add unmodified Ari64 drc to track it's changes
authornotaz <notasas@gmail.com>
Fri, 19 Nov 2010 22:51:15 +0000 (00:51 +0200)
committernotaz <notasas@gmail.com>
Sat, 20 Nov 2010 13:49:31 +0000 (15:49 +0200)
13 files changed:
libpcsxcore/new_dynarec/assem_arm.c [new file with mode: 0644]
libpcsxcore/new_dynarec/assem_arm.h [new file with mode: 0644]
libpcsxcore/new_dynarec/assem_x64.c [new file with mode: 0644]
libpcsxcore/new_dynarec/assem_x64.h [new file with mode: 0644]
libpcsxcore/new_dynarec/assem_x86.c [new file with mode: 0644]
libpcsxcore/new_dynarec/assem_x86.h [new file with mode: 0644]
libpcsxcore/new_dynarec/fpu.c [new file with mode: 0644]
libpcsxcore/new_dynarec/fpu.h [new file with mode: 0644]
libpcsxcore/new_dynarec/linkage_arm.s [new file with mode: 0644]
libpcsxcore/new_dynarec/linkage_x86.s [new file with mode: 0644]
libpcsxcore/new_dynarec/linkage_x86_64.s [new file with mode: 0644]
libpcsxcore/new_dynarec/new_dynarec.c [new file with mode: 0644]
libpcsxcore/new_dynarec/new_dynarec.h [new file with mode: 0644]

diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c
new file mode 100644 (file)
index 0000000..57684cc
--- /dev/null
@@ -0,0 +1,4353 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - assem_arm.c                                             *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+extern int cycle_count;
+extern int last_count;
+extern int pcaddr;
+extern int pending_exception;
+extern int branch_target;
+extern uint64_t readmem_dword;
+extern precomp_instr fake_pc;
+extern void *dynarec_local;
+extern u_int memory_map[1048576];
+extern u_int mini_ht[32][2];
+extern u_int rounding_modes[4];
+
+void indirect_jump_indexed();
+void indirect_jump();
+void do_interrupt();
+void jump_vaddr_r0();
+void jump_vaddr_r1();
+void jump_vaddr_r2();
+void jump_vaddr_r3();
+void jump_vaddr_r4();
+void jump_vaddr_r5();
+void jump_vaddr_r6();
+void jump_vaddr_r7();
+void jump_vaddr_r8();
+void jump_vaddr_r9();
+void jump_vaddr_r10();
+void jump_vaddr_r12();
+
+const u_int jump_vaddr_reg[16] = {
+  (int)jump_vaddr_r0,
+  (int)jump_vaddr_r1,
+  (int)jump_vaddr_r2,
+  (int)jump_vaddr_r3,
+  (int)jump_vaddr_r4,
+  (int)jump_vaddr_r5,
+  (int)jump_vaddr_r6,
+  (int)jump_vaddr_r7,
+  (int)jump_vaddr_r8,
+  (int)jump_vaddr_r9,
+  (int)jump_vaddr_r10,
+  0,
+  (int)jump_vaddr_r12,
+  0,
+  0,
+  0};
+
+#include "fpu.h"
+
+/* Linker */
+
+void set_jump_target(int addr,u_int target)
+{
+  u_char *ptr=(u_char *)addr;
+  u_int *ptr2=(u_int *)ptr;
+  if(ptr[3]==0xe2) {
+    assert((target-(u_int)ptr2-8)<1024);
+    assert((addr&3)==0);
+    assert((target&3)==0);
+    *ptr2=(*ptr2&0xFFFFF000)|((target-(u_int)ptr2-8)>>2)|0xF00;
+    //printf("target=%x addr=%x insn=%x\n",target,addr,*ptr2);
+  }
+  else if(ptr[3]==0x72) {
+    // generated by emit_jno_unlikely
+    if((target-(u_int)ptr2-8)<1024) {
+      assert((addr&3)==0);
+      assert((target&3)==0);
+      *ptr2=(*ptr2&0xFFFFF000)|((target-(u_int)ptr2-8)>>2)|0xF00;
+    }
+    else if((target-(u_int)ptr2-8)<4096&&!((target-(u_int)ptr2-8)&15)) {
+      assert((addr&3)==0);
+      assert((target&3)==0);
+      *ptr2=(*ptr2&0xFFFFF000)|((target-(u_int)ptr2-8)>>4)|0xE00;
+    }
+    else *ptr2=(0x7A000000)|(((target-(u_int)ptr2-8)<<6)>>8);
+  }
+  else {
+    assert((ptr[3]&0x0e)==0xa);
+    *ptr2=(*ptr2&0xFF000000)|(((target-(u_int)ptr2-8)<<6)>>8);
+  }
+}
+
+// This optionally copies the instruction from the target of the branch into
+// the space before the branch.  Works, but the difference in speed is
+// usually insignificant.
+void set_jump_target_fillslot(int addr,u_int target,int copy)
+{
+  u_char *ptr=(u_char *)addr;
+  u_int *ptr2=(u_int *)ptr;
+  assert(!copy||ptr2[-1]==0xe28dd000);
+  if(ptr[3]==0xe2) {
+    assert(!copy);
+    assert((target-(u_int)ptr2-8)<4096);
+    *ptr2=(*ptr2&0xFFFFF000)|(target-(u_int)ptr2-8);
+  }
+  else {
+    assert((ptr[3]&0x0e)==0xa);
+    u_int target_insn=*(u_int *)target;
+    if((target_insn&0x0e100000)==0) { // ALU, no immediate, no flags
+      copy=0;
+    }
+    if((target_insn&0x0c100000)==0x04100000) { // Load
+      copy=0;
+    }
+    if(target_insn&0x08000000) {
+      copy=0;
+    }
+    if(copy) {
+      ptr2[-1]=target_insn;
+      target+=4;
+    }
+    *ptr2=(*ptr2&0xFF000000)|(((target-(u_int)ptr2-8)<<6)>>8);
+  }
+}
+
+/* Literal pool */
+add_literal(int addr,int val)
+{
+  literals[literalcount][0]=addr;
+  literals[literalcount][1]=val;
+  literalcount++; 
+} 
+
+void kill_pointer(void *stub)
+{
+  int *ptr=(int *)(stub+4);
+  assert((*ptr&0x0ff00000)==0x05900000);
+  u_int offset=*ptr&0xfff;
+  int **l_ptr=(void *)ptr+offset+8;
+  int *i_ptr=*l_ptr;
+  set_jump_target((int)i_ptr,(int)stub);
+}
+
+int get_pointer(void *stub)
+{
+  //printf("get_pointer(%x)\n",(int)stub);
+  int *ptr=(int *)(stub+4);
+  assert((*ptr&0x0ff00000)==0x05900000);
+  u_int offset=*ptr&0xfff;
+  int **l_ptr=(void *)ptr+offset+8;
+  int *i_ptr=*l_ptr;
+  assert((*i_ptr&0x0f000000)==0x0a000000);
+  return (int)i_ptr+((*i_ptr<<8)>>6)+8;
+}
+
+// Find the "clean" entry point from a "dirty" entry point
+// by skipping past the call to verify_code
+u_int get_clean_addr(int addr)
+{
+  int *ptr=(int *)addr;
+  #ifdef ARMv5_ONLY
+  ptr+=4;
+  #else
+  ptr+=6;
+  #endif
+  if((*ptr&0xFF000000)!=0xeb000000) ptr++;
+  assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
+  ptr++;
+  if((*ptr&0xFF000000)==0xea000000) {
+    return (int)ptr+((*ptr<<8)>>6)+8; // follow jump
+  }
+  return (u_int)ptr;
+}
+
+int verify_dirty(int addr)
+{
+  u_int *ptr=(u_int *)addr;
+  #ifdef ARMv5_ONLY
+  // get from literal pool
+  assert((*ptr&0xFFF00000)==0xe5900000);
+  u_int offset=*ptr&0xfff;
+  u_int *l_ptr=(void *)ptr+offset+8;
+  u_int source=l_ptr[0];
+  u_int copy=l_ptr[1];
+  u_int len=l_ptr[2];
+  ptr+=4;
+  #else
+  // ARMv7 movw/movt
+  assert((*ptr&0xFFF00000)==0xe3000000);
+  u_int source=(ptr[0]&0xFFF)+((ptr[0]>>4)&0xF000)+((ptr[2]<<16)&0xFFF0000)+((ptr[2]<<12)&0xF0000000);
+  u_int copy=(ptr[1]&0xFFF)+((ptr[1]>>4)&0xF000)+((ptr[3]<<16)&0xFFF0000)+((ptr[3]<<12)&0xF0000000);
+  u_int len=(ptr[4]&0xFFF)+((ptr[4]>>4)&0xF000);
+  ptr+=6;
+  #endif
+  if((*ptr&0xFF000000)!=0xeb000000) ptr++;
+  assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
+  u_int verifier=(int)ptr+((*ptr<<8)>>6)+8; // get target of bl
+  if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) {
+    unsigned int page=source>>12;
+    unsigned int map_value=memory_map[page];
+    if(map_value>=0x80000000) return 0;
+    while(page<((source+len-1)>>12)) {
+      if((memory_map[++page]<<2)!=(map_value<<2)) return 0;
+    }
+    source = source+(map_value<<2);
+  }
+  //printf("verify_dirty: %x %x %x\n",source,copy,len);
+  return !memcmp((void *)source,(void *)copy,len);
+}
+
+// This doesn't necessarily find all clean entry points, just
+// guarantees that it's not dirty
+int isclean(int addr)
+{
+  #ifdef ARMv5_ONLY
+  int *ptr=((u_int *)addr)+4;
+  #else
+  int *ptr=((u_int *)addr)+6;
+  #endif
+  if((*ptr&0xFF000000)!=0xeb000000) ptr++;
+  if((*ptr&0xFF000000)!=0xeb000000) return 1; // bl instruction
+  if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code) return 0;
+  if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_vm) return 0;
+  if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_ds) return 0;
+  return 1;
+}
+
+void get_bounds(int addr,u_int *start,u_int *end)
+{
+  u_int *ptr=(u_int *)addr;
+  #ifdef ARMv5_ONLY
+  // get from literal pool
+  assert((*ptr&0xFFF00000)==0xe5900000);
+  u_int offset=*ptr&0xfff;
+  u_int *l_ptr=(void *)ptr+offset+8;
+  u_int source=l_ptr[0];
+  //u_int copy=l_ptr[1];
+  u_int len=l_ptr[2];
+  ptr+=4;
+  #else
+  // ARMv7 movw/movt
+  assert((*ptr&0xFFF00000)==0xe3000000);
+  u_int source=(ptr[0]&0xFFF)+((ptr[0]>>4)&0xF000)+((ptr[2]<<16)&0xFFF0000)+((ptr[2]<<12)&0xF0000000);
+  //u_int copy=(ptr[1]&0xFFF)+((ptr[1]>>4)&0xF000)+((ptr[3]<<16)&0xFFF0000)+((ptr[3]<<12)&0xF0000000);
+  u_int len=(ptr[4]&0xFFF)+((ptr[4]>>4)&0xF000);
+  ptr+=6;
+  #endif
+  if((*ptr&0xFF000000)!=0xeb000000) ptr++;
+  assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
+  u_int verifier=(int)ptr+((*ptr<<8)>>6)+8; // get target of bl
+  if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) {
+    if(memory_map[source>>12]>=0x80000000) source = 0;
+    else source = source+(memory_map[source>>12]<<2);
+  }
+  *start=source;
+  *end=source+len;
+}
+
+/* Register allocation */
+
+// Note: registers are allocated clean (unmodified state)
+// if you intend to modify the register, you must call dirty_reg().
+void alloc_reg(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = (reg&7);
+  if(reg==CCREG) preferred_reg=HOST_CCREG;
+  if(reg==PTEMP||reg==FTEMP) preferred_reg=12;
+  
+  // Don't allocate unused registers
+  if((cur->u>>reg)&1) return;
+  
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg) return;
+  }
+  
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+  
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  if(r<64&&((cur->u>>r)&1)) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  if(r>=64&&((cur->uu>>(r&63))&1)) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;}
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {cur->regmap[hr]=-1;break;}
+      }
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen (alloc_reg)");exit(1);
+}
+
+void alloc_reg64(struct regstat *cur,int i,signed char reg)
+{
+  int preferred_reg = 8+(reg&1);
+  int r,hr;
+  
+  // allocate the lower 32 bits
+  alloc_reg(cur,i,reg);
+  
+  // Don't allocate unused registers
+  if((cur->uu>>reg)&1) return;
+  
+  // see if the upper half is already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg+64) return;
+  }
+  
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+  
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  if(r<64&&((cur->u>>r)&1)) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  if(r>=64&&((cur->uu>>(r&63))&1)) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;}
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {cur->regmap[hr]=-1;break;}
+      }
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg|64;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg|64;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg|64;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg|64;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg|64;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg|64;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg|64;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen");exit(1);
+}
+
+// Allocate a temporary register.  This is done without regard to
+// dirty status or whether the register we request is on the unneeded list
+// Note: This will only allocate one register, even if called multiple times
+void alloc_reg_temp(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = -1;
+  
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==reg) return;
+  }
+  
+  // Try to allocate any available register
+  for(hr=HOST_REGS-1;hr>=0;hr--) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Find an unneeded register
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {
+          if(i==0||((unneeded_reg[i-1]>>r)&1)) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {
+          if(i==0||((unneeded_reg_upper[i-1]>>(r&63))&1)) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  // TODO: we might want to follow unconditional jumps here
+  // TODO: get rid of dupe code and make this into a function
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("hsn: %d %d %d %d %d %d %d\n",hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen");exit(1);
+}
+// Allocate a specific ARM register.
+void alloc_arm_reg(struct regstat *cur,int i,signed char reg,char hr)
+{
+  int n;
+  
+  // see if it's already allocated (and dealloc it)
+  for(n=0;n<HOST_REGS;n++)
+  {
+    if(n!=EXCLUDE_REG&&cur->regmap[n]==reg) {cur->regmap[n]=-1;}
+  }
+  
+  cur->regmap[hr]=reg;
+  cur->dirty&=~(1<<hr);
+  cur->isconst&=~(1<<hr);
+}
+
+// Alloc cycle count into dedicated register
+alloc_cc(struct regstat *cur,int i)
+{
+  alloc_arm_reg(cur,i,CCREG,HOST_CCREG);
+}
+
+/* Special alloc */
+
+
+/* Assembler */
+
+char regname[16][4] = {
+ "r0",
+ "r1",
+ "r2",
+ "r3",
+ "r4",
+ "r5",
+ "r6",
+ "r7",
+ "r8",
+ "r9",
+ "r10",
+ "fp",
+ "r12",
+ "sp",
+ "lr",
+ "pc"};
+
+void output_byte(u_char byte)
+{
+  *(out++)=byte;
+}
+void output_modrm(u_char mod,u_char rm,u_char ext)
+{
+  assert(mod<4);
+  assert(rm<8);
+  assert(ext<8);
+  u_char byte=(mod<<6)|(ext<<3)|rm;
+  *(out++)=byte;
+}
+void output_sib(u_char scale,u_char index,u_char base)
+{
+  assert(scale<4);
+  assert(index<8);
+  assert(base<8);
+  u_char byte=(scale<<6)|(index<<3)|base;
+  *(out++)=byte;
+}
+void output_w32(u_int word)
+{
+  *((u_int *)out)=word;
+  out+=4;
+}
+u_int rd_rn_rm(u_int rd, u_int rn, u_int rm)
+{
+  assert(rd<16);
+  assert(rn<16);
+  assert(rm<16);
+  return((rn<<16)|(rd<<12)|rm);
+}
+u_int rd_rn_imm_shift(u_int rd, u_int rn, u_int imm, u_int shift)
+{
+  assert(rd<16);
+  assert(rn<16);
+  assert(imm<256);
+  assert((shift&1)==0);
+  return((rn<<16)|(rd<<12)|(((32-shift)&30)<<7)|imm);
+}
+u_int genimm(u_int imm,u_int *encoded)
+{
+  if(imm==0) {*encoded=0;return 1;}
+  int i=32;
+  while(i>0)
+  {
+    if(imm<256) {
+      *encoded=((i&30)<<7)|imm;
+      return 1;
+    }
+    imm=(imm>>2)|(imm<<30);i-=2;
+  }
+  return 0;
+}
+u_int genjmp(u_int addr)
+{
+  int offset=addr-(int)out-8;
+  if(offset<-33554432||offset>=33554432) return 0;
+  return ((u_int)offset>>2)&0xffffff;
+}
+
+void emit_mov(int rs,int rt)
+{
+  assem_debug("mov %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs));
+}
+
+void emit_movs(int rs,int rt)
+{
+  assem_debug("movs %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0xe1b00000|rd_rn_rm(rt,0,rs));
+}
+
+void emit_add(int rs1,int rs2,int rt)
+{
+  assem_debug("add %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0800000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_adds(int rs1,int rs2,int rt)
+{
+  assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0900000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_adcs(int rs1,int rs2,int rt)
+{
+  assem_debug("adcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0b00000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_sbc(int rs1,int rs2,int rt)
+{
+  assem_debug("sbc %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0c00000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_sbcs(int rs1,int rs2,int rt)
+{
+  assem_debug("sbcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0d00000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_neg(int rs, int rt)
+{
+  assem_debug("rsb %s,%s,#0\n",regname[rt],regname[rs]);
+  output_w32(0xe2600000|rd_rn_rm(rt,rs,0));
+}
+
+void emit_negs(int rs, int rt)
+{
+  assem_debug("rsbs %s,%s,#0\n",regname[rt],regname[rs]);
+  output_w32(0xe2700000|rd_rn_rm(rt,rs,0));
+}
+
+void emit_sub(int rs1,int rs2,int rt)
+{
+  assem_debug("sub %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0400000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_subs(int rs1,int rs2,int rt)
+{
+  assem_debug("subs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0500000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_zeroreg(int rt)
+{
+  assem_debug("mov %s,#0\n",regname[rt]);
+  output_w32(0xe3a00000|rd_rn_rm(rt,0,0));
+}
+
+void emit_loadreg(int r, int hr)
+{
+  if((r&63)==0)
+    emit_zeroreg(hr);
+  else {
+    int addr=((int)reg)+((r&63)<<3)+((r&64)>>4);
+    if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
+    if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
+    if(r==CCREG) addr=(int)&cycle_count;
+    if(r==CSREG) addr=(int)&Status;
+    if(r==FSREG) addr=(int)&FCR31;
+    if(r==INVCP) addr=(int)&invc_ptr;
+    u_int offset = addr-(u_int)&dynarec_local;
+    assert(offset<4096);
+    assem_debug("ldr %s,fp+%d\n",regname[hr],offset);
+    output_w32(0xe5900000|rd_rn_rm(hr,FP,0)|offset);
+  }
+}
+void emit_storereg(int r, int hr)
+{
+  int addr=((int)reg)+((r&63)<<3)+((r&64)>>4);
+  if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
+  if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
+  if(r==CCREG) addr=(int)&cycle_count;
+  if(r==FSREG) addr=(int)&FCR31;
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<4096);
+  assem_debug("str %s,fp+%d\n",regname[hr],offset);
+  output_w32(0xe5800000|rd_rn_rm(hr,FP,0)|offset);
+}
+
+void emit_test(int rs, int rt)
+{
+  assem_debug("tst %s,%s\n",regname[rs],regname[rt]);
+  output_w32(0xe1100000|rd_rn_rm(0,rs,rt));
+}
+
+void emit_testimm(int rs,int imm)
+{
+  u_int armval;
+  assem_debug("tst %s,$%d\n",regname[rs],imm);
+  assert(genimm(imm,&armval));
+  output_w32(0xe3100000|rd_rn_rm(0,rs,0)|armval);
+}
+
+void emit_not(int rs,int rt)
+{
+  assem_debug("mvn %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0xe1e00000|rd_rn_rm(rt,0,rs));
+}
+
+void emit_and(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("and %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0000000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_or(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("orr %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe1800000|rd_rn_rm(rt,rs1,rs2));
+}
+void emit_or_and_set_flags(int rs1,int rs2,int rt)
+{
+  assem_debug("orrs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe1900000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_xor(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("eor %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0200000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_loadlp(u_int imm,u_int rt)
+{
+  add_literal((int)out,imm);
+  assem_debug("ldr %s,pc+? [=%x]\n",regname[rt],imm);
+  output_w32(0xe5900000|rd_rn_rm(rt,15,0));
+}
+void emit_movw(u_int imm,u_int rt)
+{
+  assert(imm<65536);
+  assem_debug("movw %s,#%d (0x%x)\n",regname[rt],imm,imm);
+  output_w32(0xe3000000|rd_rn_rm(rt,0,0)|(imm&0xfff)|((imm<<4)&0xf0000));
+}
+void emit_movt(u_int imm,u_int rt)
+{
+  assem_debug("movt %s,#%d (0x%x)\n",regname[rt],imm&0xffff0000,imm&0xffff0000);
+  output_w32(0xe3400000|rd_rn_rm(rt,0,0)|((imm>>16)&0xfff)|((imm>>12)&0xf0000));
+}
+void emit_movimm(u_int imm,u_int rt)
+{
+  u_int armval;
+  if(genimm(imm,&armval)) {
+    assem_debug("mov %s,#%d\n",regname[rt],imm);
+    output_w32(0xe3a00000|rd_rn_rm(rt,0,0)|armval);
+  }else if(genimm(~imm,&armval)) {
+    assem_debug("mvn %s,#%d\n",regname[rt],imm);
+    output_w32(0xe3e00000|rd_rn_rm(rt,0,0)|armval);
+  }else if(imm<65536) {
+    #ifdef ARMv5_ONLY
+    assem_debug("mov %s,#%d\n",regname[rt],imm&0xFF00);
+    output_w32(0xe3a00000|rd_rn_imm_shift(rt,0,imm>>8,8));
+    assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF);
+    output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm&0xff,0));
+    #else
+    emit_movw(imm,rt);
+    #endif
+  }else{
+    #ifdef ARMv5_ONLY
+    emit_loadlp(imm,rt);
+    #else
+    emit_movw(imm&0x0000FFFF,rt);
+    emit_movt(imm&0xFFFF0000,rt);
+    #endif
+  }
+}
+void emit_pcreladdr(u_int rt)
+{
+  assem_debug("add %s,pc,#?\n",regname[rt]);
+  output_w32(0xe2800000|rd_rn_rm(rt,15,0));
+}
+
+void emit_addimm(u_int rs,int imm,u_int rt)
+{
+  assert(rs<16);
+  assert(rt<16);
+  if(imm!=0) {
+    assert(imm>-65536&&imm<65536);
+    u_int armval;
+    if(genimm(imm,&armval)) {
+      assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],imm);
+      output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval);
+    }else if(genimm(-imm,&armval)) {
+      assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],imm);
+      output_w32(0xe2400000|rd_rn_rm(rt,rs,0)|armval);
+    }else if(imm<0) {
+      assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],(-imm)&0xFF00);
+      assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF);
+      output_w32(0xe2400000|rd_rn_imm_shift(rt,rs,(-imm)>>8,8));
+      output_w32(0xe2400000|rd_rn_imm_shift(rt,rt,(-imm)&0xff,0));
+    }else{
+      assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF00);
+      assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF);
+      output_w32(0xe2800000|rd_rn_imm_shift(rt,rs,imm>>8,8));
+      output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm&0xff,0));
+    }
+  }
+  else if(rs!=rt) emit_mov(rs,rt);
+}
+
+void emit_addimm_and_set_flags(int imm,int rt)
+{
+  assert(imm>-65536&&imm<65536);
+  u_int armval;
+  if(genimm(imm,&armval)) {
+    assem_debug("adds %s,%s,#%d\n",regname[rt],regname[rt],imm);
+    output_w32(0xe2900000|rd_rn_rm(rt,rt,0)|armval);
+  }else if(genimm(-imm,&armval)) {
+    assem_debug("subs %s,%s,#%d\n",regname[rt],regname[rt],imm);
+    output_w32(0xe2500000|rd_rn_rm(rt,rt,0)|armval);
+  }else if(imm<0) {
+    assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF00);
+    assem_debug("subs %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF);
+    output_w32(0xe2400000|rd_rn_imm_shift(rt,rt,(-imm)>>8,8));
+    output_w32(0xe2500000|rd_rn_imm_shift(rt,rt,(-imm)&0xff,0));
+  }else{
+    assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF00);
+    assem_debug("adds %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF);
+    output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm>>8,8));
+    output_w32(0xe2900000|rd_rn_imm_shift(rt,rt,imm&0xff,0));
+  }
+}
+void emit_addimm_no_flags(u_int imm,u_int rt)
+{
+  emit_addimm(rt,imm,rt);
+}
+
+void emit_addnop(u_int r)
+{
+  assert(r<16);
+  assem_debug("add %s,%s,#0 (nop)\n",regname[r],regname[r]);
+  output_w32(0xe2800000|rd_rn_rm(r,r,0));
+}
+
+void emit_adcimm(u_int rs,int imm,u_int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("adc %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe2a00000|rd_rn_rm(rt,rs,0)|armval);
+}
+/*void emit_sbcimm(int imm,u_int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("sbc %s,%s,#%d\n",regname[rt],regname[rt],imm);
+  output_w32(0xe2c00000|rd_rn_rm(rt,rt,0)|armval);
+}*/
+void emit_sbbimm(int imm,u_int rt)
+{
+  assem_debug("sbb $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<8);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,3);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,3);
+    output_w32(imm);
+  }
+}
+void emit_rscimm(int rs,int imm,u_int rt)
+{
+  assert(0);
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("rsc %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe2e00000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+void emit_addimm64_32(int rsh,int rsl,int imm,int rth,int rtl)
+{
+  // TODO: if(genimm(imm,&armval)) ...
+  // else
+  emit_movimm(imm,HOST_TEMPREG);
+  emit_adds(HOST_TEMPREG,rsl,rtl);
+  emit_adcimm(rsh,0,rth);
+}
+
+void emit_sbb(int rs1,int rs2)
+{
+  assem_debug("sbb %%%s,%%%s\n",regname[rs2],regname[rs1]);
+  output_byte(0x19);
+  output_modrm(3,rs1,rs2);
+}
+
+void emit_andimm(int rs,int imm,int rt)
+{
+  u_int armval;
+  if(genimm(imm,&armval)) {
+    assem_debug("and %s,%s,#%d\n",regname[rt],regname[rs],imm);
+    output_w32(0xe2000000|rd_rn_rm(rt,rs,0)|armval);
+  }else if(genimm(~imm,&armval)) {
+    assem_debug("bic %s,%s,#%d\n",regname[rt],regname[rs],imm);
+    output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|armval);
+  }else if(imm==65535) {
+    #ifdef ARMv5_ONLY
+    assem_debug("bic %s,%s,#FF000000\n",regname[rt],regname[rs]);
+    output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|0x4FF);
+    assem_debug("bic %s,%s,#00FF0000\n",regname[rt],regname[rt]);
+    output_w32(0xe3c00000|rd_rn_rm(rt,rt,0)|0x8FF);
+    #else
+    assem_debug("uxth %s,%s\n",regname[rt],regname[rs]);
+    output_w32(0xe6ff0070|rd_rn_rm(rt,0,rs));
+    #endif
+  }else{
+    assert(imm>0&&imm<65535);
+    #ifdef ARMv5_ONLY
+    assem_debug("mov r14,#%d\n",imm&0xFF00);
+    output_w32(0xe3a00000|rd_rn_imm_shift(HOST_TEMPREG,0,imm>>8,8));
+    assem_debug("add r14,r14,#%d\n",imm&0xFF);
+    output_w32(0xe2800000|rd_rn_imm_shift(HOST_TEMPREG,HOST_TEMPREG,imm&0xff,0));
+    #else
+    emit_movw(imm,HOST_TEMPREG);
+    #endif
+    assem_debug("and %s,%s,r14\n",regname[rt],regname[rs]);
+    output_w32(0xe0000000|rd_rn_rm(rt,rs,HOST_TEMPREG));
+  }
+}
+
+void emit_orimm(int rs,int imm,int rt)
+{
+  u_int armval;
+  if(genimm(imm,&armval)) {
+    assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm);
+    output_w32(0xe3800000|rd_rn_rm(rt,rs,0)|armval);
+  }else{
+    assert(imm>0&&imm<65536);
+    assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF00);
+    assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF);
+    output_w32(0xe3800000|rd_rn_imm_shift(rt,rs,imm>>8,8));
+    output_w32(0xe3800000|rd_rn_imm_shift(rt,rt,imm&0xff,0));
+  }
+}
+
+void emit_xorimm(int rs,int imm,int rt)
+{
+  assert(imm>0&&imm<65536);
+  u_int armval;
+  if(genimm(imm,&armval)) {
+    assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm);
+    output_w32(0xe2200000|rd_rn_rm(rt,rs,0)|armval);
+  }else{
+    assert(imm>0);
+    assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF00);
+    assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF);
+    output_w32(0xe2200000|rd_rn_imm_shift(rt,rs,imm>>8,8));
+    output_w32(0xe2200000|rd_rn_imm_shift(rt,rt,imm&0xff,0));
+  }
+}
+
+void emit_shlimm(int rs,u_int imm,int rt)
+{
+  assert(imm>0);
+  assert(imm<32);
+  //if(imm==1) ...
+  assem_debug("lsl %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|(imm<<7));
+}
+
+void emit_shrimm(int rs,u_int imm,int rt)
+{
+  assert(imm>0);
+  assert(imm<32);
+  assem_debug("lsr %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7));
+}
+
+void emit_sarimm(int rs,u_int imm,int rt)
+{
+  assert(imm>0);
+  assert(imm<32);
+  assem_debug("asr %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x40|(imm<<7));
+}
+
+void emit_rorimm(int rs,u_int imm,int rt)
+{
+  assert(imm>0);
+  assert(imm<32);
+  assem_debug("ror %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x60|(imm<<7));
+}
+
+void emit_shldimm(int rs,int rs2,u_int imm,int rt)
+{
+  assem_debug("shld %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm);
+  assert(imm>0);
+  assert(imm<32);
+  //if(imm==1) ...
+  assem_debug("lsl %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|(imm<<7));
+  assem_debug("orr %s,%s,%s,lsr #%d\n",regname[rt],regname[rt],regname[rs2],32-imm);
+  output_w32(0xe1800020|rd_rn_rm(rt,rt,rs2)|((32-imm)<<7));
+}
+
+void emit_shrdimm(int rs,int rs2,u_int imm,int rt)
+{
+  assem_debug("shrd %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm);
+  assert(imm>0);
+  assert(imm<32);
+  //if(imm==1) ...
