *next;
+       for(entry = *enclose; entry; entry = next) {
+               next = entry->next;
+               do_triple_unset(enclose, entry->member);
+       }
+}
+
+static int lookup_closure_index(struct compile_state *state,
+       struct triple *me, struct triple *val)
+{
+       struct triple *first, *ins, *next;
+       first = RHS(me, 0);
+       ins = next = first;
+       do {
+               struct triple *result;
+               struct triple *index0, *index1, *index2, *read, *write;
+               ins = next;
+               next = ins->next;
+               if (ins->op != OP_CALL) {
+                       continue;
+               }
+               /* I am at a previous call point examine it closely */
+               if (ins->next->op != OP_LABEL) {
+                       internal_error(state, ins, "call not followed by label");
+               }
+               /* Does this call does not enclose any variables? */
+               if ((ins->next->next->op != OP_INDEX) ||
+                       (ins->next->next->u.cval != 0) ||
+                       (result = MISC(ins->next->next, 0)) ||
+                       (result->id & TRIPLE_FLAG_LOCAL)) {
+                       continue;
+               }
+               index0 = ins->next->next;
+               /* The pattern is:
+                * 0 index result < 0 >
+                * 1 index 0 < ? >
+                * 2 index 1 < ? >
+                * 3 read  2
+                * 4 write 3 var
+                */
+               for(index0 = ins->next->next;
+                       (index0->op == OP_INDEX) &&
+                               (MISC(index0, 0) == result) &&
+                               (index0->u.cval == 0) ;
+                       index0 = write->next)
+               {
+                       index1 = index0->next;
+                       index2 = index1->next;
+                       read   = index2->next;
+                       write  = read->next;
+                       if ((index0->op != OP_INDEX) ||
+                               (index1->op != OP_INDEX) ||
+                               (index2->op != OP_INDEX) ||
+                               (read->op != OP_READ) ||
+                               (write->op != OP_WRITE) ||
+                               (MISC(index1, 0) != index0) ||
+                               (MISC(index2, 0) != index1) ||
+                               (RHS(read, 0) != index2) ||
+                               (RHS(write, 0) != read)) {
+                               internal_error(state, index0, "bad var read");
+                       }
+                       if (MISC(write, 0) == val) {
+                               return index2->u.cval;
+                       }
+               }
+       } while(next != first);
+       return -1;
+}
+
+static inline int enclose_triple(struct triple *ins)
+{
+       return (ins && ((ins->type->type & TYPE_MASK) != TYPE_VOID));
+}
+
+static void compute_closure_variables(struct compile_state *state,
+       struct triple *me, struct triple *fcall, struct triple_reg_set **enclose)
+{
+       struct triple_reg_set *set, *vars, **last_var;
+       struct basic_blocks bb;
+       struct reg_block *rb;
+       struct block *block;
+       struct triple *old_result, *first, *ins;
+       size_t count, idx;
+       unsigned long used_indicies;
+       int i, max_index;
+#define MAX_INDICIES (sizeof(used_indicies)*CHAR_BIT)
+#define ID_BITS(X) ((X) & (TRIPLE_FLAG_LOCAL -1))
+       struct {
+               unsigned id;
+               int index;
+       } *info;
+
+
+       /* Find the basic blocks of this function */
+       bb.func = me;
+       bb.first = RHS(me, 0);
+       old_result = 0;
+       if (!triple_is_ret(state, bb.first->prev)) {
+               bb.func = 0;
+       } else {
+               old_result = fresult(state, me);
+       }
+       analyze_basic_blocks(state, &bb);
+
+       /* Find which variables are currently alive in a given block */
+       rb = compute_variable_lifetimes(state, &bb);
+
+       /* Find the variables that are currently alive */
+       block = block_of_triple(state, fcall);
+       if (!block || (block->vertex <= 0) || (block->vertex > bb.last_vertex)) {
+               internal_error(state, fcall, "No reg block? block: %p", block);
+       }
+
+#if DEBUG_EXPLICIT_CLOSURES
+       print_live_variables(state, &bb, rb, state->dbgout);
+       fflush(state->dbgout);
+#endif
+
+       /* Count the number of triples in the function */
+       first = RHS(me, 0);
+       ins = first;
+       count = 0;
+       do {
+               count++;
+               ins = ins->next;
+       } while(ins != first);
+
+       /* Allocate some memory to temorary hold the id info */
+       info = xcmalloc(sizeof(*info) * (count +1), "info");
+
+       /* Mark the local function */
+       first = RHS(me, 0);
+       ins = first;
+       idx = 1;
+       do {
+               info[idx].id = ins->id;
+               ins->id = TRIPLE_FLAG_LOCAL | idx;
+               idx++;
+               ins = ins->next;
+       } while(ins != first);
+
+       /*
+        * Build the list of variables to enclose.
+        *
+        * A target it to put the same variable in the
+        * same slot for ever call of a given function.
+        * After coloring this removes all of the variable
+        * manipulation code.
+        *
+        * The list of variables to enclose is built ordered
+        * program order because except in corner cases this
+        * gives me the stability of assignment I need.
+        *
+        * To gurantee that stability I lookup the variables
+        * to see where they have been used before and
+        * I build my final list with the assigned indicies.
+        */
+       vars = 0;
+       if (enclose_triple(old_result)) {
+               ordered_triple_set(&vars, old_result);
+       }
+       for(set = rb[block->vertex].out; set; set = set->next) {
+               if (!enclose_triple(set->member)) {
+                       continue;
+               }
+               if ((set->member == fcall) || (set->member == old_result)) {
+                       continue;
+               }
+               if (!local_triple(state, me, set->member)) {
+                       internal_error(state, set->member, "not local?");
+               }
+               ordered_triple_set(&vars, set->member);
+       }
+
+       /* Lookup the current indicies of the live varialbe */
+       used_indicies = 0;
+       max_index = -1;
+       for(set = vars; set ; set = set->next) {
+               struct triple *ins;
+               int index;
+               ins = set->member;
+               index  = lookup_closure_index(state, me, ins);
+               info[ID_BITS(ins->id)].index = index;
+               if (index < 0) {
+                       continue;
+               }
+               if (index >= MAX_INDICIES) {
+                       internal_error(state, ins, "index unexpectedly large");
+               }
+               if (used_indicies & (1 << index)) {
+                       internal_error(state, ins, "index previously used?");
+               }
+               /* Remember which indicies have been used */
+               used_indicies |= (1 << index);
+               if (index > max_index) {
+                       max_index = index;
+               }
+       }
+
+       /* Walk through the live variables and make certain
+        * everything is assigned an index.
+        */
+       for(set = vars; set; set = set->next) {
+               struct triple *ins;
+               int index;
+               ins = set->member;
+               index = info[ID_BITS(ins->id)].index;
+               if (index >= 0) {
+                       continue;
+               }
+               /* Find the lowest unused index value */
+               for(index = 0; index < MAX_INDICIES; index++) {
+                       if (!(used_indicies & (1 << index))) {
+                               break;
+                       }
+               }
+               if (index == MAX_INDICIES) {
+                       internal_error(state, ins, "no free indicies?");
+               }
+               info[ID_BITS(ins->id)].index = index;
+               /* Remember which indicies have been used */
+               used_indicies |= (1 << index);
+               if (index > max_index) {
+                       max_index = index;
+               }
+       }
+
+       /* Build the return list of variables with positions matching
+        * their indicies.
+        */
+       *enclose = 0;
+       last_var = enclose;
+       for(i = 0; i <= max_index; i++) {
+               struct triple *var;
+               var = 0;
+               if (used_indicies & (1 << i)) {
+                       for(set = vars; set; set = set->next) {
+                               int index;
+                               index = info[ID_BITS(set->member->id)].index;
+                               if (index == i) {
+                                       var = set->member;
+                                       break;
+                               }
+                       }
+                       if (!var) {
+                               internal_error(state, me, "missing variable");
+                       }
+               }
+               insert_triple_set(last_var, var);
+               last_var = &(*last_var)->next;
+       }
+
+#if DEBUG_EXPLICIT_CLOSURES
+       /* Print out the variables to be enclosed */
+       loc(state->dbgout, state, fcall);
+       fprintf(state->dbgout, "Alive: \n");
+       for(set = *enclose; set; set = set->next) {
+               display_triple(state->dbgout, set->member);
+       }
+       fflush(state->dbgout);
+#endif
+
+       /* Clear the marks */
+       ins = first;
+       do {
+               ins->id = info[ID_BITS(ins->id)].id;
+               ins = ins->next;
+       } while(ins != first);
+
+       /* Release the ordered list of live variables */
+       free_closure_variables(state, &vars);
+
+       /* Release the storage of the old ids */
+       xfree(info);
+
+       /* Release the variable lifetime information */
+       free_variable_lifetimes(state, &bb, rb);
+
+       /* Release the basic blocks of this function */
+       free_basic_blocks(state, &bb);
+}
+
+static void expand_function_call(
+       struct compile_state *state, struct triple *me, struct triple *fcall)
+{
+       /* Generate an ordinary function call */
+       struct type *closure_type, **closure_next;
+       struct triple *func, *func_first, *func_last, *retvar;
+       struct triple *first;
+       struct type *ptype, *rtype;
+       struct triple *ret_addr, *ret_loc;
+       struct triple_reg_set *enclose, *set;
+       int closure_idx, pvals, i;
+
+#if DEBUG_EXPLICIT_CLOSURES
+       FILE *fp = state->dbgout;
+       fprintf(fp, "\ndisplay_func(me) ptr: %p\n", fcall);
+       display_func(state, fp, MISC(fcall, 0));
+       display_func(state, fp, me);
+       fprintf(fp, "__________ %s _________ done\n\n", __FUNCTION__);
+#endif
+
+       /* Find the triples */
+       func = MISC(fcall, 0);
+       func_first = RHS(func, 0);
+       retvar = fretaddr(state, func);
+       func_last  = func_first->prev;
+       first = fcall->next;
+
+       /* Find what I need to enclose */
+       compute_closure_variables(state, me, fcall, &enclose);
+
+       /* Compute the closure type */
+       closure_type = new_type(TYPE_TUPLE, 0, 0);
+       closure_type->elements = 0;
+       closure_next = &closure_type->left;
+       for(set = enclose; set ; set = set->next) {
+               struct type *type;
+               type = &void_type;
+               if (set->member) {
+                       type = set->member->type;
+               }
+               if (!*closure_next) {
+                       *closure_next = type;
+               } else {
+                       *closure_next = new_type(TYPE_PRODUCT, *closure_next,
+                               type);
+                       closure_next = &(*closure_next)->right;
+               }
+               closure_type->elements += 1;
+       }
+       if (closure_type->elements == 0) {
+               closure_type->type = TYPE_VOID;
+       }
+
+
+#if DEBUG_EXPLICIT_CLOSURES
+       fprintf(state->dbgout, "closure type: ");
+       name_of(state->dbgout, closure_type);
+       fprintf(state->dbgout, "\n");
+#endif
+
+       /* Update the called functions closure variable */
+       closure_idx = add_closure_type(state, func, closure_type);
+
+       /* Generate some needed triples */
+       ret_loc = label(state);
+       ret_addr = triple(state, OP_ADDRCONST, &void_ptr_type, ret_loc, 0);
+
+       /* Pass the parameters to the new function */
+       ptype = func->type->right;
+       pvals = fcall->rhs;
+       for(i = 0; i < pvals; i++) {
+               struct type *atype;
+               struct triple *arg, *param;
+               atype = ptype;
+               if ((ptype->type & TYPE_MASK) == TYPE_PRODUCT) {
+                       atype = ptype->left;
+               }
+               param = farg(state, func, i);
+               if ((param->type->type & TYPE_MASK) != (atype->type & TYPE_MASK)) {
+                       internal_error(state, fcall, "param type mismatch");
+               }
+               arg = RHS(fcall, i);
+               flatten(state, first, write_expr(state, param, arg));
+               ptype = ptype->right;
+       }
+       rtype = func->type->left;
+
+       /* Thread the triples together */
+       ret_loc       = flatten(state, first, ret_loc);
+
+       /* Save the active variables in the result variable */
+       for(i = 0, set = enclose; set ; set = set->next, i++) {
+               if (!set->member) {
+                       continue;
+               }
+               flatten(state, ret_loc,
+                       write_expr(state,
+                               closure_expr(state, func, closure_idx, i),
+                               read_expr(state, set->member)));
+       }
+
+       /* Initialize the return value */
+       if ((rtype->type & TYPE_MASK) != TYPE_VOID) {
+               flatten(state, ret_loc,
+                       write_expr(state,
+                               deref_index(state, fresult(state, func), 1),
+                               new_triple(state, OP_UNKNOWNVAL, rtype,  0, 0)));
+       }
+
+       ret_addr      = flatten(state, ret_loc, ret_addr);
+       flatten(state, ret_loc, write_expr(state, retvar, ret_addr));
+       flatten(state, ret_loc,
+               call(state, retvar, ret_addr, func_first, func_last));
+
+       /* Find the result */
+       if ((rtype->type & TYPE_MASK) != TYPE_VOID) {
+               struct triple * result;
+               result = flatten(state, first,
+                       read_expr(state,
+                               deref_index(state, fresult(state, func), 1)));
+
+               propogate_use(state, fcall, result);
+       }
+
+       /* Release the original fcall instruction */
+       release_triple(state, fcall);
+
+       /* Restore the active variables from the result variable */
+       for(i = 0, set = enclose; set ; set = set->next, i++) {
+               struct triple_set *use, *next;
+               struct triple *new;
+               struct basic_blocks bb;
+               if (!set->member || (set->member == fcall)) {
+                       continue;
+               }
+               /* Generate an expression for the value */
+               new = flatten(state, first,
+                       read_expr(state,
+                               closure_expr(state, func, closure_idx, i)));
+
+
+               /* If the original is an lvalue restore the preserved value */
+               if (is_lvalue(state, set->member)) {
+                       flatten(state, first,
+                               write_expr(state, set->member, new));
+                       continue;
+               }
+               /*
+                * If the original is a value update the dominated uses.
+                */
+
+               /* Analyze the basic blocks so I can see who dominates whom */
+               bb.func = me;
+               bb.first = RHS(me, 0);
+               if (!triple_is_ret(state, bb.first->prev)) {
+                       bb.func = 0;
+               }
+               analyze_basic_blocks(state, &bb);
+
+
+#if DEBUG_EXPLICIT_CLOSURES
+               fprintf(state->errout, "Updating domindated uses: %p -> %p\n",
+                       set->member, new);
+#endif
+               /* If fcall dominates the use update the expression */
+               for(use = set->member->use; use; use = next) {
+                       /* Replace use modifies the use chain and
+                        * removes use, so I must take a copy of the
+                        * next entry early.
+                        */
+                       next = use->next;
+                       if (!tdominates(state, fcall, use->member)) {
+                               continue;
+                       }
+                       replace_use(state, set->member, new, use->member);
+               }
+
+               /* Release the basic blocks, the instructions will be
+                * different next time, and flatten/insert_triple does
+                * not update the block values so I can't cache the analysis.
+                */
+               free_basic_blocks(state, &bb);
+       }
+
+       /* Release the closure variable list */
+       free_closure_variables(state, &enclose);
+
+       if (state->compiler->debug & DEBUG_INLINE) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "\n");
+               loc(fp, state, 0);
+               fprintf(fp, "\n__________ %s _________\n", __FUNCTION__);
+               display_func(state, fp, func);
+               display_func(state, fp, me);
+               fprintf(fp, "__________ %s _________ done\n\n", __FUNCTION__);
+       }
+
+       return;
+}
+
+static int do_inline(struct compile_state *state, struct triple *func)
+{
+       int do_inline;
+       int policy;
+
+       policy = state->compiler->flags & COMPILER_INLINE_MASK;
+       switch(policy) {
+       case COMPILER_INLINE_ALWAYS:
+               do_inline = 1;
+               if (func->type->type & ATTRIB_NOINLINE) {
+                       error(state, func, "noinline with always_inline compiler option");
+               }
+               break;
+       case COMPILER_INLINE_NEVER:
+               do_inline = 0;
+               if (func->type->type & ATTRIB_ALWAYS_INLINE) {
+                       error(state, func, "always_inline with noinline compiler option");
+               }
+               break;
+       case COMPILER_INLINE_DEFAULTON:
+               switch(func->type->type & STOR_MASK) {
+               case STOR_STATIC | STOR_INLINE:
+               case STOR_LOCAL  | STOR_INLINE:
+               case STOR_EXTERN | STOR_INLINE:
+                       do_inline = 1;
+                       break;
+               default:
+                       do_inline = 1;
+                       break;
+               }
+               break;
+       case COMPILER_INLINE_DEFAULTOFF:
+               switch(func->type->type & STOR_MASK) {
+               case STOR_STATIC | STOR_INLINE:
+               case STOR_LOCAL  | STOR_INLINE:
+               case STOR_EXTERN | STOR_INLINE:
+                       do_inline = 1;
+                       break;
+               default:
+                       do_inline = 0;
+                       break;
+               }
+               break;
+       case COMPILER_INLINE_NOPENALTY:
+               switch(func->type->type & STOR_MASK) {
+               case STOR_STATIC | STOR_INLINE:
+               case STOR_LOCAL  | STOR_INLINE:
+               case STOR_EXTERN | STOR_INLINE:
+                       do_inline = 1;
+                       break;
+               default:
+                       do_inline = (func->u.cval == 1);
+                       break;
+               }
+               break;
+       default:
+               do_inline = 0;
+               internal_error(state, 0, "Unimplemented inline policy");
+               break;
+       }
+       /* Force inlining */
+       if (func->type->type & ATTRIB_NOINLINE) {
+               do_inline = 0;
+       }
+       if (func->type->type & ATTRIB_ALWAYS_INLINE) {
+               do_inline = 1;
+       }
+       return do_inline;
+}
+
+static void inline_function(struct compile_state *state, struct triple *me, void *arg)
+{
+       struct triple *first, *ptr, *next;
+       /* If the function is not used don't bother */
+       if (me->u.cval <= 0) {
+               return;
+       }
+       if (state->compiler->debug & DEBUG_CALLS2) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "in: %s\n",
+                       me->type->type_ident->name);
+       }
+
+       first = RHS(me, 0);
+       ptr = next = first;
+       do {
+               struct triple *func, *prev;
+               ptr = next;
+               prev = ptr->prev;
+               next = ptr->next;
+               if (ptr->op != OP_FCALL) {
+                       continue;
+               }
+               func = MISC(ptr, 0);
+               /* See if the function should be inlined */
+               if (!do_inline(state, func)) {
+                       /* Put a label after the fcall */
+                       post_triple(state, ptr, OP_LABEL, &void_type, 0, 0);
+                       continue;
+               }
+               if (state->compiler->debug & DEBUG_CALLS) {
+                       FILE *fp = state->dbgout;
+                       if (state->compiler->debug & DEBUG_CALLS2) {
+                               loc(fp, state, ptr);
+                       }
+                       fprintf(fp, "inlining %s\n",
+                               func->type->type_ident->name);
+                       fflush(fp);
+               }
+
+               /* Update the function use counts */
+               func->u.cval -= 1;
+
+               /* Replace the fcall with the called function */
+               expand_inline_call(state, me, ptr);
+
+               next = prev->next;
+       } while (next != first);
+
+       ptr = next = first;
+       do {
+               struct triple *prev, *func;
+               ptr = next;
+               prev = ptr->prev;
+               next = ptr->next;
+               if (ptr->op != OP_FCALL) {
+                       continue;
+               }
+               func = MISC(ptr, 0);
+               if (state->compiler->debug & DEBUG_CALLS) {
+                       FILE *fp = state->dbgout;
+                       if (state->compiler->debug & DEBUG_CALLS2) {
+                               loc(fp, state, ptr);
+                       }
+                       fprintf(fp, "calling %s\n",
+                               func->type->type_ident->name);
+                       fflush(fp);
+               }
+               /* Replace the fcall with the instruction sequence
+                * needed to make the call.
+                */
+               expand_function_call(state, me, ptr);
+               next = prev->next;
+       } while(next != first);
+}
+
+static void inline_functions(struct compile_state *state, struct triple *func)
+{
+       inline_function(state, func, 0);
+       reverse_walk_functions(state, inline_function, 0);
+}
+
+static void insert_function(struct compile_state *state,
+       struct triple *func, void *arg)
+{
+       struct triple *first, *end, *ffirst, *fend;
+
+       if (state->compiler->debug & DEBUG_INLINE) {
+               FILE *fp = state->errout;
+               fprintf(fp, "%s func count: %d\n",
+                       func->type->type_ident->name, func->u.cval);
+       }
+       if (func->u.cval == 0) {
+               return;
+       }
+
+       /* Find the end points of the lists */
+       first  = arg;
+       end    = first->prev;
+       ffirst = RHS(func, 0);
+       fend   = ffirst->prev;
+
+       /* splice the lists together */
+       end->next    = ffirst;
+       ffirst->prev = end;
+       fend->next   = first;
+       first->prev  = fend;
+}
+
+struct triple *input_asm(struct compile_state *state)
+{
+       struct asm_info *info;
+       struct triple *def;
+       int i, out;
+
+       info = xcmalloc(sizeof(*info), "asm_info");
+       info->str = "";
+
+       out = sizeof(arch_input_regs)/sizeof(arch_input_regs[0]);
+       memcpy(&info->tmpl.lhs, arch_input_regs, sizeof(arch_input_regs));
 
-       /* Remove the parameter scope */
-       end_scope(state);
-#if 0
-       fprintf(stdout, "\n");
-       loc(stdout, state, 0);
-       fprintf(stdout, "\n__________ function_definition _________\n");
-       print_triple(state, def);
-       fprintf(stdout, "__________ function_definition _________ done\n\n");
-#endif
+       def = new_triple(state, OP_ASM, &void_type, out, 0);
+       def->u.ainfo = info;
+       def->id |= TRIPLE_FLAG_VOLATILE;
+
+       for(i = 0; i < out; i++) {
+               struct triple *piece;
+               piece = triple(state, OP_PIECE, &int_type, def, 0);
+               piece->u.cval = i;
+               LHS(def, i) = piece;
+       }
 
        return def;
 }
 
-static struct triple *do_decl(struct compile_state *state, 
-       struct type *type, struct hash_entry *ident)
+struct triple *output_asm(struct compile_state *state)
 {
+       struct asm_info *info;
        struct triple *def;
-       def = 0;
-       /* Clean up the storage types used */
-       switch (type->type & STOR_MASK) {
-       case STOR_AUTO:
-       case STOR_STATIC:
-               /* These are the good types I am aiming for */
-               break;
-       case STOR_REGISTER:
-               type->type &= ~STOR_MASK;
-               type->type |= STOR_AUTO;
-               break;
-       case STOR_EXTERN:
-               type->type &= ~STOR_MASK;
-               type->type |= STOR_STATIC;
-               break;
-       case STOR_TYPEDEF:
-               if (!ident) {
-                       error(state, 0, "typedef without name");
-               }
-               symbol(state, ident, &ident->sym_ident, 0, type);
-               ident->tok = TOK_TYPE_NAME;
-               return 0;
-               break;
-       default:
-               internal_error(state, 0, "Undefined storage class");
-       }
-       if (((type->type & STOR_MASK) == STOR_STATIC) &&
-               ((type->type & QUAL_CONST) == 0)) {
-               error(state, 0, "non const static variables not supported");
-       }
-       if (ident) {
-               def = variable(state, type);
-               symbol(state, ident, &ident->sym_ident, def, type);
-       }
+       int in;
+
+       info = xcmalloc(sizeof(*info), "asm_info");
+       info->str = "";
+
+       in = sizeof(arch_output_regs)/sizeof(arch_output_regs[0]);
+       memcpy(&info->tmpl.rhs, arch_output_regs, sizeof(arch_output_regs));
+
+       def = new_triple(state, OP_ASM, &void_type, 0, in);
+       def->u.ainfo = info;
+       def->id |= TRIPLE_FLAG_VOLATILE;
+
        return def;
 }
 
-static void decl(struct compile_state *state, struct triple *first)
+static void join_functions(struct compile_state *state)
 {
-       struct type *base_type, *type;
-       struct hash_entry *ident;
-       struct triple *def;
-       int global;
-       global = (state->scope_depth <= GLOBAL_SCOPE_DEPTH);
-       base_type = decl_specifiers(state);
-       ident = 0;
-       type = declarator(state, base_type, &ident, 0);
-       if (global && ident && (peek(state) == TOK_LBRACE)) {
-               /* function */
-               def = function_definition(state, type);
-               symbol(state, ident, &ident->sym_ident, def, type);
+       struct triple *start, *end, *call, *in, *out, *func;
+       struct file_state file;
+       struct type *pnext, *param;
+       struct type *result_type, *args_type;
+       int idx;
+
+       /* Be clear the functions have not been joined yet */
+       state->functions_joined = 0;
+
+       /* Dummy file state to get debug handing right */
+       memset(&file, 0, sizeof(file));
+       file.basename = "";
+       file.line = 0;
+       file.report_line = 0;
+       file.report_name = file.basename;
+       file.prev = state->file;
+       state->file = &file;
+       state->function = "";
+
+       if (!state->main_function) {
+               error(state, 0, "No functions to compile\n");
        }
-       else {
-               int done;
-               flatten(state, first, do_decl(state, type, ident));
-               /* type or variable definition */
-               do {
-                       done = 1;
-                       if (peek(state) == TOK_EQ) {
-                               if (!ident) {
-                                       error(state, 0, "cannot assign to a type");
-                               }
-                               eat(state, TOK_EQ);
-                               flatten(state, first,
-                                       init_expr(state, 
-                                               ident->sym_ident->def, 
-                                               initializer(state, type)));
-                       }
-                       arrays_complete(state, type);
-                       if (peek(state) == TOK_COMMA) {
-                               eat(state, TOK_COMMA);
-                               ident = 0;
-                               type = declarator(state, base_type, &ident, 0);
-                               flatten(state, first, do_decl(state, type, ident));
-                               done = 0;
-                       }
-               } while(!done);
-               eat(state, TOK_SEMI);
+
+       /* The type of arguments */
+       args_type   = state->main_function->type->right;
+       /* The return type without any specifiers */
+       result_type = clone_type(0, state->main_function->type->left);
+
+
+       /* Verify the external arguments */
+       if (registers_of(state, args_type) > ARCH_INPUT_REGS) {
+               error(state, state->main_function,
+                       "Too many external input arguments");
+       }
+       if (registers_of(state, result_type) > ARCH_OUTPUT_REGS) {
+               error(state, state->main_function,
+                       "Too many external output arguments");
        }
-}
 
-static void decls(struct compile_state *state)
-{
-       struct triple *list;
-       int tok;
-       list = label(state);
-       while(1) {
-               tok = peek(state);
-               if (tok == TOK_EOF) {
-                       return;
+       /* Lay down the basic program structure */
+       end           = label(state);
+       start         = label(state);
+       start         = flatten(state, state->first, start);
+       end           = flatten(state, state->first, end);
+       in            = input_asm(state);
+       out           = output_asm(state);
+       call          = new_triple(state, OP_FCALL, result_type, -1, registers_of(state, args_type));
+       MISC(call, 0) = state->main_function;
+       in            = flatten(state, state->first, in);
+       call          = flatten(state, state->first, call);
+       out           = flatten(state, state->first, out);
+
+
+       /* Read the external input arguments */
+       pnext = args_type;
+       idx = 0;
+       while(pnext && ((pnext->type & TYPE_MASK) != TYPE_VOID)) {
+               struct triple *expr;
+               param = pnext;
+               pnext = 0;
+               if ((param->type & TYPE_MASK) == TYPE_PRODUCT) {
+                       pnext = param->right;
+                       param = param->left;
                }
-               if (tok == TOK_SPACE) {
-                       eat(state, TOK_SPACE);
+               if (registers_of(state, param) != 1) {
+                       error(state, state->main_function,
+                               "Arg: %d %s requires multiple registers",
+                               idx + 1, param->field_ident->name);
                }
-               decl(state, list);
-               if (list->next != list) {
-                       error(state, 0, "global variables not supported");
+               expr = read_expr(state, LHS(in, idx));
+               RHS(call, idx) = expr;
+               expr = flatten(state, call, expr);
+               use_triple(expr, call);
+
+               idx++;
+       }
+
+
+       /* Write the external output arguments */
+       pnext = result_type;
+       if ((pnext->type & TYPE_MASK) == TYPE_STRUCT) {
+               pnext = result_type->left;
+       }
+       for(idx = 0; idx < out->rhs; idx++) {
+               struct triple *expr;
+               param = pnext;
+               pnext = 0;
+               if (param && ((param->type & TYPE_MASK) == TYPE_PRODUCT)) {
+                       pnext = param->right;
+                       param = param->left;
+               }
+               if (param && ((param->type & TYPE_MASK) == TYPE_VOID)) {
+                       param = 0;
+               }
+               if (param) {
+                       if (registers_of(state, param) != 1) {
+                               error(state, state->main_function,
+                                       "Result: %d %s requires multiple registers",
+                                       idx, param->field_ident->name);
+                       }
+                       expr = read_expr(state, call);
+                       if ((result_type->type & TYPE_MASK) == TYPE_STRUCT) {
+                               expr = deref_field(state, expr, param->field_ident);
+                       }
+               } else {
+                       expr = triple(state, OP_UNKNOWNVAL, &int_type, 0, 0);
                }
+               flatten(state, out, expr);
+               RHS(out, idx) = expr;
+               use_triple(expr, out);
+       }
+
+       /* Allocate a dummy containing function */
+       func = triple(state, OP_LIST,
+               new_type(TYPE_FUNCTION, &void_type, &void_type), 0, 0);
+       func->type->type_ident = lookup(state, "", 0);
+       RHS(func, 0) = state->first;
+       func->u.cval = 1;
+
+       /* See which functions are called, and how often */
+       mark_live_functions(state);
+       inline_functions(state, func);
+       walk_functions(state, insert_function, end);
+
+       if (start->next != end) {
+               flatten(state, start, branch(state, end, 0));
        }
+
+       /* OK now the functions have been joined. */
+       state->functions_joined = 1;
+
+       /* Done now cleanup */
+       state->file = file.prev;
+       state->function = 0;
 }
 
 /*
  * Data structurs for optimation.
  */
 
-static void do_use_block(
-       struct block *used, struct block_set **head, struct block *user, 
+
+static int do_use_block(
+       struct block *used, struct block_set **head, struct block *user,
        int front)
 {
        struct block_set **ptr, *new;
        if (!used)
-               return;
+               return 0;
        if (!user)
-               return;
+               return 0;
        ptr = head;
        while(*ptr) {
                if ((*ptr)->member == user) {
-                       return;
+                       return 0;
                }
                ptr = &(*ptr)->next;
        }
@@ -8777,11 +14905,14 @@ static void do_use_block(
                new->next = 0;
                *ptr = new;
        }
+       return 1;
 }
-static void do_unuse_block(
+static int do_unuse_block(
        struct block *used, struct block_set **head, struct block *unuser)
 {
        struct block_set *use, **ptr;
+       int count;
+       count = 0;
        ptr = head;
        while(*ptr) {
                use = *ptr;
@@ -8789,25 +14920,43 @@ static void do_unuse_block(
                        *ptr = use->next;
                        memset(use, -1, sizeof(*use));
                        xfree(use);
+                       count += 1;
                }
                else {
                        ptr = &use->next;
                }
        }
+       return count;
 }
 
 static void use_block(struct block *used, struct block *user)
 {
+       int count;
        /* Append new to the head of the list, print_block
         * depends on this.
         */
-       do_use_block(used, &used->use, user, 1); 
-       used->users++;
+       count = do_use_block(used, &used->use, user, 1);
+       used->users += count;
 }
 static void unuse_block(struct block *used, struct block *unuser)
 {
-       do_unuse_block(used, &used->use, unuser); 
-       used->users--;
+       int count;
+       count = do_unuse_block(used, &used->use, unuser);
+       used->users -= count;
+}
+
+static void add_block_edge(struct block *block, struct block *edge, int front)
+{
+       int count;
+       count = do_use_block(block, &block->edges, edge, front);
+       block->edge_count += count;
+}
+
+static void remove_block_edge(struct block *block, struct block *edge)
+{
+       int count;
+       count = do_unuse_block(block, &block->edges, edge);
+       block->edge_count -= count;
 }
 
 static void idom_block(struct block *idom, struct block *user)
@@ -8850,49 +14999,28 @@ static void unipdomf_block(struct block *block, struct block *unipdomf)
        do_unuse_block(block, &block->ipdomfrontier, unipdomf);
 }
 
-
-
-static int do_walk_triple(struct compile_state *state,
-       struct triple *ptr, int depth,
-       int (*cb)(struct compile_state *state, struct triple *ptr, int depth)) 
+static int walk_triples(
+       struct compile_state *state,
+       int (*cb)(struct compile_state *state, struct triple *ptr, void *arg),
+       void *arg)
 {
+       struct triple *ptr;
        int result;
-       result = cb(state, ptr, depth);
-       if ((result == 0) && (ptr->op == OP_LIST)) {
-               struct triple *list;
-               list = ptr;
-               ptr = RHS(list, 0);
-               do {
-                       result = do_walk_triple(state, ptr, depth + 1, cb);
-                       if (ptr->next->prev != ptr) {
-                               internal_error(state, ptr->next, "bad prev");
-                       }
-                       ptr = ptr->next;
-                       
-               } while((result == 0) && (ptr != RHS(list, 0)));
-       }
+       ptr = state->first;
+       do {
+               result = cb(state, ptr, arg);
+               if (ptr->next->prev != ptr) {
+                       internal_error(state, ptr->next, "bad prev");
+               }
+               ptr = ptr->next;
+       } while((result == 0) && (ptr != state->first));
        return result;
 }
 
-static int walk_triple(
-       struct compile_state *state, 
-       struct triple *ptr, 
-       int (*cb)(struct compile_state *state, struct triple *ptr, int depth))
-{
-       return do_walk_triple(state, ptr, 0, cb);
-}
-
-static void do_print_prefix(int depth)
-{
-       int i;
-       for(i = 0; i < depth; i++) {
-               printf("  ");
-       }
-}
-
 #define PRINT_LIST 1
-static int do_print_triple(struct compile_state *state, struct triple *ins, int depth)
+static int do_print_triple(struct compile_state *state, struct triple *ins, void *arg)
 {
+       FILE *fp = arg;
        int op;
        op = ins->op;
        if (op == OP_LIST) {
@@ -8901,36 +15029,28 @@ static int do_print_triple(struct compile_state *state, struct triple *ins, int
 #endif
        }
        if ((op == OP_LABEL) && (ins->use)) {
-               printf("\n%p:\n", ins);
+               fprintf(fp, "\n%p:\n", ins);
        }
-       do_print_prefix(depth);
-       display_triple(stdout, ins);
+       display_triple(fp, ins);
 
-       if ((ins->op == OP_BRANCH) && ins->use) {
+       if (triple_is_branch(state, ins) && ins->use &&
+               (ins->op != OP_RET) && (ins->op != OP_FCALL)) {
                internal_error(state, ins, "branch used?");
        }
-#if 0
-       {
-               struct triple_set *user;
-               for(user = ins->use; user; user = user->next) {
-                       printf("use: %p\n", user->member);
-               }
-       }
-#endif
        if (triple_is_branch(state, ins)) {
-               printf("\n");
+               fprintf(fp, "\n");
        }
        return 0;
 }
 
-static void print_triple(struct compile_state *state, struct triple *ins)
-{
-       walk_triple(state, ins, do_print_triple);
-}
-
 static void print_triples(struct compile_state *state)
 {
-       print_triple(state, state->main_function);
+       if (state->compiler->debug & DEBUG_TRIPLES) {
+               FILE *fp = state->dbgout;
+               fprintf(fp, "--------------- triples ---------------\n");
+               walk_triples(state, do_print_triple, fp);
+               fprintf(fp, "\n");
+       }
 }
 
 struct cf_block {
@@ -8938,62 +15058,151 @@ struct cf_block {
 };
 static void find_cf_blocks(struct cf_block *cf, struct block *block)
 {
+       struct block_set *edge;
        if (!block || (cf[block->vertex].block == block)) {
                return;
        }
        cf[block->vertex].block = block;
-       find_cf_blocks(cf, block->left);
-       find_cf_blocks(cf, block->right);
+       for(edge = block->edges; edge; edge = edge->next) {
+               find_cf_blocks(cf, edge->member);
+       }
+}
+
+static void print_control_flow(struct compile_state *state,
+       FILE *fp, struct basic_blocks *bb)
+{
+       struct cf_block *cf;
+       int i;
+       fprintf(fp, "\ncontrol flow\n");
+       cf = xcmalloc(sizeof(*cf) * (bb->last_vertex + 1), "cf_block");
+       find_cf_blocks(cf, bb->first_block);
+
+       for(i = 1; i <= bb->last_vertex; i++) {
+               struct block *block;
+               struct block_set *edge;
+               block = cf[i].block;
+               if (!block)
+                       continue;
+               fprintf(fp, "(%p) %d:", block, block->vertex);
+               for(edge = block->edges; edge; edge = edge->next) {
+                       fprintf(fp, " %d", edge->member->vertex);
+               }
+               fprintf(fp, "\n");
+       }
+
+       xfree(cf);
+}
+
+static void free_basic_block(struct compile_state *state, struct block *block)
+{
+       struct block_set *edge, *entry;
+       struct block *child;
+       if (!block) {
+               return;
+       }
+       if (block->vertex == -1) {
+               return;
+       }
+       block->vertex = -1;
+       for(edge = block->edges; edge; edge = edge->next) {
+               if (edge->member) {
+                       unuse_block(edge->member, block);
+               }
+       }
+       if (block->idom) {
+               unidom_block(block->idom, block);
+       }
+       block->idom = 0;
+       if (block->ipdom) {
+               unipdom_block(block->ipdom, block);
+       }
+       block->ipdom = 0;
+       while((entry = block->use)) {
+               child = entry->member;
+               unuse_block(block, child);
+               if (child && (child->vertex != -1)) {
+                       for(edge = child->edges; edge; edge = edge->next) {
+                               edge->member = 0;
+                       }
+               }
+       }
+       while((entry = block->idominates)) {
+               child = entry->member;
+               unidom_block(block, child);
+               if (child && (child->vertex != -1)) {
+                       child->idom = 0;
+               }
+       }
+       while((entry = block->domfrontier)) {
+               child = entry->member;
+               undomf_block(block, child);
+       }
+       while((entry = block->ipdominates)) {
+               child = entry->member;
+               unipdom_block(block, child);
+               if (child && (child->vertex != -1)) {
+                       child->ipdom = 0;
+               }
+       }
+       while((entry = block->ipdomfrontier)) {
+               child = entry->member;
+               unipdomf_block(block, child);
+       }
+       if (block->users != 0) {
+               internal_error(state, 0, "block still has users");
+       }
+       while((edge = block->edges)) {
+               child = edge->member;
+               remove_block_edge(block, child);
+
+               if (child && (child->vertex != -1)) {
+                       free_basic_block(state, child);
+               }
+       }
+       memset(block, -1, sizeof(*block));
+#ifndef WIN32
+       xfree(block);
+#endif
 }
 
-static void print_control_flow(struct compile_state *state)
+static void free_basic_blocks(struct compile_state *state,
+       struct basic_blocks *bb)
 {
-       struct cf_block *cf;
-       int i;
-       printf("\ncontrol flow\n");
-       cf = xcmalloc(sizeof(*cf) * (state->last_vertex + 1), "cf_block");
-       find_cf_blocks(cf, state->first_block);
-
-       for(i = 1; i <= state->last_vertex; i++) {
-               struct block *block;
-               block = cf[i].block;
-               if (!block)
-                       continue;
-               printf("(%p) %d:", block, block->vertex);
-               if (block->left) {
-                       printf(" %d", block->left->vertex);
-               }
-               if (block->right && (block->right != block->left)) {
-                       printf(" %d", block->right->vertex);
+       struct triple *first, *ins;
+       free_basic_block(state, bb->first_block);
+       bb->last_vertex = 0;
+       bb->first_block = bb->last_block = 0;
+       first = bb->first;
+       ins = first;
+       do {
+               if (triple_stores_block(state, ins)) {
+                       ins->u.block = 0;
                }
-               printf("\n");
-       }
+               ins = ins->next;
+       } while(ins != first);
 
-       xfree(cf);
 }
 
-
 static struct block *basic_block(struct compile_state *state,
-       struct triple *first)
+       struct basic_blocks *bb, struct triple *first)
 {
        struct block *block;
        struct triple *ptr;
-       int op;
-       if (first->op != OP_LABEL) {
-               internal_error(state, 0, "block does not start with a label");
+       if (!triple_is_label(state, first)) {
+               internal_error(state, first, "block does not start with a label");
        }
        /* See if this basic block has already been setup */
        if (first->u.block != 0) {
                return first->u.block;
        }
        /* Allocate another basic block structure */
-       state->last_vertex += 1;
+       bb->last_vertex += 1;
        block = xcmalloc(sizeof(*block), "block");
        block->first = block->last = first;
-       block->vertex = state->last_vertex;
+       block->vertex = bb->last_vertex;
        ptr = first;
        do {
-               if ((ptr != first) && (ptr->op == OP_LABEL) && ptr->use) {
+               if ((ptr != first) && triple_is_label(state, ptr) && (ptr->use)) {
                        break;
                }
                block->last = ptr;
@@ -9001,49 +15210,104 @@ static struct block *basic_block(struct compile_state *state,
                if (triple_stores_block(state, ptr)) {
                        ptr->u.block = block;
                }
-               if (ptr->op == OP_BRANCH) {
+               if (triple_is_branch(state, ptr)) {
                        break;
                }
                ptr = ptr->next;
-       } while (ptr != RHS(state->main_function, 0));
-       if (ptr == RHS(state->main_function, 0))
-               return block;
-       op = ptr->op;
-       if (op == OP_LABEL) {
-               block->left = basic_block(state, ptr);
-               block->right = 0;
-               use_block(block->left, block);
-       }
-       else if (op == OP_BRANCH) {
-               block->left = 0;
-               /* Trace the branch target */
-               block->right = basic_block(state, TARG(ptr, 0));
-               use_block(block->right, block);
-               /* If there is a test trace the branch as well */
-               if (TRIPLE_RHS(ptr->sizes)) {
-                       block->left = basic_block(state, ptr->next);
-                       use_block(block->left, block);
+       } while (ptr != bb->first);
+       if ((ptr == bb->first) ||
+               ((ptr->next == bb->first) && (
+                       triple_is_end(state, ptr) ||
+                       triple_is_ret(state, ptr))))
+       {
+               /* The block has no outflowing edges */
+       }
+       else if (triple_is_label(state, ptr)) {
+               struct block *next;
+               next = basic_block(state, bb, ptr);
+               add_block_edge(block, next, 0);
+               use_block(next, block);
+       }
+       else if (triple_is_branch(state, ptr)) {
+               struct triple **expr, *first;
+               struct block *child;
+               /* Find the branch targets.
+                * I special case the first branch as that magically
+                * avoids some difficult cases for the register allocator.
+                */
+               expr = triple_edge_targ(state, ptr, 0);
+               if (!expr) {
+                       internal_error(state, ptr, "branch without targets");
+               }
+               first = *expr;
+               expr = triple_edge_targ(state, ptr, expr);
+               for(; expr; expr = triple_edge_targ(state, ptr, expr)) {
+                       if (!*expr) continue;
+                       child = basic_block(state, bb, *expr);
+                       use_block(child, block);
+                       add_block_edge(block, child, 0);
+               }
+               if (first) {
+                       child = basic_block(state, bb, first);
+                       use_block(child, block);
+                       add_block_edge(block, child, 1);
+
+                       /* Be certain the return block of a call is
+                        * in a basic block.  When it is not find
+                        * start of the block, insert a label if
+                        * necessary and build the basic block.
+                        * Then add a fake edge from the start block
+                        * to the return block of the function.
+                        */
+                       if (state->functions_joined && triple_is_call(state, ptr)
+                               && !block_of_triple(state, MISC(ptr, 0))) {
+                               struct block *tail;
+                               struct triple *start;
+                               start = triple_to_block_start(state, MISC(ptr, 0));
+                               if (!triple_is_label(state, start)) {
+                                       start = pre_triple(state,
+                                               start, OP_LABEL, &void_type, 0, 0);
+                               }
+                               tail = basic_block(state, bb, start);
+                               add_block_edge(child, tail, 0);
+                               use_block(tail, child);
+                       }
                }
        }
        else {
                internal_error(state, 0, "Bad basic block split");
        }
+#if 0
+{
+       struct block_set *edge;
+       FILE *fp = state->errout;
+       fprintf(fp, "basic_block: %10p [%2d] ( %10p - %10p )",
+               block, block->vertex,
+               block->first, block->last);
+       for(edge = block->edges; edge; edge = edge->next) {
+               fprintf(fp, " %10p [%2d]",
+                       edge->member ? edge->member->first : 0,
+                       edge->member ? edge->member->vertex : -1);
+       }
+       fprintf(fp, "\n");
+}
+#endif
        return block;
 }
 
 
-static void walk_blocks(struct compile_state *state,
+static void walk_blocks(struct compile_state *state, struct basic_blocks *bb,
        void (*cb)(struct compile_state *state, struct block *block, void *arg),
        void *arg)
 {
        struct triple *ptr, *first;
        struct block *last_block;
        last_block = 0;
-       first = RHS(state->main_function, 0);
+       first = bb->first;
        ptr = first;
        do {
-               struct block *block;
-               if (ptr->op == OP_LABEL) {
+               if (triple_stores_block(state, ptr)) {
+                       struct block *block;
                        block = ptr->u.block;
                        if (block && (block != last_block)) {
                                cb(state, block, arg);
@@ -9057,81 +15321,69 @@ static void walk_blocks(struct compile_state *state,
 static void print_block(
        struct compile_state *state, struct block *block, void *arg)
 {
+       struct block_set *user, *edge;
        struct triple *ptr;
        FILE *fp = arg;
 
-       fprintf(fp, "\nblock: %p (%d), %p<-%p %p<-%p\n", 
-               block, 
-               block->vertex,
-               block->left, 
-               block->left && block->left->use?block->left->use->member : 0,
-               block->right, 
-               block->right && block->right->use?block->right->use->member : 0);
+       fprintf(fp, "\nblock: %p (%d) ",
+               block,
+               block->vertex);
+
+       for(edge = block->edges; edge; edge = edge->next) {
+               fprintf(fp, " %p<-%p",
+                       edge->member,
+                       (edge->member && edge->member->use)?
+                       edge->member->use->member : 0);
+       }
+       fprintf(fp, "\n");
        if (block->first->op == OP_LABEL) {
                fprintf(fp, "%p:\n", block->first);
        }
-       for(ptr = block->first; ; ptr = ptr->next) {
-               struct triple_set *user;
-               int op = ptr->op;
-               
-               if (triple_stores_block(state, ptr)) {
-                       if (ptr->u.block != block) {
-                               internal_error(state, ptr, 
-                                       "Wrong block pointer: %p\n",
-                                       ptr->u.block);
-                       }
-               }
-               if (op == OP_ADECL) {
-                       for(user = ptr->use; user; user = user->next) {
-                               if (!user->member->u.block) {
-                                       internal_error(state, user->member, 
-                                               "Use %p not in a block?\n",
-                                               user->member);
-                               }
-                       }
-               }
+       for(ptr = block->first; ; ) {
                display_triple(fp, ptr);
-
-#if 0
-               for(user = ptr->use; user; user = user->next) {
-                       fprintf(fp, "use: %p\n", user->member);
-               }
-#endif
-
-               /* Sanity checks... */
-               valid_ins(state, ptr);
-               for(user = ptr->use; user; user = user->next) {
-                       struct triple *use;
-                       use = user->member;
-                       valid_ins(state, use);
-                       if (triple_stores_block(state, user->member) &&
-                               !user->member->u.block) {
-                               internal_error(state, user->member,
-                                       "Use %p not in a block?",
-                                       user->member);
-                       }
-               }
-
                if (ptr == block->last)
                        break;
+               ptr = ptr->next;
+               if (ptr == block->first) {
+                       internal_error(state, 0, "missing block last?");
+               }
        }
-       fprintf(fp,"\n");
+       fprintf(fp, "users %d: ", block->users);
+       for(user = block->use; user; user = user->next) {
+               fprintf(fp, "%p (%d) ",
+                       user->member,
+                       user->member->vertex);
+       }
+       fprintf(fp,"\n\n");
 }
 
 
-static void print_blocks(struct compile_state *state, FILE *fp)
+static void romcc_print_blocks(struct compile_state *state, FILE *fp)
 {
        fprintf(fp, "--------------- blocks ---------------\n");
-       walk_blocks(state, print_block, fp);
+       walk_blocks(state, &state->bb, print_block, fp);
+}
+static void print_blocks(struct compile_state *state, const char *func, FILE *fp)
+{
+       if (state->compiler->debug & DEBUG_BASIC_BLOCKS) {
+               fprintf(fp, "After %s\n", func);
+               romcc_print_blocks(state, fp);
+               if (state->compiler->debug & DEBUG_FDOMINATORS) {
+                       print_dominators(state, fp, &state->bb);
+                       print_dominance_frontiers(state, fp, &state->bb);
+               }
+               print_control_flow(state, fp, &state->bb);
+       }
 }
 
-static void prune_nonblock_triples(struct compile_state *state)
+static void prune_nonblock_triples(struct compile_state *state,
+       struct basic_blocks *bb)
 {
        struct block *block;
        struct triple *first, *ins, *next;
        /* Delete the triples not in a basic block */
-       first = RHS(state->main_function, 0);
        block = 0;
+       first = bb->first;
        ins = first;
        do {
                next = ins->next;
@@ -9139,124 +15391,72 @@ static void prune_nonblock_triples(struct compile_state *state)
                        block = ins->u.block;
                }
                if (!block) {
+                       struct triple_set *use;
+                       for(use = ins->use; use; use = use->next) {
+                               struct block *block;
+                               block = block_of_triple(state, use->member);
+                               if (block != 0) {
+                                       internal_error(state, ins, "pruning used ins?");
+                               }
+                       }
                        release_triple(state, ins);
                }
+               if (block && block->last == ins) {
+                       block = 0;
+               }
                ins = next;
        } while(ins != first);
 }
 
