* Removed all Id tags.
[cacao.git] / src / vm / jit / sparc64 / codegen.c
index 16ddd5c94bffb07f030c2055b578a32c4e7c3153..d4f9483f382fbd84632355e2f528055c9d9545da 100644 (file)
@@ -1,6 +1,6 @@
 /* src/vm/jit/sparc64/codegen.c - machine code generator for Sparc
 
-   Copyright (C) 1996-2005, 2006 R. Grafl, A. Krall, C. Kruegel,
+   Copyright (C) 1996-2005, 2006, 2007 R. Grafl, A. Krall, C. Kruegel,
    C. Oates, R. Obermaisser, M. Platter, M. Probst, S. Ring,
    E. Steiner, C. Thalinger, D. Thuernbeck, P. Tomsich, C. Ullrich,
    J. Wenninger, Institut f. Computersprachen - TU Wien
    Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
    02110-1301, USA.
 
-   Contact: cacao@cacaojvm.org
-
-   Authors: Andreas Krall
-            Reinhard Grafl
-            Alexander Jordan
-
-   Changes: Edwin Steiner
-
-   $Id: codegen.c 4644 2006-03-16 18:44:46Z edwin $
-
 */
 
 
 #include "config.h"
 
-#include <stdio.h>
 #include <assert.h>
-
+#include <stdint.h>
+#include <stdio.h>
 
 #include "vm/types.h"
 
 #include "mm/memory.h"
 
 #include "native/jni.h"
+#include "native/localref.h"
 #include "native/native.h"
 #include "vm/builtin.h"
 #include "vm/exceptions.h"
 #include "vm/global.h"
-#include "vm/loader.h"
-#include "vm/options.h"
-#include "vm/stringlocal.h"
+
+#include "vm/jit/abi.h"
 #include "vm/jit/asmpart.h"
 #include "vm/jit/codegen-common.h"
 #include "vm/jit/dseg.h"
 #include "vm/jit/emit-common.h"
+#include "vm/jit/sparc64/emit.h"
 #include "vm/jit/jit.h"
 #include "vm/jit/parse.h"
 #include "vm/jit/patcher.h"
 #include "vm/jit/reg.h"
+#include "vm/jit/replace.h"
+#include "vm/jit/stacktrace.h"
+#include "vmcore/loader.h"
+#include "vmcore/options.h"
+
+#include "vm/jit/sparc64/solaris/macro_rename.h"
+
+#define BUILTIN_FLOAT_ARGS 1
 
 /* XXX use something like this for window control ? 
  * #define REG_PV (own_window?REG_PV_CALLEE:REG_PV_CALLER)
  */
 #define REG_PV REG_PV_CALLEE
 
+bool fits_13(s4 disp)
+{
+       /*
+       if ((disp < -4096) || (disp > 4095))
+               printf("disp %d\n", disp);
+       */
+
+       return (disp >= -4096) && (disp <= 4095);
+}
+
+s4 get_lopart_disp(disp)
+{
+       s4 lodisp;
+       
+       if (disp > 0)
+               lodisp = setlo_part(disp);
+       else {
+               if (setlo_part(disp) == 0)
+                       lodisp = 0;
+               else
+                       lodisp = setlo_part(disp) | 0x1c00;
+       }
+               
+       return lodisp;
+}
+
+#ifndef NDEBUG
+bool check_13bit_imm(s8 imm)
+{
+       s4 sign = (imm >> 12) & 0x1;
+
+       if (sign == 0) {
+               if ((imm & ~0xfff) == 0) return true; /* pos imm. */
+       }
+       else
+               if ((imm & ~0xfff) + 0xfff == -1) return true; /* neg imm. */
+       
+       printf("immediate out-of-bounds: %ld\n", imm);
+       return false;
+}
+#endif
+       
 
-/* codegen *********************************************************************
+/* codegen_emit ****************************************************************
 
    Generates machine code.
 
 *******************************************************************************/
 
-bool codegen(jitdata *jd)
+bool codegen_emit(jitdata *jd)
 {
        methodinfo         *m;
        codeinfo           *code;
        codegendata        *cd;
        registerdata       *rd;
-       s4                  len, s1, s2, s3, d, disp;
+       s4                  len, s1, s2, s3, d, disp, slots;
        varinfo            *var;
        basicblock         *bptr;
        instruction        *iptr;
@@ -98,7 +139,6 @@ bool codegen(jitdata *jd)
        methoddesc         *md;
        fieldinfo          *fi;
        unresolved_field   *uf;
-       rplpoint           *replacementpoint;
        s4                  fieldtype;
        s4                  varindex;
 
@@ -118,12 +158,13 @@ bool codegen(jitdata *jd)
 
        {
        s4 i, p, t, l;
-       s4 savedregs_num, localbase;
+       s4 savedregs_num;
+       s4 framesize_disp;
 
 #if 0 /* no leaf optimization yet */
        savedregs_num = (jd->isleafmethod) ? 0 : 1;       /* space to save the RA */
 #endif
-       savedregs_num = 16;                          /* register-window save area */ 
+       savedregs_num = WINSAVE_CNT + ABIPARAMS_CNT; /* register-window save area */ 
 
 
        /* space to save used callee saved registers */
@@ -138,10 +179,15 @@ bool codegen(jitdata *jd)
                cd->stackframesize++;
 #endif
 
+       /* keep stack 16-byte aligned (ABI requirement) */
+
+       if (cd->stackframesize & 1)
+               cd->stackframesize++;
+
        /* create method header */
 
        (void) dseg_add_unique_address(cd, code);              /* CodeinfoPointer */
-       (void) dseg_add_unique_s4(cd, cd->stackframesize * 8); /* FrameSize       */
+       framesize_disp = dseg_add_unique_s4(cd, cd->stackframesize * 8); /* FrameSize       */
 
 #if defined(ENABLE_THREADS)
        /* IsSync contains the offset relative to the stack pointer for the
@@ -151,7 +197,7 @@ bool codegen(jitdata *jd)
        */
 
        if (checksync && (m->flags & ACC_SYNCHRONIZED))
-               (void) dseg_add_unique_s4(cd, (rd->memuse + 1) * 8); /* IsSync        */
+               (void) dseg_add_unique_s4(cd, JITSTACK + (rd->memuse + 1) * 8); /* IsSync */
        else
 #endif
                (void) dseg_add_unique_s4(cd, 0);                  /* IsSync          */
@@ -173,9 +219,15 @@ bool codegen(jitdata *jd)
 
        /* save register window and create stack frame (if necessary) */
 
-       if (cd->stackframesize)
-               M_SAVE(REG_SP, -cd->stackframesize * 8, REG_SP);
-
+       if (cd->stackframesize) {
+               if (cd->stackframesize <= 4095)
+                       M_SAVE(REG_SP, -cd->stackframesize * 8, REG_SP);
+               else {
+                       M_ILD_INTERN(REG_ITMP3, REG_PV_CALLER, framesize_disp);
+                       M_SUB(REG_ZERO, REG_ITMP3, REG_ITMP3);
+                       M_SAVE_REG(REG_SP, REG_ITMP3, REG_SP);
+               }
+       }
 
        /* save callee saved float registers (none right now) */
 #if 0
@@ -191,27 +243,58 @@ bool codegen(jitdata *jd)
 #endif
        
        
-       
-       /* take arguments out of register or stack frame */
-       
-       md = m->parseddesc;
+               /* call monitorenter function */
+#if defined(ENABLE_THREADS)
+       if (checksync && (m->flags & ACC_SYNCHRONIZED)) {
+               /* stack offset for monitor argument */
 
-       /* when storing locals, use this as base */
-       localbase = USESTACK;
-       
-       /* since the register allocator does not know about the shifting window
-        * arg regs need to be copied via the stack
-        */
-       if (md->argintreguse > 0) {
-               /* allocate scratch space for copying in to save(i&l) regs */
-               M_SUB_IMM(REG_SP, INT_ARG_CNT * 8, REG_SP);
-               
-               localbase += INT_ARG_CNT * 8;
-               
-               for (p = 0; p < INT_ARG_CNT; p++)
-                       M_STX(REG_WINDOW_TRANSPOSE(rd->argintregs[p]), REG_SP, USESTACK + (p * 8));
+               s1 = rd->memuse;
+
+               /* save float argument registers */
+
+               /* XXX jit-c-call */
+               slots = FLT_ARG_CNT;
+               ALIGN_STACK_SLOTS(slots);
+
+               M_LDA(REG_SP, REG_SP, -(slots * 8));
+               for (i = 0; i < FLT_ARG_CNT; i++)
+                       M_DST(abi_registers_float_argument[i], REG_SP, CSTACK +  i * 8);
+
+               s1 += slots;
+
+               /* get correct lock object */
+
+               if (m->flags & ACC_STATIC) {
+                       disp = dseg_add_address(cd, &m->class->object.header);
+                       M_ALD(REG_OUT0, REG_PV, disp);
+                       disp = dseg_add_functionptr(cd, LOCK_monitor_enter);
+                       M_ALD(REG_ITMP3, REG_PV, disp);
+               }
+               else {
+                       /* copy class pointer: $i0 -> $o0 */
+                       M_MOV(REG_RESULT_CALLEE, REG_OUT0);
+                       M_BNEZ(REG_OUT0, 3);
+                       disp = dseg_add_functionptr(cd, LOCK_monitor_enter);
+                       M_ALD(REG_ITMP3, REG_PV, disp);                   /* branch delay */
+                       M_ALD_INTERN(REG_ZERO, REG_ZERO, EXCEPTION_HARDWARE_NULLPOINTER);
+               }
+
+               M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
+               M_AST(REG_OUT0, REG_SP, CSTACK + s1 * 8);             /* branch delay */
+
+               /* restore float argument registers */
+
+               for (i = 0; i < FLT_ARG_CNT; i++)
+                       M_DLD(abi_registers_float_argument[i], REG_SP, CSTACK + i * 8);
+
+               M_LDA(REG_SP, REG_SP, slots * 8);
        }
+#endif
+
+
+       /* take arguments out of register or stack frame */
        
+       md = m->parseddesc;
 
        for (p = 0, l = 0; p < md->paramcount; p++) {
                t = md->paramtypes[p].type;
@@ -226,71 +309,96 @@ bool codegen(jitdata *jd)
                        continue;
 
                var = VAR(varindex);
+               s1 = md->params[p].regoff;
+               
+               if (IS_INT_LNG_TYPE(t)) {                    /* integer args          */                        
 
-               s1 = md->params[p].regoff;
-               if (IS_INT_LNG_TYPE(t)) {                    /* integer args          */
+                       s2 = var->vv.regoff;
+                       
                        if (!md->params[p].inmemory) {           /* register arguments    */
-                               /*s2 = rd->argintregs[s1];*/
-                               /*s2 = REG_WINDOW_TRANSPOSE(s2);*/
+                               s1 = REG_WINDOW_TRANSPOSE(s1);
+                               
                                if (!(var->flags & INMEMORY)) {      /* reg arg -> register   */
-                                       /*M_INTMOVE(s2, var->vv.regoff);*/
-                                       M_LDX(var->vv.regoff, REG_SP, USESTACK + (s1 * 8));
 
-                               } else {                             /* reg arg -> spilled    */
-                                       /*M_STX(s2, REG_SP, (WINSAVE_CNT + var->vv.regoff) * 8);*/
-                                       
-                                       M_LDX(REG_ITMP1, REG_SP, USESTACK + (s1 * 8));
-                                       M_STX(REG_ITMP1, REG_SP, localbase + (var->vv.regoff * 8));
+                                       /* the register allocator does not know about the window. */
+                                       /* avoid copying the locals from save to save regs by     */
+                                       /* swapping variables.                                    */
+
+                                       {
+                                       int old_dest = var->vv.regoff;
+                                       int new_dest = p + 24;
+
+                                       /* run through all variables */
+
+                                       for (i = 0; i < jd->varcount; i++) {
+                                               varinfo* uvar = VAR(i);
+
+                                               if (IS_FLT_DBL_TYPE(uvar->type) || IS_INMEMORY(uvar->flags))
+                                                       continue;
+
+                                               s2 = uvar->vv.regoff;
+
+                                               /* free the in reg by moving all other references */
+
+                                               if (s2 == new_dest) {
+                                                       uvar->vv.regoff = old_dest;
+                                                       /*printf("p%d-var[%d]: moved %d -> %d (to free save reg)\n", p, i, s2, old_dest);*/
+                                               }
+
+                                               /* move all variables to the in reg */
+
+                                               if (s2 == old_dest) {
+                                                       uvar->vv.regoff = new_dest;
+                                                       /*printf("p%d-var[%d]: moved %d -> %d (to avoid copy)\n", p, i, s2, new_dest);*/
+                                               }
+                                       }
+                                       }
+
+
+
+                               } 
+                               else {                             /* reg arg -> spilled    */
+                                       M_STX(s1, REG_SP, JITSTACK + var->vv.regoff);
                                }
 
                        } else {                                 /* stack arguments       */
                                if (!(var->flags & INMEMORY)) {      /* stack arg -> register */
-                                       M_LDX(var->vv.regoff, REG_FP, (WINSAVE_CNT + s1) * 8);
+                                       M_LDX(var->vv.regoff, REG_FP, JITSTACK + s1);
 
                                } else {                             /* stack arg -> spilled  */
-                                       assert(0); /* XXX winsave area in between */
-                                       var->vv.regoff = cd->stackframesize + s1;
+                                       /* add the callers window save registers */
+                                       var->vv.regoff = cd->stackframesize * 8 + s1;
                                }
                        }
                