+  assem_debug("lsr %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe1a00020|rd_rn_rm(rt,0,rs)|(imm<<7));
+  assem_debug("orr %s,%s,%s,lsl #%d\n",regname[rt],regname[rt],regname[rs2],32-imm);
+  output_w32(0xe1800000|rd_rn_rm(rt,rt,rs2)|((32-imm)<<7));
+}
+
+void emit_shl(u_int rs,u_int shift,u_int rt)
+{
+  assert(rs<16);
+  assert(rt<16);
+  assert(shift<16);
+  //if(imm==1) ...
+  assem_debug("lsl %s,%s,%s\n",regname[rt],regname[rs],regname[shift]);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x10|(shift<<8));
+}
+void emit_shr(u_int rs,u_int shift,u_int rt)
+{
+  assert(rs<16);
+  assert(rt<16);
+  assert(shift<16);
+  assem_debug("lsr %s,%s,%s\n",regname[rt],regname[rs],regname[shift]);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x30|(shift<<8));
+}
+void emit_sar(u_int rs,u_int shift,u_int rt)
+{
+  assert(rs<16);
+  assert(rt<16);
+  assert(shift<16);
+  assem_debug("asr %s,%s,%s\n",regname[rt],regname[rs],regname[shift]);
+  output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|0x50|(shift<<8));
+}
+void emit_shlcl(int r)
+{
+  assem_debug("shl %%%s,%%cl\n",regname[r]);
+  assert(0);
+}
+void emit_shrcl(int r)
+{
+  assem_debug("shr %%%s,%%cl\n",regname[r]);
+  assert(0);
+}
+void emit_sarcl(int r)
+{
+  assem_debug("sar %%%s,%%cl\n",regname[r]);
+  assert(0);
+}
+
+void emit_shldcl(int r1,int r2)
+{
+  assem_debug("shld %%%s,%%%s,%%cl\n",regname[r1],regname[r2]);
+  assert(0);
+}
+void emit_shrdcl(int r1,int r2)
+{
+  assem_debug("shrd %%%s,%%%s,%%cl\n",regname[r1],regname[r2]);
+  assert(0);
+}
+void emit_orrshl(u_int rs,u_int shift,u_int rt)
+{
+  assert(rs<16);
+  assert(rt<16);
+  assert(shift<16);
+  assem_debug("orr %s,%s,%s,lsl %s\n",regname[rt],regname[rt],regname[rs],regname[shift]);
+  output_w32(0xe1800000|rd_rn_rm(rt,rt,rs)|0x10|(shift<<8));
+}
+void emit_orrshr(u_int rs,u_int shift,u_int rt)
+{
+  assert(rs<16);
+  assert(rt<16);
+  assert(shift<16);
+  assem_debug("orr %s,%s,%s,lsr %s\n",regname[rt],regname[rt],regname[rs],regname[shift]);
+  output_w32(0xe1800000|rd_rn_rm(rt,rt,rs)|0x30|(shift<<8));
+}
+
+void emit_cmpimm(int rs,int imm)
+{
+  u_int armval;
+  if(genimm(imm,&armval)) {
+    assem_debug("cmp %s,$%d\n",regname[rs],imm);
+    output_w32(0xe3500000|rd_rn_rm(0,rs,0)|armval);
+  }else if(genimm(-imm,&armval)) {
+    assem_debug("cmn %s,$%d\n",regname[rs],imm);
+    output_w32(0xe3700000|rd_rn_rm(0,rs,0)|armval);
+  }else if(imm>0) {
+    assert(imm<65536);
+    #ifdef ARMv5_ONLY
+    emit_movimm(imm,HOST_TEMPREG);
+    #else
+    emit_movw(imm,HOST_TEMPREG);
+    #endif
+    assem_debug("cmp %s,r14\n",regname[rs]);
+    output_w32(0xe1500000|rd_rn_rm(0,rs,HOST_TEMPREG));
+  }else{
+    assert(imm>-65536);
+    #ifdef ARMv5_ONLY
+    emit_movimm(-imm,HOST_TEMPREG);
+    #else
+    emit_movw(-imm,HOST_TEMPREG);
+    #endif
+    assem_debug("cmn %s,r14\n",regname[rs]);
+    output_w32(0xe1700000|rd_rn_rm(0,rs,HOST_TEMPREG));
+  }
+}
+
+void emit_cmovne(u_int *addr,int rt)
+{
+  assem_debug("cmovne %x,%%%s",(int)addr,regname[rt]);
+  assert(0);
+}
+void emit_cmovl(u_int *addr,int rt)
+{
+  assem_debug("cmovl %x,%%%s",(int)addr,regname[rt]);
+  assert(0);
+}
+void emit_cmovs(u_int *addr,int rt)
+{
+  assem_debug("cmovs %x,%%%s",(int)addr,regname[rt]);
+  assert(0);
+}
+void emit_cmovne_imm(int imm,int rt)
+{
+  assem_debug("movne %s,#%d\n",regname[rt],imm);
+  u_int armval;
+  assert(genimm(imm,&armval));
+  output_w32(0x13a00000|rd_rn_rm(rt,0,0)|armval);
+}
+void emit_cmovl_imm(int imm,int rt)
+{
+  assem_debug("movlt %s,#%d\n",regname[rt],imm);
+  u_int armval;
+  assert(genimm(imm,&armval));
+  output_w32(0xb3a00000|rd_rn_rm(rt,0,0)|armval);
+}
+void emit_cmovb_imm(int imm,int rt)
+{
+  assem_debug("movcc %s,#%d\n",regname[rt],imm);
+  u_int armval;
+  assert(genimm(imm,&armval));
+  output_w32(0x33a00000|rd_rn_rm(rt,0,0)|armval);
+}
+void emit_cmovs_imm(int imm,int rt)
+{
+  assem_debug("movmi %s,#%d\n",regname[rt],imm);
+  u_int armval;
+  assert(genimm(imm,&armval));
+  output_w32(0x43a00000|rd_rn_rm(rt,0,0)|armval);
+}
+void emit_cmove_reg(int rs,int rt)
+{
+  assem_debug("moveq %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0x01a00000|rd_rn_rm(rt,0,rs));
+}
+void emit_cmovne_reg(int rs,int rt)
+{
+  assem_debug("movne %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0x11a00000|rd_rn_rm(rt,0,rs));
+}
+void emit_cmovl_reg(int rs,int rt)
+{
+  assem_debug("movlt %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0xb1a00000|rd_rn_rm(rt,0,rs));
+}
+void emit_cmovs_reg(int rs,int rt)
+{
+  assem_debug("movmi %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0x41a00000|rd_rn_rm(rt,0,rs));
+}
+
+void emit_slti32(int rs,int imm,int rt)
+{
+  if(rs!=rt) emit_zeroreg(rt);
+  emit_cmpimm(rs,imm);
+  if(rs==rt) emit_movimm(0,rt);
+  emit_cmovl_imm(1,rt);
+}
+void emit_sltiu32(int rs,int imm,int rt)
+{
+  if(rs!=rt) emit_zeroreg(rt);
+  emit_cmpimm(rs,imm);
+  if(rs==rt) emit_movimm(0,rt);
+  emit_cmovb_imm(1,rt);
+}
+void emit_slti64_32(int rsh,int rsl,int imm,int rt)
+{
+  assert(rsh!=rt);
+  emit_slti32(rsl,imm,rt);
+  if(imm>=0)
+  {
+    emit_test(rsh,rsh);
+    emit_cmovne_imm(0,rt);
+    emit_cmovs_imm(1,rt);
+  }
+  else
+  {
+    emit_cmpimm(rsh,-1);
+    emit_cmovne_imm(0,rt);
+    emit_cmovl_imm(1,rt);
+  }
+}
+void emit_sltiu64_32(int rsh,int rsl,int imm,int rt)
+{
+  assert(rsh!=rt);
+  emit_sltiu32(rsl,imm,rt);
+  if(imm>=0)
+  {
+    emit_test(rsh,rsh);
+    emit_cmovne_imm(0,rt);
+  }
+  else
+  {
+    emit_cmpimm(rsh,-1);
+    emit_cmovne_imm(1,rt);
+  }
+}
+
+void emit_cmp(int rs,int rt)
+{
+  assem_debug("cmp %s,%s\n",regname[rs],regname[rt]);
+  output_w32(0xe1500000|rd_rn_rm(0,rs,rt));
+}
+void emit_set_gz32(int rs, int rt)
+{
+  //assem_debug("set_gz32\n");
+  emit_cmpimm(rs,1);
+  emit_movimm(1,rt);
+  emit_cmovl_imm(0,rt);
+}
+void emit_set_nz32(int rs, int rt)
+{
+  //assem_debug("set_nz32\n");
+  if(rs!=rt) emit_movs(rs,rt);
+  else emit_test(rs,rs);
+  emit_cmovne_imm(1,rt);
+}
+void emit_set_gz64_32(int rsh, int rsl, int rt)
+{
+  //assem_debug("set_gz64\n");
+  emit_set_gz32(rsl,rt);
+  emit_test(rsh,rsh);
+  emit_cmovne_imm(1,rt);
+  emit_cmovs_imm(0,rt);
+}
+void emit_set_nz64_32(int rsh, int rsl, int rt)
+{
+  //assem_debug("set_nz64\n");
+  emit_or_and_set_flags(rsh,rsl,rt);
+  emit_cmovne_imm(1,rt);
+}
+void emit_set_if_less32(int rs1, int rs2, int rt)
+{
+  //assem_debug("set if less (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+  if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt);
+  emit_cmp(rs1,rs2);
+  if(rs1==rt||rs2==rt) emit_movimm(0,rt);
+  emit_cmovl_imm(1,rt);
+}
+void emit_set_if_carry32(int rs1, int rs2, int rt)
+{
+  //assem_debug("set if carry (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+  if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt);
+  emit_cmp(rs1,rs2);
+  if(rs1==rt||rs2==rt) emit_movimm(0,rt);
+  emit_cmovb_imm(1,rt);
+}
+void emit_set_if_less64_32(int u1, int l1, int u2, int l2, int rt)
+{
+  //assem_debug("set if less64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]);
+  assert(u1!=rt);
+  assert(u2!=rt);
+  emit_cmp(l1,l2);
+  emit_movimm(0,rt);
+  emit_sbcs(u1,u2,HOST_TEMPREG);
+  emit_cmovl_imm(1,rt);
+}
+void emit_set_if_carry64_32(int u1, int l1, int u2, int l2, int rt)
+{
+  //assem_debug("set if carry64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]);
+  assert(u1!=rt);
+  assert(u2!=rt);
+  emit_cmp(l1,l2);
+  emit_movimm(0,rt);
+  emit_sbcs(u1,u2,HOST_TEMPREG);
+  emit_cmovb_imm(1,rt);
+}
+
+void emit_call(int a)
+{
+  assem_debug("bl %x (%x+%x)\n",a,(int)out,a-(int)out-8);
+  u_int offset=genjmp(a);
+  output_w32(0xeb000000|offset);
+}
+void emit_jmp(int a)
+{
+  assem_debug("b %x (%x+%x)\n",a,(int)out,a-(int)out-8);
+  u_int offset=genjmp(a);
+  output_w32(0xea000000|offset);
+}
+void emit_jne(int a)
+{
+  assem_debug("bne %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x1a000000|offset);
+}
+void emit_jeq(int a)
+{
+  assem_debug("beq %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x0a000000|offset);
+}
+void emit_js(int a)
+{
+  assem_debug("bmi %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x4a000000|offset);
+}
+void emit_jns(int a)
+{
+  assem_debug("bpl %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x5a000000|offset);
+}
+void emit_jl(int a)
+{
+  assem_debug("blt %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0xba000000|offset);
+}
+void emit_jge(int a)
+{
+  assem_debug("bge %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0xaa000000|offset);
+}
+void emit_jno(int a)
+{
+  assem_debug("bvc %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x7a000000|offset);
+}
+void emit_jc(int a)
+{
+  assem_debug("bcs %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x2a000000|offset);
+}
+void emit_jcc(int a)
+{
+  assem_debug("bcc %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x3a000000|offset);
+}
+
+void emit_pushimm(int imm)
+{
+  assem_debug("push $%x\n",imm);
+  assert(0);
+}
+void emit_pusha()
+{
+  assem_debug("pusha\n");
+  assert(0);
+}
+void emit_popa()
+{
+  assem_debug("popa\n");
+  assert(0);
+}
+void emit_pushreg(u_int r)
+{
+  assem_debug("push %%%s\n",regname[r]);
+  assert(0);
+}
+void emit_popreg(u_int r)
+{
+  assem_debug("pop %%%s\n",regname[r]);
+  assert(0);
+}
+void emit_callreg(u_int r)
+{
+  assem_debug("call *%%%s\n",regname[r]);
+  assert(0);
+}
+void emit_jmpreg(u_int r)
+{
+  assem_debug("mov pc,%s\n",regname[r]);
+  output_w32(0xe1a00000|rd_rn_rm(15,0,r));
+}
+
+void emit_readword_indexed(int offset, int rs, int rt)
+{
+  assert(offset>-4096&&offset<4096);
+  assem_debug("ldr %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe5900000|rd_rn_rm(rt,rs,0)|offset);
+  }else{
+    output_w32(0xe5100000|rd_rn_rm(rt,rs,0)|(-offset));
+  }
+}
+void emit_readword_dualindexedx4(int rs1, int rs2, int rt)
+{
+  assem_debug("ldr %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7900000|rd_rn_rm(rt,rs1,rs2)|0x100);
+}
+void emit_readword_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_readword_indexed(addr, rs, rt);
+  else {
+    assert(addr==0);
+    emit_readword_dualindexedx4(rs, map, rt);
+  }
+}
+void emit_readdword_indexed_tlb(int addr, int rs, int map, int rh, int rl)
+{
+  if(map<0) {
+    if(rh>=0) emit_readword_indexed(addr, rs, rh);
+    emit_readword_indexed(addr+4, rs, rl);
+  }else{
+    assert(rh!=rs);
+    if(rh>=0) emit_readword_indexed_tlb(addr, rs, map, rh);
+    emit_addimm(map,1,map);
+    emit_readword_indexed_tlb(addr, rs, map, rl);
+  }
+}
+void emit_movsbl_indexed(int offset, int rs, int rt)
+{
+  assert(offset>-256&&offset<256);
+  assem_debug("ldrsb %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe1d000d0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf));
+  }else{
+    output_w32(0xe15000d0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf));
+  }
+}
+void emit_movsbl_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_movsbl_indexed(addr, rs, rt);
+  else {
+    if(addr==0) {
+      emit_shlimm(map,2,map);
+      assem_debug("ldrsb %s,%s+%s\n",regname[rt],regname[rs],regname[map]);
+      output_w32(0xe19000d0|rd_rn_rm(rt,rs,map));
+    }else{
+      assert(addr>-256&&addr<256);
+      assem_debug("add %s,%s,%s,lsl #2\n",regname[rt],regname[rs],regname[map]);
+      output_w32(0xe0800000|rd_rn_rm(rt,rs,map)|(2<<7));
+      emit_movsbl_indexed(addr, rt, rt);
+    }
+  }
+}
+void emit_movswl_indexed(int offset, int rs, int rt)
+{
+  assert(offset>-256&&offset<256);
+  assem_debug("ldrsh %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe1d000f0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf));
+  }else{
+    output_w32(0xe15000f0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf));
+  }
+}
+void emit_movzbl_indexed(int offset, int rs, int rt)
+{
+  assert(offset>-4096&&offset<4096);
+  assem_debug("ldrb %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe5d00000|rd_rn_rm(rt,rs,0)|offset);
+  }else{
+    output_w32(0xe5500000|rd_rn_rm(rt,rs,0)|(-offset));
+  }
+}
+void emit_movzbl_dualindexedx4(int rs1, int rs2, int rt)
+{
+  assem_debug("ldrb %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7d00000|rd_rn_rm(rt,rs1,rs2)|0x100);
+}
+void emit_movzbl_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_movzbl_indexed(addr, rs, rt);
+  else {
+    if(addr==0) {
+      emit_movzbl_dualindexedx4(rs, map, rt);
+    }else{
+      emit_addimm(rs,addr,rt);
+      emit_movzbl_dualindexedx4(rt, map, rt);
+    }
+  }
+}
+void emit_movzwl_indexed(int offset, int rs, int rt)
+{
+  assert(offset>-256&&offset<256);
+  assem_debug("ldrh %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe1d000b0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf));
+  }else{
+    output_w32(0xe15000b0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf));
+  }
+}
+void emit_readword(int addr, int rt)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<4096);
+  assem_debug("ldr %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe5900000|rd_rn_rm(rt,FP,0)|offset);
+}
+void emit_movsbl(int addr, int rt)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<256);
+  assem_debug("ldrsb %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe1d000d0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf));
+}
+void emit_movswl(int addr, int rt)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<256);
+  assem_debug("ldrsh %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe1d000f0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf));
+}
+void emit_movzbl(int addr, int rt)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<4096);
+  assem_debug("ldrb %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe5d00000|rd_rn_rm(rt,FP,0)|offset);
+}
+void emit_movzwl(int addr, int rt)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<256);
+  assem_debug("ldrh %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe1d000b0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf));
+}
+void emit_movzwl_reg(int rs, int rt)
+{
+  assem_debug("movzwl %%%s,%%%s\n",regname[rs]+1,regname[rt]);
+  assert(0);
+}
+
+void emit_xchg(int rs, int rt)
+{
+  assem_debug("xchg %%%s,%%%s\n",regname[rs],regname[rt]);
+  assert(0);
+}
+void emit_writeword_indexed(int rt, int offset, int rs)
+{
+  assert(offset>-4096&&offset<4096);
+  assem_debug("str %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe5800000|rd_rn_rm(rt,rs,0)|offset);
+  }else{
+    output_w32(0xe5000000|rd_rn_rm(rt,rs,0)|(-offset));
+  }
+}
+void emit_writeword_dualindexedx4(int rt, int rs1, int rs2)
+{
+  assem_debug("str %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7800000|rd_rn_rm(rt,rs1,rs2)|0x100);
+}
+void emit_writeword_indexed_tlb(int rt, int addr, int rs, int map, int temp)
+{
+  if(map<0) emit_writeword_indexed(rt, addr, rs);
+  else {
+    assert(addr==0);
+    emit_writeword_dualindexedx4(rt, rs, map);
+  }
+}
+void emit_writedword_indexed_tlb(int rh, int rl, int addr, int rs, int map, int temp)
+{
+  if(map<0) {
+    if(rh>=0) emit_writeword_indexed(rh, addr, rs);
+    emit_writeword_indexed(rl, addr+4, rs);
+  }else{
+    assert(rh>=0);
+    if(temp!=rs) emit_addimm(map,1,temp);
+    emit_writeword_indexed_tlb(rh, addr, rs, map, temp);
+    if(temp!=rs) emit_writeword_indexed_tlb(rl, addr, rs, temp, temp);
+    else {
+      emit_addimm(rs,4,rs);
+      emit_writeword_indexed_tlb(rl, addr, rs, map, temp);
+    }
+  }
+}
+void emit_writehword_indexed(int rt, int offset, int rs)
+{
+  assert(offset>-256&&offset<256);
+  assem_debug("strh %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe1c000b0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf));
+  }else{
+    output_w32(0xe14000b0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf));
+  }
+}
+void emit_writebyte_indexed(int rt, int offset, int rs)
+{
+  assert(offset>-4096&&offset<4096);
+  assem_debug("strb %s,%s+%d\n",regname[rt],regname[rs],offset);
+  if(offset>=0) {
+    output_w32(0xe5c00000|rd_rn_rm(rt,rs,0)|offset);
+  }else{
+    output_w32(0xe5400000|rd_rn_rm(rt,rs,0)|(-offset));
+  }
+}
+void emit_writebyte_dualindexedx4(int rt, int rs1, int rs2)
+{
+  assem_debug("strb %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7c00000|rd_rn_rm(rt,rs1,rs2)|0x100);
+}
+void emit_writebyte_indexed_tlb(int rt, int addr, int rs, int map, int temp)
+{
+  if(map<0) emit_writebyte_indexed(rt, addr, rs);
+  else {
+    if(addr==0) {
+      emit_writebyte_dualindexedx4(rt, rs, map);
+    }else{
+      emit_addimm(rs,addr,temp);
+      emit_writebyte_dualindexedx4(rt, temp, map);
+    }
+  }
+}
+void emit_writeword(int rt, int addr)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<4096);
+  assem_debug("str %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe5800000|rd_rn_rm(rt,FP,0)|offset);
+}
+void emit_writehword(int rt, int addr)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<256);
+  assem_debug("strh %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe1c000b0|rd_rn_rm(rt,FP,0)|((offset<<4)&0xf00)|(offset&0xf));
+}
+void emit_writebyte(int rt, int addr)
+{
+  u_int offset = addr-(u_int)&dynarec_local;
+  assert(offset<4096);
+  assem_debug("str %s,fp+%d\n",regname[rt],offset);
+  output_w32(0xe5c00000|rd_rn_rm(rt,FP,0)|offset);
+}
+void emit_writeword_imm(int imm, int addr)
+{
+  assem_debug("movl $%x,%x\n",imm,addr);
+  assert(0);
+}
+void emit_writebyte_imm(int imm, int addr)
+{
+  assem_debug("movb $%x,%x\n",imm,addr);
+  assert(0);
+}
+
+void emit_mul(int rs)
+{
+  assem_debug("mul %%%s\n",regname[rs]);
+  assert(0);
+}
+void emit_imul(int rs)
+{
+  assem_debug("imul %%%s\n",regname[rs]);
+  assert(0);
+}
+void emit_umull(u_int rs1, u_int rs2, u_int hi, u_int lo)
+{
+  assem_debug("umull %s, %s, %s, %s\n",regname[lo],regname[hi],regname[rs1],regname[rs2]);
+  assert(rs1<16);
+  assert(rs2<16);
+  assert(hi<16);
+  assert(lo<16);
+  output_w32(0xe0800090|(hi<<16)|(lo<<12)|(rs2<<8)|rs1);
+}
+void emit_smull(u_int rs1, u_int rs2, u_int hi, u_int lo)
+{
+  assem_debug("smull %s, %s, %s, %s\n",regname[lo],regname[hi],regname[rs1],regname[rs2]);
+  assert(rs1<16);
+  assert(rs2<16);
+  assert(hi<16);
+  assert(lo<16);
+  output_w32(0xe0c00090|(hi<<16)|(lo<<12)|(rs2<<8)|rs1);
+}
+
+void emit_div(int rs)
+{
+  assem_debug("div %%%s\n",regname[rs]);
+  assert(0);
+}
+void emit_idiv(int rs)
+{
+  assem_debug("idiv %%%s\n",regname[rs]);
+  assert(0);
+}
+void emit_cdq()
+{
+  assem_debug("cdq\n");
+  assert(0);
+}
+
+void emit_clz(int rs,int rt)
+{
+  assem_debug("clz %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0xe16f0f10|rd_rn_rm(rt,0,rs));
+}
+
+void emit_subcs(int rs1,int rs2,int rt)
+{
+  assem_debug("subcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x20400000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_shrcc_imm(int rs,u_int imm,int rt)
+{
+  assert(imm>0);
+  assert(imm<32);
+  assem_debug("lsrcc %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x31a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7));
+}
+
+void emit_negmi(int rs, int rt)
+{
+  assem_debug("rsbmi %s,%s,#0\n",regname[rt],regname[rs]);
+  output_w32(0x42600000|rd_rn_rm(rt,rs,0));
+}
+
+void emit_negsmi(int rs, int rt)
+{
+  assem_debug("rsbsmi %s,%s,#0\n",regname[rt],regname[rs]);
+  output_w32(0x42700000|rd_rn_rm(rt,rs,0));
+}
+
+void emit_orreq(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("orreq %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x01800000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_orrne(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("orrne %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x11800000|rd_rn_rm(rt,rs1,rs2));
+}
+
+void emit_bic_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt)
+{
+  assem_debug("bic %s,%s,%s lsl %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]);
+  output_w32(0xe1C00000|rd_rn_rm(rt,rs1,rs2)|0x10|(shift<<8));
+}
+
+void emit_biceq_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt)
+{
+  assem_debug("biceq %s,%s,%s lsl %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]);
+  output_w32(0x01C00000|rd_rn_rm(rt,rs1,rs2)|0x10|(shift<<8));
+}
+
+void emit_bicne_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt)
+{
+  assem_debug("bicne %s,%s,%s lsl %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]);
+  output_w32(0x11C00000|rd_rn_rm(rt,rs1,rs2)|0x10|(shift<<8));
+}
+
+void emit_bic_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt)
+{
+  assem_debug("bic %s,%s,%s lsr %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]);
+  output_w32(0xe1C00000|rd_rn_rm(rt,rs1,rs2)|0x30|(shift<<8));
+}
+
+void emit_biceq_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt)
+{
+  assem_debug("biceq %s,%s,%s lsr %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]);
+  output_w32(0x01C00000|rd_rn_rm(rt,rs1,rs2)|0x30|(shift<<8));
+}
+
+void emit_bicne_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt)
+{
+  assem_debug("bicne %s,%s,%s lsr %s\n",regname[rt],regname[rs1],regname[rs2],regname[shift]);
+  output_w32(0x11C00000|rd_rn_rm(rt,rs1,rs2)|0x30|(shift<<8));
+}
+
+void emit_teq(int rs, int rt)
+{
+  assem_debug("teq %s,%s\n",regname[rs],regname[rt]);
+  output_w32(0xe1300000|rd_rn_rm(0,rs,rt));
+}
+
+void emit_rsbimm(int rs, int imm, int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("rsb %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0xe2600000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+// Load 2 immediates optimizing for small code size
+void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
+{
+  emit_movimm(imm1,rt1);
+  u_int armval;
+  if(genimm(imm2-imm1,&armval)) {
+    assem_debug("add %s,%s,#%d\n",regname[rt2],regname[rt1],imm2-imm1);
+    output_w32(0xe2800000|rd_rn_rm(rt2,rt1,0)|armval);
+  }else if(genimm(imm1-imm2,&armval)) {
+    assem_debug("sub %s,%s,#%d\n",regname[rt2],regname[rt1],imm1-imm2);
+    output_w32(0xe2400000|rd_rn_rm(rt2,rt1,0)|armval);
+  }
+  else emit_movimm(imm2,rt2);
+}
+
+// Conditionally select one of two immediates, optimizing for small code size
+// This will only be called if HAVE_CMOV_IMM is defined
+void emit_cmov2imm_e_ne_compact(int imm1,int imm2,u_int rt)
+{
+  u_int armval;
+  if(genimm(imm2-imm1,&armval)) {
+    emit_movimm(imm1,rt);
+    assem_debug("addne %s,%s,#%d\n",regname[rt],regname[rt],imm2-imm1);
+    output_w32(0x12800000|rd_rn_rm(rt,rt,0)|armval);
+  }else if(genimm(imm1-imm2,&armval)) {
+    emit_movimm(imm1,rt);
+    assem_debug("subne %s,%s,#%d\n",regname[rt],regname[rt],imm1-imm2);
+    output_w32(0x12400000|rd_rn_rm(rt,rt,0)|armval);
+  }
+  else {
+    #ifdef ARMv5_ONLY
+    emit_movimm(imm1,rt);
+    add_literal((int)out,imm2);
+    assem_debug("ldrne %s,pc+? [=%x]\n",regname[rt],imm2);
+    output_w32(0x15900000|rd_rn_rm(rt,15,0));
+    #else
+    emit_movw(imm1&0x0000FFFF,rt);
+    if((imm1&0xFFFF)!=(imm2&0xFFFF)) {
+      assem_debug("movwne %s,#%d (0x%x)\n",regname[rt],imm2&0xFFFF,imm2&0xFFFF);
+      output_w32(0x13000000|rd_rn_rm(rt,0,0)|(imm2&0xfff)|((imm2<<4)&0xf0000));
+    }
+    emit_movt(imm1&0xFFFF0000,rt);
+    if((imm1&0xFFFF0000)!=(imm2&0xFFFF0000)) {
+      assem_debug("movtne %s,#%d (0x%x)\n",regname[rt],imm2&0xffff0000,imm2&0xffff0000);
+      output_w32(0x13400000|rd_rn_rm(rt,0,0)|((imm2>>16)&0xfff)|((imm2>>12)&0xf0000));
+    }
+    #endif
+  }
+}
+
+// special case for checking invalid_code
+void emit_cmpmem_indexedsr12_imm(int addr,int r,int imm)
+{
+  assert(0);
+}
+
+// special case for checking invalid_code
+void emit_cmpmem_indexedsr12_reg(int base,int r,int imm)
+{
+  assert(imm<128&&imm>=0);
+  assert(r>=0&&r<16);
+  assem_debug("ldrb lr,%s,%s lsr #12\n",regname[base],regname[r]);
+  output_w32(0xe7d00000|rd_rn_rm(HOST_TEMPREG,base,r)|0x620);
+  emit_cmpimm(HOST_TEMPREG,imm);
+}
+
+// special case for tlb mapping
+void emit_addsr12(int rs1,int rs2,int rt)
+{
+  assem_debug("add %s,%s,%s lsr #12\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe0800620|rd_rn_rm(rt,rs1,rs2));
+}
+
+// Used to preload hash table entries
+void emit_prefetch(void *addr)
+{
+  assem_debug("prefetch %x\n",(int)addr);
+  output_byte(0x0F);
+  output_byte(0x18);
+  output_modrm(0,5,1);
+  output_w32((int)addr);
+}
+void emit_prefetchreg(int r)
+{
+  assem_debug("pld %s\n",regname[r]);
+  output_w32(0xf5d0f000|rd_rn_rm(0,r,0));
+}
+
+// Special case for mini_ht
+void emit_ldreq_indexed(int rs, u_int offset, int rt)
+{
+  assert(offset<4096);
+  assem_debug("ldreq %s,[%s, #%d]\n",regname[rt],regname[rs],offset);
