* Exception.cs: Cosmetic change to default exception message on
[mono.git] / mono / mini / mini-arm.c
index 7a100ad9d92c19c0e8c6842fdb03d03af6906899..3a194e370f1cbb643951cae50a75b0e2a12487aa 100644 (file)
 #include "inssel.h"
 #include "cpu-arm.h"
 #include "trace.h"
+#ifdef ARM_FPU_FPA
 #include "mono/arch/arm/arm-fpa-codegen.h"
+#elif defined(ARM_FPU_VFP)
+#include "mono/arch/arm/arm-vfp-codegen.h"
+#endif
+
+/* This mutex protects architecture specific caches */
+#define mono_mini_arch_lock() EnterCriticalSection (&mini_arch_mutex)
+#define mono_mini_arch_unlock() LeaveCriticalSection (&mini_arch_mutex)
+static CRITICAL_SECTION mini_arch_mutex;
+
+static int v5_supported = 0;
+static int thumb_supported = 0;
+
+static int mono_arm_is_rotated_imm8 (guint32 val, gint *rot_amount);
 
 /*
  * TODO:
@@ -39,6 +53,14 @@ int mono_exc_esp_offset = 0;
 #define arm_is_imm8(v) ((v) > -256 && (v) < 256)
 #define arm_is_fpimm8(v) ((v) >= -1020 && (v) <= 1020)
 
+#define LDR_MASK ((0xf << ARMCOND_SHIFT) | (3 << 26) | (1 << 22) | (1 << 20) | (15 << 12))
+#define LDR_PC_VAL ((ARMCOND_AL << ARMCOND_SHIFT) | (1 << 26) | (0 << 22) | (1 << 20) | (15 << 12))
+#define IS_LDR_PC(val) (((val) & LDR_MASK) == LDR_PC_VAL)
+
+#define ADD_LR_PC_4 ((ARMCOND_AL << ARMCOND_SHIFT) | (1 << 25) | (1 << 23) | (ARMREG_PC << 16) | (ARMREG_LR << 12) | 4)
+#define MOV_LR_PC ((ARMCOND_AL << ARMCOND_SHIFT) | (1 << 24) | (0xa << 20) |  (ARMREG_LR << 12) | ARMREG_PC)
+#define DEBUG_IMT 0
+
 const char*
 mono_arch_regname (int reg) {
        static const char * rnames[] = {
@@ -100,21 +122,61 @@ emit_memcpy (guint8 *code, int size, int dreg, int doffset, int sreg, int soffse
                arm_patch (code - 4, start_loop);
                return code;
        }
-       g_assert (arm_is_imm12 (doffset));
-       g_assert (arm_is_imm12 (doffset + size));
-       g_assert (arm_is_imm12 (soffset));
-       g_assert (arm_is_imm12 (soffset + size));
-       while (size >= 4) {
-               ARM_LDR_IMM (code, ARMREG_LR, sreg, soffset);
-               ARM_STR_IMM (code, ARMREG_LR, dreg, doffset);
-               doffset += 4;
-               soffset += 4;
-               size -= 4;
+       if (arm_is_imm12 (doffset) && arm_is_imm12 (doffset + size) &&
+                       arm_is_imm12 (soffset) && arm_is_imm12 (soffset + size)) {
+               while (size >= 4) {
+                       ARM_LDR_IMM (code, ARMREG_LR, sreg, soffset);
+                       ARM_STR_IMM (code, ARMREG_LR, dreg, doffset);
+                       doffset += 4;
+                       soffset += 4;
+                       size -= 4;
+               }
+       } else if (size) {
+               code = emit_big_add (code, ARMREG_R0, sreg, soffset);
+               code = emit_big_add (code, ARMREG_R1, dreg, doffset);
+               doffset = soffset = 0;
+               while (size >= 4) {
+                       ARM_LDR_IMM (code, ARMREG_LR, ARMREG_R0, soffset);
+                       ARM_STR_IMM (code, ARMREG_LR, ARMREG_R1, doffset);
+                       doffset += 4;
+                       soffset += 4;
+                       size -= 4;
+               }
        }
        g_assert (size == 0);
        return code;
 }
 
+static guint8*
+emit_call_reg (guint8 *code, int reg)
+{
+       if (v5_supported) {
+               ARM_BLX_REG (code, reg);
+       } else {
+               ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
+               if (thumb_supported)
+                       ARM_BX (code, reg);
+               else
+                       ARM_MOV_REG_REG (code, ARMREG_PC, reg);
+       }
+       return code;
+}
+
+static guint8*
+emit_call_seq (MonoCompile *cfg, guint8 *code)
+{
+       if (cfg->method->dynamic) {
+               ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
+               ARM_B (code, 0);
+               *(gpointer*)code = NULL;
+               code += 4;
+               code = emit_call_reg (code, ARMREG_IP);
+       } else {
+               ARM_BL (code, 0);
+       }
+       return code;
+}
+
 /*
  * mono_arch_get_argument_info:
  * @csig:  a method signature
@@ -152,7 +214,7 @@ mono_arch_get_argument_info (MonoMethodSignature *csig, int param_count, MonoJit
                if (csig->pinvoke)
                        size = mono_type_native_stack_size (csig->params [k], &align);
                else
-                       size = mono_type_stack_size (csig->params [k], &align);
+                       size = mini_type_stack_size (NULL, csig->params [k], &align);
 
                /* ignore alignment for now */
                align = 1;
@@ -174,6 +236,162 @@ mono_arch_get_argument_info (MonoMethodSignature *csig, int param_count, MonoJit
        return frame_size;
 }
 