                } else {                                     /* floating args         */
                        if (!md->params[p].inmemory) {           /* register arguments    */
-                               s2 = rd->argfltregs[s1];
                                if (!(var->flags & INMEMORY)) {      /* reg arg -> register   */
-                                       M_FLTMOVE(s2, var->vv.regoff);
+                                       M_FLTMOVE(s1, var->vv.regoff);
 
                                } else {                                         /* reg arg -> spilled    */
-                                       M_DST(s2, REG_SP, localbase + (var->vv.regoff) * 8);
+                                       M_DST(s1, REG_SP, JITSTACK + var->vv.regoff);
                                }
 
                        } else {                                 /* stack arguments       */
                                if (!(var->flags & INMEMORY)) {      /* stack-arg -> register */
-                                       M_DLD(var->vv.regoff, REG_FP, (WINSAVE_CNT + s1) * 8);
+                                       M_DLD(var->vv.regoff, REG_FP, JITSTACK + s1);
 
                                } else {                             /* stack-arg -> spilled  */
-                                       assert(0); /* XXX winsave area in between */
-                                       var->vv.regoff = cd->stackframesize + s1;
+                                       var->vv.regoff = cd->stackframesize * 8 + s1;
                                }
                        }
                }
        } /* end for */
        
-       if (md->argintreguse > 0) {
-               /* release scratch space */
-               M_ADD_IMM(REG_SP, INT_ARG_CNT * 8, REG_SP);
-       }
-       
-       
-       /* XXX monitor enter */
-
-
-
        
        }
        
        /* end of header generation */ 
        
-       replacementpoint = jd->code->rplpoints;
+       /* create replacement points */
+
+       REPLACEMENT_POINTS_INIT(cd, jd);
 
        /* walk through all basic blocks */
 
@@ -320,6 +428,7 @@ bool codegen(jitdata *jd)
                MCODECHECK(64+len);
                
 #if defined(ENABLE_LSRA)
+#error XXX LSRA not tested yet
                if (opt_lsra) {
                while (len) {
                        len--;
@@ -337,17 +446,17 @@ bool codegen(jitdata *jd)
                } else {
 #endif
                while (len) {
-                               len--;
+                       len--;
                        var = VAR(bptr->invars[len]);
                        if ((len == bptr->indepth-1) && (bptr->type == BBTYPE_EXH)) {
                                d = codegen_reg_of_var(0, var, REG_ITMP1);
-                                       M_INTMOVE(REG_ITMP1, d);
+                               M_INTMOVE(REG_ITMP2_XPTR, d);
                                emit_store(jd, NULL, var, d);
-                                                       }
+                       }
                        else {
                                assert((var->flags & INOUT));
-                                               }
-                                       }
+                       }
+               }
 #if defined(ENABLE_LSRA)
                }
 #endif
@@ -362,6 +471,7 @@ bool codegen(jitdata *jd)
                        }
 
                MCODECHECK(64);       /* an instruction usually needs < 64 words      */
+
                switch (iptr->opc) {
 
                case ICMD_INLINE_START:
@@ -370,6 +480,12 @@ bool codegen(jitdata *jd)
 
                case ICMD_NOP:        /* ...  ==> ...                                 */
                        break;
+
+               case ICMD_CHECKNULL:  /* ..., objectref  ==> ..., objectref           */
+
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       emit_nullpointer_check(cd, iptr, s1);
+                       break;
        
                /* constant operations ************************************************/
 
@@ -432,23 +548,23 @@ bool codegen(jitdata *jd)
                /* load/store/copy/move operations ************************************/
 
                case ICMD_ILOAD:      /* ...  ==> ..., content of local variable      */
-               case ICMD_LLOAD:      /* ...  ==> ..., content of local variable      */
-               case ICMD_ALOAD:      /* ...  ==> ..., content of local variable      */
-               case ICMD_FLOAD:      /* ...  ==> ..., content of local variable      */
-               case ICMD_DLOAD:      /* ...  ==> ..., content of local variable      */
+               case ICMD_LLOAD:
+               case ICMD_ALOAD:
+               case ICMD_FLOAD:
+               case ICMD_DLOAD:
                case ICMD_ISTORE:     /* ..., value  ==> ...                          */
-               case ICMD_LSTORE:     /* ..., value  ==> ...                          */
-               case ICMD_FSTORE:     /* ..., value  ==> ...                          */
-               case ICMD_DSTORE:     /* ..., value  ==> ...                          */
+               case ICMD_LSTORE:
+               case ICMD_FSTORE:
+               case ICMD_DSTORE:
                case ICMD_COPY:
                case ICMD_MOVE:
 
-                       emit_copy(jd, iptr, VAROP(iptr->s1), VAROP(iptr->dst));
+                       emit_copy(jd, iptr);
                        break;
        
                case ICMD_ASTORE:
                        if (!(iptr->flags.bits & INS_FLAG_RETADDR))
-                               emit_copy(jd, iptr, VAROP(iptr->s1), VAROP(iptr->dst));
+                               emit_copy(jd, iptr);
                        break;
 
 
@@ -498,7 +614,15 @@ bool codegen(jitdata *jd)
                        break;
 
                case ICMD_INT2CHAR:   /* ..., value  ==> ..., value                   */
-               case ICMD_INT2SHORT:
+               
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
+                       M_SLLX_IMM(s1, 48, d);
+                       M_SRLX_IMM( d, 48, d);
+                       emit_store_dst(jd, iptr, d);
+                       break;
+                       
+               case ICMD_INT2SHORT:   /* ..., value  ==> ..., value                   */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
@@ -627,7 +751,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       gen_div_check(s2);
+                       emit_arithmetic_check(cd, iptr, s2);
                        M_ISEXT(s1, s1);
                        /* XXX trim s2 like s1 ? */
                        M_DIVX(s1, s2, d);
@@ -639,7 +763,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       gen_div_check(s2);
+                       emit_arithmetic_check(cd, iptr, s2);
                        M_DIVX(s1, s2, d);
                        emit_store_dst(jd, iptr, d);
                        break;
@@ -649,12 +773,12 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
-                       gen_div_check(s2);
+                       emit_arithmetic_check(cd, iptr, s2);
                        M_ISEXT(s1, s1);
                        /* XXX trim s2 like s1 ? */
-                       M_DIVX(s1, s2, d);
-                       M_MULX(s2, d, d);
-                       M_SUB(s1, d, d);
+                       M_DIVX(s1, s2, REG_ITMP3);
+                       M_MULX(s2, REG_ITMP3, REG_ITMP3);
+                       M_SUB(s1, REG_ITMP3, d);
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -663,10 +787,10 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
-                       gen_div_check(s2);
-                       M_DIVX(s1, s2, d);
-                       M_MULX(s2, d, d);
-                       M_SUB(s1, d, d);
+                       emit_arithmetic_check(cd, iptr, s2);
+                       M_DIVX(s1, s2, REG_ITMP3);
+                       M_MULX(s2, REG_ITMP3, REG_ITMP3);
+                       M_SUB(s1, REG_ITMP3, d);
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -683,7 +807,15 @@ bool codegen(jitdata *jd)
                        break;
 
                case ICMD_ISHL:       /* ..., val1, val2  ==> ..., val1 << val2       */
-               case ICMD_LSHL:
+
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       s2 = emit_load_s2(jd, iptr, REG_ITMP2);
+                       d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
+                       M_SLL(s1, s2, d);
+                       emit_store_dst(jd, iptr, d);
+                       break;
+                       
+               case ICMD_LSHL:       /* ..., val1, val2  ==> ..., val1 << val2       */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
@@ -693,7 +825,16 @@ bool codegen(jitdata *jd)
                        break;
 
                case ICMD_ISHLCONST:  /* ..., value  ==> ..., value << constant       */
-               case ICMD_LSHLCONST:  /* val.i = constant                             */
+                                     /* val.i = constant                             */
+
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
+                       M_SLL_IMM(s1, iptr->sx.val.i, d);
+                       emit_store_dst(jd, iptr, d);
+                       break;
+                       
+               case ICMD_LSHLCONST:  /* ..., value  ==> ..., value << constant       */
+                                     /* val.i = constant                             */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
@@ -797,26 +938,27 @@ bool codegen(jitdata *jd)
                        emit_store_dst(jd, iptr, d);
                        break;
 
-               case ICMD_IREMPOW2:   /* ..., value  ==> ..., value % constant        */
+               case ICMD_IREMPOW2:   /* ..., value  ==> ..., value % constant           */
                                      /* sx.val.i = constant                             */
+                                     /* constant is actually constant - 1               */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       M_ISEXT(s1, s1); /* trim for 32-bit compare (BGEZ) */
                        if (s1 == d) {
                                M_MOV(s1, REG_ITMP1);
                                s1 = REG_ITMP1;
                        }
-                       if ((iptr->sx.val.i >= 0) && (iptr->sx.val.i <= 0xffff)) {
+                       M_ISEXT(s1, s1); /* trim for 32-bit compare (BGEZ) */
+                       if ((iptr->sx.val.i >= 0) && (iptr->sx.val.i <= 4095)) {
                                M_AND_IMM(s1, iptr->sx.val.i, d);
-                               M_BGEZ(s1, 4);
+                               M_BGEZ(s1, 5);
                                M_NOP;
                                M_SUB(REG_ZERO, s1, d);
                                M_AND_IMM(d, iptr->sx.val.i, d);
                        } else {
                                ICONST(REG_ITMP2, iptr->sx.val.i);
                                M_AND(s1, REG_ITMP2, d);
-                               M_BGEZ(s1, 4);
+                               M_BGEZ(s1, 5);
                                M_NOP;
                                M_SUB(REG_ZERO, s1, d);
                                M_AND(d, REG_ITMP2, d);
@@ -1049,7 +1191,7 @@ bool codegen(jitdata *jd)
                case ICMD_I2F:
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_FTMP3);
-                       disp = dseg_add_float(cd, 0.0);
+                       disp = dseg_add_unique_float(cd, 0.0);
                        M_IST (s1, REG_PV_CALLEE, disp);
                        M_FLD (d, REG_PV_CALLEE, disp);
                        M_CVTIF (d, d); /* rd gets translated to double target register */
@@ -1059,17 +1201,43 @@ bool codegen(jitdata *jd)
                case ICMD_I2D:
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_FTMP3);
-                       disp = dseg_add_float(cd, 0.0);
-                       M_IST (s1, REG_PV_CALLEE, disp);
-                       M_FLD (REG_FTMP2, REG_PV_CALLEE, disp); /* REG_FTMP2 needs to be a double temp */
+                       disp = dseg_add_unique_float(cd, 0.0);
+                       M_IST(s1, REG_PV_CALLEE, disp);
+                       M_FLD(REG_FTMP2, REG_PV_CALLEE, disp); /* REG_FTMP2 needs to be a double temp */
                        M_CVTID (REG_FTMP2, d); /* rd gets translated to double target register */
                        emit_store_dst(jd, iptr, d);
                        break;
+                       
+               case ICMD_L2F:
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       d = codegen_reg_of_dst(jd, iptr, REG_FTMP3);
+                       disp = dseg_add_unique_double(cd, 0.0);
+                       M_STX(s1, REG_PV_CALLEE, disp);
+                       M_DLD(REG_FTMP3, REG_PV_CALLEE, disp);
+                       M_CVTLF(REG_FTMP3, d);
+                       emit_store_dst(jd, iptr, d);
+                       break;
+                       
+               case ICMD_L2D:
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       d = codegen_reg_of_dst(jd, iptr, REG_FTMP3);
+                       disp = dseg_add_unique_double(cd, 0.0);
+                       M_STX(s1, REG_PV_CALLEE, disp);
+                       M_DLD(d, REG_PV_CALLEE, disp);
+                       M_CVTLD(d, d);
+                       emit_store_dst(jd, iptr, d);
+                       break;
 