+  output_w32(0x05900000|rd_rn_rm(rt,rs,0)|offset);
+}
+
+void emit_flds(int r,int sr)
+{
+  assem_debug("flds s%d,[%s]\n",sr,regname[r]);
+  output_w32(0xed900a00|((sr&14)<<11)|((sr&1)<<22)|(r<<16));
+} 
+
+void emit_vldr(int r,int vr)
+{
+  assem_debug("vldr d%d,[%s]\n",vr,regname[r]);
+  output_w32(0xed900b00|(vr<<12)|(r<<16));
+} 
+
+void emit_fsts(int sr,int r)
+{
+  assem_debug("fsts s%d,[%s]\n",sr,regname[r]);
+  output_w32(0xed800a00|((sr&14)<<11)|((sr&1)<<22)|(r<<16));
+} 
+
+void emit_vstr(int vr,int r)
+{
+  assem_debug("vstr d%d,[%s]\n",vr,regname[r]);
+  output_w32(0xed800b00|(vr<<12)|(r<<16));
+} 
+
+void emit_ftosizs(int s,int d)
+{
+  assem_debug("ftosizs s%d,s%d\n",d,s);
+  output_w32(0xeebd0ac0|((d&14)<<11)|((d&1)<<22)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_ftosizd(int s,int d)
+{
+  assem_debug("ftosizd s%d,d%d\n",d,s);
+  output_w32(0xeebd0bc0|((d&14)<<11)|((d&1)<<22)|(s&7));
+} 
+
+void emit_fsitos(int s,int d)
+{
+  assem_debug("fsitos s%d,s%d\n",d,s);
+  output_w32(0xeeb80ac0|((d&14)<<11)|((d&1)<<22)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_fsitod(int s,int d)
+{
+  assem_debug("fsitod d%d,s%d\n",d,s);
+  output_w32(0xeeb80bc0|((d&7)<<12)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_fcvtds(int s,int d)
+{
+  assem_debug("fcvtds d%d,s%d\n",d,s);
+  output_w32(0xeeb70ac0|((d&7)<<12)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_fcvtsd(int s,int d)
+{
+  assem_debug("fcvtsd s%d,d%d\n",d,s);
+  output_w32(0xeeb70bc0|((d&14)<<11)|((d&1)<<22)|(s&7));
+} 
+
+void emit_fsqrts(int s,int d)
+{
+  assem_debug("fsqrts d%d,s%d\n",d,s);
+  output_w32(0xeeb10ac0|((d&14)<<11)|((d&1)<<22)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_fsqrtd(int s,int d)
+{
+  assem_debug("fsqrtd s%d,d%d\n",d,s);
+  output_w32(0xeeb10bc0|((d&7)<<12)|(s&7));
+} 
+
+void emit_fabss(int s,int d)
+{
+  assem_debug("fabss d%d,s%d\n",d,s);
+  output_w32(0xeeb00ac0|((d&14)<<11)|((d&1)<<22)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_fabsd(int s,int d)
+{
+  assem_debug("fabsd s%d,d%d\n",d,s);
+  output_w32(0xeeb00bc0|((d&7)<<12)|(s&7));
+} 
+
+void emit_fnegs(int s,int d)
+{
+  assem_debug("fnegs d%d,s%d\n",d,s);
+  output_w32(0xeeb10a40|((d&14)<<11)|((d&1)<<22)|((s&14)>>1)|((s&1)<<5));
+} 
+
+void emit_fnegd(int s,int d)
+{
+  assem_debug("fnegd s%d,d%d\n",d,s);
+  output_w32(0xeeb10b40|((d&7)<<12)|(s&7));
+} 
+
+void emit_fadds(int s1,int s2,int d)
+{
+  assem_debug("fadds s%d,s%d,s%d\n",d,s1,s2);
+  output_w32(0xee300a00|((d&14)<<11)|((d&1)<<22)|((s1&14)<<15)|((s1&1)<<7)|((s2&14)>>1)|((s2&1)<<5));
+} 
+
+void emit_faddd(int s1,int s2,int d)
+{
+  assem_debug("faddd d%d,d%d,d%d\n",d,s1,s2);
+  output_w32(0xee300b00|((d&7)<<12)|((s1&7)<<16)|(s2&7));
+} 
+
+void emit_fsubs(int s1,int s2,int d)
+{
+  assem_debug("fsubs s%d,s%d,s%d\n",d,s1,s2);
+  output_w32(0xee300a40|((d&14)<<11)|((d&1)<<22)|((s1&14)<<15)|((s1&1)<<7)|((s2&14)>>1)|((s2&1)<<5));
+} 
+
+void emit_fsubd(int s1,int s2,int d)
+{
+  assem_debug("fsubd d%d,d%d,d%d\n",d,s1,s2);
+  output_w32(0xee300b40|((d&7)<<12)|((s1&7)<<16)|(s2&7));
+} 
+
+void emit_fmuls(int s1,int s2,int d)
+{
+  assem_debug("fmuls s%d,s%d,s%d\n",d,s1,s2);
+  output_w32(0xee200a00|((d&14)<<11)|((d&1)<<22)|((s1&14)<<15)|((s1&1)<<7)|((s2&14)>>1)|((s2&1)<<5));
+} 
+
+void emit_fmuld(int s1,int s2,int d)
+{
+  assem_debug("fmuld d%d,d%d,d%d\n",d,s1,s2);
+  output_w32(0xee200b00|((d&7)<<12)|((s1&7)<<16)|(s2&7));
+} 
+
+void emit_fdivs(int s1,int s2,int d)
+{
+  assem_debug("fdivs s%d,s%d,s%d\n",d,s1,s2);
+  output_w32(0xee800a00|((d&14)<<11)|((d&1)<<22)|((s1&14)<<15)|((s1&1)<<7)|((s2&14)>>1)|((s2&1)<<5));
+} 
+
+void emit_fdivd(int s1,int s2,int d)
+{
+  assem_debug("fdivd d%d,d%d,d%d\n",d,s1,s2);
+  output_w32(0xee800b00|((d&7)<<12)|((s1&7)<<16)|(s2&7));
+} 
+
+void emit_fcmps(int x,int y)
+{
+  assem_debug("fcmps s14, s15\n");
+  output_w32(0xeeb47a67);
+} 
+
+void emit_fcmpd(int x,int y)
+{
+  assem_debug("fcmpd d6, d7\n");
+  output_w32(0xeeb46b47);
+} 
+
+void emit_fmstat()
+{
+  assem_debug("fmstat\n");
+  output_w32(0xeef1fa10);
+} 
+
+void emit_bicne_imm(int rs,int imm,int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("bicne %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x13c00000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+void emit_biccs_imm(int rs,int imm,int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("biccs %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x23c00000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+void emit_bicvc_imm(int rs,int imm,int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("bicvc %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x73c00000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+void emit_bichi_imm(int rs,int imm,int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("bichi %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x83c00000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+void emit_orrvs_imm(int rs,int imm,int rt)
+{
+  u_int armval;
+  assert(genimm(imm,&armval));
+  assem_debug("orrvs %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x63800000|rd_rn_rm(rt,rs,0)|armval);
+}
+
+void emit_jno_unlikely(int a)
+{
+  //emit_jno(a);
+  assem_debug("addvc pc,pc,#? (%x)\n",/*a-(int)out-8,*/a);
+  output_w32(0x72800000|rd_rn_rm(15,15,0));
+}
+
+// Save registers before function call
+void save_regs(u_int reglist)
+{
+  reglist&=0x100f; // only save the caller-save registers, r0-r3, r12
+  if(!reglist) return;
+  assem_debug("stmia fp,{");
+  if(reglist&1) assem_debug("r0, ");
+  if(reglist&2) assem_debug("r1, ");
+  if(reglist&4) assem_debug("r2, ");
+  if(reglist&8) assem_debug("r3, ");
+  if(reglist&0x1000) assem_debug("r12");
+  assem_debug("}\n");
+  output_w32(0xe88b0000|reglist);
+}
+// Restore registers after function call
+void restore_regs(u_int reglist)
+{
+  reglist&=0x100f; // only restore the caller-save registers, r0-r3, r12
+  if(!reglist) return;
+  assem_debug("ldmia fp,{");
+  if(reglist&1) assem_debug("r0, ");
+  if(reglist&2) assem_debug("r1, ");
+  if(reglist&4) assem_debug("r2, ");
+  if(reglist&8) assem_debug("r3, ");
+  if(reglist&0x1000) assem_debug("r12");
+  assem_debug("}\n");
+  output_w32(0xe89b0000|reglist);
+}
+
+// Write back consts using r14 so we don't disturb the other registers
+void wb_consts(signed char i_regmap[],uint64_t i_is32,u_int i_dirty,int i)
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&i_regmap[hr]>=0&&((i_dirty>>hr)&1)) {
+      if(((regs[i].isconst>>hr)&1)&&i_regmap[hr]>0) {
+        if(i_regmap[hr]<64 || !((i_is32>>(i_regmap[hr]&63))&1) ) {
+          int value=constmap[i][hr];
+          if(value==0) {
+            emit_zeroreg(HOST_TEMPREG);
+          }
+          else {
+            emit_movimm(value,HOST_TEMPREG);
+          }
+          emit_storereg(i_regmap[hr],HOST_TEMPREG);
+          if((i_is32>>i_regmap[hr])&1) {
+            if(value!=-1&&value!=0) emit_sarimm(HOST_TEMPREG,31,HOST_TEMPREG);
+            emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
+          }
+        }
+      }
+    }
+  }
+}
+
+/* Stubs/epilogue */
+
+void literal_pool(int n)
+{
+  if(!literalcount) return;
+  if(n) {
+    if((int)out-literals[0][0]<4096-n) return;
+  }
+  u_int *ptr;
+  int i;
+  for(i=0;i<literalcount;i++)
+  {
+    ptr=(u_int *)literals[i][0];
+    u_int offset=(u_int)out-(u_int)ptr-8;
+    assert(offset<4096);
+    assert(!(offset&3));
+    *ptr|=offset;
+    output_w32(literals[i][1]);
+  }
+  literalcount=0;
+}
+
+void literal_pool_jumpover(int n)
+{
+  if(!literalcount) return;
+  if(n) {
+    if((int)out-literals[0][0]<4096-n) return;
+  }
+  int jaddr=(int)out;
+  emit_jmp(0);
+  literal_pool(0);
+  set_jump_target(jaddr,(int)out);
+}
+
+emit_extjump2(int addr, int target, int linker)
+{
+  u_char *ptr=(u_char *)addr;
+  assert((ptr[3]&0x0e)==0xa);
+  emit_loadlp(target,0);
+  emit_loadlp(addr,1);
+  assert(addr>=0x7000000&&addr<0x7FFFFFF);
+  //assert((target>=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000));
+//DEBUG >
+#ifdef DEBUG_CYCLE_COUNT
+  emit_readword((int)&last_count,ECX);
+  emit_add(HOST_CCREG,ECX,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_writeword(HOST_CCREG,(int)&Count);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+#endif
+//DEBUG <
+  emit_jmp(linker);
+}
+
+emit_extjump(int addr, int target)
+{
+  emit_extjump2(addr, target, (int)dyna_linker);
+}
+emit_extjump_ds(int addr, int target)
+{
+  emit_extjump2(addr, target, (int)dyna_linker_ds);
+}
+
+do_readstub(int n)
+{
+  assem_debug("do_readstub %x\n",start+stubs[n][3]*4);
+  literal_pool(256);
+  set_jump_target(stubs[n][1],(int)out);
+  int type=stubs[n][0];
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  u_int reglist=stubs[n][7];
+  signed char *i_regmap=i_regs->regmap;
+  int addr=get_reg(i_regmap,AGEN1+(i&1));
+  int rth,rt;
+  int ds;
+  if(itype[i]==C1LS||itype[i]==LOADLR) {
+    rth=get_reg(i_regmap,FTEMP|64);
+    rt=get_reg(i_regmap,FTEMP);
+  }else{
+    rth=get_reg(i_regmap,rt1[i]|64);
+    rt=get_reg(i_regmap,rt1[i]);
+  }
+  assert(rs>=0);
+  assert(rt>=0);
+  if(addr<0) addr=rt;
+  assert(addr>=0);
+  int ftable=0;
+  if(type==LOADB_STUB||type==LOADBU_STUB)
+    ftable=(int)readmemb;
+  if(type==LOADH_STUB||type==LOADHU_STUB)
+    ftable=(int)readmemh;
+  if(type==LOADW_STUB)
+    ftable=(int)readmem;
+  if(type==LOADD_STUB)
+    ftable=(int)readmemd;
+  emit_writeword(rs,(int)&address);
+  //emit_pusha();
+  save_regs(reglist);
+  ds=i_regs!=&regs[i];
+  int real_rs=(itype[i]==LOADLR)?-1:get_reg(i_regmap,rs1[i]);
+  u_int cmask=ds?-1:(0x100f|~i_regs->wasconst);
+  if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs))&0x100f,i);
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty&cmask&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)));
+  if(!ds) wb_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs))&~0x100f,i);
+  emit_shrimm(rs,16,1);
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    emit_loadreg(CCREG,2);
+  }
+  emit_movimm(ftable,0);
+  emit_addimm(cc<0?2:cc,2*stubs[n][6]+2,2);
+  emit_movimm(start+stubs[n][3]*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3);
+  //emit_readword((int)&last_count,temp);
+  //emit_add(cc,temp,cc);
+  //emit_writeword(cc,(int)&Count);
+  //emit_mov(15,14);
+  emit_call((int)&indirect_jump_indexed);
+  //emit_callreg(rs);
+  //emit_readword_dualindexedx4(rs,HOST_TEMPREG,15);
+  // We really shouldn't need to update the count here,
+  // but not doing so causes random crashes...
+  emit_readword((int)&Count,HOST_TEMPREG);
+  emit_readword((int)&next_interupt,2);
+  emit_addimm(HOST_TEMPREG,-2*stubs[n][6]-2,HOST_TEMPREG);
+  emit_writeword(2,(int)&last_count);
+  emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc);
+  if(cc<0) {
+    emit_storereg(CCREG,HOST_TEMPREG);
+  }
+  //emit_popa();
+  restore_regs(reglist);
+  //if((cc=get_reg(regmap,CCREG))>=0) {
+  //  emit_loadreg(CCREG,cc);
+  //}
+  if(type==LOADB_STUB)
+    emit_movsbl((int)&readmem_dword,rt);
+  if(type==LOADBU_STUB)
+    emit_movzbl((int)&readmem_dword,rt);
+  if(type==LOADH_STUB)
+    emit_movswl((int)&readmem_dword,rt);
+  if(type==LOADHU_STUB)
+    emit_movzwl((int)&readmem_dword,rt);
+  if(type==LOADW_STUB)
+    emit_readword((int)&readmem_dword,rt);
+  if(type==LOADD_STUB) {
+    emit_readword((int)&readmem_dword,rt);
+    if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth);
+  }
+  emit_jmp(stubs[n][2]); // return address
+}
+
+inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+{
+  int rs=get_reg(regmap,target);
+  int rth=get_reg(regmap,target|64);
+  int rt=get_reg(regmap,target);
+  assert(rs>=0);
+  assert(rt>=0);
+  int ftable=0;
+  if(type==LOADB_STUB||type==LOADBU_STUB)
+    ftable=(int)readmemb;
+  if(type==LOADH_STUB||type==LOADHU_STUB)
+    ftable=(int)readmemh;
+  if(type==LOADW_STUB)
+    ftable=(int)readmem;
+  if(type==LOADD_STUB)
+    ftable=(int)readmemd;
+  emit_writeword(rs,(int)&address);
+  //emit_pusha();
+  save_regs(reglist);
+  //emit_shrimm(rs,16,1);
+  int cc=get_reg(regmap,CCREG);
+  if(cc<0) {
+    emit_loadreg(CCREG,2);
+  }
+  //emit_movimm(ftable,0);
+  emit_movimm(((u_int *)ftable)[addr>>16],0);
+  //emit_readword((int)&last_count,12);
+  emit_addimm(cc<0?2:cc,CLOCK_DIVIDER*(adj+1),2);
+  if((signed int)addr>=(signed int)0xC0000000) {
+    // Pagefault address
+    int ds=regmap!=regs[i].regmap;
+    emit_movimm(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3);
+  }
+  //emit_add(12,2,2);
+  //emit_writeword(2,(int)&Count);
+  //emit_call(((u_int *)ftable)[addr>>16]);
+  emit_call((int)&indirect_jump);
+  // We really shouldn't need to update the count here,
+  // but not doing so causes random crashes...
+  emit_readword((int)&Count,HOST_TEMPREG);
+  emit_readword((int)&next_interupt,2);
+  emit_addimm(HOST_TEMPREG,-CLOCK_DIVIDER*(adj+1),HOST_TEMPREG);
+  emit_writeword(2,(int)&last_count);
+  emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc);
+  if(cc<0) {
+    emit_storereg(CCREG,HOST_TEMPREG);
+  }
+  //emit_popa();
+  restore_regs(reglist);
+  if(type==LOADB_STUB)
+    emit_movsbl((int)&readmem_dword,rt);
+  if(type==LOADBU_STUB)
+    emit_movzbl((int)&readmem_dword,rt);
+  if(type==LOADH_STUB)
+    emit_movswl((int)&readmem_dword,rt);
+  if(type==LOADHU_STUB)
+    emit_movzwl((int)&readmem_dword,rt);
+  if(type==LOADW_STUB)
+    emit_readword((int)&readmem_dword,rt);
+  if(type==LOADD_STUB) {
+    emit_readword((int)&readmem_dword,rt);
+    if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth);
+  }
+}
+
+do_writestub(int n)
+{
+  assem_debug("do_writestub %x\n",start+stubs[n][3]*4);
+  literal_pool(256);
+  set_jump_target(stubs[n][1],(int)out);
+  int type=stubs[n][0];
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  u_int reglist=stubs[n][7];
+  signed char *i_regmap=i_regs->regmap;
+  int addr=get_reg(i_regmap,AGEN1+(i&1));
+  int rth,rt,r;
+  int ds;
+  if(itype[i]==C1LS) {
+    rth=get_reg(i_regmap,FTEMP|64);
+    rt=get_reg(i_regmap,r=FTEMP);
+  }else{
+    rth=get_reg(i_regmap,rs2[i]|64);
+    rt=get_reg(i_regmap,r=rs2[i]);
+  }
+  assert(rs>=0);
+  assert(rt>=0);
+  if(addr<0) addr=get_reg(i_regmap,-1);
+  assert(addr>=0);
+  int ftable=0;
+  if(type==STOREB_STUB)
+    ftable=(int)writememb;
+  if(type==STOREH_STUB)
+    ftable=(int)writememh;
+  if(type==STOREW_STUB)
+    ftable=(int)writemem;
+  if(type==STORED_STUB)
+    ftable=(int)writememd;
+  emit_writeword(rs,(int)&address);
+  //emit_shrimm(rs,16,rs);
+  //emit_movmem_indexedx4(ftable,rs,rs);
+  if(type==STOREB_STUB)
+    emit_writebyte(rt,(int)&byte);
+  if(type==STOREH_STUB)
+    emit_writehword(rt,(int)&hword);
+  if(type==STOREW_STUB)
+    emit_writeword(rt,(int)&word);
+  if(type==STORED_STUB) {
+    emit_writeword(rt,(int)&dword);
+    emit_writeword(r?rth:rt,(int)&dword+4);
+  }
+  //emit_pusha();
+  save_regs(reglist);
+  ds=i_regs!=&regs[i];
+  int real_rs=get_reg(i_regmap,rs1[i]);
+  u_int cmask=ds?-1:(0x100f|~i_regs->wasconst);
+  if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs))&0x100f,i);
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty&cmask&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)));
+  if(!ds) wb_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs))&~0x100f,i);
+  emit_shrimm(rs,16,1);
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    emit_loadreg(CCREG,2);
+  }
+  emit_movimm(ftable,0);
+  emit_addimm(cc<0?2:cc,2*stubs[n][6]+2,2);
+  emit_movimm(start+stubs[n][3]*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3);
+  //emit_readword((int)&last_count,temp);
+  //emit_addimm(cc,2*stubs[n][5]+2,cc);
+  //emit_add(cc,temp,cc);
+  //emit_writeword(cc,(int)&Count);
+  emit_call((int)&indirect_jump_indexed);
+  //emit_callreg(rs);
+  emit_readword((int)&Count,HOST_TEMPREG);
+  emit_readword((int)&next_interupt,2);
+  emit_addimm(HOST_TEMPREG,-2*stubs[n][6]-2,HOST_TEMPREG);
+  emit_writeword(2,(int)&last_count);
+  emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc);
+  if(cc<0) {
+    emit_storereg(CCREG,HOST_TEMPREG);
+  }
+  //emit_popa();
+  restore_regs(reglist);
+  //if((cc=get_reg(regmap,CCREG))>=0) {
+  //  emit_loadreg(CCREG,cc);
+  //}
+  emit_jmp(stubs[n][2]); // return address
+}
+
+inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+{
+  int rs=get_reg(regmap,-1);
+  int rth=get_reg(regmap,target|64);
+  int rt=get_reg(regmap,target);
+  assert(rs>=0);
+  assert(rt>=0);
+  int ftable=0;
+  if(type==STOREB_STUB)
+    ftable=(int)writememb;
+  if(type==STOREH_STUB)
+    ftable=(int)writememh;
+  if(type==STOREW_STUB)
+    ftable=(int)writemem;
+  if(type==STORED_STUB)
+    ftable=(int)writememd;
+  emit_writeword(rs,(int)&address);
+  //emit_shrimm(rs,16,rs);
+  //emit_movmem_indexedx4(ftable,rs,rs);
+  if(type==STOREB_STUB)
+    emit_writebyte(rt,(int)&byte);
+  if(type==STOREH_STUB)
+    emit_writehword(rt,(int)&hword);
+  if(type==STOREW_STUB)
+    emit_writeword(rt,(int)&word);
+  if(type==STORED_STUB) {
+    emit_writeword(rt,(int)&dword);
+    emit_writeword(target?rth:rt,(int)&dword+4);
+  }
+  //emit_pusha();
+  save_regs(reglist);
+  //emit_shrimm(rs,16,1);
+  int cc=get_reg(regmap,CCREG);
+  if(cc<0) {
+    emit_loadreg(CCREG,2);
+  }
+  //emit_movimm(ftable,0);
+  emit_movimm(((u_int *)ftable)[addr>>16],0);
+  //emit_readword((int)&last_count,12);
+  emit_addimm(cc<0?2:cc,CLOCK_DIVIDER*(adj+1),2);
+  if((signed int)addr>=(signed int)0xC0000000) {
+    // Pagefault address
+    int ds=regmap!=regs[i].regmap;
+    emit_movimm(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3);
+  }
+  //emit_add(12,2,2);
+  //emit_writeword(2,(int)&Count);
+  //emit_call(((u_int *)ftable)[addr>>16]);
+  emit_call((int)&indirect_jump);
+  emit_readword((int)&Count,HOST_TEMPREG);
+  emit_readword((int)&next_interupt,2);
+  emit_addimm(HOST_TEMPREG,-CLOCK_DIVIDER*(adj+1),HOST_TEMPREG);
+  emit_writeword(2,(int)&last_count);
+  emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc);
+  if(cc<0) {
+    emit_storereg(CCREG,HOST_TEMPREG);
+  }
+  //emit_popa();
+  restore_regs(reglist);
+}
+
+do_unalignedwritestub(int n)
+{
+  set_jump_target(stubs[n][1],(int)out);
+  output_w32(0xef000000);
+  emit_jmp(stubs[n][2]); // return address
+}
+
+void printregs(int edi,int esi,int ebp,int esp,int b,int d,int c,int a)
+{
+  printf("regs: %x %x %x %x %x %x %x (%x)\n",a,b,c,d,ebp,esi,edi,(&edi)[-1]);
+}
+
+do_invstub(int n)
+{
+  literal_pool(20);
+  u_int reglist=stubs[n][3];
+  set_jump_target(stubs[n][1],(int)out);
+  save_regs(reglist);
+  if(stubs[n][4]!=0) emit_mov(stubs[n][4],0);
+  emit_call((int)&invalidate_addr);
+  restore_regs(reglist);
+  emit_jmp(stubs[n][2]); // return address
+}
+
+int do_dirty_stub(int i)
+{
+  assem_debug("do_dirty_stub %x\n",start+i*4);
+  // Careful about the code output here, verify_dirty needs to parse it.
+  #ifdef ARMv5_ONLY
+  emit_loadlp((int)start<(int)0xC0000000?(int)source:(int)start,1);
+  emit_loadlp((int)copy,2);
+  emit_loadlp(slen*4,3);
+  #else
+  emit_movw(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0x0000FFFF,1);
+  emit_movw(((u_int)copy)&0x0000FFFF,2);
+  emit_movt(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0xFFFF0000,1);
+  emit_movt(((u_int)copy)&0xFFFF0000,2);
+  emit_movw(slen*4,3);
+  #endif
+  emit_movimm(start+i*4,0);
+  emit_call((int)start<(int)0xC0000000?(int)&verify_code:(int)&verify_code_vm);
+  int entry=(int)out;
+  load_regs_entry(i);
+  if(entry==(int)out) entry=instr_addr[i];
+  emit_jmp(instr_addr[i]);
+  return entry;
+}
+
+void do_dirty_stub_ds()
+{
+  // Careful about the code output here, verify_dirty needs to parse it.
+  #ifdef ARMv5_ONLY
+  emit_loadlp((int)start<(int)0xC0000000?(int)source:(int)start,1);
+  emit_loadlp((int)copy,2);
+  emit_loadlp(slen*4,3);
+  #else
+  emit_movw(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0x0000FFFF,1);
+  emit_movw(((u_int)copy)&0x0000FFFF,2);
+  emit_movt(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0xFFFF0000,1);
+  emit_movt(((u_int)copy)&0xFFFF0000,2);
+  emit_movw(slen*4,3);
+  #endif
+  emit_movimm(start+1,0);
+  emit_call((int)&verify_code_ds);
+}
+
+do_cop1stub(int n)
+{
+  literal_pool(256);
+  assem_debug("do_cop1stub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  int ds=stubs[n][6];
+  if(!ds) {
+    load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i);
+    //if(i_regs!=&regs[i]) printf("oops: regs[i]=%x i_regs=%x",(int)&regs[i],(int)i_regs);
+  }
+  //else {printf("fp exception in delay slot\n");}
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty);
+  if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+  emit_movimm(start+(i-ds)*4,EAX); // Get PC
+  emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
+  emit_jmp(ds?(int)fp_exception_ds:(int)fp_exception);
+}
+
+/* TLB */
+
+int do_tlb_r(int s,int ar,int map,int x,int a,int shift,int c,u_int addr)
+{
+  if(c) {
+    if((signed int)addr>=(signed int)0xC0000000) {
+      // address_generation already loaded the const
+      emit_readword_dualindexedx4(FP,map,map);
+    }
+    else
+      return -1; // No mapping
+  }
+  else {
+    assert(s!=map);
+    emit_movimm(((int)memory_map-(int)&dynarec_local)>>2,map);
+    emit_addsr12(map,s,map);
+    // Schedule this while we wait on the load
+    //if(x) emit_xorimm(s,x,ar);
+    if(shift>=0) emit_shlimm(s,3,shift);
+    if(~a) emit_andimm(s,a,ar);
+    emit_readword_dualindexedx4(FP,map,map);
+  }
+  return map;
+}
+int do_tlb_r_branch(int map, int c, u_int addr, int *jaddr)
+{
+  if(!c||(signed int)addr>=(signed int)0xC0000000) {
+    emit_test(map,map);
+    *jaddr=(int)out;
+    emit_js(0);
+  }
+  return map;
+}
+
+int gen_tlb_addr_r(int ar, int map) {
+  if(map>=0) {
+    assem_debug("add %s,%s,%s lsl #2\n",regname[ar],regname[ar],regname[map]);
+    output_w32(0xe0800100|rd_rn_rm(ar,ar,map));
+  }
+}
+
+int do_tlb_w(int s,int ar,int map,int x,int c,u_int addr)
+{
+  if(c) {
+    if(addr<0x80800000||addr>=0xC0000000) {
+      // address_generation already loaded the const
+      emit_readword_dualindexedx4(FP,map,map);
+    }
+    else
+      return -1; // No mapping
+  }
+  else {
+    assert(s!=map);
+    emit_movimm(((int)memory_map-(int)&dynarec_local)>>2,map);
+    emit_addsr12(map,s,map);
+    // Schedule this while we wait on the load
+    //if(x) emit_xorimm(s,x,ar);
+    emit_readword_dualindexedx4(FP,map,map);
+  }
+  return map;
+}
+int do_tlb_w_branch(int map, int c, u_int addr, int *jaddr)
+{
+  if(!c||addr<0x80800000||addr>=0xC0000000) {
+    emit_testimm(map,0x40000000);
+    *jaddr=(int)out;
+    emit_jne(0);
+  }
+}
+
+int gen_tlb_addr_w(int ar, int map) {
+  if(map>=0) {
+    assem_debug("add %s,%s,%s lsl #2\n",regname[ar],regname[ar],regname[map]);
+    output_w32(0xe0800100|rd_rn_rm(ar,ar,map));
+  }
+}
+
+// Generate the address of the memory_map entry, relative to dynarec_local
+generate_map_const(u_int addr,int reg) {
+  //printf("generate_map_const(%x,%s)\n",addr,regname[reg]);
+  emit_movimm((addr>>12)+(((u_int)memory_map-(u_int)&dynarec_local)>>2),reg);
+}
+
+/* Special assem */
+
+void shift_assemble_arm(int i,struct regstat *i_regs)
+{
+  if(rt1[i]) {
+    if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV
+    {
+      signed char s,t,shift;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      s=get_reg(i_regs->regmap,rs1[i]);
+      shift=get_reg(i_regs->regmap,rs2[i]);
+      if(t>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(t);
+        }
+        else if(rs2[i]==0)
+        {
+          assert(s>=0);
+          if(s!=t) emit_mov(s,t);
+        }
+        else
+        {
+          emit_andimm(shift,31,HOST_TEMPREG);
+          if(opcode2[i]==4) // SLLV
+          {
+            emit_shl(s,HOST_TEMPREG,t);
+          }
+          if(opcode2[i]==6) // SRLV
+          {
+            emit_shr(s,HOST_TEMPREG,t);
+          }
+          if(opcode2[i]==7) // SRAV
+          {
+            emit_sar(s,HOST_TEMPREG,t);
+          }
+        }
+      }
+    } else { // DSLLV/DSRLV/DSRAV
+      signed char sh,sl,th,tl,shift;
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      shift=get_reg(i_regs->regmap,rs2[i]);
+      if(tl>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(tl);
+          if(th>=0) emit_zeroreg(th);
+        }
+        else if(rs2[i]==0)
+        {
+          assert(sl>=0);
+          if(sl!=tl) emit_mov(sl,tl);
+          if(th>=0&&sh!=th) emit_mov(sh,th);
+        }
+        else
+        {
+          // FIXME: What if shift==tl ?