+static gpointer*
+decode_vcall_slot_from_ldr (guint32 ldr, gpointer *regs)
+{
+       char *o = NULL;
+       int reg, offset = 0;
+       reg = (ldr >> 16 ) & 0xf;
+       offset = ldr & 0xfff;
+       if (((ldr >> 23) & 1) == 0) /*U bit, 0 means negative and 1 positive*/
+               offset = -offset;
+       /*g_print ("found vcall at r%d + %d for code at %p 0x%x\n", reg, offset, code, *code);*/
+       o = regs [reg];
+       return (gpointer*)(o + offset);
+}
+
+gpointer*
+mono_arch_get_vcall_slot_addr (guint8 *code_ptr, gpointer *regs)
+{
+       guint32* code = (guint32*)code_ptr;
+
+       /* Locate the address of the method-specific trampoline. The call using
+       the vtable slot that took the processing flow to 'arch_create_jit_trampoline' 
+       looks something like this:
+
+               ldr rA, rX, #offset
+               mov lr, pc
+               mov pc, rA
+       or better:
+               mov lr, pc
+               ldr pc, rX, #offset
+
+       The call sequence could be also:
+               ldr ip, pc, 0
+               b skip
+               function pointer literal
+               skip:
+               mov lr, pc
+               mov pc, ip
+       Note that on ARM5+ we can use one instruction instead of the last two.
+       Therefore, we need to locate the 'ldr rA' instruction to know which
+       register was used to hold the method addrs.
+       */
+
+       /* This is the instruction after "ldc pc, xxx", "mov pc, xxx" or "bl xxx" could be either the IMT value or some other instruction*/
+       --code;
+
+       /* Three possible code sequences can happen here:
+        * interface call:
+        * 
+        * add lr, [pc + #4]
+        * ldr pc, [rX - #offset]
+        * .word IMT value
+        * 
+        * virtual call:
+        * 
+        * mov lr, pc
+        * ldr pc, [rX - #offset] 
+        * 
+        * direct branch with bl:
+        * 
+        * bl #offset
+        * 
+        * direct branch with mov: 
+        * 
+        * mv pc, rX
+        * 
+        * We only need to identify interface and virtual calls, the others can be ignored.
+        * 
+        */
+       if (IS_LDR_PC (code [-1]) && code [-2] == ADD_LR_PC_4)
+               return decode_vcall_slot_from_ldr (code [-1], regs);
+
+       if (IS_LDR_PC (code [0]) && code [-1] == MOV_LR_PC)
+               return decode_vcall_slot_from_ldr (code [0], regs);
+
+       return NULL;
+}
+
+#define MAX_ARCH_DELEGATE_PARAMS 3
+
+gpointer
+mono_arch_get_delegate_invoke_impl (MonoMethodSignature *sig, gboolean has_target)
+{
+       guint8 *code, *start;
+
+       /* FIXME: Support more cases */
+       if (MONO_TYPE_ISSTRUCT (sig->ret))
+               return NULL;
+
+       if (has_target) {
+               static guint8* cached = NULL;
+               mono_mini_arch_lock ();
+               if (cached) {
+                       mono_mini_arch_unlock ();
+                       return cached;
+               }
+               
+               start = code = mono_global_codeman_reserve (12);
+
+               /* Replace the this argument with the target */
+               ARM_LDR_IMM (code, ARMREG_IP, ARMREG_R0, G_STRUCT_OFFSET (MonoDelegate, method_ptr));
+               ARM_LDR_IMM (code, ARMREG_R0, ARMREG_R0, G_STRUCT_OFFSET (MonoDelegate, target));
+               ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
+
+               g_assert ((code - start) <= 12);
+
+               mono_arch_flush_icache (code, 12);
+               cached = start;
+               mono_mini_arch_unlock ();
+               return cached;
+       } else {
+               static guint8* cache [MAX_ARCH_DELEGATE_PARAMS + 1] = {NULL};
+               int size, i;
+
+               if (sig->param_count > MAX_ARCH_DELEGATE_PARAMS)
+                       return NULL;
+               for (i = 0; i < sig->param_count; ++i)
+                       if (!mono_is_regsize_var (sig->params [i]))
+                               return NULL;
+
+               mono_mini_arch_lock ();
+               code = cache [sig->param_count];
+               if (code) {
+                       mono_mini_arch_unlock ();
+                       return code;
+               }
+
+               size = 8 + sig->param_count * 4;
+               start = code = mono_global_codeman_reserve (size);
+
+               ARM_LDR_IMM (code, ARMREG_IP, ARMREG_R0, G_STRUCT_OFFSET (MonoDelegate, method_ptr));
+               /* slide down the arguments */
+               for (i = 0; i < sig->param_count; ++i) {
+                       ARM_MOV_REG_REG (code, (ARMREG_R0 + i), (ARMREG_R0 + i + 1));
+               }
+               ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
+
+               g_assert ((code - start) <= size);
+
+               mono_arch_flush_icache (code, size);
+               cache [sig->param_count] = start;
+               mono_mini_arch_unlock ();
+               return start;
+       }
+
+       return NULL;
+}
+
+gpointer
+mono_arch_get_this_arg_from_call (MonoMethodSignature *sig, gssize *regs, guint8 *code)
+{
+       /* FIXME: handle returning a struct */
+       if (MONO_TYPE_ISSTRUCT (sig->ret))
+               return (gpointer)regs [ARMREG_R1];
+       return (gpointer)regs [ARMREG_R0];
+}
+
 /*
  * Initialize the cpu to execute managed code.
  */
@@ -182,6 +400,23 @@ mono_arch_cpu_init (void)
 {
 }
 
+/*
+ * Initialize architecture specific code.
+ */
+void
+mono_arch_init (void)
+{
+       InitializeCriticalSection (&mini_arch_mutex);   
+}
+
+/*
+ * Cleanup architecture specific code.
+ */
+void
+mono_arch_cleanup (void)
+{
+}
+
 /*
  * This function returns the optimizations supported on this cpu.
  */
@@ -189,6 +424,31 @@ guint32
 mono_arch_cpu_optimizazions (guint32 *exclude_mask)
 {
        guint32 opts = 0;
+       char buf [512];
+       char *line;
+       FILE *file = fopen ("/proc/cpuinfo", "r");
+       if (file) {
+               while ((line = fgets (buf, 512, file))) {
+                       if (strncmp (line, "Processor", 9) == 0) {
+                               char *ver = strstr (line, "(v");
+                               if (ver && (ver [2] == '5' || ver [2] == '6' || ver [2] == '7')) {
+                                       v5_supported = TRUE;
+                               }
+                               continue;
+                       }
+                       if (strncmp (line, "Features", 8) == 0) {
+                               char *th = strstr (line, "thumb");
+                               if (th) {
+                                       thumb_supported = TRUE;
+                                       if (v5_supported)
+                                               break;
+                               }
+                               continue;
+                       }
+               }
+               fclose (file);
+               /*printf ("features: v5: %d, thumb: %d\n", v5_supported, thumb_supported);*/
+       }
 
        /* no arm-specific optimizations yet */
        *exclude_mask = 0;
@@ -214,6 +474,10 @@ is_regsize_var (MonoType *t) {
        case MONO_TYPE_SZARRAY:
        case MONO_TYPE_ARRAY:
                return TRUE;
+       case MONO_TYPE_GENERICINST:
+               if (!mono_type_generic_inst_is_valuetype (t))
+                       return TRUE;
+               return FALSE;
        case MONO_TYPE_VALUETYPE:
                return FALSE;
        }
@@ -298,6 +562,7 @@ mono_arch_flush_icache (guint8 *code, gint size)
 enum {
        RegTypeGeneral,
        RegTypeBase,
+       RegTypeBaseGen,
        RegTypeFP,
        RegTypeStructByVal,
        RegTypeStructByAddr
@@ -335,16 +600,30 @@ add_general (guint *gr, guint *stack_size, ArgInfo *ainfo, gboolean simple)
                        ainfo->reg = *gr;
                }
        } else {
-               if (*gr > ARMREG_R2) {
-                       /**stack_size += 7;
-                       *stack_size &= ~7;*/
+               if (*gr == ARMREG_R3
+#ifdef __ARM_EABI__
+                               && 0
+#endif
+                                       ) {
+                       /* first word in r3 and the second on the stack */
+                       ainfo->offset = *stack_size;
+                       ainfo->reg = ARMREG_SP; /* in the caller */
+                       ainfo->regtype = RegTypeBaseGen;
+                       *stack_size += 4;
+               } else if (*gr > ARMREG_R3) {
+#ifdef __ARM_EABI__
+                       *stack_size += 7;
+                       *stack_size &= ~7;
+#endif
                        ainfo->offset = *stack_size;
                        ainfo->reg = ARMREG_SP; /* in the caller */
                        ainfo->regtype = RegTypeBase;
                        *stack_size += 8;
                } else {
-                       /*if ((*gr) & 1)
-                               (*gr) ++;*/
+#ifdef __ARM_EABI__
+                       if ((*gr) & 1)
+                               (*gr) ++;
+#endif
                        ainfo->reg = *gr;
                }
                (*gr) ++;
@@ -426,6 +705,14 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke)
                        add_general (&gr, &stack_size, cinfo->args + n, TRUE);
                        n++;
                        break;
+               case MONO_TYPE_GENERICINST:
+                       if (!mono_type_generic_inst_is_valuetype (sig->params [i])) {
+                               cinfo->args [n].size = sizeof (gpointer);
+                               add_general (&gr, &stack_size, cinfo->args + n, TRUE);
+                               n++;
+                               break;
+                       }
+                       /* Fall through */
                case MONO_TYPE_TYPEDBYREF:
                case MONO_TYPE_VALUETYPE: {
                        gint size;
@@ -511,6 +798,12 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke)
                        /* FIXME: cinfo->ret.reg = ???;
                        cinfo->ret.regtype = RegTypeFP;*/
                        break;
+               case MONO_TYPE_GENERICINST:
+                       if (!mono_type_generic_inst_is_valuetype (sig->ret)) {
+                               cinfo->ret.reg = ARMREG_R0;
+                               break;
+                       }
+                       break;
                case MONO_TYPE_VALUETYPE:
                        break;
                case MONO_TYPE_TYPEDBYREF:
@@ -631,9 +924,9 @@ mono_arch_allocate_vars (MonoCompile *m)
                if ((inst->flags & MONO_INST_IS_DEAD) || inst->opcode == OP_REGVAR)
                        continue;
 