                case ICMD_F2I:       /* ..., value  ==> ..., (int) value              */
                        s1 = emit_load_s1(jd, iptr, REG_FTMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
-                       disp = dseg_add_float(cd, 0.0);
+                       disp = dseg_add_unique_float(cd, 0.0);
+                       
+                       /* check for NaN, SPARC overflow is noncompliant (see V9 spec B.5)  */
+                       M_FCMP(s1, s1);
+                       M_FBU(5);
+                       M_MOV(REG_ZERO, d); /* delay slot */
+                       
                        M_CVTFI(s1, REG_FTMP2);
                        M_FST(REG_FTMP2, REG_PV_CALLEE, disp);
                        M_ILD(d, REG_PV, disp);
@@ -1080,7 +1248,13 @@ bool codegen(jitdata *jd)
                case ICMD_D2I:       /* ..., value  ==> ..., (int) value             */
                        s1 = emit_load_s1(jd, iptr, REG_FTMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
-                       disp = dseg_add_float(cd, 0.0);
+                       disp = dseg_add_unique_float(cd, 0.0);
+                       
+                       /* check for NaN, SPARC overflow is noncompliant (see V9 spec B.5)  */
+                       M_DCMP(s1, s1);
+                       M_FBU(5);
+                       M_MOV(REG_ZERO, d); /* delay slot */
+                       
                        M_CVTDI(s1, REG_FTMP2);
                        M_FST(REG_FTMP2, REG_PV, disp);
                        M_ILD(d, REG_PV, disp);
@@ -1090,7 +1264,13 @@ bool codegen(jitdata *jd)
                case ICMD_F2L:       /* ..., value  ==> ..., (long) value             */
                        s1 = emit_load_s1(jd, iptr, REG_FTMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
-                       disp = dseg_add_double(cd, 0.0);
+                       disp = dseg_add_unique_double(cd, 0.0);
+                       
+                       /* check for NaN, SPARC overflow is noncompliant (see V9 spec B.5)  */
+                       M_FCMP(s1, s1);
+                       M_FBU(5);
+                       M_MOV(REG_ZERO, d); /* delay slot */
+                       
                        M_CVTFL(s1, REG_FTMP2); /* FTMP2 needs to be double reg */
                        M_DST(REG_FTMP2, REG_PV, disp);
                        M_LDX(d, REG_PV, disp);
@@ -1100,7 +1280,13 @@ bool codegen(jitdata *jd)
                case ICMD_D2L:       /* ..., value  ==> ..., (long) value             */
                        s1 = emit_load_s1(jd, iptr, REG_FTMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
-                       disp = dseg_add_double(cd, 0.0);
+                       disp = dseg_add_unique_double(cd, 0.0);
+                       
+                       /* check for NaN, SPARC overflow is noncompliant (see V9 spec B.5)  */
+                       M_DCMP(s1, s1);
+                       M_FBU(5);
+                       M_MOV(REG_ZERO, d); /* delay slot */
+                       
                        M_CVTDL(s1, REG_FTMP2); /* FTMP2 needs to be double reg */
                        M_DST(REG_FTMP2, REG_PV, disp);
                        M_LDX(d, REG_PV, disp);
@@ -1130,9 +1316,9 @@ bool codegen(jitdata *jd)
                        s2 = emit_load_s2(jd, iptr, REG_FTMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
                        M_FCMP(s1,s2);
-                       M_OR_IMM(REG_ZERO, -1, REG_ITMP3); /* less by default (less or unordered) */
-                       M_CMOVFEQ_IMM(0, REG_ITMP3); /* 0 if equal */
-                       M_CMOVFGT_IMM(1, REG_ITMP3); /* 1 if greater */
+                       M_OR_IMM(REG_ZERO, -1, d); /* less by default (less or unordered) */
+                       M_CMOVFEQ_IMM(0, d); /* 0 if equal */
+                       M_CMOVFGT_IMM(1, d); /* 1 if greater */
                        emit_store_dst(jd, iptr, d);
                        break;
                        
@@ -1142,9 +1328,9 @@ bool codegen(jitdata *jd)
                        s2 = emit_load_s2(jd, iptr, REG_FTMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);
                        M_DCMP(s1,s2);
-                       M_OR_IMM(REG_ZERO, -1, REG_ITMP3); /* less by default (less or unordered) */
-                       M_CMOVFEQ_IMM(0, REG_ITMP3); /* 0 if equal */
-                       M_CMOVFGT_IMM(1, REG_ITMP3); /* 1 if greater */
+                       M_OR_IMM(REG_ZERO, -1, d); /* less by default (less or unordered) */
+                       M_CMOVFEQ_IMM(0, d); /* 0 if equal */
+                       M_CMOVFGT_IMM(1, d); /* 1 if greater */
                        emit_store_dst(jd, iptr, d);
                        break;
                        
@@ -1154,9 +1340,9 @@ bool codegen(jitdata *jd)
                        s2 = emit_load_s2(jd, iptr, REG_FTMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);    
                        M_FCMP(s1,s2);
-                       M_OR_IMM(REG_ZERO, 1, REG_ITMP3); /* greater by default (greater or unordered) */
-                       M_CMOVFEQ_IMM(0, REG_ITMP3); /* 0 if equal */
-                       M_CMOVFLT_IMM(-1, REG_ITMP3); /* -1 if less */
+                       M_OR_IMM(REG_ZERO, 1, d); /* greater by default (greater or unordered) */
+                       M_CMOVFEQ_IMM(0, d); /* 0 if equal */
+                       M_CMOVFLT_IMM(-1, d); /* -1 if less */
                        emit_store_dst(jd, iptr, d);
                        break;
                        
@@ -1166,9 +1352,9 @@ bool codegen(jitdata *jd)
                        s2 = emit_load_s2(jd, iptr, REG_FTMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP3);    
                        M_DCMP(s1,s2);
-                       M_OR_IMM(REG_ZERO, 1, REG_ITMP3); /* greater by default (greater or unordered) */
-                       M_CMOVFEQ_IMM(0, REG_ITMP3); /* 0 if equal */
-                       M_CMOVFLT_IMM(-1, REG_ITMP3); /* -1 if less */
+                       M_OR_IMM(REG_ZERO, 1, d); /* greater by default (greater or unordered) */
+                       M_CMOVFEQ_IMM(0, d); /* 0 if equal */
+                       M_CMOVFLT_IMM(-1, d); /* -1 if less */
                        emit_store_dst(jd, iptr, d);
                        break;
                        
@@ -1179,8 +1365,8 @@ bool codegen(jitdata *jd)
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       gen_nullptr_check(s1);
-                       M_ILD(d, s1, OFFSET(java_arrayheader, size));
+                       emit_nullpointer_check(cd, iptr, s1);
+                       M_ILD(d, s1, OFFSET(java_array_t, size));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1189,12 +1375,10 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP3);
-                       M_BLDS(d, REG_ITMP3, OFFSET(java_chararray, data[0]));
+                       M_BLDS(d, REG_ITMP3, OFFSET(java_bytearray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1203,13 +1387,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP3);
                        M_AADD(s2, REG_ITMP3, REG_ITMP3);
-                       M_SLDU(d, REG_ITMP3, OFFSET(java_chararray, data[0]));
+                       M_SLDU(d, REG_ITMP3, OFFSET(java_chararray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;                  
 
@@ -1218,13 +1400,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP3);
                        M_AADD(s2, REG_ITMP3, REG_ITMP3);
-                       M_SLDS(d, REG_ITMP3, OFFSET(java_chararray, data[0]));
+                       M_SLDS(d, REG_ITMP3, OFFSET(java_shortarray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1233,13 +1413,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 2, REG_ITMP3);
                        M_AADD(REG_ITMP3, s1, REG_ITMP3);
-                       M_ILD(d, REG_ITMP3, OFFSET(java_intarray, data[0]));
+                       M_ILD(d, REG_ITMP3, OFFSET(java_intarray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1248,13 +1426,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 3, REG_ITMP3);
                        M_AADD(REG_ITMP3, s1, REG_ITMP3);
-                       M_LDX(d, REG_ITMP3, OFFSET(java_longarray, data[0]));
+                       M_LDX(d, REG_ITMP3, OFFSET(java_longarray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1263,13 +1439,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_FTMP1);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 2, REG_ITMP3);
                        M_AADD(REG_ITMP3, s1, REG_ITMP3);
-                       M_FLD(d, REG_ITMP3, OFFSET(java_floatarray, data[0]));
+                       M_FLD(d, REG_ITMP3, OFFSET(java_floatarray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1278,13 +1452,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_FTMP1);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 3, REG_ITMP3);
                        M_AADD(REG_ITMP3, s1, REG_ITMP3);
-                       M_DLD(d, REG_ITMP3, OFFSET(java_doublearray, data[0]));
+                       M_DLD(d, REG_ITMP3, OFFSET(java_doublearray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1293,13 +1465,11 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, POINTERSHIFT, REG_ITMP3);
                        M_AADD(REG_ITMP3, s1, REG_ITMP3);
-                       M_ALD(d, REG_ITMP3, OFFSET(java_objectarray, data[0]));
+                       M_ALD(d, REG_ITMP3, OFFSET(java_objectarray_t, data[0]));
                        emit_store_dst(jd, iptr, d);
                        break;
 
@@ -1308,13 +1478,11 @@ bool codegen(jitdata *jd)
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_ITMP3);
-                       M_BST(s3, REG_ITMP1, OFFSET(java_bytearray, data[0]));
+                       M_BST(s3, REG_ITMP1, OFFSET(java_bytearray_t, data[0]));
                        break;
 
                case ICMD_CASTORE:    /* ..., arrayref, index, value  ==> ...         */
@@ -1322,70 +1490,60 @@ bool codegen(jitdata *jd)
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP1);
                        M_AADD(s2, REG_ITMP1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_ITMP3);
-                       M_SST(s3, REG_ITMP1, OFFSET(java_chararray, data[0]));
+                       M_SST(s3, REG_ITMP1, OFFSET(java_chararray_t, data[0]));
                        break;
 
                case ICMD_IASTORE:    /* ..., arrayref, index, value  ==> ...         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 2, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_ITMP3);
-                       M_IST_INTERN(s3, REG_ITMP1, OFFSET(java_intarray, data[0]));
+                       M_IST_INTERN(s3, REG_ITMP1, OFFSET(java_intarray_t, data[0]));
                        break;
 
                case ICMD_LASTORE:    /* ..., arrayref, index, value  ==> ...         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 3, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_ITMP3);
-                       M_STX_INTERN(s3, REG_ITMP1, OFFSET(java_longarray, data[0]));
+                       M_STX_INTERN(s3, REG_ITMP1, OFFSET(java_longarray_t, data[0]));
                        break;
 
                case ICMD_FASTORE:    /* ..., arrayref, index, value  ==> ...         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 2, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_FTMP1);
-                       M_FST_INTERN(s3, REG_ITMP1, OFFSET(java_floatarray, data[0]));
+                       M_FST_INTERN(s3, REG_ITMP1, OFFSET(java_floatarray_t, data[0]));
                        break;
 
                case ICMD_DASTORE:    /* ..., arrayref, index, value  ==> ...         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 3, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_FTMP1);
-                       M_DST_INTERN(s3, REG_ITMP1, OFFSET(java_doublearray, data[0]));
+                       M_DST_INTERN(s3, REG_ITMP1, OFFSET(java_doublearray_t, data[0]));
                        break;
 
 
@@ -1393,29 +1551,25 @@ bool codegen(jitdata *jd)
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        s3 = emit_load_s3(jd, iptr, REG_ITMP3);
 
-                       M_MOV(s1, rd->argintregs[0]);
-                       M_MOV(s3, rd->argintregs[1]);
+                       M_MOV(s1, REG_OUT0);
+                       M_MOV(s3, REG_OUT1);
                        disp = dseg_add_functionptr(cd, BUILTIN_canstore);
                        M_ALD(REG_ITMP3, REG_PV, disp);
                        M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
                        M_NOP;
-
-                       M_BEQZ(REG_RESULT_CALLER, 0);
-                       codegen_add_arraystoreexception_ref(cd);
-                       M_NOP;
+                       emit_exception_check(cd, iptr);
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_ASLL_IMM(s2, POINTERSHIFT, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
                        s3 = emit_load_s3(jd, iptr, REG_ITMP3);
-                       M_AST_INTERN(s3, REG_ITMP1, OFFSET(java_objectarray, data[0]));
+                       /* implicit null-pointer check */
+                       M_AST_INTERN(s3, REG_ITMP1, OFFSET(java_objectarray_t, data[0]));
                        break;
 
 
@@ -1423,12 +1577,10 @@ bool codegen(jitdata *jd)
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP1);
-                       M_BST(REG_ZERO, REG_ITMP1, OFFSET(java_bytearray, data[0]));
+                       M_BST(REG_ZERO, REG_ITMP1, OFFSET(java_bytearray_t, data[0]));
                        break;
 
                case ICMD_CASTORECONST:   /* ..., arrayref, index  ==> ...            */
@@ -1436,68 +1588,60 @@ bool codegen(jitdata *jd)
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_AADD(s2, s1, REG_ITMP1);
                        M_AADD(s2, REG_ITMP1, REG_ITMP1);
-                       M_SST(REG_ZERO, REG_ITMP1, OFFSET(java_chararray, data[0]));
+                       M_SST(REG_ZERO, REG_ITMP1, OFFSET(java_chararray_t, data[0]));
                        break;
 
                case ICMD_IASTORECONST:   /* ..., arrayref, index  ==> ...            */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 2, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
-                       M_IST_INTERN(REG_ZERO, REG_ITMP1, OFFSET(java_intarray, data[0]));
+                       M_IST_INTERN(REG_ZERO, REG_ITMP1, OFFSET(java_intarray_t, data[0]));
                        break;
 
                case ICMD_LASTORECONST:   /* ..., arrayref, index  ==> ...            */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, 3, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
-                       M_STX_INTERN(REG_ZERO, REG_ITMP1, OFFSET(java_longarray, data[0]));
+                       M_STX_INTERN(REG_ZERO, REG_ITMP1, OFFSET(java_longarray_t, data[0]));
                        break;
 
                case ICMD_AASTORECONST:   /* ..., arrayref, index  ==> ...            */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               gen_nullptr_check(s1);
-                               gen_bound_check;
-                       }
+                       /* implicit null-pointer check */
+                       emit_arrayindexoutofbounds_check(cd, iptr, s1, s2);
                        M_ASLL_IMM(s2, POINTERSHIFT, REG_ITMP2);
                        M_AADD(REG_ITMP2, s1, REG_ITMP1);
-                       M_AST_INTERN(REG_ZERO, REG_ITMP1, OFFSET(java_objectarray, data[0]));
+                       M_AST_INTERN(REG_ZERO, REG_ITMP1, OFFSET(java_objectarray_t, data[0]));
                        break;
                
 
                case ICMD_GETSTATIC:  /* ...  ==> ..., value                          */
 
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
-                               uf = iptr->sx.s23.s3.uf;
+                               uf        = iptr->sx.s23.s3.uf;
                                fieldtype = uf->fieldref->parseddesc.fd->type;
                                disp      = dseg_add_unique_address(cd, uf);
 