+          assert(shift!=tl);
+          int temp=get_reg(i_regs->regmap,-1);
+          int real_th=th;
+          if(th<0&&opcode2[i]!=0x14) {th=temp;} // DSLLV doesn't need a temporary register
+          assert(sl>=0);
+          assert(sh>=0);
+          emit_andimm(shift,31,HOST_TEMPREG);
+          if(opcode2[i]==0x14) // DSLLV
+          {
+            if(th>=0) emit_shl(sh,HOST_TEMPREG,th);
+            emit_rsbimm(HOST_TEMPREG,32,HOST_TEMPREG);
+            emit_orrshr(sl,HOST_TEMPREG,th);
+            emit_andimm(shift,31,HOST_TEMPREG);
+            emit_testimm(shift,32);
+            emit_shl(sl,HOST_TEMPREG,tl);
+            if(th>=0) emit_cmovne_reg(tl,th);
+            emit_cmovne_imm(0,tl);
+          }
+          if(opcode2[i]==0x16) // DSRLV
+          {
+            assert(th>=0);
+            emit_shr(sl,HOST_TEMPREG,tl);
+            emit_rsbimm(HOST_TEMPREG,32,HOST_TEMPREG);
+            emit_orrshl(sh,HOST_TEMPREG,tl);
+            emit_andimm(shift,31,HOST_TEMPREG);
+            emit_testimm(shift,32);
+            emit_shr(sh,HOST_TEMPREG,th);
+            emit_cmovne_reg(th,tl);
+            if(real_th>=0) emit_cmovne_imm(0,th);
+          }
+          if(opcode2[i]==0x17) // DSRAV
+          {
+            assert(th>=0);
+            emit_shr(sl,HOST_TEMPREG,tl);
+            emit_rsbimm(HOST_TEMPREG,32,HOST_TEMPREG);
+            if(real_th>=0) {
+              assert(temp>=0);
+              emit_sarimm(th,31,temp);
+            }
+            emit_orrshl(sh,HOST_TEMPREG,tl);
+            emit_andimm(shift,31,HOST_TEMPREG);
+            emit_testimm(shift,32);
+            emit_sar(sh,HOST_TEMPREG,th);
+            emit_cmovne_reg(th,tl);
+            if(real_th>=0) emit_cmovne_reg(temp,th);
+          }
+        }
+      }
+    }
+  }
+}
+#define shift_assemble shift_assemble_arm
+
+void loadlr_assemble_arm(int i,struct regstat *i_regs)
+{
+  int s,th,tl,temp,temp2,addr,map=-1;
+  int offset;
+  int jaddr=0;
+  int memtarget,c=0;
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,rt1[i]|64);
+  tl=get_reg(i_regs->regmap,rt1[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,-1);
+  temp2=get_reg(i_regs->regmap,FTEMP);
+  addr=get_reg(i_regs->regmap,AGEN1+(i&1));
+  assert(addr<0);
+  offset=imm[i];
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  reglist|=1<<temp;
+  if(offset||s<0||c) addr=temp2;
+  else addr=s;
+  if(s>=0) {
+    c=(i_regs->wasconst>>s)&1;
+    memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+    if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
+  }
+  if(tl>=0) {
+    //assert(tl>=0);
+    //assert(rt1[i]);
+    if(!using_tlb) {
+      if(!c) {
+        emit_shlimm(addr,3,temp);
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR
+        }else{
+          emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
+        }
+        emit_cmpimm(addr,0x800000);
+        jaddr=(int)out;
+        emit_jno(0);
+      }
+      else {
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+        }else{
+          emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+        }
+      }
+    }else{ // using tlb
+      int a;
+      if(c) {
+        a=-1;
+      }else if (opcode[i]==0x22||opcode[i]==0x26) {
+        a=0xFFFFFFFC; // LWL/LWR
+      }else{
+        a=0xFFFFFFF8; // LDL/LDR
+      }
+      map=get_reg(i_regs->regmap,TLREG);
+      assert(map>=0);
+      map=do_tlb_r(addr,temp2,map,0,a,c?-1:temp,c,constmap[i][s]+offset);
+      if(c) {
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+        }else{
+          emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+        }
+      }
+      do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
+    }
+    if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR
+      if(!c||memtarget) {
+        //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2);
+        emit_readword_indexed_tlb((int)rdram-0x80000000,temp2,map,temp2);
+        if(jaddr) add_stub(LOADW_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist);
+      emit_andimm(temp,24,temp);
+      if (opcode[i]==0x26) emit_xorimm(temp,24,temp); // LWR
+      emit_movimm(-1,HOST_TEMPREG);
+      if (opcode[i]==0x26) {
+        emit_shr(temp2,temp,temp2);
+        emit_bic_lsr(tl,HOST_TEMPREG,temp,tl);
+      }else{
+        emit_shl(temp2,temp,temp2);
+        emit_bic_lsl(tl,HOST_TEMPREG,temp,tl);
+      }
+      emit_or(temp2,tl,tl);
+      //emit_storereg(rt1[i],tl); // DEBUG
+    }
+    if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR
+      int temp2h=get_reg(i_regs->regmap,FTEMP|64);
+      if(!c||memtarget) {
+        //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,temp2,temp2h);
+        //emit_readword_indexed((int)rdram-0x7FFFFFFC,temp2,temp2);
+        emit_readdword_indexed_tlb((int)rdram-0x80000000,temp2,map,temp2h,temp2);
+        if(jaddr) add_stub(LOADD_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADD_STUB,i,(constmap[i][s]+offset)&0xFFFFFFF8,i_regs->regmap,FTEMP,ccadj[i],reglist);
+      emit_testimm(temp,32);
+      emit_andimm(temp,24,temp);
+      if (opcode[i]==0x1A) { // LDL
+        emit_rsbimm(temp,32,HOST_TEMPREG);
+        emit_shl(temp2h,temp,temp2h);
+        emit_orrshr(temp2,HOST_TEMPREG,temp2h);
+        emit_movimm(-1,HOST_TEMPREG);
+        emit_shl(temp2,temp,temp2);
+        emit_cmove_reg(temp2h,th);
+        emit_biceq_lsl(tl,HOST_TEMPREG,temp,tl);
+        emit_bicne_lsl(th,HOST_TEMPREG,temp,th);
+        emit_orreq(temp2,tl,tl);
+        emit_orrne(temp2,th,th);
+      }
+      if (opcode[i]==0x1B) { // LDR
+        emit_xorimm(temp,24,temp);
+        emit_rsbimm(temp,32,HOST_TEMPREG);
+        emit_shr(temp2,temp,temp2);
+        emit_orrshl(temp2h,HOST_TEMPREG,temp2);
+        emit_movimm(-1,HOST_TEMPREG);
+        emit_shr(temp2h,temp,temp2h);
+        emit_cmovne_reg(temp2,tl);
+        emit_bicne_lsr(th,HOST_TEMPREG,temp,th);
+        emit_biceq_lsr(tl,HOST_TEMPREG,temp,tl);
+        emit_orrne(temp2h,th,th);
+        emit_orreq(temp2h,tl,tl);
+      }
+    }
+  }
+}
+#define loadlr_assemble loadlr_assemble_arm
+
+void cop0_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]==0) // MFC0
+  {
+    signed char t=get_reg(i_regs->regmap,rt1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    //assert(t>=0); // Why does this happen?  OOT is weird
+    if(t>=0) {
+      emit_addimm(FP,(int)&fake_pc-(int)&dynarec_local,0);
+      emit_movimm((source[i]>>11)&0x1f,1);
+      emit_writeword(0,(int)&PC);
+      emit_writebyte(1,(int)&(fake_pc.f.r.nrd));
+      if(copr==9) {
+        emit_readword((int)&last_count,ECX);
+        emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+      }
+      emit_call((int)MFC0);
+      emit_readword((int)&readmem_dword,t);
+    }
+  }
+  else if(opcode2[i]==4) // MTC0
+  {
+    signed char s=get_reg(i_regs->regmap,rs1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    assert(s>=0);
+    emit_writeword(s,(int)&readmem_dword);
+    wb_register(rs1[i],i_regs->regmap,i_regs->dirty,i_regs->is32);
+    emit_addimm(FP,(int)&fake_pc-(int)&dynarec_local,0);
+    emit_movimm((source[i]>>11)&0x1f,1);
+    emit_writeword(0,(int)&PC);
+    emit_writebyte(1,(int)&(fake_pc.f.r.nrd));
+    if(copr==9||copr==11||copr==12) {
+      emit_readword((int)&last_count,ECX);
+      emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+      emit_add(HOST_CCREG,ECX,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_writeword(HOST_CCREG,(int)&Count);
+    }
+    // What a mess.  The status register (12) can enable interrupts,
+    // so needs a special case to handle a pending interrupt.
+    // The interrupt must be taken immediately, because a subsequent
+    // instruction might disable interrupts again.
+    if(copr==12&&!is_delayslot) {
+      emit_movimm(start+i*4+4,0);
+      emit_movimm(0,1);
+      emit_writeword(0,(int)&pcaddr);
+      emit_writeword(1,(int)&pending_exception);
+    }
+    //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
+    //else
+    emit_call((int)MTC0);
+    if(copr==9||copr==11||copr==12) {
+      emit_readword((int)&Count,HOST_CCREG);
+      emit_readword((int)&next_interupt,ECX);
+      emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+      emit_writeword(ECX,(int)&last_count);
+      emit_storereg(CCREG,HOST_CCREG);
+    }
+    if(copr==12) {
+      assert(!is_delayslot);
+      emit_readword((int)&pending_exception,14);
+    }
+    emit_loadreg(rs1[i],s);
+    if(get_reg(i_regs->regmap,rs1[i]|64)>=0)
+      emit_loadreg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64));
+    if(copr==12) {
+      emit_test(14,14);
+      emit_jne((int)&do_interrupt);
+    }
+    cop1_usable=0;
+  }
+  else
+  {
+    assert(opcode2[i]==0x10);
+    if((source[i]&0x3f)==0x01) // TLBR
+      emit_call((int)TLBR);
+    if((source[i]&0x3f)==0x02) // TLBWI
+      emit_call((int)TLBWI_new);
+    if((source[i]&0x3f)==0x06) { // TLBWR
+      // The TLB entry written by TLBWR is dependent on the count,
+      // so update the cycle count
+      emit_readword((int)&last_count,ECX);
+      if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+      emit_add(HOST_CCREG,ECX,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_writeword(HOST_CCREG,(int)&Count);
+      emit_call((int)TLBWR_new);
+    }
+    if((source[i]&0x3f)==0x08) // TLBP
+      emit_call((int)TLBP);
+    if((source[i]&0x3f)==0x18) // ERET
+    {
+      int count=ccadj[i];
+      if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*count,HOST_CCREG); // TODO: Should there be an extra cycle here?
+      emit_jmp((int)jump_eret);
+    }
+  }
+}
+
+void cop1_assemble(int i,struct regstat *i_regs)
+{
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  if (opcode2[i]==0) { // MFC1
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],tl);
+      emit_readword_indexed(0,tl,tl);
+    }
+  }
+  else if (opcode2[i]==1) { // DMFC1
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    signed char th=get_reg(i_regs->regmap,rt1[i]|64);
+    if(tl>=0) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],tl);
+      if(th>=0) emit_readword_indexed(4,tl,th);
+      emit_readword_indexed(0,tl,tl);
+    }
+  }
+  else if (opcode2[i]==4) { // MTC1
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    signed char temp=get_reg(i_regs->regmap,-1);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_writeword_indexed(sl,0,temp);
+  }
+  else if (opcode2[i]==5) { // DMTC1
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    signed char sh=rs1[i]>0?get_reg(i_regs->regmap,rs1[i]|64):sl;
+    signed char temp=get_reg(i_regs->regmap,-1);
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_writeword_indexed(sh,4,temp);
+    emit_writeword_indexed(sl,0,temp);
+  }
+  else if (opcode2[i]==2) // CFC1
+  {
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0) {
+      u_int copr=(source[i]>>11)&0x1f;
+      if(copr==0) emit_readword((int)&FCR0,tl);
+      if(copr==31) emit_readword((int)&FCR31,tl);
+    }
+  }
+  else if (opcode2[i]==6) // CTC1
+  {
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    u_int copr=(source[i]>>11)&0x1f;
+    assert(sl>=0);
+    if(copr==31)
+    {
+      emit_writeword(sl,(int)&FCR31);
+      // Set the rounding mode
+      //FIXME
+      //char temp=get_reg(i_regs->regmap,-1);
+      //emit_andimm(sl,3,temp);
+      //emit_fldcw_indexed((int)&rounding_modes,temp);
+    }
+  }
+}
+
+void fconv_assemble_arm(int i,struct regstat *i_regs)
+{
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  #if(defined(__VFP_FP__) && !defined(__SOFTFP__)) 
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) { // trunc_w_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp,15);
+    emit_ftosizs(15,15); // float->int, truncate
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fsts(15,temp);
+    return;
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) { // trunc_w_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_vldr(temp,7);
+    emit_ftosizd(7,13); // double->int, truncate
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fsts(13,temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x20) { // cvt_s_w
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp,13);
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fsitos(13,15);
+    emit_fsts(15,temp);
+    return;
+  }
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) { // cvt_d_w
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp,13);
+    emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fsitod(13,7);
+    emit_vstr(7,temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) { // cvt_d_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp,13);
+    emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fcvtds(13,7);
+    emit_vstr(7,temp);
+    return;
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) { // cvt_s_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_vldr(temp,7);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fcvtsd(7,13);
+    emit_fsts(13,temp);
+    return;
+  }
+  #endif
+  
+  // C emulation code
+  
+  u_int hr,reglist=0;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  save_regs(reglist);
+  
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x20) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_s_w);
+  }
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_d_w);
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x20) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_s_l);
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x21) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_d_l);
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_d_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x24) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x25) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_l_s);
+  }
+  
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_s_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x24) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x25) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)cvt_l_d);
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x08) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)round_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x09) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)trunc_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0a) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)ceil_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0b) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)floor_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0c) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)round_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)trunc_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0e) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)ceil_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0f) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)floor_w_s);
+  }
+  
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x08) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)round_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x09) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)trunc_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0a) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)ceil_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0b) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)floor_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0c) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)round_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)trunc_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0e) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)ceil_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0f) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    emit_call((int)floor_w_d);
+  }
+  
+  restore_regs(reglist);
+}
+#define fconv_assemble fconv_assemble_arm
+
+void fcomp_assemble(int i,struct regstat *i_regs)
+{
+  signed char fs=get_reg(i_regs->regmap,FSREG);
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char cs=get_reg(i_regs->regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  if((source[i]&0x3f)==0x30) {
+    emit_andimm(fs,~0x800000,fs);
+    return;
+  }
+  
+  if((source[i]&0x3e)==0x38) {
+    // sf/ngle - these should throw exceptions for NaNs
+    emit_andimm(fs,~0x800000,fs);
+    return;
+  }
+  
+  #if(defined(__VFP_FP__) && !defined(__SOFTFP__)) 
+  if(opcode2[i]==0x10) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],HOST_TEMPREG);
+    emit_orimm(fs,0x800000,fs);
+    emit_flds(temp,14);
+    emit_flds(HOST_TEMPREG,15);
+    emit_fcmps(14,15);
+    emit_fmstat();
+    if((source[i]&0x3f)==0x31) emit_bicvc_imm(fs,0x800000,fs); // c_un_s
+    if((source[i]&0x3f)==0x32) emit_bicne_imm(fs,0x800000,fs); // c_eq_s
+    if((source[i]&0x3f)==0x33) {emit_bicne_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ueq_s
+    if((source[i]&0x3f)==0x34) emit_biccs_imm(fs,0x800000,fs); // c_olt_s
+    if((source[i]&0x3f)==0x35) {emit_biccs_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ult_s 
+    if((source[i]&0x3f)==0x36) emit_bichi_imm(fs,0x800000,fs); // c_ole_s
+    if((source[i]&0x3f)==0x37) {emit_bichi_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ule_s
+    if((source[i]&0x3f)==0x3a) emit_bicne_imm(fs,0x800000,fs); // c_seq_s
+    if((source[i]&0x3f)==0x3b) emit_bicne_imm(fs,0x800000,fs); // c_ngl_s
+    if((source[i]&0x3f)==0x3c) emit_biccs_imm(fs,0x800000,fs); // c_lt_s
+    if((source[i]&0x3f)==0x3d) emit_biccs_imm(fs,0x800000,fs); // c_nge_s
+    if((source[i]&0x3f)==0x3e) emit_bichi_imm(fs,0x800000,fs); // c_le_s
+    if((source[i]&0x3f)==0x3f) emit_bichi_imm(fs,0x800000,fs); // c_ngt_s
+    return;
+  }
+  if(opcode2[i]==0x11) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],HOST_TEMPREG);
+    emit_orimm(fs,0x800000,fs);
+    emit_vldr(temp,6);
+    emit_vldr(HOST_TEMPREG,7);
+    emit_fcmpd(6,7);
+    emit_fmstat();
+    if((source[i]&0x3f)==0x31) emit_bicvc_imm(fs,0x800000,fs); // c_un_d
+    if((source[i]&0x3f)==0x32) emit_bicne_imm(fs,0x800000,fs); // c_eq_d
+    if((source[i]&0x3f)==0x33) {emit_bicne_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ueq_d
+    if((source[i]&0x3f)==0x34) emit_biccs_imm(fs,0x800000,fs); // c_olt_d
+    if((source[i]&0x3f)==0x35) {emit_biccs_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ult_d
+    if((source[i]&0x3f)==0x36) emit_bichi_imm(fs,0x800000,fs); // c_ole_d
+    if((source[i]&0x3f)==0x37) {emit_bichi_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ule_d
+    if((source[i]&0x3f)==0x3a) emit_bicne_imm(fs,0x800000,fs); // c_seq_d
+    if((source[i]&0x3f)==0x3b) emit_bicne_imm(fs,0x800000,fs); // c_ngl_d
+    if((source[i]&0x3f)==0x3c) emit_biccs_imm(fs,0x800000,fs); // c_lt_d
+    if((source[i]&0x3f)==0x3d) emit_biccs_imm(fs,0x800000,fs); // c_nge_d
+    if((source[i]&0x3f)==0x3e) emit_bichi_imm(fs,0x800000,fs); // c_le_d
+    if((source[i]&0x3f)==0x3f) emit_bichi_imm(fs,0x800000,fs); // c_ngt_d
+    return;
+  }
+  #endif
+  
+  // C only
+  
+  u_int hr,reglist=0;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  reglist&=~(1<<fs);
+  save_regs(reglist);
+  if(opcode2[i]==0x10) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],ARG2_REG);
+    if((source[i]&0x3f)==0x30) emit_call((int)c_f_s);
+    if((source[i]&0x3f)==0x31) emit_call((int)c_un_s);
+    if((source[i]&0x3f)==0x32) emit_call((int)c_eq_s);
+    if((source[i]&0x3f)==0x33) emit_call((int)c_ueq_s);
+    if((source[i]&0x3f)==0x34) emit_call((int)c_olt_s);
+    if((source[i]&0x3f)==0x35) emit_call((int)c_ult_s);
+    if((source[i]&0x3f)==0x36) emit_call((int)c_ole_s);
+    if((source[i]&0x3f)==0x37) emit_call((int)c_ule_s);
+    if((source[i]&0x3f)==0x38) emit_call((int)c_sf_s);
+    if((source[i]&0x3f)==0x39) emit_call((int)c_ngle_s);
+    if((source[i]&0x3f)==0x3a) emit_call((int)c_seq_s);
+    if((source[i]&0x3f)==0x3b) emit_call((int)c_ngl_s);
+    if((source[i]&0x3f)==0x3c) emit_call((int)c_lt_s);
+    if((source[i]&0x3f)==0x3d) emit_call((int)c_nge_s);
+    if((source[i]&0x3f)==0x3e) emit_call((int)c_le_s);
+    if((source[i]&0x3f)==0x3f) emit_call((int)c_ngt_s);
+  }
+  if(opcode2[i]==0x11) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],ARG2_REG);
+    if((source[i]&0x3f)==0x30) emit_call((int)c_f_d);
+    if((source[i]&0x3f)==0x31) emit_call((int)c_un_d);
+    if((source[i]&0x3f)==0x32) emit_call((int)c_eq_d);
+    if((source[i]&0x3f)==0x33) emit_call((int)c_ueq_d);
+    if((source[i]&0x3f)==0x34) emit_call((int)c_olt_d);
+    if((source[i]&0x3f)==0x35) emit_call((int)c_ult_d);
+    if((source[i]&0x3f)==0x36) emit_call((int)c_ole_d);
+    if((source[i]&0x3f)==0x37) emit_call((int)c_ule_d);
+    if((source[i]&0x3f)==0x38) emit_call((int)c_sf_d);
+    if((source[i]&0x3f)==0x39) emit_call((int)c_ngle_d);
+    if((source[i]&0x3f)==0x3a) emit_call((int)c_seq_d);
+    if((source[i]&0x3f)==0x3b) emit_call((int)c_ngl_d);
+    if((source[i]&0x3f)==0x3c) emit_call((int)c_lt_d);
+    if((source[i]&0x3f)==0x3d) emit_call((int)c_nge_d);
+    if((source[i]&0x3f)==0x3e) emit_call((int)c_le_d);
+    if((source[i]&0x3f)==0x3f) emit_call((int)c_ngt_d);
+  }
+  restore_regs(reglist);
+  emit_loadreg(FSREG,fs);
+}
+
+void float_assemble(int i,struct regstat *i_regs)
+{
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char cs=get_reg(i_regs->regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  #if(defined(__VFP_FP__) && !defined(__SOFTFP__)) 
+  if((source[i]&0x3f)==6) // mov
+  {
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+      if(opcode2[i]==0x10) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],HOST_TEMPREG);
+        emit_readword_indexed(0,temp,temp);
+        emit_writeword_indexed(temp,0,HOST_TEMPREG);
+      }
+      if(opcode2[i]==0x11) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],HOST_TEMPREG);
+        emit_vldr(temp,7);
+        emit_vstr(7,HOST_TEMPREG);
+      }
+    }
+    return;
+  }
+  
+  if((source[i]&0x3f)>3)
+  {
+    if(opcode2[i]==0x10) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+      emit_flds(temp,15);
+      if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+      }
+      if((source[i]&0x3f)==4) // sqrt
+        emit_fsqrts(15,15);
+      if((source[i]&0x3f)==5) // abs
+        emit_fabss(15,15);
+      if((source[i]&0x3f)==7) // neg
+        emit_fnegs(15,15);
+      emit_fsts(15,temp);
+    }
+    if(opcode2[i]==0x11) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+      emit_vldr(temp,7);
+      if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+      }
+      if((source[i]&0x3f)==4) // sqrt
+        emit_fsqrtd(7,7);
+      if((source[i]&0x3f)==5) // abs
+        emit_fabsd(7,7);
+      if((source[i]&0x3f)==7) // neg
+        emit_fnegd(7,7);
+      emit_vstr(7,temp);
+    }
+    return;
+  }
+  if((source[i]&0x3f)<4)
+  {
+    if(opcode2[i]==0x10) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    }
+    if(opcode2[i]==0x11) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    }
+    if(((source[i]>>11)&0x1f)!=((source[i]>>16)&0x1f)) {
+      if(opcode2[i]==0x10) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],HOST_TEMPREG);
+        emit_flds(temp,15);
+        emit_flds(HOST_TEMPREG,13);
+        if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+          if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) {
+            emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+          }
+        }
+        if((source[i]&0x3f)==0) emit_fadds(15,13,15);
+        if((source[i]&0x3f)==1) emit_fsubs(15,13,15);
+        if((source[i]&0x3f)==2) emit_fmuls(15,13,15);
+        if((source[i]&0x3f)==3) emit_fdivs(15,13,15);
+        if(((source[i]>>16)&0x1f)==((source[i]>>6)&0x1f)) {
+          emit_fsts(15,HOST_TEMPREG);
+        }else{
+          emit_fsts(15,temp);
+        }
+      }
+      else if(opcode2[i]==0x11) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],HOST_TEMPREG);
+        emit_vldr(temp,7);
+        emit_vldr(HOST_TEMPREG,6);
+        if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+          if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) {
+            emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+          }
+        }
+        if((source[i]&0x3f)==0) emit_faddd(7,6,7);
+        if((source[i]&0x3f)==1) emit_fsubd(7,6,7);
+        if((source[i]&0x3f)==2) emit_fmuld(7,6,7);
+        if((source[i]&0x3f)==3) emit_fdivd(7,6,7);
+        if(((source[i]>>16)&0x1f)==((source[i]>>6)&0x1f)) {
+          emit_vstr(7,HOST_TEMPREG);
+        }else{
+          emit_vstr(7,temp);
+        }
+      }
+    }
+    else {
+      if(opcode2[i]==0x10) {
+        emit_flds(temp,15);
+        if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+          emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+        }
+        if((source[i]&0x3f)==0) emit_fadds(15,15,15);
+        if((source[i]&0x3f)==1) emit_fsubs(15,15,15);
+        if((source[i]&0x3f)==2) emit_fmuls(15,15,15);
+        if((source[i]&0x3f)==3) emit_fdivs(15,15,15);
+        emit_fsts(15,temp);
+      }
+      else if(opcode2[i]==0x11) {
+        emit_vldr(temp,7);
+        if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+          emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+        }
+        if((source[i]&0x3f)==0) emit_faddd(7,7,7);
+        if((source[i]&0x3f)==1) emit_fsubd(7,7,7);
+        if((source[i]&0x3f)==2) emit_fmuld(7,7,7);
+        if((source[i]&0x3f)==3) emit_fdivd(7,7,7);
+        emit_vstr(7,temp);
+      }
+    }
+    return;
+  }
+  #endif
+  
+  u_int hr,reglist=0;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  if(opcode2[i]==0x10) { // Single precision
+    save_regs(reglist);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG);
+    if((source[i]&0x3f)<4) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],ARG2_REG);
+      emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG3_REG);
+    }else{
+      emit_readword((int)&reg_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG);
+    }
+    switch(source[i]&0x3f)
+    {
+      case 0x00: emit_call((int)add_s);break;
+      case 0x01: emit_call((int)sub_s);break;
+      case 0x02: emit_call((int)mul_s);break;
+      case 0x03: emit_call((int)div_s);break;
+      case 0x04: emit_call((int)sqrt_s);break;
+      case 0x05: emit_call((int)abs_s);break;
+      case 0x06: emit_call((int)mov_s);break;
+      case 0x07: emit_call((int)neg_s);break;
+    }
+    restore_regs(reglist);
+  }
+  if(opcode2[i]==0x11) { // Double precision
+    save_regs(reglist);
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],ARG1_REG);
+    if((source[i]&0x3f)<4) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],ARG2_REG);
+      emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG3_REG);
+    }else{
+      emit_readword((int)&reg_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG);
+    }
+    switch(source[i]&0x3f)
+    {
+      case 0x00: emit_call((int)add_d);break;
+      case 0x01: emit_call((int)sub_d);break;
+      case 0x02: emit_call((int)mul_d);break;
+      case 0x03: emit_call((int)div_d);break;
+      case 0x04: emit_call((int)sqrt_d);break;
+      case 0x05: emit_call((int)abs_d);break;
+      case 0x06: emit_call((int)mov_d);break;
+      case 0x07: emit_call((int)neg_d);break;
+    }
+    restore_regs(reglist);
+  }
+}
+
+void multdiv_assemble_arm(int i,struct regstat *i_regs)
+{
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  //  case 0x1C: DMULT
+  //  case 0x1D: DMULTU
+  //  case 0x1E: DDIV
+  //  case 0x1F: DDIVU
+  if(rs1[i]&&rs2[i])
+  {
+    if((opcode2[i]&4)==0) // 32-bit
+    {
+      if(opcode2[i]==0x18) // MULT
+      {
+        signed char m1=get_reg(i_regs->regmap,rs1[i]);
+        signed char m2=get_reg(i_regs->regmap,rs2[i]);
+        signed char hi=get_reg(i_regs->regmap,HIREG);
+        signed char lo=get_reg(i_regs->regmap,LOREG);
+        assert(m1>=0);
+        assert(m2>=0);
+        assert(hi>=0);
+        assert(lo>=0);
+        emit_smull(m1,m2,hi,lo);
+      }
+      if(opcode2[i]==0x19) // MULTU
+      {
+        signed char m1=get_reg(i_regs->regmap,rs1[i]);
+        signed char m2=get_reg(i_regs->regmap,rs2[i]);
+        signed char hi=get_reg(i_regs->regmap,HIREG);
+        signed char lo=get_reg(i_regs->regmap,LOREG);
+        assert(m1>=0);
+        assert(m2>=0);
+        assert(hi>=0);
+        assert(lo>=0);
+        emit_umull(m1,m2,hi,lo);
+      }
+      if(opcode2[i]==0x1A) // DIV
+      {
+        signed char d1=get_reg(i_regs->regmap,rs1[i]);
+        signed char d2=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1>=0);
+        assert(d2>=0);
+        signed char quotient=get_reg(i_regs->regmap,LOREG);
+        signed char remainder=get_reg(i_regs->regmap,HIREG);
+        assert(quotient>=0);
+        assert(remainder>=0);
+        emit_movs(d1,remainder);
+        emit_negmi(remainder,remainder);
+        emit_movs(d2,HOST_TEMPREG);
+        emit_jeq((int)out+52); // Division by zero
+        emit_negmi(HOST_TEMPREG,HOST_TEMPREG);
+        emit_clz(HOST_TEMPREG,quotient);
+        emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG);
+        emit_orimm(quotient,1<<31,quotient);
+        emit_shr(quotient,quotient,quotient);
+        emit_cmp(remainder,HOST_TEMPREG);
+        emit_subcs(remainder,HOST_TEMPREG,remainder);
+        emit_adcs(quotient,quotient,quotient);
+        emit_shrimm(HOST_TEMPREG,1,HOST_TEMPREG);
+        emit_jcc((int)out-16); // -4
+        emit_teq(d1,d2);
+        emit_negmi(quotient,quotient);
+        emit_test(d1,d1);
+        emit_negmi(remainder,remainder);
+      }
+      if(opcode2[i]==0x1B) // DIVU
+      {
+        signed char d1=get_reg(i_regs->regmap,rs1[i]); // dividend
+        signed char d2=get_reg(i_regs->regmap,rs2[i]); // divisor
+        assert(d1>=0);
+        assert(d2>=0);
+        signed char quotient=get_reg(i_regs->regmap,LOREG);
+        signed char remainder=get_reg(i_regs->regmap,HIREG);
+        assert(quotient>=0);
+        assert(remainder>=0);
+        emit_test(d2,d2);
+        emit_jeq((int)out+44); // Division by zero
+        emit_clz(d2,HOST_TEMPREG);
+        emit_movimm(1<<31,quotient);
+        emit_shl(d2,HOST_TEMPREG,d2);
+        emit_mov(d1,remainder);
+        emit_shr(quotient,HOST_TEMPREG,quotient);
+        emit_cmp(remainder,d2);
+        emit_subcs(remainder,d2,remainder);
+        emit_adcs(quotient,quotient,quotient);
+        emit_shrcc_imm(d2,1,d2);
+        emit_jcc((int)out-16); // -4
+      }
+    }
+    else // 64-bit
+    {
+      if(opcode2[i]==0x1C) // DMULT
+      {
+        assert(opcode2[i]!=0x1C);
+        signed char m1h=get_reg(i_regs->regmap,rs1[i]|64);
+        signed char m1l=get_reg(i_regs->regmap,rs1[i]);
+        signed char m2h=get_reg(i_regs->regmap,rs2[i]|64);
+        signed char m2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        emit_pushreg(m2h);
+        emit_pushreg(m2l);
+        emit_pushreg(m1h);
+        emit_pushreg(m1l);
+        emit_call((int)&mult64);
+        emit_popreg(m1l);
+        emit_popreg(m1h);
+        emit_popreg(m2l);
+        emit_popreg(m2h);
+        signed char hih=get_reg(i_regs->regmap,HIREG|64);
+        signed char hil=get_reg(i_regs->regmap,HIREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        signed char loh=get_reg(i_regs->regmap,LOREG|64);
+        signed char lol=get_reg(i_regs->regmap,LOREG);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1D) // DMULTU
+      {
+        signed char m1h=get_reg(i_regs->regmap,rs1[i]|64);
+        signed char m1l=get_reg(i_regs->regmap,rs1[i]);
+        signed char m2h=get_reg(i_regs->regmap,rs2[i]|64);
+        signed char m2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        save_regs(0x100f);
+        if(m1l!=0) emit_mov(m1l,0);
+        if(m1h==0) emit_readword((int)&dynarec_local,1);
+        else if(m1h>1) emit_mov(m1h,1);
+        if(m2l<2) emit_readword((int)&dynarec_local+m2l*4,2);
+        else if(m2l>2) emit_mov(m2l,2);
+        if(m2h<3) emit_readword((int)&dynarec_local+m2h*4,3);
+        else if(m2h>3) emit_mov(m2h,3);
+        emit_call((int)&multu64);
+        restore_regs(0x100f);
+        signed char hih=get_reg(i_regs->regmap,HIREG|64);
+        signed char hil=get_reg(i_regs->regmap,HIREG);
+        signed char loh=get_reg(i_regs->regmap,LOREG|64);
+        signed char lol=get_reg(i_regs->regmap,LOREG);
+        /*signed char temp=get_reg(i_regs->regmap,-1);
+        signed char rh=get_reg(i_regs->regmap,HIREG|64);
+        signed char rl=get_reg(i_regs->regmap,HIREG);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        assert(temp>=0);
+        //emit_mov(m1l,EAX);
+        //emit_mul(m2l);
+        emit_umull(rl,rh,m1l,m2l);
+        emit_storereg(LOREG,rl);
+        emit_mov(rh,temp);
+        //emit_mov(m1h,EAX);
+        //emit_mul(m2l);
+        emit_umull(rl,rh,m1h,m2l);
+        emit_adds(rl,temp,temp);
+        emit_adcimm(rh,0,rh);
+        emit_storereg(HIREG,rh);
+        //emit_mov(m2h,EAX);
+        //emit_mul(m1l);
+        emit_umull(rl,rh,m1l,m2h);
+        emit_adds(rl,temp,temp);
+        emit_adcimm(rh,0,rh);
+        emit_storereg(LOREG|64,temp);
+        emit_mov(rh,temp);
+        //emit_mov(m2h,EAX);
+        //emit_mul(m1h);
+        emit_umull(rl,rh,m1h,m2h);
+        emit_adds(rl,temp,rl);
+        emit_loadreg(HIREG,temp);
+        emit_adcimm(rh,0,rh);
+        emit_adds(rl,temp,rl);
+        emit_adcimm(rh,0,rh);
+        // DEBUG
+        /*
+        emit_pushreg(m2h);
+        emit_pushreg(m2l);
+        emit_pushreg(m1h);
+        emit_pushreg(m1l);
+        emit_call((int)&multu64);
+        emit_popreg(m1l);
+        emit_popreg(m1h);
+        emit_popreg(m2l);
+        emit_popreg(m2h);
+        signed char hih=get_reg(i_regs->regmap,HIREG|64);
+        signed char hil=get_reg(i_regs->regmap,HIREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);  // DEBUG
+        if(hil>=0) emit_loadreg(HIREG,hil);  // DEBUG
+        */
+        // Shouldn't be necessary
+        //char loh=get_reg(i_regs->regmap,LOREG|64);
+        //char lol=get_reg(i_regs->regmap,LOREG);
+        //if(loh>=0) emit_loadreg(LOREG|64,loh);
+        //if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1E) // DDIV
+      {
+        signed char d1h=get_reg(i_regs->regmap,rs1[i]|64);
+        signed char d1l=get_reg(i_regs->regmap,rs1[i]);
+        signed char d2h=get_reg(i_regs->regmap,rs2[i]|64);
+        signed char d2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1h>=0);
+        assert(d2h>=0);
+        assert(d1l>=0);
+        assert(d2l>=0);
+        save_regs(0x100f);
+        if(d1l!=0) emit_mov(d1l,0);
+        if(d1h==0) emit_readword((int)&dynarec_local,1);
+        else if(d1h>1) emit_mov(d1h,1);
+        if(d2l<2) emit_readword((int)&dynarec_local+d2l*4,2);
+        else if(d2l>2) emit_mov(d2l,2);
+        if(d2h<3) emit_readword((int)&dynarec_local+d2h*4,3);
+        else if(d2h>3) emit_mov(d2h,3);
+        emit_call((int)&div64);
+        restore_regs(0x100f);
+        signed char hih=get_reg(i_regs->regmap,HIREG|64);
+        signed char hil=get_reg(i_regs->regmap,HIREG);
+        signed char loh=get_reg(i_regs->regmap,LOREG|64);
+        signed char lol=get_reg(i_regs->regmap,LOREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1F) // DDIVU
+      {
+      //u_int hr,reglist=0;
+      //for(hr=0;hr<HOST_REGS;hr++) {
+      //  if(i_regs->regmap[hr]>=0 && (i_regs->regmap[hr]&62)!=HIREG) reglist|=1<<hr;
+      //}
+        signed char d1h=get_reg(i_regs->regmap,rs1[i]|64);
+        signed char d1l=get_reg(i_regs->regmap,rs1[i]);
+        signed char d2h=get_reg(i_regs->regmap,rs2[i]|64);
+        signed char d2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1h>=0);
+        assert(d2h>=0);
+        assert(d1l>=0);
+        assert(d2l>=0);
+        save_regs(0x100f);
+        if(d1l!=0) emit_mov(d1l,0);
+        if(d1h==0) emit_readword((int)&dynarec_local,1);
+        else if(d1h>1) emit_mov(d1h,1);
+        if(d2l<2) emit_readword((int)&dynarec_local+d2l*4,2);
+        else if(d2l>2) emit_mov(d2l,2);
+        if(d2h<3) emit_readword((int)&dynarec_local+d2h*4,3);
+        else if(d2h>3) emit_mov(d2h,3);
+        emit_call((int)&divu64);
+        restore_regs(0x100f);
+        signed char hih=get_reg(i_regs->regmap,HIREG|64);
+        signed char hil=get_reg(i_regs->regmap,HIREG);
+        signed char loh=get_reg(i_regs->regmap,LOREG|64);
+        signed char lol=get_reg(i_regs->regmap,LOREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+    }
+  }
+  else
+  {
+    // Multiply by zero is zero.