-static void setup_basic_blocks(struct compile_state *state)
+static void setup_basic_blocks(struct compile_state *state,
+       struct basic_blocks *bb)
 {
-       if (!triple_stores_block(state, RHS(state->main_function, 0)) ||
-               !triple_stores_block(state, RHS(state->main_function,0)->prev)) {
+       if (!triple_stores_block(state, bb->first)) {
                internal_error(state, 0, "ins will not store block?");
        }
+       /* Initialize the state */
+       bb->first_block = bb->last_block = 0;
+       bb->last_vertex = 0;
+       free_basic_blocks(state, bb);
+
        /* Find the basic blocks */
-       state->last_vertex = 0;
-       state->first_block = basic_block(state, RHS(state->main_function,0));
-       /* Delete the triples not in a basic block */
-       prune_nonblock_triples(state);
-       /* Find the last basic block */
-       state->last_block = RHS(state->main_function, 0)->prev->u.block;
-       if (!state->last_block) {
-               internal_error(state, 0, "end not used?");
-       }
-       /* Insert an extra unused edge from start to the end 
-        * This helps with reverse control flow calculations.
-        */
-       use_block(state->first_block, state->last_block);
-       /* If we are debugging print what I have just done */
-       if (state->debug & DEBUG_BASIC_BLOCKS) {
-               print_blocks(state, stdout);
-               print_control_flow(state);
-       }
-}
+       bb->first_block = basic_block(state, bb, bb->first);
 
-static void free_basic_block(struct compile_state *state, struct block *block)
-{
-       struct block_set *entry, *next;
-       struct block *child;
-       if (!block) {
-               return;
-       }
-       if (block->vertex == -1) {
-               return;
-       }
-       block->vertex = -1;
-       if (block->left) {
-               unuse_block(block->left, block);
-       }
-       if (block->right) {
-               unuse_block(block->right, block);
-       }
-       if (block->idom) {
-               unidom_block(block->idom, block);
-       }
-       block->idom = 0;
-       if (block->ipdom) {
-               unipdom_block(block->ipdom, block);
-       }
-       block->ipdom = 0;
-       for(entry = block->use; entry; entry = next) {
-               next = entry->next;
-               child = entry->member;
-               unuse_block(block, child);
-               if (child->left == block) {
-                       child->left = 0;
-               }
-               if (child->right == block) {
-                       child->right = 0;
+       /* Be certain the last instruction of a function, or the
+        * entire program is in a basic block.  When it is not find
+        * the start of the block, insert a label if necessary and build
+        * basic block.  Then add a fake edge from the start block
+        * to the final block.
+        */
+       if (!block_of_triple(state, bb->first->prev)) {
+               struct triple *start;
+               struct block *tail;
+               start = triple_to_block_start(state, bb->first->prev);
+               if (!triple_is_label(state, start)) {
+                       start = pre_triple(state,
+                               start, OP_LABEL, &void_type, 0, 0);
                }
+               tail = basic_block(state, bb, start);
+               add_block_edge(bb->first_block, tail, 0);
+               use_block(tail, bb->first_block);
        }
-       for(entry = block->idominates; entry; entry = next) {
-               next = entry->next;
-               child = entry->member;
-               unidom_block(block, child);
-               child->idom = 0;
-       }
-       for(entry = block->domfrontier; entry; entry = next) {
-               next = entry->next;
-               child = entry->member;
-               undomf_block(block, child);
-       }
-       for(entry = block->ipdominates; entry; entry = next) {
-               next = entry->next;
-               child = entry->member;
-               unipdom_block(block, child);
-               child->ipdom = 0;
-       }
-       for(entry = block->ipdomfrontier; entry; entry = next) {
-               next = entry->next;
-               child = entry->member;
-               unipdomf_block(block, child);
-       }
-       if (block->users != 0) {
-               internal_error(state, 0, "block still has users");
+
+       /* Find the last basic block.
+        */
+       bb->last_block = block_of_triple(state, bb->first->prev);
+
+       /* Delete the triples not in a basic block */
+       prune_nonblock_triples(state, bb);
+
+#if 0
+       /* If we are debugging print what I have just done */
+       if (state->compiler->debug & DEBUG_BASIC_BLOCKS) {
+               print_blocks(state, state->dbgout);
+               print_control_flow(state, bb);
        }
-       free_basic_block(state, block->left);
-       block->left = 0;
-       free_basic_block(state, block->right);
-       block->right = 0;
-       memset(block, -1, sizeof(*block));
-       xfree(block);
+#endif
 }
 
-static void free_basic_blocks(struct compile_state *state)
-{
-       struct triple *first, *ins;
-       free_basic_block(state, state->first_block);
-       state->last_vertex = 0;
-       state->first_block = state->last_block = 0;
-       first = RHS(state->main_function, 0);
-       ins = first;
-       do {
-               if (triple_stores_block(state, ins)) {
-                       ins->u.block = 0;
-               }
-               ins = ins->next;
-       } while(ins != first);
-       
-}
 
 struct sdom_block {
        struct block *block;
@@ -9299,6 +15499,7 @@ static void sdom_block(struct sdom_block *sdom, struct sdom_block *block)
 static int initialize_sdblock(struct sdom_block *sd,
        struct block *parent, struct block *block, int vertex)
 {
+       struct block_set *edge;
        if (!block || (sd[block->vertex].block == block)) {
                return vertex;
        }
@@ -9311,12 +15512,14 @@ static int initialize_sdblock(struct sdom_block *sd,
        sd[vertex].parent   = parent? &sd[parent->vertex] : 0;
        sd[vertex].ancestor = 0;
        sd[vertex].vertex   = vertex;
-       vertex = initialize_sdblock(sd, block, block->left, vertex);
-       vertex = initialize_sdblock(sd, block, block->right, vertex);
+       for(edge = block->edges; edge; edge = edge->next) {
+               vertex = initialize_sdblock(sd, block, edge->member, vertex);
+       }
        return vertex;
 }
 
-static int initialize_sdpblock(struct sdom_block *sd,
+static int initialize_spdblock(
+       struct compile_state *state, struct sdom_block *sd,
        struct block *parent, struct block *block, int vertex)
 {
        struct block_set *user;
@@ -9333,7 +15536,38 @@ static int initialize_sdpblock(struct sdom_block *sd,
        sd[vertex].ancestor = 0;
        sd[vertex].vertex   = vertex;
        for(user = block->use; user; user = user->next) {
-               vertex = initialize_sdpblock(sd, block, user->member, vertex);
+               vertex = initialize_spdblock(state, sd, block, user->member, vertex);
+       }
+       return vertex;
+}
+
+static int setup_spdblocks(struct compile_state *state,
+       struct basic_blocks *bb, struct sdom_block *sd)
+{
+       struct block *block;
+       int vertex;
+       /* Setup as many sdpblocks as possible without using fake edges */
+       vertex = initialize_spdblock(state, sd, 0, bb->last_block, 0);
+
+       /* Walk through the graph and find unconnected blocks.  Add a
+        * fake edge from the unconnected blocks to the end of the
+        * graph.
+        */
+       block = bb->first_block->last->next->u.block;
+       for(; block && block != bb->first_block; block = block->last->next->u.block) {
+               if (sd[block->vertex].block == block) {
+                       continue;
+               }
+#if DEBUG_SDP_BLOCKS
+               {
+                       FILE *fp = state->errout;
+                       fprintf(fp, "Adding %d\n", vertex +1);
+               }
+#endif
+               add_block_edge(block, bb->last_block, 0);
+               use_block(bb->last_block, block);
+
+               vertex = initialize_spdblock(state, sd, bb->last_block, block, vertex);
        }
        return vertex;
 }
@@ -9361,15 +15595,16 @@ static void compress_ancestors(struct sdom_block *v)
        }
 }
 
-static void compute_sdom(struct compile_state *state, struct sdom_block *sd)
+static void compute_sdom(struct compile_state *state,
+       struct basic_blocks *bb, struct sdom_block *sd)
 {
        int i;
-       /* // step 2 
+       /* // step 2
         *  for each v <= pred(w) {
         *      u = EVAL(v);
-        *      if (semi[u] < semi[w] { 
-        *              semi[w] = semi[u]; 
-        *      } 
+        *      if (semi[u] < semi[w] {
+        *              semi[w] = semi[u];
+        *      }
         * }
         * add w to bucket(vertex(semi[w]));
         * LINK(parent(w), w);
@@ -9381,7 +15616,7 @@ static void compute_sdom(struct compile_state *state, struct sdom_block *sd)
         *      dom(v) = (semi[u] < semi[v]) ? u : parent(w);
         * }
         */
-       for(i = state->last_vertex; i >= 2; i--) {
+       for(i = bb->last_vertex; i >= 2; i--) {
                struct sdom_block *v, *parent, *next;
                struct block_set *user;
                struct block *block;
@@ -9404,21 +15639,22 @@ static void compute_sdom(struct compile_state *state, struct sdom_block *sd)
                        next = v->sdom_next;
                        unsdom_block(v);
                        u = (!v->ancestor) ? v : (compress_ancestors(v), v->label);
-                       v->block->idom = (u->sdom->vertex < v->sdom->vertex)? 
+                       v->block->idom = (u->sdom->vertex < v->sdom->vertex)?
                                u->block : parent->block;
                }
        }
 }
 
-static void compute_spdom(struct compile_state *state, struct sdom_block *sd)
+static void compute_spdom(struct compile_state *state,
+       struct basic_blocks *bb, struct sdom_block *sd)
 {
        int i;
-       /* // step 2 
+       /* // step 2
         *  for each v <= pred(w) {
         *      u = EVAL(v);
-        *      if (semi[u] < semi[w] { 
-        *              semi[w] = semi[u]; 
-        *      } 
+        *      if (semi[u] < semi[w] {
+        *              semi[w] = semi[u];
+        *      }
         * }
         * add w to bucket(vertex(semi[w]));
         * LINK(parent(w), w);
@@ -9430,21 +15666,15 @@ static void compute_spdom(struct compile_state *state, struct sdom_block *sd)
         *      dom(v) = (semi[u] < semi[v]) ? u : parent(w);
         * }
         */
-       for(i = state->last_vertex; i >= 2; i--) {
+       for(i = bb->last_vertex; i >= 2; i--) {
                struct sdom_block *u, *v, *parent, *next;
+               struct block_set *edge;
                struct block *block;
                block = sd[i].block;
                parent = sd[i].parent;
                /* Step 2 */
-               if (block->left) {
-                       v = &sd[block->left->vertex];
-                       u = !(v->ancestor)? v : (compress_ancestors(v), v->label);
-                       if (u->sdom->vertex < sd[i].sdom->vertex) {
-                               sd[i].sdom = u->sdom;
-                       }
-               }
-               if (block->right && (block->right != block->left)) {
-                       v = &sd[block->right->vertex];
+               for(edge = block->edges; edge; edge = edge->next) {
+                       v = &sd[edge->member->vertex];
                        u = !(v->ancestor)? v : (compress_ancestors(v), v->label);
                        if (u->sdom->vertex < sd[i].sdom->vertex) {
                                sd[i].sdom = u->sdom;
@@ -9458,16 +15688,17 @@ static void compute_spdom(struct compile_state *state, struct sdom_block *sd)
                        next = v->sdom_next;
                        unsdom_block(v);
                        u = (!v->ancestor) ? v : (compress_ancestors(v), v->label);
-                       v->block->ipdom = (u->sdom->vertex < v->sdom->vertex)? 
+                       v->block->ipdom = (u->sdom->vertex < v->sdom->vertex)?
                                u->block : parent->block;
                }
        }
 }
 
-static void compute_idom(struct compile_state *state, struct sdom_block *sd)
+static void compute_idom(struct compile_state *state,
+       struct basic_blocks *bb, struct sdom_block *sd)
 {
        int i;
-       for(i = 2; i <= state->last_vertex; i++) {
+       for(i = 2; i <= bb->last_vertex; i++) {
                struct block *block;
                block = sd[i].block;
                if (block->idom->vertex != sd[i].sdom->vertex) {
@@ -9478,10 +15709,11 @@ static void compute_idom(struct compile_state *state, struct sdom_block *sd)
        sd[1].block->idom = 0;
 }
 
-static void compute_ipdom(struct compile_state *state, struct sdom_block *sd)
+static void compute_ipdom(struct compile_state *state,
+       struct basic_blocks *bb, struct sdom_block *sd)
 {
        int i;
-       for(i = 2; i <= state->last_vertex; i++) {
+       for(i = 2; i <= bb->last_vertex; i++) {
                struct block *block;
                block = sd[i].block;
                if (block->ipdom->vertex != sd[i].sdom->vertex) {
@@ -9494,13 +15726,13 @@ static void compute_ipdom(struct compile_state *state, struct sdom_block *sd)
 
        /* Theorem 1:
         *   Every vertex of a flowgraph G = (V, E, r) except r has
-        *   a unique immediate dominator.  
+        *   a unique immediate dominator.
         *   The edges {(idom(w), w) |w <= V - {r}} form a directed tree
-        *   rooted at r, called the dominator tree of G, such that 
+        *   rooted at r, called the dominator tree of G, such that
         *   v dominates w if and only if v is a proper ancestor of w in
         *   the dominator tree.
         */
-       /* Lemma 1:  
+       /* Lemma 1:
         *   If v and w are vertices of G such that v <= w,
         *   than any path from v to w must contain a common ancestor
         *   of v and w in T.
@@ -9513,7 +15745,7 @@ static void compute_ipdom(struct compile_state *state, struct sdom_block *sd)
         *   sdom(u) >= sdom(w).  Then idom(w) = sdom(w).
         */
        /* Theorem 3:
-        *   Let w != r and let u be a vertex for which sdom(u) is 
+        *   Let w != r and let u be a vertex for which sdom(u) is
         *   minimum amoung vertices u satisfying sdom(w) -> u -> w.
         *   Then sdom(u) <= sdom(w) and idom(u) = idom(w).
         */
@@ -9521,7 +15753,8 @@ static void compute_ipdom(struct compile_state *state, struct sdom_block *sd)
         *           Then v -> idom(w) or idom(w) -> idom(v)
         */
 
-static void find_immediate_dominators(struct compile_state *state)
+static void find_immediate_dominators(struct compile_state *state,
+       struct basic_blocks *bb)
 {
        struct sdom_block *sd;
        /* w->sdom = min{v| there is a path v = v0,v1,...,vk = w such that:
@@ -9530,11 +15763,11 @@ static void find_immediate_dominators(struct compile_state *state)
        /* Theorem 4:
         *   For any vertex w != r.
         *   sdom(w) = min(
-        *                 {v|(v,w) <= E  and v < w } U 
+        *                 {v|(v,w) <= E  and v < w } U
         *                 {sdom(u) | u > w and there is an edge (v, w) such that u -> v})
         */
        /* Corollary 1:
-        *   Let w != r and let u be a vertex for which sdom(u) is 
+        *   Let w != r and let u be a vertex for which sdom(u) is
         *   minimum amoung vertices u satisfying sdom(w) -> u -> w.
         *   Then:
         *                   { sdom(w) if sdom(w) = sdom(u),
@@ -9542,7 +15775,7 @@ static void find_immediate_dominators(struct compile_state *state)
         *                   { idom(u) otherwise
         */
        /* The algorithm consists of the following 4 steps.
-        * Step 1.  Carry out a depth-first search of the problem graph.  
+        * Step 1.  Carry out a depth-first search of the problem graph.
         *    Number the vertices from 1 to N as they are reached during
         *    the search.  Initialize the variables used in succeeding steps.
         * Step 2.  Compute the semidominators of all vertices by applying
@@ -9555,8 +15788,8 @@ static void find_immediate_dominators(struct compile_state *state)
         *    by number.
         */
        /* Step 1 initialize the basic block information */
-       sd = xcmalloc(sizeof(*sd) * (state->last_vertex + 1), "sdom_state");
-       initialize_sdblock(sd, 0, state->first_block, 0);
+       sd = xcmalloc(sizeof(*sd) * (bb->last_vertex + 1), "sdom_state");
+       initialize_sdblock(sd, 0, bb->first_block, 0);
 #if 0
        sd[1].size  = 0;
        sd[1].label = 0;
@@ -9564,25 +15797,31 @@ static void find_immediate_dominators(struct compile_state *state)
 #endif
        /* Step 2 compute the semidominators */
        /* Step 3 implicitly define the immediate dominator of each vertex */
-       compute_sdom(state, sd);
+       compute_sdom(state, bb, sd);
        /* Step 4 explicitly define the immediate dominator of each vertex */
-       compute_idom(state, sd);
+       compute_idom(state, bb, sd);
        xfree(sd);
 }
 
-static void find_post_dominators(struct compile_state *state)
+static void find_post_dominators(struct compile_state *state,
+       struct basic_blocks *bb)
 {
        struct sdom_block *sd;
+       int vertex;
        /* Step 1 initialize the basic block information */
-       sd = xcmalloc(sizeof(*sd) * (state->last_vertex + 1), "sdom_state");
+       sd = xcmalloc(sizeof(*sd) * (bb->last_vertex + 1), "sdom_state");
 
-       initialize_sdpblock(sd, 0, state->last_block, 0);
+       vertex = setup_spdblocks(state, bb, sd);
+       if (vertex != bb->last_vertex) {
+               internal_error(state, 0, "missing %d blocks",
+                       bb->last_vertex - vertex);
+       }
 
        /* Step 2 compute the semidominators */
        /* Step 3 implicitly define the immediate dominator of each vertex */
-       compute_spdom(state, sd);
+       compute_spdom(state, bb, sd);
        /* Step 4 explicitly define the immediate dominator of each vertex */
-       compute_ipdom(state, sd);
+       compute_ipdom(state, bb, sd);
        xfree(sd);
 }
 
@@ -9591,7 +15830,7 @@ static void find_post_dominators(struct compile_state *state)
 static void find_block_domf(struct compile_state *state, struct block *block)
 {
        struct block *child;
-       struct block_set *user;
+       struct block_set *user, *edge;
        if (block->domfrontier != 0) {
                internal_error(state, block->first, "domfrontier present?");
        }
@@ -9602,11 +15841,10 @@ static void find_block_domf(struct compile_state *state, struct block *block)
                }
                find_block_domf(state, child);
        }
-       if (block->left && block->left->idom != block) {
-               domf_block(block, block->left);
-       }
-       if (block->right && block->right->idom != block) {
-               domf_block(block, block->right);
+       for(edge = block->edges; edge; edge = edge->next) {
+               if (edge->member->idom != block) {
+                       domf_block(block, edge->member);
+               }
        }
        for(user = block->idominates; user; user = user->next) {
                struct block_set *frontier;
@@ -9633,13 +15871,12 @@ static void find_block_ipdomf(struct compile_state *state, struct block *block)
                }
                find_block_ipdomf(state, child);
        }
-       if (block->left && block->left->ipdom != block) {
-               ipdomf_block(block, block->left);
-       }
-       if (block->right && block->right->ipdom != block) {
-               ipdomf_block(block, block->right);
+       for(user = block->use; user; user = user->next) {
+               if (user->member->ipdom != block) {
+                       ipdomf_block(block, user->member);
+               }
        }
-       for(user = block->idominates; user; user = user->next) {
+       for(user = block->ipdominates; user; user = user->next) {
                struct block_set *frontier;
                child = user->member;
                for(frontier = child->ipdomfrontier; frontier; frontier = frontier->next) {
@@ -9666,56 +15903,98 @@ static void print_dominated(
        fprintf(fp,"\n");
 }
 
-static void print_dominators(struct compile_state *state, FILE *fp)
+static void print_dominated2(
+       struct compile_state *state, FILE *fp, int depth, struct block *block)
+{
+       struct block_set *user;
+       struct triple *ins;
+       struct occurance *ptr, *ptr2;
+       const char *filename1, *filename2;
+       int equal_filenames;
+       int i;
+       for(i = 0; i < depth; i++) {
+               fprintf(fp, "   ");
+       }
+       fprintf(fp, "%3d: %p (%p - %p) @",
+               block->vertex, block, block->first, block->last);
+       ins = block->first;
+       while(ins != block->last && (ins->occurance->line == 0)) {
+               ins = ins->next;
+       }
+       ptr = ins->occurance;
+       ptr2 = block->last->occurance;
+       filename1 = ptr->filename? ptr->filename : "";
+       filename2 = ptr2->filename? ptr2->filename : "";
+       equal_filenames = (strcmp(filename1, filename2) == 0);
+       if ((ptr == ptr2) || (equal_filenames && ptr->line == ptr2->line)) {
+               fprintf(fp, " %s:%d", ptr->filename, ptr->line);
+       } else if (equal_filenames) {
+               fprintf(fp, " %s:(%d - %d)",
+                       ptr->filename, ptr->line, ptr2->line);
+       } else {
+               fprintf(fp, " (%s:%d - %s:%d)",
+                       ptr->filename, ptr->line,
+                       ptr2->filename, ptr2->line);
+       }
+       fprintf(fp, "\n");
+       for(user = block->idominates; user; user = user->next) {
+               print_dominated2(state, fp, depth + 1, user->member);
+       }
+}
+
+static void print_dominators(struct compile_state *state, FILE *fp, struct basic_blocks *bb)
 {
        fprintf(fp, "\ndominates\n");
-       walk_blocks(state, print_dominated, fp);
+       walk_blocks(state, bb, print_dominated, fp);
+       fprintf(fp, "dominates\n");
+       print_dominated2(state, fp, 0, bb->first_block);
 }
 
 
 static int print_frontiers(
-       struct compile_state *state, struct block *block, int vertex)
+       struct compile_state *state, FILE *fp, struct block *block, int vertex)
 {
-       struct block_set *user;
+       struct block_set *user, *edge;
 
        if (!block || (block->vertex != vertex + 1)) {
                return vertex;
        }
        vertex += 1;
 
-       printf("%d:", block->vertex);
+       fprintf(fp, "%d:", block->vertex);
        for(user = block->domfrontier; user; user = user->next) {
-               printf(" %d", user->member->vertex);
+               fprintf(fp, " %d", user->member->vertex);
        }
-       printf("\n");
+       fprintf(fp, "\n");
 
-       vertex = print_frontiers(state, block->left, vertex);
-       vertex = print_frontiers(state, block->right, vertex);
+       for(edge = block->edges; edge; edge = edge->next) {
+               vertex = print_frontiers(state, fp, edge->member, vertex);
+       }
        return vertex;
 }
-static void print_dominance_frontiers(struct compile_state *state)
+static void print_dominance_frontiers(struct compile_state *state,
+       FILE *fp, struct basic_blocks *bb)
 {
-       printf("\ndominance frontiers\n");
-       print_frontiers(state, state->first_block, 0);
-       
+       fprintf(fp, "\ndominance frontiers\n");
+       print_frontiers(state, fp, bb->first_block, 0);
+
 }
 
-static void analyze_idominators(struct compile_state *state)
+static void analyze_idominators(struct compile_state *state, struct basic_blocks *bb)
 {
        /* Find the immediate dominators */
-       find_immediate_dominators(state);
+       find_immediate_dominators(state, bb);
        /* Find the dominance frontiers */
-       find_block_domf(state, state->first_block);
+       find_block_domf(state, bb->first_block);
        /* If debuging print the print what I have just found */
-       if (state->debug & DEBUG_FDOMINATORS) {
-               print_dominators(state, stdout);
-               print_dominance_frontiers(state);
-               print_control_flow(state);
+       if (state->compiler->debug & DEBUG_FDOMINATORS) {
+               print_dominators(state, state->dbgout, bb);
+               print_dominance_frontiers(state, state->dbgout, bb);
+               print_control_flow(state, state->dbgout, bb);
        }
 }
 
 
-
 static void print_ipdominated(
        struct compile_state *state, struct block *block, void *arg)
 {
@@ -9732,14 +16011,15 @@ static void print_ipdominated(
        fprintf(fp, "\n");
 }
 
-static void print_ipdominators(struct compile_state *state, FILE *fp)
+static void print_ipdominators(struct compile_state *state, FILE *fp,
+       struct basic_blocks *bb)
 {
        fprintf(fp, "\nipdominates\n");
-       walk_blocks(state, print_ipdominated, fp);
+       walk_blocks(state, bb, print_ipdominated, fp);
 }
 
 static int print_pfrontiers(
-       struct compile_state *state, struct block *block, int vertex)
+       struct compile_state *state, FILE *fp, struct block *block, int vertex)
 {
        struct block_set *user;
 
@@ -9748,34 +16028,36 @@ static int print_pfrontiers(
        }
        vertex += 1;
 
-       printf("%d:", block->vertex);
+       fprintf(fp, "%d:", block->vertex);
        for(user = block->ipdomfrontier; user; user = user->next) {
-               printf(" %d", user->member->vertex);
+               fprintf(fp, " %d", user->member->vertex);
        }
-       printf("\n");
+       fprintf(fp, "\n");
        for(user = block->use; user; user = user->next) {
-               vertex = print_pfrontiers(state, user->member, vertex);
+               vertex = print_pfrontiers(state, fp, user->member, vertex);
        }
        return vertex;
 }
-static void print_ipdominance_frontiers(struct compile_state *state)
+static void print_ipdominance_frontiers(struct compile_state *state,
+       FILE *fp, struct basic_blocks *bb)
 {
-       printf("\nipdominance frontiers\n");
-       print_pfrontiers(state, state->last_block, 0);
-       
+       fprintf(fp, "\nipdominance frontiers\n");
+       print_pfrontiers(state, fp, bb->last_block, 0);
+
 }
 
-static void analyze_ipdominators(struct compile_state *state)
+static void analyze_ipdominators(struct compile_state *state,
+       struct basic_blocks *bb)
 {
        /* Find the post dominators */
-       find_post_dominators(state);
+       find_post_dominators(state, bb);
        /* Find the control dependencies (post dominance frontiers) */
-       find_block_ipdomf(state, state->last_block);
+       find_block_ipdomf(state, bb->last_block);
        /* If debuging print the print what I have just found */
-       if (state->debug & DEBUG_RDOMINATORS) {
-               print_ipdominators(state, stdout);
-               print_ipdominance_frontiers(state);
-               print_control_flow(state);
+       if (state->compiler->debug & DEBUG_RDOMINATORS) {
+               print_ipdominators(state, state->dbgout, bb);
+               print_ipdominance_frontiers(state, state->dbgout, bb);
+               print_control_flow(state, state->dbgout, bb);
        }
 }
 
@@ -9797,9 +16079,12 @@ static int tdominates(struct compile_state *state,
        bsub = block_of_triple(state, sub);
        if (bdom != bsub) {
                result = bdominates(state, bdom, bsub);
-       } 
+       }
        else {
                struct triple *ins;
+               if (!bdom || !bsub) {
+                       internal_error(state, dom, "huh?");
+               }
                ins = sub;
                while((ins != bsub->first) && (ins != dom)) {
                        ins = ins->prev;
@@ -9809,6 +16094,14 @@ static int tdominates(struct compile_state *state,
        return result;
 }
 
+static void analyze_basic_blocks(
+       struct compile_state *state, struct basic_blocks *bb)
+{
+       setup_basic_blocks(state, bb);
+       analyze_idominators(state, bb);
+       analyze_ipdominators(state, bb);
+}
+
 static void insert_phi_operations(struct compile_state *state)
 {
        size_t size;
@@ -9816,34 +16109,43 @@ static void insert_phi_operations(struct compile_state *state)
        int *has_already, *work;
        struct block *work_list, **work_list_tail;
        int iter;
-       struct triple *var;
+       struct triple *var, *vnext;
 
-       size = sizeof(int) * (state->last_vertex + 1);
+       size = sizeof(int) * (state->bb.last_vertex + 1);
        has_already = xcmalloc(size, "has_already");
        work =        xcmalloc(size, "work");
        iter = 0;
 
-       first = RHS(state->main_function, 0);
-       for(var = first->next; var != first ; var = var->next) {
+       first = state->first;
+       for(var = first->next; var != first ; var = vnext) {
                struct block *block;
-               struct triple_set *user;
-               if ((var->op != OP_ADECL) || !var->use) {
+               struct triple_set *user, *unext;
+               vnext = var->next;
+
+               if (!triple_is_auto_var(state, var) || !var->use) {
                        continue;
                }
+
                iter += 1;
                work_list = 0;
                work_list_tail = &work_list;
-               for(user = var->use; user; user = user->next) {
+               for(user = var->use; user; user = unext) {
+                       unext = user->next;
+                       if (MISC(var, 0) == user->member) {
+                               continue;
+                       }
                        if (user->member->op == OP_READ) {
                                continue;
                        }
                        if (user->member->op != OP_WRITE) {
-                               internal_error(state, user->member, 
+                               internal_error(state, user->member,
                                        "bad variable access");
                        }
                        block = user->member->u.block;
                        if (!block) {
                                warning(state, user->member, "dead code");
+                               release_triple(state, user->member);
+                               continue;
                        }
                        if (work[block->vertex] >= iter) {
                                continue;
@@ -9867,34 +16169,132 @@ static void insert_phi_operations(struct compile_state *state)
                                /* Count how many edges flow into this block */
                                in_edges = front->users;
                                /* Insert a phi function for this variable */
+                               get_occurance(var->occurance);
                                phi = alloc_triple(
-                                       state, OP_PHI, var->type, -1, in_edges, 
-                                       front->first->filename, 
-                                       front->first->line,
-                                       front->first->col);
+                                       state, OP_PHI, var->type, -1, in_edges,
+                                       var->occurance);
                                phi->u.block = front;
                                MISC(phi, 0) = var;
                                use_triple(var, phi);
+#if 1
+                               if (phi->rhs != in_edges) {
+                                       internal_error(state, phi, "phi->rhs: %d != in_edges: %d",
+                                               phi->rhs, in_edges);
+                               }
+#endif
                                /* Insert the phi functions immediately after the label */
                                insert_triple(state, front->first->next, phi);
                                if (front->first == front->last) {
                                        front->last = front->first->next;
                                }
                                has_already[front->vertex] = iter;
+                               transform_to_arch_instruction(state, phi);
+
+                               /* If necessary plan to visit the basic block */
+                               if (work[front->vertex] >= iter) {
+                                       continue;
+                               }
+                               work[front->vertex] = iter;
+                               *work_list_tail = front;
+                               front->work_next = 0;
+                               work_list_tail = &front->work_next;
+                       }
+               }
+       }
+       xfree(has_already);
+       xfree(work);
+}
+
+
+struct stack {
+       struct triple_set *top;
+       unsigned orig_id;
+};
+
+static int count_auto_vars(struct compile_state *state)
+{
+       struct triple *first, *ins;
+       int auto_vars = 0;
+       first = state->first;
+       ins = first;
+       do {
+               if (triple_is_auto_var(state, ins)) {
+                       auto_vars += 1;
+               }
+               ins = ins->next;
+       } while(ins != first);
+       return auto_vars;
+}
+
+static void number_auto_vars(struct compile_state *state, struct stack *stacks)
+{
+       struct triple *first, *ins;
+       int auto_vars = 0;
+       first = state->first;
+       ins = first;
+       do {
+               if (triple_is_auto_var(state, ins)) {
+                       auto_vars += 1;
+                       stacks[auto_vars].orig_id = ins->id;
+                       ins->id = auto_vars;
+               }
+               ins = ins->next;
+       } while(ins != first);
+}
+
+static void restore_auto_vars(struct compile_state *state, struct stack *stacks)
+{
+       struct triple *first, *ins;
+       first = state->first;
+       ins = first;
+       do {
+               if (triple_is_auto_var(state, ins)) {
+                       ins->id = stacks[ins->id].orig_id;
+               }
+               ins = ins->next;
+       } while(ins != first);
+}
 
-                               /* If necessary plan to visit the basic block */
-                               if (work[front->vertex] >= iter) {
-                                       continue;
-                               }
-                               work[front->vertex] = iter;
-                               *work_list_tail = front;
-                               front->work_next = 0;
-                               work_list_tail = &front->work_next;
-                       }
+static struct triple *peek_triple(struct stack *stacks, struct triple *var)
+{
+       struct triple_set *head;
+       struct triple *top_val;
+       top_val = 0;
+       head = stacks[var->id].top;
+       if (head) {
+               top_val = head->member;
+       }
+       return top_val;
+}
+
+static void push_triple(struct stack *stacks, struct triple *var, struct triple *val)
+{
+       struct triple_set *new;
+       /* Append new to the head of the list,
+        * it's the only sensible behavoir for a stack.
+        */
+       new = xcmalloc(sizeof(*new), "triple_set");
+       new->member = val;
+       new->next   = stacks[var->id].top;
+       stacks[var->id].top = new;
+}
+
+static void pop_triple(struct stack *stacks, struct triple *var, struct triple *oldval)
+{
+       struct triple_set *set, **ptr;
+       ptr = &stacks[var->id].top;
+       while(*ptr) {
+               set = *ptr;
+               if (set->member == oldval) {
+                       *ptr = set->next;
+                       xfree(set);
+                       /* Only free one occurance from the stack */
+                       return;
+               }
+               else {
+                       ptr = &set->next;
                }
        }
-       xfree(has_already);
-       xfree(work);
 }
 
 /*
@@ -9902,7 +16302,7 @@ static void insert_phi_operations(struct compile_state *state)
  * S(V)
  */
 static void fixup_block_phi_variables(
-       struct compile_state *state, struct block *parent, struct block *block)
+       struct compile_state *state, struct stack *stacks, struct block *parent, struct block *block)
 {
        struct block_set *set;
        struct triple *ptr;
@@ -9927,11 +16327,11 @@ static void fixup_block_phi_variables(
                                internal_error(state, ptr, "no var???");
                        }
                        /* Find the current value of the variable */
-                       val = var->use->member;
-                       if ((val->op == OP_WRITE) || (val->op == OP_READ)) {
+                       val = peek_triple(stacks, var);
+                       if (val && ((val->op == OP_WRITE) || (val->op == OP_READ))) {
                                internal_error(state, val, "bad value in phi");
                        }
-                       if (edge >= TRIPLE_RHS(ptr->sizes)) {
+                       if (edge >= ptr->rhs) {
                                internal_error(state, ptr, "edges > phi rhs");
                        }
                        slot = &RHS(ptr, edge);
@@ -9949,9 +16349,9 @@ static void fixup_block_phi_variables(
 
 
 static void rename_block_variables(
-       struct compile_state *state, struct block *block)
+       struct compile_state *state, struct stack *stacks, struct block *block)
 {
-       struct block_set *user;
+       struct block_set *user, *edge;
        struct triple *ptr, *next, *last;
        int done;
        if (!block)
@@ -9967,12 +16367,28 @@ static void rename_block_variables(
                if (ptr->op == OP_READ) {
                        struct triple *var, *val;
                        var = RHS(ptr, 0);
+                       if (!triple_is_auto_var(state, var)) {
+                               internal_error(state, ptr, "read of non auto var!");
+                       }
                        unuse_triple(var, ptr);
-                       if (!var->use) {
+                       /* Find the current value of the variable */
+                       val = peek_triple(stacks, var);
+                       if (!val) {
+                               /* Let the optimizer at variables that are not initially
+                                * set.  But give it a bogus value so things seem to
+                                * work by accident.  This is useful for bitfields because
+                                * setting them always involves a read-modify-write.
+                                */
+                               if (TYPE_ARITHMETIC(ptr->type->type)) {
+                                       val = pre_triple(state, ptr, OP_INTCONST, ptr->type, 0, 0);
+                                       val->u.cval = 0xdeadbeaf;
+                               } else {
+                                       val = pre_triple(state, ptr, OP_UNKNOWNVAL, ptr->type, 0, 0);
+                               }
+                       }
+                       if (!val) {
                                error(state, ptr, "variable used without being set");
                        }
-                       /* Find the current value of the variable */
-                       val = var->use->member;
                        if ((val->op == OP_WRITE) || (val->op == OP_READ)) {
                                internal_error(state, val, "bad value in read");
                        }
@@ -9982,33 +16398,56 @@ static void rename_block_variables(
                }
                /* LHS(A) */
                if (ptr->op == OP_WRITE) {
-                       struct triple *var, *val;
-                       var = LHS(ptr, 0);
-                       val = RHS(ptr, 0);
-                       if ((val->op == OP_WRITE) || (val->op == OP_READ)) {
-                               internal_error(state, val, "bad value in write");
+                       struct triple *var, *val, *tval;
+                       var = MISC(ptr, 0);
+                       if (!triple_is_auto_var(state, var)) {
+                               internal_error(state, ptr, "write to non auto var!");
                        }
-                       propogate_use(state, ptr, val);
+                       tval = val = RHS(ptr, 0);
+                       if ((val->op == OP_WRITE) || (val->op == OP_READ) ||
+                               triple_is_auto_var(state, val)) {
+                               internal_error(state, ptr, "bad value in write");
+                       }
+                       /* Insert a cast if the types differ */
+                       if (!is_subset_type(ptr->type, val->type)) {
+                               if (val->op == OP_INTCONST) {
+                                       tval = pre_triple(state, ptr, OP_INTCONST, ptr->type, 0, 0);
+                                       tval->u.cval = val->u.cval;
+                               }
+                               else {
+                                       tval = pre_triple(state, ptr, OP_CONVERT, ptr->type, val, 0);
+                                       use_triple(val, tval);
+                               }
+                               transform_to_arch_instruction(state, tval);
+                               unuse_triple(val, ptr);
+                               RHS(ptr, 0) = tval;
+                               use_triple(tval, ptr);
+                       }
+                       propogate_use(state, ptr, tval);
                        unuse_triple(var, ptr);
                        /* Push OP_WRITE ptr->right onto a stack of variable uses */
-                       push_triple(var, val);
+                       push_triple(stacks, var, tval);
                }
                if (ptr->op == OP_PHI) {
                        struct triple *var;
                        var = MISC(ptr, 0);
+                       if (!triple_is_auto_var(state, var)) {
+                               internal_error(state, ptr, "phi references non auto var!");
+                       }
                        /* Push OP_PHI onto a stack of variable uses */
-                       push_triple(var, ptr);
+                       push_triple(stacks, var, ptr);
                }
                last = ptr;
        }
        block->last = last;
 
        /* Fixup PHI functions in the cf successors */
-       fixup_block_phi_variables(state, block, block->left);
-       fixup_block_phi_variables(state, block, block->right);
+       for(edge = block->edges; edge; edge = edge->next) {
+               fixup_block_phi_variables(state, stacks, block, edge->member);
+       }
        /* rename variables in the dominated nodes */
        for(user = block->idominates; user; user = user->next) {
-               rename_block_variables(state, user->member);
+               rename_block_variables(state, stacks, user->member);
        }
        /* pop the renamed variable stack */
        last = block->first;
@@ -10020,9 +16459,9 @@ static void rename_block_variables(
                }
                if (ptr->op == OP_WRITE) {
                        struct triple *var;
-                       var = LHS(ptr, 0);
+                       var = MISC(ptr, 0);
                        /* Pop OP_WRITE ptr->right from the stack of variable uses */
-                       pop_triple(var, RHS(ptr, 0));
+                       pop_triple(stacks, var, RHS(ptr, 0));
                        release_triple(state, ptr);
                        continue;
                }
@@ -10030,32 +16469,58 @@ static void rename_block_variables(
                        struct triple *var;
                        var = MISC(ptr, 0);
                        /* Pop OP_WRITE ptr->right from the stack of variable uses */
-                       pop_triple(var, ptr);
+                       pop_triple(stacks, var, ptr);
                }
                last = ptr;
        }
        block->last = last;
 }
 
+static void rename_variables(struct compile_state *state)
+{
+       struct stack *stacks;
+       int auto_vars;
+
+       /* Allocate stacks for the Variables */
+       auto_vars = count_auto_vars(state);
+       stacks = xcmalloc(sizeof(stacks[0])*(auto_vars + 1), "auto var stacks");
+
+       /* Give each auto_var a stack */
+       number_auto_vars(state, stacks);
+
+       /* Rename the variables */
+       rename_block_variables(state, stacks, state->bb.first_block);
+
+       /* Remove the stacks from the auto_vars */
+       restore_auto_vars(state, stacks);
+       xfree(stacks);
+}
+
 static void prune_block_variables(struct compile_state *state,
        struct block *block)
 {
        struct block_set *user;
-       struct triple *next, *last, *ptr;
+       struct triple *next, *ptr;
        int done;
-       last = block->first;
+
        done = 0;
        for(ptr = block->first; !done; ptr = next) {
+               /* Be extremely careful I am deleting the list
+                * as I walk trhough it.
+                */
                next = ptr->next;
                if (ptr == block->last) {
                        done = 1;
                }
-               if (ptr->op == OP_ADECL) {
+               if (triple_is_auto_var(state, ptr)) {
                        struct triple_set *user, *next;
                        for(user = ptr->use; user; user = next) {
                                struct triple *use;
                                next = user->next;
                                use = user->member;
+                               if (MISC(ptr, 0) == user->member) {
+                                       continue;
+                               }
                                if (use->op != OP_PHI) {
                                        internal_error(state, use, "decl still used");
                                }
@@ -10065,33 +16530,144 @@ static void prune_block_variables(struct compile_state *state,
                                unuse_triple(ptr, use);
                                MISC(use, 0) = 0;
                        }
-                       release_triple(state, ptr);
+                       if ((ptr->u.cval == 0) && (MISC(ptr, 0)->lhs == 1)) {
+                               /* Delete the adecl */
+                               release_triple(state, MISC(ptr, 0));
+                               /* And the piece */
+                               release_triple(state, ptr);
+                       }
                        continue;
                }
-               last = ptr;
        }
-       block->last = last;
        for(user = block->idominates; user; user = user->next) {
                prune_block_variables(state, user->member);
        }
 }
 
-static void transform_to_ssa_form(struct compile_state *state)
+struct phi_triple {
+       struct triple *phi;
+       unsigned orig_id;
+       int alive;
+};
+
+static void keep_phi(struct compile_state *state, struct phi_triple *live, struct triple *phi)
 {
-       insert_phi_operations(state);
+       struct triple **slot;
+       int zrhs, i;
+       if (live[phi->id].alive) {
+               return;
+       }
+       live[phi->id].alive = 1;
+       zrhs = phi->rhs;
+       slot = &RHS(phi, 0);
+       for(i = 0; i < zrhs; i++) {
+               struct triple *used;
+               used = slot[i];
+               if (used && (used->op == OP_PHI)) {
+                       keep_phi(state, live, used);
+               }
+       }
+}
+
+static void prune_unused_phis(struct compile_state *state)
+{
+       struct triple *first, *phi;
+       struct phi_triple *live;
+       int phis, i;
+
+       /* Find the first instruction */
+       first = state->first;
+
+       /* Count how many phi functions I need to process */
+       phis = 0;
+       for(phi = first->next; phi != first; phi = phi->next) {
+               if (phi->op == OP_PHI) {
+                       phis += 1;
+               }
+       }
+
+       /* Mark them all dead */
+       live = xcmalloc(sizeof(*live) * (phis + 1), "phi_triple");
+       phis = 0;
+       for(phi = first->next; phi != first; phi = phi->next) {
+               if (phi->op != OP_PHI) {
+                       continue;
+               }
+               live[phis].alive   = 0;
+               live[phis].orig_id = phi->id;
+               live[phis].phi     = phi;
+               phi->id = phis;
+               phis += 1;
+       }
+
+       /* Mark phis alive that are used by non phis */
+       for(i = 0; i < phis; i++) {
+               struct triple_set *set;
+               for(set = live[i].phi->use; !live[i].alive && set; set = set->next) {
+                       if (set->member->op != OP_PHI) {
+                               keep_phi(state, live, live[i].phi);
+                               break;
+                       }
+               }
+       }
+
+       /* Delete the extraneous phis */
+       for(i = 0; i < phis; i++) {
+               struct triple **slot;
+               int zrhs, j;
+               if (!live[i].alive) {
+                       release_triple(state, live[i].phi);
+                       continue;
+               }
+               phi = live[i].phi;
+               slot = &RHS(phi, 0);
+               zrhs = phi->rhs;
+               for(j = 0; j < zrhs; j++) {
+                       if(!slot[j]) {
+                               struct triple *unknown;
+                               get_occurance(phi->occurance);
+                               unknown = flatten(state, state->global_pool,
+                                       alloc_triple(state, OP_UNKNOWNVAL,
+                                               phi->type, 0, 0, phi->occurance));
+                               slot[j] = unknown;
+                               use_triple(unknown, phi);
+                               transform_to_arch_instruction(state, unknown);
 #if 0
-       printf("@%s:%d\n", __FILE__, __LINE__);
-       print_blocks(state, stdout);
+                               warning(state, phi, "variable not set at index %d on all paths to use", j);
 #endif
-       rename_block_variables(state, state->first_block);
-       prune_block_variables(state, state->first_block);
+                       }
+               }
+       }
+       xfree(live);
+}
+
+static void transform_to_ssa_form(struct compile_state *state)
+{
+       insert_phi_operations(state);
+       rename_variables(state);
+
+       prune_block_variables(state, state->bb.first_block);
+       prune_unused_phis(state);
+
+       print_blocks(state, __func__, state->dbgout);
 }
 
 
 static void clear_vertex(
        struct compile_state *state, struct block *block, void *arg)
 {
+       /* Clear the current blocks vertex and the vertex of all
+        * of the current blocks neighbors in case there are malformed
+        * blocks with now instructions at this point.
+        */
+       struct block_set *user, *edge;
        block->vertex = 0;
+       for(edge = block->edges; edge; edge = edge->next) {
+               edge->member->vertex = 0;
+       }
+       for(user = block->use; user; user = user->next) {
+               user->member->vertex = 0;
+       }
 }
 
 static void mark_live_block(
@@ -10105,8 +16681,8 @@ static void mark_live_block(
        *next_vertex += 1;
        if (triple_is_branch(state, block->last)) {
                struct triple **targ;
-               targ = triple_targ(state, block->last, 0);
-               for(; targ; targ = triple_targ(state, block->last, targ)) {
+               targ = triple_edge_targ(state, block->last, 0);
+               for(; targ; targ = triple_edge_targ(state, block->last, targ)) {
                        if (!*targ) {
                                continue;
                        }
@@ -10115,8 +16691,12 @@ static void mark_live_block(
                        }
                        mark_live_block(state, (*targ)->u.block, next_vertex);
                }
+               /* Ensure the last block of a function remains alive */
+               if (triple_is_call(state, block->last)) {
+                       mark_live_block(state, MISC(block->last, 0)->u.block, next_vertex);
+               }
        }
-       else if (block->last->next != RHS(state->main_function, 0)) {
+       else if (block->last->next != state->first) {
                struct triple *ins;
                ins = block->last->next;
                if (!triple_stores_block(state, ins)) {
@@ -10132,30 +16712,37 @@ static void transform_from_ssa_form(struct compile_state *state)
         * edges to blocks containting phi functions.
         */
        struct triple *first;
-       struct triple *phi, *next;
+       struct triple *phi, *var, *next;
        int next_vertex;
 
        /* Walk the control flow to see which blocks remain alive */
-       walk_blocks(state, clear_vertex, 0);
+       walk_blocks(state, &state->bb, clear_vertex, 0);
        next_vertex = 1;
-       mark_live_block(state, state->first_block, &next_vertex);
+       mark_live_block(state, state->bb.first_block, &next_vertex);
 
        /* Walk all of the operations to find the phi functions */
-       first = RHS(state->main_function, 0);
+       first = state->first;
        for(phi = first->next; phi != first ; phi = next) {
                struct block_set *set;
                struct block *block;
                struct triple **slot;
-               struct triple *var, *read;
+               struct triple *var;
                struct triple_set *use, *use_next;
-               int edge, used;
+               int edge, writers, readers;
                next = phi->next;
                if (phi->op != OP_PHI) {
                        continue;
                }
+
                block = phi->u.block;
                slot  = &RHS(phi, 0);
 
+               /* If this phi is in a dead block just forget it */
+               if (block->vertex == 0) {
+                       release_triple(state, phi);
+                       continue;
+               }
+
                /* Forget uses from code in dead blocks */
                for(use = phi->use; use; use = use_next) {
                        struct block *ublock;
@@ -10173,64 +16760,184 @@ static void transform_from_ssa_form(struct compile_state *state)
                        }
                        unuse_triple(phi, use->member);
                }
-
                /* A variable to replace the phi function */
-               var = post_triple(state, phi, OP_ADECL, phi->type, 0,0);
-               /* A read of the single value that is set into the variable */
-               read = post_triple(state, var, OP_READ, phi->type, var, 0);
-               use_triple(var, read);
+               if (registers_of(state, phi->type) != 1) {
+                       internal_error(state, phi, "phi->type does not fit in a single register!");
+               }
+               var = post_triple(state, phi, OP_ADECL, phi->type, 0, 0);
+               var = var->next; /* point at the var */
+
+               /* Replaces use of phi with var */
+               propogate_use(state, phi, var);
 
-               /* Replaces uses of the phi with variable reads */
-               propogate_use(state, phi, read);
+               /* Count the readers */
+               readers = 0;
+               for(use = var->use; use; use = use->next) {
+                       if (use->member != MISC(var, 0)) {
+                               readers++;
+                       }
+               }
 
                /* Walk all of the incoming edges/blocks and insert moves.
                 */
+               writers = 0;
                for(edge = 0, set = block->use; set; set = set->next, edge++) {
-                       struct block *eblock;
+                       struct block *eblock, *vblock;
                        struct triple *move;
-                       struct triple *val;
+                       struct triple *val, *base;
                        eblock = set->member;
                        val = slot[edge];
                        slot[edge] = 0;
                        unuse_triple(val, phi);
+                       vblock = block_of_triple(state, val);
 