-               /* inst->unused indicates native sized value types, this is used by the
+               /* inst->backend.is_pinvoke indicates native sized value types, this is used by the
                * pinvoke wrappers when they call functions returning structure */
-               if (inst->unused && MONO_TYPE_ISSTRUCT (inst->inst_vtype) && inst->inst_vtype->type != MONO_TYPE_TYPEDBYREF)
+               if (inst->backend.is_pinvoke && MONO_TYPE_ISSTRUCT (inst->inst_vtype) && inst->inst_vtype->type != MONO_TYPE_TYPEDBYREF)
                        size = mono_class_native_size (mono_class_from_mono_type (inst->inst_vtype), &align);
                else
                        size = mono_type_size (inst->inst_vtype, &align);
@@ -654,7 +947,7 @@ mono_arch_allocate_vars (MonoCompile *m)
 
        curinst = 0;
        if (sig->hasthis) {
-               inst = m->varinfo [curinst];
+               inst = m->args [curinst];
                if (inst->opcode != OP_REGVAR) {
                        inst->opcode = OP_REGOFFSET;
                        inst->inst_basereg = frame_reg;
@@ -669,7 +962,7 @@ mono_arch_allocate_vars (MonoCompile *m)
        }
 
        for (i = 0; i < sig->param_count; ++i) {
-               inst = m->varinfo [curinst];
+               inst = m->args [curinst];
                if (inst->opcode != OP_REGVAR) {
                        inst->opcode = OP_REGOFFSET;
                        inst->inst_basereg = frame_reg;
@@ -756,13 +1049,15 @@ mono_arch_call_opcode (MonoCompile *cfg, MonoBasicBlock* bb, MonoCallInst *call,
                        arg->next = call->out_args;
                        call->out_args = arg;
                        if (ainfo->regtype == RegTypeGeneral) {
-                               arg->unused = ainfo->reg;
+                               arg->backend.reg3 = ainfo->reg;
                                call->used_iregs |= 1 << ainfo->reg;
                                if (arg->type == STACK_I8)
                                        call->used_iregs |= 1 << (ainfo->reg + 1);
                                if (arg->type == STACK_R8) {
                                        if (ainfo->size == 4) {
+#ifndef MONO_ARCH_SOFT_FLOAT
                                                arg->opcode = OP_OUTARG_R4;
+#endif
                                        } else {
                                                call->used_iregs |= 1 << (ainfo->reg + 1);
                                        }
@@ -770,7 +1065,7 @@ mono_arch_call_opcode (MonoCompile *cfg, MonoBasicBlock* bb, MonoCallInst *call,
                                }
                        } else if (ainfo->regtype == RegTypeStructByAddr) {
                                /* FIXME: where si the data allocated? */
-                               arg->unused = ainfo->reg;
+                               arg->backend.reg3 = ainfo->reg;
                                call->used_iregs |= 1 << ainfo->reg;
                                g_assert_not_reached ();
                        } else if (ainfo->regtype == RegTypeStructByVal) {
@@ -782,13 +1077,19 @@ mono_arch_call_opcode (MonoCompile *cfg, MonoBasicBlock* bb, MonoCallInst *call,
                                arg->opcode = OP_OUTARG_VT;
                                /* vtsize and offset have just 12 bits of encoding in number of words */
                                g_assert (((ainfo->vtsize | (ainfo->offset / 4)) & 0xfffff000) == 0);
-                               arg->unused = ainfo->reg | (ainfo->size << 4) | (ainfo->vtsize << 8) | ((ainfo->offset / 4) << 20);
+                               arg->backend.arg_info = ainfo->reg | (ainfo->size << 4) | (ainfo->vtsize << 8) | ((ainfo->offset / 4) << 20);
                        } else if (ainfo->regtype == RegTypeBase) {
                                arg->opcode = OP_OUTARG_MEMBASE;
-                               arg->unused = (ainfo->offset << 8) | ainfo->size;
+                               arg->backend.arg_info = (ainfo->offset << 8) | ainfo->size;
+                       } else if (ainfo->regtype == RegTypeBaseGen) {
+                               call->used_iregs |= 1 << ARMREG_R3;
+                               arg->opcode = OP_OUTARG_MEMBASE;
+                               arg->backend.arg_info = (ainfo->offset << 8) | 0xff;
+                               if (arg->type == STACK_R8)
+                                       cfg->flags |= MONO_CFG_HAS_FPOUT;
                        } else if (ainfo->regtype == RegTypeFP) {
-                               arg->unused = ainfo->reg;
-                               /* FPA args are passed in int regs */
+                               arg->backend.reg3 = ainfo->reg;
+                               /* FP args are passed in int regs */
                                call->used_iregs |= 1 << ainfo->reg;
                                if (ainfo->size == 8) {
                                        arg->opcode = OP_OUTARG_R8;
@@ -839,8 +1140,7 @@ mono_arch_instrument_prolog (MonoCompile *cfg, void *func, void *p, gboolean ena
        code = mono_arm_emit_load_imm (code, ARMREG_R0, (guint32)cfg->method);
        ARM_MOV_REG_IMM8 (code, ARMREG_R1, 0); /* NULL ebp for now */
        code = mono_arm_emit_load_imm (code, ARMREG_R2, (guint32)func);
-       ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-       ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_R2);
+       code = emit_call_reg (code, ARMREG_R2);
        return code;
 }
 
@@ -871,7 +1171,6 @@ mono_arch_instrument_epilog (MonoCompile *cfg, void *func, void *p, gboolean ena
                cfg->native_code = g_realloc (cfg->native_code, cfg->code_size);
                code = cfg->native_code + offset;
        }
-handle_enum:
        switch (rtype) {
        case MONO_TYPE_VOID:
                /* special case string .ctor icall */
@@ -930,8 +1229,7 @@ handle_enum:
 
        code = mono_arm_emit_load_imm (code, ARMREG_R0, (guint32)cfg->method);
        code = mono_arm_emit_load_imm (code, ARMREG_IP, (guint32)func);
-       ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-       ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
+       code = emit_call_reg (code, ARMREG_IP);
 
        switch (save_mode) {
        case SAVE_TWO:
@@ -1084,15 +1382,8 @@ peephole_pass (MonoCompile *cfg, MonoBasicBlock *bb)
                        if (last_ins && (last_ins->opcode == OP_STOREI1_MEMBASE_REG) &&
                                        ins->inst_basereg == last_ins->inst_destbasereg &&
                                        ins->inst_offset == last_ins->inst_offset) {
-                               if (ins->dreg == last_ins->sreg1) {
-                                       last_ins->next = ins->next;                             
-                                       ins = ins->next;                                
-                                       continue;
-                               } else {
-                                       //static int c = 0; printf ("MATCHX %s %d\n", cfg->method->name,c++);
-                                       ins->opcode = OP_MOVE;
-                                       ins->sreg1 = last_ins->sreg1;
-                               }
+                               ins->opcode = (ins->opcode == OP_LOADI1_MEMBASE) ? CEE_CONV_I1 : CEE_CONV_U1;
+                               ins->sreg1 = last_ins->sreg1;                           
                        }
                        break;
                case OP_LOADU2_MEMBASE:
@@ -1100,15 +1391,8 @@ peephole_pass (MonoCompile *cfg, MonoBasicBlock *bb)
                        if (last_ins && (last_ins->opcode == OP_STOREI2_MEMBASE_REG) &&
                                        ins->inst_basereg == last_ins->inst_destbasereg &&
                                        ins->inst_offset == last_ins->inst_offset) {
-                               if (ins->dreg == last_ins->sreg1) {
-                                       last_ins->next = ins->next;                             
-                                       ins = ins->next;                                
-                                       continue;
-                               } else {
-                                       //static int c = 0; printf ("MATCHX %s %d\n", cfg->method->name,c++);
-                                       ins->opcode = OP_MOVE;
-                                       ins->sreg1 = last_ins->sreg1;
-                               }
+                               ins->opcode = (ins->opcode == OP_LOADI2_MEMBASE) ? CEE_CONV_I2 : CEE_CONV_U2;
+                               ins->sreg1 = last_ins->sreg1;                           
                        }
                        break;
                case CEE_CONV_I4:
@@ -1263,12 +1547,12 @@ map_to_reg_reg_op (int op)
 static void
 mono_arch_lowering_pass (MonoCompile *cfg, MonoBasicBlock *bb)
 {
-       MonoInst *ins, *next, *temp, *last_ins = NULL;
+       MonoInst *ins, *temp, *last_ins = NULL;
        int rot_amount, imm8, low_imm;
 