+    // MIPS does not have a divide by zero exception.
+    // The result is undefined, we return zero.
+    signed char hr=get_reg(i_regs->regmap,HIREG);
+    signed char lr=get_reg(i_regs->regmap,LOREG);
+    if(hr>=0) emit_zeroreg(hr);
+    if(lr>=0) emit_zeroreg(lr);
+  }
+}
+#define multdiv_assemble multdiv_assemble_arm
+
+void do_preload_rhash(int r) {
+  // Don't need this for ARM.  On x86, this puts the value 0xf8 into the
+  // register.  On ARM the hash can be done with a single instruction (below)
+}
+
+void do_preload_rhtbl(int ht) {
+  emit_addimm(FP,(int)&mini_ht-(int)&dynarec_local,ht);
+}
+
+void do_rhash(int rs,int rh) {
+  emit_andimm(rs,0xf8,rh);
+}
+
+void do_miniht_load(int ht,int rh) {
+  assem_debug("ldr %s,[%s,%s]!\n",regname[rh],regname[ht],regname[rh]);
+  output_w32(0xe7b00000|rd_rn_rm(rh,ht,rh));
+}
+
+void do_miniht_jump(int rs,int rh,int ht) {
+  emit_cmp(rh,rs);
+  emit_ldreq_indexed(ht,4,15);
+  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+  emit_mov(rs,7);
+  emit_jmp(jump_vaddr_reg[7]);
+  #else
+  emit_jmp(jump_vaddr_reg[rs]);
+  #endif
+}
+
+void do_miniht_insert(u_int return_address,int rt,int temp) {
+  #ifdef ARMv5_ONLY
+  emit_movimm(return_address,rt); // PC into link register
+  add_to_linker((int)out,return_address,1);
+  emit_pcreladdr(temp);
+  emit_writeword(rt,(int)&mini_ht[(return_address&0xFF)>>3][0]);
+  emit_writeword(temp,(int)&mini_ht[(return_address&0xFF)>>3][1]);
+  #else
+  emit_movw(return_address&0x0000FFFF,rt);
+  add_to_linker((int)out,return_address,1);
+  emit_pcreladdr(temp);
+  emit_writeword(temp,(int)&mini_ht[(return_address&0xFF)>>3][1]);
+  emit_movt(return_address&0xFFFF0000,rt);
+  emit_writeword(rt,(int)&mini_ht[(return_address&0xFF)>>3][0]);
+  #endif
+}
+
+// Sign-extend to 64 bits and write out upper half of a register
+// This is useful where we have a 32-bit value in a register, and want to
+// keep it in a 32-bit register, but can't guarantee that it won't be read
+// as a 64-bit value later.
+void wb_sx(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32_pre,uint64_t is32,uint64_t u,uint64_t uu)
+{
+  if(is32_pre==is32) return;
+  int hr,reg;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      //if(pre[hr]==entry[hr]) {
+        if((reg=pre[hr])>=0) {
+          if((dirty>>hr)&1) {
+            if( ((is32_pre&~is32&~uu)>>reg)&1 ) {
+              emit_sarimm(hr,31,HOST_TEMPREG);
+              emit_storereg(reg|64,HOST_TEMPREG);
+            }
+          }
+        }
+      //}
+    }
+  }
+}
+
+void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t is32_pre,uint64_t u,uint64_t uu)
+{
+  //if(dirty_pre==dirty) return;
+  int hr,reg,new_hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      reg=pre[hr];
+      if(((~u)>>(reg&63))&1) {
+        if(reg==entry[hr]||(reg>0&&entry[hr]<0)) {
+          if(((dirty_pre&~dirty)>>hr)&1) {
+            if(reg>0&&reg<34) {
+              emit_storereg(reg,hr);
+              if( ((is32_pre&~uu)>>reg)&1 ) {
+                emit_sarimm(hr,31,HOST_TEMPREG);
+                emit_storereg(reg|64,HOST_TEMPREG);
+              }
+            }
+            else if(reg>=64) {
+              emit_storereg(reg,hr);
+            }
+          }
+        }
+        else // Check if register moved to a different register
+        if((new_hr=get_reg(entry,reg))>=0) {
+          if((dirty_pre>>hr)&(~dirty>>new_hr)&1) {
+            if(reg>0&&reg<34) {
+              emit_storereg(reg,hr);
+              if( ((is32_pre&~uu)>>reg)&1 ) {
+                emit_sarimm(hr,31,HOST_TEMPREG);
+                emit_storereg(reg|64,HOST_TEMPREG);
+              }
+            }
+            else if(reg>=64) {
+              emit_storereg(reg,hr);
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+
+/* using strd could possibly help but you'd have to allocate registers in pairs
+void wb_invalidate_arm(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32,uint64_t u,uint64_t uu)
+{
+  int hr;
+  int wrote=-1;
+  for(hr=HOST_REGS-1;hr>=0;hr--) {
+    if(hr!=EXCLUDE_REG) {
+      if(pre[hr]!=entry[hr]) {
+        if(pre[hr]>=0) {
+          if((dirty>>hr)&1) {
+            if(get_reg(entry,pre[hr])<0) {
+              if(pre[hr]<64) {
+                if(!((u>>pre[hr])&1)) {
+                  if(hr<10&&(~hr&1)&&(pre[hr+1]<0||wrote==hr+1)) {
+                    if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) {
+                      emit_sarimm(hr,31,hr+1);
+                      emit_strdreg(pre[hr],hr);
+                    }
+                    else
+                      emit_storereg(pre[hr],hr);
+                  }else{
+                    emit_storereg(pre[hr],hr);
+                    if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) {
+                      emit_sarimm(hr,31,hr);
+                      emit_storereg(pre[hr]|64,hr);
+                    }
+                  }
+                }
+              }else{
+                if(!((uu>>(pre[hr]&63))&1) && !((is32>>(pre[hr]&63))&1)) {
+                  emit_storereg(pre[hr],hr);
+                }
+              }
+              wrote=hr;
+            }
+          }
+        }
+      }
+    }
+  }
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(pre[hr]!=entry[hr]) {
+        if(pre[hr]>=0) {
+          int nr;
+          if((nr=get_reg(entry,pre[hr]))>=0) {
+            emit_mov(hr,nr);
+          }
+        }
+      }
+    }
+  }
+}
+#define wb_invalidate wb_invalidate_arm
+*/
+
+// CPU-architecture-specific initialization
+void arch_init() {
+  rounding_modes[0]=0x0<<22; // round
+  rounding_modes[1]=0x3<<22; // trunc
+  rounding_modes[2]=0x1<<22; // ceil
+  rounding_modes[3]=0x2<<22; // floor
+}
diff --git a/libpcsxcore/new_dynarec/assem_arm.h b/libpcsxcore/new_dynarec/assem_arm.h
new file mode 100644 (file)
index 0000000..93066a2
--- /dev/null
@@ -0,0 +1,42 @@
+#define HOST_REGS 13
+#define HOST_CCREG 10
+#define HOST_BTREG 8
+#define EXCLUDE_REG 11
+
+#define HOST_IMM8 1
+#define HAVE_CMOV_IMM 1
+#define CORTEX_A8_BRANCH_PREDICTION_HACK 1
+#define USE_MINI_HT 1
+//#define REG_PREFETCH 1
+
+/* ARM calling convention:
+   r0-r3, r12: caller-save
+   r4-r11: callee-save */
+
+#define ARG1_REG 0
+#define ARG2_REG 1
+#define ARG3_REG 2
+#define ARG4_REG 3
+
+/* GCC register naming convention:
+   r10 = sl (base)
+   r11 = fp (frame pointer)
+   r12 = ip (scratch)
+   r13 = sp (stack pointer)
+   r14 = lr (link register)
+   r15 = pc (program counter) */
+
+#define FP 11
+#define LR 14
+#define HOST_TEMPREG 14
+
+// Note: FP is set to &dynarec_local when executing generated code.
+// Thus the local variables are actually global and not on the stack.
+
+extern char *invc_ptr;
+
+#define BASE_ADDR 0x7000000 // Code generator target address
+#define TARGET_SIZE_2 24 // 2^24 = 16 megabytes
+
+// This is defined in linkage_arm.s, but gcc -O3 likes this better
+#define rdram ((unsigned int *)0x80000000)
diff --git a/libpcsxcore/new_dynarec/assem_x64.c b/libpcsxcore/new_dynarec/assem_x64.c
new file mode 100644 (file)
index 0000000..bd1f8f8
--- /dev/null
@@ -0,0 +1,4287 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - assem_x64.c                                             *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+int cycle_count;
+int last_count;
+int pcaddr;
+int pending_exception;
+int branch_target;
+uint64_t readmem_dword;
+precomp_instr fake_pc;
+u_int memory_map[1048576];
+u_int mini_ht[32][2]  __attribute__((aligned(8)));
+u_char restore_candidate[512]  __attribute__((aligned(4)));
+
+void do_interrupt();
+void jump_vaddr_eax();
+void jump_vaddr_ecx();
+void jump_vaddr_edx();
+void jump_vaddr_ebx();
+void jump_vaddr_ebp();
+void jump_vaddr_edi();
+
+const void * jump_vaddr_reg[8] = {
+  jump_vaddr_eax,
+  jump_vaddr_ecx,
+  jump_vaddr_edx,
+  jump_vaddr_ebx,
+  0,
+  jump_vaddr_ebp,
+  0,
+  jump_vaddr_edi };
+
+const u_short rounding_modes[4] = {
+  0x33F, // round
+  0xF3F, // trunc
+  0xB3F, // ceil
+  0x73F};// floor
+
+#include "fpu.h"
+
+// We need these for cmovcc instructions on x86
+u_int const_zero=0;
+u_int const_one=1;
+
+/* Linker */
+
+void set_jump_target(int addr,int target)
+{
+  u_char *ptr=(u_char *)addr;
+  if(*ptr==0x0f)
+  {
+    assert(ptr[1]>=0x80&&ptr[1]<=0x8f);
+    u_int *ptr2=(u_int *)(ptr+2);
+    *ptr2=target-(int)ptr2-4;
+  }
+  else if(*ptr==0xe8||*ptr==0xe9) {
+    u_int *ptr2=(u_int *)(ptr+1);
+    *ptr2=target-(int)ptr2-4;
+  }
+  else
+  {
+    assert(*ptr==0xc7); /* mov immediate (store address) */
+    u_int *ptr2=(u_int *)(ptr+6);
+    *ptr2=target;
+  }
+}
+
+void kill_pointer(void *stub)
+{
+  int i_ptr=*((int *)(stub+6));
+  *((int *)i_ptr)=(int)stub-(int)i_ptr-4;
+}
+int get_pointer(void *stub)
+{
+  int i_ptr=*((int *)(stub+6));
+  return *((int *)i_ptr)+(int)i_ptr+4;
+}
+
+// Find the "clean" entry point from a "dirty" entry point
+// by skipping past the call to verify_code
+u_int get_clean_addr(int addr)
+{
+  u_char *ptr=(u_char *)addr;
+  assert(ptr[21]==0xE8); // call instruction
+  if(ptr[26]==0xE9) return *(u_int *)(ptr+27)+addr+31; // follow jmp
+  else return(addr+26);
+}
+
+int verify_dirty(int addr)
+{
+  u_char *ptr=(u_char *)addr;
+  assert(ptr[0]==0xB8);
+  u_int source=*(u_int *)(ptr+1);
+  u_int copy=*(u_int *)(ptr+6);
+  u_int len=*(u_int *)(ptr+11);
+  //printf("source=%x source-rdram=%x\n",source,source-(int)rdram);
+  assert(ptr[21]==0xE8); // call instruction
+  u_int verifier=*(u_int *)(ptr+22)+(u_int)ptr+26;
+  if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) {
+    unsigned int page=source>>12;
+    unsigned int map_value=memory_map[page];
+    if(map_value>=0x80000000) return 0;
+    while(page<((source+len-1)>>12)) {
+      if((memory_map[++page]<<2)!=(map_value<<2)) return 0;
+    }
+    source = source+(map_value<<2);
+  }
+  //printf("verify_dirty: %x %x %x\n",source,copy,len);
+  return !memcmp((void *)source,(void *)copy,len);
+}
+
+// This doesn't necessarily find all clean entry points, just
+// guarantees that it's not dirty
+int isclean(int addr)
+{
+  u_char *ptr=(u_char *)addr;
+  if(ptr[0]!=0xB8) return 1; // mov imm,%eax
+  if(ptr[5]!=0xBB) return 1; // mov imm,%ebx
+  if(ptr[10]!=0xB9) return 1; // mov imm,%ecx
+  if(ptr[15]!=0x41) return 1; // rex prefix
+  if(ptr[16]!=0xBC) return 1; // mov imm,%r12d
+  if(ptr[21]!=0xE8) return 1; // call instruction
+  return 0;
+}
+
+void get_bounds(int addr,u_int *start,u_int *end)
+{
+  u_char *ptr=(u_char *)addr;
+  assert(ptr[0]==0xB8);
+  u_int source=*(u_int *)(ptr+1);
+  //u_int copy=*(u_int *)(ptr+6);
+  u_int len=*(u_int *)(ptr+11);
+  assert(ptr[21]==0xE8); // call instruction
+  u_int verifier=*(u_int *)(ptr+22)+(u_int)ptr+26;
+  if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) {
+    if(memory_map[source>>12]>=0x80000000) source = 0;
+    else source = source+(memory_map[source>>12]<<2);
+  }
+  *start=source;
+  *end=source+len;
+}
+
+/* Register allocation */
+
+// Note: registers are allocated clean (unmodified state)
+// if you intend to modify the register, you must call dirty_reg().
+void alloc_reg(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = (reg&3)+(reg>28)*4-(reg==32)+2*(reg==36)-(reg==40);
+  
+  // Don't allocate unused registers
+  if((cur->u>>reg)&1) return;
+  
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg) return;
+  }
+  
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+  
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  if(r<64&&((cur->u>>r)&1)) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  if(r>=64&&((cur->uu>>(r&63))&1)) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  
+  // Try to allocate EAX, EBX, ECX, or EDX
+  // We prefer these because they can do byte and halfword loads
+  for(hr=0;hr<4;hr++) {
+    if(cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1)
+          if(i==0||(unneeded_reg[i-1]>>r)&1) {cur->regmap[hr]=-1;break;}
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1)
+          if(i==0||(unneeded_reg_upper[i-1]>>(r&63))&1) {cur->regmap[hr]=-1;break;}
+      }
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen (alloc_reg)");exit(1);
+}
+
+void alloc_reg64(struct regstat *cur,int i,signed char reg)
+{
+  int preferred_reg = 5+reg%3;
+  int r,hr;
+  
+  // allocate the lower 32 bits
+  alloc_reg(cur,i,reg);
+  
+  // Don't allocate unused registers
+  if((cur->uu>>reg)&1) return;
+  
+  // see if the upper half is already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg+64) return;
+  }
+  
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+  
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  if(r<64&&((cur->u>>r)&1)) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  if(r>=64&&((cur->uu>>(r&63))&1)) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  
+  // Try to allocate EBP, ESI or EDI
+  for(hr=5;hr<8;hr++) {
+    if(cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg|64;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;}
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {cur->regmap[hr]=-1;break;}
+      }
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg|64;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg|64;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg|64;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg|64;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg|64;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg|64;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg|64;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen");exit(1);
+}
+
+// Allocate a temporary register.  This is done without regard to
+// dirty status or whether the register we request is on the unneeded list
+// Note: This will only allocate one register, even if called multiple times
+void alloc_reg_temp(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = -1;
+  
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==reg) return;
+  }
+  
+  // Try to allocate any available register, starting with EDI, ESI, EBP...
+  // We prefer EDI, ESI, EBP since the others are used for byte/halfword stores
+  for(hr=HOST_REGS-1;hr>=0;hr--) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Find an unneeded register
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {
+          if(i==0||((unneeded_reg[i-1]>>r)&1)) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {
+          if(i==0||((unneeded_reg_upper[i-1]>>(r&63))&1)) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  // TODO: we might want to follow unconditional jumps here
+  // TODO: get rid of dupe code and make this into a function
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("hsn: %d %d %d %d %d %d %d\n",hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen");exit(1);
+}
+// Allocate a specific x86 register.
+void alloc_x86_reg(struct regstat *cur,int i,signed char reg,char hr)
+{
+  int n;
+  
+  // see if it's already allocated (and dealloc it)
+  for(n=0;n<HOST_REGS;n++)
+  {
+    if(n!=ESP&&cur->regmap[n]==reg) {cur->regmap[n]=-1;}
+  }
+  
+  cur->regmap[hr]=reg;
+  cur->dirty&=~(1<<hr);
+  cur->isconst&=~(1<<hr);
+}
+
+// Alloc cycle count into dedicated register
+alloc_cc(struct regstat *cur,int i)
+{
+  alloc_x86_reg(cur,i,CCREG,ESI);
+}
+
+/* Special alloc */
+
+void multdiv_alloc_x86(struct regstat *current,int i)
+{
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  //  case 0x1C: DMULT
+  //  case 0x1D: DMULTU
+  //  case 0x1E: DDIV
+  //  case 0x1F: DDIVU
+  clear_const(current,rs1[i]);
+  clear_const(current,rs2[i]);
+  if(rs1[i]&&rs2[i])
+  {
+    if((opcode2[i]&4)==0) // 32-bit
+    {
+      current->u&=~(1LL<<HIREG);
+      current->u&=~(1LL<<LOREG);
+      alloc_x86_reg(current,i,HIREG,EDX);
+      alloc_x86_reg(current,i,LOREG,EAX);
+      alloc_reg(current,i,rs1[i]);
+      alloc_reg(current,i,rs2[i]);
+      current->is32|=1LL<<HIREG;
+      current->is32|=1LL<<LOREG;
+      dirty_reg(current,HIREG);
+      dirty_reg(current,LOREG);
+    }
+    else // 64-bit
+    {
+      alloc_x86_reg(current,i,HIREG|64,EDX);
+      alloc_x86_reg(current,i,HIREG,EAX);
+      alloc_reg64(current,i,rs1[i]);
+      alloc_reg64(current,i,rs2[i]);
+      alloc_all(current,i);
+      current->is32&=~(1LL<<HIREG);
+      current->is32&=~(1LL<<LOREG);
+      dirty_reg(current,HIREG);
+      dirty_reg(current,LOREG);
+    }
+  }
+  else
+  {
+    // Multiply by zero is zero.
+    // MIPS does not have a divide by zero exception.
+    // The result is undefined, we return zero.