                                codegen_add_patch_ref(cd, PATCHER_get_putstatic, uf, disp);
                        } 
                        else {
-                               fi = iptr->sx.s23.s3.fmiref->p.field;
+                               fi        = iptr->sx.s23.s3.fmiref->p.field;
                                fieldtype = fi->type;
-                               disp = dseg_add_address(cd, &(fi->value));
+                               disp      = dseg_add_address(cd, fi->value);
 
                                if (!CLASS_IS_OR_ALMOST_INITIALIZED(fi->class))
                                        codegen_add_patch_ref(cd, PATCHER_clinit, fi->class, disp);
@@ -1533,16 +1677,16 @@ bool codegen(jitdata *jd)
                case ICMD_PUTSTATIC:  /* ..., value  ==> ...                          */
 
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
-                               uf = iptr->sx.s23.s3.uf;
+                               uf        = iptr->sx.s23.s3.uf;
                                fieldtype = uf->fieldref->parseddesc.fd->type;
                                disp      = dseg_add_unique_address(cd, uf);
 
                                codegen_add_patch_ref(cd, PATCHER_get_putstatic, uf, disp);
                        } 
                        else {
-                               fi = iptr->sx.s23.s3.fmiref->p.field;
+                               fi        = iptr->sx.s23.s3.fmiref->p.field;
                                fieldtype = fi->type;
-                               disp = dseg_add_address(cd, &(fi->value));
+                               disp      = dseg_add_address(cd, fi->value);
 
                                if (!CLASS_IS_OR_ALMOST_INITIALIZED(fi->class))
                                        codegen_add_patch_ref(cd, PATCHER_clinit, fi->class, disp);
@@ -1581,14 +1725,14 @@ bool codegen(jitdata *jd)
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
                                uf        = iptr->sx.s23.s3.uf;
                                fieldtype = uf->fieldref->parseddesc.fd->type;
-                               disp = dseg_add_unique_address(cd, uf);
+                               disp      = dseg_add_unique_address(cd, uf);
 
                                codegen_add_patch_ref(cd, PATCHER_get_putstatic, uf, disp);
                        } 
                        else {
                                fi        = iptr->sx.s23.s3.fmiref->p.field;
                                fieldtype = fi->type;
-                               disp      = dseg_add_address(cd, &(fi->value));
+                               disp      = dseg_add_address(cd, fi->value);
 
                                if (!CLASS_IS_OR_ALMOST_INITIALIZED(fi->class))
                                        codegen_add_patch_ref(cd, PATCHER_clinit, fi->class, disp);
@@ -1619,7 +1763,7 @@ bool codegen(jitdata *jd)
                case ICMD_GETFIELD:   /* ...  ==> ..., value                          */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       gen_nullptr_check(s1);
+                       emit_nullpointer_check(cd, iptr, s1);
 
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
                                uf = iptr->sx.s23.s3.uf;
@@ -1666,7 +1810,7 @@ bool codegen(jitdata *jd)
                case ICMD_PUTFIELD:   /* ..., objectref, value  ==> ...               */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       gen_nullptr_check(s1);
+                       emit_nullpointer_check(cd, iptr, s1);
 
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
                                uf = iptr->sx.s23.s3.uf;
@@ -1715,7 +1859,7 @@ bool codegen(jitdata *jd)
                                          /* following NOP)                           */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       gen_nullptr_check(s1);
+                       emit_nullpointer_check(cd, iptr, s1);
 
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
                                unresolved_field *uf = iptr->sx.s23.s3.uf;
@@ -1770,18 +1914,15 @@ bool codegen(jitdata *jd)
 
 #ifdef ENABLE_VERIFIER
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
-                               codegen_addpatchref(cd, PATCHER_athrow_areturn,
-                                                                       iptr->sx.s23.s2.uc, 0);
+                               unresolved_class *uc = iptr->sx.s23.s2.uc;
 
-                               if (opt_showdisassemble) {
-                                       M_NOP; M_NOP;
-                               }
+                               codegen_add_patch_ref(cd, PATCHER_athrow_areturn, uc, 0);
                        }
 #endif /* ENABLE_VERIFIER */
 
                        disp = dseg_add_functionptr(cd, asm_handle_exception);
-                       M_ALD(REG_ITMP2, REG_PV, disp);
-                       M_JMP(REG_ITMP3_XPC, REG_ITMP2, REG_ZERO);
+                       M_ALD(REG_ITMP1, REG_PV, disp);
+                       M_JMP(REG_ITMP3_XPC, REG_ITMP1, REG_ZERO);
                        M_NOP;
                        M_NOP;              /* nop ensures that XPC is less than the end */
                                            /* of basic block                            */
@@ -1791,158 +1932,115 @@ bool codegen(jitdata *jd)
                case ICMD_GOTO:         /* ... ==> ...                                */
                case ICMD_RET:          /* ... ==> ...                                */
 
-                       M_BR(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_br(cd, iptr->dst.block);
                        ALIGNCODENOP;
                        break;
 
                case ICMD_JSR:          /* ... ==> ...                                */
 
-                       M_BR(0);
-                       codegen_add_branch_ref(cd, iptr->sx.s23.s3.jsrtarget.block);
-                       M_NOP;
+                       emit_br(cd, iptr->sx.s23.s3.jsrtarget.block);
                        ALIGNCODENOP;
                        break;
 
                case ICMD_IFNULL:       /* ..., value ==> ...                         */
+               case ICMD_IFNONNULL:
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       M_BEQZ(s1, 0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
-                       break;
-
-               case ICMD_IFNONNULL:    /* ..., value ==> ...                         */
-
-                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       M_BNEZ(s1, 0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bccz(cd, iptr->dst.block, iptr->opc - ICMD_IFNULL, s1, BRANCH_OPT_NONE);
                        break;
+                       
+               /* Note: int compares must not branch on the register directly.       */
+               /* Reason is, that register content is not 32-bit clean.              */
 
                case ICMD_IFEQ:         /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.i == 0) {
-                               M_BEQZ(s1, 0);
-                       } else {
-                               if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
-                                       }
-                               else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.i);
-                                       M_CMP(s1, REG_ITMP2);
-                                       }
-                               M_BEQ(0);
-                               }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       
+                       if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
+                               M_CMP_IMM(s1, iptr->sx.val.i);
+                       }
+                       else {
+                               ICONST(REG_ITMP2, iptr->sx.val.i);
+                               M_CMP(s1, REG_ITMP2);
+                       }
+                       emit_beq(cd, iptr->dst.block);
                        break;
 
                case ICMD_IFLT:         /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.i == 0) {
-                               M_BLTZ(s1, 0);
-                       } else {
-                               if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
-                               } else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.i);
-                                       M_CMP(s1, REG_ITMP2);
-                               }
-                               M_BLT(0);
+                       
+                       if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
+                               M_CMP_IMM(s1, iptr->sx.val.i);
+                       } 
+                       else {
+                               ICONST(REG_ITMP2, iptr->sx.val.i);
+                               M_CMP(s1, REG_ITMP2);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_blt(cd, iptr->dst.block);
                        break;
 
                case ICMD_IFLE:         /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.i == 0) {
-                               M_BLEZ(s1, 0);
-                               }
+
+                       if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
+                               M_CMP_IMM(s1, iptr->sx.val.i);
+                       }
                        else {
-                               if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
-                                       }
-                               else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.i);
-                                       M_CMP(s1, REG_ITMP2);
-                               }
-                               M_BLE(0);
+                               ICONST(REG_ITMP2, iptr->sx.val.i);
+                               M_CMP(s1, REG_ITMP2);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_ble(cd, iptr->dst.block);
                        break;
 
                case ICMD_IFNE:         /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.i == 0) {
-                               M_BNEZ(s1, 0);
-                               }
+               
+                       if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
+                               M_CMP_IMM(s1, iptr->sx.val.i);
+                       }
                        else {
-                               if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
-                               }
-                               else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.i);
-                                       M_CMP(s1, REG_ITMP2);
-                               }
-                               M_BNE(0);
+                               ICONST(REG_ITMP2, iptr->sx.val.i);
+                               M_CMP(s1, REG_ITMP2);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bne(cd, iptr->dst.block);
                        break;
                                                
                case ICMD_IFGT:         /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.i == 0) {
-                               M_BGTZ(s1, 0);
+               
+                       if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
+                               M_CMP_IMM(s1, iptr->sx.val.i);
                        } 
                        else {
-                               if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
-                               } else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.i);
-                                       M_CMP(s1, REG_ITMP2);
-                               }
-                               M_BGT(0);
+                               ICONST(REG_ITMP2, iptr->sx.val.i);
+                               M_CMP(s1, REG_ITMP2);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bgt(cd, iptr->dst.block);          
                        break;
 
                case ICMD_IFGE:         /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.i == 0) {
-                               M_BGEZ(s1, 0);
-                               }
+
+                       if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
+                               M_CMP_IMM(s1, iptr->sx.val.i);
+                       }
                        else {
-                               if ((iptr->sx.val.i >= -4096) && (iptr->sx.val.i <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
-                                       }
-                               else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.i);
-                                       M_CMP(s1, REG_ITMP2);
-                               }
-                               M_BGE(0);
+                               ICONST(REG_ITMP2, iptr->sx.val.i);
+                               M_CMP(s1, REG_ITMP2);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bge(cd, iptr->dst.block);
                        break;
                        
                case ICMD_IF_LEQ:       /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.l == 0) {
-                               M_BEQZ(s1, 0);
-                       }
+                       if (iptr->sx.val.l == 0)
+                               emit_beqz(cd, iptr->dst.block, s1);
                        else {
                                if ((iptr->sx.val.l >= -4096) && (iptr->sx.val.l <= 4095)) {
                                        M_CMP_IMM(s1, iptr->sx.val.l);
@@ -1951,108 +2049,93 @@ bool codegen(jitdata *jd)
                                        LCONST(REG_ITMP2, iptr->sx.val.l);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_XBEQ(0);
+                               emit_beq_xcc(cd, iptr->dst.block);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
                        break;
                        
                case ICMD_IF_LLT:       /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.l == 0) {
-                               M_BLTZ(s1, 0);
-                       } 
+                       if (iptr->sx.val.l == 0)
+                               emit_bltz(cd, iptr->dst.block, s1);
                        else {
                                if ((iptr->sx.val.l >= -4096) && (iptr->sx.val.l <= 4095)) {
                                        M_CMP_IMM(s1, iptr->sx.val.l);
                                } 
                                else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.l);
+                                       LCONST(REG_ITMP2, iptr->sx.val.l);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_XBLT(0);
+                               emit_blt_xcc(cd, iptr->dst.block);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
                        break;
 
                case ICMD_IF_LLE:       /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.l == 0) {
-                               M_BLEZ(s1, 0);
-                               }
+                       if (iptr->sx.val.l == 0)
+                               emit_blez(cd, iptr->dst.block, s1);
                        else {
                                if ((iptr->sx.val.l >= -4096) && (iptr->sx.val.l <= 4095)) {
                                        M_CMP_IMM(s1, iptr->sx.val.l);
-                                       }
+                               }
                                else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.l);
+                                       LCONST(REG_ITMP2, iptr->sx.val.l);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_XBLE(0);
+                               emit_ble_xcc(cd, iptr->dst.block);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
                        break;
                        
                case ICMD_IF_LNE:       /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.l == 0) {
-                               M_BNEZ(s1, 0);
-                               }
+                       if (iptr->sx.val.l == 0)
+                               emit_bnez(cd, iptr->dst.block, s1);
                        else {
                                if ((iptr->sx.val.l >= -4096) && (iptr->sx.val.l <= 4095)) {
-                                       M_CMP_IMM(s1, iptr->sx.val.i);
+                                       M_CMP_IMM(s1, iptr->sx.val.l);
                                }
                                else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.l);
+                                       LCONST(REG_ITMP2, iptr->sx.val.l);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_XBNE(0);
+                               emit_bne_xcc(cd, iptr->dst.block);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
                        break;
                                                
                case ICMD_IF_LGT:       /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.l == 0) {
-                               M_BLTZ(s1, 0);
-                       else {
+                       if (iptr->sx.val.l == 0)
+                               emit_bgtz(cd, iptr->dst.block, s1);
+                       else {
                                if ((iptr->sx.val.l >= -4096) && (iptr->sx.val.l <= 4095)) {
                                        M_CMP_IMM(s1, iptr->sx.val.l);
-                               } else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.l);
+                               } 
+                               else {
+                                       LCONST(REG_ITMP2, iptr->sx.val.l);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_XBGT(0);
+                               emit_bgt_xcc(cd, iptr->dst.block);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
                        break;
 
                case ICMD_IF_LGE:       /* ..., value ==> ...                         */
 
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                       if (iptr->sx.val.l == 0) {
-                               M_BLEZ(s1, 0);
-                               }
+                       if (iptr->sx.val.l == 0)
+                               emit_bgez(cd, iptr->dst.block, s1);
                        else {
                                if ((iptr->sx.val.l >= -4096) && (iptr->sx.val.l <= 4095)) {
                                        M_CMP_IMM(s1, iptr->sx.val.l);
-                                       }
+                               }
                                else {
-                                       ICONST(REG_ITMP2, iptr->sx.val.l);
+                                       LCONST(REG_ITMP2, iptr->sx.val.l);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_XBGE(0);
+                               emit_bge_xcc(cd, iptr->dst.block);
                        }
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
                        break;                  
                        