-                       if (!val || (val == &zero_triple) ||
-                               (block->vertex == 0) || (eblock->vertex == 0) ||
-                               (val == phi) || (val == read)) {
+                       /* If we don't have a value that belongs in an OP_WRITE
+                        * continue on.
+                        */
+                       if (!val || (val == &unknown_triple) || (val == phi)
+                               || (vblock && (vblock->vertex == 0))) {
                                continue;
                        }
-                       
-                       move = post_triple(state, 
-                               val, OP_WRITE, phi->type, var, val);
-                       use_triple(val, move);
-                       use_triple(var, move);
-               }               
-               /* See if there are any writers of var */
-               used = 0;
-               for(use = var->use; use; use = use->next) {
-                       struct triple **expr;
-                       expr = triple_lhs(state, use->member, 0);
-                       for(; expr; expr = triple_lhs(state, use->member, expr)) {
-                               if (*expr == var) {
-                                       used = 1;
+                       /* If the value should never occur error */
+                       if (!vblock) {
+                               internal_error(state, val, "no vblock?");
+                               continue;
+                       }
+
+                       /* If the value occurs in a dead block see if a replacement
+                        * block can be found.
+                        */
+                       while(eblock && (eblock->vertex == 0)) {
+                               eblock = eblock->idom;
+                       }
+                       /* If not continue on with the next value. */
+                       if (!eblock || (eblock->vertex == 0)) {
+                               continue;
+                       }
+
+                       /* If we have an empty incoming block ignore it. */
+                       if (!eblock->first) {
+                               internal_error(state, 0, "empty block?");
+                       }
+
+                       /* Make certain the write is placed in the edge block... */
+                       /* Walk through the edge block backwards to find an
+                        * appropriate location for the OP_WRITE.
+                        */
+                       for(base = eblock->last; base != eblock->first; base = base->prev) {
+                               struct triple **expr;
+                               if (base->op == OP_PIECE) {
+                                       base = MISC(base, 0);
+                               }
+                               if ((base == var) || (base == val)) {
+                                       goto out;
+                               }
+                               expr = triple_lhs(state, base, 0);
+                               for(; expr; expr = triple_lhs(state, base, expr)) {
+                                       if ((*expr) == val) {
+                                               goto out;
+                                       }
                                }
+                               expr = triple_rhs(state, base, 0);
+                               for(; expr; expr = triple_rhs(state, base, expr)) {
+                                       if ((*expr) == var) {
+                                               goto out;
+                                       }
+                               }
+                       }
+               out:
+                       if (triple_is_branch(state, base)) {
+                               internal_error(state, base,
+                                       "Could not insert write to phi");
                        }
+                       move = post_triple(state, base, OP_WRITE, var->type, val, var);
+                       use_triple(val, move);
+                       use_triple(var, move);
+                       writers++;
+               }
+               if (!writers && readers) {
+                       internal_error(state, var, "no value written to in use phi?");
                }
                /* If var is not used free it */
-               if (!used) {
-                       unuse_triple(var, read);
-                       free_triple(state, read);
-                       free_triple(state, var);
+               if (!writers) {
+                       release_triple(state, MISC(var, 0));
+                       release_triple(state, var);
                }
-
                /* Release the phi function */
                release_triple(state, phi);
        }
-       
+
+       /* Walk all of the operations to find the adecls */
+       for(var = first->next; var != first ; var = var->next) {
+               struct triple_set *use, *use_next;
+               if (!triple_is_auto_var(state, var)) {
+                       continue;
+               }
+
+               /* Walk through all of the rhs uses of var and
+                * replace them with read of var.
+                */
+               for(use = var->use; use; use = use_next) {
+                       struct triple *read, *user;
+                       struct triple **slot;
+                       int zrhs, i, used;
+                       use_next = use->next;
+                       user = use->member;
+
+                       /* Generate a read of var */
+                       read = pre_triple(state, user, OP_READ, var->type, var, 0);
+                       use_triple(var, read);
+
+                       /* Find the rhs uses and see if they need to be replaced */
+                       used = 0;
+                       zrhs = user->rhs;
+                       slot = &RHS(user, 0);
+                       for(i = 0; i < zrhs; i++) {
+                               if (slot[i] == var) {
+                                       slot[i] = read;
+                                       used = 1;
+                               }
+                       }
+                       /* If we did use it cleanup the uses */
+                       if (used) {
+                               unuse_triple(var, user);
+                               use_triple(read, user);
+                       }
+                       /* If we didn't use it release the extra triple */
+                       else {
+                               release_triple(state, read);
+                       }
+               }
+       }
 }
 
+#define HI() if (state->compiler->debug & DEBUG_REBUILD_SSA_FORM) { \
+       FILE *fp = state->dbgout; \
+       fprintf(fp, "@ %s:%d\n", __FILE__, __LINE__); romcc_print_blocks(state, fp); \
+       }
+
+static void rebuild_ssa_form(struct compile_state *state)
+{
+HI();
+       transform_from_ssa_form(state);
+HI();
+       state->bb.first = state->first;
+       free_basic_blocks(state, &state->bb);
+       analyze_basic_blocks(state, &state->bb);
+HI();
+       insert_phi_operations(state);
+HI();
+       rename_variables(state);
+HI();
+
+       prune_block_variables(state, state->bb.first_block);
+HI();
+       prune_unused_phis(state);
+HI();
+}
+#undef HI
 
-/* 
+/*
  * Register conflict resolution
  * =========================================================
  */
@@ -10260,7 +16967,7 @@ static struct reg_info find_def_color(
                if (tinfo.reg >= MAX_REGISTERS) {
                        tinfo.reg = REG_UNSET;
                }
-               if ((tinfo.reg != REG_UNSET) && 
+               if ((tinfo.reg != REG_UNSET) &&
                        (info.reg != REG_UNSET) &&
                        (tinfo.reg != info.reg)) {
                        internal_error(state, def, "register conflict");
@@ -10285,8 +16992,8 @@ static struct reg_info find_lhs_pre_color(
 {
        struct reg_info info;
        int zlhs, zrhs, i;
-       zrhs = TRIPLE_RHS(ins->sizes);
-       zlhs = TRIPLE_LHS(ins->sizes);
+       zrhs = ins->rhs;
+       zlhs = ins->lhs;
        if (!zlhs && triple_is_def(state, ins)) {
                zlhs = 1;
        }
@@ -10321,14 +17028,14 @@ static struct reg_info find_lhs_post_color(
        struct triple_set *set;
        struct reg_info info;
        struct triple *lhs;
-#if 0
-       fprintf(stderr, "find_lhs_post_color(%p, %d)\n",
+#if DEBUG_TRIPLE_COLOR
+       fprintf(state->errout, "find_lhs_post_color(%p, %d)\n",
                ins, index);
 #endif
        if ((index == 0) && triple_is_def(state, ins)) {
                lhs = ins;
        }
-       else if (index < TRIPLE_LHS(ins->sizes)) {
+       else if (index < ins->lhs) {
                lhs = LHS(ins, index);
        }
        else {
@@ -10344,7 +17051,7 @@ static struct reg_info find_lhs_post_color(
                struct triple *user;
                int zrhs, i;
                user = set->member;
-               zrhs = TRIPLE_RHS(user->sizes);
+               zrhs = user->rhs;
                for(i = 0; i < zrhs; i++) {
                        if (RHS(user, i) != lhs) {
                                continue;
@@ -10365,8 +17072,8 @@ static struct reg_info find_lhs_post_color(
                        info.regcm &= rinfo.regcm;
                }
        }
-#if 0
-       fprintf(stderr, "find_lhs_post_color(%p, %d) -> ( %d, %x)\n",
+#if DEBUG_TRIPLE_COLOR
+       fprintf(state->errout, "find_lhs_post_color(%p, %d) -> ( %d, %x)\n",
                ins, index, info.reg, info.regcm);
 #endif
        return info;
@@ -10377,12 +17084,12 @@ static struct reg_info find_rhs_post_color(
 {
        struct reg_info info, rinfo;
        int zlhs, i;
-#if 0
-       fprintf(stderr, "find_rhs_post_color(%p, %d)\n",
+#if DEBUG_TRIPLE_COLOR
+       fprintf(state->errout, "find_rhs_post_color(%p, %d)\n",
                ins, index);
 #endif
        rinfo = arch_reg_rhs(state, ins, index);
-       zlhs = TRIPLE_LHS(ins->sizes);
+       zlhs = ins->lhs;
        if (!zlhs && triple_is_def(state, ins)) {
                zlhs = 1;
        }
@@ -10400,7 +17107,7 @@ static struct reg_info find_rhs_post_color(
                        if (tinfo.reg >= MAX_REGISTERS) {
                                tinfo.reg = REG_UNSET;
                        }
-                       info.regcm &= linfo.reg;
+                       info.regcm &= linfo.regcm;
                        info.regcm &= tinfo.regcm;
                        if (info.reg != REG_UNSET) {
                                internal_error(state, ins, "register conflict");
@@ -10411,8 +17118,8 @@ static struct reg_info find_rhs_post_color(
                        info.reg = tinfo.reg;
                }
        }
-#if 0
-       fprintf(stderr, "find_rhs_post_color(%p, %d) -> ( %d, %x)\n",
+#if DEBUG_TRIPLE_COLOR
+       fprintf(state->errout, "find_rhs_post_color(%p, %d) -> ( %d, %x)\n",
                ins, index, info.reg, info.regcm);
 #endif
        return info;
@@ -10422,8 +17129,8 @@ static struct reg_info find_lhs_color(
        struct compile_state *state, struct triple *ins, int index)
 {
        struct reg_info pre, post, info;
-#if 0
-       fprintf(stderr, "find_lhs_color(%p, %d)\n",
+#if DEBUG_TRIPLE_COLOR
+       fprintf(state->errout, "find_lhs_color(%p, %d)\n",
                ins, index);
 #endif
        pre = find_lhs_pre_color(state, ins, index);
@@ -10438,9 +17145,10 @@ static struct reg_info find_lhs_color(
        if (info.reg == REG_UNSET) {
                info.reg = post.reg;
        }
-#if 0
-       fprintf(stderr, "find_lhs_color(%p, %d) -> ( %d, %x)\n",
-               ins, index, info.reg, info.regcm);
+#if DEBUG_TRIPLE_COLOR
+       fprintf(state->errout, "find_lhs_color(%p, %d) -> ( %d, %x) ... (%d, %x) (%d, %x)\n",
+               ins, index, info.reg, info.regcm,
+               pre.reg, pre.regcm, post.reg, post.regcm);
 #endif
        return info;
 }
@@ -10475,22 +17183,49 @@ static struct triple *post_copy(struct compile_state *state, struct triple *ins)
        return out;
 }
 
-static struct triple *pre_copy(
-       struct compile_state *state, struct triple *ins, int index)
+static struct triple *typed_pre_copy(
+       struct compile_state *state, struct type *type, struct triple *ins, int index)
 {
        /* Carefully insert enough operations so that I can
         * enter any operation with a GPR32.
         */
        struct triple *in;
        struct triple **expr;
+       unsigned classes;
+       struct reg_info info;
+       int op;
+       if (ins->op == OP_PHI) {
+               internal_error(state, ins, "pre_copy on a phi?");
+       }
+       classes = arch_type_to_regcm(state, type);
+       info = arch_reg_rhs(state, ins, index);
        expr = &RHS(ins, index);
-       in = pre_triple(state, ins, OP_COPY, (*expr)->type, *expr, 0);
+       if ((info.regcm & classes) == 0) {
+               FILE *fp = state->errout;
+               fprintf(fp, "src_type: ");
+               name_of(fp, ins->type);
+               fprintf(fp, "\ndst_type: ");
+               name_of(fp, type);
+               fprintf(fp, "\n");
+               internal_error(state, ins, "pre_copy with no register classes");
+       }
+       op = OP_COPY;
+       if (!equiv_types(type, (*expr)->type)) {
+               op = OP_CONVERT;
+       }
+       in = pre_triple(state, ins, op, type, *expr, 0);
        unuse_triple(*expr, ins);
        *expr = in;
        use_triple(RHS(in, 0), in);
        use_triple(in, ins);
        transform_to_arch_instruction(state, in);
        return in;
+
+}
+static struct triple *pre_copy(
+       struct compile_state *state, struct triple *ins, int index)
+{
+       return typed_pre_copy(state, RHS(ins, index)->type, ins, index);
 }
 
 
@@ -10503,11 +17238,11 @@ static void insert_copies_to_phi(struct compile_state *state)
        struct triple *phi;
 
        /* Walk all of the operations to find the phi functions */
-       first = RHS(state->main_function, 0);
+       first = state->first;
        for(phi = first->next; phi != first ; phi = phi->next) {
                struct block_set *set;
                struct block *block;
-               struct triple **slot;
+               struct triple **slot, *copy;
                int edge;
                if (phi->op != OP_PHI) {
                        continue;
@@ -10515,6 +17250,13 @@ static void insert_copies_to_phi(struct compile_state *state)
                phi->id |= TRIPLE_FLAG_POST_SPLIT;
                block = phi->u.block;
                slot  = &RHS(phi, 0);
+               /* Phi's that feed into mandatory live range joins
+                * cause nasty complications.  Insert a copy of
+                * the phi value so I never have to deal with
+                * that in the rest of the code.
+                */
+               copy = post_copy(state, phi);
+               copy->id |= TRIPLE_FLAG_PRE_SPLIT;
                /* Walk all of the incoming edges/blocks and insert moves.
                 */
                for(edge = 0, set = block->use; set; set = set->next, edge++) {
@@ -10529,24 +17271,43 @@ static void insert_copies_to_phi(struct compile_state *state)
                                continue;
                        }
 
-                       move = build_triple(state, OP_COPY, phi->type, val, 0,
-                               val->filename, val->line, val->col);
+                       get_occurance(val->occurance);
+                       move = build_triple(state, OP_COPY, val->type, val, 0,
+                               val->occurance);
                        move->u.block = eblock;
                        move->id |= TRIPLE_FLAG_PRE_SPLIT;
                        use_triple(val, move);
-                       
+
                        slot[edge] = move;
                        unuse_triple(val, phi);
                        use_triple(move, phi);
 
+                       /* Walk up the dominator tree until I have found the appropriate block */
+                       while(eblock && !tdominates(state, val, eblock->last)) {
+                               eblock = eblock->idom;
+                       }
+                       if (!eblock) {
+                               internal_error(state, phi, "Cannot find block dominated by %p",
+                                       val);
+                       }
+
                        /* Walk through the block backwards to find
                         * an appropriate location for the OP_COPY.
                         */
                        for(ptr = eblock->last; ptr != eblock->first; ptr = ptr->prev) {
                                struct triple **expr;
+                               if (ptr->op == OP_PIECE) {
+                                       ptr = MISC(ptr, 0);
+                               }
                                if ((ptr == phi) || (ptr == val)) {
                                        goto out;
                                }
+                               expr = triple_lhs(state, ptr, 0);
+                               for(;expr; expr = triple_lhs(state, ptr, expr)) {
+                                       if ((*expr) == val) {
+                                               goto out;
+                                       }
+                               }
                                expr = triple_rhs(state, ptr, 0);
                                for(;expr; expr = triple_rhs(state, ptr, expr)) {
                                        if ((*expr) == phi) {
@@ -10559,29 +17320,21 @@ static void insert_copies_to_phi(struct compile_state *state)
                                internal_error(state, ptr,
                                        "Could not insert write to phi");
                        }
-                       insert_triple(state, ptr->next, move);
-                       if (eblock->last == ptr) {
+                       insert_triple(state, after_lhs(state, ptr), move);
+                       if (eblock->last == after_lhs(state, ptr)->prev) {
                                eblock->last = move;
                        }
                        transform_to_arch_instruction(state, move);
                }
        }
+       print_blocks(state, __func__, state->dbgout);
 }
 
-struct triple_reg_set {
-       struct triple_reg_set *next;
-       struct triple *member;
-       struct triple *new;
-};
+struct triple_reg_set;
+struct reg_block;
 
-struct reg_block {
-       struct block *block;
-       struct triple_reg_set *in;
-       struct triple_reg_set *out;
-       int vertex;
-};
 
-static int do_triple_set(struct triple_reg_set **head, 
+static int do_triple_set(struct triple_reg_set **head,
        struct triple *member, struct triple *new_member)
 {
        struct triple_reg_set **ptr, *new;
@@ -10623,36 +17376,92 @@ static int in_triple(struct reg_block *rb, struct triple *in)
 {
        return do_triple_set(&rb->in, in, 0);
 }
+
+#if DEBUG_ROMCC_WARNING
 static void unin_triple(struct reg_block *rb, struct triple *unin)
 {
        do_triple_unset(&rb->in, unin);
 }
+#endif
 
 static int out_triple(struct reg_block *rb, struct triple *out)
 {
        return do_triple_set(&rb->out, out, 0);
 }
+#if DEBUG_ROMCC_WARNING
 static void unout_triple(struct reg_block *rb, struct triple *unout)
 {
        do_triple_unset(&rb->out, unout);
 }
+#endif
+
+static int initialize_regblock(struct reg_block *blocks,
+       struct block *block, int vertex)
+{
+       struct block_set *user;
+       if (!block || (blocks[block->vertex].block == block)) {
+               return vertex;
+       }
+       vertex += 1;
+       /* Renumber the blocks in a convinient fashion */
+       block->vertex = vertex;
+       blocks[vertex].block    = block;
+       blocks[vertex].vertex   = vertex;
+       for(user = block->use; user; user = user->next) {
+               vertex = initialize_regblock(blocks, user->member, vertex);
+       }
+       return vertex;
+}
+
+static struct triple *part_to_piece(struct compile_state *state, struct triple *ins)
+{
+/* Part to piece is a best attempt and it cannot be correct all by
+ * itself.  If various values are read as different sizes in different
+ * parts of the code this function cannot work.  Or rather it cannot
+ * work in conjunction with compute_variable_liftimes.  As the
+ * analysis will get confused.
+ */
+       struct triple *base;
+       unsigned reg;
+       if (!is_lvalue(state, ins)) {
+               return ins;
+       }
+       base = 0;
+       reg = 0;
+       while(ins && triple_is_part(state, ins) && (ins->op != OP_PIECE)) {
+               base = MISC(ins, 0);
+               switch(ins->op) {
+               case OP_INDEX:
+                       reg += index_reg_offset(state, base->type, ins->u.cval)/REG_SIZEOF_REG;
+                       break;
+               case OP_DOT:
+                       reg += field_reg_offset(state, base->type, ins->u.field)/REG_SIZEOF_REG;
+                       break;
+               default:
+                       internal_error(state, ins, "unhandled part");
+                       break;
+               }
+               ins = base;
+       }
+       if (base) {
+               if (reg > base->lhs) {
+                       internal_error(state, base, "part out of range?");
+               }
+               ins = LHS(base, reg);
+       }
+       return ins;
+}
 
-static int initialize_regblock(struct reg_block *blocks,
-       struct block *block, int vertex)
+static int this_def(struct compile_state *state,
+       struct triple *ins, struct triple *other)
 {
-       struct block_set *user;
-       if (!block || (blocks[block->vertex].block == block)) {
-               return vertex;
+       if (ins == other) {
+               return 1;
        }
-       vertex += 1;
-       /* Renumber the blocks in a convinient fashion */
-       block->vertex = vertex;
-       blocks[vertex].block    = block;
-       blocks[vertex].vertex   = vertex;
-       for(user = block->use; user; user = user->next) {
-               vertex = initialize_regblock(blocks, user->member, vertex);
+       if (ins->op == OP_WRITE) {
+               ins = part_to_piece(state, MISC(ins, 0));
        }
-       return vertex;
+       return ins == other;
 }
 
 static int phi_in(struct compile_state *state, struct reg_block *blocks,
@@ -10694,7 +17503,7 @@ static int phi_in(struct compile_state *state, struct reg_block *blocks,
                 */
                ptr2 = rb->block->first;
                for(done2 = 0; !done2; ptr2 = ptr2->next) {
-                       if (ptr2 == expr) {
+                       if (this_def(state, ptr2, expr)) {
                                break;
                        }
                        done2 = (ptr2 == rb->block->last);
@@ -10731,7 +17540,7 @@ static int reg_in(struct compile_state *state, struct reg_block *blocks,
                last = rb->block->last;
                done = 0;
                for(ptr = first; !done; ptr = ptr->next) {
-                       if (ptr == in_set->member) {
+                       if (this_def(state, ptr, in_set->member)) {
                                break;
                        }
                        done = (ptr == last);
@@ -10745,13 +17554,14 @@ static int reg_in(struct compile_state *state, struct reg_block *blocks,
        return change;
 }
 
-
 static int use_in(struct compile_state *state, struct reg_block *rb)
 {
        /* Find the variables we use but don't define and add
         * it to the current blocks input set.
         */
+#if DEBUG_ROMCC_WARNINGS
 #warning "FIXME is this O(N^2) algorithm bad?"
+#endif
        struct block *block;
        struct triple *ptr;
        int done;
@@ -10772,14 +17582,17 @@ static int use_in(struct compile_state *state, struct reg_block *rb)
                for(;expr; expr = triple_rhs(state, ptr, expr)) {
                        struct triple *rhs, *test;
                        int tdone;
-                       rhs = *expr;
+                       rhs = part_to_piece(state, *expr);
                        if (!rhs) {
                                continue;
                        }
-                       /* See if rhs is defined in this block */
+
+                       /* See if rhs is defined in this block.
+                        * A write counts as a definition.
+                        */
                        for(tdone = 0, test = ptr; !tdone; test = test->prev) {
                                tdone = (test == block->first);
-                               if (test == rhs) {
+                               if (this_def(state, test, rhs)) {
                                        rhs = 0;
                                        break;
                                }
@@ -10792,27 +17605,23 @@ static int use_in(struct compile_state *state, struct reg_block *rb)
 }
 
 static struct reg_block *compute_variable_lifetimes(
-       struct compile_state *state)
+       struct compile_state *state, struct basic_blocks *bb)
 {
        struct reg_block *blocks;
        int change;
        blocks = xcmalloc(
-               sizeof(*blocks)*(state->last_vertex + 1), "reg_block");
-       initialize_regblock(blocks, state->last_block, 0);
+               sizeof(*blocks)*(bb->last_vertex + 1), "reg_block");
+       initialize_regblock(blocks, bb->last_block, 0);
        do {
                int i;
                change = 0;
-               for(i = 1; i <= state->last_vertex; i++) {
+               for(i = 1; i <= bb->last_vertex; i++) {
+                       struct block_set *edge;
                        struct reg_block *rb;
                        rb = &blocks[i];
-                       /* Add the left successor's input set to in */
-                       if (rb->block->left) {
-                               change |= reg_in(state, blocks, rb, rb->block->left);
-                       }
-                       /* Add the right successor's input set to in */
-                       if ((rb->block->right) && 
-                               (rb->block->right != rb->block->left)) {
-                               change |= reg_in(state, blocks, rb, rb->block->right);
+                       /* Add the all successor's input set to in */
+                       for(edge = rb->block->edges; edge; edge = edge->next) {
+                               change |= reg_in(state, blocks, rb, edge->member);
                        }
                        /* Add use to in... */
                        change |= use_in(state, rb);
@@ -10821,12 +17630,12 @@ static struct reg_block *compute_variable_lifetimes(
        return blocks;
 }
 
-static void free_variable_lifetimes(
-       struct compile_state *state, struct reg_block *blocks)
+static void free_variable_lifetimes(struct compile_state *state,
+       struct basic_blocks *bb, struct reg_block *blocks)
 {
        int i;
        /* free in_set && out_set on each block */
-       for(i = 1; i <= state->last_vertex; i++) {
+       for(i = 1; i <= bb->last_vertex; i++) {
                struct triple_reg_set *entry, *next;
                struct reg_block *rb;
                rb = &blocks[i];
@@ -10843,17 +17652,18 @@ static void free_variable_lifetimes(
 
 }
 
-typedef struct triple *(*wvl_cb_t)(
-       struct compile_state *state, 
-       struct reg_block *blocks, struct triple_reg_set *live, 
+typedef void (*wvl_cb_t)(
+       struct compile_state *state,
+       struct reg_block *blocks, struct triple_reg_set *live,
        struct reg_block *rb, struct triple *ins, void *arg);
 
 static void walk_variable_lifetimes(struct compile_state *state,
-       struct reg_block *blocks, wvl_cb_t cb, void *arg)
+       struct basic_blocks *bb, struct reg_block *blocks,
+       wvl_cb_t cb, void *arg)
 {
        int i;
-       
-       for(i = 1; i <= state->last_vertex; i++) {
+
+       for(i = 1; i <= state->bb.last_vertex; i++) {
                struct triple_reg_set *live;
                struct triple_reg_set *entry, *next;
                struct triple *ptr, *prev;
@@ -10873,7 +17683,7 @@ static void walk_variable_lifetimes(struct compile_state *state,
                }
                /* Walk through the basic block calculating live */
                for(done = 0, ptr = block->last; !done; ptr = prev) {
-                       struct triple **expr, *result;
+                       struct triple **expr;
 
                        prev = ptr->prev;
                        done = (ptr == block->first);
@@ -10886,20 +17696,11 @@ static void walk_variable_lifetimes(struct compile_state *state,
                        /* Inform the callback function of what is
                         * going on.
                         */
-                       result = cb(state, blocks, live, rb, ptr, arg);
-                       
+                        cb(state, blocks, live, rb, ptr, arg);
+
                        /* Remove the current definition from live */
                        do_triple_unset(&live, ptr);
 
-                       /* If the current instruction was deleted continue */
-                       if (!result) {
-                               if (block->last == ptr) {
-                                       block->last = prev;
-                               }
-                               continue;
-                       }
-                       
-
                        /* Add the current uses to live.
                         *
                         * It is safe to skip phi functions because they do
@@ -10927,11 +17728,99 @@ static void walk_variable_lifetimes(struct compile_state *state,
        }
 }
 
+struct print_live_variable_info {
+       struct reg_block *rb;
+       FILE *fp;
+};
+#if DEBUG_EXPLICIT_CLOSURES
+static void print_live_variables_block(
+       struct compile_state *state, struct block *block, void *arg)
+
+{
+       struct print_live_variable_info *info = arg;
+       struct block_set *edge;
+       FILE *fp = info->fp;
+       struct reg_block *rb;
+       struct triple *ptr;
+       int phi_present;
+       int done;
+       rb = &info->rb[block->vertex];
+
+       fprintf(fp, "\nblock: %p (%d),",
+               block,  block->vertex);
+       for(edge = block->edges; edge; edge = edge->next) {
+               fprintf(fp, " %p<-%p",
+                       edge->member,
+                       edge->member && edge->member->use?edge->member->use->member : 0);
+       }
+       fprintf(fp, "\n");
+       if (rb->in) {
+               struct triple_reg_set *in_set;
+               fprintf(fp, "        in:");
+               for(in_set = rb->in; in_set; in_set = in_set->next) {
+                       fprintf(fp, " %-10p", in_set->member);
+               }
+               fprintf(fp, "\n");
+       }
+       phi_present = 0;
+       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
+               done = (ptr == block->last);
+               if (ptr->op == OP_PHI) {
+                       phi_present = 1;
+                       break;
+               }
+       }
+       if (phi_present) {
+               int edge;
+               for(edge = 0; edge < block->users; edge++) {
+                       fprintf(fp, "     in(%d):", edge);
+                       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
+                               struct triple **slot;
+                               done = (ptr == block->last);
+                               if (ptr->op != OP_PHI) {
+                                       continue;
+                               }
+                               slot = &RHS(ptr, 0);
+                               fprintf(fp, " %-10p", slot[edge]);
+                       }
+                       fprintf(fp, "\n");
+               }
+       }
+       if (block->first->op == OP_LABEL) {
+               fprintf(fp, "%p:\n", block->first);
+       }
+       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
+               done = (ptr == block->last);
+               display_triple(fp, ptr);
+       }
+       if (rb->out) {
+               struct triple_reg_set *out_set;
+               fprintf(fp, "       out:");
+               for(out_set = rb->out; out_set; out_set = out_set->next) {
+                       fprintf(fp, " %-10p", out_set->member);
+               }
+               fprintf(fp, "\n");
+       }
+       fprintf(fp, "\n");
+}
+
+static void print_live_variables(struct compile_state *state,
+       struct basic_blocks *bb, struct reg_block *rb, FILE *fp)
+{
+       struct print_live_variable_info info;
+       info.rb = rb;
+       info.fp = fp;
+       fprintf(fp, "\nlive variables by block\n");
+       walk_blocks(state, bb, print_live_variables_block, &info);
+
+}
+#endif
+
 static int count_triples(struct compile_state *state)
 {
        struct triple *first, *ins;
        int triples = 0;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                triples++;
@@ -10939,15 +17828,47 @@ static int count_triples(struct compile_state *state)
        } while (ins != first);
        return triples;
 }
+
+
 struct dead_triple {
        struct triple *triple;
        struct dead_triple *work_next;
        struct block *block;
-       int color;
+       int old_id;
        int flags;
 #define TRIPLE_FLAG_ALIVE 1
+#define TRIPLE_FLAG_FREE  1
 };
 
+static void print_dead_triples(struct compile_state *state,
+       struct dead_triple *dtriple)
+{
+       struct triple *first, *ins;
+       struct dead_triple *dt;
+       FILE *fp;
+       if (!(state->compiler->debug & DEBUG_TRIPLES)) {
+               return;
+       }
+       fp = state->dbgout;
+       fprintf(fp, "--------------- dtriples ---------------\n");
+       first = state->first;
+       ins = first;
+       do {
+               dt = &dtriple[ins->id];
+               if ((ins->op == OP_LABEL) && (ins->use)) {
+                       fprintf(fp, "\n%p:\n", ins);
+               }
+               fprintf(fp, "%c",
+                       (dt->flags & TRIPLE_FLAG_ALIVE)?' ': '-');
+               display_triple(fp, ins);
+               if (triple_is_branch(state, ins)) {
+                       fprintf(fp, "\n");
+               }
+               ins = ins->next;
+       } while(ins != first);
+       fprintf(fp, "\n");
+}
+
 
 static void awaken(
        struct compile_state *state,
@@ -10968,7 +17889,7 @@ static void awaken(
                        triple->id);
        }
        if (triple->op == OP_NOOP) {
-               internal_warning(state, triple, "awakening noop?");
+               internal_error(state, triple, "awakening noop?");
                return;
        }
        dt = &dtriple[triple->id];
@@ -10983,44 +17904,43 @@ static void awaken(
 
 static void eliminate_inefectual_code(struct compile_state *state)
 {
-       struct block *block;
        struct dead_triple *dtriple, *work_list, **work_list_tail, *dt;
        int triples, i;
        struct triple *first, *ins;
 
+       if (!(state->compiler->flags & COMPILER_ELIMINATE_INEFECTUAL_CODE)) {
+               return;
+       }
+
        /* Setup the work list */
        work_list = 0;
        work_list_tail = &work_list;
 
-       first = RHS(state->main_function, 0);
+       first = state->first;
 
        /* Count how many triples I have */
        triples = count_triples(state);
 
        /* Now put then in an array and mark all of the triples dead */
        dtriple = xcmalloc(sizeof(*dtriple) * (triples + 1), "dtriples");
-       
+
        ins = first;
        i = 1;
-       block = 0;
        do {
-               if (ins->op == OP_LABEL) {
-                       block = ins->u.block;
-               }
                dtriple[i].triple = ins;
-               dtriple[i].block  = block;
+               dtriple[i].block  = block_of_triple(state, ins);
                dtriple[i].flags  = 0;
-               dtriple[i].color  = ins->id;
+               dtriple[i].old_id = ins->id;
                ins->id = i;
                /* See if it is an operation we always keep */
-#warning "FIXME handle the case of killing a branch instruction"
-               if (!triple_is_pure(state, ins) || triple_is_branch(state, ins)) {
+               if (!triple_is_pure(state, ins, dtriple[i].old_id)) {
                        awaken(state, dtriple, &ins, &work_list_tail);
                }
                i++;
                ins = ins->next;
        } while(ins != first);
        while(work_list) {
+               struct block *block;
                struct dead_triple *dt;
                struct block_set *user;
                struct triple **expr;
@@ -11029,6 +17949,15 @@ static void eliminate_inefectual_code(struct compile_state *state)
                if (!work_list) {
                        work_list_tail = &work_list;
                }
+               /* Make certain the block the current instruction is in lives */
+               block = block_of_triple(state, dt->triple);
+               awaken(state, dtriple, &block->first, &work_list_tail);
+               if (triple_is_branch(state, block->last)) {
+                       awaken(state, dtriple, &block->last, &work_list_tail);
+               } else {
+                       awaken(state, dtriple, &block->last->next, &work_list_tail);
+               }
+
                /* Wake up the data depencencies of this triple */
                expr = 0;
                do {
@@ -11050,27 +17979,34 @@ static void eliminate_inefectual_code(struct compile_state *state)
                } while(expr);
                /* Wake up the reverse control dependencies of this triple */
                for(user = dt->block->ipdomfrontier; user; user = user->next) {
-                       awaken(state, dtriple, &user->member->last, &work_list_tail);
+                       struct triple *last;
+                       last = user->member->last;
+                       while((last->op == OP_NOOP) && (last != user->member->first)) {
+#if DEBUG_ROMCC_WARNINGS
+#warning "Should we bring the awakening noops back?"
+#endif
+                               // internal_warning(state, last, "awakening noop?");
+                               last = last->prev;
+                       }
+                       awaken(state, dtriple, &last, &work_list_tail);
                }
        }
+       print_dead_triples(state, dtriple);
        for(dt = &dtriple[1]; dt <= &dtriple[triples]; dt++) {
-               if ((dt->triple->op == OP_NOOP) && 
+               if ((dt->triple->op == OP_NOOP) &&
                        (dt->flags & TRIPLE_FLAG_ALIVE)) {
                        internal_error(state, dt->triple, "noop effective?");
                }
-               dt->triple->id = dt->color;     /* Restore the color */
+               dt->triple->id = dt->old_id;    /* Restore the color */
                if (!(dt->flags & TRIPLE_FLAG_ALIVE)) {
-#warning "FIXME handle the case of killing a basic block"
-                       if (dt->block->first == dt->triple) {
-                               continue;
-                       }
-                       if (dt->block->last == dt->triple) {
-                               dt->block->last = dt->triple->prev;
-                       }
                        release_triple(state, dt->triple);
                }
        }
        xfree(dtriple);
+
+       rebuild_ssa_form(state);
+
+       print_blocks(state, __func__, state->dbgout);
 }
 
 
@@ -11085,7 +18021,7 @@ static void insert_mandatory_copies(struct compile_state *state)
         * are inserting copies before instructions but that
         * case should be rare.
         */
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                struct triple_set *entry, *next;
@@ -11098,11 +18034,11 @@ static void insert_mandatory_copies(struct compile_state *state)
                        goto next;
                }
                /* Find the architecture specific color information */
-               info = arch_reg_lhs(state, ins, 0);
+               info = find_lhs_pre_color(state, ins, 0);
                if (info.reg >= MAX_REGISTERS) {
                        info.reg = REG_UNSET;
                }
-               
+
                reg = REG_UNSET;
                regcm = arch_type_to_regcm(state, ins->type);
                do_post_copy = do_pre_copy = 0;
@@ -11116,13 +18052,13 @@ static void insert_mandatory_copies(struct compile_state *state)
                        if (i < 0) {
                                continue;
                        }
-                       
+
                        /* Find the users color requirements */
                        rinfo = arch_reg_rhs(state, entry->member, i);
                        if (rinfo.reg >= MAX_REGISTERS) {
                                rinfo.reg = REG_UNSET;
                        }
-                       
+
                        /* See if I need a pre_copy */
                        if (rinfo.reg != REG_UNSET) {
                                if ((reg != REG_UNSET) && (reg != rinfo.reg)) {
@@ -11135,17 +18071,25 @@ static void insert_mandatory_copies(struct compile_state *state)
                        if (regcm == 0) {
                                do_pre_copy = 1;
                        }
+                       /* Always use pre_copies for constants.
+                        * They do not take up any registers until a
+                        * copy places them in one.
+                        */
+                       if ((info.reg == REG_UNNEEDED) &&
+                               (rinfo.reg != REG_UNNEEDED)) {
+                               do_pre_copy = 1;
+                       }
                }
                do_post_copy =
                        !do_pre_copy &&
-                       (((info.reg != REG_UNSET) && 
+                       (((info.reg != REG_UNSET) &&
                                (reg != REG_UNSET) &&
                                (info.reg != reg)) ||
                        ((info.regcm & regcm) == 0));
 
                reg = info.reg;
                regcm = info.regcm;
-               /* Walk through the uses of insert and do a pre_copy or see if a post_copy is warranted */
+               /* Walk through the uses of ins and do a pre_copy or see if a post_copy is warranted */
                for(entry = ins->use; entry; entry = next) {
                        struct reg_info rinfo;
                        int i;
@@ -11154,7 +18098,7 @@ static void insert_mandatory_copies(struct compile_state *state)
                        if (i < 0) {
                                continue;
                        }
-                       
+
                        /* Find the users color requirements */
                        rinfo = arch_reg_rhs(state, entry->member, i);
                        if (rinfo.reg >= MAX_REGISTERS) {
@@ -11180,6 +18124,7 @@ static void insert_mandatory_copies(struct compile_state *state)
                                                        internal_error(state, user, "bad rhs");
                                                }
                                                tmp = pre_copy(state, user, i);
+                                               tmp->id |= TRIPLE_FLAG_PRE_SPLIT;
                                                continue;
                                        } else {
                                                do_post_copy = 1;
@@ -11195,17 +18140,19 @@ static void insert_mandatory_copies(struct compile_state *state)
                                                internal_error(state, user, "bad rhs");
                                        }
                                        tmp = pre_copy(state, user, i);
+                                       tmp->id |= TRIPLE_FLAG_PRE_SPLIT;
                                        continue;
                                } else {
                                        do_post_copy = 1;
                                }
                        }
                        regcm &= rinfo.regcm;
-                       
+
                }
                if (do_post_copy) {
                        struct reg_info pre, post;
                        tmp = post_copy(state, ins);
+                       tmp->id |= TRIPLE_FLAG_PRE_SPLIT;
                        pre = arch_reg_lhs(state, ins, 0);
                        post = arch_reg_lhs(state, tmp, 0);
                        if ((pre.reg == post.reg) && (pre.regcm == post.regcm)) {
@@ -11215,6 +18162,8 @@ static void insert_mandatory_copies(struct compile_state *state)
        next:
                ins = ins->next;
        } while(ins != first);
+
+       print_blocks(state, __func__, state->dbgout);
 }
 
 
@@ -11266,10 +18215,142 @@ struct reg_state {
        unsigned defs;
        unsigned ranges;
        int passes, max_passes;
-#define MAX_ALLOCATION_PASSES 100
 };
 
 
+struct print_interference_block_info {
+       struct reg_state *rstate;
+       FILE *fp;
+       int need_edges;
+};
+static void print_interference_block(
+       struct compile_state *state, struct block *block, void *arg)
+
+{
+       struct print_interference_block_info *info = arg;
+       struct reg_state *rstate = info->rstate;
+       struct block_set *edge;
+       FILE *fp = info->fp;
+       struct reg_block *rb;
+       struct triple *ptr;
+       int phi_present;
+       int done;
+       rb = &rstate->blocks[block->vertex];
+
+       fprintf(fp, "\nblock: %p (%d),",
+               block,  block->vertex);
+       for(edge = block->edges; edge; edge = edge->next) {
+               fprintf(fp, " %p<-%p",
+                       edge->member,
+                       edge->member && edge->member->use?edge->member->use->member : 0);
+       }
+       fprintf(fp, "\n");
+       if (rb->in) {
+               struct triple_reg_set *in_set;
+               fprintf(fp, "        in:");
+               for(in_set = rb->in; in_set; in_set = in_set->next) {
+                       fprintf(fp, " %-10p", in_set->member);
+               }
+               fprintf(fp, "\n");
+       }
+       phi_present = 0;
+       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
+               done = (ptr == block->last);
+               if (ptr->op == OP_PHI) {
+                       phi_present = 1;
+                       break;
+               }
+       }
+       if (phi_present) {
+               int edge;
+               for(edge = 0; edge < block->users; edge++) {
+                       fprintf(fp, "     in(%d):", edge);
+                       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
+                               struct triple **slot;
+                               done = (ptr == block->last);
+                               if (ptr->op != OP_PHI) {
+                                       continue;
+                               }
+                               slot = &RHS(ptr, 0);
+                               fprintf(fp, " %-10p", slot[edge]);
+                       }
+                       fprintf(fp, "\n");
+               }
+       }
+       if (block->first->op == OP_LABEL) {
+               fprintf(fp, "%p:\n", block->first);
+       }
+       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
+               struct live_range *lr;
+               unsigned id;
+               done = (ptr == block->last);
+               lr = rstate->lrd[ptr->id].lr;
+
+               id = ptr->id;
+               ptr->id = rstate->lrd[id].orig_id;
+               SET_REG(ptr->id, lr->color);
+               display_triple(fp, ptr);
+               ptr->id = id;
+
+               if (triple_is_def(state, ptr) && (lr->defs == 0)) {
+                       internal_error(state, ptr, "lr has no defs!");
+               }
+               if (info->need_edges) {
+                       if (lr->defs) {
+                               struct live_range_def *lrd;
+                               fprintf(fp, "       range:");
+                               lrd = lr->defs;
+                               do {
+                                       fprintf(fp, " %-10p", lrd->def);
+                                       lrd = lrd->next;
+                               } while(lrd != lr->defs);
+                               fprintf(fp, "\n");
+                       }
+                       if (lr->edges > 0) {
+                               struct live_range_edge *edge;
+                               fprintf(fp, "       edges:");
+                               for(edge = lr->edges; edge; edge = edge->next) {
+                                       struct live_range_def *lrd;
+                                       lrd = edge->node->defs;
+                                       do {
+                                               fprintf(fp, " %-10p", lrd->def);
+                                               lrd = lrd->next;
+                                       } while(lrd != edge->node->defs);
+                                       fprintf(fp, "|");
+                               }
+                               fprintf(fp, "\n");
+                       }
+               }
+               /* Do a bunch of sanity checks */
+               valid_ins(state, ptr);
+               if ((ptr->id < 0) || (ptr->id > rstate->defs)) {
+                       internal_error(state, ptr, "Invalid triple id: %d",
+                               ptr->id);
+               }
+       }
+       if (rb->out) {
+               struct triple_reg_set *out_set;
+               fprintf(fp, "       out:");
+               for(out_set = rb->out; out_set; out_set = out_set->next) {
+                       fprintf(fp, " %-10p", out_set->member);
+               }
+               fprintf(fp, "\n");
+       }
+       fprintf(fp, "\n");
+}
+
+static void print_interference_blocks(
+       struct compile_state *state, struct reg_state *rstate, FILE *fp, int need_edges)
+{
+       struct print_interference_block_info info;
+       info.rstate = rstate;
+       info.fp = fp;
+       info.need_edges = need_edges;
+       fprintf(fp, "\nlive variables by block\n");
+       walk_blocks(state, &state->bb, print_interference_block, &info);
+
+}
+
 static unsigned regc_max_size(struct compile_state *state, int classes)
 {
        unsigned max_size;
@@ -11369,9 +18450,12 @@ static struct lre_hash **lre_probe(struct reg_state *rstate,
                right = tmp;
        }
        index = hash_live_edge(left, right);
-       
+
        ptr = &rstate->hash[index];
-       while((*ptr) && ((*ptr)->left != left) && ((*ptr)->right != right)) {
+       while(*ptr) {
+               if (((*ptr)->left == left) && ((*ptr)->right == right)) {
+                       break;
+               }
                ptr = &(*ptr)->next;
        }
        return ptr;
@@ -11385,7 +18469,7 @@ static int interfere(struct reg_state *rstate,
        return ptr && *ptr;
 }
 
-static void add_live_edge(struct reg_state *rstate, 
+static void add_live_edge(struct reg_state *rstate,
        struct live_range *left, struct live_range *right)
 {
        /* FIXME the memory allocation overhead is noticeable here... */
@@ -11409,6 +18493,10 @@ static void add_live_edge(struct reg_state *rstate,
        if (*ptr) {
                return;
        }
+#if 0
+       fprintf(state->errout, "new_live_edge(%p, %p)\n",
+               left, right);
+#endif
        new_hash = xmalloc(sizeof(*new_hash), "lre_hash");
        new_hash->next  = *ptr;
        new_hash->left  = left;
@@ -11420,7 +18508,7 @@ static void add_live_edge(struct reg_state *rstate,
        edge->node   = right;
        left->edges  = edge;
        left->degree += 1;
-       
+
        edge = xmalloc(sizeof(*edge), "live_range_edge");
        edge->next    = right->edges;
        edge->node    = left;
@@ -11447,6 +18535,7 @@ static void remove_live_edge(struct reg_state *rstate,
                        *ptr = edge->next;
                        memset(edge, 0, sizeof(*edge));
                        xfree(edge);
+                       right->degree--;
                        break;
                }
        }
@@ -11456,6 +18545,7 @@ static void remove_live_edge(struct reg_state *rstate,
                        *ptr = edge->next;
                        memset(edge, 0, sizeof(*edge));
                        xfree(edge);
+                       left->degree--;
                        break;
                }
        }
@@ -11470,9 +18560,22 @@ static void remove_live_edges(struct reg_state *rstate, struct live_range *range
        }
 }
 
+static void transfer_live_edges(struct reg_state *rstate,
+       struct live_range *dest, struct live_range *src)
+{
+       struct live_range_edge *edge, *next;
+       for(edge = src->edges; edge; edge = next) {
+               struct live_range *other;
+               next = edge->next;
+               other = edge->node;
+               remove_live_edge(rstate, src, other);
+               add_live_edge(rstate, dest, other);
+       }
+}
+
 
 /* Interference graph...
- * 
+ *
  * new(n) --- Return a graph with n nodes but no edges.
  * add(g,x,y) --- Return a graph including g with an between x and y
  * interfere(g, x, y) --- Return true if there exists an edge between the nodes
@@ -11485,11 +18588,11 @@ static void remove_live_edges(struct reg_state *rstate, struct live_range *range
  * The adjacency vectors support an efficient implementation of neighbors.
  */
 
-/* 
+/*
  *     +---------------------------------------------------+
  *     |         +--------------+                          |
  *     v         v              |                          |
- * renumber -> build graph -> colalesce -> spill_costs -> simplify -> select 
+ * renumber -> build graph -> colalesce -> spill_costs -> simplify -> select
  *
  * -- In simplify implment optimistic coloring... (No backtracking)
  * -- Implement Rematerialization it is the only form of spilling we can perform
@@ -11503,8 +18606,9 @@ static void remove_live_edges(struct reg_state *rstate, struct live_range *range
  *
  */
 
+#if DEBUG_ROMCC_WARNING
 static void different_colored(
-       struct compile_state *state, struct reg_state *rstate, 
+       struct compile_state *state, struct reg_state *rstate,
        struct triple *parent, struct triple *ins)
 {
        struct live_range *lr;
@@ -11522,7 +18626,7 @@ static void different_colored(
                }
        }
 }
-
+#endif
 
 static struct live_range *coalesce_ranges(
        struct compile_state *state, struct reg_state *rstate,
@@ -11540,13 +18644,13 @@ static struct live_range *coalesce_ranges(
        }
        if ((lr1->color == REG_UNNEEDED) ||
                (lr2->color == REG_UNNEEDED)) {
-               internal_error(state, 0, 
+               internal_error(state, 0,
                        "cannot coalesce live ranges without a possible color");
        }
        if ((lr1->color != lr2->color) &&
                (lr1->color != REG_UNSET) &&
                (lr2->color != REG_UNSET)) {
-               internal_error(state, lr1->defs->def, 
+               internal_error(state, lr1->defs->def,
                        "cannot coalesce live ranges of different colors");
        }
        color = lr1->color;
@@ -11558,10 +18662,26 @@ static struct live_range *coalesce_ranges(
                internal_error(state, lr1->defs->def,
                        "cannot coalesce live ranges with dissimilar register classes");
        }
+       if (state->compiler->debug & DEBUG_COALESCING) {
+               FILE *fp = state->errout;
+               fprintf(fp, "coalescing:");
+               lrd = lr1->defs;
+               do {
+                       fprintf(fp, " %p", lrd->def);
+                       lrd = lrd->next;
+               } while(lrd != lr1->defs);
+               fprintf(fp, " |");
+               lrd = lr2->defs;
+               do {
+                       fprintf(fp, " %p", lrd->def);
+                       lrd = lrd->next;
+               } while(lrd != lr2->defs);
+               fprintf(fp, "\n");
+       }
        /* If there is a clear dominate live range put it in lr1,
         * For purposes of this test phi functions are
         * considered dominated by the definitions that feed into
-        * them. 
+        * them.
         */
        if ((lr1->defs->prev->def->op == OP_PHI) ||
                ((lr2->defs->prev->def->op != OP_PHI) &&
@@ -11573,34 +18693,41 @@ static struct live_range *coalesce_ranges(
        }
 #if 0
        if (lr1->defs->orig_id  & TRIPLE_FLAG_POST_SPLIT) {
-               fprintf(stderr, "lr1 post\n");
+               fprintf(state->errout, "lr1 post\n");
        }
        if (lr1->defs->orig_id & TRIPLE_FLAG_PRE_SPLIT) {
-               fprintf(stderr, "lr1 pre\n");
+               fprintf(state->errout, "lr1 pre\n");
        }
        if (lr2->defs->orig_id  & TRIPLE_FLAG_POST_SPLIT) {
-               fprintf(stderr, "lr2 post\n");
+               fprintf(state->errout, "lr2 post\n");
        }
        if (lr2->defs->orig_id & TRIPLE_FLAG_PRE_SPLIT) {
-               fprintf(stderr, "lr2 pre\n");
+               fprintf(state->errout, "lr2 pre\n");
        }
 #endif
 #if 0
-       fprintf(stderr, "coalesce color1(%p): %3d color2(%p) %3d\n",
+       fprintf(state->errout, "coalesce color1(%p): %3d color2(%p) %3d\n",
                lr1->defs->def,
                lr1->color,
                lr2->defs->def,
                lr2->color);
 #endif
-       
-       lr1->classes = classes;
+
        /* Append lr2 onto lr1 */
+#if DEBUG_ROMCC_WARNINGS
 #warning "FIXME should this be a merge instead of a splice?"
+#endif
+       /* This FIXME item applies to the correctness of live_range_end
+        * and to the necessity of making multiple passes of coalesce_live_ranges.
+        * A failure to find some coalesce opportunities in coaleace_live_ranges
+        * does not impact the correct of the compiler just the efficiency with
+        * which registers are allocated.
+        */
        head = lr1->defs;
        mid1 = lr1->defs->prev;
        mid2 = lr2->defs;
        end  = lr2->defs->prev;
-       
+
        head->prev = end;
        end->next  = head;
 
@@ -11626,6 +18753,9 @@ static struct live_range *coalesce_ranges(
        lr1->color   = color;
        lr1->classes = classes;
 
+       /* Keep the graph in sync by transfering the edges from lr2 to lr1 */
+       transfer_live_edges(rstate, lr1, lr2);
+
        return lr1;
 }
 
@@ -11667,14 +18797,14 @@ static void initialize_live_ranges(
        size_t count, size;
        int i, j;
 
-       first = RHS(state->main_function, 0);
+       first = state->first;
        /* First count how many instructions I have.
         */
        count = count_triples(state);
        /* Potentially I need one live range definitions for each
-        * instruction, plus an extra for the split routines.
+        * instruction.
         */
-       rstate->defs = count + 1;
+       rstate->defs = count;
        /* Potentially I need one live range for each instruction
         * plus an extra for the dummy live range.
         */
@@ -11696,14 +18826,13 @@ static void initialize_live_ranges(
                        struct reg_info info;
                        /* Find the architecture specific color information */
                        info = find_def_color(state, ins);
-
                        i++;
                        rstate->lr[i].defs    = &rstate->lrd[j];
                        rstate->lr[i].color   = info.reg;
                        rstate->lr[i].classes = info.regcm;
                        rstate->lr[i].degree  = 0;
                        rstate->lrd[j].lr = &rstate->lr[i];
-               } 
+               }
                /* Otherwise give the triple the dummy live range. */
                else {
                        rstate->lrd[j].lr = &rstate->lr[0];
@@ -11720,7 +18849,6 @@ static void initialize_live_ranges(
                ins = ins->next;
        } while(ins != first);
        rstate->ranges = i;
-       rstate->defs -= 1;
 