        /* setup the virtual reg allocator */
-       if (bb->max_ireg > cfg->rs->next_vireg)
-               cfg->rs->next_vireg = bb->max_ireg;
+       if (bb->max_vreg > cfg->rs->next_vreg)
+               cfg->rs->next_vreg = bb->max_vreg;
 
        ins = bb->code;
        while (ins) {
@@ -1356,7 +1640,7 @@ loop_start:
                                ins->inst_offset = low_imm;
                                break;
                        }
-                       /* FPA doesn't have indexed load instructions */
+                       /* VFP/FPA doesn't have indexed load instructions */
                        g_assert_not_reached ();
                        break;
                case OP_STORE_MEMBASE_REG:
@@ -1394,7 +1678,7 @@ loop_start:
                                break;
                        }
                        /*g_print ("fail with: %d (%d, %d)\n", ins->inst_offset, ins->inst_offset & ~0x1ff, low_imm);*/
-                       /* FPA doesn't have indexed store instructions */
+                       /* VFP/FPA doesn't have indexed store instructions */
                        g_assert_not_reached ();
                        break;
                case OP_STORE_MEMBASE_IMM:
@@ -1413,7 +1697,7 @@ loop_start:
                ins = ins->next;
        }
        bb->last_ins = last_ins;
-       bb->max_ireg = cfg->rs->next_vireg;
+       bb->max_vreg = cfg->rs->next_vreg;
 
 }
 
@@ -1430,7 +1714,15 @@ static guchar*
 emit_float_to_int (MonoCompile *cfg, guchar *code, int dreg, int sreg, int size, gboolean is_signed)
 {
        /* sreg is a float, dreg is an integer reg  */
+#ifdef ARM_FPU_FPA
        ARM_FIXZ (code, dreg, sreg);
+#elif defined(ARM_FPU_VFP)
+       if (is_signed)
+               ARM_TOSIZD (code, ARM_VFP_F0, sreg);
+       else
+               ARM_TOUIZD (code, ARM_VFP_F0, sreg);
+       ARM_FMRS (code, dreg, ARM_VFP_F0);
+#endif
        if (!is_signed) {
                if (size == 1)
                        ARM_AND_REG_IMM8 (code, dreg, dreg, 0xff);
@@ -1465,7 +1757,7 @@ search_thunk_slot (void *data, int csize, int bsize, void *user_data) {
        guchar *code = data;
        guint32 *thunks = data;
        guint32 *endthunks = (guint32*)(code + bsize);
-       int i, count = 0;
+       int count = 0;
        int difflow, diffhigh;
 
        /* always ensure a call from pdata->code can reach to the thunks without further thunks */
@@ -1492,9 +1784,15 @@ search_thunk_slot (void *data, int csize, int bsize, void *user_data) {
                                return 1;
                        } else if ((thunks [0] == 0) && (thunks [1] == 0) && (thunks [2] == 0)) {
                                /* found a free slot instead: emit thunk */
+                               /* ARMREG_IP is fine to use since this can't be an IMT call
+                                * which is indirect
+                                */
                                code = (guchar*)thunks;
                                ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
-                               ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
+                               if (thumb_supported)
+                                       ARM_BX (code, ARMREG_IP);
+                               else
+                                       ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
                                thunks [2] = (guint32)pdata->target;
                                mono_arch_flush_icache ((guchar*)thunks, 12);
 
@@ -1540,19 +1838,37 @@ handle_thunk (int absolute, guchar *code, const guchar *target) {
 void
 arm_patch (guchar *code, const guchar *target)
 {
-       guint32 ins = *(guint32*)code;
+       guint32 *code32 = (void*)code;
+       guint32 ins = *code32;
        guint32 prim = (ins >> 25) & 7;
-       guint32 ovf;
+       guint32 tval = GPOINTER_TO_UINT (target);
 
        //g_print ("patching 0x%08x (0x%08x) to point to 0x%08x\n", code, ins, target);
        if (prim == 5) { /* 101b */
                /* the diff starts 8 bytes from the branch opcode */
                gint diff = target - code - 8;
+               gint tbits;
+               gint tmask = 0xffffffff;
+               if (tval & 1) { /* entering thumb mode */
+                       diff = target - 1 - code - 8;
+                       g_assert (thumb_supported);
+                       tbits = 0xf << 28; /* bl->blx bit pattern */
+                       g_assert ((ins & (1 << 24))); /* it must be a bl, not b instruction */
+                       /* this low bit of the displacement is moved to bit 24 in the instruction encoding */
+                       if (diff & 2) {
+                               tbits |= 1 << 24;
+                       }
+                       tmask = ~(1 << 24); /* clear the link bit */
+                       /*g_print ("blx to thumb: target: %p, code: %p, diff: %d, mask: %x\n", target, code, diff, tmask);*/
+               } else {
+                       tbits = 0;
+               }
                if (diff >= 0) {
                        if (diff <= 33554431) {
                                diff >>= 2;
                                ins = (ins & 0xff000000) | diff;
-                               *(guint32*)code = ins;
+                               ins &= tmask;
+                               *code32 = ins | tbits;
                                return;
                        }
                } else {
@@ -1560,7 +1876,8 @@ arm_patch (guchar *code, const guchar *target)
                        if (diff >= -33554432) {
                                diff >>= 2;
                                ins = (ins & 0xff000000) | (diff & ~0xff000000);
-                               *(guint32*)code = ins;
+                               ins &= tmask;
+                               *code32 = ins | tbits;
                                return;
                        }
                }
@@ -1569,24 +1886,47 @@ arm_patch (guchar *code, const guchar *target)
                return;
        }
 
-
+       /*
+        * The alternative call sequences looks like this:
+        *
+        *      ldr ip, [pc] // loads the address constant
+        *      b 1f         // jumps around the constant
+        *      address constant embedded in the code
+        *   1f:
+        *      mov lr, pc
+        *      mov pc, ip
+        *
+        * There are two cases for patching:
+        * a) at the end of method emission: in this case code points to the start
+        *    of the call sequence
+        * b) during runtime patching of the call site: in this case code points
+        *    to the mov pc, ip instruction
+        *
+        * We have to handle also the thunk jump code sequence:
+        *
+        *      ldr ip, [pc]
+        *      mov pc, ip
+        *      address constant // execution never reaches here
+        */
        if ((ins & 0x0ffffff0) == 0x12fff10) {
                /* branch and exchange: the address is constructed in a reg */
                g_assert_not_reached ();
        } else {
-               guint32 ccode [3];
+               guint32 ccode [4];
                guint32 *tmp = ccode;
-               ARM_LDR_IMM (tmp, ARMREG_IP, ARMREG_PC, 0);
-               ARM_MOV_REG_REG (tmp, ARMREG_LR, ARMREG_PC);
-               ARM_MOV_REG_REG (tmp, ARMREG_PC, ARMREG_IP);
+               guint8 *emit = (guint8*)tmp;
+               ARM_LDR_IMM (emit, ARMREG_IP, ARMREG_PC, 0);
+               ARM_MOV_REG_REG (emit, ARMREG_LR, ARMREG_PC);
+               ARM_MOV_REG_REG (emit, ARMREG_PC, ARMREG_IP);
+               ARM_BX (emit, ARMREG_IP);
                if (ins == ccode [2]) {
-                       tmp = (guint32*)code;
-                       tmp [-1] = (guint32)target;
+                       g_assert_not_reached (); // should be -2 ...
+                       code32 [-1] = (guint32)target;
                        return;
                }
                if (ins == ccode [0]) {
-                       tmp = (guint32*)code;
-                       tmp [2] = (guint32)target;
+                       /* handles both thunk jump code and the far call sequence */
+                       code32 [2] = (guint32)target;
                        return;
                }
                g_assert_not_reached ();
@@ -1600,7 +1940,7 @@ arm_patch (guchar *code, const guchar *target)
  * to be used with the emit macros.
  * Return -1 otherwise.
  */
-int
+static int
 mono_arm_is_rotated_imm8 (guint32 val, gint *rot_amount)
 {
        guint32 res, i;
@@ -1702,7 +2042,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
        while (ins) {
                offset = code - cfg->native_code;
 