 
@@ -2062,9 +2145,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_XBEQ(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_beq_xcc(cd, iptr->dst.block);
                        break;
 
                case ICMD_IF_ICMPEQ:    /* 32-bit compare                             */
@@ -2072,9 +2153,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BEQ(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_beq(cd, iptr->dst.block);
                        break;
 
                case ICMD_IF_ACMPNE:    /* ..., value, value ==> ...                  */
@@ -2083,9 +2162,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_XBNE(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bne_xcc(cd, iptr->dst.block);
                        break;
                        
                case ICMD_IF_ICMPNE:    /* 32-bit compare                             */
@@ -2093,9 +2170,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BNE(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bne(cd, iptr->dst.block);
                        break;
 
                case ICMD_IF_LCMPLT:    /* ..., value, value ==> ...                  */
@@ -2103,9 +2178,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_XBLT(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_blt_xcc(cd, iptr->dst.block);
                        break;
                        
                case ICMD_IF_ICMPLT:    /* 32-bit compare                             */
@@ -2113,9 +2186,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BLT(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_blt(cd, iptr->dst.block);
                        break;
 
                case ICMD_IF_LCMPGT:    /* ..., value, value ==> ...                  */
@@ -2123,9 +2194,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_XBGT(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bgt_xcc(cd, iptr->dst.block);
                        break;
                        
                case ICMD_IF_ICMPGT:    /* 32-bit compare                             */
@@ -2133,9 +2202,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BGT(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bgt(cd, iptr->dst.block);
                        break;
 
                case ICMD_IF_LCMPLE:    /* ..., value, value ==> ...                  */
@@ -2143,9 +2210,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BLE(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_ble_xcc(cd, iptr->dst.block);
                        break;
                        
                case ICMD_IF_ICMPLE:    /* 32-bit compare                             */
@@ -2153,9 +2218,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BLE(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_ble(cd, iptr->dst.block);
                        break;                  
        
 
@@ -2164,9 +2227,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BGE(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bge_xcc(cd, iptr->dst.block);
                        break;
                        
                case ICMD_IF_ICMPGE:    /* 32-bit compare                             */
@@ -2174,9 +2235,7 @@ bool codegen(jitdata *jd)
                        s1 = emit_load_s1(jd, iptr, REG_ITMP1);
                        s2 = emit_load_s2(jd, iptr, REG_ITMP2);
                        M_CMP(s1, s2);
-                       M_BGE(0);
-                       codegen_add_branch_ref(cd, iptr->dst.block);
-                       M_NOP;
+                       emit_bge(cd, iptr->dst.block);
                        break;
 
 
@@ -2194,12 +2253,9 @@ bool codegen(jitdata *jd)
 
 #ifdef ENABLE_VERIFIER
                        if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
-                               codegen_addpatchref(cd, PATCHER_athrow_areturn,
-                                                                       iptr->sx.s23.s2.uc, 0);
+                               unresolved_class *uc = iptr->sx.s23.s2.uc;
 
-                               if (opt_showdisassemble) {
-                                       M_NOP; M_NOP;
-                               }
+                               codegen_add_patch_ref(cd, PATCHER_athrow_areturn, uc, 0);
                        }
 #endif /* ENABLE_VERIFIER */
                        goto nowperformreturn;
@@ -2208,7 +2264,7 @@ bool codegen(jitdata *jd)
                case ICMD_DRETURN:
 
                        s1 = emit_load_s1(jd, iptr, REG_FRESULT);
-                       M_FLTMOVE(s1, REG_FRESULT);
+                       M_DBLMOVE(s1, REG_FRESULT);
                        goto nowperformreturn;
 
                case ICMD_RETURN:       /* ...  ==> ...                               */
@@ -2226,24 +2282,32 @@ nowperformreturn:
 
 #if defined(ENABLE_THREADS)
                        if (checksync && (m->flags & ACC_SYNCHRONIZED)) {
-/* XXX: REG_RESULT is save, but what about FRESULT? */
-                               M_ALD(rd->argintregs[0], REG_SP, rd->memuse * 8); /* XXX: what for ? */
+                               /* XXX jit-c-call */
+                               disp = dseg_add_functionptr(cd, LOCK_monitor_exit);
+                               M_ALD(REG_ITMP3, REG_PV, disp);
+
+                               /* we need to save fp return value (int saved by window) */
 
                                switch (iptr->opc) {
                                case ICMD_FRETURN:
                                case ICMD_DRETURN:
-                                       M_DST(REG_FRESULT, REG_SP, rd->memuse * 8);
-                                       break;
-                               }
+                                       M_ALD(REG_OUT0, REG_SP, CSTACK + rd->memuse * 8);
+                                       M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
+                                       M_DST(REG_FRESULT, REG_SP, CSTACK + rd->memuse * 8); /* delay */
 
-                               disp = dseg_add_functionptr(cd, BUILTIN_monitorexit);
-                               M_ALD(REG_ITMP3, REG_PV, disp);
-                               M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO); /*REG_RA_CALLER */
+                                       /* restore the fp return value */
 
-                               switch (iptr->opc) {
-                               case ICMD_FRETURN:
-                               case ICMD_DRETURN:
-                                       M_DLD(REG_FRESULT, REG_SP, rd->memuse * 8);
+                                       M_DLD(REG_FRESULT, REG_SP, CSTACK + rd->memuse * 8);
+                                       break;
+                               case ICMD_IRETURN:
+                               case ICMD_LRETURN:
+                               case ICMD_ARETURN:
+                               case ICMD_RETURN:
+                                       M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
+                                       M_ALD(REG_OUT0, REG_SP, CSTACK + rd->memuse * 8); /* delay */
+                                       break;
+                               default:
+                                       assert(false);
                                        break;
                                }
                        }
@@ -2271,29 +2335,27 @@ nowperformreturn:
                        if (l == 0) {
                                M_INTMOVE(s1, REG_ITMP1);
                        }
-                       else if (l <= 4095) {
+                       else if (-l >= 4096 && -l <= 4095) {
                                M_ADD_IMM(s1, -l, REG_ITMP1);
                        }
                        else {
                                ICONST(REG_ITMP2, l);
-                               /* XXX: do I need to truncate s1 to 32-bit ? */
                                M_SUB(s1, REG_ITMP2, REG_ITMP1);
                        }
-                       i = i - l + 1;
+
+                       i = i - l + 1; /* number of targets (>0) */
 
 
                        /* range check */
                                        
                        if (i <= 4095) {
-                               M_CMP_IMM(REG_ITMP1, i);
+                               M_CMP_IMM(REG_ITMP1, i - 1);
                        }
                        else {
-                               ICONST(REG_ITMP2, i);
+                               ICONST(REG_ITMP2, i - 1);
                                M_CMP(REG_ITMP1, REG_ITMP2);
                        }               
-                       M_XBULT(0);
-                       codegen_add_branch_ref(cd, table[0].block); /* default target */
-                       M_ASLL_IMM(REG_ITMP1, POINTERSHIFT, REG_ITMP1);      /* delay slot*/
+                       emit_bugt(cd, table[0].block); /* default target */
 
                        /* build jump table top down and use address of lowest entry */
 
@@ -2307,6 +2369,7 @@ nowperformreturn:
 
                        /* length of dataseg after last dseg_addtarget is used by load */
 
+                       M_ASLL_IMM(REG_ITMP1, POINTERSHIFT, REG_ITMP1);
                        M_AADD(REG_ITMP1, REG_PV, REG_ITMP2);
                        M_ALD(REG_ITMP2, REG_ITMP2, -(cd->dseglen));
                        M_JMP(REG_ZERO, REG_ITMP2, REG_ZERO);
@@ -2333,15 +2396,11 @@ nowperformreturn:
                                        ICONST(REG_ITMP2, lookup->value);
                                        M_CMP(s1, REG_ITMP2);
                                }
-                               M_BEQ(0);
-                               codegen_add_branch_ref(cd, lookup->target.block); 
-                               M_NOP;
+                               emit_beq(cd, lookup->target.block);
                                ++lookup;
                        }
 
-                       M_BR(0);
-                       codegen_add_branch_ref(cd, iptr->sx.s23.s3.lookupdefault.block);
-                       M_NOP;
+                       emit_br(cd, iptr->sx.s23.s3.lookupdefault.block);
                        ALIGNCODENOP;
                        break;
                        }
@@ -2352,8 +2411,50 @@ nowperformreturn:
                        bte = iptr->sx.s23.s3.bte;
                        md = bte->md;
                        
-                       /* XXX: proper builtin calling and float args are so not implemented */
-                       assert(md->paramcount <= 5 && md->argfltreguse < 1);
+                       /* XXX: builtin calling with stack arguments not implemented */
+                       assert(md->paramcount <= 5 && md->argfltreguse <= 16);
+                       
+                       s3 = md->paramcount;
+
+                       MCODECHECK((s3 << 1) + 64);
+
+#ifdef BUILTIN_FLOAT_ARGS /* float args for builtins disabled */
+
+                       /* copy float arguments according to ABI convention */
+
+                       int num_fltregargs = 0;
+                       int fltregarg_inswap[16];
+
+                       for (s3 = s3 - 1; s3 >= 0; s3--) {
+                               var = VAR(iptr->sx.s23.s2.args[s3]);
+
+                               if (IS_FLT_DBL_TYPE(var->type)) {
+                                       if (!md->params[s3].inmemory) {
+                                               s1 = s3; /*native flt args use argument index directly*/
+                                               d = emit_load(jd, iptr, var, REG_FTMP1);
+                                               
+                                               M_DMOV(d, s1 + 16);
+                                               fltregarg_inswap[num_fltregargs] = s1;
+                                               num_fltregargs++;
+                                               /*printf("builtin: flt arg swap to %d\n", s1 + 16);*/
+                                       }
+                                       else {
+                                               assert(0);
+                                       }
+                               }
+                       }
+                       
+                       int i;
+                       /* move swapped float args to target regs */
+                       for (i = 0; i < num_fltregargs; i++) {
+                               s1 = fltregarg_inswap[i];
+                               M_DMOV(s1 + 16, s1);
+                               /*printf("builtin float arg to target reg: %d ==> %d\n", s1+16, s1);*/
+                       }
+                       
+#else
+                       assert(md->argfltreguse == 0);
+#endif
                        
                        goto gen_method;
 
@@ -2383,36 +2484,37 @@ gen_method:
 
                        for (s3 = s3 - 1; s3 >= 0; s3--) {
                                var = VAR(iptr->sx.s23.s2.args[s3]);
+                               d  = md->params[s3].regoff;
 
                                if (var->flags & PREALLOC)
                                        continue;
 
                                if (IS_INT_LNG_TYPE(var->type)) {
                                        if (!md->params[s3].inmemory) {
-                                               s1 = rd->argintregs[md->params[s3].regoff];
-                                               d = emit_load(jd, iptr, var, s1);
-                                               M_INTMOVE(d, s1);
+                                               s1 = emit_load(jd, iptr, var, d);
+                                               M_INTMOVE(s1, d);
                                        } 
                                        else {
-                                               d = emit_load(jd, iptr, var, REG_ITMP1);
-                                               M_STX(d, REG_SP, md->params[s3].regoff * 8);
+                                               s1 = emit_load(jd, iptr, var, REG_ITMP1);
+                                               M_STX(s1, REG_SP, JITSTACK + d);
                                        }
                                }
                                else {
+#ifdef BUILTIN_FLOAT_ARGS
+                                       if (iptr->opc == ICMD_BUILTIN)
+                                               continue;
+#endif
+                                               
                                        if (!md->params[s3].inmemory) {
-                                               s1 = rd->argfltregs[md->params[s3].regoff];
-                                               d = emit_load(jd, iptr, var, s1);
+                                               s1 = emit_load(jd, iptr, var, d);
                                                if (IS_2_WORD_TYPE(var->type))
-                                                       M_DMOV(d, s1);
+                                                       M_DMOV(s1, d);
                                                else
-                                                       M_FMOV(d, s1);
+                                                       M_FMOV(s1, d);
                                        }
                                        else {
-                                               d = emit_load(jd, iptr, var, REG_FTMP1);
-                                               if (IS_2_WORD_TYPE(var->type))
-                                                       M_DST(d, REG_SP, md->params[s3].regoff * 8);
-                                               else
-                                                       M_FST(d, REG_SP, md->params[s3].regoff * 8);
+                                               s1 = emit_load(jd, iptr, var, REG_FTMP1);
+                                               M_DST(s1, REG_SP, JITSTACK + d);
                                        }
                                }
                        }
@@ -2422,18 +2524,26 @@ gen_method:
                                disp = dseg_add_functionptr(cd, bte->fp);
 
                                M_ALD(REG_PV_CALLER, REG_PV, disp);  /* built-in-function pointer */
-                               s1 = REG_PV_CALLER;
-
-                               /* c call, allocate parameter array */
-                               M_LDA(REG_SP, REG_SP, -(ABI_PARAMARRAY_SLOTS) * 8);
 
+                               /* XXX jit-c-call */
+                               /* generate the actual call */
+    
+                           M_JMP(REG_RA_CALLER, REG_PV_CALLER, REG_ZERO);
+                           M_NOP;
+                           disp = (s4) (cd->mcodeptr - cd->mcodebase);
+                           /* REG_RA holds the value of the jmp instruction, therefore +8 */
+                           M_LDA(REG_ZERO, REG_RA_CALLER, -disp + 8); 
+
+                               emit_exception_check(cd, iptr);
+                               if (md->returntype.type == TYPE_FLT) {
+                                       /* special handling for float return value in %f0 */
+                                       M_FMOV_INTERN(0,1);
+                               }
                                break;
 