        /* Make a second pass to handle achitecture specific register
         * constraints.
@@ -11731,14 +18859,19 @@ static void initialize_live_ranges(
                if (ins->id > rstate->defs) {
                        internal_error(state, ins, "bad id");
                }
-               
+
                /* Walk through the template of ins and coalesce live ranges */
-               zlhs = TRIPLE_LHS(ins->sizes);
+               zlhs = ins->lhs;
                if ((zlhs == 0) && triple_is_def(state, ins)) {
                        zlhs = 1;
                }
-               zrhs = TRIPLE_RHS(ins->sizes);
-               
+               zrhs = ins->rhs;
+
+               if (state->compiler->debug & DEBUG_COALESCING2) {
+                       fprintf(state->errout, "mandatory coalesce: %p %d %d\n",
+                               ins, zlhs, zrhs);
+               }
+
                for(i = 0; i < zlhs; i++) {
                        struct reg_info linfo;
                        struct live_range_def *lhs;
@@ -11751,6 +18884,12 @@ static void initialize_live_ranges(
                        } else {
                                lhs = &rstate->lrd[LHS(ins, i)->id];
                        }
+
+                       if (state->compiler->debug & DEBUG_COALESCING2) {
+                               fprintf(state->errout, "coalesce lhs(%d): %p %d\n",
+                                       i, lhs, linfo.reg);
+                       }
+
                        for(j = 0; j < zrhs; j++) {
                                struct reg_info rinfo;
                                struct live_range_def *rhs;
@@ -11758,9 +18897,15 @@ static void initialize_live_ranges(
                                if (rinfo.reg < MAX_REGISTERS) {
                                        continue;
                                }
-                               rhs = &rstate->lrd[RHS(ins, i)->id];
+                               rhs = &rstate->lrd[RHS(ins, j)->id];
+
+                               if (state->compiler->debug & DEBUG_COALESCING2) {
+                                       fprintf(state->errout, "coalesce rhs(%d): %p %d\n",
+                                               j, rhs, rinfo.reg);
+                               }
+
                                if (rinfo.reg == linfo.reg) {
-                                       coalesce_ranges(state, rstate, 
+                                       coalesce_ranges(state, rstate,
                                                lhs->lr, rhs->lr);
                                }
                        }
@@ -11769,9 +18914,9 @@ static void initialize_live_ranges(
        } while(ins != first);
 }
 
-static struct triple *graph_ins(
-       struct compile_state *state, 
-       struct reg_block *blocks, struct triple_reg_set *live, 
+static void graph_ins(
+       struct compile_state *state,
+       struct reg_block *blocks, struct triple_reg_set *live,
        struct reg_block *rb, struct triple *ins, void *arg)
 {
        struct reg_state *rstate = arg;
@@ -11783,10 +18928,10 @@ static struct triple *graph_ins(
         * the interference graph.
         */
        if (!triple_is_def(state, ins)) {
-               return ins;
+               return;
        }
        def = rstate->lrd[ins->id].lr;
-       
+
        /* Create an edge between ins and everything that is
         * alive, unless the live_range cannot share
         * a physical register with ins.
@@ -11805,57 +18950,143 @@ static struct triple *graph_ins(
                }
                add_live_edge(rstate, def, lr);
        }
-       return ins;
+       return;
+}
+
+#if DEBUG_CONSISTENCY > 1
+static struct live_range *get_verify_live_range(
+       struct compile_state *state, struct reg_state *rstate, struct triple *ins)
+{
+       struct live_range *lr;
+       struct live_range_def *lrd;
+       int ins_found;
+       if ((ins->id < 0) || (ins->id > rstate->defs)) {
+               internal_error(state, ins, "bad ins?");
+       }
+       lr = rstate->lrd[ins->id].lr;
+       ins_found = 0;
+       lrd = lr->defs;
+       do {
+               if (lrd->def == ins) {
+                       ins_found = 1;
+               }
+               lrd = lrd->next;
+       } while(lrd != lr->defs);
+       if (!ins_found) {
+               internal_error(state, ins, "ins not in live range");
+       }
+       return lr;
 }
 
+static void verify_graph_ins(
+       struct compile_state *state,
+       struct reg_block *blocks, struct triple_reg_set *live,
+       struct reg_block *rb, struct triple *ins, void *arg)
+{
+       struct reg_state *rstate = arg;
+       struct triple_reg_set *entry1, *entry2;
+
+
+       /* Compare live against edges and make certain the code is working */
+       for(entry1 = live; entry1; entry1 = entry1->next) {
+               struct live_range *lr1;
+               lr1 = get_verify_live_range(state, rstate, entry1->member);
+               for(entry2 = live; entry2; entry2 = entry2->next) {
+                       struct live_range *lr2;
+                       struct live_range_edge *edge2;
+                       int lr1_found;
+                       int lr2_degree;
+                       if (entry2 == entry1) {
+                               continue;
+                       }
+                       lr2 = get_verify_live_range(state, rstate, entry2->member);
+                       if (lr1 == lr2) {
+                               internal_error(state, entry2->member,
+                                       "live range with 2 values simultaneously alive");
+                       }
+                       if (!arch_regcm_intersect(lr1->classes, lr2->classes)) {
+                               continue;
+                       }
+                       if (!interfere(rstate, lr1, lr2)) {
+                               internal_error(state, entry2->member,
+                                       "edges don't interfere?");
+                       }
+
+                       lr1_found = 0;
+                       lr2_degree = 0;
+                       for(edge2 = lr2->edges; edge2; edge2 = edge2->next) {
+                               lr2_degree++;
+                               if (edge2->node == lr1) {
+                                       lr1_found = 1;
+                               }
+                       }
+                       if (lr2_degree != lr2->degree) {
+                               internal_error(state, entry2->member,
+                                       "computed degree: %d does not match reported degree: %d\n",
+                                       lr2_degree, lr2->degree);
+                       }
+                       if (!lr1_found) {
+                               internal_error(state, entry2->member, "missing edge");
+                       }
+               }
+       }
+       return;
+}
+#endif
 
-static struct triple *print_interference_ins(
-       struct compile_state *state, 
-       struct reg_block *blocks, struct triple_reg_set *live, 
+static void print_interference_ins(
+       struct compile_state *state,
+       struct reg_block *blocks, struct triple_reg_set *live,
        struct reg_block *rb, struct triple *ins, void *arg)
 {
        struct reg_state *rstate = arg;
        struct live_range *lr;
+       unsigned id;
+       FILE *fp = state->dbgout;
 
        lr = rstate->lrd[ins->id].lr;
-       display_triple(stdout, ins);
+       id = ins->id;
+       ins->id = rstate->lrd[id].orig_id;
+       SET_REG(ins->id, lr->color);
+       display_triple(state->dbgout, ins);
+       ins->id = id;
 
        if (lr->defs) {
                struct live_range_def *lrd;
-               printf("       range:");
+               fprintf(fp, "       range:");
                lrd = lr->defs;
                do {
-                       printf(" %-10p", lrd->def);
+                       fprintf(fp, " %-10p", lrd->def);
                        lrd = lrd->next;
                } while(lrd != lr->defs);
-               printf("\n");
+               fprintf(fp, "\n");
        }
        if (live) {
                struct triple_reg_set *entry;
-               printf("        live:");
+               fprintf(fp, "        live:");
                for(entry = live; entry; entry = entry->next) {
-                       printf(" %-10p", entry->member);
+                       fprintf(fp, " %-10p", entry->member);
                }
-               printf("\n");
+               fprintf(fp, "\n");
        }
        if (lr->edges) {
                struct live_range_edge *entry;
-               printf("       edges:");
+               fprintf(fp, "       edges:");
                for(entry = lr->edges; entry; entry = entry->next) {
                        struct live_range_def *lrd;
                        lrd = entry->node->defs;
                        do {
-                               printf(" %-10p", lrd->def);
+                               fprintf(fp, " %-10p", lrd->def);
                                lrd = lrd->next;
                        } while(lrd != entry->node->defs);
-                       printf("|");
+                       fprintf(fp, "|");
                }
-               printf("\n");
+               fprintf(fp, "\n");
        }
        if (triple_is_branch(state, ins)) {
-               printf("\n");
+               fprintf(fp, "\n");
        }
-       return ins;
+       return;
 }
 
 static int coalesce_live_ranges(
@@ -11875,7 +19106,7 @@ static int coalesce_live_ranges(
         * Forcing a value to stay in a single register
         * for an extended period of time does have
         * limitations when applied to non homogenous
-        * register pool.  
+        * register pool.
         *
         * The two cases I have identified are:
         * 1) Two forced register assignments may
@@ -11893,7 +19124,7 @@ static int coalesce_live_ranges(
         *   functions.  This creates a 2 headed live
         *   range that cannot be sanely split.
         *
-        * - phi functions (coalesced in initialize_live_ranges) 
+        * - phi functions (coalesced in initialize_live_ranges)
         *   are handled as pre split live ranges so we will
         *   never attempt to split them.
         */
@@ -11943,11 +19174,11 @@ static int coalesce_live_ranges(
                                if ((lr1->classes & lr2->classes) == 0) {
                                        continue;
                                }
-                               
+
                                if (interfere(rstate, lr1, lr2)) {
                                        continue;
                                }
-                               
+
                                res = coalesce_ranges(state, rstate, lr1, lr2);
                                coalesced += 1;
                                if (res != lr1) {
@@ -11962,17 +19193,12 @@ static int coalesce_live_ranges(
 }
 
 
-struct coalesce_conflict {
-       struct triple *ins;
-       int index;
-};
-static struct triple *spot_coalesce_conflict(struct compile_state *state,
+static void fix_coalesce_conflicts(struct compile_state *state,
        struct reg_block *blocks, struct triple_reg_set *live,
        struct reg_block *rb, struct triple *ins, void *arg)
 {
-       struct coalesce_conflict *conflict = arg;
+       int *conflicts = arg;
        int zlhs, zrhs, i, j;
-       int found;
 
        /* See if we have a mandatory coalesce operation between
         * a lhs and a rhs value.  If so and the rhs value is also
@@ -11980,411 +19206,405 @@ static struct triple *spot_coalesce_conflict(struct compile_state *state,
         * we would have two definitions in the same live range simultaneously
         * alive.
         */
-       found = -1;
-       zlhs = TRIPLE_LHS(ins->sizes);
+       zlhs = ins->lhs;
        if ((zlhs == 0) && triple_is_def(state, ins)) {
                zlhs = 1;
        }
-       zrhs = TRIPLE_RHS(ins->sizes);
-       for(i = 0; (i < zlhs) && (found == -1); i++) {
+       zrhs = ins->rhs;
+       for(i = 0; i < zlhs; i++) {
                struct reg_info linfo;
                linfo = arch_reg_lhs(state, ins, i);
                if (linfo.reg < MAX_REGISTERS) {
                        continue;
                }
-               for(j = 0; (j < zrhs) && (found == -1); j++) {
+               for(j = 0; j < zrhs; j++) {
                        struct reg_info rinfo;
                        struct triple *rhs;
                        struct triple_reg_set *set;
+                       int found;
+                       found = 0;
                        rinfo = arch_reg_rhs(state, ins, j);
                        if (rinfo.reg != linfo.reg) {
                                continue;
                        }
                        rhs = RHS(ins, j);
-                       for(set = live; set && (found == -1); set = set->next) {
+                       for(set = live; set && !found; set = set->next) {
                                if (set->member == rhs) {
-                                       found = j;
+                                       found = 1;
                                }
                        }
+                       if (found) {
+                               struct triple *copy;
+                               copy = pre_copy(state, ins, j);
+                               copy->id |= TRIPLE_FLAG_PRE_SPLIT;
+                               (*conflicts)++;
+                       }
                }
        }
-       /* Only update conflict if we are the least dominated conflict */
-       if ((found != -1) &&
-               (!conflict->ins || tdominates(state, ins, conflict->ins))) {
-               conflict->ins = ins;
-               conflict->index = found;
-       }
-       return ins;
+       return;
 }
 
-static void resolve_coalesce_conflict(
-       struct compile_state *state, struct coalesce_conflict *conflict)
+static int correct_coalesce_conflicts(
+       struct compile_state *state, struct reg_block *blocks)
 {
-       struct triple *copy;
-       copy = pre_copy(state, conflict->ins, conflict->index);
-       copy->id |= TRIPLE_FLAG_PRE_SPLIT;
+       int conflicts;
+       conflicts = 0;
+       walk_variable_lifetimes(state, &state->bb, blocks,
+               fix_coalesce_conflicts, &conflicts);
+       return conflicts;
 }
-       
 
-static struct triple *spot_tangle(struct compile_state *state,
-       struct reg_block *blocks, struct triple_reg_set *live,
-       struct reg_block *rb, struct triple *ins, void *arg)
+static void replace_set_use(struct compile_state *state,
+       struct triple_reg_set *head, struct triple *orig, struct triple *new)
 {
-       struct triple **tangle = arg;
-       char used[MAX_REGISTERS];
        struct triple_reg_set *set;
-
-       /* Find out which registers have multiple uses at this point */
-       memset(used, 0, sizeof(used));
-       for(set = live; set; set = set->next) {
-               struct reg_info info;
-               info = find_lhs_color(state, set->member, 0);
-               if (info.reg == REG_UNSET) {
-                       continue;
+       for(set = head; set; set = set->next) {
+               if (set->member == orig) {
+                       set->member = new;
                }
-               reg_inc_used(state, used, info.reg);
        }
+}
 
-       /* Now find the least dominated definition of a register in
-        * conflict I have seen so far.
-        */
-       for(set = live; set; set = set->next) {
-               struct reg_info info;
-               info = find_lhs_color(state, set->member, 0);
-               if (used[info.reg] < 2) {
-                       continue;
-               }
-               if (!*tangle || tdominates(state, set->member, *tangle)) {
-                       *tangle = set->member;
+static void replace_block_use(struct compile_state *state,
+       struct reg_block *blocks, struct triple *orig, struct triple *new)
+{
+       int i;
+#if DEBUG_ROMCC_WARNINGS
+#warning "WISHLIST visit just those blocks that need it *"
+#endif
+       for(i = 1; i <= state->bb.last_vertex; i++) {
+               struct reg_block *rb;
+               rb = &blocks[i];
+               replace_set_use(state, rb->in, orig, new);
+               replace_set_use(state, rb->out, orig, new);
+       }
+}
+
+static void color_instructions(struct compile_state *state)
+{
+       struct triple *ins, *first;
+       first = state->first;
+       ins = first;
+       do {
+               if (triple_is_def(state, ins)) {
+                       struct reg_info info;
+                       info = find_lhs_color(state, ins, 0);
+                       if (info.reg >= MAX_REGISTERS) {
+                               info.reg = REG_UNSET;
+                       }
+                       SET_INFO(ins->id, info);
                }
+               ins = ins->next;
+       } while(ins != first);
+}
+
+static struct reg_info read_lhs_color(
+       struct compile_state *state, struct triple *ins, int index)
+{
+       struct reg_info info;
+       if ((index == 0) && triple_is_def(state, ins)) {
+               info.reg   = ID_REG(ins->id);
+               info.regcm = ID_REGCM(ins->id);
        }
-       return ins;
+       else if (index < ins->lhs) {
+               info = read_lhs_color(state, LHS(ins, index), 0);
+       }
+       else {
+               internal_error(state, ins, "Bad lhs %d", index);
+               info.reg = REG_UNSET;
+               info.regcm = 0;
+       }
+       return info;
 }
 
-static void resolve_tangle(struct compile_state *state, struct triple *tangle)
+static struct triple *resolve_tangle(
+       struct compile_state *state, struct triple *tangle)
 {
        struct reg_info info, uinfo;
        struct triple_set *set, *next;
        struct triple *copy;
 
-#if 0
-       fprintf(stderr, "Resolving tangle: %p\n", tangle);
-       print_blocks(state, stderr);
+#if DEBUG_ROMCC_WARNINGS
+#warning "WISHLIST recalculate all affected instructions colors"
 #endif
        info = find_lhs_color(state, tangle, 0);
-#if 0
-       fprintf(stderr, "color: %d\n", info.reg);
-#endif
        for(set = tangle->use; set; set = next) {
                struct triple *user;
                int i, zrhs;
                next = set->next;
                user = set->member;
-               zrhs = TRIPLE_RHS(user->sizes);
+               zrhs = user->rhs;
                for(i = 0; i < zrhs; i++) {
                        if (RHS(user, i) != tangle) {
                                continue;
                        }
                        uinfo = find_rhs_post_color(state, user, i);
-#if 0
-                       fprintf(stderr, "%p rhs %d color: %d\n", 
-                               user, i, uinfo.reg);
-#endif
                        if (uinfo.reg == info.reg) {
                                copy = pre_copy(state, user, i);
                                copy->id |= TRIPLE_FLAG_PRE_SPLIT;
+                               SET_INFO(copy->id, uinfo);
                        }
                }
        }
+       copy = 0;
        uinfo = find_lhs_pre_color(state, tangle, 0);
-#if 0
-       fprintf(stderr, "pre color: %d\n", uinfo.reg);
-#endif
        if (uinfo.reg == info.reg) {
+               struct reg_info linfo;
                copy = post_copy(state, tangle);
                copy->id |= TRIPLE_FLAG_PRE_SPLIT;
+               linfo = find_lhs_color(state, copy, 0);
+               SET_INFO(copy->id, linfo);
        }
+       info = find_lhs_color(state, tangle, 0);
+       SET_INFO(tangle->id, info);
+
+       return copy;
 }
 
 
-struct least_conflict {
-       struct reg_state *rstate;
-       struct live_range *ref_range;
-       struct triple *ins;
-       struct triple_reg_set *live;
-       size_t count;
-};
-static struct triple *least_conflict(struct compile_state *state,
+static void fix_tangles(struct compile_state *state,
        struct reg_block *blocks, struct triple_reg_set *live,
        struct reg_block *rb, struct triple *ins, void *arg)
 {
-       struct least_conflict *conflict = arg;
-       struct live_range_edge *edge;
-       struct triple_reg_set *set;
-       size_t count;
-
-#warning "FIXME handle instructions with left hand sides..."
-       /* Only instructions that introduce a new definition
-        * can be the conflict instruction.
-        */
-       if (!triple_is_def(state, ins)) {
-               return ins;
-       }
+       int *tangles = arg;
+       struct triple *tangle;
+       do {
+               char used[MAX_REGISTERS];
+               struct triple_reg_set *set;
+               tangle = 0;
 
-       /* See if live ranges at this instruction are a
-        * strict subset of the live ranges that are in conflict.
-        */
-       count = 0;
-       for(set = live; set; set = set->next) {
-               struct live_range *lr;
-               lr = conflict->rstate->lrd[set->member->id].lr;
-               for(edge = conflict->ref_range->edges; edge; edge = edge->next) {
-                       if (edge->node == lr) {
-                               break;
+               /* Find out which registers have multiple uses at this point */
+               memset(used, 0, sizeof(used));
+               for(set = live; set; set = set->next) {
+                       struct reg_info info;
+                       info = read_lhs_color(state, set->member, 0);
+                       if (info.reg == REG_UNSET) {
+                               continue;
                        }
+                       reg_inc_used(state, used, info.reg);
                }
-               if (!edge && (lr != conflict->ref_range)) {
-                       return ins;
-               }
-               count++;
-       }
-       if (count <= 1) {
-               return ins;
-       }
-
-       /* See if there is an uncolored member in this subset. 
-        */
-        for(set = live; set; set = set->next) {
-               struct live_range *lr;
-               lr = conflict->rstate->lrd[set->member->id].lr;
-               if (lr->color == REG_UNSET) {
-                       break;
-               }
-       }
-       if (!set && (conflict->ref_range != REG_UNSET)) {
-               return ins;
-       }
 
-
-       /* Find the instruction with the largest possible subset of
-        * conflict ranges and that dominates any other instruction
-        * with an equal sized set of conflicting ranges.
-        */
-       if ((count > conflict->count) ||
-               ((count == conflict->count) &&
-                       tdominates(state, ins, conflict->ins))) {
-               struct triple_reg_set *next;
-               /* Remember the canidate instruction */
-               conflict->ins = ins;
-               conflict->count = count;
-               /* Free the old collection of live registers */
-               for(set = conflict->live; set; set = next) {
-                       next = set->next;
-                       do_triple_unset(&conflict->live, set->member);
-               }
-               conflict->live = 0;
-               /* Rember the registers that are alive but do not feed
-                * into or out of conflict->ins.
+               /* Now find the least dominated definition of a register in
+                * conflict I have seen so far.
                 */
                for(set = live; set; set = set->next) {
-                       struct triple **expr;
-                       if (set->member == ins) {
-                               goto next;
+                       struct reg_info info;
+                       info = read_lhs_color(state, set->member, 0);
+                       if (used[info.reg] < 2) {
+                               continue;
                        }
-                       expr = triple_rhs(state, ins, 0);
-                       for(;expr; expr = triple_rhs(state, ins, expr)) {
-                               if (*expr == set->member) {
-                                       goto next;
-                               }
+                       /* Changing copies that feed into phi functions
+                        * is incorrect.
+                        */
+                       if (set->member->use &&
+                               (set->member->use->member->op == OP_PHI)) {
+                               continue;
                        }
-                       expr = triple_lhs(state, ins, 0);
-                       for(; expr; expr = triple_lhs(state, ins, expr)) {
-                               if (*expr == set->member) {
-                                       goto next;
-                               }
+                       if (!tangle || tdominates(state, set->member, tangle)) {
+                               tangle = set->member;
                        }
-                       do_triple_set(&conflict->live, set->member, set->new);
-               next:
-                       ;
                }
-       }
-       return ins;
+               /* If I have found a tangle resolve it */
+               if (tangle) {
+                       struct triple *post_copy;
+                       (*tangles)++;
+                       post_copy = resolve_tangle(state, tangle);
+                       if (post_copy) {
+                               replace_block_use(state, blocks, tangle, post_copy);
+                       }
+                       if (post_copy && (tangle != ins)) {
+                               replace_set_use(state, live, tangle, post_copy);
+                       }
+               }
+       } while(tangle);
+       return;
 }
 
-static void find_range_conflict(struct compile_state *state,
-       struct reg_state *rstate, char *used, struct live_range *ref_range,
-       struct least_conflict *conflict)
+static int correct_tangles(
+       struct compile_state *state, struct reg_block *blocks)
 {
-       /* there are 3 kinds ways conflicts can occure.
-        * 1) the life time of 2 values simply overlap.
-        * 2) the 2 values feed into the same instruction.
-        * 3) the 2 values feed into a phi function.
-        */
+       int tangles;
+       tangles = 0;
+       color_instructions(state);
+       walk_variable_lifetimes(state, &state->bb, blocks,
+               fix_tangles, &tangles);
+       return tangles;
+}
 
-       /* find the instruction where the problematic conflict comes
-        * into existance.  that the instruction where all of
-        * the values are alive, and among such instructions it is
-        * the least dominated one.
-        *
-        * a value is alive an an instruction if either;
-        * 1) the value defintion dominates the instruction and there
-        *    is a use at or after that instrction
-        * 2) the value definition feeds into a phi function in the
-        *    same block as the instruction.  and the phi function
-        *    is at or after the instruction.
-        */
-       memset(conflict, 0, sizeof(*conflict));
-       conflict->rstate    = rstate;
-       conflict->ref_range = ref_range;
-       conflict->ins       = 0;
-       conflict->count     = 0;
-       conflict->live      = 0;
-       walk_variable_lifetimes(state, rstate->blocks, least_conflict, conflict);
 
-       if (!conflict->ins) {
-               internal_error(state, 0, "No conflict ins?");
-       }
-       if (!conflict->live) {
-               internal_error(state, 0, "No conflict live?");
-       }
-       return;
-}
+static void ids_from_rstate(struct compile_state *state, struct reg_state *rstate);
+static void cleanup_rstate(struct compile_state *state, struct reg_state *rstate);
 
-static struct triple *split_constrained_range(struct compile_state *state, 
-       struct reg_state *rstate, char *used, struct least_conflict *conflict)
+struct triple *find_constrained_def(
+       struct compile_state *state, struct live_range *range, struct triple *constrained)
 {
-       unsigned constrained_size;
-       struct triple *new, *constrained;
-       struct triple_reg_set *cset;
-       /* Find a range that is having problems because it is
-        * artificially constrained.
-        */
-       constrained_size = ~0;
-       constrained = 0;
-       new = 0;
-       for(cset = conflict->live; cset; cset = cset->next) {
-               struct triple_set *set;
+       struct live_range_def *lrd, *lrd_next;
+       lrd_next = range->defs;
+       do {
                struct reg_info info;
-               unsigned classes;
-               unsigned cur_size, size;
-               /* Skip the live range that starts with conflict->ins */
-               if (cset->member == conflict->ins) {
-                       continue;
-               }
-               /* Find how many registers this value can potentially
-                * be assigned to.
-                */
-               classes = arch_type_to_regcm(state, cset->member->type);
-               size = regc_max_size(state, classes);
+               unsigned regcm;
 
-               /* Find how many registers we allow this value to
-                * be assigned to.
+               lrd = lrd_next;
+               lrd_next = lrd->next;
+
+               regcm = arch_type_to_regcm(state, lrd->def->type);
+               info = find_lhs_color(state, lrd->def, 0);
+               regcm      = arch_regcm_reg_normalize(state, regcm);
+               info.regcm = arch_regcm_reg_normalize(state, info.regcm);
+               /* If the 2 register class masks are equal then
+                * the current register class is not constrained.
                 */
-               info = arch_reg_lhs(state, cset->member, 0);
-#warning "FIXME do I need a call to arch_reg_rhs around here somewhere?"
-               if ((info.reg == REG_UNSET) || (info.reg >= MAX_REGISTERS)) {
-                       cur_size = regc_max_size(state, info.regcm);
-               } else {
-                       cur_size = 1;
+               if (regcm == info.regcm) {
+                       continue;
                }
-               /* If this live_range feeds into conflict->ins
-                * splitting it is unlikely to help.
+
+               /* If there is just one use.
+                * That use cannot accept a larger register class.
+                * There are no intervening definitions except
+                * definitions that feed into that use.
+                * Then a triple is not constrained.
+                * FIXME handle this case!
                 */
-               for(set = cset->member->use; set; set = set->next) {
-                       if (set->member == conflict->ins) {
-                               goto next;
-                       }
-               }
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME ignore cases that cannot be fixed (a definition followed by a use)"
+#endif
+
 
-               /* If there is no difference between potential and
-                * actual register count there is nothing to do.
+               /* Of the constrained live ranges deal with the
+                * least dominated one first.
                 */
-               if (cur_size >= size) {
-                       continue;
+               if (state->compiler->debug & DEBUG_RANGE_CONFLICTS) {
+                       fprintf(state->errout, "canidate: %p %-8s regcm: %x %x\n",
+                               lrd->def, tops(lrd->def->op), regcm, info.regcm);
                }
-               /* Of the constrained registers deal with the
-                * most constrained one first.
-                */
                if (!constrained ||
-                       (size < constrained_size)) {
-                       constrained = cset->member;
-                       constrained_size = size;
+                       tdominates(state, lrd->def, constrained))
+               {
+                       constrained = lrd->def;
                }
-       next:
-               ;
+       } while(lrd_next != range->defs);
+       return constrained;
+}
+
+static int split_constrained_ranges(
+       struct compile_state *state, struct reg_state *rstate,
+       struct live_range *range)
+{
+       /* Walk through the edges in conflict and our current live
+        * range, and find definitions that are more severly constrained
+        * than they type of data they contain require.
+        *
+        * Then pick one of those ranges and relax the constraints.
+        */
+       struct live_range_edge *edge;
+       struct triple *constrained;
+
+       constrained = 0;
+       for(edge = range->edges; edge; edge = edge->next) {
+               constrained = find_constrained_def(state, edge->node, constrained);
+       }
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME should I call find_constrained_def here only if no previous constrained def was found?"
+#endif
+       if (!constrained) {
+               constrained = find_constrained_def(state, range, constrained);
+       }
+
+       if (state->compiler->debug & DEBUG_RANGE_CONFLICTS) {
+               fprintf(state->errout, "constrained: ");
+               display_triple(state->errout, constrained);
        }
        if (constrained) {
-               new = post_copy(state, constrained);
-               new->id |= TRIPLE_FLAG_POST_SPLIT;
+               ids_from_rstate(state, rstate);
+               cleanup_rstate(state, rstate);
+               resolve_tangle(state, constrained);
+       }
+       return !!constrained;
+}
+
+static int split_ranges(
+       struct compile_state *state, struct reg_state *rstate,
+       char *used, struct live_range *range)
+{
+       int split;
+       if (state->compiler->debug & DEBUG_RANGE_CONFLICTS) {
+               fprintf(state->errout, "split_ranges %d %s %p\n",
+                       rstate->passes, tops(range->defs->def->op), range->defs->def);
+       }
+       if ((range->color == REG_UNNEEDED) ||
+               (rstate->passes >= rstate->max_passes)) {
+               return 0;
+       }
+       split = split_constrained_ranges(state, rstate, range);
+
+       /* Ideally I would split the live range that will not be used
+        * for the longest period of time in hopes that this will
+        * (a) allow me to spill a register or
+        * (b) allow me to place a value in another register.
+        *
+        * So far I don't have a test case for this, the resolving
+        * of mandatory constraints has solved all of my
+        * know issues.  So I have choosen not to write any
+        * code until I cat get a better feel for cases where
+        * it would be useful to have.
+        *
+        */
+#if DEBUG_ROMCC_WARNINGS
+#warning "WISHLIST implement live range splitting..."
+#endif
+
+       if (!split && (state->compiler->debug & DEBUG_RANGE_CONFLICTS2)) {
+               FILE *fp = state->errout;
+               print_interference_blocks(state, rstate, fp, 0);
+               print_dominators(state, fp, &state->bb);
+       }
+       return split;
+}
+
+static FILE *cgdebug_fp(struct compile_state *state)
+{
+       FILE *fp;
+       fp = 0;
+       if (!fp && (state->compiler->debug & DEBUG_COLOR_GRAPH2)) {
+               fp = state->errout;
+       }
+       if (!fp && (state->compiler->debug & DEBUG_COLOR_GRAPH)) {
+               fp = state->dbgout;
        }
-       return new;
+       return fp;
 }
 
-static int split_ranges(
-       struct compile_state *state, struct reg_state *rstate, 
-       char *used, struct live_range *range)
+static void cgdebug_printf(struct compile_state *state, const char *fmt, ...)
 {
-       struct triple *new;
-
-       if ((range->color == REG_UNNEEDED) ||
-               (rstate->passes >= rstate->max_passes)) {
-               return 0;
+       FILE *fp;
+       fp = cgdebug_fp(state);
+       if (fp) {
+               va_list args;
+               va_start(args, fmt);
+               vfprintf(fp, fmt, args);
+               va_end(args);
        }
-       new = 0;
-       /* If I can't allocate a register something needs to be split */
-       if (arch_select_free_register(state, used, range->classes) == REG_UNSET) {
-               struct least_conflict conflict;
+}
 
-               /* Find where in the set of registers the conflict
-                * actually occurs.
-                */
-               find_range_conflict(state, rstate, used, range, &conflict);
-
-               /* If a range has been artifically constrained split it */
-               new = split_constrained_range(state, rstate, used, &conflict);
-               
-               if (!new) {
-               /* Ideally I would split the live range that will not be used
-                * for the longest period of time in hopes that this will 
-                * (a) allow me to spill a register or
-                * (b) allow me to place a value in another register.
-                *
-                * So far I don't have a test case for this, the resolving
-                * of mandatory constraints has solved all of my
-                * know issues.  So I have choosen not to write any
-                * code until I cat get a better feel for cases where
-                * it would be useful to have.
-                *
-                */
-#warning "WISHLIST implement live range splitting..."
-                       return 0;
-               }
-       }
-       if (new) {
-               rstate->lrd[rstate->defs].orig_id = new->id;
-               new->id = rstate->defs;
-               rstate->defs++;
-#if 0
-               fprintf(stderr, "new: %p\n", new);
-#endif
-               return 1;
+static void cgdebug_flush(struct compile_state *state)
+{
+       FILE *fp;
+       fp = cgdebug_fp(state);
+       if (fp) {
+               fflush(fp);
        }
-       return 0;
 }
 
-#if DEBUG_COLOR_GRAPH > 1
-#define cgdebug_printf(...) fprintf(stdout, __VA_ARGS__)
-#define cgdebug_flush() fflush(stdout)
-#elif DEBUG_COLOR_GRAPH == 1
-#define cgdebug_printf(...) fprintf(stderr, __VA_ARGS__)
-#define cgdebug_flush() fflush(stderr)
-#else
-#define cgdebug_printf(...)
-#define cgdebug_flush()
-#endif
+static void cgdebug_loc(struct compile_state *state, struct triple *ins)
+{
+       FILE *fp;
+       fp = cgdebug_fp(state);
+       if (fp) {
+               loc(fp, state, ins);
+       }
+}
 
-       
-static int select_free_color(struct compile_state *state, 
+static int select_free_color(struct compile_state *state,
        struct reg_state *rstate, struct live_range *range)
 {
        struct triple_set *entry;
@@ -12407,26 +19627,24 @@ static int select_free_color(struct compile_state *state,
                }
                reg_fill_used(state, used, edge->node->color);
        }
-#if DEBUG_COLOR_GRAPH > 1
-       {
+
+       if (state->compiler->debug & DEBUG_COLOR_GRAPH2) {
                int i;
                i = 0;
                for(edge = range->edges; edge; edge = edge->next) {
                        i++;
                }
-               cgdebug_printf("\n%s edges: %d @%s:%d.%d\n", 
-                       tops(range->def->op), i, 
-                       range->def->filename, range->def->line, range->def->col);
+               cgdebug_printf(state, "\n%s edges: %d",
+                       tops(range->defs->def->op), i);
+               cgdebug_loc(state, range->defs->def);
+               cgdebug_printf(state, "\n");
                for(i = 0; i < MAX_REGISTERS; i++) {
                        if (used[i]) {
-                               cgdebug_printf("used: %s\n",
+                               cgdebug_printf(state, "used: %s\n",
                                        arch_reg_str(i));
                        }
                }
-       }       
-#endif
-
-#warning "FIXME detect conflicts caused by the source and destination being the same register"
+       }
 
        /* If a color is already assigned see if it will work */
        if (range->color != REG_UNSET) {
@@ -12468,6 +19686,7 @@ static int select_free_color(struct compile_state *state,
                entry = lrd->def->use;
                for(;(range->color == REG_UNSET) && entry; entry = entry->next) {
                        struct live_range_def *insd;
+                       unsigned regcm;
                        insd = &rstate->lrd[entry->member->id];
                        if (insd->lr->defs == 0) {
                                continue;
@@ -12476,8 +19695,11 @@ static int select_free_color(struct compile_state *state,
                                !interfere(rstate, range, insd->lr)) {
                                phi = insd;
                        }
-                       if ((insd->lr->color == REG_UNSET) ||
-                               ((insd->lr->classes & range->classes) == 0) ||
+                       if (insd->lr->color == REG_UNSET) {
+                               continue;
+                       }
+                       regcm = insd->lr->classes;
+                       if (((regcm & range->classes) == 0) ||
                                (used[insd->lr->color])) {
                                continue;
                        }
@@ -12500,12 +19722,16 @@ static int select_free_color(struct compile_state *state,
                        expr = triple_rhs(state, phi->def, 0);
                        for(; expr; expr = triple_rhs(state, phi->def, expr)) {
                                struct live_range *lr;
+                               unsigned regcm;
                                if (!*expr) {
                                        continue;
                                }
                                lr = rstate->lrd[(*expr)->id].lr;
-                               if ((lr->color == REG_UNSET) || 
-                                       ((lr->classes & range->classes) == 0) ||
+                               if (lr->color == REG_UNSET) {
+                                       continue;
+                               }
+                               regcm = lr->classes;
+                               if (((regcm & range->classes) == 0) ||
                                        (used[lr->color])) {
                                        continue;
                                }
@@ -12522,12 +19748,16 @@ static int select_free_color(struct compile_state *state,
                expr = triple_rhs(state, lrd->def, 0);
                for(; expr; expr = triple_rhs(state, lrd->def, expr)) {
                        struct live_range *lr;
+                       unsigned regcm;
                        if (!*expr) {
                                continue;
                        }
                        lr = rstate->lrd[(*expr)->id].lr;
-                       if ((lr->color == -1) || 
-                               ((lr->classes & range->classes) == 0) ||
+                       if (lr->color == REG_UNSET) {
+                               continue;
+                       }
+                       regcm = lr->classes;
+                       if (((regcm & range->classes) == 0) ||
                                (used[lr->color])) {
                                continue;
                        }
@@ -12542,21 +19772,33 @@ static int select_free_color(struct compile_state *state,
         * pick the first color that is free.
         */
        if (range->color == REG_UNSET) {
-               range->color = 
+               range->color =
                        arch_select_free_register(state, used, range->classes);
        }
        if (range->color == REG_UNSET) {
+               struct live_range_def *lrd;
                int i;
                if (split_ranges(state, rstate, used, range)) {
                        return 0;
                }
                for(edge = range->edges; edge; edge = edge->next) {
-                       if (edge->node->color == REG_UNSET) {
-                               continue;
-                       }
-                       warning(state, edge->node->defs->def, "reg %s", 
+                       warning(state, edge->node->defs->def, "edge reg %s",
                                arch_reg_str(edge->node->color));
+                       lrd = edge->node->defs;
+                       do {
+                               warning(state, lrd->def, " %s %p",
+                                       tops(lrd->def->op), lrd->def);
+                               lrd = lrd->next;
+                       } while(lrd != edge->node->defs);
                }
+               warning(state, range->defs->def, "range: ");
+               lrd = range->defs;
+               do {
+                       warning(state, lrd->def, " %s %p",
+                               tops(lrd->def->op), lrd->def);
+                       lrd = lrd->next;
+               } while(lrd != range->defs);
+
                warning(state, range->defs->def, "classes: %x",
                        range->classes);
                for(i = 0; i < MAX_REGISTERS; i++) {
@@ -12565,14 +19807,10 @@ static int select_free_color(struct compile_state *state,
                                        arch_reg_str(i));
                        }
                }
-#if DEBUG_COLOR_GRAPH < 2
                error(state, range->defs->def, "too few registers");
-#else
-               internal_error(state, range->defs->def, "too few registers");
-#endif
        }
-       range->classes = arch_reg_regcm(state, range->color);
-       if (range->color == -1) {
+       range->classes &= arch_reg_regcm(state, range->color);
+       if ((range->color == REG_UNSET) || (range->classes == 0)) {
                internal_error(state, range->defs->def, "select_free_color did not?");
        }
        return 1;
@@ -12584,7 +19822,7 @@ static int color_graph(struct compile_state *state, struct reg_state *rstate)
        struct live_range_edge *edge;
        struct live_range *range;
        if (rstate->low) {
-               cgdebug_printf("Lo: ");
+               cgdebug_printf(state, "Lo: ");
                range = rstate->low;
                if (*range->group_prev != range) {
                        internal_error(state, 0, "lo: *prev != range?");
@@ -12601,7 +19839,7 @@ static int color_graph(struct compile_state *state, struct reg_state *rstate)
                }
        }
        else if (rstate->high) {
-               cgdebug_printf("Hi: ");
+               cgdebug_printf(state, "Hi: ");
                range = rstate->high;
                if (*range->group_prev != range) {
                        internal_error(state, 0, "hi: *prev != range?");
@@ -12620,7 +19858,7 @@ static int color_graph(struct compile_state *state, struct reg_state *rstate)
        else {
                return 1;
        }
-       cgdebug_printf(" %d\n", range - rstate->lr);
+       cgdebug_printf(state, " %d\n", range - rstate->lr);
        range->group_prev = 0;
        for(edge = range->edges; edge; edge = edge->next) {
                struct live_range *node;
@@ -12638,7 +19876,7 @@ static int color_graph(struct compile_state *state, struct reg_state *rstate)
                        if (&node->group_next == rstate->high_tail) {
                                rstate->high_tail = node->group_prev;
                        }
-                       cgdebug_printf("Moving...%d to low\n", node - rstate->lr);
+                       cgdebug_printf(state, "Moving...%d to low\n", node - rstate->lr);
                        node->group_prev  = rstate->low_tail;
                        node->group_next  = 0;
                        *rstate->low_tail = node;
@@ -12651,12 +19889,13 @@ static int color_graph(struct compile_state *state, struct reg_state *rstate)
        }
        colored = color_graph(state, rstate);
        if (colored) {
-               cgdebug_printf("Coloring %d @%s:%d.%d:", 
-                       range - rstate->lr,
-                       range->def->filename, range->def->line, range->def->col);
-               cgdebug_flush();
+               cgdebug_printf(state, "Coloring %d @", range - rstate->lr);
+               cgdebug_loc(state, range->defs->def);
+               cgdebug_flush(state);
                colored = select_free_color(state, rstate, range);
-               cgdebug_printf(" %s\n", arch_reg_str(range->color));
+               if (colored) {
+                       cgdebug_printf(state, " %s\n", arch_reg_str(range->color));
+               }
        }
        return colored;
 }
@@ -12667,12 +19906,12 @@ static void verify_colors(struct compile_state *state, struct reg_state *rstate)
        struct live_range_edge *edge;
        struct triple *ins, *first;
        char used[MAX_REGISTERS];
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                if (triple_is_def(state, ins)) {
                        if ((ins->id < 0) || (ins->id > rstate->defs)) {
-                               internal_error(state, ins, 
+                               internal_error(state, ins,
                                        "triple without a live range def");
                        }
                        lr = rstate->lrd[ins->id].lr;
@@ -12700,168 +19939,24 @@ static void verify_colors(struct compile_state *state, struct reg_state *rstate)
 
 static void color_triples(struct compile_state *state, struct reg_state *rstate)
 {
+       struct live_range_def *lrd;
        struct live_range *lr;
        struct triple *first, *ins;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                if ((ins->id < 0) || (ins->id > rstate->defs)) {
-                       internal_error(state, ins, 
+                       internal_error(state, ins,
                                "triple without a live range");
                }
-               lr = rstate->lrd[ins->id].lr;
+               lrd = &rstate->lrd[ins->id];
+               lr = lrd->lr;
+               ins->id = lrd->orig_id;
                SET_REG(ins->id, lr->color);
                ins = ins->next;
        } while (ins != first);
 }
 
-static void print_interference_block(
-       struct compile_state *state, struct block *block, void *arg)
-
-{
-       struct reg_state *rstate = arg;
-       struct reg_block *rb;
-       struct triple *ptr;
-       int phi_present;
-       int done;
-       rb = &rstate->blocks[block->vertex];
-
-       printf("\nblock: %p (%d), %p<-%p %p<-%p\n", 
-               block, 
-               block->vertex,
-               block->left, 
-               block->left && block->left->use?block->left->use->member : 0,
-               block->right, 
-               block->right && block->right->use?block->right->use->member : 0);
-       if (rb->in) {
-               struct triple_reg_set *in_set;
-               printf("        in:");
-               for(in_set = rb->in; in_set; in_set = in_set->next) {
-                       printf(" %-10p", in_set->member);
-               }
-               printf("\n");
-       }
-       phi_present = 0;
-       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
-               done = (ptr == block->last);
-               if (ptr->op == OP_PHI) {
-                       phi_present = 1;
-                       break;
-               }
-       }
-       if (phi_present) {
-               int edge;
-               for(edge = 0; edge < block->users; edge++) {
-                       printf("     in(%d):", edge);
-                       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
-                               struct triple **slot;
-                               done = (ptr == block->last);
-                               if (ptr->op != OP_PHI) {
-                                       continue;
-                               }
-                               slot = &RHS(ptr, 0);
-                               printf(" %-10p", slot[edge]);
-                       }
-                       printf("\n");
-               }
-       }
-       if (block->first->op == OP_LABEL) {
-               printf("%p:\n", block->first);
-       }
-       for(done = 0, ptr = block->first; !done; ptr = ptr->next) {
-               struct triple_set *user;
-               struct live_range *lr;
-               unsigned id;
-               int op;
-               op = ptr->op;
-               done = (ptr == block->last);
-               lr = rstate->lrd[ptr->id].lr;
-               
-               if (triple_stores_block(state, ptr)) {
-                       if (ptr->u.block != block) {
-                               internal_error(state, ptr, 
-                                       "Wrong block pointer: %p",
-                                       ptr->u.block);
-                       }
-               }
-               if (op == OP_ADECL) {
-                       for(user = ptr->use; user; user = user->next) {
-                               if (!user->member->u.block) {
-                                       internal_error(state, user->member, 
-                                               "Use %p not in a block?",
-                                               user->member);
-                               }
-                               
-                       }
-               }
-               id = ptr->id;
-               SET_REG(ptr->id, lr->color);
-               display_triple(stdout, ptr);
-               ptr->id = id;
-
-               if (triple_is_def(state, ptr) && (lr->defs == 0)) {
-                       internal_error(state, ptr, "lr has no defs!");
-               }
-
-               if (lr->defs) {
-                       struct live_range_def *lrd;
-                       printf("       range:");
-                       lrd = lr->defs;
-                       do {
-                               printf(" %-10p", lrd->def);
-                               lrd = lrd->next;
-                       } while(lrd != lr->defs);
-                       printf("\n");
-               }
-               if (lr->edges > 0) {
-                       struct live_range_edge *edge;
-                       printf("       edges:");
-                       for(edge = lr->edges; edge; edge = edge->next) {
-                               struct live_range_def *lrd;
-                               lrd = edge->node->defs;
-                               do {
-                                       printf(" %-10p", lrd->def);
-                                       lrd = lrd->next;
-                               } while(lrd != edge->node->defs);
-                               printf("|");
-                       }
-                       printf("\n");
-               }
-               /* Do a bunch of sanity checks */
-               valid_ins(state, ptr);
-               if ((ptr->id < 0) || (ptr->id > rstate->defs)) {
-                       internal_error(state, ptr, "Invalid triple id: %d",
-                               ptr->id);
-               }
-               for(user = ptr->use; user; user = user->next) {
-                       struct triple *use;
-                       struct live_range *ulr;
-                       use = user->member;
-                       valid_ins(state, use);
-                       if ((use->id < 0) || (use->id > rstate->defs)) {
-                               internal_error(state, use, "Invalid triple id: %d",
-                                       use->id);
-                       }
-                       ulr = rstate->lrd[user->member->id].lr;
-                       if (triple_stores_block(state, user->member) &&
-                               !user->member->u.block) {
-                               internal_error(state, user->member,
-                                       "Use %p not in a block?",
-                                       user->member);
-                       }
-               }
-       }
-       if (rb->out) {
-               struct triple_reg_set *out_set;
-               printf("       out:");
-               for(out_set = rb->out; out_set; out_set = out_set->next) {
-                       printf(" %-10p", out_set->member);
-               }
-               printf("\n");
-       }
-       printf("\n");
-}
-
 static struct live_range *merge_sort_lr(
        struct live_range *first, struct live_range *last)
 {
@@ -12872,7 +19967,7 @@ static struct live_range *merge_sort_lr(
                mid = first + size/2;
                first = merge_sort_lr(first, mid -1);
                mid   = merge_sort_lr(mid, last);
-               
+
                join = 0;
                join_tail = &join;
                /* merge the two lists */
@@ -12901,7 +19996,7 @@ static struct live_range *merge_sort_lr(
                /* Splice the remaining list */
                pick = (first)? first : mid;
                *join_tail = pick;
-               if (pick) { 
+               if (pick) {
                        pick->group_prev = join_tail;
                }
        }
@@ -12914,7 +20009,7 @@ static struct live_range *merge_sort_lr(
        return join;
 }
 
-static void ids_from_rstate(struct compile_state *state, 
+static void ids_from_rstate(struct compile_state *state,
        struct reg_state *rstate)
 {
        struct triple *ins, *first;
@@ -12922,11 +20017,13 @@ static void ids_from_rstate(struct compile_state *state,
                return;
        }
        /* Display the graph if desired */
-       if (state->debug & DEBUG_INTERFERENCE) {
-               print_blocks(state, stdout);
-               print_control_flow(state);
+       if (state->compiler->debug & DEBUG_INTERFERENCE) {
+               FILE *fp = state->dbgout;
+               print_interference_blocks(state, rstate, fp, 0);
+               print_control_flow(state, fp, &state->bb);
+               fflush(fp);
        }
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                if (ins->id) {
@@ -12955,7 +20052,7 @@ static void cleanup_rstate(struct compile_state *state, struct reg_state *rstate
 
        /* Free the variable lifetime information */
        if (rstate->blocks) {
-               free_variable_lifetimes(state, rstate->blocks);
+               free_variable_lifetimes(state, &state->bb, rstate->blocks);
        }
        rstate->defs = 0;
        rstate->ranges = 0;
@@ -12964,6 +20061,7 @@ static void cleanup_rstate(struct compile_state *state, struct reg_state *rstate
        rstate->blocks = 0;
 }
 
+static void verify_consistency(struct compile_state *state);
 static void allocate_registers(struct compile_state *state)
 {
        struct reg_state rstate;
@@ -12971,85 +20069,95 @@ static void allocate_registers(struct compile_state *state)
 
        /* Clear out the reg_state */
        memset(&rstate, 0, sizeof(rstate));
-       rstate.max_passes = MAX_ALLOCATION_PASSES;
+       rstate.max_passes = state->compiler->max_allocation_passes;
 
        do {
                struct live_range **point, **next;
-               struct triple *tangle;
-               struct coalesce_conflict conflict;
+               int tangles;
                int coalesced;
 
+               if (state->compiler->debug & DEBUG_RANGE_CONFLICTS) {
+                       FILE *fp = state->errout;
+                       fprintf(fp, "pass: %d\n", rstate.passes);
+                       fflush(fp);
+               }
+
                /* Restore ids */
                ids_from_rstate(state, &rstate);
 
-               do {
-                       /* Cleanup the temporary data structures */
-                       cleanup_rstate(state, &rstate);
+               /* Cleanup the temporary data structures */
+               cleanup_rstate(state, &rstate);
 