-               max_len = ((guint8 *)arm_cpu_desc [ins->opcode])[MONO_INST_LEN];
+               max_len = ((guint8 *)ins_get_spec (ins->opcode))[MONO_INST_LEN];
 
                if (offset > (cfg->code_size - max_len - 16)) {
                        cfg->code_size *= 2;
@@ -1714,6 +2054,8 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                mono_debug_record_line_number (cfg, ins, offset);
 
                switch (ins->opcode) {
+               case OP_MEMORY_BARRIER:
+                       break;
                case OP_TLS_GET:
                        g_assert_not_reached ();
                        break;
@@ -1848,10 +2190,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                        g_assert (imm8 >= 0);
                        ARM_CMP_REG_IMM (code, ins->sreg1, imm8, rot_amount);
                        break;
-               case OP_X86_TEST_NULL:
-                       g_assert_not_reached ();
-                       break;
-               case CEE_BREAK:
+               case OP_BREAK:
                        *(int*)code = 0xe7f001f0;
                        *(int*)code = 0xef9f0001;
                        code += 4;
@@ -2052,12 +2391,21 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                }
                case OP_SETFREG:
                case OP_FMOVE:
+#ifdef ARM_FPU_FPA
                        ARM_MVFD (code, ins->dreg, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CPYD (code, ins->dreg, ins->sreg1);
+#endif
                        break;
                case OP_FCONV_TO_R4:
+#ifdef ARM_FPU_FPA
                        ARM_MVFS (code, ins->dreg, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CVTD (code, ins->dreg, ins->sreg1);
+                       ARM_CVTS (code, ins->dreg, ins->dreg);
+#endif
                        break;
-               case CEE_JMP:
+               case OP_JMP:
                        /*
                         * Keep in sync with mono_arch_emit_epilog
                         */
@@ -2093,24 +2441,14 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                                mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_METHOD, call->method);
                        else
                                mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_ABS, call->fptr);
-                       if (cfg->method->dynamic) {
-                               ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
-                               ARM_B (code, 0);
-                               *(gpointer*)code = NULL;
-                               code += 4;
-                               ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-                               ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
-                       } else {
-                               ARM_BL (code, 0);
-                       }
+                       code = emit_call_seq (cfg, code);
                        break;
                case OP_FCALL_REG:
                case OP_LCALL_REG:
                case OP_VCALL_REG:
                case OP_VOIDCALL_REG:
                case OP_CALL_REG:
-                       ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-                       ARM_MOV_REG_REG (code, ARMREG_PC, ins->sreg1);
+                       code = emit_call_reg (code, ins->sreg1);
                        break;
                case OP_FCALL_MEMBASE:
                case OP_LCALL_MEMBASE:
@@ -2119,8 +2457,16 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                case OP_CALL_MEMBASE:
                        g_assert (arm_is_imm12 (ins->inst_offset));
                        g_assert (ins->sreg1 != ARMREG_LR);
-                       ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-                       ARM_LDR_IMM (code, ARMREG_PC, ins->sreg1, ins->inst_offset);
+                       call = (MonoCallInst*)ins;
+                       if (call->method->klass->flags & TYPE_ATTRIBUTE_INTERFACE) {
+                               ARM_ADD_REG_IMM8 (code, ARMREG_LR, ARMREG_PC, 4);
+                               ARM_LDR_IMM (code, ARMREG_PC, ins->sreg1, ins->inst_offset);
+                               *((gpointer*)code) = (gpointer)call->method;
+                               code += 4;
+                       } else {
+                               ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
+                               ARM_LDR_IMM (code, ARMREG_PC, ins->sreg1, ins->inst_offset);
+                       }
                        break;
                case OP_OUTARG:
                        g_assert_not_reached ();
@@ -2156,21 +2502,12 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                        g_assert_not_reached ();
                        ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_LR);
                        break;
-               case CEE_THROW: {
+               case OP_THROW: {
                        if (ins->sreg1 != ARMREG_R0)
                                ARM_MOV_REG_REG (code, ARMREG_R0, ins->sreg1);
                        mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, 
                                             (gpointer)"mono_arch_throw_exception");
-                       if (cfg->method->dynamic) {
-                               ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
-                               ARM_B (code, 0);
-                               *(gpointer*)code = NULL;
-                               code += 4;
-                               ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-                               ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
-                       } else {
-                               ARM_BL (code, 0);
-                       }
+                       code = emit_call_seq (cfg, code);
                        break;
                }
                case OP_RETHROW: {
@@ -2178,16 +2515,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                                ARM_MOV_REG_REG (code, ARMREG_R0, ins->sreg1);
                        mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, 
                                             (gpointer)"mono_arch_rethrow_exception");
-                       if (cfg->method->dynamic) {
-                               ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
-                               ARM_B (code, 0);
-                               *(gpointer*)code = NULL;
-                               code += 4;
-                               ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-                               ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
-                       } else {
-                               ARM_BL (code, 0);
-                       }
+                       code = emit_call_seq (cfg, code);
                        break;
                }
                case OP_START_HANDLER:
@@ -2210,7 +2538,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                        }
                        ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
                        break;
-               case CEE_ENDFINALLY:
+               case OP_ENDFINALLY:
                        if (arm_is_imm12 (ins->inst_left->inst_offset)) {
                                ARM_LDR_IMM (code, ARMREG_IP, ins->inst_left->inst_basereg, ins->inst_left->inst_offset);
                        } else {
@@ -2227,7 +2555,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                case OP_LABEL:
                        ins->inst_c0 = code - cfg->native_code;
                        break;
-               case CEE_BR:
+               case OP_BR:
                        if (ins->flags & MONO_INST_BRLABEL) {
                                /*if (ins->inst_i0->inst_c0) {
                                        ARM_B (code, 0);
@@ -2272,7 +2600,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                        code += 4 * GPOINTER_TO_INT (ins->klass);
                        break;
                case OP_CEQ:
-                       ARM_MOV_REG_IMM8 (code, ins->dreg, 0);
+                       ARM_MOV_REG_IMM8_COND (code, ins->dreg, 0, ARMCOND_NE);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_EQ);
                        break;
                case OP_CLT:
@@ -2323,6 +2651,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                        break;
 