                        case ICMD_INVOKESPECIAL:
-                               M_BEQZ(REG_OUT0, 0);
-                               codegen_add_nullpointerexception_ref(cd);
-                               M_NOP;
-                               /* fall through */
+                               emit_nullpointer_check(cd, iptr, REG_OUT0);
+                               /* fall-through */
 
                        case ICMD_INVOKESTATIC:
                                if (lm == NULL) {
@@ -2446,11 +2556,18 @@ gen_method:
                                        disp = dseg_add_address(cd, lm->stubroutine);
 
                                M_ALD(REG_PV_CALLER, REG_PV, disp);          /* method pointer in pv */
-                               s1 = REG_PV_CALLER;
+                               
+                               /* generate the actual call */
+    
+                           M_JMP(REG_RA_CALLER, REG_PV_CALLER, REG_ZERO);
+                           M_NOP;
+                           disp = (s4) (cd->mcodeptr - cd->mcodebase);
+                           /* REG_RA holds the value of the jmp instruction, therefore +8 */
+                           M_LDA(REG_ZERO, REG_RA_CALLER, -disp + 8); 
                                break;
 
                        case ICMD_INVOKEVIRTUAL:
-                               gen_nullptr_check(REG_OUT0);
+                               emit_nullpointer_check(cd, iptr, REG_OUT0);
 
                                if (lm == NULL) {
                                        codegen_add_patch_ref(cd, PATCHER_invokevirtual, um, 0);
@@ -2461,21 +2578,24 @@ gen_method:
                                        s1 = OFFSET(vftbl_t, table[0]) +
                                                sizeof(methodptr) * lm->vftblindex;
 
-                               M_ALD(REG_METHODPTR, REG_OUT0,
-                                         OFFSET(java_objectheader, vftbl));
+                               /* implicit null-pointer check */
+                               M_ALD(REG_METHODPTR, REG_OUT0,OFFSET(java_object_t, vftbl));
                                M_ALD(REG_PV_CALLER, REG_METHODPTR, s1);
-                               s1 = REG_PV_CALLER;
+                               
+                               /* generate the actual call */
+    
+                           M_JMP(REG_RA_CALLER, REG_PV_CALLER, REG_ZERO);
+                           M_NOP;
+                           disp = (s4) (cd->mcodeptr - cd->mcodebase);
+                           /* REG_RA holds the value of the jmp instruction, therefore +8 */
+                           M_LDA(REG_ZERO, REG_RA_CALLER, -disp + 8); 
                                break;
 
                        case ICMD_INVOKEINTERFACE:
-                               gen_nullptr_check(rd->argintregs[0]);
+                               emit_nullpointer_check(cd, iptr, REG_OUT0);
 
                                if (lm == NULL) {
-                                       codegen_addpatchref(cd, PATCHER_invokeinterface, um, 0);
-
-                                       if (opt_showdisassemble) {
-                                               M_NOP; M_NOP;
-                                       }
+                                       codegen_add_patch_ref(cd, PATCHER_invokeinterface, um, 0);
 
                                        s1 = 0;
                                        s2 = 0;
@@ -2487,34 +2607,19 @@ gen_method:
                                        s2 = sizeof(methodptr) * (lm - lm->class->methods);
                                }
 
-                               M_ALD(REG_METHODPTR, REG_OUT0,
-                                         OFFSET(java_objectheader, vftbl));
+                               /* implicit null-pointer check */
+                               M_ALD(REG_METHODPTR, REG_OUT0, OFFSET(java_object_t, vftbl));
                                M_ALD(REG_METHODPTR, REG_METHODPTR, s1);
                                M_ALD(REG_PV_CALLER, REG_METHODPTR, s2);
-                               s1 = REG_PV_CALLER;
-                               break;
-                       }
-
-                       /* generate the actual call */
-
-                       M_JMP(REG_RA_CALLER, s1, REG_ZERO);
-                       M_NOP;
-                       disp = (s4) (cd->mcodeptr - cd->mcodebase);
-                       /* REG_RA holds the value of the jmp instruction, therefore +8 */
-                       M_LDA(REG_ZERO, REG_RA_CALLER, -disp + 8); 
-
-                       if (iptr->opc == ICMD_BUILTIN) {
-                               /* remove param slots */
-                               M_LDA(REG_SP, REG_SP, (ABI_PARAMARRAY_SLOTS) * 8);
-                       }
-
-
-                       /* actually only used for ICMD_BUILTIN */
 
-                       if (INSTRUCTION_MUST_CHECK(iptr)) {
-                               M_BEQZ(REG_RESULT_CALLER, 0);
-                               codegen_add_fillinstacktrace_ref(cd);
-                               M_NOP;
+                           /* generate the actual call */
+    
+                           M_JMP(REG_RA_CALLER, REG_PV_CALLER, REG_ZERO);
+                           M_NOP;
+                           disp = (s4) (cd->mcodeptr - cd->mcodebase);
+                           /* REG_RA holds the value of the jmp instruction, therefore +8 */
+                           M_LDA(REG_ZERO, REG_RA_CALLER, -disp + 8);
+                               break;
                        }
 
                        /* store return value */
@@ -2540,7 +2645,6 @@ gen_method:
 
 
                case ICMD_CHECKCAST:  /* ..., objectref ==> ..., objectref            */
-               /* XXX needs manual attention! */
                                      /* val.a: (classinfo*) superclass               */
 
                        /*  superclass is an interface:
@@ -2558,146 +2662,111 @@ gen_method:
 
                        if (!(iptr->flags.bits & INS_FLAG_ARRAY)) {
                                classinfo *super;
-                               vftbl_t   *supervftbl;
                                s4         superindex;
 
-                               super = iptr->sx.s23.s3.c.cls;
-
-                               if (super == NULL) {
+                               if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
+                                       super      = NULL;
                                        superindex = 0;
-                                       supervftbl = NULL;
                                }
                                else {
+                                       super = iptr->sx.s23.s3.c.cls;
                                        superindex = super->index;
-                                       supervftbl = super->vftbl;
                                }
 
-#if defined(ENABLE_THREADS)
-                               codegen_threadcritrestart(cd, cd->mcodeptr - cd->mcodebase);
-#endif
+                               if ((super == NULL) || !(super->flags & ACC_INTERFACE))
+                                       CODEGEN_CRITICAL_SECTION_NEW;
 
                                s1 = emit_load_s1(jd, iptr, REG_ITMP1);
 
-                               /* calculate interface checkcast code size */
-
-                               s2 = 8;
-                               if (super == NULL)
-                                       s2 += (opt_showdisassemble ? 2 : 0);
-
-                               /* calculate class checkcast code size */
-
-                               s3 = 10 /* 10 + (s1 == REG_ITMP1) */;
-                               if (super == NULL)
-                                       s3 += (opt_showdisassemble ? 2 : 0);
-
                                /* if class is not resolved, check which code to call */
 
                                if (super == NULL) {
-                                       M_BEQZ(s1, 5 + (opt_showdisassemble ? 2 : 0) + s2 + 2 + s3);
-                                       M_NOP;
+                                       emit_label_beqz(cd, BRANCH_LABEL_1, s1);
 
+                                       cr   = iptr->sx.s23.s3.c.ref;
                                        disp = dseg_add_unique_s4(cd, 0);         /* super->flags */
 
-                                       codegen_addpatchref(cd, PATCHER_checkcast_instanceof_flags,
-                                                                               iptr->sx.s23.s3.c.ref,
-                                                                               disp);
-
-                                       if (opt_showdisassemble) {
-                                               M_NOP; M_NOP;
-                                       }
+                                       codegen_add_patch_ref(cd, PATCHER_checkcast_instanceof_flags,
+                                                                                 cr, disp);
 
                                        M_ILD(REG_ITMP2, REG_PV, disp);
                                        M_AND_IMM(REG_ITMP2, ACC_INTERFACE, REG_ITMP2);
-                                       M_BEQZ(REG_ITMP2, 1 + s2 + 2);
-                                       M_NOP;
+                                       emit_label_beqz(cd, BRANCH_LABEL_2, REG_ITMP2);
                                }
 
                                /* interface checkcast code */
 
                                if ((super == NULL) || (super->flags & ACC_INTERFACE)) {
                                        if (super == NULL) {
-                                               codegen_addpatchref(cd,
-                                                                                       PATCHER_checkcast_instanceof_interface,
-                                                                                       iptr->sx.s23.s3.c.ref,
-                                                                                       0);
+                                               cr = iptr->sx.s23.s3.c.ref;
 
-                                               if (opt_showdisassemble) {
-                                                       M_NOP; M_NOP;
-                                               }
+                                               codegen_add_patch_ref(cd, PATCHER_checkcast_interface,
+                                                                                         cr, 0);
                                        }
                                        else {
-                                               M_BEQZ(s1, 1 + s2);
-                                               M_NOP;
+                                               emit_label_beqz(cd, BRANCH_LABEL_3, s1);
                                        }
 
-                                       M_ALD(REG_ITMP2, s1, OFFSET(java_objectheader, vftbl));
-                                       M_ILD(REG_ITMP3, REG_ITMP2, OFFSET(vftbl_t, interfacetablelength));
-                                       M_LDA(REG_ITMP3, REG_ITMP3, -superindex);
-                                       M_BLEZ(REG_ITMP3, 0);
-                                       codegen_add_classcastexception_ref(cd, s1);
-                                       M_NOP;
+                                       M_ALD(REG_ITMP2, s1, OFFSET(java_object_t, vftbl));
+                                       M_ILD(REG_ITMP3, REG_ITMP2,
+                                                       OFFSET(vftbl_t, interfacetablelength));
+                                       M_ADD_IMM(REG_ITMP3, -superindex, REG_ITMP3);
+                                       emit_classcast_check(cd, iptr, ICMD_IFLE, REG_ITMP3, s1);
+
                                        M_ALD(REG_ITMP3, REG_ITMP2,
-                                                 (s4) (OFFSET(vftbl_t, interfacetable[0]) -
-                                                               superindex * sizeof(methodptr*)));
-                                       M_BEQZ(REG_ITMP3, 0);
-                                       codegen_add_classcastexception_ref(cd, s1);
-                                       M_NOP;
+                                                 OFFSET(vftbl_t, interfacetable[0]) -
+                                                 superindex * sizeof(methodptr*));
+                                       emit_classcast_check(cd, iptr, ICMD_IFEQ, REG_ITMP3, s1);
 
-                                       if (super == NULL) {
-                                               M_BR(1 + s3);
-                                               M_NOP;
-                                       }
+                                       if (super == NULL)
+                                               emit_label_br(cd, BRANCH_LABEL_4);
+                                       else
+                                               emit_label(cd, BRANCH_LABEL_3);
                                }
 
                                /* class checkcast code */
 
                                if ((super == NULL) || !(super->flags & ACC_INTERFACE)) {
                                        if (super == NULL) {
+                                               emit_label(cd, BRANCH_LABEL_2);
+
+                                               cr   = iptr->sx.s23.s3.c.ref;
                                                disp = dseg_add_unique_address(cd, NULL);
 
-                                               codegen_addpatchref(cd,
+                                               codegen_add_patch_ref(cd,
                                                                                        PATCHER_checkcast_instanceof_class,
-                                                                                       iptr->sx.s23.s3.c.ref,
-                                                                                       disp);
-
-                                               if (opt_showdisassemble) {
-                                                       M_NOP; M_NOP;
-                                               }
+                                                                                         cr, disp);
                                        }
                                        else {
-                                               disp = dseg_add_address(cd, supervftbl);
+                                               disp = dseg_add_address(cd, super->vftbl);
 
-                                               M_BEQZ(s1, 1 + s3);
-                                               M_NOP;
+                                               emit_label_beqz(cd, BRANCH_LABEL_5, s1);
                                        }
 
-                                       M_ALD(REG_ITMP2, s1, OFFSET(java_objectheader, vftbl));
+                                       M_ALD(REG_ITMP2, s1, OFFSET(java_object_t, vftbl));
                                        M_ALD(REG_ITMP3, REG_PV, disp);
-#if defined(ENABLE_THREADS)
-                                       codegen_threadcritstart(cd, cd->mcodeptr - cd->mcodebase);
-#endif
+                                       
+                                       CODEGEN_CRITICAL_SECTION_START;
+
                                        M_ILD(REG_ITMP2, REG_ITMP2, OFFSET(vftbl_t, baseval));
-                                       /*                              if (s1 != REG_ITMP1) { */
-                                       /*                                      M_ILD(REG_ITMP1, REG_ITMP3, OFFSET(vftbl_t, baseval)); */
-                                       /*                                      M_ILD(REG_ITMP3, REG_ITMP3, OFFSET(vftbl_t, diffval)); */
-                                       /*  #if defined(ENABLE_THREADS) */
-                                       /*                                      codegen_threadcritstop(cd, (u1 *) mcodeptr - cd->mcodebase); */
-                                       /*  #endif */
-                                       /*                                      M_ISUB(REG_ITMP2, REG_ITMP1, REG_ITMP2); */
-
-                                       /*                              } else { */
                                        M_ILD(REG_ITMP3, REG_ITMP3, OFFSET(vftbl_t, baseval));
                                        M_SUB(REG_ITMP2, REG_ITMP3, REG_ITMP2);
                                        M_ALD(REG_ITMP3, REG_PV, disp);
                                        M_ILD(REG_ITMP3, REG_ITMP3, OFFSET(vftbl_t, diffval));
-#if defined(ENABLE_THREADS)
-                                       codegen_threadcritstop(cd, cd->mcodeptr - cd->mcodebase);
-#endif
+
+                                       CODEGEN_CRITICAL_SECTION_END;
+
                                        /*                              } */
                                        M_CMP(REG_ITMP3, REG_ITMP2);
-                                       M_BULT(0);                         /* branch if ITMP3 < ITMP2 */ 
-                                       codegen_add_classcastexception_ref(cd, s1);
-                                       M_NOP;
+                                       emit_classcast_check(cd, iptr, BRANCH_ULT, REG_ITMP3, s1);
+
+                                       if (super != NULL)
+                                               emit_label(cd, BRANCH_LABEL_5);
+                               }
+
+                               if (super == NULL) {
+                                       emit_label(cd, BRANCH_LABEL_1);
+                                       emit_label(cd, BRANCH_LABEL_4);
                                }
 