-                       /* Compute the variable lifetimes */
-                       rstate.blocks = compute_variable_lifetimes(state);
+               /* Compute the variable lifetimes */
+               rstate.blocks = compute_variable_lifetimes(state, &state->bb);
 
-                       /* Find an invalid mandatory live range coalesce */
-                       conflict.ins = 0;
-                       conflict.index = -1;
-                       walk_variable_lifetimes(
-                               state, rstate.blocks, spot_coalesce_conflict, &conflict);
-                       
-                       /* If a tangle was found resolve it */
-                       if (conflict.ins) {
-                               resolve_coalesce_conflict(state, &conflict);
-                       }
-               } while(conflict.ins);
+               /* Fix invalid mandatory live range coalesce conflicts */
+               correct_coalesce_conflicts(state, rstate.blocks);
 
+               /* Fix two simultaneous uses of the same register.
+                * In a few pathlogical cases a partial untangle moves
+                * the tangle to a part of the graph we won't revisit.
+                * So we keep looping until we have no more tangle fixes
+                * to apply.
+                */
                do {
-                       /* Cleanup the temporary data structures */
-                       cleanup_rstate(state, &rstate);
-
-                       /* Compute the variable lifetimes */
-                       rstate.blocks = compute_variable_lifetimes(state);
+                       tangles = correct_tangles(state, rstate.blocks);
+               } while(tangles);
 
-                       /* Find two simultaneous uses of the same register */
-                       tangle = 0;
-                       walk_variable_lifetimes(
-                               state, rstate.blocks, spot_tangle, &tangle);
-                       
-                       /* If a tangle was found resolve it */
-                       if (tangle) {
-                               resolve_tangle(state, tangle);
-                       }
-               } while(tangle);
 
-               if (state->debug & DEBUG_INSERTED_COPIES) {
-                       printf("After resolve_tangles\n");
-                       print_blocks(state, stdout);
-                       print_control_flow(state);
-               }
+               print_blocks(state, "resolve_tangles", state->dbgout);
+               verify_consistency(state);
 
-               
                /* Allocate and initialize the live ranges */
                initialize_live_ranges(state, &rstate);
-               
+
+               /* Note currently doing coalescing in a loop appears to
+                * buys me nothing.  The code is left this way in case
+                * there is some value in it.  Or if a future bugfix
+                * yields some benefit.
+                */
                do {
-                       /* Forget previous live range edge calculations */
+                       if (state->compiler->debug & DEBUG_COALESCING) {
+                               fprintf(state->errout, "coalescing\n");
+                       }
+
+                       /* Remove any previous live edge calculations */
                        cleanup_live_edges(&rstate);
 
                        /* Compute the interference graph */
                        walk_variable_lifetimes(
-                               state, rstate.blocks, graph_ins, &rstate);
-               
+                               state, &state->bb, rstate.blocks,
+                               graph_ins, &rstate);
+
                        /* Display the interference graph if desired */
-                       if (state->debug & DEBUG_INTERFERENCE) {
-                               printf("\nlive variables by block\n");
-                               walk_blocks(state, print_interference_block, &rstate);
-                               printf("\nlive variables by instruction\n");
+                       if (state->compiler->debug & DEBUG_INTERFERENCE) {
+                               print_interference_blocks(state, &rstate, state->dbgout, 1);
+                               fprintf(state->dbgout, "\nlive variables by instruction\n");
                                walk_variable_lifetimes(
-                                       state, rstate.blocks, 
+                                       state, &state->bb, rstate.blocks,
                                        print_interference_ins, &rstate);
                        }
-                       
+
                        coalesced = coalesce_live_ranges(state, &rstate);
+
+                       if (state->compiler->debug & DEBUG_COALESCING) {
+                               fprintf(state->errout, "coalesced: %d\n", coalesced);
+                       }
                } while(coalesced);
-                       
+
+#if DEBUG_CONSISTENCY > 1
+# if 0
+               fprintf(state->errout, "verify_graph_ins...\n");
+# endif
+               /* Verify the interference graph */
+               walk_variable_lifetimes(
+                       state, &state->bb, rstate.blocks,
+                       verify_graph_ins, &rstate);
+# if 0
+               fprintf(state->errout, "verify_graph_ins done\n");
+#endif
+#endif
+
                /* Build the groups low and high.  But with the nodes
                 * first sorted by degree order.
                 */
@@ -13069,13 +20177,13 @@ static void allocate_registers(struct compile_state *state)
                        struct live_range *range;
                        next = &(*point)->group_next;
                        range = *point;
-                       
+
                        /* If it has a low degree or it already has a color
                         * place the node in low.
                         */
                        if ((range->degree < regc_max_size(state, range->classes)) ||
                                (range->color != REG_UNSET)) {
-                               cgdebug_printf("Lo: %5d degree %5d%s\n", 
+                               cgdebug_printf(state, "Lo: %5d degree %5d%s\n",
                                        range - rstate.lr, range->degree,
                                        (range->color != REG_UNSET) ? " (colored)": "");
                                *range->group_prev = range->group_next;
@@ -13092,7 +20200,7 @@ static void allocate_registers(struct compile_state *state)
                                next = point;
                        }
                        else {
-                               cgdebug_printf("hi: %5d degree %5d%s\n", 
+                               cgdebug_printf(state, "hi: %5d degree %5d%s\n",
                                        range - rstate.lr, range->degree,
                                        (range->color != REG_UNSET) ? " (colored)": "");
                        }
@@ -13110,6 +20218,9 @@ static void allocate_registers(struct compile_state *state)
 
        /* Cleanup the temporary data structures */
        cleanup_rstate(state, &rstate);
+
+       /* Display the new graph */
+       print_blocks(state, __func__, state->dbgout);
 }
 
 /* Sparce Conditional Constant Propogation
@@ -13123,9 +20234,9 @@ struct lattice_node {
        struct ssa_edge *out;
        struct flow_block *fblock;
        struct triple *val;
-       /* lattice high   val && !is_const(val) 
+       /* lattice high   val == def
         * lattice const  is_const(val)
-        * lattice low    val == 0
+        * lattice low    other
         */
 };
 struct ssa_edge {
@@ -13144,11 +20255,12 @@ struct flow_edge {
        struct flow_edge *out_next;
        int executable;
 };
+#define MAX_FLOW_BLOCK_EDGES 3
 struct flow_block {
        struct block *block;
        struct flow_edge *in;
        struct flow_edge *out;
-       struct flow_edge left, right;
+       struct flow_edge *edges;
 };
 
 struct scc_state {
@@ -13161,9 +20273,45 @@ struct scc_state {
 };
 
 
-static void scc_add_fedge(struct compile_state *state, struct scc_state *scc, 
+static int is_scc_const(struct compile_state *state, struct triple *ins)
+{
+       return ins && (triple_is_ubranch(state, ins) || is_const(ins));
+}
+
+static int is_lattice_hi(struct compile_state *state, struct lattice_node *lnode)
+{
+       return !is_scc_const(state, lnode->val) && (lnode->val == lnode->def);
+}
+
+static int is_lattice_const(struct compile_state *state, struct lattice_node *lnode)
+{
+       return is_scc_const(state, lnode->val);
+}
+
+static int is_lattice_lo(struct compile_state *state, struct lattice_node *lnode)
+{
+       return (lnode->val != lnode->def) && !is_scc_const(state, lnode->val);
+}
+
+static void scc_add_fedge(struct compile_state *state, struct scc_state *scc,
        struct flow_edge *fedge)
 {
+       if (state->compiler->debug & DEBUG_SCC_TRANSFORM2) {
+               fprintf(state->errout, "adding fedge: %p (%4d -> %5d)\n",
+                       fedge,
+                       fedge->src->block?fedge->src->block->last->id: 0,
+                       fedge->dst->block?fedge->dst->block->first->id: 0);
+       }
+       if ((fedge == scc->flow_work_list) ||
+               (fedge->work_next != fedge) ||
+               (fedge->work_prev != fedge)) {
+
+               if (state->compiler->debug & DEBUG_SCC_TRANSFORM2) {
+                       fprintf(state->errout, "dupped fedge: %p\n",
+                               fedge);
+               }
+               return;
+       }
        if (!scc->flow_work_list) {
                scc->flow_work_list = fedge;
                fedge->work_next = fedge->work_prev = fedge;
@@ -13191,6 +20339,7 @@ static struct flow_edge *scc_next_fedge(
                } else {
                        scc->flow_work_list = 0;
                }
+               fedge->work_next = fedge->work_prev = fedge;
        }
        return fedge;
 }
@@ -13198,6 +20347,22 @@ static struct flow_edge *scc_next_fedge(
 static void scc_add_sedge(struct compile_state *state, struct scc_state *scc,
        struct ssa_edge *sedge)
 {
+       if (state->compiler->debug & DEBUG_SCC_TRANSFORM2) {
+               fprintf(state->errout, "adding sedge: %5ld (%4d -> %5d)\n",
+                       (long)(sedge - scc->ssa_edges),
+                       sedge->src->def->id,
+                       sedge->dst->def->id);
+       }
+       if ((sedge == scc->ssa_work_list) ||
+               (sedge->work_next != sedge) ||
+               (sedge->work_prev != sedge)) {
+
+               if (state->compiler->debug & DEBUG_SCC_TRANSFORM2) {
+                       fprintf(state->errout, "dupped sedge: %5ld\n",
+                               (long)(sedge - scc->ssa_edges));
+               }
+               return;
+       }
        if (!scc->ssa_work_list) {
                scc->ssa_work_list = sedge;
                sedge->work_next = sedge->work_prev = sedge;
@@ -13225,6 +20390,7 @@ static struct ssa_edge *scc_next_sedge(
                } else {
                        scc->ssa_work_list = 0;
                }
+               sedge->work_next = sedge->work_prev = sedge;
        }
        return sedge;
 }
@@ -13241,7 +20407,7 @@ static void initialize_scc_state(
        memset(scc, 0, sizeof(*scc));
 
        /* Inialize pass zero find out how much memory we need */
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        ins_count = ssa_edge_count = 0;
        do {
@@ -13252,17 +20418,17 @@ static void initialize_scc_state(
                }
                ins = ins->next;
        } while(ins != first);
-#if DEBUG_SCC
-       fprintf(stderr, "ins_count: %d ssa_edge_count: %d vertex_count: %d\n",
-               ins_count, ssa_edge_count, state->last_vertex);
-#endif
+       if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+               fprintf(state->errout, "ins_count: %d ssa_edge_count: %d vertex_count: %d\n",
+                       ins_count, ssa_edge_count, state->bb.last_vertex);
+       }
        scc->ins_count   = ins_count;
-       scc->lattice     = 
+       scc->lattice     =
                xcmalloc(sizeof(*scc->lattice)*(ins_count + 1), "lattice");
-       scc->ssa_edges   = 
+       scc->ssa_edges   =
                xcmalloc(sizeof(*scc->ssa_edges)*(ssa_edge_count + 1), "ssa_edges");
-       scc->flow_blocks = 
-               xcmalloc(sizeof(*scc->flow_blocks)*(state->last_vertex + 1), 
+       scc->flow_blocks =
+               xcmalloc(sizeof(*scc->flow_blocks)*(state->bb.last_vertex + 1),
                        "flow_blocks");
 
        /* Initialize pass one collect up the nodes */
@@ -13280,6 +20446,8 @@ static void initialize_scc_state(
                        block->vertex = fblock_index;
                        fblock = &scc->flow_blocks[fblock_index];
                        fblock->block = block;
+                       fblock->edges = xcmalloc(sizeof(*fblock->edges)*block->edge_count,
+                               "flow_edges");
                }
                {
                        struct lattice_node *lnode;
@@ -13289,6 +20457,9 @@ static void initialize_scc_state(
                        lnode->out = 0;
                        lnode->fblock = fblock;
                        lnode->val = ins; /* LATTICE HIGH */
+                       if (lnode->val->op == OP_UNKNOWNVAL) {
+                               lnode->val = 0; /* LATTICE LOW by definition */
+                       }
                        lnode->old_id = ins->id;
                        ins->id = ins_index;
                }
@@ -13299,6 +20470,25 @@ static void initialize_scc_state(
        fblock = 0;
        ins = first;
        do {
+               {
+                       struct triple_set *edge;
+                       struct ssa_edge **stail;
+                       struct lattice_node *lnode;
+                       lnode = &scc->lattice[ins->id];
+                       lnode->out = 0;
+                       stail = &lnode->out;
+                       for(edge = ins->use; edge; edge = edge->next) {
+                               struct ssa_edge *sedge;
+                               ssa_edge_index += 1;
+                               sedge = &scc->ssa_edges[ssa_edge_index];
+                               *stail = sedge;
+                               stail = &sedge->out_next;
+                               sedge->src = lnode;
+                               sedge->dst = &scc->lattice[edge->member->id];
+                               sedge->work_next = sedge->work_prev = sedge;
+                               sedge->out_next = 0;
+                       }
+               }
                if ((ins->op == OP_LABEL) && (block != ins->u.block)) {
                        struct flow_edge *fedge, **ftail;
                        struct block_set *bedge;
@@ -13307,64 +20497,53 @@ static void initialize_scc_state(
                        fblock->in = 0;
                        fblock->out = 0;
                        ftail = &fblock->out;
-                       if (block->left) {
-                               fblock->left.dst = &scc->flow_blocks[block->left->vertex];
-                               if (fblock->left.dst->block != block->left) {
-                                       internal_error(state, 0, "block mismatch");
-                               }
-                               fblock->left.out_next = 0;
-                               *ftail = &fblock->left;
-                               ftail = &fblock->left.out_next;
-                       }
-                       if (block->right) {
-                               fblock->right.dst = &scc->flow_blocks[block->right->vertex];
-                               if (fblock->right.dst->block != block->right) {
+
+                       fedge = fblock->edges;
+                       bedge = block->edges;
+                       for(; bedge; bedge = bedge->next, fedge++) {
+                               fedge->dst = &scc->flow_blocks[bedge->member->vertex];
+                               if (fedge->dst->block != bedge->member) {
                                        internal_error(state, 0, "block mismatch");
                                }
-                               fblock->right.out_next = 0;
-                               *ftail = &fblock->right;
-                               ftail = &fblock->right.out_next;
+                               *ftail = fedge;
+                               ftail = &fedge->out_next;
+                               fedge->out_next = 0;
                        }
                        for(fedge = fblock->out; fedge; fedge = fedge->out_next) {
                                fedge->src = fblock;
                                fedge->work_next = fedge->work_prev = fedge;
                                fedge->executable = 0;
                        }
+               }
+               ins = ins->next;
+       } while (ins != first);
+       block = 0;
+       fblock = 0;
+       ins = first;
+       do {
+               if ((ins->op  == OP_LABEL) && (block != ins->u.block)) {
+                       struct flow_edge **ftail;
+                       struct block_set *bedge;
+                       block = ins->u.block;
+                       fblock = &scc->flow_blocks[block->vertex];
                        ftail = &fblock->in;
                        for(bedge = block->use; bedge; bedge = bedge->next) {
-                               struct block *src_block;
-                               struct flow_block *sfblock;
-                               struct flow_edge *sfedge;
-                               src_block = bedge->member;
-                               sfblock = &scc->flow_blocks[src_block->vertex];
-                               sfedge = 0;
-                               if (src_block->left == block) {
-                                       sfedge = &sfblock->left;
-                               } else {
-                                       sfedge = &sfblock->right;
-                               }
-                               *ftail = sfedge;
-                               ftail = &sfedge->in_next;
-                               sfedge->in_next = 0;
-                       }
-               }
-               {
-                       struct triple_set *edge;
-                       struct ssa_edge **stail;
-                       struct lattice_node *lnode;
-                       lnode = &scc->lattice[ins->id];
-                       lnode->out = 0;
-                       stail = &lnode->out;
-                       for(edge = ins->use; edge; edge = edge->next) {
-                               struct ssa_edge *sedge;
-                               ssa_edge_index += 1;
-                               sedge = &scc->ssa_edges[ssa_edge_index];
-                               *stail = sedge;
-                               stail = &sedge->out_next;
-                               sedge->src = lnode;
-                               sedge->dst = &scc->lattice[edge->member->id];
-                               sedge->work_next = sedge->work_prev = sedge;
-                               sedge->out_next = 0;
+                               struct block *src_block;
+                               struct flow_block *sfblock;
+                               struct flow_edge *sfedge;
+                               src_block = bedge->member;
+                               sfblock = &scc->flow_blocks[src_block->vertex];
+                               for(sfedge = sfblock->out; sfedge; sfedge = sfedge->out_next) {
+                                       if (sfedge->dst == fblock) {
+                                               break;
+                                       }
+                               }
+                               if (!sfedge) {
+                                       internal_error(state, 0, "edge mismatch");
+                               }
+                               *ftail = sfedge;
+                               ftail = &sfedge->in_next;
+                               sfedge->in_next = 0;
                        }
                }
                ins = ins->next;
@@ -13375,10 +20554,11 @@ static void initialize_scc_state(
                struct flow_edge *fedge;
                fblock = &scc->flow_blocks[0];
                fblock->block = 0;
+               fblock->edges = xcmalloc(sizeof(*fblock->edges)*1, "flow_edges");
                fblock->in = 0;
-               fblock->out = &fblock->left;
-               dst = &scc->flow_blocks[state->first_block->vertex];
-               fedge = &fblock->left;
+               fblock->out = fblock->edges;
+               dst = &scc->flow_blocks[state->bb.first_block->vertex];
+               fedge = fblock->edges;
                fedge->src        = fblock;
                fedge->dst        = dst;
                fedge->work_next  = fedge;
@@ -13387,26 +20567,35 @@ static void initialize_scc_state(
                fedge->out_next   = 0;
                fedge->executable = 0;
                fedge->dst->in = fedge;
-               
+
                /* Initialize the work lists */
                scc->flow_work_list = 0;
                scc->ssa_work_list  = 0;
                scc_add_fedge(state, scc, fedge);
        }
-#if DEBUG_SCC
-       fprintf(stderr, "ins_index: %d ssa_edge_index: %d fblock_index: %d\n",
-               ins_index, ssa_edge_index, fblock_index);
-#endif
+       if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+               fprintf(state->errout, "ins_index: %d ssa_edge_index: %d fblock_index: %d\n",
+                       ins_index, ssa_edge_index, fblock_index);
+       }
 }
 
-       
+
 static void free_scc_state(
        struct compile_state *state, struct scc_state *scc)
 {
+       int i;
+       for(i = 0; i < state->bb.last_vertex + 1; i++) {
+               struct flow_block *fblock;
+               fblock = &scc->flow_blocks[i];
+               if (fblock->edges) {
+                       xfree(fblock->edges);
+                       fblock->edges = 0;
+               }
+       }
        xfree(scc->flow_blocks);
        xfree(scc->ssa_edges);
        xfree(scc->lattice);
-       
+
 }
 
 static struct lattice_node *triple_to_lattice(
@@ -13435,7 +20624,7 @@ static struct triple *preserve_lval(
        return old;
 }
 
-static int lval_changed(struct compile_state *state, 
+static int lval_changed(struct compile_state *state,
        struct triple *old, struct lattice_node *lnode)
 {
        int changed;
@@ -13444,13 +20633,10 @@ static int lval_changed(struct compile_state *state,
        if (!old && !lnode->val) {
                changed = 0;
        }
-       if (changed && lnode->val && !is_const(lnode->val)) {
-               changed = 0;
-       }
        if (changed &&
                lnode->val && old &&
                (memcmp(lnode->val->param, old->param,
-                       TRIPLE_SIZE(lnode->val->sizes) * sizeof(lnode->val->param[0])) == 0) &&
+                       TRIPLE_SIZE(lnode->val) * sizeof(lnode->val->param[0])) == 0) &&
                (memcmp(&lnode->val->u, &old->u, sizeof(old->u)) == 0)) {
                changed = 0;
        }
@@ -13461,63 +20647,36 @@ static int lval_changed(struct compile_state *state,
 
 }
 
-static void scc_visit_phi(struct compile_state *state, struct scc_state *scc, 
-       struct lattice_node *lnode)
+static void scc_debug_lnode(
+       struct compile_state *state, struct scc_state *scc,
+       struct lattice_node *lnode, int changed)
 {
-       struct lattice_node *tmp;
-       struct triple **slot, *old;
-       struct flow_edge *fedge;
-       int index;
-       if (lnode->def->op != OP_PHI) {
-               internal_error(state, lnode->def, "not phi");
+       if ((state->compiler->debug & DEBUG_SCC_TRANSFORM2) && lnode->val) {
+               display_triple_changes(state->errout, lnode->val, lnode->def);
        }
-       /* Store the original value */
-       old = preserve_lval(state, lnode);
-
-       /* default to lattice high */
-       lnode->val = lnode->def;
-       slot = &RHS(lnode->def, 0);
-       index = 0;
-       for(fedge = lnode->fblock->in; fedge; index++, fedge = fedge->in_next) {
-               if (!fedge->executable) {
-                       continue;
-               }
-               if (!slot[index]) {
-                       internal_error(state, lnode->def, "no phi value");
-               }
-               tmp = triple_to_lattice(state, scc, slot[index]);
-               /* meet(X, lattice low) = lattice low */
-               if (!tmp->val) {
-                       lnode->val = 0;
-               }
-               /* meet(X, lattice high) = X */
-               else if (!tmp->val) {
-                       lnode->val = lnode->val;
-               }
-               /* meet(lattice high, X) = X */
-               else if (!is_const(lnode->val)) {
-                       lnode->val = dup_triple(state, tmp->val);
-                       lnode->val->type = lnode->def->type;
-               }
-               /* meet(const, const) = const or lattice low */
-               else if (!constants_equal(state, lnode->val, tmp->val)) {
-                       lnode->val = 0;
-               }
-               if (!lnode->val) {
-                       break;
+       if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+               FILE *fp = state->errout;
+               struct triple *val, **expr;
+               val = lnode->val? lnode->val : lnode->def;
+               fprintf(fp, "%p %s %3d %10s (",
+                       lnode->def,
+                       ((lnode->def->op == OP_PHI)? "phi: ": "expr:"),
+                       lnode->def->id,
+                       tops(lnode->def->op));
+               expr = triple_rhs(state, lnode->def, 0);
+               for(;expr;expr = triple_rhs(state, lnode->def, expr)) {
+                       if (*expr) {
+                               fprintf(fp, " %d", (*expr)->id);
+                       }
                }
-       }
-#if DEBUG_SCC
-       fprintf(stderr, "phi: %d -> %s\n",
-               lnode->def->id,
-               (!lnode->val)? "lo": is_const(lnode->val)? "const": "hi");
-#endif
-       /* If the lattice value has changed update the work lists. */
-       if (lval_changed(state, old, lnode)) {
-               struct ssa_edge *sedge;
-               for(sedge = lnode->out; sedge; sedge = sedge->out_next) {
-                       scc_add_sedge(state, scc, sedge);
+               if (val->op == OP_INTCONST) {
+                       fprintf(fp, " <0x%08lx>", (unsigned long)(val->u.cval));
                }
+               fprintf(fp, " ) -> %s %s\n",
+                       (is_lattice_hi(state, lnode)? "hi":
+                               is_lattice_const(state, lnode)? "const" : "lo"),
+                       changed? "changed" : ""
+                       );
        }
 }
 
@@ -13528,7 +20687,7 @@ static int compute_lnode_val(struct compile_state *state, struct scc_state *scc,
        struct triple *old, *scratch;
        struct triple **dexpr, **vexpr;
        int count, i;
-       
+
        /* Store the original value */
        old = preserve_lval(state, lnode);
 
@@ -13539,24 +20698,26 @@ static int compute_lnode_val(struct compile_state *state, struct scc_state *scc,
        scratch->prev     = scratch;
        scratch->use      = 0;
 
-       count = TRIPLE_SIZE(scratch->sizes);
+       count = TRIPLE_SIZE(scratch);
        for(i = 0; i < count; i++) {
                dexpr = &lnode->def->param[i];
                vexpr = &scratch->param[i];
                *vexpr = *dexpr;
-               if (((i < TRIPLE_MISC_OFF(scratch->sizes)) ||
-                       (i >= TRIPLE_TARG_OFF(scratch->sizes))) &&
+               if (((i < TRIPLE_MISC_OFF(scratch)) ||
+                       (i >= TRIPLE_TARG_OFF(scratch))) &&
                        *dexpr) {
                        struct lattice_node *tmp;
                        tmp = triple_to_lattice(state, scc, *dexpr);
                        *vexpr = (tmp->val)? tmp->val : tmp->def;
                }
        }
-       if (scratch->op == OP_BRANCH) {
+       if (triple_is_branch(state, scratch)) {
                scratch->next = lnode->def->next;
        }
        /* Recompute the value */
+#if DEBUG_ROMCC_WARNINGS
 #warning "FIXME see if simplify does anything bad"
+#endif
        /* So far it looks like only the strength reduction
         * optimization are things I need to worry about.
         */
@@ -13567,134 +20728,246 @@ static int compute_lnode_val(struct compile_state *state, struct scc_state *scc,
        }
        if ((scratch->prev != scratch) ||
                ((scratch->next != scratch) &&
-                       ((lnode->def->op != OP_BRANCH) ||
+                       (!triple_is_branch(state, lnode->def) ||
                                (scratch->next != lnode->def->next)))) {
                internal_error(state, lnode->def, "scratch in list?");
        }
        /* undo any uses... */
-       count = TRIPLE_SIZE(scratch->sizes);
+       count = TRIPLE_SIZE(scratch);
        for(i = 0; i < count; i++) {
                vexpr = &scratch->param[i];
                if (*vexpr) {
                        unuse_triple(*vexpr, scratch);
                }
        }
-       if (!is_const(scratch)) {
-               for(i = 0; i < count; i++) {
-                       dexpr = &lnode->def->param[i];
-                       if (((i < TRIPLE_MISC_OFF(scratch->sizes)) ||
-                               (i >= TRIPLE_TARG_OFF(scratch->sizes))) &&
-                               *dexpr) {
-                               struct lattice_node *tmp;
-                               tmp = triple_to_lattice(state, scc, *dexpr);
-                               if (!tmp->val) {
-                                       lnode->val = 0;
-                               }
-                       }
-               }
+       if (lnode->val->op == OP_UNKNOWNVAL) {
+               lnode->val = 0; /* Lattice low by definition */
        }
-       if (lnode->val && 
+       /* Find the case when I am lattice high */
+       if (lnode->val &&
                (lnode->val->op == lnode->def->op) &&
-               (memcmp(lnode->val->param, lnode->def->param, 
+               (memcmp(lnode->val->param, lnode->def->param,
                        count * sizeof(lnode->val->param[0])) == 0) &&
                (memcmp(&lnode->val->u, &lnode->def->u, sizeof(lnode->def->u)) == 0)) {
                lnode->val = lnode->def;
        }
+       /* Only allow lattice high when all of my inputs
+        * are also lattice high.  Occassionally I can
+        * have constants with a lattice low input, so
+        * I do not need to check that case.
+        */
+       if (is_lattice_hi(state, lnode)) {
+               struct lattice_node *tmp;
+               int rhs;
+               rhs = lnode->val->rhs;
+               for(i = 0; i < rhs; i++) {
+                       tmp = triple_to_lattice(state, scc, RHS(lnode->val, i));
+                       if (!is_lattice_hi(state, tmp)) {
+                               lnode->val = 0;
+                               break;
+                       }
+               }
+       }
        /* Find the cases that are always lattice lo */
-       if (lnode->val && 
+       if (lnode->val &&
                triple_is_def(state, lnode->val) &&
-               !triple_is_pure(state, lnode->val)) {
+               !triple_is_pure(state, lnode->val, lnode->old_id)) {
                lnode->val = 0;
        }
-       if (lnode->val && 
-               (lnode->val->op == OP_SDECL) && 
-               (lnode->val != lnode->def)) {
-               internal_error(state, lnode->def, "bad sdecl");
-       }
        /* See if the lattice value has changed */
        changed = lval_changed(state, old, lnode);
+       /* See if this value should not change */
+       if ((lnode->val != lnode->def) &&
+               ((      !triple_is_def(state, lnode->def)  &&
+                       !triple_is_cbranch(state, lnode->def)) ||
+                       (lnode->def->op == OP_PIECE))) {
+#if DEBUG_ROMCC_WARNINGS
+#warning "FIXME constant propogate through expressions with multiple left hand sides"
+#endif
+               if (changed) {
+                       internal_warning(state, lnode->def, "non def changes value?");
+               }
+               lnode->val = 0;
+       }
+
+       /* See if we need to free the scratch value */
        if (lnode->val != scratch) {
                xfree(scratch);
        }
+
        return changed;
 }
 
-static void scc_visit_branch(struct compile_state *state, struct scc_state *scc,
+
+static void scc_visit_cbranch(struct compile_state *state, struct scc_state *scc,
        struct lattice_node *lnode)
 {
        struct lattice_node *cond;
-#if DEBUG_SCC
-       {
+       struct flow_edge *left, *right;
+       int changed;
+
+       /* Update the branch value */
+       changed = compute_lnode_val(state, scc, lnode);
+       scc_debug_lnode(state, scc, lnode, changed);
+
+       /* This only applies to conditional branches */
+       if (!triple_is_cbranch(state, lnode->def)) {
+               internal_error(state, lnode->def, "not a conditional branch");
+       }
+
+       if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
                struct flow_edge *fedge;
-               fprintf(stderr, "branch: %d (",
+               FILE *fp = state->errout;
+               fprintf(fp, "%s: %d (",
+                       tops(lnode->def->op),
                        lnode->def->id);
-               
+
                for(fedge = lnode->fblock->out; fedge; fedge = fedge->out_next) {
-                       fprintf(stderr, " %d", fedge->dst->block->vertex);
+                       fprintf(fp, " %d", fedge->dst->block->vertex);
                }
-               fprintf(stderr, " )");
-               if (TRIPLE_RHS(lnode->def->sizes) > 0) {
-                       fprintf(stderr, " <- %d",
+               fprintf(fp, " )");
+               if (lnode->def->rhs > 0) {
+                       fprintf(fp, " <- %d",
                                RHS(lnode->def, 0)->id);
                }
-               fprintf(stderr, "\n");
+               fprintf(fp, "\n");
+       }
+       cond = triple_to_lattice(state, scc, RHS(lnode->def,0));
+       for(left = cond->fblock->out; left; left = left->out_next) {
+               if (left->dst->block->first == lnode->def->next) {
+                       break;
+               }
        }
-#endif
-       if (lnode->def->op != OP_BRANCH) {
-               internal_error(state, lnode->def, "not branch");
+       if (!left) {
+               internal_error(state, lnode->def, "Cannot find left branch edge");
        }
-       /* This only applies to conditional branches */
-       if (TRIPLE_RHS(lnode->def->sizes) == 0) {
-               return;
+       for(right = cond->fblock->out; right; right = right->out_next) {
+               if (right->dst->block->first == TARG(lnode->def, 0)) {
+                       break;
+               }
        }
-       cond = triple_to_lattice(state, scc, RHS(lnode->def,0));
-       if (cond->val && !is_const(cond->val)) {
-#warning "FIXME do I need to do something here?"
-               warning(state, cond->def, "condition not constant?");
+       if (!right) {
+               internal_error(state, lnode->def, "Cannot find right branch edge");
+       }
+       /* I should only come here if the controlling expressions value
+        * has changed, which means it must be either a constant or lo.
+        */
+       if (is_lattice_hi(state, cond)) {
+               internal_error(state, cond->def, "condition high?");
                return;
        }
-       if (cond->val == 0) {
-               scc_add_fedge(state, scc, cond->fblock->out);
-               scc_add_fedge(state, scc, cond->fblock->out->out_next);
+       if (is_lattice_lo(state, cond)) {
+               scc_add_fedge(state, scc, left);
+               scc_add_fedge(state, scc, right);
        }
        else if (cond->val->u.cval) {
-               scc_add_fedge(state, scc, cond->fblock->out->out_next);
-               
+               scc_add_fedge(state, scc, right);
        } else {
-               scc_add_fedge(state, scc, cond->fblock->out);
+               scc_add_fedge(state, scc, left);
        }
 
 }
 
-static void scc_visit_expr(struct compile_state *state, struct scc_state *scc,
+
+static void scc_add_sedge_dst(struct compile_state *state,
+       struct scc_state *scc, struct ssa_edge *sedge)
+{
+       if (triple_is_cbranch(state, sedge->dst->def)) {
+               scc_visit_cbranch(state, scc, sedge->dst);
+       }
+       else if (triple_is_def(state, sedge->dst->def)) {
+               scc_add_sedge(state, scc, sedge);
+       }
+}
+
+static void scc_visit_phi(struct compile_state *state, struct scc_state *scc,
        struct lattice_node *lnode)
 {
+       struct lattice_node *tmp;
+       struct triple **slot, *old;
+       struct flow_edge *fedge;
        int changed;
+       int index;
+       if (lnode->def->op != OP_PHI) {
+               internal_error(state, lnode->def, "not phi");
+       }
+       /* Store the original value */
+       old = preserve_lval(state, lnode);
 
-       changed = compute_lnode_val(state, scc, lnode);
-#if DEBUG_SCC
-       {
-               struct triple **expr;
-               fprintf(stderr, "expr: %3d %10s (",
-                       lnode->def->id, tops(lnode->def->op));
-               expr = triple_rhs(state, lnode->def, 0);
-               for(;expr;expr = triple_rhs(state, lnode->def, expr)) {
-                       if (*expr) {
-                               fprintf(stderr, " %d", (*expr)->id);
+       /* default to lattice high */
+       lnode->val = lnode->def;
+       slot = &RHS(lnode->def, 0);
+       index = 0;
+       for(fedge = lnode->fblock->in; fedge; index++, fedge = fedge->in_next) {
+               if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+                       fprintf(state->errout, "Examining edge: %d vertex: %d executable: %d\n",
+                               index,
+                               fedge->dst->block->vertex,
+                               fedge->executable
+                               );
+               }
+               if (!fedge->executable) {
+                       continue;
+               }
+               if (!slot[index]) {
+                       internal_error(state, lnode->def, "no phi value");
+               }
+               tmp = triple_to_lattice(state, scc, slot[index]);
+               /* meet(X, lattice low) = lattice low */
+               if (is_lattice_lo(state, tmp)) {
+                       lnode->val = 0;
+               }
+               /* meet(X, lattice high) = X */
+               else if (is_lattice_hi(state, tmp)) {
+                       lnode->val = lnode->val;
+               }
+               /* meet(lattice high, X) = X */
+               else if (is_lattice_hi(state, lnode)) {
+                       lnode->val = dup_triple(state, tmp->val);
+                       /* Only change the type if necessary */
+                       if (!is_subset_type(lnode->def->type, tmp->val->type)) {
+                               lnode->val->type = lnode->def->type;
                        }
                }
-               fprintf(stderr, " ) -> %s\n",
-                       (!lnode->val)? "lo": is_const(lnode->val)? "const": "hi");
+               /* meet(const, const) = const or lattice low */
+               else if (!constants_equal(state, lnode->val, tmp->val)) {
+                       lnode->val = 0;
+               }
+
+               /* meet(lattice low, X) = lattice low */
+               if (is_lattice_lo(state, lnode)) {
+                       lnode->val = 0;
+                       break;
+               }
        }
-#endif
-       if (lnode->def->op == OP_BRANCH) {
-               scc_visit_branch(state, scc, lnode);
+       changed = lval_changed(state, old, lnode);
+       scc_debug_lnode(state, scc, lnode, changed);
+
+       /* If the lattice value has changed update the work lists. */
+       if (changed) {
+               struct ssa_edge *sedge;
+               for(sedge = lnode->out; sedge; sedge = sedge->out_next) {
+                       scc_add_sedge_dst(state, scc, sedge);
+               }
+       }
+}
 
+
+static void scc_visit_expr(struct compile_state *state, struct scc_state *scc,
+       struct lattice_node *lnode)
+{
+       int changed;
+
+       if (!triple_is_def(state, lnode->def)) {
+               internal_warning(state, lnode->def, "not visiting an expression?");
        }
-       else if (changed) {
+       changed = compute_lnode_val(state, scc, lnode);
+       scc_debug_lnode(state, scc, lnode, changed);
+
+       if (changed) {
                struct ssa_edge *sedge;
                for(sedge = lnode->out; sedge; sedge = sedge->out_next) {
-                       scc_add_sedge(state, scc, sedge);
+                       scc_add_sedge_dst(state, scc, sedge);
                }
        }
 }
@@ -13703,19 +20976,34 @@ static void scc_writeback_values(
        struct compile_state *state, struct scc_state *scc)
 {
        struct triple *first, *ins;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                struct lattice_node *lnode;
                lnode = triple_to_lattice(state, scc, ins);
+               if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+                       if (is_lattice_hi(state, lnode) &&
+                               (lnode->val->op != OP_NOOP))
+                       {
+                               struct flow_edge *fedge;
+                               int executable;
+                               executable = 0;
+                               for(fedge = lnode->fblock->in;
+                                   !executable && fedge; fedge = fedge->in_next) {
+                                       executable |= fedge->executable;
+                               }
+                               if (executable) {
+                                       internal_warning(state, lnode->def,
+                                               "lattice node %d %s->%s still high?",
+                                               ins->id,
+                                               tops(lnode->def->op),
+                                               tops(lnode->val->op));
+                               }
+                       }
+               }
+
                /* Restore id */
                ins->id = lnode->old_id;
-#if DEBUG_SCC
-               if (lnode->val && !is_const(lnode->val)) {
-                       warning(state, lnode->def, 
-                               "lattice node still high?");
-               }
-#endif
                if (lnode->val && (lnode->val != ins)) {
                        /* See if it something I know how to write back */
                        switch(lnode->val->op) {
@@ -13723,7 +21011,7 @@ static void scc_writeback_values(
                                mkconst(state, ins, lnode->val->u.cval);
                                break;
                        case OP_ADDRCONST:
-                               mkaddr_const(state, ins, 
+                               mkaddr_const(state, ins,
                                        MISC(lnode->val, 0), lnode->val->u.cval);
                                break;
                        default:
@@ -13748,6 +21036,9 @@ static void scc_writeback_values(
 static void scc_transform(struct compile_state *state)
 {
        struct scc_state scc;
+       if (!(state->compiler->flags & COMPILER_SCC_TRANSFORM)) {
+               return;
+       }
 
        initialize_scc_state(state, &scc);
 
@@ -13759,7 +21050,7 @@ static void scc_transform(struct compile_state *state)
                        struct block *block;
                        struct triple *ptr;
                        struct flow_block *fblock;
-                       int time;
+                       int reps;
                        int done;
                        if (fedge->executable) {
                                continue;
@@ -13773,17 +21064,18 @@ static void scc_transform(struct compile_state *state)
                        fedge->executable = 1;
                        fblock = fedge->dst;
                        block = fblock->block;
-                       time = 0;
+                       reps = 0;
                        for(fptr = fblock->in; fptr; fptr = fptr->in_next) {
                                if (fptr->executable) {
-                                       time++;
+                                       reps++;
                                }
                        }
-#if DEBUG_SCC
-                       fprintf(stderr, "vertex: %d time: %d\n", 
-                               block->vertex, time);
-                       
-#endif
+
+                       if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+                               fprintf(state->errout, "vertex: %d reps: %d\n",
+                                       block->vertex, reps);
+                       }
+
                        done = 0;
                        for(ptr = block->first; !done; ptr = ptr->next) {
                                struct lattice_node *lnode;
@@ -13792,12 +21084,17 @@ static void scc_transform(struct compile_state *state)
                                if (ptr->op == OP_PHI) {
                                        scc_visit_phi(state, &scc, lnode);
                                }
-                               else if (time == 1) {
+                               else if ((reps == 1) && triple_is_def(state, ptr))
+                               {
                                        scc_visit_expr(state, &scc, lnode);
                                }
                        }
-                       if (fblock->out && !fblock->out->out_next) {
-                               scc_add_fedge(state, &scc, fblock->out);
+                       /* Add unconditional branch edges */
+                       if (!triple_is_cbranch(state, fblock->block->last)) {
+                               struct flow_edge *out;
+                               for(out = fblock->out; out; out = out->out_next) {
+                                       scc_add_fedge(state, &scc, out);
+                               }
                        }
                }
                while((sedge = scc_next_sedge(state, &scc))) {
@@ -13805,12 +21102,14 @@ static void scc_transform(struct compile_state *state)
                        struct flow_block *fblock;
                        lnode = sedge->dst;
                        fblock = lnode->fblock;
-#if DEBUG_SCC
-                       fprintf(stderr, "sedge: %5d (%5d -> %5d)\n",
-                               sedge - scc.ssa_edges,
-                               sedge->src->def->id,
-                               sedge->dst->def->id);
-#endif
+
+                       if (state->compiler->debug & DEBUG_SCC_TRANSFORM) {
+                               fprintf(state->errout, "sedge: %5ld (%5d -> %5d)\n",
+                                       (unsigned long)sedge - (unsigned long)scc.ssa_edges,
+                                       sedge->src->def->id,
+                                       sedge->dst->def->id);
+                       }
+
                        if (lnode->def->op == OP_PHI) {
                                scc_visit_phi(state, &scc, lnode);
                        }
@@ -13826,20 +21125,25 @@ static void scc_transform(struct compile_state *state)
                        }
                }
        }
-       
+
        scc_writeback_values(state, &scc);
        free_scc_state(state, &scc);
+       rebuild_ssa_form(state);
+
+       print_blocks(state, __func__, state->dbgout);
 }
 
 
 static void transform_to_arch_instructions(struct compile_state *state)
 {
        struct triple *ins, *first;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                ins = transform_to_arch_instruction(state, ins);
        } while(ins != first);
+
+       print_blocks(state, __func__, state->dbgout);
 }
 
 #if DEBUG_CONSISTENCY
@@ -13847,13 +21151,15 @@ static void verify_uses(struct compile_state *state)
 {
        struct triple *first, *ins;
        struct triple_set *set;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                struct triple **expr;
                expr = triple_rhs(state, ins, 0);
                for(; expr; expr = triple_rhs(state, ins, expr)) {
-                       for(set = *expr?(*expr)->use:0; set; set = set->next) {
+                       struct triple *rhs;
+                       rhs = *expr;
+                       for(set = rhs?rhs->use:0; set; set = set->next) {
                                if (set->member == ins) {
                                        break;
                                }
@@ -13864,7 +21170,9 @@ static void verify_uses(struct compile_state *state)
                }
                expr = triple_lhs(state, ins, 0);
                for(; expr; expr = triple_lhs(state, ins, expr)) {
-                       for(set =  *expr?(*expr)->use:0; set; set = set->next) {
+                       struct triple *lhs;
+                       lhs = *expr;
+                       for(set =  lhs?lhs->use:0; set; set = set->next) {
                                if (set->member == ins) {
                                        break;
                                }
@@ -13873,29 +21181,150 @@ static void verify_uses(struct compile_state *state)
                                internal_error(state, ins, "lhs not used");
                        }
                }
+               expr = triple_misc(state, ins, 0);
+               if (ins->op != OP_PHI) {
+                       for(; expr; expr = triple_targ(state, ins, expr)) {
+                               struct triple *misc;
+                               misc = *expr;
+                               for(set = misc?misc->use:0; set; set = set->next) {
+                                       if (set->member == ins) {
+                                               break;
+                                       }
+                               }
+                               if (!set) {
+                                       internal_error(state, ins, "misc not used");
+                               }
+                       }
+               }
+               if (!triple_is_ret(state, ins)) {
+                       expr = triple_targ(state, ins, 0);
+                       for(; expr; expr = triple_targ(state, ins, expr)) {
+                               struct triple *targ;
+                               targ = *expr;
+                               for(set = targ?targ->use:0; set; set = set->next) {
+                                       if (set->member == ins) {
+                                               break;
+                                       }
+                               }
+                               if (!set) {
+                                       internal_error(state, ins, "targ not used");
+                               }
+                       }
+               }
+               ins = ins->next;
+       } while(ins != first);
+
+}
+static void verify_blocks_present(struct compile_state *state)
+{
+       struct triple *first, *ins;
+       if (!state->bb.first_block) {
+               return;
+       }
+       first = state->first;
+       ins = first;
+       do {
+               valid_ins(state, ins);
+               if (triple_stores_block(state, ins)) {
+                       if (!ins->u.block) {
+                               internal_error(state, ins,
+                                       "%p not in a block?", ins);
+                       }
+               }
                ins = ins->next;
        } while(ins != first);
-       
+
+
+}
+
+static int edge_present(struct compile_state *state, struct block *block, struct triple *edge)
+{
+       struct block_set *bedge;
+       struct block *targ;
+       targ = block_of_triple(state, edge);
+       for(bedge = block->edges; bedge; bedge = bedge->next) {
+               if (bedge->member == targ) {
+                       return 1;
+               }
+       }
+       return 0;
 }
+
 static void verify_blocks(struct compile_state *state)
 {
        struct triple *ins;
        struct block *block;
-       block = state->first_block;
+       int blocks;
+       block = state->bb.first_block;
        if (!block) {
                return;
        }
+       blocks = 0;
        do {
+               int users;
+               struct block_set *user, *edge;
+               blocks++;
                for(ins = block->first; ins != block->last->next; ins = ins->next) {
-                       if (!triple_stores_block(state, ins)) {
+                       if (triple_stores_block(state, ins) && (ins->u.block != block)) {
+                               internal_error(state, ins, "inconsitent block specified");
+                       }
+                       valid_ins(state, ins);
+               }
+               users = 0;
+               for(user = block->use; user; user = user->next) {
+                       users++;
+                       if (!user->member->first) {
+                               internal_error(state, block->first, "user is empty");
+                       }
+                       if ((block == state->bb.last_block) &&
+                               (user->member == state->bb.first_block)) {
                                continue;
                        }
-                       if (ins->u.block != block) {
-                               internal_error(state, ins, "inconsitent block specified");
+                       for(edge = user->member->edges; edge; edge = edge->next) {
+                               if (edge->member == block) {
+                                       break;
+                               }
+                       }
+                       if (!edge) {
+                               internal_error(state, user->member->first,
+                                       "user does not use block");
+                       }
+               }
+               if (triple_is_branch(state, block->last)) {
+                       struct triple **expr;
+                       expr = triple_edge_targ(state, block->last, 0);
+                       for(;expr; expr = triple_edge_targ(state, block->last, expr)) {
+                               if (*expr && !edge_present(state, block, *expr)) {
+                                       internal_error(state, block->last, "no edge to targ");
+                               }
                        }
                }
+               if (!triple_is_ubranch(state, block->last) &&
+                       (block != state->bb.last_block) &&
+                       !edge_present(state, block, block->last->next)) {
+                       internal_error(state, block->last, "no edge to block->last->next");
+               }
+               for(edge = block->edges; edge; edge = edge->next) {
+                       for(user = edge->member->use; user; user = user->next) {
+                               if (user->member == block) {
+                                       break;
+                               }
+                       }
+                       if (!user || user->member != block) {
+                               internal_error(state, block->first,
+                                       "block does not use edge");
+                       }
+                       if (!edge->member->first) {
+                               internal_error(state, block->first, "edge block is empty");
+                       }
+               }
+               if (block->users != users) {
+                       internal_error(state, block->first,
+                               "computed users %d != stored users %d",
+                               users, block->users);
+               }
                if (!triple_stores_block(state, block->last->next)) {
-                       internal_error(state, block->last->next, 
+                       internal_error(state, block->last->next,
                                "cannot find next block");
                }
                block = block->last->next->u.block;
@@ -13903,36 +21332,89 @@ static void verify_blocks(struct compile_state *state)
                        internal_error(state, block->last->next,
                                "bad next block");
                }
-       } while(block != state->first_block);
+       } while(block != state->bb.first_block);
+       if (blocks != state->bb.last_vertex) {
+               internal_error(state, 0, "computed blocks: %d != stored blocks %d",
+                       blocks, state->bb.last_vertex);
+       }
 }
 