                /* floating point opcodes */
+#ifdef ARM_FPU_FPA
                case OP_R8CONST:
                        /* FIXME: we can optimize the imm load by dealing with part of 
                         * the displacement in LDFD (aligning to 512).
@@ -2382,14 +2711,48 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                case CEE_CONV_R8:
                        ARM_FLTD (code, ins->dreg, ins->sreg1);
                        break;
-               case OP_X86_FP_LOAD_I8:
+#elif defined(ARM_FPU_VFP)
+               case OP_R8CONST:
+                       /* FIXME: we can optimize the imm load by dealing with part of 
+                        * the displacement in LDFD (aligning to 512).
+                        */
+                       code = mono_arm_emit_load_imm (code, ARMREG_LR, (guint32)ins->inst_p0);
+                       ARM_FLDD (code, ins->dreg, ARMREG_LR, 0);
+                       break;
+               case OP_R4CONST:
+                       code = mono_arm_emit_load_imm (code, ARMREG_LR, (guint32)ins->inst_p0);
+                       ARM_FLDS (code, ins->dreg, ARMREG_LR, 0);
+                       ARM_CVTS (code, ins->dreg, ins->dreg);
+                       break;
+               case OP_STORER8_MEMBASE_REG:
+                       g_assert (arm_is_fpimm8 (ins->inst_offset));
+                       ARM_FSTD (code, ins->sreg1, ins->inst_destbasereg, ins->inst_offset);
+                       break;
+               case OP_LOADR8_MEMBASE:
+                       g_assert (arm_is_fpimm8 (ins->inst_offset));
+                       ARM_FLDD (code, ins->dreg, ins->inst_basereg, ins->inst_offset);
+                       break;
+               case OP_STORER4_MEMBASE_REG:
+                       g_assert (arm_is_fpimm8 (ins->inst_offset));
+                       ARM_FSTS (code, ins->sreg1, ins->inst_destbasereg, ins->inst_offset);
+                       break;
+               case OP_LOADR4_MEMBASE:
+                       g_assert (arm_is_fpimm8 (ins->inst_offset));
+                       ARM_FLDS (code, ins->dreg, ins->inst_basereg, ins->inst_offset);
+                       break;
+               case CEE_CONV_R_UN: {
                        g_assert_not_reached ();
-                       /*x86_fild_membase (code, ins->inst_basereg, ins->inst_offset, TRUE);*/
                        break;
-               case OP_X86_FP_LOAD_I4:
+               }
+               case CEE_CONV_R4:
                        g_assert_not_reached ();
-                       /*x86_fild_membase (code, ins->inst_basereg, ins->inst_offset, FALSE);*/
+                       //ARM_FLTS (code, ins->dreg, ins->sreg1);
                        break;
+               case CEE_CONV_R8:
+                       g_assert_not_reached ();
+                       //ARM_FLTD (code, ins->dreg, ins->sreg1);
+                       break;
+#endif
                case OP_FCONV_TO_I1:
                        code = emit_float_to_int (cfg, code, ins->dreg, ins->sreg1, 1, TRUE);
                        break;
@@ -2448,6 +2811,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                                ARM_MOV_REG_REG (code, ins->dreg, ins->sreg1);
                        break;
                }
+#ifdef ARM_FPU_FPA
                case OP_FADD:
                        ARM_FPA_ADFD (code, ins->dreg, ins->sreg1, ins->sreg2);
                        break;
@@ -2462,7 +2826,24 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                        break;          
                case OP_FNEG:
                        ARM_MNFD (code, ins->dreg, ins->sreg1);
+                       break;
+#elif defined(ARM_FPU_VFP)
+               case OP_FADD:
+                       ARM_VFP_ADDD (code, ins->dreg, ins->sreg1, ins->sreg2);
+                       break;
+               case OP_FSUB:
+                       ARM_VFP_SUBD (code, ins->dreg, ins->sreg1, ins->sreg2);
                        break;          
+               case OP_FMUL:
+                       ARM_VFP_MULD (code, ins->dreg, ins->sreg1, ins->sreg2);
+                       break;          
+               case OP_FDIV:
+                       ARM_VFP_DIVD (code, ins->dreg, ins->sreg1, ins->sreg2);
+                       break;          
+               case OP_FNEG:
+                       ARM_NEGD (code, ins->dreg, ins->sreg1);
+                       break;
+#endif
                case OP_FREM:
                        /* emulated */
                        g_assert_not_reached ();
@@ -2470,33 +2851,53 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                case OP_FCOMPARE:
                        /* each fp compare op needs to do its own */
                        g_assert_not_reached ();
-                       ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+                       //ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
                        break;
                case OP_FCEQ:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
-                       ARM_MOV_REG_IMM8 (code, ins->dreg, 0);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
+                       ARM_MOV_REG_IMM8_COND (code, ins->dreg, 0, ARMCOND_NE);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_EQ);
                        break;
                case OP_FCLT:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        ARM_MOV_REG_IMM8 (code, ins->dreg, 0);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_MI);
                        break;
                case OP_FCLT_UN:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        ARM_MOV_REG_IMM8 (code, ins->dreg, 0);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_MI);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_VS);
                        break;
                case OP_FCGT:
                        /* swapped */
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg2, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg2, ins->sreg1);
+#endif
                        ARM_MOV_REG_IMM8 (code, ins->dreg, 0);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_MI);
                        break;
                case OP_FCGT_UN:
                        /* swapped */
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg2, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg2, ins->sreg1);
+#endif
                        ARM_MOV_REG_IMM8 (code, ins->dreg, 0);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_MI);
                        ARM_MOV_REG_IMM8_COND (code, ins->dreg, 1, ARMCOND_VS);
@@ -2508,50 +2909,90 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
                 * V        Unordered               ARMCOND_VS
                 */
                case OP_FBEQ:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        EMIT_COND_BRANCH (ins, CEE_BEQ - CEE_BEQ);
                        break;
                case OP_FBNE_UN:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        EMIT_COND_BRANCH (ins, CEE_BNE_UN - CEE_BEQ);
                        break;
                case OP_FBLT:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_MI); /* N set */
                        break;
                case OP_FBLT_UN:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_VS); /* V set */
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_MI); /* N set */
                        break;
                case OP_FBGT:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg2, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg2, ins->sreg1);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_MI); /* N set, swapped args */
                        break;
                case OP_FBGT_UN:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg2, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg2, ins->sreg1);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_VS); /* V set */
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_MI); /* N set, swapped args */
                        break;
                case OP_FBGE:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_CS);
                        break;
                case OP_FBGE_UN:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg1, ins->sreg2);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg1, ins->sreg2);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_VS); /* V set */
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_GE);
                        break;
                case OP_FBLE:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg2, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg2, ins->sreg1);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_CS); /* swapped */
                        break;
                case OP_FBLE_UN:
+#ifdef ARM_FPU_FPA
                        ARM_FCMP (code, ARM_FPA_CMF, ins->sreg2, ins->sreg1);
+#elif defined(ARM_FPU_VFP)
+                       ARM_CMPD (code, ins->sreg2, ins->sreg1);
+#endif
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_VS); /* V set */
                        EMIT_COND_BRANCH_FLAGS (ins, ARMCOND_GE); /* swapped */
                        break;
-               case CEE_CKFINITE: {
+               case OP_CKFINITE: {
                        /*ppc_stfd (code, ins->sreg1, -8, ppc_sp);
                        ppc_lwz (code, ppc_r11, -8, ppc_sp);
                        ppc_rlwinm (code, ppc_r11, ppc_r11, 0, 1, 31);
@@ -2604,7 +3045,6 @@ mono_arch_patch_code (MonoMethod *method, MonoDomain *domain, guint8 *code, Mono
                const unsigned char *target;
 
                if (patch_info->type == MONO_PATCH_INFO_SWITCH) {
-                       gpointer *table = (gpointer *)patch_info->data.table->table;
                        gpointer *jt = (gpointer*)(ip + 8);
                        int i;
                        /* jt is the inlined jump table, 2 instructions after ip
@@ -2753,7 +3193,7 @@ mono_arch_emit_prolog (MonoCompile *cfg)
                        max_offset += 6; 
 
                while (ins) {
-                       max_offset += ((guint8 *)arm_cpu_desc [ins->opcode])[MONO_INST_LEN];
+                       max_offset += ((guint8 *)ins_get_spec (ins->opcode))[MONO_INST_LEN];
                        ins = ins->next;
                }
        }
@@ -2771,7 +3211,7 @@ mono_arch_emit_prolog (MonoCompile *cfg)
        }
        for (i = 0; i < sig->param_count + sig->hasthis; ++i) {
                ArgInfo *ainfo = cinfo->args + i;
-               inst = cfg->varinfo [pos];
+               inst = cfg->args [pos];
                