                                d = codegen_reg_of_dst(jd, iptr, s1);
@@ -2705,31 +2774,30 @@ gen_method:
                        else {
                                /* array type cast-check */
 
-                               s1 = emit_load_s1(jd, iptr, rd->argintregs[0]);
-                               M_INTMOVE(s1, rd->argintregs[0]);
+                               s1 = emit_load_s1(jd, iptr, REG_OUT0);
+                               M_INTMOVE(s1, REG_OUT0);
 
                                disp = dseg_add_address(cd, iptr->sx.s23.s3.c.cls);
 
                                if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
-                                       codegen_addpatchref(cd, PATCHER_builtin_arraycheckcast,
-                                                                               iptr->sx.s23.s3.c.ref,
-                                                                               disp);
+                                       cr   = iptr->sx.s23.s3.c.ref;
+                                       disp = dseg_add_unique_address(cd, NULL);
 
-                                       if (opt_showdisassemble) {
-                                               M_NOP; M_NOP;
-                                       }
+                                       codegen_add_patch_ref(cd, PATCHER_builtin_arraycheckcast,
+                                                                                 cr, disp);
                                }
+                               else
+                                       disp = dseg_add_address(cd, iptr->sx.s23.s3.c.cls);
 
-                               M_ALD(rd->argintregs[1], REG_PV, disp);
+                               M_ALD(REG_OUT1, REG_PV, disp);
                                disp = dseg_add_functionptr(cd, BUILTIN_arraycheckcast);
                                M_ALD(REG_ITMP3, REG_PV, disp);
+                               /* XXX jit-c-call */
                                M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
                                M_NOP;
 
                                s1 = emit_load_s1(jd, iptr, REG_ITMP1);
-                               M_BEQZ(REG_RESULT_CALLER, 0);
-                               codegen_add_classcastexception_ref(cd, s1);
-                               M_NOP;
+                               emit_classcast_check(cd, iptr, ICMD_IFEQ, REG_RESULT_CALLER, s1);
 
                                d = codegen_reg_of_dst(jd, iptr, s1);
                        }
@@ -2738,30 +2806,230 @@ gen_method:
                        emit_store_dst(jd, iptr, d);
                        break;
 
+               case ICMD_INSTANCEOF: /* ..., objectref ==> ..., intresult            */
+                                     /* val.a: (classinfo*) superclass               */
+
+                       /*  superclass is an interface:
+                        *      
+                        *  return (sub != NULL) &&
+                        *         (sub->vftbl->interfacetablelength > super->index) &&
+                        *         (sub->vftbl->interfacetable[-super->index] != NULL);
+                        *      
+                        *  superclass is a class:
+                        *      
+                        *  return ((sub != NULL) && (0
+                        *          <= (sub->vftbl->baseval - super->vftbl->baseval) <=
+                        *          super->vftbl->diffvall));
+                        */
+
+                       {
+                       classinfo *super;
+                       vftbl_t   *supervftbl;
+                       s4         superindex;
+
+                       if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
+                               super = NULL;
+                               superindex = 0;
+                               supervftbl = NULL;
+
+                       } else {
+                               super = iptr->sx.s23.s3.c.cls;
+                               superindex = super->index;
+                               supervftbl = super->vftbl;
+                       }
+
+                       if ((super == NULL) || !(super->flags & ACC_INTERFACE))
+                               CODEGEN_CRITICAL_SECTION_NEW;
+
+                       s1 = emit_load_s1(jd, iptr, REG_ITMP1);
+                       d = codegen_reg_of_dst(jd, iptr, REG_ITMP2);
+                       if (s1 == d) {
+                               M_MOV(s1, REG_ITMP1);
+                               s1 = REG_ITMP1;
+                       }
+
+                       M_CLR(d);
+
+                       /* if class is not resolved, check which code to call */
+
+                       if (super == NULL) {
+                               emit_label_beqz(cd, BRANCH_LABEL_1, s1);
+
+                               cr   = iptr->sx.s23.s3.c.ref;
+                               disp = dseg_add_unique_s4(cd, 0);             /* super->flags */
+
+                               codegen_add_patch_ref(cd, PATCHER_checkcast_instanceof_flags,
+                                                                         cr, disp);
+
+                               M_ILD(REG_ITMP3, REG_PV, disp);
+                               M_AND_IMM(REG_ITMP3, ACC_INTERFACE, REG_ITMP3);
+                               emit_label_beqz(cd, BRANCH_LABEL_2, REG_ITMP3);
+                       }
+
+                       /* interface instanceof code */
+
+                       if ((super == NULL) || (super->flags & ACC_INTERFACE)) {
+                               if (super == NULL) {
+                                       cr = iptr->sx.s23.s3.c.ref;
+
+                                       codegen_add_patch_ref(cd, PATCHER_instanceof_interface,
+                                                                                 cr, 0);
+                               }
+                               else {
+                                       emit_label_beqz(cd, BRANCH_LABEL_3, s1);
+                               }
+
+                               M_ALD(REG_ITMP1, s1, OFFSET(java_object_t, vftbl));
+                               M_ILD(REG_ITMP3, REG_ITMP1, OFFSET(vftbl_t, interfacetablelength));
+                               M_CMP_IMM(REG_ITMP3, superindex);
+                               M_BLE(4);
+                               M_NOP;
+                               M_ALD(REG_ITMP1, REG_ITMP1,
+                                         (s4) (OFFSET(vftbl_t, interfacetable[0]) -
+                                                       superindex * sizeof(methodptr*)));
+                               M_CMOVRNE_IMM(REG_ITMP1, 1, d);      /* REG_ITMP1 != 0  */
+
+                               if (super == NULL)
+                                       emit_label_br(cd, BRANCH_LABEL_4);
+                               else
+                                       emit_label(cd, BRANCH_LABEL_3);
+                       }
+
+                       /* class instanceof code */
+
+                       if ((super == NULL) || !(super->flags & ACC_INTERFACE)) {
+                               if (super == NULL) {
+                                       emit_label(cd, BRANCH_LABEL_2);
+
+                                       cr   = iptr->sx.s23.s3.c.ref;
+                                       disp = dseg_add_unique_address(cd, NULL);
+
+                                       codegen_add_patch_ref(cd, PATCHER_checkcast_instanceof_class,
+                                                                                 cr, disp);
+                               }
+                               else {
+                                       disp = dseg_add_address(cd, supervftbl);
+
+                                       emit_label_beqz(cd, BRANCH_LABEL_5, s1);
+                               }
+
+                               M_ALD(REG_ITMP1, s1, OFFSET(java_object_t, vftbl));
+                               M_ALD(REG_ITMP2, REG_PV, disp);
+
+                               CODEGEN_CRITICAL_SECTION_START;
+
+                               M_ILD(REG_ITMP1, REG_ITMP1, OFFSET(vftbl_t, baseval));
+                               M_ILD(REG_ITMP3, REG_ITMP2, OFFSET(vftbl_t, baseval));
+                               M_ILD(REG_ITMP2, REG_ITMP2, OFFSET(vftbl_t, diffval));
+
+                               CODEGEN_CRITICAL_SECTION_END;
+
+                               M_SUB(REG_ITMP1, REG_ITMP3, REG_ITMP1);
+                               M_CMP(REG_ITMP1, REG_ITMP2);
+                               M_XCMOVULE_IMM(1, d);
+
+                               if (super != NULL)
+                                       emit_label(cd, BRANCH_LABEL_5);
+                       }
+
+                       if (super == NULL) {
+                               emit_label(cd, BRANCH_LABEL_1);
+                               emit_label(cd, BRANCH_LABEL_4);
+                       }
+
+                       emit_store_dst(jd, iptr, d);
+                       }
+                       break;
+
+               case ICMD_MULTIANEWARRAY:/* ..., cnt1, [cnt2, ...] ==> ..., arrayref  */
+
+                       /* check for negative sizes and copy sizes to stack if necessary  */
+
+                       MCODECHECK((iptr->s1.argcount << 1) + 64);
+
+                       for (s1 = iptr->s1.argcount; --s1 >= 0; ) {
+
+                               var = VAR(iptr->sx.s23.s2.args[s1]);
+       
+                               /* copy SAVEDVAR sizes to stack */
+
+                               /* Already Preallocated? */
+
+                               if (!(var->flags & PREALLOC)) {
+                                       s2 = emit_load(jd, iptr, var, REG_ITMP1);
+                                       M_STX(s2, REG_SP, CSTACK + (s1 * 8));
+                               }
+                       }
+
+                       /* arg 0 = dimension count */
+
+                       ICONST(REG_OUT0, iptr->s1.argcount);
+
+                       /* is patcher function set? */
+
+                       if (INSTRUCTION_IS_UNRESOLVED(iptr)) {
+                               disp = dseg_add_unique_address(cd, 0);
+
+                               codegen_add_patch_ref(cd, PATCHER_builtin_multianewarray,
+                                                                         iptr->sx.s23.s3.c.ref,
+                                                                         disp);
+                       }
+                       else
+                               disp = dseg_add_address(cd, iptr->sx.s23.s3.c.cls);
+
+                       /* arg 1 = arraydescriptor */
+
+                       M_ALD(REG_OUT1, REG_PV, disp);
 
+                       /* arg 2 = pointer to dimensions = stack pointer (absolute) */
+
+                       M_ADD_IMM(REG_SP, CSTACK, REG_OUT2);
+
+                       /* XXX c abi call */
+                       disp = dseg_add_functionptr(cd, BUILTIN_multianewarray);
+                       M_ALD(REG_ITMP3, REG_PV, disp);
+                       M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
+                       M_NOP;
+
+                       /* check for exception before result assignment */
+
+                       emit_exception_check(cd, iptr);
+
+                       d = codegen_reg_of_dst(jd, iptr, REG_RESULT_CALLER);
+                       M_INTMOVE(REG_RESULT_CALLER, d);
+                       emit_store_dst(jd, iptr, d);
+                       break;
 
                default:
-                       *exceptionptr = new_internalerror("Unknown ICMD %d", iptr->opc);
+                       exceptions_throw_internalerror("Unknown ICMD %d during code generation",
+                                                                                  iptr->opc);
                        return false;
                        
        } /* switch */
                
        } /* for instruction */
+
+       MCODECHECK(64);
        
+       /* At the end of a basic block we may have to append some nops,
+          because the patcher stub calling code might be longer than the
+          actual instruction. So codepatching does not change the
+          following block unintentionally. */
 
+       if (cd->mcodeptr < cd->lastmcodeptr) {
+               while (cd->mcodeptr < cd->lastmcodeptr) {
+                       M_NOP;
+               }
+       }
                
        } /* if (bptr -> flags >= BBREACHED) */
        } /* for basic block */
        
        dseg_createlinenumbertable(cd);
 
-       /* generate exception and patcher stubs */
+       /* generate stubs */
 
-       emit_exception_stubs(jd);
        emit_patcher_stubs(jd);
-       emit_replacement_stubs(jd);
-
-       codegen_finish(jd);
        
        /* everything's ok */
 
@@ -2769,102 +3037,56 @@ gen_method:
 }
 
 
+/* codegen_emit_stub_compiler **************************************************
 
-
-
-/* createcompilerstub **********************************************************
-
-   Creates a stub routine which calls the compiler.
+   Emits a stub routine which calls the compiler.
        
 *******************************************************************************/
 
-#define COMPILERSTUB_DATASIZE    3 * SIZEOF_VOID_P
-#define COMPILERSTUB_CODESIZE    4 * 4
-
-#define COMPILERSTUB_SIZE        COMPILERSTUB_DATASIZE + COMPILERSTUB_CODESIZE
-
-
-u1 *createcompilerstub(methodinfo *m)
+void codegen_emit_stub_compiler(jitdata *jd)
 {
-       u1     *s;                          /* memory to hold the stub            */
-       ptrint      *d;
-       codeinfo    *code;
+       methodinfo  *m;
        codegendata *cd;
-       s4           dumpsize;
-       
-       s = CNEW(u1, COMPILERSTUB_SIZE);
-
-       /* set data pointer and code pointer */
-
-       d = (ptrint *) s;
-       s = s + COMPILERSTUB_DATASIZE;
 
-       /* mark start of dump memory area */
-
-       dumpsize = dump_size();
-
-       cd = DNEW(codegendata);
-       cd->mcodeptr = s;
-       
-       /* Store the codeinfo pointer in the same place as in the
-          methodheader for compiled methods. */
-
-       code = code_codeinfo_new(m);
+       /* get required compiler data */
 
-       d[0] = (ptrint) asm_call_jit_compiler;
-       d[1] = (ptrint) m;
-       d[2] = (ptrint) code;
+       m  = jd->m;
+       cd = jd->cd;
 
        /* code for the stub */
+
        /* no window save yet, user caller's PV */
        M_ALD_INTERN(REG_ITMP1, REG_PV_CALLER, -2 * SIZEOF_VOID_P);  /* codeinfo pointer */
        M_ALD_INTERN(REG_PV_CALLER, REG_PV_CALLER, -3 * SIZEOF_VOID_P);  /* pointer to compiler */
        M_JMP(REG_ZERO, REG_PV_CALLER, REG_ZERO);  /* jump to the compiler, RA is wasted */
        M_NOP;
-
-#if defined(ENABLE_STATISTICS)
-       if (opt_stat)
-               count_cstub_len += COMPILERSTUB_SIZE;
-#endif
-
-       /* release dump area */
-
-       dump_release(dumpsize);
-
-       return s;
 }
 
 
+/* codegen_emit_stub_native ****************************************************
 
-/* createnativestub ************************************************************
-
-   Creates a stub routine which calls a native method.
+   Emits a stub routine which calls a native method.
 