 static void verify_domination(struct compile_state *state)
 {
        struct triple *first, *ins;
        struct triple_set *set;
-       if (!state->first_block) {
+       if (!state->bb.first_block) {
                return;
        }
-       
-       first = RHS(state->main_function, 0);
+
+       first = state->first;
        ins = first;
        do {
                for(set = ins->use; set; set = set->next) {
-                       struct triple **expr;
-                       if (set->member->op == OP_PHI) {
-                               continue;
-                       }
-                       /* See if the use is on the righ hand side */
-                       expr = triple_rhs(state, set->member, 0);
-                       for(; expr ; expr = triple_rhs(state, set->member, expr)) {
-                               if (*expr == ins) {
+                       struct triple **slot;
+                       struct triple *use_point;
+                       int i, zrhs;
+                       use_point = 0;
+                       zrhs = set->member->rhs;
+                       slot = &RHS(set->member, 0);
+                       /* See if the use is on the right hand side */
+                       for(i = 0; i < zrhs; i++) {
+                               if (slot[i] == ins) {
                                        break;
                                }
                        }
-                       if (expr &&
-                               !tdominates(state, ins, set->member)) {
-                               internal_error(state, set->member, 
-                                       "non dominated rhs use?");
+                       if (i < zrhs) {
+                               use_point = set->member;
+                               if (set->member->op == OP_PHI) {
+                                       struct block_set *bset;
+                                       int edge;
+                                       bset = set->member->u.block->use;
+                                       for(edge = 0; bset && (edge < i); edge++) {
+                                               bset = bset->next;
+                                       }
+                                       if (!bset) {
+                                               internal_error(state, set->member,
+                                                       "no edge for phi rhs %d", i);
+                                       }
+                                       use_point = bset->member->last;
+                               }
+                       }
+                       if (use_point &&
+                               !tdominates(state, ins, use_point)) {
+                               if (is_const(ins)) {
+                                       internal_warning(state, ins,
+                                       "non dominated rhs use point %p?", use_point);
+                               }
+                               else {
+                                       internal_error(state, ins,
+                                               "non dominated rhs use point %p?", use_point);
+                               }
+                       }
+               }
+               ins = ins->next;
+       } while(ins != first);
+}
+
+static void verify_rhs(struct compile_state *state)
+{
+       struct triple *first, *ins;
+       first = state->first;
+       ins = first;
+       do {
+               struct triple **slot;
+               int zrhs, i;
+               zrhs = ins->rhs;
+               slot = &RHS(ins, 0);
+               for(i = 0; i < zrhs; i++) {
+                       if (slot[i] == 0) {
+                               internal_error(state, ins,
+                                       "missing rhs %d on %s",
+                                       i, tops(ins->op));
+                       }
+                       if ((ins->op != OP_PHI) && (slot[i] == ins)) {
+                               internal_error(state, ins,
+                                       "ins == rhs[%d] on %s",
+                                       i, tops(ins->op));
                        }
                }
                ins = ins->next;
@@ -13942,15 +21424,12 @@ static void verify_domination(struct compile_state *state)
 static void verify_piece(struct compile_state *state)
 {
        struct triple *first, *ins;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                struct triple *ptr;
                int lhs, i;
-               lhs = TRIPLE_LHS(ins->sizes);
-               if ((ins->op == OP_WRITE) || (ins->op == OP_STORE)) {
-                       lhs = 0;
-               }
+               lhs = ins->lhs;
                for(ptr = ins->next, i = 0; i < lhs; i++, ptr = ptr->next) {
                        if (ptr != LHS(ins, i)) {
                                internal_error(state, ins, "malformed lhs on %s",
@@ -13968,114 +21447,198 @@ static void verify_piece(struct compile_state *state)
                ins = ins->next;
        } while(ins != first);
 }
+
 static void verify_ins_colors(struct compile_state *state)
 {
        struct triple *first, *ins;
-       
-       first = RHS(state->main_function, 0);
+
+       first = state->first;
+       ins = first;
+       do {
+               ins = ins->next;
+       } while(ins != first);
+}
+
+static void verify_unknown(struct compile_state *state)
+{
+       struct triple *first, *ins;
+       if (    (unknown_triple.next != &unknown_triple) ||
+               (unknown_triple.prev != &unknown_triple) ||
+#if 0
+               (unknown_triple.use != 0) ||
+#endif
+               (unknown_triple.op != OP_UNKNOWNVAL) ||
+               (unknown_triple.lhs != 0) ||
+               (unknown_triple.rhs != 0) ||
+               (unknown_triple.misc != 0) ||
+               (unknown_triple.targ != 0) ||
+               (unknown_triple.template_id != 0) ||
+               (unknown_triple.id != -1) ||
+               (unknown_triple.type != &unknown_type) ||
+               (unknown_triple.occurance != &dummy_occurance) ||
+               (unknown_triple.param[0] != 0) ||
+               (unknown_triple.param[1] != 0)) {
+               internal_error(state, &unknown_triple, "unknown_triple corrupted!");
+       }
+       if (    (dummy_occurance.count != 2) ||
+               (strcmp(dummy_occurance.filename, __FILE__) != 0) ||
+               (strcmp(dummy_occurance.function, "") != 0) ||
+               (dummy_occurance.col != 0) ||
+               (dummy_occurance.parent != 0)) {
+               internal_error(state, &unknown_triple, "dummy_occurance corrupted!");
+       }
+       if (    (unknown_type.type != TYPE_UNKNOWN)) {
+               internal_error(state, &unknown_triple, "unknown_type corrupted!");
+       }
+       first = state->first;
        ins = first;
        do {
+               int params, i;
+               if (ins == &unknown_triple) {
+                       internal_error(state, ins, "unknown triple in list");
+               }
+               params = TRIPLE_SIZE(ins);
+               for(i = 0; i < params; i++) {
+                       if (ins->param[i] == &unknown_triple) {
+                               internal_error(state, ins, "unknown triple used!");
+                       }
+               }
                ins = ins->next;
        } while(ins != first);
 }
+
+static void verify_types(struct compile_state *state)
+{
+       struct triple *first, *ins;
+       first = state->first;
+       ins = first;
+       do {
+               struct type *invalid;
+               invalid = invalid_type(state, ins->type);
+               if (invalid) {
+                       FILE *fp = state->errout;
+                       fprintf(fp, "type: ");
+                       name_of(fp, ins->type);
+                       fprintf(fp, "\n");
+                       fprintf(fp, "invalid type: ");
+                       name_of(fp, invalid);
+                       fprintf(fp, "\n");
+                       internal_error(state, ins, "invalid ins type");
+               }
+       } while(ins != first);
+}
+
+static void verify_copy(struct compile_state *state)
+{
+       struct triple *first, *ins, *next;
+       first = state->first;
+       next = ins = first;
+       do {
+               ins = next;
+               next = ins->next;
+               if (ins->op != OP_COPY) {
+                       continue;
+               }
+               if (!equiv_types(ins->type, RHS(ins, 0)->type)) {
+                       FILE *fp = state->errout;
+                       fprintf(fp, "src type: ");
+                       name_of(fp, RHS(ins, 0)->type);
+                       fprintf(fp, "\n");
+                       fprintf(fp, "dst type: ");
+                       name_of(fp, ins->type);
+                       fprintf(fp, "\n");
+                       internal_error(state, ins, "type mismatch in copy");
+               }
+       } while(next != first);
+}
+
 static void verify_consistency(struct compile_state *state)
 {
+       verify_unknown(state);
        verify_uses(state);
+       verify_blocks_present(state);
        verify_blocks(state);
        verify_domination(state);
+       verify_rhs(state);
        verify_piece(state);
        verify_ins_colors(state);
+       verify_types(state);
+       verify_copy(state);
+       if (state->compiler->debug & DEBUG_VERIFICATION) {
+               fprintf(state->dbgout, "consistency verified\n");
+       }
 }
-#else 
-#define verify_consistency(state) do {} while(0)
-#endif /* DEBUG_USES */
+#else
+static void verify_consistency(struct compile_state *state) {}
+#endif /* DEBUG_CONSISTENCY */
 
 static void optimize(struct compile_state *state)
 {
-       if (state->debug & DEBUG_TRIPLES) {
-               print_triples(state);
-       }
+       /* Join all of the functions into one giant function */
+       join_functions(state);
+
+       /* Dump what the instruction graph intially looks like */
+       print_triples(state);
+
        /* Replace structures with simpler data types */
-       flatten_structures(state);
-       if (state->debug & DEBUG_TRIPLES) {
-               print_triples(state);
-       }
+       decompose_compound_types(state);
+       print_triples(state);
+
        verify_consistency(state);
-       /* Analize the intermediate code */
-       setup_basic_blocks(state);
-       analyze_idominators(state);
-       analyze_ipdominators(state);
-       /* Transform the code to ssa form */
+       /* Analyze the intermediate code */
+       state->bb.first = state->first;
+       analyze_basic_blocks(state, &state->bb);
+
+       /* Transform the code to ssa form. */
+       /*
+        * The transformation to ssa form puts a phi function
+        * on each of edge of a dominance frontier where that
+        * phi function might be needed.  At -O2 if we don't
+        * eleminate the excess phi functions we can get an
+        * exponential code size growth.  So I kill the extra
+        * phi functions early and I kill them often.
+        */
        transform_to_ssa_form(state);
        verify_consistency(state);
-       if (state->debug & DEBUG_CODE_ELIMINATION) {
-               fprintf(stdout, "After transform_to_ssa_form\n");
-               print_blocks(state, stdout);
-       }
+
+       /* Remove dead code */
+       eliminate_inefectual_code(state);
+       verify_consistency(state);
+
        /* Do strength reduction and simple constant optimizations */
-       if (state->optimize >= 1) {
-               simplify_all(state);
-       }
+       simplify_all(state);
        verify_consistency(state);
        /* Propogate constants throughout the code */
-       if (state->optimize >= 2) {
-#warning "FIXME fix scc_transform"
-               scc_transform(state);
-               transform_from_ssa_form(state);
-               free_basic_blocks(state);
-               setup_basic_blocks(state);
-               analyze_idominators(state);
-               analyze_ipdominators(state);
-               transform_to_ssa_form(state);
-       }
+       scc_transform(state);
        verify_consistency(state);
+#if DEBUG_ROMCC_WARNINGS
 #warning "WISHLIST implement single use constants (least possible register pressure)"
 #warning "WISHLIST implement induction variable elimination"
+#endif
        /* Select architecture instructions and an initial partial
         * coloring based on architecture constraints.
         */
        transform_to_arch_instructions(state);
        verify_consistency(state);
-       if (state->debug & DEBUG_ARCH_CODE) {
-               printf("After transform_to_arch_instructions\n");
-               print_blocks(state, stdout);
-               print_control_flow(state);
-       }
+
+       /* Remove dead code */
        eliminate_inefectual_code(state);
        verify_consistency(state);
-       if (state->debug & DEBUG_CODE_ELIMINATION) {
-               printf("After eliminate_inefectual_code\n");
-               print_blocks(state, stdout);
-               print_control_flow(state);
-       }
-       verify_consistency(state);
+
        /* Color all of the variables to see if they will fit in registers */
        insert_copies_to_phi(state);
-       if (state->debug & DEBUG_INSERTED_COPIES) {
-               printf("After insert_copies_to_phi\n");
-               print_blocks(state, stdout);
-               print_control_flow(state);
-       }
        verify_consistency(state);
+
        insert_mandatory_copies(state);
-       if (state->debug & DEBUG_INSERTED_COPIES) {
-               printf("After insert_mandatory_copies\n");
-               print_blocks(state, stdout);
-               print_control_flow(state);
-       }
        verify_consistency(state);
+
        allocate_registers(state);
        verify_consistency(state);
-       if (state->debug & DEBUG_INTERMEDIATE_CODE) {
-               print_blocks(state, stdout);
-       }
-       if (state->debug & DEBUG_CONTROL_FLOW) {
-               print_control_flow(state);
-       }
+
        /* Remove the optimization information.
         * This is more to check for memory consistency than to free memory.
         */
-       free_basic_blocks(state);
+       free_basic_blocks(state, &state->bb);
 }
 
 static void print_op_asm(struct compile_state *state,
@@ -14085,8 +21648,8 @@ static void print_op_asm(struct compile_state *state,
        const char *ptr;
        unsigned lhs, rhs, i;
        info = ins->u.ainfo;
-       lhs = TRIPLE_LHS(ins->sizes);
-       rhs = TRIPLE_RHS(ins->sizes);
+       lhs = ins->lhs;
+       rhs = ins->rhs;
        /* Don't count the clobbers in lhs */
        for(i = 0; i < lhs; i++) {
                if (LHS(ins, i)->type == &void_type) {
@@ -14094,6 +21657,7 @@ static void print_op_asm(struct compile_state *state,
                }
        }
        lhs = i;
+       fprintf(fp, "#ASM\n");
        fputc('\t', fp);
        for(ptr = info->str; *ptr; ptr++) {
                char *next;
@@ -14117,11 +21681,11 @@ static void print_op_asm(struct compile_state *state,
                                param);
                }
                piece = (param < lhs)? LHS(ins, param) : RHS(ins, param - lhs);
-               fprintf(fp, "%s", 
+               fprintf(fp, "%s",
                        arch_reg_str(ID_REG(piece->id)));
-               ptr = next;
+               ptr = next -1;
        }
-       fputc('\n', fp);
+       fprintf(fp, "\n#NOT ASM\n");
 }
 
 
@@ -14130,48 +21694,48 @@ static void print_op_asm(struct compile_state *state,
  */
 #define X86_4_8BIT_GPRS 1
 
-/* Recognized x86 cpu variants */
-#define BAD_CPU      0
-#define CPU_I386     1
-#define CPU_P3       2
-#define CPU_P4       3
-#define CPU_K7       4
-#define CPU_K8       5
-
-#define CPU_DEFAULT  CPU_I386
+/* x86 featrues */
+#define X86_MMX_REGS  (1<<0)
+#define X86_XMM_REGS  (1<<1)
+#define X86_NOOP_COPY (1<<2)
 
 /* The x86 register classes */
-#define REGC_FLAGS    0
-#define REGC_GPR8     1
-#define REGC_GPR16    2
-#define REGC_GPR32    3
-#define REGC_GPR64    4
-#define REGC_MMX      5
-#define REGC_XMM      6
-#define REGC_GPR32_8  7
-#define REGC_GPR16_8  8
-#define REGC_IMM32    9
-#define REGC_IMM16   10
-#define REGC_IMM8    11
+#define REGC_FLAGS       0
+#define REGC_GPR8        1
+#define REGC_GPR16       2
+#define REGC_GPR32       3
+#define REGC_DIVIDEND64  4
+#define REGC_DIVIDEND32  5
+#define REGC_MMX         6
+#define REGC_XMM         7
+#define REGC_GPR32_8     8
+#define REGC_GPR16_8     9
+#define REGC_GPR8_LO    10
+#define REGC_IMM32      11
+#define REGC_IMM16      12
+#define REGC_IMM8       13
 #define LAST_REGC  REGC_IMM8
 #if LAST_REGC >= MAX_REGC
 #error "MAX_REGC is to low"
 #endif
 
 /* Register class masks */
-#define REGCM_FLAGS   (1 << REGC_FLAGS)
-#define REGCM_GPR8    (1 << REGC_GPR8)
-#define REGCM_GPR16   (1 << REGC_GPR16)
-#define REGCM_GPR32   (1 << REGC_GPR32)
-#define REGCM_GPR64   (1 << REGC_GPR64)
-#define REGCM_MMX     (1 << REGC_MMX)
-#define REGCM_XMM     (1 << REGC_XMM)
-#define REGCM_GPR32_8 (1 << REGC_GPR32_8)
-#define REGCM_GPR16_8 (1 << REGC_GPR16_8)
-#define REGCM_IMM32   (1 << REGC_IMM32)
-#define REGCM_IMM16   (1 << REGC_IMM16)
-#define REGCM_IMM8    (1 << REGC_IMM8)
-#define REGCM_ALL     ((1 << (LAST_REGC + 1)) - 1)
+#define REGCM_FLAGS      (1 << REGC_FLAGS)
+#define REGCM_GPR8       (1 << REGC_GPR8)
+#define REGCM_GPR16      (1 << REGC_GPR16)
+#define REGCM_GPR32      (1 << REGC_GPR32)
+#define REGCM_DIVIDEND64 (1 << REGC_DIVIDEND64)
+#define REGCM_DIVIDEND32 (1 << REGC_DIVIDEND32)
+#define REGCM_MMX        (1 << REGC_MMX)
+#define REGCM_XMM        (1 << REGC_XMM)
+#define REGCM_GPR32_8    (1 << REGC_GPR32_8)
+#define REGCM_GPR16_8    (1 << REGC_GPR16_8)
+#define REGCM_GPR8_LO    (1 << REGC_GPR8_LO)
+#define REGCM_IMM32      (1 << REGC_IMM32)
+#define REGCM_IMM16      (1 << REGC_IMM16)
+#define REGCM_IMM8       (1 << REGC_IMM8)
+#define REGCM_ALL        ((1 << (LAST_REGC + 1)) - 1)
+#define REGCM_IMMALL   (REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8)
 
 /* The x86 registers */
 #define REG_EFLAGS  2
@@ -14185,12 +21749,10 @@ static void print_op_asm(struct compile_state *state,
 #define REG_BH      8
 #define REG_CH      9
 #define REG_DH      10
+#define REGC_GPR8_LO_FIRST REG_AL
+#define REGC_GPR8_LO_LAST  REG_DL
 #define REGC_GPR8_FIRST  REG_AL
-#if X86_4_8BIT_GPRS
-#define REGC_GPR8_LAST   REG_DL
-#else 
 #define REGC_GPR8_LAST   REG_DH
-#endif
 #define REG_AX     11
 #define REG_BX     12
 #define REG_CX     13
@@ -14212,29 +21774,36 @@ static void print_op_asm(struct compile_state *state,
 #define REGC_GPR32_FIRST REG_EAX
 #define REGC_GPR32_LAST  REG_ESP
 #define REG_EDXEAX 27
-#define REGC_GPR64_FIRST REG_EDXEAX
-#define REGC_GPR64_LAST  REG_EDXEAX
-#define REG_MMX0   28
-#define REG_MMX1   29
-#define REG_MMX2   30
-#define REG_MMX3   31
-#define REG_MMX4   32
-#define REG_MMX5   33
-#define REG_MMX6   34
-#define REG_MMX7   35
+#define REGC_DIVIDEND64_FIRST REG_EDXEAX
+#define REGC_DIVIDEND64_LAST  REG_EDXEAX
+#define REG_DXAX   28
+#define REGC_DIVIDEND32_FIRST REG_DXAX
+#define REGC_DIVIDEND32_LAST  REG_DXAX
+#define REG_MMX0   29
+#define REG_MMX1   30
+#define REG_MMX2   31
+#define REG_MMX3   32
+#define REG_MMX4   33
+#define REG_MMX5   34
+#define REG_MMX6   35
+#define REG_MMX7   36
 #define REGC_MMX_FIRST REG_MMX0
 #define REGC_MMX_LAST  REG_MMX7
-#define REG_XMM0   36
-#define REG_XMM1   37
-#define REG_XMM2   38
-#define REG_XMM3   39
-#define REG_XMM4   40
-#define REG_XMM5   41
-#define REG_XMM6   42
-#define REG_XMM7   43
+#define REG_XMM0   37
+#define REG_XMM1   38
+#define REG_XMM2   39
+#define REG_XMM3   40
+#define REG_XMM4   41
+#define REG_XMM5   42
+#define REG_XMM6   43
+#define REG_XMM7   44
 #define REGC_XMM_FIRST REG_XMM0
 #define REGC_XMM_LAST  REG_XMM7
+
+#if DEBUG_ROMCC_WARNINGS
 #warning "WISHLIST figure out how to use pinsrw and pextrw to better use extended regs"
+#endif
+
 #define LAST_REG   REG_XMM7
 
 #define REGC_GPR32_8_FIRST REG_EAX
@@ -14255,57 +21824,109 @@ static void print_op_asm(struct compile_state *state,
 
 
 static unsigned regc_size[LAST_REGC +1] = {
-       [REGC_FLAGS]   = REGC_FLAGS_LAST   - REGC_FLAGS_FIRST + 1,
-       [REGC_GPR8]    = REGC_GPR8_LAST    - REGC_GPR8_FIRST + 1,
-       [REGC_GPR16]   = REGC_GPR16_LAST   - REGC_GPR16_FIRST + 1,
-       [REGC_GPR32]   = REGC_GPR32_LAST   - REGC_GPR32_FIRST + 1,
-       [REGC_GPR64]   = REGC_GPR64_LAST   - REGC_GPR64_FIRST + 1,
-       [REGC_MMX]     = REGC_MMX_LAST     - REGC_MMX_FIRST + 1,
-       [REGC_XMM]     = REGC_XMM_LAST     - REGC_XMM_FIRST + 1,
-       [REGC_GPR32_8] = REGC_GPR32_8_LAST - REGC_GPR32_8_FIRST + 1,
-       [REGC_GPR16_8] = REGC_GPR16_8_LAST - REGC_GPR16_8_FIRST + 1,
-       [REGC_IMM32]   = 0,
-       [REGC_IMM16]   = 0,
-       [REGC_IMM8]    = 0,
+       [REGC_FLAGS]      = REGC_FLAGS_LAST      - REGC_FLAGS_FIRST + 1,
+       [REGC_GPR8]       = REGC_GPR8_LAST       - REGC_GPR8_FIRST + 1,
+       [REGC_GPR16]      = REGC_GPR16_LAST      - REGC_GPR16_FIRST + 1,
+       [REGC_GPR32]      = REGC_GPR32_LAST      - REGC_GPR32_FIRST + 1,
+       [REGC_DIVIDEND64] = REGC_DIVIDEND64_LAST - REGC_DIVIDEND64_FIRST + 1,
+       [REGC_DIVIDEND32] = REGC_DIVIDEND32_LAST - REGC_DIVIDEND32_FIRST + 1,
+       [REGC_MMX]        = REGC_MMX_LAST        - REGC_MMX_FIRST + 1,
+       [REGC_XMM]        = REGC_XMM_LAST        - REGC_XMM_FIRST + 1,
+       [REGC_GPR32_8]    = REGC_GPR32_8_LAST    - REGC_GPR32_8_FIRST + 1,
+       [REGC_GPR16_8]    = REGC_GPR16_8_LAST    - REGC_GPR16_8_FIRST + 1,
+       [REGC_GPR8_LO]    = REGC_GPR8_LO_LAST    - REGC_GPR8_LO_FIRST + 1,
+       [REGC_IMM32]      = 0,
+       [REGC_IMM16]      = 0,
+       [REGC_IMM8]       = 0,
 };
 
 static const struct {
        int first, last;
 } regcm_bound[LAST_REGC + 1] = {
-       [REGC_FLAGS]   = { REGC_FLAGS_FIRST,   REGC_FLAGS_LAST },
-       [REGC_GPR8]    = { REGC_GPR8_FIRST,    REGC_GPR8_LAST },
-       [REGC_GPR16]   = { REGC_GPR16_FIRST,   REGC_GPR16_LAST },
-       [REGC_GPR32]   = { REGC_GPR32_FIRST,   REGC_GPR32_LAST },
-       [REGC_GPR64]   = { REGC_GPR64_FIRST,   REGC_GPR64_LAST },
-       [REGC_MMX]     = { REGC_MMX_FIRST,     REGC_MMX_LAST },
-       [REGC_XMM]     = { REGC_XMM_FIRST,     REGC_XMM_LAST },
-       [REGC_GPR32_8] = { REGC_GPR32_8_FIRST, REGC_GPR32_8_LAST },
-       [REGC_GPR16_8] = { REGC_GPR16_8_FIRST, REGC_GPR16_8_LAST },
-       [REGC_IMM32]   = { REGC_IMM32_FIRST,   REGC_IMM32_LAST },
-       [REGC_IMM16]   = { REGC_IMM16_FIRST,   REGC_IMM16_LAST },
-       [REGC_IMM8]    = { REGC_IMM8_FIRST,    REGC_IMM8_LAST },
+       [REGC_FLAGS]      = { REGC_FLAGS_FIRST,      REGC_FLAGS_LAST },
+       [REGC_GPR8]       = { REGC_GPR8_FIRST,       REGC_GPR8_LAST },
+       [REGC_GPR16]      = { REGC_GPR16_FIRST,      REGC_GPR16_LAST },
+       [REGC_GPR32]      = { REGC_GPR32_FIRST,      REGC_GPR32_LAST },
+       [REGC_DIVIDEND64] = { REGC_DIVIDEND64_FIRST, REGC_DIVIDEND64_LAST },
+       [REGC_DIVIDEND32] = { REGC_DIVIDEND32_FIRST, REGC_DIVIDEND32_LAST },
+       [REGC_MMX]        = { REGC_MMX_FIRST,        REGC_MMX_LAST },
+       [REGC_XMM]        = { REGC_XMM_FIRST,        REGC_XMM_LAST },
+       [REGC_GPR32_8]    = { REGC_GPR32_8_FIRST,    REGC_GPR32_8_LAST },
+       [REGC_GPR16_8]    = { REGC_GPR16_8_FIRST,    REGC_GPR16_8_LAST },
+       [REGC_GPR8_LO]    = { REGC_GPR8_LO_FIRST,    REGC_GPR8_LO_LAST },
+       [REGC_IMM32]      = { REGC_IMM32_FIRST,      REGC_IMM32_LAST },
+       [REGC_IMM16]      = { REGC_IMM16_FIRST,      REGC_IMM16_LAST },
+       [REGC_IMM8]       = { REGC_IMM8_FIRST,       REGC_IMM8_LAST },
 };
 
-static int arch_encode_cpu(const char *cpu)
-{
-       struct cpu {
-               const char *name;
-               int cpu;
-       } cpus[] = {
-               { "i386", CPU_I386 },
-               { "p3",   CPU_P3 },
-               { "p4",   CPU_P4 },
-               { "k7",   CPU_K7 },
-               { "k8",   CPU_K8 },
-               {  0,     BAD_CPU }
-       };
-       struct cpu *ptr;
-       for(ptr = cpus; ptr->name; ptr++) {
-               if (strcmp(ptr->name, cpu) == 0) {
-                       break;
-               }
+#if ARCH_INPUT_REGS != 4
+#error ARCH_INPUT_REGS size mismatch
+#endif
+static const struct reg_info arch_input_regs[ARCH_INPUT_REGS] = {
+       { .reg = REG_EAX, .regcm = REGCM_GPR32 },
+       { .reg = REG_EBX, .regcm = REGCM_GPR32 },
+       { .reg = REG_ECX, .regcm = REGCM_GPR32 },
+       { .reg = REG_EDX, .regcm = REGCM_GPR32 },
+};
+
+#if ARCH_OUTPUT_REGS != 4
+#error ARCH_INPUT_REGS size mismatch
+#endif
+static const struct reg_info arch_output_regs[ARCH_OUTPUT_REGS] = {
+       { .reg = REG_EAX, .regcm = REGCM_GPR32 },
+       { .reg = REG_EBX, .regcm = REGCM_GPR32 },
+       { .reg = REG_ECX, .regcm = REGCM_GPR32 },
+       { .reg = REG_EDX, .regcm = REGCM_GPR32 },
+};
+
+static void init_arch_state(struct arch_state *arch)
+{
+       memset(arch, 0, sizeof(*arch));
+       arch->features = 0;
+}
+
+static const struct compiler_flag arch_flags[] = {
+       { "mmx",       X86_MMX_REGS },
+       { "sse",       X86_XMM_REGS },
+       { "noop-copy", X86_NOOP_COPY },
+       { 0,     0 },
+};
+static const struct compiler_flag arch_cpus[] = {
+       { "i386", 0 },
+       { "p2",   X86_MMX_REGS },
+       { "p3",   X86_MMX_REGS | X86_XMM_REGS },
+       { "p4",   X86_MMX_REGS | X86_XMM_REGS },
+       { "k7",   X86_MMX_REGS },
+       { "k8",   X86_MMX_REGS | X86_XMM_REGS },
+       { "c3",   X86_MMX_REGS },
+       { "c3-2", X86_MMX_REGS | X86_XMM_REGS }, /* Nehemiah */
+       {  0,     0 }
+};
+static int arch_encode_flag(struct arch_state *arch, const char *flag)
+{
+       int result;
+       int act;
+
+       act = 1;
+       result = -1;
+       if (strncmp(flag, "no-", 3) == 0) {
+               flag += 3;
+               act = 0;
+       }
+       if (act && strncmp(flag, "cpu=", 4) == 0) {
+               flag += 4;
+               result = set_flag(arch_cpus, &arch->features, 1, flag);
+       }
+       else {
+               result = set_flag(arch_flags, &arch->features, act, flag);
        }
-       return ptr->cpu;
+       return result;
+}
+
+static void arch_usage(FILE *fp)
+{
+       flag_usage(fp, arch_flags, "-m", "-mno-");
+       flag_usage(fp, arch_cpus, "-mcpu=", 0);
 }
 
 static unsigned arch_regc_size(struct compile_state *state, int class)
@@ -14315,17 +21936,19 @@ static unsigned arch_regc_size(struct compile_state *state, int class)
        }
        return regc_size[class];
 }
+
 static int arch_regcm_intersect(unsigned regcm1, unsigned regcm2)
 {
        /* See if two register classes may have overlapping registers */
-       unsigned gpr_mask = REGCM_GPR8 | REGCM_GPR16_8 | REGCM_GPR16 |
-               REGCM_GPR32_8 | REGCM_GPR32 | REGCM_GPR64;
+       unsigned gpr_mask = REGCM_GPR8 | REGCM_GPR8_LO | REGCM_GPR16_8 | REGCM_GPR16 |
+               REGCM_GPR32_8 | REGCM_GPR32 |
+               REGCM_DIVIDEND32 | REGCM_DIVIDEND64;
 
        /* Special case for the immediates */
        if ((regcm1 & (REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8)) &&
                ((regcm1 & ~(REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8)) == 0) &&
                (regcm2 & (REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8)) &&
-               ((regcm2 & ~(REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8)) == 0)) { 
+               ((regcm2 & ~(REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8)) == 0)) {
                return 0;
        }
        return (regcm1 & regcm2) ||
@@ -14346,6 +21969,7 @@ static void arch_reg_equivs(
 #endif
                *equiv++ = REG_AX;
                *equiv++ = REG_EAX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
        case REG_AH:
@@ -14354,9 +21978,10 @@ static void arch_reg_equivs(
 #endif
                *equiv++ = REG_AX;
                *equiv++ = REG_EAX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
-       case REG_BL:  
+       case REG_BL:
 #if X86_4_8BIT_GPRS
                *equiv++ = REG_BH;
 #endif
@@ -14392,6 +22017,7 @@ static void arch_reg_equivs(
 #endif
                *equiv++ = REG_DX;
                *equiv++ = REG_EDX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
        case REG_DH:
@@ -14400,12 +22026,14 @@ static void arch_reg_equivs(
 #endif
                *equiv++ = REG_DX;
                *equiv++ = REG_EDX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
        case REG_AX:
                *equiv++ = REG_AL;
                *equiv++ = REG_AH;
                *equiv++ = REG_EAX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
        case REG_BX:
@@ -14413,18 +22041,19 @@ static void arch_reg_equivs(
                *equiv++ = REG_BH;
                *equiv++ = REG_EBX;
                break;
-       case REG_CX:  
+       case REG_CX:
                *equiv++ = REG_CL;
                *equiv++ = REG_CH;
                *equiv++ = REG_ECX;
                break;
-       case REG_DX:  
+       case REG_DX:
                *equiv++ = REG_DL;
                *equiv++ = REG_DH;
                *equiv++ = REG_EDX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
-       case REG_SI:  
+       case REG_SI:
                *equiv++ = REG_ESI;
                break;
        case REG_DI:
@@ -14440,6 +22069,7 @@ static void arch_reg_equivs(
                *equiv++ = REG_AL;
                *equiv++ = REG_AH;
                *equiv++ = REG_AX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
        case REG_EBX:
@@ -14456,21 +22086,33 @@ static void arch_reg_equivs(
                *equiv++ = REG_DL;
                *equiv++ = REG_DH;
                *equiv++ = REG_DX;
+               *equiv++ = REG_DXAX;
                *equiv++ = REG_EDXEAX;
                break;
-       case REG_ESI: 
+       case REG_ESI:
                *equiv++ = REG_SI;
                break;
-       case REG_EDI: 
+       case REG_EDI:
                *equiv++ = REG_DI;
                break;
-       case REG_EBP: 
+       case REG_EBP:
                *equiv++ = REG_BP;
                break;
-       case REG_ESP: 
+       case REG_ESP:
                *equiv++ = REG_SP;
                break;
-       case REG_EDXEAX: 
+       case REG_DXAX:
+               *equiv++ = REG_AL;
+               *equiv++ = REG_AH;
+               *equiv++ = REG_DL;
+               *equiv++ = REG_DH;
+               *equiv++ = REG_AX;
+               *equiv++ = REG_DX;
+               *equiv++ = REG_EAX;
+               *equiv++ = REG_EDX;
+               *equiv++ = REG_EDXEAX;
+               break;
+       case REG_EDXEAX:
                *equiv++ = REG_AL;
                *equiv++ = REG_AH;
                *equiv++ = REG_DL;
@@ -14479,33 +22121,26 @@ static void arch_reg_equivs(
                *equiv++ = REG_DX;
                *equiv++ = REG_EAX;
                *equiv++ = REG_EDX;
+               *equiv++ = REG_DXAX;
                break;
        }
-       *equiv++ = REG_UNSET; 
+       *equiv++ = REG_UNSET;
 }
 
 static unsigned arch_avail_mask(struct compile_state *state)
 {
        unsigned avail_mask;
-       avail_mask = REGCM_GPR8 | REGCM_GPR16_8 | REGCM_GPR16 | 
-               REGCM_GPR32 | REGCM_GPR32_8 | REGCM_GPR64 |
+       /* REGCM_GPR8 is not available */
+       avail_mask = REGCM_GPR8_LO | REGCM_GPR16_8 | REGCM_GPR16 |
+               REGCM_GPR32 | REGCM_GPR32_8 |
+               REGCM_DIVIDEND32 | REGCM_DIVIDEND64 |
                REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8 | REGCM_FLAGS;
-       switch(state->cpu) {
-       case CPU_P3:
-       case CPU_K7:
+       if (state->arch->features & X86_MMX_REGS) {
                avail_mask |= REGCM_MMX;
-               break;
-       case CPU_P4:
-       case CPU_K8:
-               avail_mask |= REGCM_MMX | REGCM_XMM;
-               break;
        }
-#if 0
-       /* Don't enable 8 bit values until I can force both operands
-        * to be 8bits simultaneously.
-        */
-       avail_mask &= ~(REGCM_GPR8 | REGCM_GPR16_8 | REGCM_GPR16);
-#endif
+       if (state->arch->features & X86_XMM_REGS) {
+               avail_mask |= REGCM_XMM;
+       }
        return avail_mask;
 }
 
@@ -14514,7 +22149,6 @@ static unsigned arch_regcm_normalize(struct compile_state *state, unsigned regcm
        unsigned mask, result;
        int class, class2;
        result = regcm;
-       result &= arch_avail_mask(state);
 
        for(class = 0, mask = 1; mask; mask <<= 1, class++) {
                if ((result & mask) == 0) {
@@ -14530,9 +22164,20 @@ static unsigned arch_regcm_normalize(struct compile_state *state, unsigned regcm
                        }
                }
        }
+       result &= arch_avail_mask(state);
        return result;
 }
 
+static unsigned arch_regcm_reg_normalize(struct compile_state *state, unsigned regcm)
+{
+       /* Like arch_regcm_normalize except immediate register classes are excluded */
+       regcm = arch_regcm_normalize(state, regcm);
+       /* Remove the immediate register classes */
+       regcm &= ~(REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8);
+       return regcm;
+
+}
+
 static unsigned arch_reg_regcm(struct compile_state *state, int reg)
 {
        unsigned mask;
@@ -14558,19 +22203,19 @@ static struct reg_info arch_reg_constraint(
                unsigned int mask;
                unsigned int reg;
        } constraints[] = {
-               { 'r', REGCM_GPR32, REG_UNSET },
-               { 'g', REGCM_GPR32, REG_UNSET },
-               { 'p', REGCM_GPR32, REG_UNSET },
-               { 'q', REGCM_GPR8 REG_UNSET },
+               { 'r', REGCM_GPR32,   REG_UNSET },
+               { 'g', REGCM_GPR32,   REG_UNSET },
+               { 'p', REGCM_GPR32,   REG_UNSET },
+               { 'q', REGCM_GPR8_LO, REG_UNSET },
                { 'Q', REGCM_GPR32_8, REG_UNSET },
-               { 'x', REGCM_XMM,   REG_UNSET },
-               { 'y', REGCM_MMX,   REG_UNSET },
-               { 'a', REGCM_GPR32, REG_EAX },
-               { 'b', REGCM_GPR32, REG_EBX },
-               { 'c', REGCM_GPR32, REG_ECX },
-               { 'd', REGCM_GPR32, REG_EDX },
-               { 'D', REGCM_GPR32, REG_EDI },
-               { 'S', REGCM_GPR32, REG_ESI },
+               { 'x', REGCM_XMM,     REG_UNSET },
+               { 'y', REGCM_MMX,     REG_UNSET },
+               { 'a', REGCM_GPR32,   REG_EAX },
+               { 'b', REGCM_GPR32,   REG_EBX },
+               { 'c', REGCM_GPR32,   REG_ECX },
+               { 'd', REGCM_GPR32,   REG_EDX },
+               { 'D', REGCM_GPR32,   REG_EDI },
+               { 'S', REGCM_GPR32,   REG_ESI },
                { '\0', 0, REG_UNSET },
        };
        unsigned int regcm;
@@ -14619,35 +22264,35 @@ static struct reg_info arch_reg_clobber(
                result.reg = REG_UNSET;
                result.regcm = 0;
        }
-       else if (strcmp(clobber, "%eax") == 0) {
+       else if (strcmp(clobber, "eax") == 0) {
                result.reg = REG_EAX;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%ebx") == 0) {
+       else if (strcmp(clobber, "ebx") == 0) {
                result.reg = REG_EBX;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%ecx") == 0) {
+       else if (strcmp(clobber, "ecx") == 0) {
                result.reg = REG_ECX;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%edx") == 0) {
+       else if (strcmp(clobber, "edx") == 0) {
                result.reg = REG_EDX;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%esi") == 0) {
+       else if (strcmp(clobber, "esi") == 0) {
                result.reg = REG_ESI;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%edi") == 0) {
+       else if (strcmp(clobber, "edi") == 0) {
                result.reg = REG_EDI;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%ebp") == 0) {
+       else if (strcmp(clobber, "ebp") == 0) {
                result.reg = REG_EBP;
                result.regcm = REGCM_GPR32;
        }
-       else if (strcmp(clobber, "%esp") == 0) {
+       else if (strcmp(clobber, "esp") == 0) {
                result.reg = REG_ESP;
                result.regcm = REGCM_GPR32;
        }
@@ -14660,20 +22305,21 @@ static struct reg_info arch_reg_clobber(
                result.reg = REG_XMM0 + octdigval(clobber[3]);
                result.regcm = REGCM_XMM;
        }
-       else if ((strncmp(clobber, "mmx", 3) == 0) &&
+       else if ((strncmp(clobber, "mm", 2) == 0) &&
                octdigitp(clobber[3]) && (clobber[4] == '\0')) {
                result.reg = REG_MMX0 + octdigval(clobber[3]);
                result.regcm = REGCM_MMX;
        }
        else {
-               error(state, 0, "Invalid register clobber");
+               error(state, 0, "unknown register name `%s' in asm",
+                       clobber);
                result.reg = REG_UNSET;
                result.regcm = 0;
        }
        return result;
 }
 
-static int do_select_reg(struct compile_state *state, 
+static int do_select_reg(struct compile_state *state,
        char *used, int reg, unsigned classes)
 {
        unsigned mask;
@@ -14687,21 +22333,28 @@ static int do_select_reg(struct compile_state *state,
 static int arch_select_free_register(
        struct compile_state *state, char *used, int classes)
 {
-       /* Preference: flags, 8bit gprs, 32bit gprs, other 32bit reg
-        * other types of registers.
+       /* Live ranges with the most neighbors are colored first.
+        *
+        * Generally it does not matter which colors are given
+        * as the register allocator attempts to color live ranges
+        * in an order where you are guaranteed not to run out of colors.
+        *
+        * Occasionally the register allocator cannot find an order
+        * of register selection that will find a free color.  To
+        * increase the odds the register allocator will work when
+        * it guesses first give out registers from register classes
+        * least likely to run out of registers.
+        *
         */
        int i, reg;
        reg = REG_UNSET;
-       for(i = REGC_FLAGS_FIRST; (reg == REG_UNSET) && (i <= REGC_FLAGS_LAST); i++) {
-               reg = do_select_reg(state, used, i, classes);
-       }
-       for(i = REGC_GPR32_FIRST; (reg == REG_UNSET) && (i <= REGC_GPR32_LAST); i++) {
+       for(i = REGC_XMM_FIRST; (reg == REG_UNSET) && (i <= REGC_XMM_LAST); i++) {
                reg = do_select_reg(state, used, i, classes);
        }
        for(i = REGC_MMX_FIRST; (reg == REG_UNSET) && (i <= REGC_MMX_LAST); i++) {
                reg = do_select_reg(state, used, i, classes);
        }
-       for(i = REGC_XMM_FIRST; (reg == REG_UNSET) && (i <= REGC_XMM_LAST); i++) {
+       for(i = REGC_GPR32_LAST; (reg == REG_UNSET) && (i >= REGC_GPR32_FIRST); i--) {
                reg = do_select_reg(state, used, i, classes);
        }
        for(i = REGC_GPR16_FIRST; (reg == REG_UNSET) && (i <= REGC_GPR16_LAST); i++) {
@@ -14710,31 +22363,41 @@ static int arch_select_free_register(
        for(i = REGC_GPR8_FIRST; (reg == REG_UNSET) && (i <= REGC_GPR8_LAST); i++) {
                reg = do_select_reg(state, used, i, classes);
        }
-       for(i = REGC_GPR64_FIRST; (reg == REG_UNSET) && (i <= REGC_GPR64_LAST); i++) {
+       for(i = REGC_GPR8_LO_FIRST; (reg == REG_UNSET) && (i <= REGC_GPR8_LO_LAST); i++) {
+               reg = do_select_reg(state, used, i, classes);
+       }
+       for(i = REGC_DIVIDEND32_FIRST; (reg == REG_UNSET) && (i <= REGC_DIVIDEND32_LAST); i++) {
+               reg = do_select_reg(state, used, i, classes);
+       }
+       for(i = REGC_DIVIDEND64_FIRST; (reg == REG_UNSET) && (i <= REGC_DIVIDEND64_LAST); i++) {
+               reg = do_select_reg(state, used, i, classes);
+       }
+       for(i = REGC_FLAGS_FIRST; (reg == REG_UNSET) && (i <= REGC_FLAGS_LAST); i++) {
                reg = do_select_reg(state, used, i, classes);
        }
        return reg;
 }
 
 
-static unsigned arch_type_to_regcm(struct compile_state *state, struct type *type) 
+static unsigned arch_type_to_regcm(struct compile_state *state, struct type *type)
 {
+
+#if DEBUG_ROMCC_WARNINGS
 #warning "FIXME force types smaller (if legal) before I get here"
-       unsigned avail_mask;
+#endif
        unsigned mask;
        mask = 0;
-       avail_mask = arch_avail_mask(state);
        switch(type->type & TYPE_MASK) {
        case TYPE_ARRAY:
-       case TYPE_VOID: 
-               mask = 0; 
+       case TYPE_VOID:
+               mask = 0;
                break;
        case TYPE_CHAR:
        case TYPE_UCHAR:
-               mask = REGCM_GPR8 | 
-                       REGCM_GPR16 | REGCM_GPR16_8 | 
+               mask = REGCM_GPR8 | REGCM_GPR8_LO |
+                       REGCM_GPR16 | REGCM_GPR16_8 |
                        REGCM_GPR32 | REGCM_GPR32_8 |
-                       REGCM_GPR64 |
+                       REGCM_DIVIDEND32 | REGCM_DIVIDEND64 |
                        REGCM_MMX | REGCM_XMM |
                        REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8;
                break;
@@ -14742,32 +22405,51 @@ static unsigned arch_type_to_regcm(struct compile_state *state, struct type *typ
        case TYPE_USHORT:
                mask =  REGCM_GPR16 | REGCM_GPR16_8 |
                        REGCM_GPR32 | REGCM_GPR32_8 |
-                       REGCM_GPR64 |
+                       REGCM_DIVIDEND32 | REGCM_DIVIDEND64 |
                        REGCM_MMX | REGCM_XMM |
                        REGCM_IMM32 | REGCM_IMM16;
                break;
+       case TYPE_ENUM:
        case TYPE_INT:
        case TYPE_UINT:
        case TYPE_LONG:
        case TYPE_ULONG:
        case TYPE_POINTER:
                mask =  REGCM_GPR32 | REGCM_GPR32_8 |
-                       REGCM_GPR64 | REGCM_MMX | REGCM_XMM |
+                       REGCM_DIVIDEND32 | REGCM_DIVIDEND64 |
+                       REGCM_MMX | REGCM_XMM |
                        REGCM_IMM32;
                break;
+       case TYPE_JOIN:
+       case TYPE_UNION:
+               mask = arch_type_to_regcm(state, type->left);
+               break;
+       case TYPE_OVERLAP:
+               mask = arch_type_to_regcm(state, type->left) &
+                       arch_type_to_regcm(state, type->right);
+               break;
+       case TYPE_BITFIELD:
+               mask = arch_type_to_regcm(state, type->left);
+               break;
        default:
+               fprintf(state->errout, "type: ");
+               name_of(state->errout, type);
+               fprintf(state->errout, "\n");
                internal_error(state, 0, "no register class for type");
                break;
        }
-       mask &= avail_mask;
+       mask = arch_regcm_normalize(state, mask);
        return mask;
 }
 
 static int is_imm32(struct triple *imm)
 {
-       return ((imm->op == OP_INTCONST) && (imm->u.cval <= 0xffffffffUL)) ||
+       // second condition commented out to prevent compiler warning:
+       // imm->u.cval is always 32bit unsigned, so the comparison is
+       // always true.
+       return ((imm->op == OP_INTCONST) /* && (imm->u.cval <= 0xffffffffUL) */ ) ||
                (imm->op == OP_ADDRCONST);
-       
+
 }
 static int is_imm16(struct triple *imm)
 {
@@ -14810,111 +22492,217 @@ static int get_imm8(struct triple *ins, struct triple **expr)
        return 1;
 }
 
-#define TEMPLATE_NOP         0
-#define TEMPLATE_INTCONST8   1
-#define TEMPLATE_INTCONST32  2
-#define TEMPLATE_COPY_REG    3
-#define TEMPLATE_COPY_IMM32  4
-#define TEMPLATE_COPY_IMM16  5
-#define TEMPLATE_COPY_IMM8   6
-#define TEMPLATE_PHI         7
-#define TEMPLATE_STORE8      8
-#define TEMPLATE_STORE16     9
-#define TEMPLATE_STORE32    10
-#define TEMPLATE_LOAD8      11
-#define TEMPLATE_LOAD16     12
-#define TEMPLATE_LOAD32     13
-#define TEMPLATE_BINARY_REG 14
-#define TEMPLATE_BINARY_IMM 15
-#define TEMPLATE_SL_CL      16
-#define TEMPLATE_SL_IMM     17
-#define TEMPLATE_UNARY      18
-#define TEMPLATE_CMP_REG    19
-#define TEMPLATE_CMP_IMM    20
-#define TEMPLATE_TEST       21
-#define TEMPLATE_SET        22
-#define TEMPLATE_JMP        23
-#define TEMPLATE_INB_DX     24
-#define TEMPLATE_INB_IMM    25
-#define TEMPLATE_INW_DX     26
-#define TEMPLATE_INW_IMM    27
-#define TEMPLATE_INL_DX     28
-#define TEMPLATE_INL_IMM    29
-#define TEMPLATE_OUTB_DX    30
-#define TEMPLATE_OUTB_IMM   31
-#define TEMPLATE_OUTW_DX    32
-#define TEMPLATE_OUTW_IMM   33
-#define TEMPLATE_OUTL_DX    34
-#define TEMPLATE_OUTL_IMM   35
-#define TEMPLATE_BSF        36
-#define TEMPLATE_RDMSR      37
-#define TEMPLATE_WRMSR      38
-#define LAST_TEMPLATE       TEMPLATE_WRMSR
+#define TEMPLATE_NOP           0
+#define TEMPLATE_INTCONST8     1
+#define TEMPLATE_INTCONST32    2
+#define TEMPLATE_UNKNOWNVAL    3
+#define TEMPLATE_COPY8_REG     5
+#define TEMPLATE_COPY16_REG    6
+#define TEMPLATE_COPY32_REG    7
+#define TEMPLATE_COPY_IMM8     8
+#define TEMPLATE_COPY_IMM16    9
+#define TEMPLATE_COPY_IMM32   10
+#define TEMPLATE_PHI8         11
+#define TEMPLATE_PHI16        12
+#define TEMPLATE_PHI32        13
+#define TEMPLATE_STORE8       14
+#define TEMPLATE_STORE16      15
+#define TEMPLATE_STORE32      16
+#define TEMPLATE_LOAD8        17
+#define TEMPLATE_LOAD16       18
+#define TEMPLATE_LOAD32       19
+#define TEMPLATE_BINARY8_REG  20
+#define TEMPLATE_BINARY16_REG 21
+#define TEMPLATE_BINARY32_REG 22
+#define TEMPLATE_BINARY8_IMM  23
+#define TEMPLATE_BINARY16_IMM 24
+#define TEMPLATE_BINARY32_IMM 25
+#define TEMPLATE_SL8_CL       26
+#define TEMPLATE_SL16_CL      27
+#define TEMPLATE_SL32_CL      28
+#define TEMPLATE_SL8_IMM      29
+#define TEMPLATE_SL16_IMM     30
+#define TEMPLATE_SL32_IMM     31
+#define TEMPLATE_UNARY8       32
+#define TEMPLATE_UNARY16      33
+#define TEMPLATE_UNARY32      34
+#define TEMPLATE_CMP8_REG     35
+#define TEMPLATE_CMP16_REG    36
+#define TEMPLATE_CMP32_REG    37
+#define TEMPLATE_CMP8_IMM     38
+#define TEMPLATE_CMP16_IMM    39
+#define TEMPLATE_CMP32_IMM    40
+#define TEMPLATE_TEST8        41
+#define TEMPLATE_TEST16       42
+#define TEMPLATE_TEST32       43
+#define TEMPLATE_SET          44
+#define TEMPLATE_JMP          45
+#define TEMPLATE_RET          46
+#define TEMPLATE_INB_DX       47
+#define TEMPLATE_INB_IMM      48
+#define TEMPLATE_INW_DX       49
+#define TEMPLATE_INW_IMM      50
+#define TEMPLATE_INL_DX       51
+#define TEMPLATE_INL_IMM      52
+#define TEMPLATE_OUTB_DX      53
+#define TEMPLATE_OUTB_IMM     54
+#define TEMPLATE_OUTW_DX      55
+#define TEMPLATE_OUTW_IMM     56
+#define TEMPLATE_OUTL_DX      57
+#define TEMPLATE_OUTL_IMM     58
+#define TEMPLATE_BSF          59
+#define TEMPLATE_RDMSR        60
+#define TEMPLATE_WRMSR        61
+#define TEMPLATE_UMUL8        62
+#define TEMPLATE_UMUL16       63
+#define TEMPLATE_UMUL32       64
+#define TEMPLATE_DIV8         65
+#define TEMPLATE_DIV16        66
+#define TEMPLATE_DIV32        67
+#define LAST_TEMPLATE       TEMPLATE_DIV32
 #if LAST_TEMPLATE >= MAX_TEMPLATES
 #error "MAX_TEMPLATES to low"
 #endif
 