+    alloc_reg(current,i,HIREG);
+    alloc_reg(current,i,LOREG);
+    current->is32|=1LL<<HIREG;
+    current->is32|=1LL<<LOREG;
+    dirty_reg(current,HIREG);
+    dirty_reg(current,LOREG);
+  }
+}
+#define multdiv_alloc multdiv_alloc_x86
+
+/* Assembler */
+
+char regname[16][4] = {
+ "eax",
+ "ecx",
+ "edx",
+ "ebx",
+ "esp",
+ "ebp",
+ "esi",
+ "edi",
+ "r8",
+ "r9",
+ "r10",
+ "r11",
+ "r12",
+ "r13",
+ "r14",
+ "r15"};
+
+void output_byte(u_char byte)
+{
+  *(out++)=byte;
+}
+void output_modrm(u_char mod,u_char rm,u_char ext)
+{
+  assert(mod<4);
+  assert(rm<8);
+  assert(ext<8);
+  u_char byte=(mod<<6)|(ext<<3)|rm;
+  *(out++)=byte;
+}
+void output_sib(u_char scale,u_char index,u_char base)
+{
+  assert(scale<4);
+  assert(index<8);
+  assert(base<8);
+  u_char byte=(scale<<6)|(index<<3)|base;
+  *(out++)=byte;
+}
+void output_rex(u_char w,u_char r,u_char x,u_char b)
+{
+  assert(w<2);
+  assert(r<2);
+  assert(x<2);
+  assert(b<2);
+  u_char byte=0x40|(w<<3)|(r<<2)|(x<<1)|b;
+  *(out++)=byte;
+}
+void output_w32(u_int word)
+{
+  *((u_int *)out)=word;
+  out+=4;
+}
+
+void emit_mov(int rs,int rt)
+{
+  assem_debug("mov %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x89);
+  output_modrm(3,rt,rs);
+}
+
+void emit_mov64(int rs,int rt)
+{
+  assem_debug("mov %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_rex(1,0,0,rt>>3);
+  output_byte(0x89);
+  output_modrm(3,rt,rs);
+}
+
+void emit_add(int rs1,int rs2,int rt)
+{
+  if(rs1==rt) {
+    assem_debug("add %%%s,%%%s\n",regname[rs2],regname[rs1]);
+    output_byte(0x01);
+    output_modrm(3,rs1,rs2);
+  }else if(rs2==rt) {
+    assem_debug("add %%%s,%%%s\n",regname[rs1],regname[rs2]);
+    output_byte(0x01);
+    output_modrm(3,rs2,rs1);
+  }else {
+    assem_debug("lea (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+    output_byte(0x8D);
+    if(rs1!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,rs2,rs1);
+    }else if(rs2!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,rs1,rs2);
+    }else /* lea 0(,%ebp,2) */{
+      output_modrm(0,4,rt);
+      output_sib(1,EBP,5);
+      output_w32(0);
+    }
+  }
+}
+
+void emit_adds(int rs1,int rs2,int rt)
+{
+  emit_add(rs1,rs2,rt);
+}
+
+void emit_lea8(int rs1,int rt)
+{
+  assem_debug("lea 0(%%%s,8),%%%s\n",regname[rs1],regname[rt]);
+  output_byte(0x8D);
+  output_modrm(0,4,rt);
+  output_sib(3,rs1,5);
+  output_w32(0);
+}
+void emit_leairrx1(int imm,int rs1,int rs2,int rt)
+{
+  assem_debug("lea %x(%%%s,%%%s,1),%%%s\n",imm,regname[rs1],regname[rs2],regname[rt]);
+  output_byte(0x8D);
+  if(imm!=0||rs1==EBP) {
+    output_modrm(2,4,rt);
+    output_sib(0,rs2,rs1);
+    output_w32(imm);
+  }else{
+    output_modrm(0,4,rt);
+    output_sib(0,rs2,rs1);
+  }
+}
+void emit_leairrx4(int imm,int rs1,int rs2,int rt)
+{
+  assem_debug("lea %x(%%%s,%%%s,4),%%%s\n",imm,regname[rs1],regname[rs2],regname[rt]);
+  output_byte(0x8D);
+  if(imm!=0||rs1==EBP) {
+    output_modrm(2,4,rt);
+    output_sib(2,rs2,rs1);
+    output_w32(imm);
+  }else{
+    output_modrm(0,4,rt);
+    output_sib(2,rs2,rs1);
+  }
+}
+
+void emit_neg(int rs, int rt)
+{
+  if(rs!=rt) emit_mov(rs,rt);
+  assem_debug("neg %%%s\n",regname[rt]);
+  output_byte(0xF7);
+  output_modrm(3,rt,3);
+}
+
+void emit_negs(int rs, int rt)
+{
+  emit_neg(rs,rt);
+}
+
+void emit_sub(int rs1,int rs2,int rt)
+{
+  if(rs1==rt) {
+    assem_debug("sub %%%s,%%%s\n",regname[rs2],regname[rs1]);
+    output_byte(0x29);
+    output_modrm(3,rs1,rs2);
+  } else if(rs2==rt) {
+    emit_neg(rs2,rs2);
+    emit_add(rs2,rs1,rs2);
+  } else {
+    emit_mov(rs1,rt);
+    emit_sub(rt,rs2,rt);
+  }
+}
+
+void emit_subs(int rs1,int rs2,int rt)
+{
+  emit_sub(rs1,rs2,rt);
+}
+
+void emit_zeroreg(int rt)
+{
+  output_byte(0x31);
+  output_modrm(3,rt,rt);
+  assem_debug("xor %%%s,%%%s\n",regname[rt],regname[rt]);
+}
+
+void emit_loadreg(int r, int hr)
+{
+  if((r&63)==0)
+    emit_zeroreg(hr);
+  else {
+    int addr=((int)reg)+((r&63)<<3)+((r&64)>>4);
+    if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
+    if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
+    if(r==CCREG) addr=(int)&cycle_count;
+    if(r==CSREG) addr=(int)&Status;
+    if(r==FSREG) addr=(int)&FCR31;
+    assem_debug("mov %x+%d,%%%s\n",addr,r,regname[hr]);
+    output_byte(0x8B);
+    output_modrm(0,5,hr);
+    output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+  }
+}
+void emit_storereg(int r, int hr)
+{
+  int addr=((int)reg)+((r&63)<<3)+((r&64)>>4);
+  if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
+  if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
+  if(r==CCREG) addr=(int)&cycle_count;
+  if(r==FSREG) addr=(int)&FCR31;
+  assem_debug("mov %%%s,%x+%d\n",regname[hr],addr,r);
+  output_byte(0x89);
+  output_modrm(0,5,hr);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+
+void emit_test(int rs, int rt)
+{
+  assem_debug("test %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x85);
+  output_modrm(3,rs,rt);
+}
+
+void emit_testimm(int rs,int imm)
+{
+  assem_debug("test $0x%x,%%%s\n",imm,regname[rs]);
+  if(imm<128&&imm>=-128&&rs<4) {
+    output_byte(0xF6);
+    output_modrm(3,rs,0);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0xF7);
+    output_modrm(3,rs,0);
+    output_w32(imm);
+  }
+}
+
+void emit_not(int rs,int rt)
+{
+  if(rs!=rt) emit_mov(rs,rt);
+  assem_debug("not %%%s\n",regname[rt]);
+  output_byte(0xF7);
+  output_modrm(3,rt,2);
+}
+
+void emit_and(u_int rs1,u_int rs2,u_int rt)
+{
+  assert(rs1<8);
+  assert(rs2<8);
+  assert(rt<8);
+  if(rs1==rt) {
+    assem_debug("and %%%s,%%%s\n",regname[rs2],regname[rt]);
+    output_byte(0x21);
+    output_modrm(3,rs1,rs2);
+  }
+  else
+  if(rs2==rt) {
+    assem_debug("and %%%s,%%%s\n",regname[rs1],regname[rt]);
+    output_byte(0x21);
+    output_modrm(3,rs2,rs1);
+  }
+  else {
+    emit_mov(rs1,rt);
+    emit_and(rt,rs2,rt);
+  }
+}
+
+void emit_or(u_int rs1,u_int rs2,u_int rt)
+{
+  assert(rs1<8);
+  assert(rs2<8);
+  assert(rt<8);
+  if(rs1==rt) {
+    assem_debug("or %%%s,%%%s\n",regname[rs2],regname[rt]);
+    output_byte(0x09);
+    output_modrm(3,rs1,rs2);
+  }
+  else
+  if(rs2==rt) {
+    assem_debug("or %%%s,%%%s\n",regname[rs1],regname[rt]);
+    output_byte(0x09);
+    output_modrm(3,rs2,rs1);
+  }
+  else {
+    emit_mov(rs1,rt);
+    emit_or(rt,rs2,rt);
+  }
+}
+void emit_or_and_set_flags(int rs1,int rs2,int rt)
+{
+  emit_or(rs1,rs2,rt);
+}
+
+void emit_xor(u_int rs1,u_int rs2,u_int rt)
+{
+  assert(rs1<8);
+  assert(rs2<8);
+  assert(rt<8);
+  if(rs1==rt) {
+    assem_debug("xor %%%s,%%%s\n",regname[rs2],regname[rt]);
+    output_byte(0x31);
+    output_modrm(3,rs1,rs2);
+  }
+  else
+  if(rs2==rt) {
+    assem_debug("xor %%%s,%%%s\n",regname[rs1],regname[rt]);
+    output_byte(0x31);
+    output_modrm(3,rs2,rs1);
+  }
+  else {
+    emit_mov(rs1,rt);
+    emit_xor(rt,rs2,rt);
+  }
+}
+
+void emit_movimm(int imm,u_int rt)
+{
+  assem_debug("mov $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<16);
+  if(rt>=8) output_rex(0,0,0,1);
+  output_byte(0xB8+(rt&7));
+  output_w32(imm);
+}
+
+void emit_addimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    if(imm!=0) {
+      assem_debug("add $%d,%%%s\n",imm,regname[rt]);
+      if(imm<128&&imm>=-128) {
+        output_byte(0x83);
+        output_modrm(3,rt,0);
+        output_byte(imm);
+      }
+      else
+      {
+        output_byte(0x81);
+        output_modrm(3,rt,0);
+        output_w32(imm);
+      }
+    }
+  }
+  else {
+    if(imm!=0) {
+      assem_debug("lea %d(%%%s),%%%s\n",imm,regname[rs],regname[rt]);
+      output_byte(0x8D);
+      if(imm<128&&imm>=-128) {
+        output_modrm(1,rs,rt);
+        output_byte(imm);
+      }else{
+        output_modrm(2,rs,rt);
+        output_w32(imm);
+      }
+    }else{
+      emit_mov(rs,rt);
+    }
+  }
+}
+
+void emit_addimm64(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    if(imm!=0) {
+      assem_debug("add $%d,%%%s\n",imm,regname[rt]);
+      if(imm<128&&imm>=-128) {
+        output_rex(1,0,0,rt>>3);
+        output_byte(0x83);
+        output_modrm(3,rt&7,0);
+        output_byte(imm);
+      }
+      else
+      {
+        output_rex(1,0,0,rt>>3);
+        output_byte(0x81);
+        output_modrm(3,rt&7,0);
+        output_w32(imm);
+      }
+    }
+  }
+  else {
+    if(imm!=0) {
+      assem_debug("lea %d(%%%s),%%%s\n",imm,regname[rs],regname[rt]);
+      output_rex(1,rt>>3,0,rs>>3);
+      output_byte(0x8D);
+      if(imm<128&&imm>=-128) {
+        output_modrm(1,rs&7,rt&7);
+        output_byte(imm);
+      }else{
+        output_modrm(2,rs&7,rt&7);
+        output_w32(imm);
+      }
+    }else{
+      emit_mov(rs,rt);
+    }
+  }
+}
+
+void emit_addimm_and_set_flags(int imm,int rt)
+{
+  assem_debug("add $%d,%%%s\n",imm,regname[rt]);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,0);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,0);
+    output_w32(imm);
+  }
+}
+void emit_addimm_no_flags(int imm,int rt)
+{
+  if(imm!=0) {
+    assem_debug("lea %d(%%%s),%%%s\n",imm,regname[rt],regname[rt]);
+    output_byte(0x8D);
+    if(imm<128&&imm>=-128) {
+      output_modrm(1,rt,rt);
+      output_byte(imm);
+    }else{
+      output_modrm(2,rt,rt);
+      output_w32(imm);
+    }
+  }
+}
+
+void emit_adcimm(int imm,u_int rt)
+{
+  assem_debug("adc $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<8);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,2);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,2);
+    output_w32(imm);
+  }
+}
+void emit_sbbimm(int imm,u_int rt)
+{
+  assem_debug("sbb $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<8);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,3);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,3);
+    output_w32(imm);
+  }
+}
+
+void emit_addimm64_32(int rsh,int rsl,int imm,int rth,int rtl)
+{
+  if(rsh==rth&&rsl==rtl) {
+    assem_debug("add $%d,%%%s\n",imm,regname[rtl]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rtl,0);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rtl,0);
+      output_w32(imm);
+    }
+    assem_debug("adc $%d,%%%s\n",imm>>31,regname[rth]);
+    output_byte(0x83);
+    output_modrm(3,rth,2);
+    output_byte(imm>>31);
+  }
+  else {
+    emit_mov(rsh,rth);
+    emit_mov(rsl,rtl);
+    emit_addimm64_32(rth,rtl,imm,rth,rtl);
+  }
+}
+
+void emit_sbb(int rs1,int rs2)
+{
+  assem_debug("sbb %%%s,%%%s\n",regname[rs2],regname[rs1]);
+  output_byte(0x19);
+  output_modrm(3,rs1,rs2);
+}
+
+void emit_andimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("and $%d,%%%s\n",imm,regname[rt]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rt,4);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rt,4);
+      output_w32(imm);
+    }
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_andimm(rt,imm,rt);
+  }
+}
+
+void emit_orimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("or $%d,%%%s\n",imm,regname[rt]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rt,1);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rt,1);
+      output_w32(imm);
+    }
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_orimm(rt,imm,rt);
+  }
+}
+
+void emit_xorimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("xor $%d,%%%s\n",imm,regname[rt]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rt,6);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rt,6);
+      output_w32(imm);
+    }
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_xorimm(rt,imm,rt);
+  }
+}
+
+void emit_shlimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shl %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,4);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shlimm(rt,imm,rt);
+  }
+}
+
+void emit_shrimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shr %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,5);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shrimm(rt,imm,rt);
+  }
+}
+
+void emit_shrimm64(int rs,u_int imm,int rt)
+{
+  assert(rs==rt);
+  if(rs==rt) {
+    assem_debug("shr %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    output_rex(1,0,0,rt>>3);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,5);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shrimm(rt,imm,rt);
+  }
+}
+
+void emit_sarimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("sar %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,7);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_sarimm(rt,imm,rt);
+  }
+}
+
+void emit_rorimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("ror %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,1);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_sarimm(rt,imm,rt);
+  }
+}
+
+void emit_shldimm(int rs,int rs2,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shld %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm);
+    assert(imm>0);
+    output_byte(0x0F);
+    output_byte(0xA4);
+    output_modrm(3,rt,rs2);
+    output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shldimm(rt,rs2,imm,rt);
+  }
+}
+
+void emit_shrdimm(int rs,int rs2,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shrd %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm);
+    assert(imm>0);
+    output_byte(0x0F);
+    output_byte(0xAC);
+    output_modrm(3,rt,rs2);
+    output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shrdimm(rt,rs2,imm,rt);
+  }
+}
+
+void emit_shlcl(int r)
+{
+  assem_debug("shl %%%s,%%cl\n",regname[r]);
+  output_byte(0xD3);
+  output_modrm(3,r,4);
+}
+void emit_shrcl(int r)
+{
+  assem_debug("shr %%%s,%%cl\n",regname[r]);
+  output_byte(0xD3);
+  output_modrm(3,r,5);
+}
+void emit_sarcl(int r)
+{
+  assem_debug("sar %%%s,%%cl\n",regname[r]);
+  output_byte(0xD3);
+  output_modrm(3,r,7);
+}
+
+void emit_shldcl(int r1,int r2)
+{
+  assem_debug("shld %%%s,%%%s,%%cl\n",regname[r1],regname[r2]);
+  output_byte(0x0F);
+  output_byte(0xA5);
+  output_modrm(3,r1,r2);
+}
+void emit_shrdcl(int r1,int r2)
+{
+  assem_debug("shrd %%%s,%%%s,%%cl\n",regname[r1],regname[r2]);
+  output_byte(0x0F);
+  output_byte(0xAD);
+  output_modrm(3,r1,r2);
+}
+
+void emit_cmpimm(int rs,int imm)
+{
+  assem_debug("cmp $%d,%%%s\n",imm,regname[rs]);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rs,7);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rs,7);
+    output_w32(imm);
+  }
+}
+
+void emit_cmovne(u_int *addr,int rt)
+{
+  assem_debug("cmovne %x,%%%s",(int)addr,regname[rt]);
+  if(addr==&const_zero) assem_debug(" [zero]\n");
+  else if(addr==&const_one) assem_debug(" [one]\n");
+  else assem_debug("\n");
+  output_byte(0x0F);
+  output_byte(0x45);
+  output_modrm(0,5,rt);
+  output_w32((int)addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_cmovl(u_int *addr,int rt)
+{
+  assem_debug("cmovl %x,%%%s",(int)addr,regname[rt]);
+  if(addr==&const_zero) assem_debug(" [zero]\n");
+  else if(addr==&const_one) assem_debug(" [one]\n");
+  else assem_debug("\n");
+  output_byte(0x0F);
+  output_byte(0x4C);
+  output_modrm(0,5,rt);
+  output_w32((int)addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_cmovs(u_int *addr,int rt)
+{
+  assem_debug("cmovs %x,%%%s",(int)addr,regname[rt]);
+  if(addr==&const_zero) assem_debug(" [zero]\n");
+  else if(addr==&const_one) assem_debug(" [one]\n");
+  else assem_debug("\n");
+  output_byte(0x0F);
+  output_byte(0x48);
+  output_modrm(0,5,rt);
+  output_w32((int)addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_cmovne_reg(int rs,int rt)
+{
+  assem_debug("cmovne %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x45);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovl_reg(int rs,int rt)
+{
+  assem_debug("cmovl %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x4C);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovs_reg(int rs,int rt)
+{
+  assem_debug("cmovs %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x48);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovnc_reg(int rs,int rt)
+{
+  assem_debug("cmovae %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x43);
+  output_modrm(3,rs,rt);
+}
+void emit_cmova_reg(int rs,int rt)
+{
+  assem_debug("cmova %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x47);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovp_reg(int rs,int rt)
+{
+  assem_debug("cmovp %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x4A);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovnp_reg(int rs,int rt)
+{
+  assem_debug("cmovnp %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x4B);
+  output_modrm(3,rs,rt);
+}
+void emit_setl(int rt)
+{
+  assem_debug("setl %%%s\n",regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x9C);
+  output_modrm(3,rt,2);
+}
+void emit_movzbl_reg(int rs, int rt)
+{
+  assem_debug("movzbl %%%s,%%%s\n",regname[rs]+1,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB6);
+  output_modrm(3,rs,rt);
+}
+
+void emit_slti32(int rs,int imm,int rt)
+{
+  if(rs!=rt) emit_zeroreg(rt);
+  emit_cmpimm(rs,imm);
+  if(rt<4) {
+    emit_setl(rt);
+    if(rs==rt) emit_movzbl_reg(rt,rt);
+  }
+  else
+  {
+    if(rs==rt) emit_movimm(0,rt);
+    emit_cmovl(&const_one,rt);
+  }
+}
+void emit_sltiu32(int rs,int imm,int rt)
+{
+  if(rs!=rt) emit_zeroreg(rt);
+  emit_cmpimm(rs,imm);
+  if(rs==rt) emit_movimm(0,rt);
+  emit_adcimm(0,rt);
+}
+void emit_slti64_32(int rsh,int rsl,int imm,int rt)
+{
+  assert(rsh!=rt);
+  emit_slti32(rsl,imm,rt);
+  if(imm>=0)
+  {
+    emit_test(rsh,rsh);
+    emit_cmovne(&const_zero,rt);
+    emit_cmovs(&const_one,rt);
+  }
+  else
+  {
+    emit_cmpimm(rsh,-1);
+    emit_cmovne(&const_zero,rt);
+    emit_cmovl(&const_one,rt);
+  }
+}
+void emit_sltiu64_32(int rsh,int rsl,int imm,int rt)
+{
+  assert(rsh!=rt);
+  emit_sltiu32(rsl,imm,rt);
+  if(imm>=0)
+  {
+    emit_test(rsh,rsh);
+    emit_cmovne(&const_zero,rt);
+  }
+  else
+  {
+    emit_cmpimm(rsh,-1);
+    emit_cmovne(&const_one,rt);
+  }
+}
+
+void emit_cmp(int rs,int rt)
+{
+  assem_debug("cmp %%%s,%%%s\n",regname[rt],regname[rs]);
+  output_byte(0x39);
+  output_modrm(3,rs,rt);
+}
+void emit_set_gz32(int rs, int rt)
+{
+  //assem_debug("set_gz32\n");
+  emit_cmpimm(rs,1);
+  emit_movimm(1,rt);
+  emit_cmovl(&const_zero,rt);
+}
+void emit_set_nz32(int rs, int rt)
+{
+  //assem_debug("set_nz32\n");
+  emit_cmpimm(rs,1);
+  emit_movimm(1,rt);
+  emit_sbbimm(0,rt);
+}
+void emit_set_gz64_32(int rsh, int rsl, int rt)
+{
+  //assem_debug("set_gz64\n");
+  emit_set_gz32(rsl,rt);
+  emit_test(rsh,rsh);
+  emit_cmovne(&const_one,rt);
+  emit_cmovs(&const_zero,rt);
+}
+void emit_set_nz64_32(int rsh, int rsl, int rt)
+{
+  //assem_debug("set_nz64\n");
+  emit_or_and_set_flags(rsh,rsl,rt);
+  emit_cmovne(&const_one,rt);
+}
+void emit_set_if_less32(int rs1, int rs2, int rt)
+{
+  //assem_debug("set if less (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+  if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt);
+  emit_cmp(rs1,rs2);
+  if(rs1==rt||rs2==rt) emit_movimm(0,rt);
+  emit_cmovl(&const_one,rt);
+}
+void emit_set_if_carry32(int rs1, int rs2, int rt)
+{
+  //assem_debug("set if carry (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+  if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt);
+  emit_cmp(rs1,rs2);
+  if(rs1==rt||rs2==rt) emit_movimm(0,rt);
+  emit_adcimm(0,rt);
+}
+void emit_set_if_less64_32(int u1, int l1, int u2, int l2, int rt)
+{
+  //assem_debug("set if less64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]);
+  assert(u1!=rt);
+  assert(u2!=rt);
+  emit_cmp(l1,l2);
+  emit_mov(u1,rt);
+  emit_sbb(rt,u2);
+  emit_movimm(0,rt);
+  emit_cmovl(&const_one,rt);
+}
+void emit_set_if_carry64_32(int u1, int l1, int u2, int l2, int rt)
+{
+  //assem_debug("set if carry64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]);
+  assert(u1!=rt);
+  assert(u2!=rt);
+  emit_cmp(l1,l2);
+  emit_mov(u1,rt);
+  emit_sbb(rt,u2);
+  emit_movimm(0,rt);
+  emit_adcimm(0,rt);
+}
+
+void emit_call(int a)
+{
+  assem_debug("call %x (%x+%x)\n",a,(int)out+5,a-(int)out-5);
+  output_byte(0xe8);
+  output_w32(a-(int)out-4);
+}
+void emit_jmp(int a)
+{
+  assem_debug("jmp %x (%x+%x)\n",a,(int)out+5,a-(int)out-5);
+  output_byte(0xe9);
+  output_w32(a-(int)out-4);
+}
+void emit_jne(int a)
+{
+  assem_debug("jne %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x85);
+  output_w32(a-(int)out-4);
+}
+void emit_jeq(int a)
+{
+  assem_debug("jeq %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x84);
+  output_w32(a-(int)out-4);
+}
+void emit_js(int a)
+{
+  assem_debug("js %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x88);
+  output_w32(a-(int)out-4);
+}
+void emit_jns(int a)
+{
+  assem_debug("jns %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x89);
+  output_w32(a-(int)out-4);
+}
+void emit_jl(int a)
+{
+  assem_debug("jl %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x8c);
+  output_w32(a-(int)out-4);
+}
+void emit_jge(int a)
+{
+  assem_debug("jge %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x8d);
+  output_w32(a-(int)out-4);
+}
+void emit_jno(int a)
+{
+  assem_debug("jno %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x81);
+  output_w32(a-(int)out-4);
+}
+void emit_jc(int a)
+{
+  assem_debug("jc %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x82);
+  output_w32(a-(int)out-4);
+}
+
+void emit_pushimm(int imm)
+{
+  assem_debug("push $%x\n",imm);
+  output_byte(0x68);
+  output_w32(imm);
+}
+//void emit_pusha()
+//{
+//  assem_debug("pusha\n");
+//  output_byte(0x60);
+//}
+//void emit_popa()
+//{
+//  assem_debug("popa\n");
+//  output_byte(0x61);
+//}
+void emit_pushreg(u_int r)
+{
+  assem_debug("push %%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0x50+r);
+}
+void emit_popreg(u_int r)
+{
+  assem_debug("pop %%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0x58+r);
+}
+void emit_callreg(u_int r)
+{
+  assem_debug("call *%%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0xFF);
+  output_modrm(3,r,2);
+}
+void emit_jmpreg(u_int r)
+{
+  assem_debug("jmp *%%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0xFF);
+  output_modrm(3,r,4);
+}
+void emit_jmpmem_indexed(u_int addr,u_int r)
+{
+  assem_debug("jmp *%x(%%%s)\n",addr,regname[r]);
+  assert(r<8);
+  output_byte(0xFF);
+  output_modrm(2,r,4);
+  output_w32(addr);
+}
+
+void emit_readword(int addr, int rt)
+{
+  assem_debug("mov %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x8B);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_readword_indexed(int addr, int rs, int rt)
+{
+  assem_debug("mov %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x8B);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_w32(addr);
+  }
+}
+void emit_readword_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_readword(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("addr32 mov %x(,%%%s,4),%%%s\n",addr,regname[map],regname[rt]);
+    output_byte(0x67);
+    output_byte(0x8B);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr);
+  }
+}
+void emit_readword_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_readword_indexed(addr+(int)rdram-0x80000000, rs, rt);
+  else {
+    assem_debug("addr32 mov %x(%%%s,%%%s,4),%%%s\n",addr,regname[rs],regname[map],regname[rt]);
+    assert(rs!=ESP);
+    output_byte(0x67);
+    output_byte(0x8B);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(2,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(2,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(2,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_movmem_indexedx4(int addr, int rs, int rt)
+{
+  assem_debug("mov (%x,%%%s,4),%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x8B);
+  output_modrm(0,4,rt);
+  output_sib(2,rs,5);
+  output_w32(addr);
+}
+void emit_movmem_indexedx4_addr32(int addr, int rs, int rt)
+{
+  assem_debug("addr32 mov (%x,%%%s,4),%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x67);
+  output_byte(0x8B);
+  output_modrm(0,4,rt);
+  output_sib(2,rs,5);
+  output_w32(addr);
+}
+void emit_movmem_indexedx8(int addr, int rs, int rt)
+{
+  assem_debug("mov (%x,%%%s,8),%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x8B);
+  output_modrm(0,4,rt);
+  output_sib(3,rs,5);
+  output_w32(addr);
+}
+void emit_readdword_tlb(int addr, int map, int rh, int rl)
+{
+  if(map<0) {
+    if(rh>=0) emit_readword(addr+(int)rdram-0x80000000, rh);
+    emit_readword(addr+(int)rdram-0x7FFFFFFC, rl);
+  }
+  else {
+    if(rh>=0) emit_movmem_indexedx4_addr32(addr, map, rh);
+    emit_movmem_indexedx4_addr32(addr+4, map, rl);
+  }
+}
+void emit_readdword_indexed(int addr, int rs, int rt)
+{
+  assem_debug("mov %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_rex(1,rt>>3,0,rs>>3);
+  output_byte(0x8B);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs&7,rt&7);
+    if(rs==ESP) output_sib(0,4,4);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs&7,rt&7);
+    if(rs==ESP) output_sib(0,4,4);
+    output_w32(addr);
+  }
+}
+void emit_readdword_indexed_tlb(int addr, int rs, int map, int rh, int rl)
+{
+  assert(rh!=rs);
+  if(rh>=0) emit_readword_indexed_tlb(addr, rs, map, rh);
+  emit_readword_indexed_tlb(addr+4, rs, map, rl);
+}
+void emit_movsbl(int addr, int rt)
+{
+  assem_debug("movsbl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBE);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_movsbl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movsbl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBE);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movsbl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movsbl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("addr32 movsbl %x(,%%%s,4),%%%s\n",addr,regname[map],regname[rt]);
+    output_byte(0x67);
+    output_byte(0x0F);
+    output_byte(0xBE);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr);
+  }
+}
+void emit_movsbl_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_movsbl_indexed(addr+(int)rdram-0x80000000, rs, rt);
+  else {
+    assem_debug("addr32 movsbl %x(%%%s,%%%s,4),%%%s\n",addr,regname[rs],regname[map],regname[rt]);
+    assert(rs!=ESP);
+    output_byte(0x67);
+    output_byte(0x0F);
+    output_byte(0xBE);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(2,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(2,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(2,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_movswl(int addr, int rt)
+{
+  assem_debug("movswl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBF);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_movswl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movswl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBF);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movswl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movswl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("addr32 movswl %x(,%%%s,4),%%%s\n",addr,regname[map],regname[rt]);
+    output_byte(0x67);
+    output_byte(0x0F);
+    output_byte(0xBF);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr);
+  }
+}
+void emit_movzbl(int addr, int rt)
+{
+  assem_debug("movzbl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB6);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_movzbl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movzbl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB6);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movzbl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movzbl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("addr32 movzbl %x(,%%%s,4),%%%s\n",addr,regname[map],regname[rt]);
+    output_byte(0x67);
+    output_byte(0x0F);
+    output_byte(0xB6);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr);
+  }
+}
+void emit_movzbl_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_movzbl_indexed(addr+(int)rdram-0x80000000, rs, rt);
+  else {
+    assem_debug("addr32 movzbl %x(%%%s,%%%s,4),%%%s\n",addr,regname[rs],regname[map],regname[rt]);
+    assert(rs!=ESP);
+    output_byte(0x67);
+    output_byte(0x0F);
+    output_byte(0xB6);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(2,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(2,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(2,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_movzwl(int addr, int rt)
+{
+  assem_debug("movzwl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB7);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_movzwl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movzwl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB7);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movzwl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movzwl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("addr32 movzwl %x(,%%%s,4),%%%s\n",addr,regname[map],regname[rt]);
+    output_byte(0x67);
+    output_byte(0x0F);
+    output_byte(0xB7);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr);
+  }
+}
+void emit_movzwl_reg(int rs, int rt)
+{