                if (cfg->verbose_level > 2)
                        g_print ("Saving argument %d (type: %d)\n", i, ainfo->regtype);
@@ -2801,8 +3241,13 @@ mono_arch_emit_prolog (MonoCompile *cfg)
                                        }
                                        break;
                                case 2:
-                                       g_assert (arm_is_imm8 (inst->inst_offset));
-                                       ARM_STRH_IMM (code, ainfo->reg, inst->inst_basereg, inst->inst_offset);
+                                       if (arm_is_imm8 (inst->inst_offset)) {
+                                               ARM_STRH_IMM (code, ainfo->reg, inst->inst_basereg, inst->inst_offset);
+                                       } else {
+                                               code = mono_arm_emit_load_imm (code, ARMREG_IP, inst->inst_offset);
+                                               ARM_ADD_REG_REG (code, ARMREG_IP, ARMREG_IP, inst->inst_basereg);
+                                               ARM_STRH_IMM (code, ainfo->reg, ARMREG_IP, 0);
+                                       }
                                        break;
                                case 8:
                                        g_assert (arm_is_imm12 (inst->inst_offset));
@@ -2819,6 +3264,12 @@ mono_arch_emit_prolog (MonoCompile *cfg)
                                        }
                                        break;
                                }
+                       } else if (ainfo->regtype == RegTypeBaseGen) {
+                               g_assert (arm_is_imm12 (prev_sp_offset + ainfo->offset));
+                               g_assert (arm_is_imm12 (inst->inst_offset));
+                               ARM_LDR_IMM (code, ARMREG_LR, ARMREG_SP, (prev_sp_offset + ainfo->offset));
+                               ARM_STR_IMM (code, ARMREG_LR, inst->inst_basereg, inst->inst_offset + 4);
+                               ARM_STR_IMM (code, ARMREG_R3, inst->inst_basereg, inst->inst_offset);
                        } else if (ainfo->regtype == RegTypeBase) {
                                g_assert (arm_is_imm12 (prev_sp_offset + ainfo->offset));
                                switch (ainfo->size) {
@@ -2829,8 +3280,13 @@ mono_arch_emit_prolog (MonoCompile *cfg)
                                        break;
                                case 2:
                                        ARM_LDR_IMM (code, ARMREG_LR, ARMREG_SP, (prev_sp_offset + ainfo->offset));
-                                       g_assert (arm_is_imm8 (inst->inst_offset));
-                                       ARM_STRH_IMM (code, ARMREG_LR, inst->inst_basereg, inst->inst_offset);
+                                       if (arm_is_imm8 (inst->inst_offset)) {
+                                               ARM_STRH_IMM (code, ARMREG_LR, inst->inst_basereg, inst->inst_offset);
+                                       } else {
+                                               code = mono_arm_emit_load_imm (code, ARMREG_IP, inst->inst_offset);
+                                               ARM_ADD_REG_REG (code, ARMREG_IP, ARMREG_IP, inst->inst_basereg);
+                                               ARM_STRH_IMM (code, ARMREG_LR, ARMREG_IP, 0);
+                                       }
                                        break;
                                case 8:
                                        g_assert (arm_is_imm12 (inst->inst_offset));
@@ -2881,16 +3337,7 @@ mono_arch_emit_prolog (MonoCompile *cfg)
 
                mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, 
                             (gpointer)"mono_get_lmf_addr");
-               if (cfg->method->dynamic) {
-                       ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
-                       ARM_B (code, 0);
-                       *(gpointer*)code = NULL;
-                       code += 4;
-                       ARM_MOV_REG_REG (code, ARMREG_LR, ARMREG_PC);
-                       ARM_MOV_REG_REG (code, ARMREG_PC, ARMREG_IP);
-               } else {
-                       ARM_BL (code, 0);
-               }
+               code = emit_call_seq (cfg, code);
                /* we build the MonoLMF structure on the stack - see mini-arm.h */
                /* lmf_offset is the offset from the previous stack pointer,
                 * alloc_size is the total stack space allocated, so the offset
@@ -2908,7 +3355,7 @@ mono_arch_emit_prolog (MonoCompile *cfg)
                /* *(lmf_addr) = r1 */
                ARM_STR_IMM (code, ARMREG_R1, ARMREG_R0, G_STRUCT_OFFSET (MonoLMF, previous_lmf));
                /* save method info */
-               code = mono_arm_emit_load_imm (code, ARMREG_R2, method);
+               code = mono_arm_emit_load_imm (code, ARMREG_R2, GPOINTER_TO_INT (method));
                ARM_STR_IMM (code, ARMREG_R2, ARMREG_R1, G_STRUCT_OFFSET (MonoLMF, method));
                ARM_STR_IMM (code, ARMREG_SP, ARMREG_R1, G_STRUCT_OFFSET (MonoLMF, ebp));
                /* save the current IP */
@@ -2929,7 +3376,6 @@ mono_arch_emit_prolog (MonoCompile *cfg)
 void
 mono_arch_emit_epilog (MonoCompile *cfg)
 {
-       MonoJumpInfo *patch_info;
        MonoMethod *method = cfg->method;
        int pos, i, rot_amount;
        int max_epilog_size = 16 + 20*4;
@@ -2951,7 +3397,7 @@ mono_arch_emit_epilog (MonoCompile *cfg)
        }
 
        /*
-        * Keep in sync with CEE_JMP
+        * Keep in sync with OP_JMP
         */
        code = cfg->native_code + cfg->code_len;
 
@@ -2987,6 +3433,7 @@ mono_arch_emit_epilog (MonoCompile *cfg)
                        code = mono_arm_emit_load_imm (code, ARMREG_IP, cfg->stack_usage);
                        ARM_ADD_REG_REG (code, ARMREG_SP, ARMREG_SP, ARMREG_IP);
                }
+               /* FIXME: add v4 thumb interworking support */
                ARM_POP_NWB (code, cfg->used_int_regs | ((1 << ARMREG_SP) | (1 << ARMREG_PC)));
        }
 
@@ -3015,18 +3462,17 @@ exception_id_by_name (const char *name)
        if (strcmp (name, "ArrayTypeMismatchException") == 0)
                return MONO_EXC_ARRAY_TYPE_MISMATCH;
        g_error ("Unknown intrinsic exception %s\n", name);
+       return -1;
 }
 
 void
 mono_arch_emit_exceptions (MonoCompile *cfg)
 {
        MonoJumpInfo *patch_info;
-       int nthrows, i;
+       int i;
        guint8 *code;
        const guint8* exc_throw_pos [MONO_EXC_INTRINS_NUM] = {NULL};
        guint8 exc_throw_found [MONO_EXC_INTRINS_NUM] = {0};
-       guint32 code_size;
-       int exc_count = 0;
        int max_epilog_size = 50;
 
        /* count the number of exception infos */
@@ -3077,7 +3523,7 @@ mono_arch_emit_exceptions (MonoCompile *cfg)
                        patch_info->data.name = "mono_arch_throw_exception_by_name";
                        patch_info->ip.i = code - cfg->native_code;
                        ARM_B (code, 0);
-                       *(gpointer*)code = ex_name;
+                       *(gconstpointer*)code = ex_name;
                        code += 4;
                        break;
                }
@@ -3120,7 +3566,7 @@ mono_arch_emit_this_vret_args (MonoCompile *cfg, MonoCallInst *inst, int this_re
                this->sreg1 = this_reg;
                this->dreg = mono_regstate_next_int (cfg->rs);
                mono_bblock_add_inst (cfg->cbb, this);
-               mono_call_inst_add_outarg_reg (inst, this->dreg, this_dreg, FALSE);
+               mono_call_inst_add_outarg_reg (cfg, inst, this->dreg, this_dreg, FALSE);
        }
 
        if (vt_reg != -1) {
@@ -3130,14 +3576,19 @@ mono_arch_emit_this_vret_args (MonoCompile *cfg, MonoCallInst *inst, int this_re
                vtarg->sreg1 = vt_reg;
                vtarg->dreg = mono_regstate_next_int (cfg->rs);
                mono_bblock_add_inst (cfg->cbb, vtarg);
-               mono_call_inst_add_outarg_reg (inst, vtarg->dreg, ARMREG_R0, FALSE);
+               mono_call_inst_add_outarg_reg (cfg, inst, vtarg->dreg, ARMREG_R0, FALSE);
        }
 }
 
 MonoInst*
 mono_arch_get_inst_for_method (MonoCompile *cfg, MonoMethod *cmethod, MonoMethodSignature *fsig, MonoInst **args)
 {
-       return NULL;
+       MonoInst *ins = NULL;
+       if (cmethod->klass == mono_defaults.thread_class &&
+                       strcmp (cmethod->name, "MemoryBarrier") == 0) {
+               MONO_INST_NEW (cfg, ins, OP_MEMORY_BARRIER);
+       }
+       return ins;
 }
 
 gboolean
@@ -3170,3 +3621,185 @@ mono_arch_fixup_jinfo (MonoCompile *cfg)
        cfg->jit_info->used_regs |= cfg->stack_usage << 14;
 }
 