-#define COPY_REGCM (REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8 | REGCM_MMX | REGCM_XMM)
-#define COPY32_REGCM (REGCM_GPR32 | REGCM_MMX | REGCM_XMM)
+#define COPY8_REGCM     (REGCM_DIVIDEND64 | REGCM_DIVIDEND32 | REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO | REGCM_MMX | REGCM_XMM)
+#define COPY16_REGCM    (REGCM_DIVIDEND64 | REGCM_DIVIDEND32 | REGCM_GPR32 | REGCM_GPR16 | REGCM_MMX | REGCM_XMM)
+#define COPY32_REGCM    (REGCM_DIVIDEND64 | REGCM_DIVIDEND32 | REGCM_GPR32 | REGCM_MMX | REGCM_XMM)
+
 
 static struct ins_template templates[] = {
-       [TEMPLATE_NOP]      = {},
-       [TEMPLATE_INTCONST8] = { 
+       [TEMPLATE_NOP]      = {
+               .lhs = {
+                       [ 0] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 1] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 2] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 3] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 4] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 5] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 6] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 7] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 8] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [ 9] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [10] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [11] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [12] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [13] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [14] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [15] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [16] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [17] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [18] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [19] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [20] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [21] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [22] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [23] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [24] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [25] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [26] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [27] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [28] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [29] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [30] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [31] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [32] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [33] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [34] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [35] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [36] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [37] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [38] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [39] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [40] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [41] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [42] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [43] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [44] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [45] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [46] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [47] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [48] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [49] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [50] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [51] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [52] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [53] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [54] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [55] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [56] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [57] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [58] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [59] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [60] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [61] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [62] = { REG_UNNEEDED, REGCM_IMMALL },
+                       [63] = { REG_UNNEEDED, REGCM_IMMALL },
+               },
+       },
+       [TEMPLATE_INTCONST8] = {
                .lhs = { [0] = { REG_UNNEEDED, REGCM_IMM8 } },
        },
-       [TEMPLATE_INTCONST32] = { 
+       [TEMPLATE_INTCONST32] = {
                .lhs = { [0] = { REG_UNNEEDED, REGCM_IMM32 } },
        },
-       [TEMPLATE_COPY_REG] = {
-               .lhs = { [0] = { REG_UNSET, COPY_REGCM } },
-               .rhs = { [0] = { REG_UNSET, COPY_REGCM }  },
-       },
-       [TEMPLATE_COPY_IMM32] = {
+       [TEMPLATE_UNKNOWNVAL] = {
                .lhs = { [0] = { REG_UNSET, COPY32_REGCM } },
-               .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM32 } },
        },
-       [TEMPLATE_COPY_IMM16] = {
-               .lhs = { [0] = { REG_UNSET, COPY32_REGCM | REGCM_GPR16 } },
-               .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM16 } },
+       [TEMPLATE_COPY8_REG] = {
+               .lhs = { [0] = { REG_UNSET, COPY8_REGCM } },
+               .rhs = { [0] = { REG_UNSET, COPY8_REGCM }  },
+       },
+       [TEMPLATE_COPY16_REG] = {
+               .lhs = { [0] = { REG_UNSET, COPY16_REGCM } },
+               .rhs = { [0] = { REG_UNSET, COPY16_REGCM }  },
+       },
+       [TEMPLATE_COPY32_REG] = {
+               .lhs = { [0] = { REG_UNSET, COPY32_REGCM } },
+               .rhs = { [0] = { REG_UNSET, COPY32_REGCM }  },
        },
        [TEMPLATE_COPY_IMM8] = {
-               .lhs = { [0] = { REG_UNSET, COPY_REGCM } },
+               .lhs = { [0] = { REG_UNSET, COPY8_REGCM } },
                .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM8 } },
        },
-       [TEMPLATE_PHI] = { 
-               .lhs = { [0] = { REG_VIRT0, COPY_REGCM } },
-               .rhs = { 
-                       [ 0] = { REG_VIRT0, COPY_REGCM },
-                       [ 1] = { REG_VIRT0, COPY_REGCM },
-                       [ 2] = { REG_VIRT0, COPY_REGCM },
-                       [ 3] = { REG_VIRT0, COPY_REGCM },
-                       [ 4] = { REG_VIRT0, COPY_REGCM },
-                       [ 5] = { REG_VIRT0, COPY_REGCM },
-                       [ 6] = { REG_VIRT0, COPY_REGCM },
-                       [ 7] = { REG_VIRT0, COPY_REGCM },
-                       [ 8] = { REG_VIRT0, COPY_REGCM },
-                       [ 9] = { REG_VIRT0, COPY_REGCM },
-                       [10] = { REG_VIRT0, COPY_REGCM },
-                       [11] = { REG_VIRT0, COPY_REGCM },
-                       [12] = { REG_VIRT0, COPY_REGCM },
-                       [13] = { REG_VIRT0, COPY_REGCM },
-                       [14] = { REG_VIRT0, COPY_REGCM },
-                       [15] = { REG_VIRT0, COPY_REGCM },
-               }, },
+       [TEMPLATE_COPY_IMM16] = {
+               .lhs = { [0] = { REG_UNSET, COPY16_REGCM } },
+               .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM16 | REGCM_IMM8 } },
+       },
+       [TEMPLATE_COPY_IMM32] = {
+               .lhs = { [0] = { REG_UNSET, COPY32_REGCM } },
+               .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM32 | REGCM_IMM16 | REGCM_IMM8 } },
+       },
+       [TEMPLATE_PHI8] = {
+               .lhs = { [0] = { REG_VIRT0, COPY8_REGCM } },
+               .rhs = { [0] = { REG_VIRT0, COPY8_REGCM } },
+       },
+       [TEMPLATE_PHI16] = {
+               .lhs = { [0] = { REG_VIRT0, COPY16_REGCM } },
+               .rhs = { [0] = { REG_VIRT0, COPY16_REGCM } },
+       },
+       [TEMPLATE_PHI32] = {
+               .lhs = { [0] = { REG_VIRT0, COPY32_REGCM } },
+               .rhs = { [0] = { REG_VIRT0, COPY32_REGCM } },
+       },
        [TEMPLATE_STORE8] = {
-               .lhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
-               .rhs = { [0] = { REG_UNSET, REGCM_GPR8 } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR32 },
+                       [1] = { REG_UNSET, REGCM_GPR8_LO },
+               },
        },
        [TEMPLATE_STORE16] = {
-               .lhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
-               .rhs = { [0] = { REG_UNSET, REGCM_GPR16 } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR32 },
+                       [1] = { REG_UNSET, REGCM_GPR16 },
+               },
        },
        [TEMPLATE_STORE32] = {
-               .lhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
-               .rhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR32 },
+                       [1] = { REG_UNSET, REGCM_GPR32 },
+               },
        },
        [TEMPLATE_LOAD8] = {
-               .lhs = { [0] = { REG_UNSET, REGCM_GPR8 } },
+               .lhs = { [0] = { REG_UNSET, REGCM_GPR8_LO } },
                .rhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
        },
        [TEMPLATE_LOAD16] = {
@@ -14925,77 +22713,180 @@ static struct ins_template templates[] = {
                .lhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
                .rhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
        },
-       [TEMPLATE_BINARY_REG] = {
+       [TEMPLATE_BINARY8_REG] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR8_LO } },
+               .rhs = {
+                       [0] = { REG_VIRT0, REGCM_GPR8_LO },
+                       [1] = { REG_UNSET, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_BINARY16_REG] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR16 } },
+               .rhs = {
+                       [0] = { REG_VIRT0, REGCM_GPR16 },
+                       [1] = { REG_UNSET, REGCM_GPR16 },
+               },
+       },
+       [TEMPLATE_BINARY32_REG] = {
                .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
-               .rhs = { 
+               .rhs = {
                        [0] = { REG_VIRT0, REGCM_GPR32 },
                        [1] = { REG_UNSET, REGCM_GPR32 },
                },
        },
-       [TEMPLATE_BINARY_IMM] = {
+       [TEMPLATE_BINARY8_IMM] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR8_LO } },
+               .rhs = {
+                       [0] = { REG_VIRT0,    REGCM_GPR8_LO },
+                       [1] = { REG_UNNEEDED, REGCM_IMM8 },
+               },
+       },
+       [TEMPLATE_BINARY16_IMM] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR16 } },
+               .rhs = {
+                       [0] = { REG_VIRT0,    REGCM_GPR16 },
+                       [1] = { REG_UNNEEDED, REGCM_IMM16 },
+               },
+       },
+       [TEMPLATE_BINARY32_IMM] = {
                .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
-               .rhs = { 
+               .rhs = {
                        [0] = { REG_VIRT0,    REGCM_GPR32 },
                        [1] = { REG_UNNEEDED, REGCM_IMM32 },
                },
        },
-       [TEMPLATE_SL_CL] = {
-               .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
-               .rhs = { 
-                       [0] = { REG_VIRT0, REGCM_GPR32 },
-                       [1] = { REG_CL, REGCM_GPR8 },
+       [TEMPLATE_SL8_CL] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR8_LO } },
+               .rhs = {
+                       [0] = { REG_VIRT0, REGCM_GPR8_LO },
+                       [1] = { REG_CL, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_SL16_CL] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR16 } },
+               .rhs = {
+                       [0] = { REG_VIRT0, REGCM_GPR16 },
+                       [1] = { REG_CL, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_SL32_CL] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
+               .rhs = {
+                       [0] = { REG_VIRT0, REGCM_GPR32 },
+                       [1] = { REG_CL, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_SL8_IMM] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR8_LO } },
+               .rhs = {
+                       [0] = { REG_VIRT0,    REGCM_GPR8_LO },
+                       [1] = { REG_UNNEEDED, REGCM_IMM8 },
+               },
+       },
+       [TEMPLATE_SL16_IMM] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR16 } },
+               .rhs = {
+                       [0] = { REG_VIRT0,    REGCM_GPR16 },
+                       [1] = { REG_UNNEEDED, REGCM_IMM8 },
+               },
+       },
+       [TEMPLATE_SL32_IMM] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
+               .rhs = {
+                       [0] = { REG_VIRT0,    REGCM_GPR32 },
+                       [1] = { REG_UNNEEDED, REGCM_IMM8 },
+               },
+       },
+       [TEMPLATE_UNARY8] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR8_LO } },
+               .rhs = { [0] = { REG_VIRT0, REGCM_GPR8_LO } },
+       },
+       [TEMPLATE_UNARY16] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR16 } },
+               .rhs = { [0] = { REG_VIRT0, REGCM_GPR16 } },
+       },
+       [TEMPLATE_UNARY32] = {
+               .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
+               .rhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
+       },
+       [TEMPLATE_CMP8_REG] = {
+               .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR8_LO },
+                       [1] = { REG_UNSET, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_CMP16_REG] = {
+               .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR16 },
+                       [1] = { REG_UNSET, REGCM_GPR16 },
+               },
+       },
+       [TEMPLATE_CMP32_REG] = {
+               .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR32 },
+                       [1] = { REG_UNSET, REGCM_GPR32 },
                },
        },
-       [TEMPLATE_SL_IMM] = {
-               .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
-               .rhs = { 
-                       [0] = { REG_VIRT0,    REGCM_GPR32 },
+       [TEMPLATE_CMP8_IMM] = {
+               .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
+               .rhs = {
+                       [0] = { REG_UNSET, REGCM_GPR8_LO },
                        [1] = { REG_UNNEEDED, REGCM_IMM8 },
                },
        },
-       [TEMPLATE_UNARY] = {
-               .lhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
-               .rhs = { [0] = { REG_VIRT0, REGCM_GPR32 } },
-       },
-       [TEMPLATE_CMP_REG] = {
+       [TEMPLATE_CMP16_IMM] = {
                .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
                .rhs = {
-                       [0] = { REG_UNSET, REGCM_GPR32 },
-                       [1] = { REG_UNSET, REGCM_GPR32 },
+                       [0] = { REG_UNSET, REGCM_GPR16 },
+                       [1] = { REG_UNNEEDED, REGCM_IMM16 },
                },
        },
-       [TEMPLATE_CMP_IMM] = {
+       [TEMPLATE_CMP32_IMM] = {
                .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
                .rhs = {
                        [0] = { REG_UNSET, REGCM_GPR32 },
                        [1] = { REG_UNNEEDED, REGCM_IMM32 },
                },
        },
-       [TEMPLATE_TEST] = {
+       [TEMPLATE_TEST8] = {
+               .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
+               .rhs = { [0] = { REG_UNSET, REGCM_GPR8_LO } },
+       },
+       [TEMPLATE_TEST16] = {
+               .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
+               .rhs = { [0] = { REG_UNSET, REGCM_GPR16 } },
+       },
+       [TEMPLATE_TEST32] = {
                .lhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
                .rhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
        },
        [TEMPLATE_SET] = {
-               .lhs = { [0] = { REG_UNSET, REGCM_GPR8 } },
+               .lhs = { [0] = { REG_UNSET, REGCM_GPR8_LO } },
                .rhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
        },
        [TEMPLATE_JMP] = {
                .rhs = { [0] = { REG_EFLAGS, REGCM_FLAGS } },
        },
+       [TEMPLATE_RET] = {
+               .rhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
+       },
        [TEMPLATE_INB_DX] = {
-               .lhs = { [0] = { REG_AL,  REGCM_GPR8 } },  
+               .lhs = { [0] = { REG_AL,  REGCM_GPR8_LO } },
                .rhs = { [0] = { REG_DX, REGCM_GPR16 } },
        },
        [TEMPLATE_INB_IMM] = {
-               .lhs = { [0] = { REG_AL,  REGCM_GPR8 } },  
+               .lhs = { [0] = { REG_AL,  REGCM_GPR8_LO } },
                .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM8 } },
        },
-       [TEMPLATE_INW_DX]  = { 
-               .lhs = { [0] = { REG_AX,  REGCM_GPR16 } }, 
+       [TEMPLATE_INW_DX]  = {
+               .lhs = { [0] = { REG_AX,  REGCM_GPR16 } },
                .rhs = { [0] = { REG_DX, REGCM_GPR16 } },
        },
-       [TEMPLATE_INW_IMM] = { 
-               .lhs = { [0] = { REG_AX,  REGCM_GPR16 } }, 
+       [TEMPLATE_INW_IMM] = {
+               .lhs = { [0] = { REG_AX,  REGCM_GPR16 } },
                .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM8 } },
        },
        [TEMPLATE_INL_DX]  = {
@@ -15006,19 +22897,19 @@ static struct ins_template templates[] = {
                .lhs = { [0] = { REG_EAX, REGCM_GPR32 } },
                .rhs = { [0] = { REG_UNNEEDED, REGCM_IMM8 } },
        },
-       [TEMPLATE_OUTB_DX] = { 
+       [TEMPLATE_OUTB_DX] = {
                .rhs = {
-                       [0] = { REG_AL,  REGCM_GPR8 },
+                       [0] = { REG_AL,  REGCM_GPR8_LO },
                        [1] = { REG_DX, REGCM_GPR16 },
                },
        },
-       [TEMPLATE_OUTB_IMM] = { 
+       [TEMPLATE_OUTB_IMM] = {
                .rhs = {
-                       [0] = { REG_AL,  REGCM_GPR8 },  
+                       [0] = { REG_AL,  REGCM_GPR8_LO },
                        [1] = { REG_UNNEEDED, REGCM_IMM8 },
                },
        },
-       [TEMPLATE_OUTW_DX] = { 
+       [TEMPLATE_OUTW_DX] = {
                .rhs = {
                        [0] = { REG_AX,  REGCM_GPR16 },
                        [1] = { REG_DX, REGCM_GPR16 },
@@ -15026,19 +22917,19 @@ static struct ins_template templates[] = {
        },
        [TEMPLATE_OUTW_IMM] = {
                .rhs = {
-                       [0] = { REG_AX,  REGCM_GPR16 }, 
+                       [0] = { REG_AX,  REGCM_GPR16 },
                        [1] = { REG_UNNEEDED, REGCM_IMM8 },
                },
        },
-       [TEMPLATE_OUTL_DX] = { 
+       [TEMPLATE_OUTL_DX] = {
                .rhs = {
                        [0] = { REG_EAX, REGCM_GPR32 },
                        [1] = { REG_DX, REGCM_GPR16 },
                },
        },
-       [TEMPLATE_OUTL_IMM] = { 
+       [TEMPLATE_OUTL_IMM] = {
                .rhs = {
-                       [0] = { REG_EAX, REGCM_GPR32 }, 
+                       [0] = { REG_EAX, REGCM_GPR32 },
                        [1] = { REG_UNNEEDED, REGCM_IMM8 },
                },
        },
@@ -15047,7 +22938,7 @@ static struct ins_template templates[] = {
                .rhs = { [0] = { REG_UNSET, REGCM_GPR32 } },
        },
        [TEMPLATE_RDMSR] = {
-               .lhs = { 
+               .lhs = {
                        [0] = { REG_EAX, REGCM_GPR32 },
                        [1] = { REG_EDX, REGCM_GPR32 },
                },
@@ -15060,8 +22951,85 @@ static struct ins_template templates[] = {
                        [2] = { REG_EDX, REGCM_GPR32 },
                },
        },
+       [TEMPLATE_UMUL8] = {
+               .lhs = { [0] = { REG_AX, REGCM_GPR16 } },
+               .rhs = {
+                       [0] = { REG_AL, REGCM_GPR8_LO },
+                       [1] = { REG_UNSET, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_UMUL16] = {
+               .lhs = { [0] = { REG_DXAX, REGCM_DIVIDEND32 } },
+               .rhs = {
+                       [0] = { REG_AX, REGCM_GPR16 },
+                       [1] = { REG_UNSET, REGCM_GPR16 },
+               },
+       },
+       [TEMPLATE_UMUL32] = {
+               .lhs = { [0] = { REG_EDXEAX, REGCM_DIVIDEND64 } },
+               .rhs = {
+                       [0] = { REG_EAX, REGCM_GPR32 },
+                       [1] = { REG_UNSET, REGCM_GPR32 },
+               },
+       },
+       [TEMPLATE_DIV8] = {
+               .lhs = {
+                       [0] = { REG_AL, REGCM_GPR8_LO },
+                       [1] = { REG_AH, REGCM_GPR8 },
+               },
+               .rhs = {
+                       [0] = { REG_AX, REGCM_GPR16 },
+                       [1] = { REG_UNSET, REGCM_GPR8_LO },
+               },
+       },
+       [TEMPLATE_DIV16] = {
+               .lhs = {
+                       [0] = { REG_AX, REGCM_GPR16 },
+                       [1] = { REG_DX, REGCM_GPR16 },
+               },
+               .rhs = {
+                       [0] = { REG_DXAX, REGCM_DIVIDEND32 },
+                       [1] = { REG_UNSET, REGCM_GPR16 },
+               },
+       },
+       [TEMPLATE_DIV32] = {
+               .lhs = {
+                       [0] = { REG_EAX, REGCM_GPR32 },
+                       [1] = { REG_EDX, REGCM_GPR32 },
+               },
+               .rhs = {
+                       [0] = { REG_EDXEAX, REGCM_DIVIDEND64 },
+                       [1] = { REG_UNSET, REGCM_GPR32 },
+               },
+       },
 };
 
+static void fixup_branch(struct compile_state *state,
+       struct triple *branch, int jmp_op, int cmp_op, struct type *cmp_type,
+       struct triple *left, struct triple *right)
+{
+       struct triple *test;
+       if (!left) {
+               internal_error(state, branch, "no branch test?");
+       }
+       test = pre_triple(state, branch,
+               cmp_op, cmp_type, left, right);
+       test->template_id = TEMPLATE_TEST32;
+       if (cmp_op == OP_CMP) {
+               test->template_id = TEMPLATE_CMP32_REG;
+               if (get_imm32(test, &RHS(test, 1))) {
+                       test->template_id = TEMPLATE_CMP32_IMM;
+               }
+       }
+       use_triple(RHS(test, 0), test);
+       use_triple(RHS(test, 1), test);
+       unuse_triple(RHS(branch, 0), branch);
+       RHS(branch, 0) = test;
+       branch->op = jmp_op;
+       branch->template_id = TEMPLATE_JMP;
+       use_triple(RHS(branch, 0), branch);
+}
+
 static void fixup_branches(struct compile_state *state,
        struct triple *cmp, struct triple *use, int jmp_op)
 {
@@ -15071,40 +23039,26 @@ static void fixup_branches(struct compile_state *state,
                if (entry->member->op == OP_COPY) {
                        fixup_branches(state, cmp, entry->member, jmp_op);
                }
-               else if (entry->member->op == OP_BRANCH) {
-                       struct triple *branch, *test;
+               else if (entry->member->op == OP_CBRANCH) {
+                       struct triple *branch;
                        struct triple *left, *right;
                        left = right = 0;
                        left = RHS(cmp, 0);
-                       if (TRIPLE_RHS(cmp->sizes) > 1) {
+                       if (cmp->rhs > 1) {
                                right = RHS(cmp, 1);
                        }
                        branch = entry->member;
-                       test = pre_triple(state, branch,
+                       fixup_branch(state, branch, jmp_op,
                                cmp->op, cmp->type, left, right);
-                       test->template_id = TEMPLATE_TEST; 
-                       if (cmp->op == OP_CMP) {
-                               test->template_id = TEMPLATE_CMP_REG;
-                               if (get_imm32(test, &RHS(test, 1))) {
-                                       test->template_id = TEMPLATE_CMP_IMM;
-                               }
-                       }
-                       use_triple(RHS(test, 0), test);
-                       use_triple(RHS(test, 1), test);
-                       unuse_triple(RHS(branch, 0), branch);
-                       RHS(branch, 0) = test;
-                       branch->op = jmp_op;
-                       branch->template_id = TEMPLATE_JMP;
-                       use_triple(RHS(branch, 0), branch);
                }
        }
 }
 
-static void bool_cmp(struct compile_state *state, 
+static void bool_cmp(struct compile_state *state,
        struct triple *ins, int cmp_op, int jmp_op, int set_op)
 {
        struct triple_set *entry, *next;
-       struct triple *set;
+       struct triple *set, *convert;
 
        /* Put a barrier up before the cmp which preceeds the
         * copy instruction.  If a set actually occurs this gives
@@ -15113,50 +23067,35 @@ static void bool_cmp(struct compile_state *state,
 
        /* Modify the comparison operator */
        ins->op = cmp_op;
-       ins->template_id = TEMPLATE_TEST;
+       ins->template_id = TEMPLATE_TEST32;
        if (cmp_op == OP_CMP) {
-               ins->template_id = TEMPLATE_CMP_REG;
+               ins->template_id = TEMPLATE_CMP32_REG;
                if (get_imm32(ins, &RHS(ins, 1))) {
-                       ins->template_id =  TEMPLATE_CMP_IMM;
+                       ins->template_id =  TEMPLATE_CMP32_IMM;
                }
        }
        /* Generate the instruction sequence that will transform the
         * result of the comparison into a logical value.
         */
-       set = post_triple(state, ins, set_op, ins->type, ins, 0);
+       set = post_triple(state, ins, set_op, &uchar_type, ins, 0);
        use_triple(ins, set);
        set->template_id = TEMPLATE_SET;
 
+       convert = set;
+       if (!equiv_types(ins->type, set->type)) {
+               convert = post_triple(state, set, OP_CONVERT, ins->type, set, 0);
+               use_triple(set, convert);
+               convert->template_id = TEMPLATE_COPY32_REG;
+       }
+
        for(entry = ins->use; entry; entry = next) {
                next = entry->next;
                if (entry->member == set) {
                        continue;
                }
-               replace_rhs_use(state, ins, set, entry->member);
-       }
-       fixup_branches(state, ins, set, jmp_op);
-}
-
-static struct triple *after_lhs(struct compile_state *state, struct triple *ins)
-{
-       struct triple *next;
-       int lhs, i;
-       lhs = TRIPLE_LHS(ins->sizes);
-       for(next = ins->next, i = 0; i < lhs; i++, next = next->next) {
-               if (next != LHS(ins, i)) {
-                       internal_error(state, ins, "malformed lhs on %s",
-                               tops(ins->op));
-               }
-               if (next->op != OP_PIECE) {
-                       internal_error(state, ins, "bad lhs op %s at %d on %s",
-                               tops(next->op), i, tops(ins->op));
-               }
-               if (next->u.cval != i) {
-                       internal_error(state, ins, "bad u.cval of %d %d expected",
-                               next->u.cval, i);
-               }
+               replace_rhs_use(state, ins, convert, entry->member);
        }
-       return next;
+       fixup_branches(state, ins, convert, jmp_op);
 }
 
 struct reg_info arch_reg_lhs(struct compile_state *state, struct triple *ins, int index)
@@ -15168,12 +23107,12 @@ struct reg_info arch_reg_lhs(struct compile_state *state, struct triple *ins, in
                index = ins->u.cval;
                ins = MISC(ins, 0);
        }
-       zlhs = TRIPLE_LHS(ins->sizes);
+       zlhs = ins->lhs;
        if (triple_is_def(state, ins)) {
                zlhs = 1;
        }
        if (index >= zlhs) {
-               internal_error(state, ins, "index %d out of range for %s\n",
+               internal_error(state, ins, "index %d out of range for %s",
                        index, tops(ins->op));
        }
        switch(ins->op) {
@@ -15182,7 +23121,7 @@ struct reg_info arch_reg_lhs(struct compile_state *state, struct triple *ins, in
                break;
        default:
                if (ins->template_id > LAST_TEMPLATE) {
-                       internal_error(state, ins, "bad template number %d", 
+                       internal_error(state, ins, "bad template number %d",
                                ins->template_id);
                }
                template = &templates[ins->template_id];
@@ -15203,7 +23142,7 @@ struct reg_info arch_reg_rhs(struct compile_state *state, struct triple *ins, in
 {
        struct reg_info result;
        struct ins_template *template;
-       if ((index > TRIPLE_RHS(ins->sizes)) ||
+       if ((index > ins->rhs) ||
                (ins->op == OP_PIECE)) {
                internal_error(state, ins, "index %d out of range for %s\n",
                        index, tops(ins->op));
@@ -15212,9 +23151,12 @@ struct reg_info arch_reg_rhs(struct compile_state *state, struct triple *ins, in
        case OP_ASM:
                template = &ins->u.ainfo->tmpl;
                break;
+       case OP_PHI:
+               index = 0;
+               /* Fall through */
        default:
                if (ins->template_id > LAST_TEMPLATE) {
-                       internal_error(state, ins, "bad template number %d", 
+                       internal_error(state, ins, "bad template number %d",
                                ins->template_id);
                }
                template = &templates[ins->template_id];
@@ -15228,16 +23170,168 @@ struct reg_info arch_reg_rhs(struct compile_state *state, struct triple *ins, in
        return result;
 }
 
+static struct triple *mod_div(struct compile_state *state,
+       struct triple *ins, int div_op, int index)
+{
+       struct triple *div, *piece1;
+
+       /* Generate the appropriate division instruction */
+       div = post_triple(state, ins, div_op, ins->type, 0, 0);
+       RHS(div, 0) = RHS(ins, 0);
+       RHS(div, 1) = RHS(ins, 1);
+       piece1 = LHS(div, 1);
+       div->template_id  = TEMPLATE_DIV32;
+       use_triple(RHS(div, 0), div);
+       use_triple(RHS(div, 1), div);
+       use_triple(LHS(div, 0), div);
+       use_triple(LHS(div, 1), div);
+
+       /* Replate uses of ins with the appropriate piece of the div */
+       propogate_use(state, ins, LHS(div, index));
+       release_triple(state, ins);
+
+       /* Return the address of the next instruction */
+       return piece1->next;
+}
+
+static int noop_adecl(struct triple *adecl)
+{
+       struct triple_set *use;
+       /* It's a noop if it doesn't specify stoorage */
+       if (adecl->lhs == 0) {
+               return 1;
+       }
+       /* Is the adecl used? If not it's a noop */
+       for(use = adecl->use; use ; use = use->next) {
+               if ((use->member->op != OP_PIECE) ||
+                       (MISC(use->member, 0) != adecl)) {
+                       return 0;
+               }
+       }
+       return 1;
+}
+
+static struct triple *x86_deposit(struct compile_state *state, struct triple *ins)
+{
+       struct triple *mask, *nmask, *shift;
+       struct triple *val, *val_mask, *val_shift;
+       struct triple *targ, *targ_mask;
+       struct triple *new;
+       ulong_t the_mask, the_nmask;
+
+       targ = RHS(ins, 0);
+       val = RHS(ins, 1);
+
+       /* Get constant for the mask value */
+       the_mask = 1;
+       the_mask <<= ins->u.bitfield.size;
+       the_mask -= 1;
+       the_mask <<= ins->u.bitfield.offset;
+       mask = pre_triple(state, ins, OP_INTCONST, &uint_type, 0, 0);
+       mask->u.cval = the_mask;
+
+       /* Get the inverted mask value */
+       the_nmask = ~the_mask;
+       nmask = pre_triple(state, ins, OP_INTCONST, &uint_type, 0, 0);
+       nmask->u.cval = the_nmask;
+
+       /* Get constant for the shift value */
+       shift = pre_triple(state, ins, OP_INTCONST, &uint_type, 0, 0);
+       shift->u.cval = ins->u.bitfield.offset;
+
+       /* Shift and mask the source value */
+       val_shift = val;
+       if (shift->u.cval != 0) {
+               val_shift = pre_triple(state, ins, OP_SL, val->type, val, shift);
+               use_triple(val, val_shift);
+               use_triple(shift, val_shift);
+       }
+       val_mask = val_shift;
+       if (is_signed(val->type)) {
+               val_mask = pre_triple(state, ins, OP_AND, val->type, val_shift, mask);
+               use_triple(val_shift, val_mask);
+               use_triple(mask, val_mask);
+       }
+
+       /* Mask the target value */
+       targ_mask = pre_triple(state, ins, OP_AND, targ->type, targ, nmask);
+       use_triple(targ, targ_mask);
+       use_triple(nmask, targ_mask);
+
+       /* Now combined them together */
+       new = pre_triple(state, ins, OP_OR, targ->type, targ_mask, val_mask);
+       use_triple(targ_mask, new);
+       use_triple(val_mask, new);
+
+       /* Move all of the users over to the new expression */
+       propogate_use(state, ins, new);
+
+       /* Delete the original triple */
+       release_triple(state, ins);
+
+       /* Restart the transformation at mask */
+       return mask;
+}
+
+static struct triple *x86_extract(struct compile_state *state, struct triple *ins)
+{
+       struct triple *mask, *shift;
+       struct triple *val, *val_mask, *val_shift;
+       ulong_t the_mask;
+
+       val = RHS(ins, 0);
+
+       /* Get constant for the mask value */
+       the_mask = 1;
+       the_mask <<= ins->u.bitfield.size;
+       the_mask -= 1;
+       mask = pre_triple(state, ins, OP_INTCONST, &int_type, 0, 0);
+       mask->u.cval = the_mask;
+
+       /* Get constant for the right shift value */
+       shift = pre_triple(state, ins, OP_INTCONST, &int_type, 0, 0);
+       shift->u.cval = ins->u.bitfield.offset;
+
+       /* Shift arithmetic right, to correct the sign */
+       val_shift = val;
+       if (shift->u.cval != 0) {
+               int op;
+               if (ins->op == OP_SEXTRACT) {
+                       op = OP_SSR;
+               } else {
+                       op = OP_USR;
+               }
+               val_shift = pre_triple(state, ins, op, val->type, val, shift);
+               use_triple(val, val_shift);
+               use_triple(shift, val_shift);
+       }
+
+       /* Finally mask the value */
+       val_mask = pre_triple(state, ins, OP_AND, ins->type, val_shift, mask);
+       use_triple(val_shift, val_mask);
+       use_triple(mask,      val_mask);
+
+       /* Move all of the users over to the new expression */
+       propogate_use(state, ins, val_mask);
+
+       /* Release the original instruction */
+       release_triple(state, ins);
+
+       return mask;
+
+}
+
 static struct triple *transform_to_arch_instruction(
        struct compile_state *state, struct triple *ins)
 {
        /* Transform from generic 3 address instructions
         * to archtecture specific instructions.
-        * And apply architecture specific constrains to instructions.
+        * And apply architecture specific constraints to instructions.
         * Copies are inserted to preserve the register flexibility
         * of 3 address instructions.
         */
-       struct triple *next;
+       struct triple *next, *value;
+       size_t size;
        next = ins->next;
        switch(ins->op) {
        case OP_INTCONST:
@@ -15249,6 +23343,9 @@ static struct triple *transform_to_arch_instruction(
        case OP_ADDRCONST:
                ins->template_id = TEMPLATE_INTCONST32;
                break;
+       case OP_UNKNOWNVAL:
+               ins->template_id = TEMPLATE_UNKNOWNVAL;
+               break;
        case OP_NOOP:
        case OP_SDECL:
        case OP_BLOBCONST:
@@ -15256,22 +23353,58 @@ static struct triple *transform_to_arch_instruction(
                ins->template_id = TEMPLATE_NOP;
                break;
        case OP_COPY:
-               ins->template_id = TEMPLATE_COPY_REG;
-               if (is_imm8(RHS(ins, 0))) {
+       case OP_CONVERT:
+               size = size_of(state, ins->type);
+               value = RHS(ins, 0);
+               if (is_imm8(value) && (size <= SIZEOF_I8)) {
                        ins->template_id = TEMPLATE_COPY_IMM8;
                }
-               else if (is_imm16(RHS(ins, 0))) {
+               else if (is_imm16(value) && (size <= SIZEOF_I16)) {
                        ins->template_id = TEMPLATE_COPY_IMM16;
                }
-               else if (is_imm32(RHS(ins, 0))) {
+               else if (is_imm32(value) && (size <= SIZEOF_I32)) {
                        ins->template_id = TEMPLATE_COPY_IMM32;
                }
-               else if (is_const(RHS(ins, 0))) {
+               else if (is_const(value)) {
                        internal_error(state, ins, "bad constant passed to copy");
                }
+               else if (size <= SIZEOF_I8) {
+                       ins->template_id = TEMPLATE_COPY8_REG;
+               }
+               else if (size <= SIZEOF_I16) {
+                       ins->template_id = TEMPLATE_COPY16_REG;
+               }
+               else if (size <= SIZEOF_I32) {
+                       ins->template_id = TEMPLATE_COPY32_REG;
+               }
+               else {
+                       internal_error(state, ins, "bad type passed to copy");
+               }
                break;
        case OP_PHI:
-               ins->template_id = TEMPLATE_PHI;
+               size = size_of(state, ins->type);
+               if (size <= SIZEOF_I8) {
+                       ins->template_id = TEMPLATE_PHI8;
+               }
+               else if (size <= SIZEOF_I16) {
+                       ins->template_id = TEMPLATE_PHI16;
+               }
+               else if (size <= SIZEOF_I32) {
+                       ins->template_id = TEMPLATE_PHI32;
+               }
+               else {
+                       internal_error(state, ins, "bad type passed to phi");
+               }
+               break;
+       case OP_ADECL:
+               /* Adecls should always be treated as dead code and
+                * removed.  If we are not optimizing they may linger.
+                */
+               if (!noop_adecl(ins)) {
+                       internal_error(state, ins, "adecl remains?");
+               }
+               ins->template_id = TEMPLATE_NOP;
+               next = after_lhs(state, ins);
                break;
        case OP_STORE:
                switch(ins->type->type & TYPE_MASK) {
@@ -15294,23 +23427,16 @@ static struct triple *transform_to_arch_instruction(
        case OP_LOAD:
                switch(ins->type->type & TYPE_MASK) {
                case TYPE_CHAR:   case TYPE_UCHAR:
-                       ins->template_id = TEMPLATE_LOAD8;
-                       break;
-               case TYPE_SHORT:
-               case TYPE_USHORT:
-                       ins->template_id = TEMPLATE_LOAD16;
-                       break;
-               case TYPE_INT:
-               case TYPE_UINT:
-               case TYPE_LONG:
-               case TYPE_ULONG:
+               case TYPE_SHORT:  case TYPE_USHORT:
+               case TYPE_INT:    case TYPE_UINT:
+               case TYPE_LONG:   case TYPE_ULONG:
                case TYPE_POINTER:
-                       ins->template_id = TEMPLATE_LOAD32;
                        break;
                default:
                        internal_error(state, ins, "unknown type in load");
                        break;
                }
+               ins->template_id = TEMPLATE_LOAD32;
                break;
        case OP_ADD:
        case OP_SUB:
@@ -15318,25 +23444,47 @@ static struct triple *transform_to_arch_instruction(
        case OP_XOR:
        case OP_OR:
        case OP_SMUL:
-               ins->template_id = TEMPLATE_BINARY_REG;
+               ins->template_id = TEMPLATE_BINARY32_REG;
                if (get_imm32(ins, &RHS(ins, 1))) {
-                       ins->template_id = TEMPLATE_BINARY_IMM;
+                       ins->template_id = TEMPLATE_BINARY32_IMM;
                }
                break;
+       case OP_SDIVT:
+       case OP_UDIVT:
+               ins->template_id = TEMPLATE_DIV32;
+               next = after_lhs(state, ins);
+               break;
+       case OP_UMUL:
+               ins->template_id = TEMPLATE_UMUL32;
+               break;
+       case OP_UDIV:
+               next = mod_div(state, ins, OP_UDIVT, 0);
+               break;
+       case OP_SDIV:
+               next = mod_div(state, ins, OP_SDIVT, 0);
+               break;
+       case OP_UMOD:
+               next = mod_div(state, ins, OP_UDIVT, 1);
+               break;
+       case OP_SMOD:
+               next = mod_div(state, ins, OP_SDIVT, 1);
+               break;
        case OP_SL:
        case OP_SSR:
        case OP_USR:
-               ins->template_id = TEMPLATE_SL_CL;
+               ins->template_id = TEMPLATE_SL32_CL;
                if (get_imm8(ins, &RHS(ins, 1))) {
-                       ins->template_id = TEMPLATE_SL_IMM;
+                       ins->template_id = TEMPLATE_SL32_IMM;
+               } else if (size_of(state, RHS(ins, 1)->type) > SIZEOF_CHAR) {
+                       typed_pre_copy(state, &uchar_type, ins, 1);
                }
                break;
        case OP_INVERT:
        case OP_NEG:
-               ins->template_id = TEMPLATE_UNARY;
+               ins->template_id = TEMPLATE_UNARY32;
                break;
-       case OP_EQ: 
-               bool_cmp(state, ins, OP_CMP, OP_JMP_EQ, OP_SET_EQ); 
+       case OP_EQ:
+               bool_cmp(state, ins, OP_CMP, OP_JMP_EQ, OP_SET_EQ);
                break;
        case OP_NOTEQ:
                bool_cmp(state, ins, OP_CMP, OP_JMP_NOTEQ, OP_SET_NOTEQ);
@@ -15372,12 +23520,19 @@ static struct triple *transform_to_arch_instruction(
                bool_cmp(state, ins, OP_TEST, OP_JMP_EQ, OP_SET_EQ);
                break;
        case OP_BRANCH:
-               if (TRIPLE_RHS(ins->sizes) > 0) {
-                       internal_error(state, ins, "bad branch test");
-               }
                ins->op = OP_JMP;
                ins->template_id = TEMPLATE_NOP;
                break;
+       case OP_CBRANCH:
+               fixup_branch(state, ins, OP_JMP_NOTEQ, OP_TEST,
+                       RHS(ins, 0)->type, RHS(ins, 0), 0);
+               break;
+       case OP_CALL:
+               ins->template_id = TEMPLATE_NOP;
+               break;
+       case OP_RET:
+               ins->template_id = TEMPLATE_RET;
+               break;
        case OP_INB:
        case OP_INW:
        case OP_INL:
@@ -15422,14 +23577,17 @@ static struct triple *transform_to_arch_instruction(
                break;
                /* Already transformed instructions */
        case OP_TEST:
-               ins->template_id = TEMPLATE_TEST;
+               ins->template_id = TEMPLATE_TEST32;
                break;
        case OP_CMP:
-               ins->template_id = TEMPLATE_CMP_REG;
+               ins->template_id = TEMPLATE_CMP32_REG;
                if (get_imm32(ins, &RHS(ins, 1))) {
-                       ins->template_id = TEMPLATE_CMP_IMM;
+                       ins->template_id = TEMPLATE_CMP32_IMM;
                }
                break;
+       case OP_JMP:
+               ins->template_id = TEMPLATE_NOP;
+               break;
        case OP_JMP_EQ:      case OP_JMP_NOTEQ:
        case OP_JMP_SLESS:   case OP_JMP_ULESS:
        case OP_JMP_SMORE:   case OP_JMP_UMORE:
@@ -15444,38 +23602,48 @@ static struct triple *transform_to_arch_instruction(
        case OP_SET_SMOREEQ: case OP_SET_UMOREEQ:
                ins->template_id = TEMPLATE_SET;
                break;
+       case OP_DEPOSIT:
+               next = x86_deposit(state, ins);
+               break;
+       case OP_SEXTRACT:
+       case OP_UEXTRACT:
+               next = x86_extract(state, ins);
+               break;
                /* Unhandled instructions */
        case OP_PIECE:
        default:
-               internal_error(state, ins, "unhandled ins: %d %s\n",
+               internal_error(state, ins, "unhandled ins: %d %s",
                        ins->op, tops(ins->op));
                break;
        }
        return next;
 }
 
+static long next_label(struct compile_state *state)
+{
+       static long label_counter = 1000;
+       return ++label_counter;
+}
 static void generate_local_labels(struct compile_state *state)
 {
        struct triple *first, *label;
-       int label_counter;
-       label_counter = 0;
-       first = RHS(state->main_function, 0);
+       first = state->first;
        label = first;
        do {
-               if ((label->op == OP_LABEL) || 
+               if ((label->op == OP_LABEL) ||
                        (label->op == OP_SDECL)) {
                        if (label->use) {
-                               label->u.cval = ++label_counter;
+                               label->u.cval = next_label(state);
                        } else {
                                label->u.cval = 0;
                        }
-                       
+
                }
                label = label->next;
        } while(label != first);
 }
 
-static int check_reg(struct compile_state *state, 
+static int check_reg(struct compile_state *state,
        struct triple *triple, int classes)
 {
        unsigned mask;
@@ -15492,27 +23660,31 @@ static int check_reg(struct compile_state *state,
        return reg;
 }
 
+
+#if REG_XMM7 != 44
+#error "Registers have renumberd fix arch_reg_str"
+#endif
+static const char *arch_regs[] = {
+       "%unset",
+       "%unneeded",
+       "%eflags",
+       "%al", "%bl", "%cl", "%dl", "%ah", "%bh", "%ch", "%dh",
+       "%ax", "%bx", "%cx", "%dx", "%si", "%di", "%bp", "%sp",
+       "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi", "%ebp", "%esp",
+       "%edx:%eax",
+       "%dx:%ax",
+       "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7",
+       "%xmm0", "%xmm1", "%xmm2", "%xmm3",
+       "%xmm4", "%xmm5", "%xmm6", "%xmm7",
+};
 static const char *arch_reg_str(int reg)
 {
-       static const char *regs[] = {
-               "%bad_register",
-               "%bad_register2",
-               "%eflags",
-               "%al", "%bl", "%cl", "%dl", "%ah", "%bh", "%ch", "%dh",
-               "%ax", "%bx", "%cx", "%dx", "%si", "%di", "%bp", "%sp",
-               "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi", "%ebp", "%esp",
-               "%edx:%eax",
-               "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7",
-               "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
-               "%xmm4", "%xmm5", "%xmm6", "%xmm7",
-       };
        if (!((reg >= REG_EFLAGS) && (reg <= REG_XMM7))) {
                reg = 0;
        }
-       return regs[reg];
+       return arch_regs[reg];
 }
 
-
 static const char *reg(struct compile_state *state, struct triple *triple,
        int classes)
 {
@@ -15521,13 +23693,56 @@ static const char *reg(struct compile_state *state, struct triple *triple,
        return arch_reg_str(reg);
 }
 
+static int arch_reg_size(int reg)
+{
+       int size;
+       size = 0;
+       if (reg == REG_EFLAGS) {
+               size = 32;
+       }
+       else if ((reg >= REG_AL) && (reg <= REG_DH)) {
+               size = 8;
+       }
+       else if ((reg >= REG_AX) && (reg <= REG_SP)) {
+               size = 16;
+       }
+       else if ((reg >= REG_EAX) && (reg <= REG_ESP)) {
+               size = 32;
+       }
+       else if (reg == REG_EDXEAX) {
+               size = 64;
+       }
+       else if (reg == REG_DXAX) {
+               size = 32;
+       }
+       else if ((reg >= REG_MMX0) && (reg <= REG_MMX7)) {
+               size = 64;
+       }
+       else if ((reg >= REG_XMM0) && (reg <= REG_XMM7)) {
+               size = 128;
+       }
+       return size;
+}
+
+static int reg_size(struct compile_state *state, struct triple *ins)
+{
+       int reg;
+       reg = ID_REG(ins->id);
+       if (reg == REG_UNSET) {
+               internal_error(state, ins, "register not set");
+       }
+       return arch_reg_size(reg);
+}
+
+
+
 const char *type_suffix(struct compile_state *state, struct type *type)
 {
        const char *suffix;
        switch(size_of(state, type)) {
-       case 1: suffix = "b"; break;
-       case 2: suffix = "w"; break;
-       case 4: suffix = "l"; break;
+       case SIZEOF_I8:  suffix = "b"; break;
+       case SIZEOF_I16: suffix = "w"; break;
+       case SIZEOF_I32: suffix = "l"; break;
        default:
                internal_error(state, 0, "unknown suffix");
                suffix = 0;
@@ -15541,14 +23756,22 @@ static void print_const_val(
 {
        switch(ins->op) {
        case OP_INTCONST:
-               fprintf(fp, " $%ld ", 
-                       (long_t)(ins->u.cval));
+               fprintf(fp, " $%ld ",
+                       (long)(ins->u.cval));
                break;
        case OP_ADDRCONST:
+               if ((MISC(ins, 0)->op != OP_SDECL) &&
+                       (MISC(ins, 0)->op != OP_LABEL))
+               {
+                       internal_error(state, ins, "bad base for addrconst");
+               }
+               if (MISC(ins, 0)->u.cval <= 0) {
+                       internal_error(state, ins, "unlabeled constant");
+               }
                fprintf(fp, " $L%s%lu+%lu ",
-                       state->label_prefix, 
-                       MISC(ins, 0)->u.cval,
-                       ins->u.cval);
+                       state->compiler->label_prefix,
+                       (unsigned long)(MISC(ins, 0)->u.cval),
+                       (unsigned long)(ins->u.cval));
                break;
        default:
                internal_error(state, ins, "unknown constant type");
@@ -15556,12 +23779,114 @@ static void print_const_val(
        }
 }
 