+  assem_debug("movzwl %%%s,%%%s\n",regname[rs]+1,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB7);
+  output_modrm(3,rs,rt);
+}
+
+void emit_xchg(int rs, int rt)
+{
+  assem_debug("xchg %%%s,%%%s\n",regname[rs],regname[rt]);
+  if(rs==EAX) {
+    output_byte(0x90+rt);
+  }
+  else
+  {
+    output_byte(0x87);
+    output_modrm(3,rs,rt);
+  }
+}
+void emit_writeword(int rt, int addr)
+{
+  assem_debug("movl %%%s,%x\n",regname[rt],addr);
+  output_byte(0x89);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_writeword_indexed(int rt, int addr, int rs)
+{
+  assem_debug("mov %%%s,%x+%%%s\n",regname[rt],addr,regname[rs]);
+  output_byte(0x89);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_w32(addr);
+  }
+}
+void emit_writeword_tlb(int rt, int addr, int map)
+{
+  if(map<0) {
+    emit_writeword(rt, addr+(int)rdram-0x80000000);
+  } else {
+    emit_writeword_indexed(rt, addr, map);
+  }
+}
+void emit_writeword_indexed_tlb(int rt, int addr, int rs, int map, int temp)
+{
+  if(map<0) emit_writeword_indexed(rt, addr+(int)rdram-0x80000000, rs);
+  else {
+    assem_debug("addr32 mov %%%s,%x(%%%s,%%%s,1)\n",regname[rt],addr,regname[rs],regname[map]);
+    assert(rs!=ESP);
+    output_byte(0x67);
+    output_byte(0x89);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(0,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(0,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_writedword_tlb(int rh, int rl, int addr, int map)
+{
+  assert(rh>=0);
+  if(map<0) {
+    emit_writeword(rh, addr+(int)rdram-0x80000000);
+    emit_writeword(rl, addr+(int)rdram-0x7FFFFFFC);
+  }
+  else {
+    emit_writeword_indexed(rh, addr, map);
+    emit_writeword_indexed(rl, addr+4, map);
+  }
+}
+void emit_writedword_indexed_tlb(int rh, int rl, int addr, int rs, int map, int temp)
+{
+  assert(rh>=0);
+  emit_writeword_indexed_tlb(rh, addr, rs, map, temp);
+  emit_writeword_indexed_tlb(rl, addr+4, rs, map, temp);
+}
+void emit_writehword(int rt, int addr)
+{
+  assem_debug("movw %%%s,%x\n",regname[rt]+1,addr);
+  output_byte(0x66);
+  output_byte(0x89);
+  output_modrm(0,5,rt);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_writehword_indexed(int rt, int addr, int rs)
+{
+  assem_debug("movw %%%s,%x+%%%s\n",regname[rt]+1,addr,regname[rs]);
+  output_byte(0x66);
+  output_byte(0x89);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs,rt);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs,rt);
+    output_w32(addr);
+  }
+}
+void emit_writehword_tlb(int rt, int addr, int map)
+{
+  if(map<0) {
+    emit_writehword(rt, addr+(int)rdram-0x80000000);
+  } else {
+    emit_writehword_indexed(rt, addr, map);
+  }
+}
+void emit_writebyte(int rt, int addr)
+{
+  if(rt<4) {
+    assem_debug("movb %%%cl,%x\n",regname[rt][1],addr);
+    output_byte(0x88);
+    output_modrm(0,5,rt);
+    output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+  }
+  else
+  {
+    emit_xchg(EAX,rt);
+    emit_writebyte(EAX,addr);
+    emit_xchg(EAX,rt);
+  }
+}
+void emit_writebyte_indexed(int rt, int addr, int rs)
+{
+  if(rt<4) {
+    assem_debug("movb %%%cl,%x+%%%s\n",regname[rt][1],addr,regname[rs]);
+    output_byte(0x88);
+    if(addr<128&&addr>=-128) {
+      output_modrm(1,rs,rt);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,rs,rt);
+      output_w32(addr);
+    }
+  }
+  else
+  {
+    emit_xchg(EAX,rt);
+    emit_writebyte_indexed(EAX,addr,rs==EAX?rt:rs);
+    emit_xchg(EAX,rt);
+  }
+}
+void emit_writebyte_tlb(int rt, int addr, int map)
+{
+  if(map<0) {
+    emit_writebyte(rt, addr+(int)rdram-0x80000000);
+  } else {
+    emit_writebyte_indexed(rt, addr, map);
+  }
+}
+void emit_writebyte_indexed_tlb(int rt, int addr, int rs, int map, int temp)
+{
+  if(map<0) emit_writebyte_indexed(rt, addr+(int)rdram-0x80000000, rs);
+  else
+  if(rt<4) {
+    assem_debug("addr32 movb %%%cl,%x(%%%s,%%%s,1)\n",regname[rt][1],addr,regname[rs],regname[map]);
+    assert(rs!=ESP);
+    output_byte(0x67);
+    output_byte(0x88);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(0,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(0,map,rs);
+      output_w32(addr);
+    }
+  }
+  else
+  {
+    emit_xchg(EAX,rt);
+    emit_writebyte_indexed_tlb(EAX,addr,rs==EAX?rt:rs,map==EAX?rt:map,temp);
+    emit_xchg(EAX,rt);
+  }
+}
+void emit_writeword_imm(int imm, int addr)
+{
+  assem_debug("movl $%x,%x\n",imm,addr);
+  output_byte(0xC7);
+  output_modrm(0,5,0);
+  output_w32(addr-(int)out-8); // Note: rip-relative in 64-bit mode
+  output_w32(imm);
+}
+void emit_writeword_imm_esp(int imm, int addr)
+{
+  assem_debug("mov $%x,%x(%%esp)\n",imm,addr);
+  assert(addr>=-128&&addr<128);
+  output_byte(0xC7);
+  output_modrm(!!addr,4,0);
+  output_sib(0,4,4);
+  if(addr) output_byte(addr);
+  output_w32(imm);
+}
+void emit_writedword_imm32(int imm, int addr)
+{
+  assem_debug("movq $%x,%x\n",imm,addr);
+  output_rex(1,0,0,0);
+  output_byte(0xC7);
+  output_modrm(0,5,0);
+  output_w32(addr-(int)out-8); // Note: rip-relative in 64-bit mode
+  output_w32(imm); // Note: This 32-bit value will be sign extended
+}
+void emit_writebyte_imm(int imm, int addr)
+{
+  assem_debug("movb $%x,%x\n",imm,addr);
+  assert(imm>=-128&&imm<128);
+  output_byte(0xC6);
+  output_modrm(0,5,0);
+  output_w32(addr-(int)out-5); // Note: rip-relative in 64-bit mode
+  output_byte(imm);
+}
+
+void emit_mul(int rs)
+{
+  assem_debug("mul %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,4);
+}
+void emit_imul(int rs)
+{
+  assem_debug("imul %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,5);
+}
+void emit_div(int rs)
+{
+  assem_debug("div %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,6);
+}
+void emit_idiv(int rs)
+{
+  assem_debug("idiv %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,7);
+}
+void emit_cdq()
+{
+  assem_debug("cdq\n");
+  output_byte(0x99);
+}
+
+// Load 2 immediates optimizing for small code size
+void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
+{
+  emit_movimm(imm1,rt1);
+  if(imm2-imm1<128&&imm2-imm1>=-128) emit_addimm(rt1,imm2-imm1,rt2);
+  else emit_movimm(imm2,rt2);
+}
+
+// special case for checking pending_exception
+void emit_cmpmem_imm_byte(int addr,int imm)
+{
+  assert(imm<128&&imm>=-127);
+  assem_debug("cmpb $%d,%x\n",imm,addr);
+  output_byte(0x80);
+  output_modrm(0,5,7);
+  output_w32(addr-(int)out-5); // Note: rip-relative in 64-bit mode
+  output_byte(imm);
+}
+
+// special case for checking invalid_code
+void emit_cmpmem_indexedsr12_imm(int addr,int r,int imm)
+{
+  assert(imm<128&&imm>=-127);
+  assert(r>=0&&r<8);
+  emit_shrimm(r,12,r);
+  assem_debug("cmp $%d,%x+%%%s\n",imm,addr,regname[r]);
+  output_byte(0x80);
+  output_modrm(2,r,7);
+  output_w32(addr);
+  output_byte(imm);
+}
+
+// special case for checking hash_table
+void emit_cmpmem_indexed(int addr,int rs,int rt)
+{
+  assert(rs>=0&&rs<8);
+  assert(rt>=0&&rt<8);
+  assem_debug("cmp %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x39);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+
+// special case for checking memory_map in verify_mapping
+void emit_cmpmem(int addr,int rt)
+{
+  assert(rt>=0&&rt<8);
+  assem_debug("cmp %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x39);
+  output_modrm(0,5,rt);
+  output_w32((int)addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+
+// Used to preload hash table entries
+void emit_prefetch(void *addr)
+{
+  assem_debug("prefetch %x\n",(int)addr);
+  output_byte(0x0F);
+  output_byte(0x18);
+  output_modrm(0,5,1);
+  output_w32((int)addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+
+/*void emit_submem(int r,int addr)
+{
+  assert(r>=0&&r<8);
+  assem_debug("sub %x,%%%s\n",addr,regname[r]);
+  output_byte(0x2B);
+  output_modrm(0,5,r);
+  output_w32((int)addr);
+}*/
+
+void emit_flds(int r)
+{
+  assem_debug("flds (%%%s)\n",regname[r]);
+  output_byte(0xd9);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fldl(int r)
+{
+  assem_debug("fldl (%%%s)\n",regname[r]);
+  output_byte(0xdd);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fucomip(u_int r)
+{
+  assem_debug("fucomip %d\n",r);
+  assert(r<8);
+  output_byte(0xdf);
+  output_byte(0xe8+r);
+}
+void emit_fchs()
+{
+  assem_debug("fchs\n");
+  output_byte(0xd9);
+  output_byte(0xe0);
+}
+void emit_fabs()
+{
+  assem_debug("fabs\n");
+  output_byte(0xd9);
+  output_byte(0xe1);
+}
+void emit_fsqrt()
+{
+  assem_debug("fsqrt\n");
+  output_byte(0xd9);
+  output_byte(0xfa);
+}
+void emit_fadds(int r)
+{
+  assem_debug("fadds (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_faddl(int r)
+{
+  assem_debug("faddl (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fadd(int r)
+{
+  assem_debug("fadd st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xc0+r);
+}
+void emit_fsubs(int r)
+{
+  assem_debug("fsubs (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,4);
+  else {output_modrm(1,EBP,4);output_byte(0);}
+}
+void emit_fsubl(int r)
+{
+  assem_debug("fsubl (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,4);
+  else {output_modrm(1,EBP,4);output_byte(0);}
+}
+void emit_fsub(int r)
+{
+  assem_debug("fsub st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xe0+r);
+}
+void emit_fmuls(int r)
+{
+  assem_debug("fmuls (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,1);
+  else {output_modrm(1,EBP,1);output_byte(0);}
+}
+void emit_fmull(int r)
+{
+  assem_debug("fmull (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,1);
+  else {output_modrm(1,EBP,1);output_byte(0);}
+}
+void emit_fmul(int r)
+{
+  assem_debug("fmul st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xc8+r);
+}
+void emit_fdivs(int r)
+{
+  assem_debug("fdivs (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,6);
+  else {output_modrm(1,EBP,6);output_byte(0);}
+}
+void emit_fdivl(int r)
+{
+  assem_debug("fdivl (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,6);
+  else {output_modrm(1,EBP,6);output_byte(0);}
+}
+void emit_fdiv(int r)
+{
+  assem_debug("fdiv st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xf0+r);
+}
+void emit_fpop()
+{
+  // fstp st(0)
+  assem_debug("fpop\n");
+  output_byte(0xdd);
+  output_byte(0xd8);
+}
+void emit_fildl(int r)
+{
+  assem_debug("fildl (%%%s)\n",regname[r]);
+  output_byte(0xdb);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fildll(int r)
+{
+  assem_debug("fildll (%%%s)\n",regname[r]);
+  output_byte(0xdf);
+  if(r!=EBP) output_modrm(0,r,5);
+  else {output_modrm(1,EBP,5);output_byte(0);}
+}
+void emit_fistpl(int r)
+{
+  assem_debug("fistpl (%%%s)\n",regname[r]);
+  output_byte(0xdb);
+  if(r!=EBP) output_modrm(0,r,3);
+  else {output_modrm(1,EBP,3);output_byte(0);}
+}
+void emit_fistpll(int r)
+{
+  assem_debug("fistpll (%%%s)\n",regname[r]);
+  output_byte(0xdf);
+  if(r!=EBP) output_modrm(0,r,7);
+  else {output_modrm(1,EBP,7);output_byte(0);}
+}
+void emit_fstps(int r)
+{
+  assem_debug("fstps (%%%s)\n",regname[r]);
+  output_byte(0xd9);
+  if(r!=EBP) output_modrm(0,r,3);
+  else {output_modrm(1,EBP,3);output_byte(0);}
+}
+void emit_fstpl(int r)
+{
+  assem_debug("fstpl (%%%s)\n",regname[r]);
+  output_byte(0xdd);
+  if(r!=EBP) output_modrm(0,r,3);
+  else {output_modrm(1,EBP,3);output_byte(0);}
+}
+void emit_fnstcw_stack()
+{
+  assem_debug("fnstcw (%%esp)\n");
+  output_byte(0xd9);
+  output_modrm(0,4,7);
+  output_sib(0,4,4);
+}
+void emit_fldcw_stack()
+{
+  assem_debug("fldcw (%%esp)\n");
+  output_byte(0xd9);
+  output_modrm(0,4,5);
+  output_sib(0,4,4);
+}
+void emit_fldcw_indexed(int addr,int r)
+{
+  assem_debug("fldcw %x(%%%s)\n",addr,regname[r]);
+  output_byte(0xd9);
+  output_modrm(0,4,5);
+  output_sib(1,r,5);
+  output_w32(addr);
+}
+void emit_fldcw(int addr)
+{
+  assem_debug("fldcw %x\n",addr);
+  output_byte(0xd9);
+  output_modrm(0,5,5);
+  output_w32(addr-(int)out-4); // Note: rip-relative in 64-bit mode
+}
+void emit_movss_load(u_int addr,u_int ssereg)
+{
+  assem_debug("movss (%%%s),xmm%d\n",regname[addr],ssereg);
+  assert(ssereg<8);
+  output_byte(0xf3);
+  output_byte(0x0f);
+  output_byte(0x10);
+  if(addr!=EBP) output_modrm(0,addr,ssereg);
+  else {output_modrm(1,EBP,ssereg);output_byte(0);}
+}
+void emit_movsd_load(u_int addr,u_int ssereg)
+{
+  assem_debug("movsd (%%%s),xmm%d\n",regname[addr],ssereg);
+  assert(ssereg<8);
+  output_byte(0xf2);
+  output_byte(0x0f);
+  output_byte(0x10);
+  if(addr!=EBP) output_modrm(0,addr,ssereg);
+  else {output_modrm(1,EBP,ssereg);output_byte(0);}
+}
+void emit_movd_store(u_int ssereg,u_int addr)
+{
+  assem_debug("movd xmm%d,(%%%s)\n",ssereg,regname[addr]);
+  assert(ssereg<8);
+  output_byte(0x66);
+  output_byte(0x0f);
+  output_byte(0x7e);
+  if(addr!=EBP) output_modrm(0,addr,ssereg);
+  else {output_modrm(1,EBP,ssereg);output_byte(0);}
+}
+void emit_cvttps2dq(u_int ssereg1,u_int ssereg2)
+{
+  assem_debug("cvttps2dq xmm%d,xmm%d\n",ssereg1,ssereg2);
+  assert(ssereg1<8);
+  assert(ssereg2<8);
+  output_byte(0xf3);
+  output_byte(0x0f);
+  output_byte(0x5b);
+  output_modrm(3,ssereg1,ssereg2);
+}
+void emit_cvttpd2dq(u_int ssereg1,u_int ssereg2)
+{
+  assem_debug("cvttpd2dq xmm%d,xmm%d\n",ssereg1,ssereg2);
+  assert(ssereg1<8);
+  assert(ssereg2<8);
+  output_byte(0x66);
+  output_byte(0x0f);
+  output_byte(0xe6);
+  output_modrm(3,ssereg1,ssereg2);
+}
+
+unsigned int count_bits(u_int reglist)
+{
+  int count=0;
+  while(reglist)
+  {
+    count+=reglist&1;
+    reglist>>=1;
+  }
+  return count;
+}
+
+// Save registers before function call
+// This code is executed infrequently so we try to minimize code size
+// by pushing registers onto the stack instead of writing them to their
+// usual locations
+void save_regs(u_int reglist)
+{
+  int hr;
+  int count=count_bits(reglist);
+  if(count) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG) {
+        if((reglist>>hr)&1) {
+          emit_pushreg(hr);
+        }
+      }
+    }
+  }
+  emit_addimm(ESP,-(8-count)*8,ESP);
+}
+// Restore registers after function call
+void restore_regs(u_int reglist)
+{
+  int hr;
+  int count=count_bits(reglist);
+  emit_addimm(ESP,(8-count)*8,ESP);
+  if(count) {
+    for(hr=HOST_REGS-1;hr>=0;hr--) {
+      if(hr!=EXCLUDE_REG) {
+        if((reglist>>hr)&1) {
+          emit_popreg(hr);
+        }
+      }
+    }
+  }
+}
+
+/* Stubs/epilogue */
+
+emit_extjump2(int addr, int target, int linker)
+{
+  u_char *ptr=(u_char *)addr;
+  if(*ptr==0x0f)
+  {
+    assert(ptr[1]>=0x80&&ptr[1]<=0x8f);
+    addr+=2;
+  }
+  else
+  {
+    assert(*ptr==0xe8||*ptr==0xe9);
+    addr++;
+  }
+  emit_movimm(target,EAX);
+  emit_movimm(addr,EBX);
+  //assert(addr>=0x7000000&&addr<0x7FFFFFF);
+  //assert((target>=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000));
+//DEBUG >
+#ifdef DEBUG_CYCLE_COUNT
+  emit_readword((int)&last_count,ECX);
+  emit_add(HOST_CCREG,ECX,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_writeword(HOST_CCREG,(int)&Count);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+#endif
+//DEBUG <
+  emit_jmp(linker);
+}
+
+emit_extjump(int addr, int target)
+{
+  emit_extjump2(addr, target, (int)dyna_linker);
+}
+emit_extjump_ds(int addr, int target)
+{
+  emit_extjump2(addr, target, (int)dyna_linker_ds);
+}
+
+do_readstub(int n)
+{
+  assem_debug("do_readstub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int type=stubs[n][0];
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  u_int reglist=stubs[n][7];
+  signed char *i_regmap=i_regs->regmap;
+  int addr=get_reg(i_regmap,AGEN1+(i&1));
+  int rth,rt;
+  int ds;
+  if(itype[i]==C1LS||itype[i]==LOADLR) {
+    rth=get_reg(i_regmap,FTEMP|64);
+    rt=get_reg(i_regmap,FTEMP);
+  }else{
+    rth=get_reg(i_regmap,rt1[i]|64);
+    rt=get_reg(i_regmap,rt1[i]);
+  }
+  assert(rs>=0);
+  assert(rt>=0);
+  if(addr<0) addr=rt;
+  assert(addr>=0);
+  int ftable=0;
+  if(type==LOADB_STUB||type==LOADBU_STUB)
+    ftable=(int)readmemb;
+  if(type==LOADH_STUB||type==LOADHU_STUB)
+    ftable=(int)readmemh;
+  if(type==LOADW_STUB)
+    ftable=(int)readmem;
+  if(type==LOADD_STUB)
+    ftable=(int)readmemd;
+  emit_writeword(rs,(int)&address);
+  emit_shrimm(rs,16,addr);
+  emit_movmem_indexedx8(ftable,addr,addr);
+  save_regs(reglist);
+  ds=i_regs!=&regs[i];
+  int real_rs=(itype[i]==LOADLR)?-1:get_reg(i_regmap,rs1[i]);
+  if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)),i);
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)));
+  
+  int temp;
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    if(addr==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!addr;
+    }
+  }
+  else
+  {
+    temp=!addr;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(stubs[n][6]+1),cc);
+  emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,0);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  emit_callreg(addr);
+  // We really shouldn't need to update the count here,
+  // but not doing so causes random crashes...
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(stubs[n][6]+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  restore_regs(reglist);
+  if((cc=get_reg(i_regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+  if(type==LOADB_STUB)
+    emit_movsbl((int)&readmem_dword,rt);
+  if(type==LOADBU_STUB)
+    emit_movzbl((int)&readmem_dword,rt);
+  if(type==LOADH_STUB)
+    emit_movswl((int)&readmem_dword,rt);
+  if(type==LOADHU_STUB)
+    emit_movzwl((int)&readmem_dword,rt);
+  if(type==LOADW_STUB)
+    emit_readword((int)&readmem_dword,rt);
+  if(type==LOADD_STUB) {
+    emit_readword((int)&readmem_dword,rt);
+    if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth);
+  }
+  emit_jmp(stubs[n][2]); // return address
+}
+
+inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+{
+  assem_debug("inline_readstub\n");
+  int rs=get_reg(regmap,target);
+  int rth=get_reg(regmap,target|64);
+  int rt=get_reg(regmap,target);
+  assert(rs>=0);
+  assert(rt>=0);
+  int ftable=0;
+  if(type==LOADB_STUB||type==LOADBU_STUB)
+    ftable=(int)readmemb;
+  if(type==LOADH_STUB||type==LOADHU_STUB)
+    ftable=(int)readmemh;
+  if(type==LOADW_STUB)
+    ftable=(int)readmem;
+  if(type==LOADD_STUB)
+    ftable=(int)readmemd;
+  #ifdef HOST_IMM_ADDR32
+  emit_writeword_imm(addr,(int)&address);
+  #else
+  emit_writeword(rs,(int)&address);
+  #endif
+  save_regs(reglist);
+  int cc=get_reg(regmap,CCREG);
+  int temp;
+  if(cc<0) {
+    if(rs==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!rs;
+    }
+  }
+  else
+  {
+    temp=!rs;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(adj+1),cc);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  if((signed int)addr>=(signed int)0xC0000000) {
+    // Pagefault address
+    int ds=regmap!=regs[i].regmap;
+    emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,0);
+  }
+  emit_call(((uint64_t *)ftable)[addr>>16]);
+  // We really shouldn't need to update the count here,
+  // but not doing so causes random crashes...
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(adj+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  restore_regs(reglist);
+  if((cc=get_reg(regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+  if(type==LOADB_STUB)
+    emit_movsbl((int)&readmem_dword,rt);
+  if(type==LOADBU_STUB)
+    emit_movzbl((int)&readmem_dword,rt);
+  if(type==LOADH_STUB)
+    emit_movswl((int)&readmem_dword,rt);
+  if(type==LOADHU_STUB)
+    emit_movzwl((int)&readmem_dword,rt);
+  if(type==LOADW_STUB)
+    emit_readword((int)&readmem_dword,rt);
+  if(type==LOADD_STUB) {
+    emit_readword((int)&readmem_dword,rt);
+    if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth);
+  }
+}
+
+do_writestub(int n)
+{
+  assem_debug("do_writestub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int type=stubs[n][0];
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  u_int reglist=stubs[n][7];
+  signed char *i_regmap=i_regs->regmap;
+  int addr=get_reg(i_regmap,AGEN1+(i&1));
+  int rth,rt,r;
+  int ds;
+  if(itype[i]==C1LS) {
+    rth=get_reg(i_regmap,FTEMP|64);
+    rt=get_reg(i_regmap,r=FTEMP);
+  }else{
+    rth=get_reg(i_regmap,rs2[i]|64);
+    rt=get_reg(i_regmap,r=rs2[i]);
+  }
+  assert(rs>=0);
+  assert(rt>=0);
+  if(addr<0) addr=get_reg(i_regmap,-1);
+  assert(addr>=0);
+  int ftable=0;
+  if(type==STOREB_STUB)
+    ftable=(int)writememb;
+  if(type==STOREH_STUB)
+    ftable=(int)writememh;
+  if(type==STOREW_STUB)
+    ftable=(int)writemem;
+  if(type==STORED_STUB)
+    ftable=(int)writememd;
+  emit_writeword(rs,(int)&address);
+  emit_shrimm(rs,16,addr);
+  emit_movmem_indexedx8(ftable,addr,addr);
+  if(type==STOREB_STUB)
+    emit_writebyte(rt,(int)&byte);
+  if(type==STOREH_STUB)
+    emit_writehword(rt,(int)&hword);
+  if(type==STOREW_STUB)
+    emit_writeword(rt,(int)&word);
+  if(type==STORED_STUB) {
+    emit_writeword(rt,(int)&dword);
+    emit_writeword(r?rth:rt,(int)&dword+4);
+  }
+  save_regs(reglist);
+  ds=i_regs!=&regs[i];
+  int real_rs=get_reg(i_regmap,rs1[i]);
+  if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)),i);
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)));
+  
+  int temp;
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    if(addr==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!addr;
+    }
+  }
+  else
+  {
+    temp=!addr;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(stubs[n][6]+1),cc);
+  emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,0);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  emit_callreg(addr);
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(stubs[n][6]+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  restore_regs(reglist);
+  if((cc=get_reg(i_regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+  emit_jmp(stubs[n][2]); // return address
+}
+
+inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+{
+  assem_debug("inline_writestub\n");
+  int rs=get_reg(regmap,-1);
+  int rth=get_reg(regmap,target|64);
+  int rt=get_reg(regmap,target);
+  assert(rs>=0);
+  assert(rt>=0);
+  int ftable=0;
+  if(type==STOREB_STUB)
+    ftable=(int)writememb;
+  if(type==STOREH_STUB)
+    ftable=(int)writememh;
+  if(type==STOREW_STUB)
+    ftable=(int)writemem;
+  if(type==STORED_STUB)
+    ftable=(int)writememd;
+  emit_writeword(rs,(int)&address);
+  if(type==STOREB_STUB)
+    emit_writebyte(rt,(int)&byte);
+  if(type==STOREH_STUB)
+    emit_writehword(rt,(int)&hword);
+  if(type==STOREW_STUB)
+    emit_writeword(rt,(int)&word);
+  if(type==STORED_STUB) {
+    emit_writeword(rt,(int)&dword);
+    emit_writeword(target?rth:rt,(int)&dword+4);
+  }
+  save_regs(reglist);
+  int cc=get_reg(regmap,CCREG);
+  int temp;
+  if(cc<0) {
+    if(rs==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!rs;
+    }
+  }
+  else
+  {
+    temp=!rs;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(adj+1),cc);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  if((signed int)addr>=(signed int)0xC0000000) {
+    // Pagefault address
+    int ds=regmap!=regs[i].regmap;
+    emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,0);
+  }
+  emit_call(((uint64_t *)ftable)[addr>>16]);
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(adj+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  restore_regs(reglist);
+  if((cc=get_reg(regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+}
+
+do_unalignedwritestub(int n)
+{
+  set_jump_target(stubs[n][1],(int)out);
+  output_byte(0xCC);
+  emit_jmp(stubs[n][2]); // return address
+}
+
+void printregs(int edi,int esi,int ebp,int esp,int b,int d,int c,int a)
+{
+  printf("regs: %x %x %x %x %x %x %x (%x)\n",a,b,c,d,ebp,esi,edi,(&edi)[-1]);
+}
+
+do_invstub(int n)
+{
+  u_int reglist=stubs[n][3];
+  set_jump_target(stubs[n][1],(int)out);
+  save_regs(reglist);
+  if(stubs[n][4]!=EDI) emit_mov(stubs[n][4],EDI);
+  emit_call((int)&invalidate_block);
+  restore_regs(reglist);
+  emit_jmp(stubs[n][2]); // return address
+}
+
+int do_dirty_stub(int i)
+{
+  assem_debug("do_dirty_stub %x\n",start+i*4);
+  emit_movimm((int)start<(int)0xC0000000?(int)source:(int)start,EAX);
+  emit_movimm((int)copy,EBX);
+  emit_movimm(slen*4,ECX);
+  emit_movimm(start+i*4,12);
+  emit_call((int)start<(int)0xC0000000?(int)&verify_code:(int)&verify_code_vm);
+  int entry=(int)out;
+  load_regs_entry(i);
+  if(entry==(int)out) entry=instr_addr[i];
+  emit_jmp(instr_addr[i]);
+  return entry;
+}
+
+void do_dirty_stub_ds()
+{
+  emit_movimm((int)start<(int)0xC0000000?(int)source:(int)start,EAX);
+  emit_movimm((int)copy,EBX);
+  emit_movimm(slen*4,ECX);
+  emit_movimm(start+1,12);
+  emit_call((int)&verify_code_ds);
+}
+
+do_cop1stub(int n)
+{
+  assem_debug("do_cop1stub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  int ds=stubs[n][6];
+  if(!ds) {
+    load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i);
+    //if(i_regs!=&regs[i]) printf("oops: regs[i]=%x i_regs=%x",(int)&regs[i],(int)i_regs);
+  }
+  //else {printf("fp exception in delay slot\n");}
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty);
+  if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+  emit_movimm(start+(i-ds)*4,EAX); // Get PC
+  emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
+  emit_jmp(ds?(int)fp_exception_ds:(int)fp_exception);
+}
+
+/* TLB */
+
+int do_tlb_r(int s,int ar,int map,int x,int a,int shift,int c,u_int addr)
+{
+  if(c) {
+    if((signed int)addr>=(signed int)0xC0000000) {
+      emit_readword((int)(memory_map+(addr>>12)),map);
+    }
+    else
+      return -1; // No mapping
+  }
+  else {
+    if(s!=map) emit_mov(s,map);
+    emit_shrimm(map,12,map);
+    // Schedule this while we wait on the load
+    //if(x) emit_xorimm(addr,x,addr);
+    if(shift>=0) emit_lea8(s,shift);
+    if(~a) emit_andimm(s,a,ar);
+    emit_movmem_indexedx4((int)memory_map,map,map);
+  }
+  return map;
+}
+int do_tlb_r_branch(int map, int c, u_int addr, int *jaddr)
+{
+  if(!c||(signed int)addr>=(signed int)0xC0000000) {
+    emit_test(map,map);
+    *jaddr=(int)out;
+    emit_js(0);
+  }
+  return map;
+}
+
+int gen_tlb_addr_r(int ar, int map) {
+  if(map>=0) {
+    emit_leairrx4(0,ar,map,ar);
+  }
+}
+
+int do_tlb_w(int s,int ar,int map,int x,int c,u_int addr)
+{
+  if(c) {
+    if(addr<0x80800000||addr>=0xC0000000) {
+      emit_readword((int)(memory_map+(addr>>12)),map);
+    }
+    else
+      return -1; // No mapping
+  }
+  else {
+    if(s!=map) emit_mov(s,map);
+    //if(s!=ar) emit_mov(s,ar);
+    emit_shrimm(map,12,map);
+    // Schedule this while we wait on the load
+    //if(x) emit_xorimm(s,x,addr);
+    emit_movmem_indexedx4((int)memory_map,map,map);
+  }
+  emit_shlimm(map,2,map);
+  return map;
+}
+int do_tlb_w_branch(int map, int c, u_int addr, int *jaddr)
+{
+  if(!c||addr<0x80800000||addr>=0xC0000000) {
+    *jaddr=(int)out;
+    emit_jc(0);
+  }
+}
+
+int gen_tlb_addr_w(int ar, int map) {
+  if(map>=0) {
+    emit_leairrx1(0,ar,map,ar);
+  }
+}
+
+// We don't need this for x86
+generate_map_const(u_int addr,int reg) {
+  // void *mapaddr=memory_map+(addr>>12);
+}
+
+/* Special assem */
+
+void shift_assemble_x86(int i,struct regstat *i_regs)
+{
+  if(rt1[i]) {
+    if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV
+    {
+      char s,t,shift;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      s=get_reg(i_regs->regmap,rs1[i]);
+      shift=get_reg(i_regs->regmap,rs2[i]);
+      if(t>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(t);
+        }
+        else if(rs2[i]==0)
+        {
+          assert(s>=0);
+          if(s!=t) emit_mov(s,t);
+        }
+        else
+        {
+          char temp=get_reg(i_regs->regmap,-1);
+          assert(s>=0);
+          if(t==ECX&&s!=ECX) {
+            if(shift!=ECX) emit_mov(shift,ECX);
+            if(rt1[i]==rs2[i]) {shift=temp;}
+            if(s!=shift) emit_mov(s,shift);
+          }
+          else
+          {
+            if(rt1[i]==rs2[i]) {emit_mov(shift,temp);shift=temp;}
+            if(s!=t) emit_mov(s,t);
+            if(shift!=ECX) {
+              if(i_regs->regmap[ECX]<0)
+                emit_mov(shift,ECX);
+              else
+                emit_xchg(shift,ECX);
+            }
+          }
+          if(opcode2[i]==4) // SLLV
+          {
+            emit_shlcl(t==ECX?shift:t);
+          }
+          if(opcode2[i]==6) // SRLV
+          {
+            emit_shrcl(t==ECX?shift:t);
+          }
+          if(opcode2[i]==7) // SRAV
+          {
+            emit_sarcl(t==ECX?shift:t);
+          }
+          if(shift!=ECX&&i_regs->regmap[ECX]>=0) emit_xchg(shift,ECX);
+        }
+      }
+    } else { // DSLLV/DSRLV/DSRAV
+      char sh,sl,th,tl,shift;
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      shift=get_reg(i_regs->regmap,rs2[i]);
+      if(tl>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(tl);
+          if(th>=0) emit_zeroreg(th);
+        }
+        else if(rs2[i]==0)
+        {
+          assert(sl>=0);
+          if(sl!=tl) emit_mov(sl,tl);
+          if(th>=0&&sh!=th) emit_mov(sh,th);
+        }
+        else
+        {
+          // FIXME: What if shift==tl ?