 *******************************************************************************/
 
-u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
+void codegen_emit_stub_native(jitdata *jd, methoddesc *nmd, functionptr f)
 {
        methodinfo   *m;
        codeinfo     *code;
        codegendata  *cd;
-       registerdata *rd;
        methoddesc   *md;
        s4            nativeparams;
        s4            i, j;                 /* count variables                    */
        s4            t;
        s4            s1, s2, disp;
        s4            funcdisp;             /* displacement of the function       */
+       s4            fltregarg_offset[FLT_ARG_CNT];
 
        /* get required compiler data */
 
        m    = jd->m;
        code = jd->code;
        cd   = jd->cd;
-       rd   = jd->rd;
-
-       /* rewrite registers and params */
-       md_native_reg_setup(jd);
-       md_native_param_alloc(nmd);
 
        /* initialize variables */
 
@@ -2877,9 +3099,15 @@ u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
                sizeof(stackframeinfo) / SIZEOF_VOID_P +
                sizeof(localref_table) / SIZEOF_VOID_P +
                md->paramcount +                /* for saving arguments over calls    */
-               nmd->memuse +  /* nmd knows about the native stackframe layout */
+               nmd->memuse +              /* nmd->memuse includes the (6) abi params */
                WINSAVE_CNT;
 
+
+       /* keep stack 16-byte aligned (ABI requirement) */
+
+       if (cd->stackframesize & 1)
+               cd->stackframesize++;
+
        /* create method header */
 
        (void) dseg_add_unique_address(cd, code);              /* CodeinfoPointer */
@@ -2906,26 +3134,26 @@ u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
 
 #if !defined(WITH_STATIC_CLASSPATH)
        if (f == NULL) {
-               codegen_addpatchref(cd, PATCHER_resolve_native, m, funcdisp);
-
-               if (opt_showdisassemble) {
-                       M_NOP; M_NOP;
-               }
+               codegen_add_patch_ref(cd, PATCHER_resolve_native, m, funcdisp);
        }
 #endif
 
        /* save float argument registers */
 
+       assert(ABIPARAMS_CNT >= FLT_ARG_CNT);
+
        for (i = 0, j = 0; i < md->paramcount && i < FLT_ARG_CNT; i++) {
                if (IS_FLT_DBL_TYPE(md->paramtypes[i].type)) {
-                       M_DST(rd->argfltregs[i], REG_SP, j * 8);
+                       s1 = WINSAVE_CNT + nmd->memuse + j;
+                       M_DST(abi_registers_float_argument[i], REG_SP, BIAS + (s1*8));
+                       fltregarg_offset[i] = s1; /* remember stack offset */
                        j++;
                }
        }
 
        /* prepare data structures for native function call */
 
-       M_ADD_IMM(REG_FP, BIAS, REG_OUT0); /* datasp == top of the stack frame (absolute == +BIAS) */
+       M_ADD_IMM(REG_FP, BIAS, REG_OUT0); /* datasp == top of the stack frame (absolute, ie. + BIAS) */
        M_MOV(REG_PV_CALLEE, REG_OUT1);
        M_MOV(REG_FP, REG_OUT2); /* java sp */
        M_MOV(REG_RA_CALLEE, REG_OUT3);
@@ -2934,85 +3162,108 @@ u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
        M_JMP(REG_RA_CALLER, REG_ITMP3, REG_ZERO);
        M_NOP; /* XXX fill me! */
 
-       /* restore float argument registers */
-
+       /* keep float arguments on stack */
+#if 0
        for (i = 0, j = 0; i < md->paramcount && i < FLT_ARG_CNT; i++) {
                if (IS_FLT_DBL_TYPE(md->paramtypes[i].type)) {
-                       M_DLD(rd->argfltregs[i], REG_SP, j * 8);
+                       M_DLD(abi_registers_float_argument[i], REG_SP, CSTACK + (j * 8));
                        j++;
                }
        }
+#endif
 
        /* copy or spill arguments to new locations */
-       int num_fltregargs = 0;
-       int fltregarg_inswap[16];
+
        for (i = md->paramcount - 1, j = i + nativeparams; i >= 0; i--, j--) {
                t = md->paramtypes[i].type;
 
                if (IS_INT_LNG_TYPE(t)) {
+
+                       /* integral types */
+
                        if (!md->params[i].inmemory) {
-                               s1 = rd->argintregs[md->params[i].regoff];
+                               s1 = md->params[i].regoff;
                                /* s1 refers to the old window, transpose */
                                s1 = REG_WINDOW_TRANSPOSE(s1);
 
                                if (!nmd->params[j].inmemory) {
-                                       s2 = nat_argintregs[nmd->params[j].regoff];
+                                       s2 = nmd->params[j].regoff;
                                        M_INTMOVE(s1, s2);
                                } else {
-                                       s2 = nmd->params[j].regoff;
-                                       M_AST(s1, REG_SP, USESTACK_PARAMS + s2 * 8);
+                                       /* nmd's regoff is relative to the start of the param array */
+                                       s2 = BIAS + WINSAVE_CNT * 8 + nmd->params[j].regoff;
+                                       M_AST(s1, REG_SP, s2);
                                }
 
                        } else {
-                               s1 = md->params[i].regoff + cd->stackframesize;
-                               s2 = nmd->params[j].regoff;
-                               M_ALD(REG_ITMP1, REG_SP, USESTACK_PARAMS + s1 * 8);
-                               M_AST(REG_ITMP1, REG_SP, USESTACK_PARAMS + s2 * 8);
+                               if (!nmd->params[j].inmemory) {
+                                       /* JIT stack arg -> NAT reg arg */
+
+                                       /* Due to the Env pointer that is always passed, the 6th JIT arg   */
+                                       /* is the 7th (or 8th w/ class ptr) NAT arg, and goes to the stack */
+
+                                       assert(false); /* path never taken */
+                               }
+
+                               s1 = md->params[i].regoff + cd->stackframesize * 8;
+                               s2 = BIAS + WINSAVE_CNT * 8 + nmd->params[j].regoff;
+                               M_ALD(REG_ITMP1, REG_SP, CSTACK + s1);
+                               M_AST(REG_ITMP1, REG_SP, s2);
                        }
 
                } else {
+
+                       /* floating point types */
+
                        if (!md->params[i].inmemory) {
-                               s1 = rd->argfltregs[md->params[i].regoff];
+                               s1 = md->params[i].regoff;
 
                                if (!nmd->params[j].inmemory) {
+
                                        /* no mapping to regs needed, native flt args use regoff */
                                        s2 = nmd->params[j].regoff;
-                                       
-                                       /* we cannot move flt regs to their native arg locations directly */
-                                       M_DMOV(s1, s2 + 16);
-                                       fltregarg_inswap[num_fltregargs] = s2;
-                                       num_fltregargs++;
-                                       printf("flt arg swap to %d\n", s2 + 16);
 
-                               } else {
+                                       /* JIT float regs are still on the stack */
+                                       M_DLD(s2, REG_SP, BIAS + (fltregarg_offset[i] * 8));
+                               } 
+                               else {
+                                       /* not supposed to happen with 16 NAT flt args */
+                                       assert(false); 
+                                       /*
                                        s2 = nmd->params[j].regoff;
                                        if (IS_2_WORD_TYPE(t))
-                                               M_DST(s1, REG_SP, USESTACK_PARAMS + s2 * 8);
+                                               M_DST(s1, REG_SP, CSTACK + (s2 * 8));
                                        else
-                                               M_FST(s1, REG_SP, USESTACK_PARAMS + s2 * 8);
+                                               M_FST(s1, REG_SP, CSTACK + (s2 * 8));
+                                       */
                                }
 
-                       } else {
-                               s1 = md->params[i].regoff + cd->stackframesize;
-                               s2 = nmd->params[j].regoff;
-                               if (IS_2_WORD_TYPE(t)) {
-                                       M_DLD(REG_FTMP1, REG_SP, USESTACK_PARAMS + s1 * 8);
-                                       M_DST(REG_FTMP1, REG_SP, USESTACK_PARAMS + s2 * 8);
-                               } else {
-                                       M_FLD(REG_FTMP1, REG_SP, USESTACK_PARAMS + s1 * 8);
-                                       M_FST(REG_FTMP1, REG_SP, USESTACK_PARAMS + s2 * 8);
+                       } 
+                       else {
+                               s1 = md->params[i].regoff;
+
+                               if (!nmd->params[j].inmemory) {
+
+                                       /* JIT stack -> NAT reg */
+
+                                       s2 = nmd->params[j].regoff;
+                                       M_DLD(s2, REG_FP, JITSTACK + s1);
+                               }
+                               else {
+
+                                       /* JIT stack -> NAT stack */
+
+                                       s2 = WINSAVE_CNT * 8 + nmd->params[j].regoff;
+
+                                       /* The FTMP register may already be loaded with args */
+                                       /* we know $f0 is unused because of the env pointer  */
+                                       M_DLD(REG_F0, REG_FP, JITSTACK + s1);
+                                       M_DST(REG_F0, REG_SP, BIAS + s2);
                                }
                        }
                }
        }
        
-       /* move swapped float args to target regs */
-       for (i = 0; i < num_fltregargs; i++) {
-               s1 = fltregarg_inswap[i];
-               M_DMOV(s1 + 16, s1);
-               printf("float arg to target reg: %d ==> %d\n", s1+16, s1);
-       }
-
 
        /* put class into second argument register */
 
@@ -3038,7 +3289,7 @@ u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
                if (IS_INT_LNG_TYPE(md->returntype.type))
                        M_MOV(REG_RESULT_CALLER, REG_RESULT_CALLEE);
                else
-                       M_DST(REG_FRESULT, REG_SP, USESTACK_PARAMS);
+                       M_DST(REG_FRESULT, REG_SP, CSTACK);
        }
        
        /* Note: native functions return float values in %f0 (see ABI) */
@@ -3048,7 +3299,7 @@ u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
        /* But for the trace function we need to put a flt result into %f1 */
        if (JITDATA_HAS_FLAG_VERBOSECALL(jd)) {
                if (!IS_2_WORD_TYPE(md->returntype.type))
-                       M_FLD(REG_FRESULT, REG_SP, USESTACK_PARAMS);
+                       M_FLD(REG_FRESULT, REG_SP, CSTACK);
                emit_verbosecall_exit(jd);
        }
 #endif
@@ -3067,34 +3318,30 @@ u1 *createnativestub(functionptr f, jitdata *jd, methoddesc *nmd)
        if (md->returntype.type != TYPE_VOID) {
                if (IS_FLT_DBL_TYPE(md->returntype.type)) {
                        if (IS_2_WORD_TYPE(md->returntype.type))
-                               M_DLD(REG_FRESULT, REG_SP, USESTACK_PARAMS);
+                               M_DLD(REG_FRESULT, REG_SP, CSTACK);
                        else
-                               M_FLD(REG_FRESULT, REG_SP, USESTACK_PARAMS);
+                               M_FLD(REG_FRESULT, REG_SP, CSTACK);
                }
        }
 
        /* check for exception */
-
        M_BNEZ(REG_ITMP2_XPTR, 4);          /* if no exception then return        */
-       M_RESTORE(REG_ZERO, 0, REG_ZERO);   /* restore callers window (DELAY)     */
+       M_NOP;
 
-       M_RET(REG_RA_CALLER, 8);            /* return to caller                   */
-       M_NOP;                              /* DELAY SLOT                         */
+       M_RETURN(REG_RA_CALLEE, 8); /* implicit window restore */
+       M_NOP;
 
        /* handle exception */
        
        disp = dseg_add_functionptr(cd, asm_handle_nat_exception);
-       M_ALD(REG_ITMP3, REG_PV, disp);     /* load asm exception handler address */
-       M_JMP(REG_ZERO, REG_ITMP3, REG_ZERO);/* jump to asm exception handler     */
-       M_MOV(REG_RA_CALLER, REG_ITMP3_XPC); /* get exception address (DELAY)    */
-
+       M_ALD(REG_ITMP1, REG_PV, disp);     /* load asm exception handler address */
+       M_MOV(REG_RA_CALLEE, REG_ITMP3_XPC); /* get exception address             */
+       M_JMP(REG_ZERO, REG_ITMP1, REG_ZERO);/* jump to asm exception handler     */
+       M_RESTORE(REG_ZERO, 0, REG_ZERO);   /* restore callers window (DELAY)     */
+       
        /* generate patcher stubs */
 
        emit_patcher_stubs(jd);
-
-       codegen_finish(jd);
-
-       return code->entrypoint;
 }
 
 /*