+#ifdef MONO_ARCH_HAVE_IMT
+
+void
+mono_arch_emit_imt_argument (MonoCompile *cfg, MonoCallInst *call)
+{
+}
+
+MonoMethod*
+mono_arch_find_imt_method (gpointer *regs, guint8 *code)
+{
+       guint32 *code_ptr = (guint32*)code;
+       code_ptr -= 2;
+       /* The IMT value is stored in the code stream right after the LDC instruction. */
+       if (!IS_LDR_PC (code_ptr [0])) {
+               g_warning ("invalid code stream, instruction before IMT value is not a LDC in %s() (code %p value 0: 0x%x -1: 0x%x -2: 0x%x)", __FUNCTION__, code, code_ptr [2], code_ptr [1], code_ptr [0]);
+               g_assert (IS_LDR_PC (code_ptr [0]));
+       }
+       return (MonoMethod*) code_ptr [1];
+}
+
+MonoObject*
+mono_arch_find_this_argument (gpointer *regs, MonoMethod *method)
+{
+       return mono_arch_get_this_arg_from_call (mono_method_signature (method), (gssize*)regs, NULL);
+}
+
+
+#define ENABLE_WRONG_METHOD_CHECK 0
+#define BASE_SIZE (4 * 4)
+#define BSEARCH_ENTRY_SIZE (4 * 4)
+#define CMP_SIZE (3 * 4)
+#define BRANCH_SIZE (1 * 4)
+#define CALL_SIZE (2 * 4)
+#define WMC_SIZE (5 * 4)
+#define DISTANCE(A, B) (((gint32)(B)) - ((gint32)(A)))
+
+static arminstr_t *
+arm_emit_value_and_patch_ldr (arminstr_t *code, arminstr_t *target, guint32 value)
+{
+       guint32 delta = DISTANCE (target, code);
+       delta -= 8;
+       g_assert (delta >= 0 && delta <= 0xFFF);
+       *target = *target | delta;
+       *code = value;
+       return code + 1;
+}
+
+gpointer
+mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckItem **imt_entries, int count)
+{
+       int size, i, extra_space = 0;
+       arminstr_t *code, *start, *vtable_target = NULL;
+       size = BASE_SIZE;
+
+       for (i = 0; i < count; ++i) {
+               MonoIMTCheckItem *item = imt_entries [i];
+               if (item->is_equals) {
+                       g_assert (arm_is_imm12 (DISTANCE (vtable, &vtable->vtable[item->vtable_slot])));
+
+                       if (item->check_target_idx) {
+                               if (!item->compare_done)
+                                       item->chunk_size += CMP_SIZE;
+                               item->chunk_size += BRANCH_SIZE;
+                       } else {
+#if ENABLE_WRONG_METHOD_CHECK
+                               item->chunk_size += WMC_SIZE;
+#endif
+                       }
+                       item->chunk_size += CALL_SIZE;
+               } else {
+                       item->chunk_size += BSEARCH_ENTRY_SIZE;
+                       imt_entries [item->check_target_idx]->compare_done = TRUE;
+               }
+               size += item->chunk_size;
+       }
+
+       start = code = mono_code_manager_reserve (domain->code_mp, size);
+
+#if DEBUG_IMT
+       printf ("building IMT thunk for class %s %s entries %d code size %d code at %p end %p vtable %p\n", vtable->klass->name_space, vtable->klass->name, count, size, start, ((guint8*)start) + size, vtable);
+       for (i = 0; i < count; ++i) {
+               MonoIMTCheckItem *item = imt_entries [i];
+               printf ("method %d (%p) %s vtable slot %p is_equals %d chunk size %d\n", i, item->method, item->method->name, &vtable->vtable [item->vtable_slot], item->is_equals, item->chunk_size);
+       }
+#endif
+
+       ARM_PUSH2 (code, ARMREG_R0, ARMREG_R1);
+       ARM_LDR_IMM (code, ARMREG_R0, ARMREG_LR, -4);
+       vtable_target = code;
+       ARM_LDR_IMM (code, ARMREG_IP, ARMREG_PC, 0);
+
+       for (i = 0; i < count; ++i) {
+               MonoIMTCheckItem *item = imt_entries [i];
+               arminstr_t *imt_method = NULL;
+               item->code_target = (guint8*)code;
+
+               if (item->is_equals) {
+                       if (item->check_target_idx) {
+                               if (!item->compare_done) {
+                                       imt_method = code;
+                                       ARM_LDR_IMM (code, ARMREG_R1, ARMREG_PC, 0);
+                                       ARM_CMP_REG_REG (code, ARMREG_R0, ARMREG_R1);
+                               }
+                               item->jmp_code = (guint8*)code;
+                               ARM_B_COND (code, ARMCOND_NE, 0);
+
+                               ARM_POP2 (code, ARMREG_R0, ARMREG_R1);
+                               ARM_LDR_IMM (code, ARMREG_PC, ARMREG_IP, DISTANCE (vtable, &vtable->vtable[item->vtable_slot]));
+                       } else {
+                               /*Enable the commented code to assert on wrong method*/
+#if ENABLE_WRONG_METHOD_CHECK
+                               imt_method = code;
+                               ARM_LDR_IMM (code, ARMREG_R1, ARMREG_PC, 0);
+                               ARM_CMP_REG_REG (code, ARMREG_R0, ARMREG_R1);
+                               ARM_B_COND (code, ARMCOND_NE, 1);
+#endif
+                               ARM_POP2 (code, ARMREG_R0, ARMREG_R1);
+                               ARM_LDR_IMM (code, ARMREG_PC, ARMREG_IP, DISTANCE (vtable, &vtable->vtable[item->vtable_slot]));
+
+#if ENABLE_WRONG_METHOD_CHECK
+                               ARM_DBRK (code);
+#endif
+                       }
+
+                       if (imt_method)
+                               code = arm_emit_value_and_patch_ldr (code, imt_method, (guint32)item->method);
+
+                       /*must emit after unconditional branch*/
+                       if (vtable_target) {
+                               code = arm_emit_value_and_patch_ldr (code, vtable_target, (guint32)vtable);
+                               item->chunk_size += 4;
+                               vtable_target = NULL;
+                       }
+
+                       /*We reserve the space for bsearch IMT values after the first entry with an absolute jump*/
+                       if (extra_space) {
+                               code += extra_space;
+                               extra_space = 0;
+                       }
+               } else {
+                       ARM_LDR_IMM (code, ARMREG_R1, ARMREG_PC, 0);
+                       ARM_CMP_REG_REG (code, ARMREG_R0, ARMREG_R1);
+
+                       item->jmp_code = (guint8*)code;
+                       ARM_B_COND (code, ARMCOND_GE, 0);
+                       ++extra_space;
+               }
+       }
+
+       for (i = 0; i < count; ++i) {
+               MonoIMTCheckItem *item = imt_entries [i];
+               if (item->jmp_code) {
+                       if (item->check_target_idx)
+                               arm_patch (item->jmp_code, imt_entries [item->check_target_idx]->code_target);
+               }
+               if (i > 0 && item->is_equals) {
+                       int j;
+                       arminstr_t *space_start = (arminstr_t*)(item->code_target + item->chunk_size);
+                       for (j = i - 1; j >= 0 && !imt_entries [j]->is_equals; --j) {
+                               space_start = arm_emit_value_and_patch_ldr (space_start, (arminstr_t*)imt_entries [j]->code_target, (guint32)imt_entries [j]->method);
+                       }
+               }
+       }
+
+#if DEBUG_IMT
+       {
+               char *buff = g_strdup_printf ("thunk_for_class_%s_%s_entries_%d", vtable->klass->name_space, vtable->klass->name, count);
+               mono_disassemble_code (NULL, (guint8*)start, size, buff);
+               g_free (buff);
+       }
+#endif
+
+       mono_arch_flush_icache ((guint8*)start, size);
+       mono_stats.imt_thunks_size += code - start;
+
+       g_assert (DISTANCE (start, code) <= size);
+       return start;
+}
+
+#endif
+
+