+static void print_const(struct compile_state *state,
+       struct triple *ins, FILE *fp)
+{
+       switch(ins->op) {
+       case OP_INTCONST:
+               switch(ins->type->type & TYPE_MASK) {
+               case TYPE_CHAR:
+               case TYPE_UCHAR:
+                       fprintf(fp, ".byte 0x%02lx\n",
+                               (unsigned long)(ins->u.cval));
+                       break;
+               case TYPE_SHORT:
+               case TYPE_USHORT:
+                       fprintf(fp, ".short 0x%04lx\n",
+                               (unsigned long)(ins->u.cval));
+                       break;
+               case TYPE_INT:
+               case TYPE_UINT:
+               case TYPE_LONG:
+               case TYPE_ULONG:
+               case TYPE_POINTER:
+                       fprintf(fp, ".int %lu\n",
+                               (unsigned long)(ins->u.cval));
+                       break;
+               default:
+                       fprintf(state->errout, "type: ");
+                       name_of(state->errout, ins->type);
+                       fprintf(state->errout, "\n");
+                       internal_error(state, ins, "Unknown constant type. Val: %lu",
+                               (unsigned long)(ins->u.cval));
+               }
+
+               break;
+       case OP_ADDRCONST:
+               if ((MISC(ins, 0)->op != OP_SDECL) &&
+                       (MISC(ins, 0)->op != OP_LABEL)) {
+                       internal_error(state, ins, "bad base for addrconst");
+               }
+               if (MISC(ins, 0)->u.cval <= 0) {
+                       internal_error(state, ins, "unlabeled constant");
+               }
+               fprintf(fp, ".int L%s%lu+%lu\n",
+                       state->compiler->label_prefix,
+                       (unsigned long)(MISC(ins, 0)->u.cval),
+                       (unsigned long)(ins->u.cval));
+               break;
+       case OP_BLOBCONST:
+       {
+               unsigned char *blob;
+               size_t size, i;
+               size = size_of_in_bytes(state, ins->type);
+               blob = ins->u.blob;
+               for(i = 0; i < size; i++) {
+                       fprintf(fp, ".byte 0x%02x\n",
+                               blob[i]);
+               }
+               break;
+       }
+       default:
+               internal_error(state, ins, "Unknown constant type");
+               break;
+       }
+}
+
+#define TEXT_SECTION ".rom.text"
+#define DATA_SECTION ".rom.data"
+
+static long get_const_pool_ref(
+       struct compile_state *state, struct triple *ins, size_t size, FILE *fp)
+{
+       size_t fill_bytes;
+       long ref;
+       ref = next_label(state);
+       fprintf(fp, ".section \"" DATA_SECTION "\"\n");
+       fprintf(fp, ".balign %ld\n", (long int)align_of_in_bytes(state, ins->type));
+       fprintf(fp, "L%s%lu:\n", state->compiler->label_prefix, ref);
+       print_const(state, ins, fp);
+       fill_bytes = bits_to_bytes(size - size_of(state, ins->type));
+       if (fill_bytes) {
+               fprintf(fp, ".fill %ld, 1, 0\n", (long int)fill_bytes);
+       }
+       fprintf(fp, ".section \"" TEXT_SECTION "\"\n");
+       return ref;
+}
+
+static long get_mask_pool_ref(
+       struct compile_state *state, struct triple *ins, unsigned long mask, FILE *fp)
+{
+       long ref;
+       if (mask == 0xff) {
+               ref = 1;
+       }
+       else if (mask == 0xffff) {
+               ref = 2;
+       }
+       else {
+               ref = 0;
+               internal_error(state, ins, "unhandled mask value");
+       }
+       return ref;
+}
+
 static void print_binary_op(struct compile_state *state,
-       const char *op, struct triple *ins, FILE *fp) 
+       const char *op, struct triple *ins, FILE *fp)
 {
        unsigned mask;
-       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8;
-       if (RHS(ins, 0)->id != ins->id) {
+       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO;
+       if (ID_REG(RHS(ins, 0)->id) != ID_REG(ins->id)) {
                internal_error(state, ins, "invalid register assignment");
        }
        if (is_const(RHS(ins, 1))) {
@@ -15584,11 +23909,11 @@ static void print_binary_op(struct compile_state *state,
                        reg(state, RHS(ins, 0), mask));
        }
 }
-static void print_unary_op(struct compile_state *state, 
+static void print_unary_op(struct compile_state *state,
        const char *op, struct triple *ins, FILE *fp)
 {
        unsigned mask;
-       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8;
+       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO;
        fprintf(fp, "\t%s %s\n",
                op,
                reg(state, RHS(ins, 0), mask));
@@ -15598,8 +23923,8 @@ static void print_op_shift(struct compile_state *state,
        const char *op, struct triple *ins, FILE *fp)
 {
        unsigned mask;
-       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8;
-       if (RHS(ins, 0)->id != ins->id) {
+       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO;
+       if (ID_REG(RHS(ins, 0)->id) != ID_REG(ins->id)) {
                internal_error(state, ins, "invalid register assignment");
        }
        if (is_const(RHS(ins, 1))) {
@@ -15611,7 +23936,7 @@ static void print_op_shift(struct compile_state *state,
        else {
                fprintf(fp, "\t%s %s, %s\n",
                        op,
-                       reg(state, RHS(ins, 1), REGCM_GPR8),
+                       reg(state, RHS(ins, 1), REGCM_GPR8_LO),
                        reg(state, RHS(ins, 0), mask));
        }
 }
@@ -15623,7 +23948,7 @@ static void print_op_in(struct compile_state *state, struct triple *ins, FILE *f
        int dreg;
        mask = 0;
        switch(ins->op) {
-       case OP_INB: op = "inb", mask = REGCM_GPR8; break;
+       case OP_INB: op = "inb", mask = REGCM_GPR8_LO; break;
        case OP_INW: op = "inw", mask = REGCM_GPR16; break;
        case OP_INL: op = "inl", mask = REGCM_GPR32; break;
        default:
@@ -15648,7 +23973,7 @@ static void print_op_in(struct compile_state *state, struct triple *ins, FILE *f
                        internal_error(state, ins, "src != %%dx");
                }
                fprintf(fp, "\t%s %s, %s\n",
-                       op, 
+                       op,
                        reg(state, RHS(ins, 0), REGCM_GPR16),
                        reg(state, ins, mask));
        }
@@ -15661,7 +23986,7 @@ static void print_op_out(struct compile_state *state, struct triple *ins, FILE *
        int lreg;
        mask = 0;
        switch(ins->op) {
-       case OP_OUTB: op = "outb", mask = REGCM_GPR8; break;
+       case OP_OUTB: op = "outb", mask = REGCM_GPR8_LO; break;
        case OP_OUTW: op = "outw", mask = REGCM_GPR16; break;
        case OP_OUTL: op = "outl", mask = REGCM_GPR32; break;
        default:
@@ -15674,7 +23999,7 @@ static void print_op_out(struct compile_state *state, struct triple *ins, FILE *
                internal_error(state, ins, "src != %%eax");
        }
        if (is_const(RHS(ins, 1))) {
-               fprintf(fp, "\t%s %s,", 
+               fprintf(fp, "\t%s %s,",
                        op, reg(state, RHS(ins, 0), mask));
                print_const_val(state, RHS(ins, 1), fp);
                fprintf(fp, "\n");
@@ -15686,7 +24011,7 @@ static void print_op_out(struct compile_state *state, struct triple *ins, FILE *
                        internal_error(state, ins, "dst != %%dx");
                }
                fprintf(fp, "\t%s %s, %s\n",
-                       op, 
+                       op,
                        reg(state, RHS(ins, 0), mask),
                        reg(state, RHS(ins, 1), REGCM_GPR16));
        }
@@ -15699,31 +24024,45 @@ static void print_op_move(struct compile_state *state,
         * of registers we can move between.
         * Because OP_COPY will be introduced in arbitrary locations
         * OP_COPY must not affect flags.
+        * OP_CONVERT can change the flags and it is the only operation
+        * where it is expected the types in the registers can change.
         */
        int omit_copy = 1; /* Is it o.k. to omit a noop copy? */
        struct triple *dst, *src;
-       if (ins->op == OP_COPY) {
-               src = RHS(ins, 0);
-               dst = ins;
+       if (state->arch->features & X86_NOOP_COPY) {
+               omit_copy = 0;
        }
-       else if (ins->op == OP_WRITE) {
-               dst = LHS(ins, 0);
+       if ((ins->op == OP_COPY) || (ins->op == OP_CONVERT)) {
                src = RHS(ins, 0);
+               dst = ins;
        }
        else {
                internal_error(state, ins, "unknown move operation");
                src = dst = 0;
        }
+       if (reg_size(state, dst) < size_of(state, dst->type)) {
+               internal_error(state, ins, "Invalid destination register");
+       }
+       if (!equiv_types(src->type, dst->type) && (dst->op == OP_COPY)) {
+               fprintf(state->errout, "src type: ");
+               name_of(state->errout, src->type);
+               fprintf(state->errout, "\n");
+               fprintf(state->errout, "dst type: ");
+               name_of(state->errout, dst->type);
+               fprintf(state->errout, "\n");
+               internal_error(state, ins, "Type mismatch for OP_COPY");
+       }
+
        if (!is_const(src)) {
                int src_reg, dst_reg;
                int src_regcm, dst_regcm;
-               src_reg = ID_REG(src->id);
+               src_reg   = ID_REG(src->id);
                dst_reg   = ID_REG(dst->id);
                src_regcm = arch_reg_regcm(state, src_reg);
-               dst_regcm   = arch_reg_regcm(state, dst_reg);
+               dst_regcm = arch_reg_regcm(state, dst_reg);
                /* If the class is the same just move the register */
-               if (src_regcm & dst_regcm & 
-                       (REGCM_GPR8 | REGCM_GPR16 | REGCM_GPR32)) {
+               if (src_regcm & dst_regcm &
+                       (REGCM_GPR8_LO | REGCM_GPR16 | REGCM_GPR32)) {
                        if ((src_reg != dst_reg) || !omit_copy) {
                                fprintf(fp, "\tmov %s, %s\n",
                                        reg(state, src, src_regcm),
@@ -15736,13 +24075,23 @@ static void print_op_move(struct compile_state *state,
                        src_reg = (src_reg - REGC_GPR32_FIRST) + REGC_GPR16_FIRST;
                        if ((src_reg != dst_reg) || !omit_copy) {
                                fprintf(fp, "\tmovw %s, %s\n",
-                                       arch_reg_str(src_reg), 
+                                       arch_reg_str(src_reg),
+                                       arch_reg_str(dst_reg));
+                       }
+               }
+               /* Move from 32bit gprs to 16bit gprs */
+               else if ((src_regcm & REGCM_GPR32) &&
+                       (dst_regcm & REGCM_GPR16)) {
+                       dst_reg = (dst_reg - REGC_GPR16_FIRST) + REGC_GPR32_FIRST;
+                       if ((src_reg != dst_reg) || !omit_copy) {
+                               fprintf(fp, "\tmov %s, %s\n",
+                                       arch_reg_str(src_reg),
                                        arch_reg_str(dst_reg));
                        }
                }
                /* Move 32bit to 8bit */
                else if ((src_regcm & REGCM_GPR32_8) &&
-                       (dst_regcm & REGCM_GPR8))
+                       (dst_regcm & REGCM_GPR8_LO))
                {
                        src_reg = (src_reg - REGC_GPR32_8_FIRST) + REGC_GPR8_FIRST;
                        if ((src_reg != dst_reg) || !omit_copy) {
@@ -15753,7 +24102,7 @@ static void print_op_move(struct compile_state *state,
                }
                /* Move 16bit to 8bit */
                else if ((src_regcm & REGCM_GPR16_8) &&
-                       (dst_regcm & REGCM_GPR8))
+                       (dst_regcm & REGCM_GPR8_LO))
                {
                        src_reg = (src_reg - REGC_GPR16_8_FIRST) + REGC_GPR8_FIRST;
                        if ((src_reg != dst_reg) || !omit_copy) {
@@ -15763,7 +24112,7 @@ static void print_op_move(struct compile_state *state,
                        }
                }
                /* Move 8/16bit to 16/32bit */
-               else if ((src_regcm & (REGCM_GPR8 | REGCM_GPR16)) && 
+               else if ((src_regcm & (REGCM_GPR8_LO | REGCM_GPR16)) &&
                        (dst_regcm & (REGCM_GPR16 | REGCM_GPR32))) {
                        const char *op;
                        op = is_signed(src->type)? "movsx": "movzx";
@@ -15780,15 +24129,26 @@ static void print_op_move(struct compile_state *state,
                                        reg(state, dst, dst_regcm));
                        }
                }
-               /* Move between mmx registers or mmx & sse  registers */
-               else if ((src_regcm & (REGCM_MMX | REGCM_XMM)) &&
-                       (dst_regcm & (REGCM_MMX | REGCM_XMM))) {
+               /* Move between mmx registers */
+               else if ((src_regcm & dst_regcm & REGCM_MMX)) {
                        if ((src_reg != dst_reg) || !omit_copy) {
                                fprintf(fp, "\tmovq %s, %s\n",
                                        reg(state, src, src_regcm),
                                        reg(state, dst, dst_regcm));
                        }
                }
+               /* Move from sse to mmx registers */
+               else if ((src_regcm & REGCM_XMM) && (dst_regcm & REGCM_MMX)) {
+                       fprintf(fp, "\tmovdq2q %s, %s\n",
+                               reg(state, src, src_regcm),
+                               reg(state, dst, dst_regcm));
+               }
+               /* Move from mmx to sse registers */
+               else if ((src_regcm & REGCM_MMX) && (dst_regcm & REGCM_XMM)) {
+                       fprintf(fp, "\tmovq2dq %s, %s\n",
+                               reg(state, src, src_regcm),
+                               reg(state, dst, dst_regcm));
+               }
                /* Move between 32bit gprs & mmx/sse registers */
                else if ((src_regcm & (REGCM_GPR32 | REGCM_MMX | REGCM_XMM)) &&
                        (dst_regcm & (REGCM_GPR32 | REGCM_MMX | REGCM_XMM))) {
@@ -15796,9 +24156,71 @@ static void print_op_move(struct compile_state *state,
                                reg(state, src, src_regcm),
                                reg(state, dst, dst_regcm));
                }
+               /* Move from 16bit gprs &  mmx/sse registers */
+               else if ((src_regcm & REGCM_GPR16) &&
+                       (dst_regcm & (REGCM_MMX | REGCM_XMM))) {
+                       const char *op;
+                       int mid_reg;
+                       op = is_signed(src->type)? "movsx":"movzx";
+                       mid_reg = (src_reg - REGC_GPR16_FIRST) + REGC_GPR32_FIRST;
+                       fprintf(fp, "\t%s %s, %s\n\tmovd %s, %s\n",
+                               op,
+                               arch_reg_str(src_reg),
+                               arch_reg_str(mid_reg),
+                               arch_reg_str(mid_reg),
+                               arch_reg_str(dst_reg));
+               }
+               /* Move from mmx/sse registers to 16bit gprs */
+               else if ((src_regcm & (REGCM_MMX | REGCM_XMM)) &&
+                       (dst_regcm & REGCM_GPR16)) {
+                       dst_reg = (dst_reg - REGC_GPR16_FIRST) + REGC_GPR32_FIRST;
+                       fprintf(fp, "\tmovd %s, %s\n",
+                               arch_reg_str(src_reg),
+                               arch_reg_str(dst_reg));
+               }
+               /* Move from gpr to 64bit dividend */
+               else if ((src_regcm & (REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO))  &&
+                       (dst_regcm & REGCM_DIVIDEND64)) {
+                       const char *extend;
+                       extend = is_signed(src->type)? "cltd":"movl $0, %edx";
+                       fprintf(fp, "\tmov %s, %%eax\n\t%s\n",
+                               arch_reg_str(src_reg),
+                               extend);
+               }
+               /* Move from 64bit gpr to gpr */
+               else if ((src_regcm & REGCM_DIVIDEND64) &&
+                       (dst_regcm & (REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO))) {
+                       if (dst_regcm & REGCM_GPR32) {
+                               src_reg = REG_EAX;
+                       }
+                       else if (dst_regcm & REGCM_GPR16) {
+                               src_reg = REG_AX;
+                       }
+                       else if (dst_regcm & REGCM_GPR8_LO) {
+                               src_reg = REG_AL;
+                       }
+                       fprintf(fp, "\tmov %s, %s\n",
+                               arch_reg_str(src_reg),
+                               arch_reg_str(dst_reg));
+               }
+               /* Move from mmx/sse registers to 64bit gpr */
+               else if ((src_regcm & (REGCM_MMX | REGCM_XMM)) &&
+                       (dst_regcm & REGCM_DIVIDEND64)) {
+                       const char *extend;
+                       extend = is_signed(src->type)? "cltd": "movl $0, %edx";
+                       fprintf(fp, "\tmovd %s, %%eax\n\t%s\n",
+                               arch_reg_str(src_reg),
+                               extend);
+               }
+               /* Move from 64bit gpr to mmx/sse register */
+               else if ((src_regcm & REGCM_DIVIDEND64) &&
+                       (dst_regcm & (REGCM_XMM | REGCM_MMX))) {
+                       fprintf(fp, "\tmovd %%eax, %s\n",
+                               arch_reg_str(dst_reg));
+               }
 #if X86_4_8BIT_GPRS
                /* Move from 8bit gprs to  mmx/sse registers */
-               else if ((src_regcm & REGCM_GPR8) && (src_reg <= REG_DL) &&
+               else if ((src_regcm & REGCM_GPR8_LO) && (src_reg <= REG_DL) &&
                        (dst_regcm & (REGCM_MMX | REGCM_XMM))) {
                        const char *op;
                        int mid_reg;
@@ -15813,26 +24235,16 @@ static void print_op_move(struct compile_state *state,
                }
                /* Move from mmx/sse registers and 8bit gprs */
                else if ((src_regcm & (REGCM_MMX | REGCM_XMM)) &&
-                       (dst_regcm & REGCM_GPR8) && (dst_reg <= REG_DL)) {
+                       (dst_regcm & REGCM_GPR8_LO) && (dst_reg <= REG_DL)) {
                        int mid_reg;
                        mid_reg = (dst_reg - REGC_GPR8_FIRST) + REGC_GPR32_FIRST;
                        fprintf(fp, "\tmovd %s, %s\n",
                                reg(state, src, src_regcm),
                                arch_reg_str(mid_reg));
                }
-               /* Move from 32bit gprs to 16bit gprs */
-               else if ((src_regcm & REGCM_GPR32) &&
-                       (dst_regcm & REGCM_GPR16)) {
-                       dst_reg = (dst_reg - REGC_GPR16_FIRST) + REGC_GPR32_FIRST;
-                       if ((src_reg != dst_reg) || !omit_copy) {
-                               fprintf(fp, "\tmov %s, %s\n",
-                                       arch_reg_str(src_reg),
-                                       arch_reg_str(dst_reg));
-                       }
-               }
                /* Move from 32bit gprs to 8bit gprs */
                else if ((src_regcm & REGCM_GPR32) &&
-                       (dst_regcm & REGCM_GPR8)) {
+                       (dst_regcm & REGCM_GPR8_LO)) {
                        dst_reg = (dst_reg - REGC_GPR8_FIRST) + REGC_GPR32_FIRST;
                        if ((src_reg != dst_reg) || !omit_copy) {
                                fprintf(fp, "\tmov %s, %s\n",
@@ -15842,7 +24254,7 @@ static void print_op_move(struct compile_state *state,
                }
                /* Move from 16bit gprs to 8bit gprs */
                else if ((src_regcm & REGCM_GPR16) &&
-                       (dst_regcm & REGCM_GPR8)) {
+                       (dst_regcm & REGCM_GPR8_LO)) {
                        dst_reg = (dst_reg - REGC_GPR8_FIRST) + REGC_GPR16_FIRST;
                        if ((src_reg != dst_reg) || !omit_copy) {
                                fprintf(fp, "\tmov %s, %s\n",
@@ -15851,15 +24263,161 @@ static void print_op_move(struct compile_state *state,
                        }
                }
 #endif /* X86_4_8BIT_GPRS */
+               /* Move from %eax:%edx to %eax:%edx */
+               else if ((src_regcm & REGCM_DIVIDEND64) &&
+                       (dst_regcm & REGCM_DIVIDEND64) &&
+                       (src_reg == dst_reg)) {
+                       if (!omit_copy) {
+                               fprintf(fp, "\t/*mov %s, %s*/\n",
+                                       arch_reg_str(src_reg),
+                                       arch_reg_str(dst_reg));
+                       }
+               }
                else {
+                       if ((src_regcm & ~REGCM_FLAGS) == 0) {
+                               internal_error(state, ins, "attempt to copy from %%eflags!");
+                       }
                        internal_error(state, ins, "unknown copy type");
                }
        }
        else {
-               fprintf(fp, "\tmov ");
-               print_const_val(state, src, fp);
-               fprintf(fp, ", %s\n",
-                       reg(state, dst, REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8));
+               size_t dst_size;
+               int dst_reg;
+               int dst_regcm;
+               dst_size = size_of(state, dst->type);
+               dst_reg = ID_REG(dst->id);
+               dst_regcm = arch_reg_regcm(state, dst_reg);
+               if (dst_regcm & (REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO)) {
+                       fprintf(fp, "\tmov ");
+                       print_const_val(state, src, fp);
+                       fprintf(fp, ", %s\n",
+                               reg(state, dst, REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO));
+               }
+               else if (dst_regcm & REGCM_DIVIDEND64) {
+                       if (dst_size > SIZEOF_I32) {
+                               internal_error(state, ins, "%dbit constant...", dst_size);
+                       }
+                       fprintf(fp, "\tmov $0, %%edx\n");
+                       fprintf(fp, "\tmov ");
+                       print_const_val(state, src, fp);
+                       fprintf(fp, ", %%eax\n");
+               }
+               else if (dst_regcm & REGCM_DIVIDEND32) {
+                       if (dst_size > SIZEOF_I16) {
+                               internal_error(state, ins, "%dbit constant...", dst_size);
+                       }
+                       fprintf(fp, "\tmov $0, %%dx\n");
+                       fprintf(fp, "\tmov ");
+                       print_const_val(state, src, fp);
+                       fprintf(fp, ", %%ax");
+               }
+               else if (dst_regcm & (REGCM_XMM | REGCM_MMX)) {
+                       long ref;
+                       if (dst_size > SIZEOF_I32) {
+                               internal_error(state, ins, "%d bit constant...", dst_size);
+                       }
+                       ref = get_const_pool_ref(state, src, SIZEOF_I32, fp);
+                       fprintf(fp, "\tmovd L%s%lu, %s\n",
+                               state->compiler->label_prefix, ref,
+                               reg(state, dst, (REGCM_XMM | REGCM_MMX)));
+               }
+               else {
+                       internal_error(state, ins, "unknown copy immediate type");
+               }
+       }
+       /* Leave now if this is not a type conversion */
+       if (ins->op != OP_CONVERT) {
+               return;
+       }
+       /* Now make certain I have not logically overflowed the destination */
+       if ((size_of(state, src->type) > size_of(state, dst->type)) &&
+               (size_of(state, dst->type) < reg_size(state, dst)))
+       {
+               unsigned long mask;
+               int dst_reg;
+               int dst_regcm;
+               if (size_of(state, dst->type) >= 32) {
+                       fprintf(state->errout, "dst type: ");
+                       name_of(state->errout, dst->type);
+                       fprintf(state->errout, "\n");
+                       internal_error(state, dst, "unhandled dst type size");
+               }
+               mask = 1;
+               mask <<= size_of(state, dst->type);
+               mask -= 1;
+
+               dst_reg = ID_REG(dst->id);
+               dst_regcm = arch_reg_regcm(state, dst_reg);
+
+               if (dst_regcm & (REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO)) {
+                       fprintf(fp, "\tand $0x%lx, %s\n",
+                               mask, reg(state, dst, REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO));
+               }
+               else if (dst_regcm & REGCM_MMX) {
+                       long ref;
+                       ref = get_mask_pool_ref(state, dst, mask, fp);
+                       fprintf(fp, "\tpand L%s%lu, %s\n",
+                               state->compiler->label_prefix, ref,
+                               reg(state, dst, REGCM_MMX));
+               }
+               else if (dst_regcm & REGCM_XMM) {
+                       long ref;
+                       ref = get_mask_pool_ref(state, dst, mask, fp);
+                       fprintf(fp, "\tpand L%s%lu, %s\n",
+                               state->compiler->label_prefix, ref,
+                               reg(state, dst, REGCM_XMM));
+               }
+               else {
+                       fprintf(state->errout, "dst type: ");
+                       name_of(state->errout, dst->type);
+                       fprintf(state->errout, "\n");
+                       fprintf(state->errout, "dst: %s\n", reg(state, dst, REGCM_ALL));
+                       internal_error(state, dst, "failed to trunc value: mask %lx", mask);
+               }
+       }
+       /* Make certain I am properly sign extended */
+       if ((size_of(state, src->type) < size_of(state, dst->type)) &&
+               (is_signed(src->type)))
+       {
+               int reg_bits, shift_bits;
+               int dst_reg;
+               int dst_regcm;
+
+               reg_bits = reg_size(state, dst);
+               if (reg_bits > 32) {
+                       reg_bits = 32;
+               }
+               shift_bits = reg_bits - size_of(state, src->type);
+               dst_reg = ID_REG(dst->id);
+               dst_regcm = arch_reg_regcm(state, dst_reg);
+
+               if (shift_bits < 0) {
+                       internal_error(state, dst, "negative shift?");
+               }
+
+               if (dst_regcm & (REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO)) {
+                       fprintf(fp, "\tshl $%d, %s\n",
+                               shift_bits,
+                               reg(state, dst, REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO));
+                       fprintf(fp, "\tsar $%d, %s\n",
+                               shift_bits,
+                               reg(state, dst, REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO));
+               }
+               else if (dst_regcm & (REGCM_MMX | REGCM_XMM)) {
+                       fprintf(fp, "\tpslld $%d, %s\n",
+                               shift_bits,
+                               reg(state, dst, REGCM_MMX | REGCM_XMM));
+                       fprintf(fp, "\tpsrad $%d, %s\n",
+                               shift_bits,
+                               reg(state, dst, REGCM_MMX | REGCM_XMM));
+               }
+               else {
+                       fprintf(state->errout, "dst type: ");
+                       name_of(state->errout, dst->type);
+                       fprintf(state->errout, "\n");
+                       fprintf(state->errout, "dst: %s\n", reg(state, dst, REGCM_ALL));
+                       internal_error(state, dst, "failed to signed extend value");
+               }
        }
 }
 
@@ -15867,14 +24425,31 @@ static void print_op_load(struct compile_state *state,
        struct triple *ins, FILE *fp)
 {
        struct triple *dst, *src;
+       const char *op;
        dst = ins;
        src = RHS(ins, 0);
        if (is_const(src) || is_const(dst)) {
                internal_error(state, ins, "unknown load operation");
        }
-       fprintf(fp, "\tmov (%s), %s\n",
+       switch(ins->type->type & TYPE_MASK) {
+       case TYPE_CHAR:   op = "movsbl"; break;
+       case TYPE_UCHAR:  op = "movzbl"; break;
+       case TYPE_SHORT:  op = "movswl"; break;
+       case TYPE_USHORT: op = "movzwl"; break;
+       case TYPE_INT:    case TYPE_UINT:
+       case TYPE_LONG:   case TYPE_ULONG:
+       case TYPE_POINTER:
+               op = "movl";
+               break;
+       default:
+               internal_error(state, ins, "unknown type in load");
+               op = "<invalid opcode>";
+               break;
+       }
+       fprintf(fp, "\t%s (%s), %s\n",
+               op,
                reg(state, src, REGCM_GPR32),
-               reg(state, dst, REGCM_GPR8 | REGCM_GPR16 | REGCM_GPR32));
+               reg(state, dst, REGCM_GPR32));
 }
 
 
@@ -15882,21 +24457,21 @@ static void print_op_store(struct compile_state *state,
        struct triple *ins, FILE *fp)
 {
        struct triple *dst, *src;
-       dst = LHS(ins, 0);
-       src = RHS(ins, 0);
+       dst = RHS(ins, 0);
+       src = RHS(ins, 1);
        if (is_const(src) && (src->op == OP_INTCONST)) {
                long_t value;
                value = (long_t)(src->u.cval);
                fprintf(fp, "\tmov%s $%ld, (%s)\n",
                        type_suffix(state, src->type),
-                       value,
+                       (long)(value),
                        reg(state, dst, REGCM_GPR32));
        }
        else if (is_const(dst) && (dst->op == OP_INTCONST)) {
                fprintf(fp, "\tmov%s %s, 0x%08lx\n",
                        type_suffix(state, src->type),
-                       reg(state, src, REGCM_GPR8 | REGCM_GPR16 | REGCM_GPR32),
-                       dst->u.cval);
+                       reg(state, src, REGCM_GPR8_LO | REGCM_GPR16 | REGCM_GPR32),
+                       (unsigned long)(dst->u.cval));
        }
        else {
                if (is_const(src) || is_const(dst)) {
@@ -15904,11 +24479,11 @@ static void print_op_store(struct compile_state *state,
                }
                fprintf(fp, "\tmov%s %s, (%s)\n",
                        type_suffix(state, src->type),
-                       reg(state, src, REGCM_GPR8 | REGCM_GPR16 | REGCM_GPR32),
+                       reg(state, src, REGCM_GPR8_LO | REGCM_GPR16 | REGCM_GPR32),
                        reg(state, dst, REGCM_GPR32));
        }
-       
-       
+
+
 }
 
 static void print_op_smul(struct compile_state *state,
@@ -15931,7 +24506,7 @@ static void print_op_cmp(struct compile_state *state,
 {
        unsigned mask;
        int dreg;
-       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8;
+       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO;
        dreg = check_reg(state, ins, REGCM_FLAGS);
        if (!reg_is_reg(state, dreg, REG_EFLAGS)) {
                internal_error(state, ins, "bad dest register for cmp");
@@ -15959,7 +24534,7 @@ static void print_op_test(struct compile_state *state,
        struct triple *ins, FILE *fp)
 {
        unsigned mask;
-       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8;
+       mask = REGCM_GPR32 | REGCM_GPR16 | REGCM_GPR8_LO;
        fprintf(fp, "\ttest %s, %s\n",
                reg(state, RHS(ins, 0), mask),
                reg(state, RHS(ins, 0), mask));
@@ -15969,15 +24544,15 @@ static void print_op_branch(struct compile_state *state,
        struct triple *branch, FILE *fp)
 {
        const char *bop = "j";
-       if (branch->op == OP_JMP) {
-               if (TRIPLE_RHS(branch->sizes) != 0) {
+       if ((branch->op == OP_JMP) || (branch->op == OP_CALL)) {
+               if (branch->rhs != 0) {
                        internal_error(state, branch, "jmp with condition?");
                }
                bop = "jmp";
        }
        else {
                struct triple *ptr;
-               if (TRIPLE_RHS(branch->sizes) != 1) {
+               if (branch->rhs != 1) {
                        internal_error(state, branch, "jmpcc without condition?");
                }
                check_reg(state, RHS(branch, 0), REGCM_FLAGS);
@@ -15985,7 +24560,9 @@ static void print_op_branch(struct compile_state *state,
                        (RHS(branch, 0)->op != OP_TEST)) {
                        internal_error(state, branch, "bad branch test");
                }
+#if DEBUG_ROMCC_WARNINGS
 #warning "FIXME I have observed instructions between the test and branch instructions"
+#endif
                ptr = RHS(branch, 0);
                for(ptr = RHS(branch, 0)->next; ptr != branch; ptr = ptr->next) {
                        if (ptr->op != OP_COPY) {
@@ -16007,19 +24584,31 @@ static void print_op_branch(struct compile_state *state,
                        internal_error(state, branch, "Invalid branch op");
                        break;
                }
-               
+
+       }
+#if 1
+       if (branch->op == OP_CALL) {
+               fprintf(fp, "\t/* call */\n");
        }
+#endif
        fprintf(fp, "\t%s L%s%lu\n",
-               bop, 
-               state->label_prefix,
-               TARG(branch, 0)->u.cval);
+               bop,
+               state->compiler->label_prefix,
+               (unsigned long)(TARG(branch, 0)->u.cval));
+}
+
+static void print_op_ret(struct compile_state *state,
+       struct triple *branch, FILE *fp)
+{
+       fprintf(fp, "\tjmp *%s\n",
+               reg(state, RHS(branch, 0), REGCM_GPR32));
 }
 
 static void print_op_set(struct compile_state *state,
        struct triple *set, FILE *fp)
 {
        const char *sop = "set";
-       if (TRIPLE_RHS(set->sizes) != 1) {
+       if (set->rhs != 1) {
                internal_error(state, set, "setcc without condition?");
        }
        check_reg(state, RHS(set, 0), REGCM_FLAGS);
@@ -16046,22 +24635,22 @@ static void print_op_set(struct compile_state *state,
                break;
        }
        fprintf(fp, "\t%s %s\n",
-               sop, reg(state, set, REGCM_GPR8));
+               sop, reg(state, set, REGCM_GPR8_LO));
 }
 
-static void print_op_bit_scan(struct compile_state *state, 
-       struct triple *ins, FILE *fp) 
+static void print_op_bit_scan(struct compile_state *state,
+       struct triple *ins, FILE *fp)
 {
        const char *op;
        switch(ins->op) {
        case OP_BSF: op = "bsf"; break;
        case OP_BSR: op = "bsr"; break;
-       default: 
+       default:
                internal_error(state, ins, "unknown bit scan");
                op = 0;
                break;
        }
-       fprintf(fp, 
+       fprintf(fp,
                "\t%s %s, %s\n"
                "\tjnz 1f\n"
                "\tmovl $-1, %s\n"
@@ -16072,67 +24661,24 @@ static void print_op_bit_scan(struct compile_state *state,
                reg(state, ins, REGCM_GPR32));
 }
 
-static void print_const(struct compile_state *state,
-       struct triple *ins, FILE *fp)
-{
-       switch(ins->op) {
-       case OP_INTCONST:
-               switch(ins->type->type & TYPE_MASK) {
-               case TYPE_CHAR:
-               case TYPE_UCHAR:
-                       fprintf(fp, ".byte 0x%02lx\n", ins->u.cval);
-                       break;
-               case TYPE_SHORT:
-               case TYPE_USHORT:
-                       fprintf(fp, ".short 0x%04lx\n", ins->u.cval);
-                       break;
-               case TYPE_INT:
-               case TYPE_UINT:
-               case TYPE_LONG:
-               case TYPE_ULONG:
-                       fprintf(fp, ".int %lu\n", ins->u.cval);
-                       break;
-               default:
-                       internal_error(state, ins, "Unknown constant type");
-               }
-               break;
-       case OP_BLOBCONST:
-       {
-               unsigned char *blob;
-               size_t size, i;
-               size = size_of(state, ins->type);
-               blob = ins->u.blob;
-               for(i = 0; i < size; i++) {
-                       fprintf(fp, ".byte 0x%02x\n",
-                               blob[i]);
-               }
-               break;
-       }
-       default:
-               internal_error(state, ins, "Unknown constant type");
-               break;
-       }
-}
-
-#define TEXT_SECTION ".rom.text"
-#define DATA_SECTION ".rom.data"
 
 static void print_sdecl(struct compile_state *state,
        struct triple *ins, FILE *fp)
 {
        fprintf(fp, ".section \"" DATA_SECTION "\"\n");
-       fprintf(fp, ".balign %d\n", align_of(state, ins->type));
-       fprintf(fp, "L%s%lu:\n", state->label_prefix, ins->u.cval);
+       fprintf(fp, ".balign %ld\n", (long int)align_of_in_bytes(state, ins->type));
+       fprintf(fp, "L%s%lu:\n",
+               state->compiler->label_prefix, (unsigned long)(ins->u.cval));
        print_const(state, MISC(ins, 0), fp);
        fprintf(fp, ".section \"" TEXT_SECTION "\"\n");
-               
+
 }
 
 static void print_instruction(struct compile_state *state,
        struct triple *ins, FILE *fp)
 {
        /* Assumption: after I have exted the register allocator
-        * everything is in a valid register. 
+        * everything is in a valid register.
         */
        switch(ins->op) {
        case OP_ASM:
@@ -16149,6 +24695,7 @@ static void print_instruction(struct compile_state *state,
        case OP_POS:    break;
        case OP_NEG:    print_unary_op(state, "neg", ins, fp); break;
        case OP_INVERT: print_unary_op(state, "not", ins, fp); break;
+       case OP_NOOP:
        case OP_INTCONST:
        case OP_ADDRCONST:
        case OP_BLOBCONST:
@@ -16156,11 +24703,15 @@ static void print_instruction(struct compile_state *state,
        case OP_PHI:
                /* Don't generate anything for variable declarations. */
                break;
+       case OP_UNKNOWNVAL:
+               fprintf(fp, " /* unknown %s */\n",
+                       reg(state, ins, REGCM_ALL));
+               break;
        case OP_SDECL:
                print_sdecl(state, ins, fp);
                break;
-       case OP_WRITE: 
-       case OP_COPY:   
+       case OP_COPY:
+       case OP_CONVERT:
                print_op_move(state, ins, fp);
                break;
        case OP_LOAD:
@@ -16180,8 +24731,12 @@ static void print_instruction(struct compile_state *state,
        case OP_JMP_SMORE:   case OP_JMP_UMORE:
        case OP_JMP_SLESSEQ: case OP_JMP_ULESSEQ:
        case OP_JMP_SMOREEQ: case OP_JMP_UMOREEQ:
+       case OP_CALL:
                print_op_branch(state, ins, fp);
                break;
+       case OP_RET:
+               print_op_ret(state, ins, fp);
+               break;
        case OP_SET_EQ:      case OP_SET_NOTEQ:
        case OP_SET_SLESS:   case OP_SET_ULESS:
        case OP_SET_SMORE:   case OP_SET_UMORE:
@@ -16190,10 +24745,10 @@ static void print_instruction(struct compile_state *state,
                print_op_set(state, ins, fp);
                break;
        case OP_INB:  case OP_INW:  case OP_INL:
-               print_op_in(state, ins, fp); 
+               print_op_in(state, ins, fp);
                break;
        case OP_OUTB: case OP_OUTW: case OP_OUTL:
-               print_op_out(state, ins, fp); 
+               print_op_out(state, ins, fp);
                break;
        case OP_BSF:
        case OP_BSR:
@@ -16209,20 +24764,34 @@ static void print_instruction(struct compile_state *state,
        case OP_HLT:
                fprintf(fp, "\thlt\n");
                break;
+       case OP_SDIVT:
+               fprintf(fp, "\tidiv %s\n", reg(state, RHS(ins, 1), REGCM_GPR32));
+               break;
+       case OP_UDIVT:
+               fprintf(fp, "\tdiv %s\n", reg(state, RHS(ins, 1), REGCM_GPR32));
+               break;
+       case OP_UMUL:
+               fprintf(fp, "\tmul %s\n", reg(state, RHS(ins, 1), REGCM_GPR32));
+               break;
        case OP_LABEL:
                if (!ins->use) {
                        return;
                }
-               fprintf(fp, "L%s%lu:\n", state->label_prefix, ins->u.cval);
+               fprintf(fp, "L%s%lu:\n",
+                       state->compiler->label_prefix, (unsigned long)(ins->u.cval));
+               break;
+       case OP_ADECL:
+               /* Ignore adecls with no registers error otherwise */
+               if (!noop_adecl(ins)) {
+                       internal_error(state, ins, "adecl remains?");
+               }
                break;
                /* Ignore OP_PIECE */
        case OP_PIECE:
                break;
-               /* Operations I am not yet certain how to handle */
-       case OP_UMUL:
+               /* Operations that should never get here */
        case OP_SDIV: case OP_UDIV:
        case OP_SMOD: case OP_UMOD:
-               /* Operations that should never get here */
        case OP_LTRUE:   case OP_LFALSE:  case OP_EQ:      case OP_NOTEQ:
        case OP_SLESS:   case OP_ULESS:   case OP_SMORE:   case OP_UMORE:
        case OP_SLESSEQ: case OP_ULESSEQ: case OP_SMOREEQ: case OP_UMOREEQ:
@@ -16237,128 +24806,270 @@ static void print_instructions(struct compile_state *state)
 {
        struct triple *first, *ins;
        int print_location;
-       int last_line;
-       int last_col;
-       const char *last_filename;
+       struct occurance *last_occurance;
        FILE *fp;
+       int max_inline_depth;
+       max_inline_depth = 0;
        print_location = 1;
-       last_line = -1;
-       last_col  = -1;
-       last_filename = 0;
+       last_occurance = 0;
        fp = state->output;
+       /* Masks for common sizes */
+       fprintf(fp, ".section \"" DATA_SECTION "\"\n");
+       fprintf(fp, ".balign 16\n");
+       fprintf(fp, "L%s1:\n", state->compiler->label_prefix);
+       fprintf(fp, ".int 0xff, 0, 0, 0\n");
+       fprintf(fp, "L%s2:\n", state->compiler->label_prefix);
+       fprintf(fp, ".int 0xffff, 0, 0, 0\n");
        fprintf(fp, ".section \"" TEXT_SECTION "\"\n");
-       first = RHS(state->main_function, 0);
+       first = state->first;
        ins = first;
        do {
                if (print_location &&
-                       ((last_filename != ins->filename) ||
-                               (last_line != ins->line) ||
-                               (last_col  != ins->col))) {
-                       fprintf(fp, "\t/* %s:%d */\n",
-                               ins->filename, ins->line);
-                       last_filename = ins->filename;
-                       last_line = ins->line;
-                       last_col  = ins->col;
+                       last_occurance != ins->occurance) {
+                       if (!ins->occurance->parent) {
+                               fprintf(fp, "\t/* %s,%s:%d.%d */\n",
+                                       ins->occurance->function?ins->occurance->function:"(null)",
+                                       ins->occurance->filename?ins->occurance->filename:"(null)",
+                                       ins->occurance->line,
+                                       ins->occurance->col);
+                       }
+                       else {
+                               struct occurance *ptr;
+                               int inline_depth;
+                               fprintf(fp, "\t/*\n");
+                               inline_depth = 0;
+                               for(ptr = ins->occurance; ptr; ptr = ptr->parent) {
+                                       inline_depth++;
+                                       fprintf(fp, "\t * %s,%s:%d.%d\n",
+                                               ptr->function,
+                                               ptr->filename,
+                                               ptr->line,
+                                               ptr->col);
+                               }
+                               fprintf(fp, "\t */\n");
+                               if (inline_depth > max_inline_depth) {
+                                       max_inline_depth = inline_depth;
+                               }
+                       }
+                       if (last_occurance) {
+                               put_occurance(last_occurance);
+                       }
+                       get_occurance(ins->occurance);
+                       last_occurance = ins->occurance;
                }
 
                print_instruction(state, ins, fp);
                ins = ins->next;
        } while(ins != first);
-       
+       if (print_location) {
+               fprintf(fp, "/* max inline depth %d */\n",
+                       max_inline_depth);
+       }
 }
+
 static void generate_code(struct compile_state *state)
 {
        generate_local_labels(state);
        print_instructions(state);
-       
+
 }
 
-static void print_tokens(struct compile_state *state)
+static void print_preprocessed_tokens(struct compile_state *state)
 {
-       struct token *tk;
-       tk = &state->token[0];
-       do {
-#if 1
-               token(state, 0);
-#else
-               next_token(state, 0);
-#endif
-               loc(stdout, state, 0);
-               printf("%s <- `%s'\n",
-                       tokens[tk->tok],
+       int tok;
+       FILE *fp;
+       int line;
+       const char *filename;
+       fp = state->output;
+       filename = 0;
+       line = 0;
+       for(;;) {
+               struct file_state *file;
+               struct token *tk;
+               const char *token_str;
+               tok = peek(state);
+               if (tok == TOK_EOF) {
+                       break;
+               }
+               tk = eat(state, tok);
+               token_str =
                        tk->ident ? tk->ident->name :
-                       tk->str_len ? tk->val.str : "");
-               
-       } while(tk->tok != TOK_EOF);
+                       tk->str_len ? tk->val.str :
+                       tokens[tk->tok];
+
+               file = state->file;
+               while(file->macro && file->prev) {
+                       file = file->prev;
+               }
+               if (!file->macro &&
+                       ((file->line != line) || (file->basename != filename)))
+               {
+                       int i, col;
+                       if ((file->basename == filename) &&
+                               (line < file->line)) {
+                               while(line < file->line) {
+                                       fprintf(fp, "\n");
+                                       line++;
+                               }
+                       }
+                       else {
+                               fprintf(fp, "\n#line %d \"%s\"\n",
+                                       file->line, file->basename);
+                       }
+                       line = file->line;
+                       filename = file->basename;
+                       col = get_col(file) - strlen(token_str);
+                       for(i = 0; i < col; i++) {
+                               fprintf(fp, " ");
+                       }
+               }
+
+               fprintf(fp, "%s ", token_str);
+
+               if (state->compiler->debug & DEBUG_TOKENS) {
+                       loc(state->dbgout, state, 0);
+                       fprintf(state->dbgout, "%s <- `%s'\n",
+                               tokens[tok], token_str);
+               }
+       }
 }
 
-static void compile(const char *filename, const char *ofilename, 
-       int cpu, int debug, int opt, const char *label_prefix)
+static void compile(const char *filename,
+       struct compiler_state *compiler, struct arch_state *arch)
 {
        int i;
        struct compile_state state;
+       struct triple *ptr;
+       struct filelist *includes = include_filelist;
        memset(&state, 0, sizeof(state));
+       state.compiler = compiler;
+       state.arch     = arch;
        state.file = 0;
        for(i = 0; i < sizeof(state.token)/sizeof(state.token[0]); i++) {
                memset(&state.token[i], 0, sizeof(state.token[i]));
                state.token[i].tok = -1;
        }
-       /* Remember the debug settings */
-       state.cpu      = cpu;
-       state.debug    = debug;
-       state.optimize = opt;
+       /* Remember the output descriptors */
+       state.errout = stderr;
+       state.dbgout = stdout;
        /* Remember the output filename */
-       state.ofilename = ofilename;
-       state.output    = fopen(state.ofilename, "w");
-       if (!state.output) {
-               error(&state, 0, "Cannot open output file %s\n",
-                       ofilename);
-       }
-       /* Remember the label prefix */
-       state.label_prefix = label_prefix;
+       if ((state.compiler->flags & COMPILER_PP_ONLY) && (strcmp("auto.inc",state.compiler->ofilename) == 0)) {
+               state.output    = stdout;
+       } else {
+               state.output    = fopen(state.compiler->ofilename, "w");
+               if (!state.output) {
+                       error(&state, 0, "Cannot open output file %s\n",
+                               state.compiler->ofilename);
+               }
+       }
+       /* Make certain a good cleanup happens */
+       exit_state = &state;
+       atexit(exit_cleanup);
+
        /* Prep the preprocessor */
        state.if_depth = 0;
-       state.if_value = 0;
+       memset(state.if_bytes, 0, sizeof(state.if_bytes));
        /* register the C keywords */
        register_keywords(&state);
        /* register the keywords the macro preprocessor knows */
        register_macro_keywords(&state);
+       /* generate some builtin macros */
+       register_builtin_macros(&state);
        /* Memorize where some special keywords are. */
-       state.i_continue = lookup(&state, "continue", 8);
-       state.i_break    = lookup(&state, "break", 5);
+       state.i_switch        = lookup(&state, "switch", 6);
+       state.i_case          = lookup(&state, "case", 4);
+       state.i_continue      = lookup(&state, "continue", 8);
+       state.i_break         = lookup(&state, "break", 5);
+       state.i_default       = lookup(&state, "default", 7);
+       state.i_return        = lookup(&state, "return", 6);
+       /* Memorize where predefined macros are. */
+       state.i___VA_ARGS__   = lookup(&state, "__VA_ARGS__", 11);
+       state.i___FILE__      = lookup(&state, "__FILE__", 8);
+       state.i___LINE__      = lookup(&state, "__LINE__", 8);
+       /* Memorize where predefined identifiers are. */
+       state.i___func__      = lookup(&state, "__func__", 8);
+       /* Memorize where some attribute keywords are. */
+       state.i_noinline      = lookup(&state, "noinline", 8);
+       state.i_always_inline = lookup(&state, "always_inline", 13);
+       state.i_noreturn      = lookup(&state, "noreturn", 8);
+
+       /* Process the command line macros */
+       process_cmdline_macros(&state);
+
+       /* Allocate beginning bounding labels for the function list */
+       state.first = label(&state);
+       state.first->id |= TRIPLE_FLAG_VOLATILE;
+       use_triple(state.first, state.first);
+       ptr = label(&state);
+       ptr->id |= TRIPLE_FLAG_VOLATILE;
+       use_triple(ptr, ptr);
+       flatten(&state, state.first, ptr);
+
+       /* Allocate a label for the pool of global variables */
+       state.global_pool = label(&state);
+       state.global_pool->id |= TRIPLE_FLAG_VOLATILE;
+       flatten(&state, state.first, state.global_pool);
+
        /* Enter the globl definition scope */
        start_scope(&state);
        register_builtins(&state);
+
        compile_file(&state, filename, 1);
-#if 0
-       print_tokens(&state);
-#endif 
+
+       while (includes) {
+               compile_file(&state, includes->filename, 1);
+               includes=includes->next;
+       }
+
+       /* Stop if all we want is preprocessor output */
+       if (state.compiler->flags & COMPILER_PP_ONLY) {
+               print_preprocessed_tokens(&state);
+               return;
+       }
+
        decls(&state);
+
        /* Exit the global definition scope */
        end_scope(&state);
 
-       /* Now that basic compilation has happened 
-        * optimize the intermediate code 
+       /* Now that basic compilation has happened
+        * optimize the intermediate code
         */
        optimize(&state);
 
        generate_code(&state);
-       if (state.debug) {
-               fprintf(stderr, "done\n");
+       if (state.compiler->debug) {
+               fprintf(state.errout, "done\n");
        }
+       exit_state = 0;
 }
 
-static void version(void)
+static void version(FILE *fp)
 {
-       printf("romcc " VERSION " released " RELEASE_DATE "\n");
+       fprintf(fp, "romcc " VERSION " released " RELEASE_DATE "\n");
 }
 
 static void usage(void)
 {
-       version();
-       printf(
-               "Usage: romcc <source>.c\n"
-               "Compile a C source file without using ram\n"
+       FILE *fp = stdout;
+       version(fp);
+       fprintf(fp,
+               "\nUsage: romcc [options] <source>.c\n"
+               "Compile a C source file generating a binary that does not implicilty use RAM\n"
+               "Options: \n"
+               "-o <output file name>\n"
+               "-f<option>            Specify a generic compiler option\n"
+               "-m<option>            Specify a arch dependent option\n"
+               "--                    Specify this is the last option\n"
+               "\nGeneric compiler options:\n"
+       );
+       compiler_usage(fp);
+       fprintf(fp,
+               "\nArchitecture compiler options:\n"
+       );
+       arch_usage(fp);
+       fprintf(fp,
+               "\n"
        );
 }
 
@@ -16375,61 +25086,93 @@ static void arg_error(char *fmt, ...)
 int main(int argc, char **argv)
 {
        const char *filename;
-       const char *ofilename;
-       const char *label_prefix;
-       int cpu;
-       int last_argc;
-       int debug;
-       int optimize;
-       cpu = CPU_DEFAULT;
-       label_prefix = "";
-       ofilename = "auto.inc";
-       optimize = 0;
-       debug = 0;
-       last_argc = -1;
-       while((argc > 1) && (argc != last_argc)) {
-               last_argc = argc;
-               if (strncmp(argv[1], "--debug=", 8) == 0) {
-                       debug = atoi(argv[1] + 8);
-                       argv++;
-                       argc--;
-               }
-               else if (strncmp(argv[1], "--label-prefix=", 15) == 0) {
-                       label_prefix= argv[1] + 15;
-                       argv++;
-                       argc--;
-               }
-               else if ((strcmp(argv[1],"-O") == 0) ||
-                       (strcmp(argv[1], "-O1") == 0)) {
-                       optimize = 1;
-                       argv++;
-                       argc--;
+       struct compiler_state compiler;
+       struct arch_state arch;
+       int all_opts;
+
+
+       /* I don't want any surprises */
+       setlocale(LC_ALL, "C");
+
+       init_compiler_state(&compiler);
+       init_arch_state(&arch);
+       filename = 0;
+       all_opts = 0;
+       while(argc > 1) {
+               if (!all_opts && (strcmp(argv[1], "-o") == 0) && (argc > 2)) {
+                       compiler.ofilename = argv[2];
+                       argv += 2;
+                       argc -= 2;
                }
-               else if (strcmp(argv[1],"-O2") == 0) {
-                       optimize = 2;
+               else if (!all_opts && argv[1][0] == '-') {
+                       int result;
+                       result = -1;
+                       if (strcmp(argv[1], "--") == 0) {
+                               result = 0;
+                               all_opts = 1;
+                       }
+                       else if (strncmp(argv[1], "-E", 2) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]);
+                       }
+                       else if (strncmp(argv[1], "-O", 2) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]);
+                       }
+                       else if (strncmp(argv[1], "-I", 2) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]);
+                       }
+                       else if (strncmp(argv[1], "-D", 2) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]);
+                       }
+                       else if (strncmp(argv[1], "-U", 2) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]);
+                       }
+                       else if (strncmp(argv[1], "--label-prefix=", 15) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]+2);
+                       }
+                       else if (strncmp(argv[1], "-f", 2) == 0) {
+                               result = compiler_encode_flag(&compiler, argv[1]+2);
+                       }
+                       else if (strncmp(argv[1], "-m", 2) == 0) {
+                               result = arch_encode_flag(&arch, argv[1]+2);
+                       }
+                       else if (strncmp(argv[1], "-c", 2) == 0) {
+                               result = 0;
+                       }
+                       else if (strncmp(argv[1], "-S", 2) == 0) {
+                               result = 0;
+                       }
+                       else if (strncmp(argv[1], "-include", 10) == 0) {
+                               struct filelist *old_head = include_filelist;
+                               include_filelist = malloc(sizeof(struct filelist));
+                               if (!include_filelist) {
+                                       die("Out of memory.\n");
+                               }
+                               argv++;
+                               argc--;
+                               include_filelist->filename = strdup(argv[1]);
+                               include_filelist->next = old_head;
+                               result = 0;
+                       }
+                       if (result < 0) {
+                               arg_error("Invalid option specified: %s\n",
+                                       argv[1]);
+                       }
                        argv++;
                        argc--;
                }
-               else if ((strcmp(argv[1], "-o") == 0) && (argc > 2)) {
-                       ofilename = argv[2];
-                       argv += 2;
-                       argc -= 2;
-               }
-               else if (strncmp(argv[1], "-mcpu=", 6) == 0) {
-                       cpu = arch_encode_cpu(argv[1] + 6);
-                       if (cpu == BAD_CPU) {
-                               arg_error("Invalid cpu specified: %s\n",
-                                       argv[1] + 6);
+               else {
+                       if (filename) {
+                               arg_error("Only one filename may be specified\n");
                        }
+                       filename = argv[1];
                        argv++;
                        argc--;
                }
        }
-       if (argc != 2) {
-               arg_error("Wrong argument count %d\n", argc);
+       if (!filename) {
+               arg_error("No filename specified\n");
        }
-       filename = argv[1];
-       compile(filename, ofilename, cpu, debug, optimize, label_prefix);
+       compile(filename, &compiler, &arch);
 
        return 0;
 }