+          assert(shift!=tl);
+          int temp=get_reg(i_regs->regmap,-1);
+          int real_th=th;
+          if(th<0&&opcode2[i]!=0x14) {th=temp;} // DSLLV doesn't need a temporary register
+          assert(sl>=0);
+          assert(sh>=0);
+          if(tl==ECX&&sl!=ECX) {
+            if(shift!=ECX) emit_mov(shift,ECX);
+            if(sl!=shift) emit_mov(sl,shift);
+            if(th>=0 && sh!=th) emit_mov(sh,th);
+          }
+          else if(th==ECX&&sh!=ECX) {
+            if(shift!=ECX) emit_mov(shift,ECX);
+            if(sh!=shift) emit_mov(sh,shift);
+            if(sl!=tl) emit_mov(sl,tl);
+          }
+          else
+          {
+            if(sl!=tl) emit_mov(sl,tl);
+            if(th>=0 && sh!=th) emit_mov(sh,th);
+            if(shift!=ECX) {
+              if(i_regs->regmap[ECX]<0)
+                emit_mov(shift,ECX);
+              else
+                emit_xchg(shift,ECX);
+            }
+          }
+          if(opcode2[i]==0x14) // DSLLV
+          {
+            if(th>=0) emit_shldcl(th==ECX?shift:th,tl==ECX?shift:tl);
+            emit_shlcl(tl==ECX?shift:tl);
+            emit_testimm(ECX,32);
+            if(th>=0) emit_cmovne_reg(tl==ECX?shift:tl,th==ECX?shift:th);
+            emit_cmovne(&const_zero,tl==ECX?shift:tl);
+          }
+          if(opcode2[i]==0x16) // DSRLV
+          {
+            assert(th>=0);
+            emit_shrdcl(tl==ECX?shift:tl,th==ECX?shift:th);
+            emit_shrcl(th==ECX?shift:th);
+            emit_testimm(ECX,32);
+            emit_cmovne_reg(th==ECX?shift:th,tl==ECX?shift:tl);
+            if(real_th>=0) emit_cmovne(&const_zero,th==ECX?shift:th);
+          }
+          if(opcode2[i]==0x17) // DSRAV
+          {
+            assert(th>=0);
+            emit_shrdcl(tl==ECX?shift:tl,th==ECX?shift:th);
+            if(real_th>=0) {
+              assert(temp>=0);
+              emit_mov(th==ECX?shift:th,temp==ECX?shift:temp);
+            }
+            emit_sarcl(th==ECX?shift:th);
+            if(real_th>=0) emit_sarimm(temp==ECX?shift:temp,31,temp==ECX?shift:temp);
+            emit_testimm(ECX,32);
+            emit_cmovne_reg(th==ECX?shift:th,tl==ECX?shift:tl);
+            if(real_th>=0) emit_cmovne_reg(temp==ECX?shift:temp,th==ECX?shift:th);
+          }
+          if(shift!=ECX&&(i_regs->regmap[ECX]>=0||temp==ECX)) emit_xchg(shift,ECX);
+        }
+      }
+    }
+  }
+}
+#define shift_assemble shift_assemble_x86
+
+void loadlr_assemble_x86(int i,struct regstat *i_regs)
+{
+  int s,th,tl,temp,temp2,addr,map=-1;
+  int offset;
+  int jaddr=0;
+  int memtarget,c=0;
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,rt1[i]|64);
+  tl=get_reg(i_regs->regmap,rt1[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,-1);
+  temp2=get_reg(i_regs->regmap,FTEMP);
+  addr=get_reg(i_regs->regmap,AGEN1+(i&1));
+  assert(addr<0);
+  offset=imm[i];
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  reglist|=1<<temp;
+  if(offset||s<0||c) addr=temp2;
+  else addr=s;
+  if(s>=0) {
+    c=(i_regs->wasconst>>s)&1;
+    memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+    if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
+  }
+  if(tl>=0) {
+    //assert(tl>=0);
+    //assert(rt1[i]);
+    if(!using_tlb) {
+      if(!c) {
+        emit_lea8(addr,temp);
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR
+        }else{
+          emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
+        }
+        emit_cmpimm(addr,0x800000);
+        jaddr=(int)out;
+        emit_jno(0);
+      }
+      else {
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+        }else{
+          emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+        }
+      }
+    }else{ // using tlb
+      int a;
+      if(c) {
+        a=-1;
+      }else if (opcode[i]==0x22||opcode[i]==0x26) {
+        a=0xFFFFFFFC; // LWL/LWR
+      }else{
+        a=0xFFFFFFF8; // LDL/LDR
+      }
+      map=get_reg(i_regs->regmap,TLREG);
+      assert(map>=0);
+      map=do_tlb_r(addr,temp2,map,0,a,c?-1:temp,c,constmap[i][s]+offset);
+      if(c) {
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+        }else{
+          emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+        }
+      }
+      do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
+    }
+    if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR
+      if(!c||memtarget) {
+        //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2);
+        emit_readword_indexed_tlb(0,temp2,map,temp2);
+        if(jaddr) add_stub(LOADW_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist);
+      emit_andimm(temp,24,temp);
+      if (opcode[i]==0x26) emit_xorimm(temp,24,temp); // LWR
+      if(temp==ECX)
+      {
+        int temp3=EDX;
+        if(temp3==temp2) temp3++;
+        emit_pushreg(temp3);
+        emit_movimm(-1,temp3);
+        if (opcode[i]==0x26) {
+          emit_shrcl(temp3);
+          emit_shrcl(temp2);
+        }else{
+          emit_shlcl(temp3);
+          emit_shlcl(temp2);
+        }
+        emit_mov(temp3,ECX);
+        emit_not(ECX,ECX);
+        emit_popreg(temp3);
+      }
+      else
+      {
+        int temp3=EBP;
+        if(temp3==temp) temp3++;
+        if(temp3==temp2) temp3++;
+        if(temp3==temp) temp3++;
+        emit_xchg(ECX,temp);
+        emit_pushreg(temp3);
+        emit_movimm(-1,temp3);
+        if (opcode[i]==0x26) {
+          emit_shrcl(temp3);
+          emit_shrcl(temp2==ECX?temp:temp2);
+        }else{
+          emit_shlcl(temp3);
+          emit_shlcl(temp2==ECX?temp:temp2);
+        }
+        emit_not(temp3,temp3);
+        emit_mov(temp,ECX);
+        emit_mov(temp3,temp);
+        emit_popreg(temp3);
+      }
+      emit_and(temp,tl,tl);
+      emit_or(temp2,tl,tl);
+      //emit_storereg(rt1[i],tl); // DEBUG
+    }
+    if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR
+      if(s>=0) 
+        if((i_regs->wasdirty>>s)&1)
+          emit_storereg(rs1[i],s);
+      if(get_reg(i_regs->regmap,rs1[i]|64)>=0) 
+        if((i_regs->wasdirty>>get_reg(i_regs->regmap,rs1[i]|64))&1)
+          emit_storereg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64));
+      int temp2h=get_reg(i_regs->regmap,FTEMP|64);
+      if(!c||memtarget) {
+        //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,temp2,temp2h);
+        //emit_readword_indexed((int)rdram-0x7FFFFFFC,temp2,temp2);
+        emit_readdword_indexed_tlb(0,temp2,map,temp2h,temp2);
+        if(jaddr) add_stub(LOADD_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADD_STUB,i,(constmap[i][s]+offset)&0xFFFFFFF8,i_regs->regmap,FTEMP,ccadj[i],reglist);
+      emit_andimm(temp,56,temp);
+      //output_byte(0xCC);
+      //emit_pushreg(temp);
+      //emit_pushreg(temp2h);
+      //emit_pushreg(temp2);
+      //emit_pushreg(th);
+      //emit_pushreg(tl);
+      emit_addimm64(ESP,-20,ESP);
+      emit_writeword_indexed(temp,16,ESP);
+      emit_writeword_indexed(temp2h,12,ESP);
+      emit_writeword_indexed(temp2,8,ESP);
+      emit_writeword_indexed(th,4,ESP);
+      emit_writeword_indexed(tl,0,ESP);
+      emit_mov(temp,EDX);
+      emit_readdword_indexed(0,ESP,ARG1_REG);
+      emit_readdword_indexed(8,ESP,ARG2_REG);
+      if(opcode[i]==0x1A) emit_call((int)ldl_merge);
+      if(opcode[i]==0x1B) emit_call((int)ldr_merge);
+      emit_addimm64(ESP,20,ESP);
+      if(th!=EAX) {
+        emit_mov64(EAX,th);
+      }
+      emit_mov(EAX,tl);
+      emit_shrimm64(th,32,th);
+      if(s>=0) emit_loadreg(rs1[i],s);
+      if(get_reg(i_regs->regmap,rs1[i]|64)>=0)
+        emit_loadreg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64));
+    }
+  }
+}
+#define loadlr_assemble loadlr_assemble_x86
+
+void cop0_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]==0) // MFC0
+  {
+    signed char t=get_reg(i_regs->regmap,rt1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    //assert(t>=0); // Why does this happen?  OOT is weird
+    if(t>=0) {
+      emit_writedword_imm32((int)&fake_pc,(int)&PC);
+      emit_writebyte_imm((source[i]>>11)&0x1f,(int)&(fake_pc.f.r.nrd));
+      if(copr==9) {
+        emit_readword((int)&last_count,ECX);
+        emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+      }
+      emit_call((int)MFC0);
+      emit_readword((int)&readmem_dword,t);
+    }
+  }
+  else if(opcode2[i]==4) // MTC0
+  {
+    signed char s=get_reg(i_regs->regmap,rs1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    assert(s>=0);
+    emit_writeword(s,(int)&readmem_dword);
+    wb_register(rs1[i],i_regs->regmap,i_regs->dirty,i_regs->was32); // FIXME
+    emit_writedword_imm32((int)&fake_pc,(int)&PC);
+    emit_writebyte_imm((source[i]>>11)&0x1f,(int)&(fake_pc.f.r.nrd));
+    if(copr==9||copr==11||copr==12) {
+      emit_readword((int)&last_count,ECX);
+      emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+      emit_add(HOST_CCREG,ECX,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_writeword(HOST_CCREG,(int)&Count);
+    }
+    // What a mess.  The status register (12) can enable interrupts,
+    // so needs a special case to handle a pending interrupt.
+    // The interrupt must be taken immediately, because a subsequent
+    // instruction might disable interrupts again.
+    if(copr==12&&!is_delayslot) {
+      emit_writeword_imm(start+i*4+4,(int)&pcaddr);
+      emit_writebyte_imm(0,(int)&pending_exception);
+    }
+    //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
+    //else
+    emit_call((int)MTC0);
+    if(copr==9||copr==11||copr==12) {
+      emit_readword((int)&Count,HOST_CCREG);
+      emit_readword((int)&next_interupt,ECX);
+      emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+      emit_writeword(ECX,(int)&last_count);
+      emit_storereg(CCREG,HOST_CCREG);
+    }
+    emit_loadreg(rs1[i],s);
+    if(copr==12) {
+      assert(!is_delayslot);
+      //if(is_delayslot) output_byte(0xcc);
+      emit_cmpmem_imm_byte((int)&pending_exception,0);
+      emit_jne((int)&do_interrupt);
+    }
+    cop1_usable=0;
+  }
+  else
+  {
+    assert(opcode2[i]==0x10);
+    if((source[i]&0x3f)==0x01) // TLBR
+      emit_call((int)TLBR);
+    if((source[i]&0x3f)==0x02) // TLBWI
+      emit_call((int)TLBWI_new);
+    if((source[i]&0x3f)==0x06) { // TLBWR
+      // The TLB entry written by TLBWR is dependent on the count,
+      // so update the cycle count
+      emit_readword((int)&last_count,ECX);
+      if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+      emit_add(HOST_CCREG,ECX,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_writeword(HOST_CCREG,(int)&Count);
+      emit_call((int)TLBWR_new);
+    }
+    if((source[i]&0x3f)==0x08) // TLBP
+      emit_call((int)TLBP);
+    if((source[i]&0x3f)==0x18) // ERET
+    {
+      int count=ccadj[i];
+      if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+      emit_addimm_and_set_flags(CLOCK_DIVIDER*count,HOST_CCREG); // TODO: Should there be an extra cycle here?
+      emit_jmp((int)jump_eret);
+    }
+  }
+}
+
+void cop1_assemble(int i,struct regstat *i_regs)
+{
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  if (opcode2[i]==0) { // MFC1
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],tl);
+      emit_readword_indexed(0,tl,tl);
+    }
+  }
+  else if (opcode2[i]==1) { // DMFC1
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    signed char th=get_reg(i_regs->regmap,rt1[i]|64);
+    if(tl>=0) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],tl);
+      if(th>=0) emit_readword_indexed(4,tl,th);
+      emit_readword_indexed(0,tl,tl);
+    }
+  }
+  else if (opcode2[i]==4) { // MTC1
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    signed char temp=get_reg(i_regs->regmap,-1);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_writeword_indexed(sl,0,temp);
+  }
+  else if (opcode2[i]==5) { // DMTC1
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    signed char sh=rs1[i]>0?get_reg(i_regs->regmap,rs1[i]|64):sl;
+    signed char temp=get_reg(i_regs->regmap,-1);
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_writeword_indexed(sh,4,temp);
+    emit_writeword_indexed(sl,0,temp);
+  }
+  else if (opcode2[i]==2) // CFC1
+  {
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0) {
+      u_int copr=(source[i]>>11)&0x1f;
+      if(copr==0) emit_readword((int)&FCR0,tl);
+      if(copr==31) emit_readword((int)&FCR31,tl);
+    }
+  }
+  else if (opcode2[i]==6) // CTC1
+  {
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    u_int copr=(source[i]>>11)&0x1f;
+    assert(sl>=0);
+    if(copr==31)
+    {
+      emit_writeword(sl,(int)&FCR31);
+      // Set the rounding mode
+      char temp=get_reg(i_regs->regmap,-1);
+      emit_movimm(3,temp);
+      emit_and(sl,temp,temp);
+      emit_fldcw_indexed((int)&rounding_modes,temp);
+    }
+  }
+}
+
+void fconv_assemble_x86(int i,struct regstat *i_regs)
+{
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) { // trunc_w_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_movss_load(temp,0);
+    emit_cvttps2dq(0,0); // float->int, truncate
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_movd_store(0,temp);
+    return;
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) { // trunc_w_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_movsd_load(temp,0);
+    emit_cvttpd2dq(0,0); // double->int, truncate
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_movd_store(0,temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x20) { // cvt_s_w
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_fildl(temp);
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fstps(temp);
+    return;
+  }
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) { // cvt_d_w
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_fildl(temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fstpl(temp);
+    return;
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x20) { // cvt_s_l
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fildll(temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fstps(temp);
+    return;
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x21) { // cvt_d_l
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fildll(temp);
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fstpl(temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) { // cvt_d_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fstpl(temp);
+    return;
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) { // cvt_s_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fldl(temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fstps(temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x10) { // cvt_*_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp);
+  }
+  if(opcode2[i]==0x11) { // cvt_*_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fldl(temp);
+  }
+  if((source[i]&0x3f)<0x10) {
+    emit_fnstcw_stack();
+    if((source[i]&3)==0) emit_fldcw((int)&round_mode); //printf("round\n");
+    if((source[i]&3)==1) emit_fldcw((int)&trunc_mode); //printf("trunc\n");
+    if((source[i]&3)==2) emit_fldcw((int)&ceil_mode); //printf("ceil\n");
+    if((source[i]&3)==3) emit_fldcw((int)&floor_mode); //printf("floor\n");
+  }
+  if((source[i]&0x3f)==0x24||(source[i]&0x3c)==0x0c) { // cvt_w_*
+    if(opcode2[i]!=0x10||((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fistpl(temp);
+  }
+  if((source[i]&0x3f)==0x25||(source[i]&0x3c)==0x08) { // cvt_l_*
+    if(opcode2[i]!=0x11||((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fistpll(temp);
+  }
+  if((source[i]&0x3f)<0x10) {
+    emit_fldcw_stack();
+  }
+  return;
+}
+#define fconv_assemble fconv_assemble_x86
+
+void fcomp_assemble(int i,struct regstat *i_regs)
+{
+  signed char fs=get_reg(i_regs->regmap,FSREG);
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char cs=get_reg(i_regs->regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  if((source[i]&0x3f)==0x30) {
+    emit_andimm(fs,~0x800000,fs);
+    return;
+  }
+  
+  if((source[i]&0x3e)==0x38) {
+    // sf/ngle - these should throw exceptions for NaNs
+    emit_andimm(fs,~0x800000,fs);
+    return;
+  }
+  
+  if(opcode2[i]==0x10) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],temp);
+    emit_flds(temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp);
+    emit_movimm(0x800000,temp);
+    emit_or(fs,temp,fs);
+    emit_xor(temp,fs,temp);
+    emit_fucomip(1);
+    emit_fpop();
+    if((source[i]&0x3f)==0x31) emit_cmovnp_reg(temp,fs); // c_un_s
+    if((source[i]&0x3f)==0x32) {emit_cmovne_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_eq_s
+    if((source[i]&0x3f)==0x33) emit_cmovne_reg(temp,fs); // c_ueq_s
+    if((source[i]&0x3f)==0x34) {emit_cmovnc_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_olt_s
+    if((source[i]&0x3f)==0x35) emit_cmovnc_reg(temp,fs); // c_ult_s
+    if((source[i]&0x3f)==0x36) {emit_cmova_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_ole_s
+    if((source[i]&0x3f)==0x37) emit_cmova_reg(temp,fs); // c_ule_s
+    if((source[i]&0x3f)==0x3a) emit_cmovne_reg(temp,fs); // c_seq_s
+    if((source[i]&0x3f)==0x3b) emit_cmovne_reg(temp,fs); // c_ngl_s
+    if((source[i]&0x3f)==0x3c) emit_cmovnc_reg(temp,fs); // c_lt_s
+    if((source[i]&0x3f)==0x3d) emit_cmovnc_reg(temp,fs); // c_nge_s
+    if((source[i]&0x3f)==0x3e) emit_cmova_reg(temp,fs); // c_le_s
+    if((source[i]&0x3f)==0x3f) emit_cmova_reg(temp,fs); // c_ngt_s
+    return;
+  }
+  if(opcode2[i]==0x11) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],temp);
+    emit_fldl(temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fldl(temp);
+    emit_movimm(0x800000,temp);
+    emit_or(fs,temp,fs);
+    emit_xor(temp,fs,temp);
+    emit_fucomip(1);
+    emit_fpop();
+    if((source[i]&0x3f)==0x31) emit_cmovnp_reg(temp,fs); // c_un_d
+    if((source[i]&0x3f)==0x32) {emit_cmovne_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_eq_d
+    if((source[i]&0x3f)==0x33) emit_cmovne_reg(temp,fs); // c_ueq_d
+    if((source[i]&0x3f)==0x34) {emit_cmovnc_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_olt_d
+    if((source[i]&0x3f)==0x35) emit_cmovnc_reg(temp,fs); // c_ult_d
+    if((source[i]&0x3f)==0x36) {emit_cmova_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_ole_d
+    if((source[i]&0x3f)==0x37) emit_cmova_reg(temp,fs); // c_ule_d
+    if((source[i]&0x3f)==0x3a) emit_cmovne_reg(temp,fs); // c_seq_d
+    if((source[i]&0x3f)==0x3b) emit_cmovne_reg(temp,fs); // c_ngl_d
+    if((source[i]&0x3f)==0x3c) emit_cmovnc_reg(temp,fs); // c_lt_d
+    if((source[i]&0x3f)==0x3d) emit_cmovnc_reg(temp,fs); // c_nge_d
+    if((source[i]&0x3f)==0x3e) emit_cmova_reg(temp,fs); // c_le_d
+    if((source[i]&0x3f)==0x3f) emit_cmova_reg(temp,fs); // c_ngt_d
+    return;
+  }
+}
+
+void float_assemble(int i,struct regstat *i_regs)
+{
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char cs=get_reg(i_regs->regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  if((source[i]&0x3f)==6) // mov
+  {
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+      if(opcode2[i]==0x10) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+        emit_flds(temp);
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+        emit_fstps(temp);
+      }
+      if(opcode2[i]==0x11) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+        emit_fldl(temp);
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+        emit_fstpl(temp);
+      }
+    }
+    return;
+  }
+  
+  if((source[i]&0x3f)>3)
+  {
+    if(opcode2[i]==0x10) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+      emit_flds(temp);
+      if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+      }
+    }
+    if(opcode2[i]==0x11) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+      emit_fldl(temp);
+      if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+      }
+    }
+    if((source[i]&0x3f)==4) // sqrt
+      emit_fsqrt();
+    if((source[i]&0x3f)==5) // abs
+      emit_fabs();
+    if((source[i]&0x3f)==7) // neg
+      emit_fchs();
+    if(opcode2[i]==0x10) {
+      emit_fstps(temp);
+    }
+    if(opcode2[i]==0x11) {
+      emit_fstpl(temp);
+    }
+    return;
+  }
+  if((source[i]&0x3f)<4)
+  {
+    if(opcode2[i]==0x10) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+      emit_flds(temp);
+    }
+    if(opcode2[i]==0x11) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+      emit_fldl(temp);
+    }
+    if(((source[i]>>11)&0x1f)!=((source[i]>>16)&0x1f)) {
+      if(opcode2[i]==0x10) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],temp);
+        if((source[i]&0x3f)==0) emit_fadds(temp);
+        if((source[i]&0x3f)==1) emit_fsubs(temp);
+        if((source[i]&0x3f)==2) emit_fmuls(temp);
+        if((source[i]&0x3f)==3) emit_fdivs(temp);
+      }
+      else if(opcode2[i]==0x11) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],temp);
+        if((source[i]&0x3f)==0) emit_faddl(temp);
+        if((source[i]&0x3f)==1) emit_fsubl(temp);
+        if((source[i]&0x3f)==2) emit_fmull(temp);
+        if((source[i]&0x3f)==3) emit_fdivl(temp);
+      }
+    }
+    else {
+      if((source[i]&0x3f)==0) emit_fadd(0);
+      if((source[i]&0x3f)==1) emit_fsub(0);
+      if((source[i]&0x3f)==2) emit_fmul(0);
+      if((source[i]&0x3f)==3) emit_fdiv(0);
+    }
+    if(opcode2[i]==0x10) {
+      if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+      }
+      emit_fstps(temp);
+    }
+    if(opcode2[i]==0x11) {
+      if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+      }
+      emit_fstpl(temp);
+    }
+    return;
+  }
+}
+
+void multdiv_assemble_x86(int i,struct regstat *i_regs)
+{
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  //  case 0x1C: DMULT
+  //  case 0x1D: DMULTU
+  //  case 0x1E: DDIV
+  //  case 0x1F: DDIVU
+  if(rs1[i]&&rs2[i])
+  {
+    if((opcode2[i]&4)==0) // 32-bit
+    {
+      if(opcode2[i]==0x18) // MULT
+      {
+        char m1=get_reg(i_regs->regmap,rs1[i]);
+        char m2=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1>=0);
+        assert(m2>=0);
+        emit_mov(m1,EAX);
+        emit_imul(m2);
+      }
+      if(opcode2[i]==0x19) // MULTU
+      {
+        char m1=get_reg(i_regs->regmap,rs1[i]);
+        char m2=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1>=0);
+        assert(m2>=0);
+        emit_mov(m1,EAX);
+        emit_mul(m2);
+      }
+      if(opcode2[i]==0x1A) // DIV
+      {
+        char d1=get_reg(i_regs->regmap,rs1[i]);
+        char d2=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1>=0);
+        assert(d2>=0);
+        emit_mov(d1,EAX);
+        emit_cdq();
+        emit_test(d2,d2);
+        emit_jeq((int)out+8);
+        emit_idiv(d2);
+      }
+      if(opcode2[i]==0x1B) // DIVU
+      {
+        char d1=get_reg(i_regs->regmap,rs1[i]);
+        char d2=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1>=0);
+        assert(d2>=0);
+        emit_mov(d1,EAX);
+        emit_zeroreg(EDX);
+        emit_test(d2,d2);
+        emit_jeq((int)out+8);
+        emit_div(d2);
+      }
+    }
+    else // 64-bit
+    {
+      if(opcode2[i]==0x1C) // DMULT
+      {
+        char m1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char m1l=get_reg(i_regs->regmap,rs1[i]);
+        char m2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char m2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        output_byte(0xCC);
+        emit_pushreg(m2h);
+        emit_pushreg(m2l);
+        emit_pushreg(m1h);
+        emit_pushreg(m1l);
+        emit_call((int)&mult64);
+        emit_popreg(m1l);
+        emit_popreg(m1h);
+        emit_popreg(m2l);
+        emit_popreg(m2h);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        char loh=get_reg(i_regs->regmap,LOREG|64);
+        char lol=get_reg(i_regs->regmap,LOREG);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1D) // DMULTU
+      {
+        char m1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char m1l=get_reg(i_regs->regmap,rs1[i]);
+        char m2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char m2l=get_reg(i_regs->regmap,rs2[i]);
+        char temp=get_reg(i_regs->regmap,-1);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        assert(temp>=0);
+        emit_mov(m1l,EAX);
+        emit_mul(m2l);
+        emit_storereg(LOREG,EAX);
+        emit_mov(EDX,temp);
+        emit_mov(m1h,EAX);
+        emit_mul(m2l);
+        emit_add(EAX,temp,temp);
+        emit_adcimm(0,EDX);
+        emit_storereg(HIREG,EDX);
+        emit_mov(m2h,EAX);
+        emit_mul(m1l);
+        emit_add(EAX,temp,temp);
+        emit_adcimm(0,EDX);
+        emit_storereg(LOREG|64,temp);
+        emit_mov(EDX,temp);
+        emit_mov(m2h,EAX);
+        emit_mul(m1h);
+        emit_add(EAX,temp,EAX);
+        emit_loadreg(HIREG,temp);
+        emit_adcimm(0,EDX);
+        emit_add(EAX,temp,EAX);
+        emit_adcimm(0,EDX);
+        // DEBUG
+        /*
+        emit_pushreg(m2h);
+        emit_pushreg(m2l);
+        emit_pushreg(m1h);
+        emit_pushreg(m1l);
+        emit_call((int)&multu64);
+        emit_popreg(m1l);
+        emit_popreg(m1h);
+        emit_popreg(m2l);
+        emit_popreg(m2h);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);  // DEBUG
+        if(hil>=0) emit_loadreg(HIREG,hil);  // DEBUG
+        */
+        // Shouldn't be necessary
+        //char loh=get_reg(i_regs->regmap,LOREG|64);
+        //char lol=get_reg(i_regs->regmap,LOREG);
+        //if(loh>=0) emit_loadreg(LOREG|64,loh);
+        //if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1E) // DDIV
+      {
+        char d1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char d1l=get_reg(i_regs->regmap,rs1[i]);
+        char d2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char d2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1h>=0);
+        assert(d2h>=0);
+        assert(d1l>=0);
+        assert(d2l>=0);
+        //emit_pushreg(d2h);
+        //emit_pushreg(d2l);
+        //emit_pushreg(d1h);
+        //emit_pushreg(d1l);
+        emit_addimm64(ESP,-16,ESP);
+        emit_writeword_indexed(d2h,12,ESP);
+        emit_writeword_indexed(d2l,8,ESP);
+        emit_writeword_indexed(d1h,4,ESP);
+        emit_writeword_indexed(d1l,0,ESP);
+        emit_readdword_indexed(0,ESP,ARG1_REG);
+        emit_readdword_indexed(8,ESP,ARG2_REG);
+        emit_call((int)&div64);
+        //emit_popreg(d1l);
+        //emit_popreg(d1h);
+        //emit_popreg(d2l);
+        //emit_popreg(d2h);
+        emit_readword_indexed(0,ESP,d1l);
+        emit_readword_indexed(4,ESP,d1h);
+        emit_readword_indexed(8,ESP,d2l);
+        emit_readword_indexed(12,ESP,d2h);
+        emit_addimm64(ESP,16,ESP);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        char loh=get_reg(i_regs->regmap,LOREG|64);
+        char lol=get_reg(i_regs->regmap,LOREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1F) // DDIVU
+      {
+        char d1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char d1l=get_reg(i_regs->regmap,rs1[i]);
+        char d2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char d2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1h>=0);
+        assert(d2h>=0);
+        assert(d1l>=0);
+        assert(d2l>=0);
+        //emit_pushreg(d2h);
+        //emit_pushreg(d2l);
+        //emit_pushreg(d1h);
+        //emit_pushreg(d1l);
+        emit_addimm64(ESP,-16,ESP);
+        emit_writeword_indexed(d2h,12,ESP);
+        emit_writeword_indexed(d2l,8,ESP);
+        emit_writeword_indexed(d1h,4,ESP);
+        emit_writeword_indexed(d1l,0,ESP);
+        emit_readdword_indexed(0,ESP,ARG1_REG);
+        emit_readdword_indexed(8,ESP,ARG2_REG);
+        emit_call((int)&divu64);
+        //emit_popreg(d1l);
+        //emit_popreg(d1h);
+        //emit_popreg(d2l);
+        //emit_popreg(d2h);
+        emit_readword_indexed(0,ESP,d1l);
+        emit_readword_indexed(4,ESP,d1h);
+        emit_readword_indexed(8,ESP,d2l);
+        emit_readword_indexed(12,ESP,d2h);
+        emit_addimm64(ESP,16,ESP);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        char loh=get_reg(i_regs->regmap,LOREG|64);
+        char lol=get_reg(i_regs->regmap,LOREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+    }
+  }
+  else
+  {
+    // Multiply by zero is zero.
+    // MIPS does not have a divide by zero exception.
+    // The result is undefined, we return zero.
+    char hr=get_reg(i_regs->regmap,HIREG);
+    char lr=get_reg(i_regs->regmap,LOREG);
+    if(hr>=0) emit_zeroreg(hr);
+    if(lr>=0) emit_zeroreg(lr);
+  }
+}
+#define multdiv_assemble multdiv_assemble_x86
+
+void do_preload_rhash(int r) {
+  emit_movimm(0xf8,r);
+}
+
+void do_preload_rhtbl(int r) {
+  // Don't need this for x86
+}
+
+void do_rhash(int rs,int rh) {
+  emit_and(rs,rh,rh);
+}
+
+void do_miniht_load(int ht,int rh) {
+  // Don't need this for x86.  The load and compare can be combined into
+  // a single instruction (below)
+}
+
+void do_miniht_jump(int rs,int rh,int ht) {
+  emit_cmpmem_indexed((int)mini_ht,rh,rs);
+  emit_jne(jump_vaddr_reg[rs]);
+  emit_readword_indexed((int)mini_ht+4,rh,rh);
+  emit_jmpreg(rh);
+}
+
+void do_miniht_insert(int return_address,int rt,int temp) {
+  emit_movimm(return_address,rt); // PC into link register
+  //emit_writeword_imm(return_address,(int)&mini_ht[(return_address&0xFF)>>8][0]);
+  emit_writeword(rt,(int)&mini_ht[(return_address&0xFF)>>3][0]);
+  add_to_linker((int)out,return_address,1);
+  emit_writeword_imm(0,(int)&mini_ht[(return_address&0xFF)>>3][1]);
+}
+
+// We don't need this for x86
+void literal_pool(int n) {}
+void literal_pool_jumpover(int n) {}
+
+// CPU-architecture-specific initialization, not needed for x86
+void arch_init() {}
diff --git a/libpcsxcore/new_dynarec/assem_x64.h b/libpcsxcore/new_dynarec/assem_x64.h
new file mode 100644 (file)
index 0000000..9c114f5
--- /dev/null
@@ -0,0 +1,24 @@
+#define HOST_REGS 8
+#define HOST_CCREG 6
+#define HOST_BTREG 5
+#define EXCLUDE_REG 4
+
+//#define IMM_PREFETCH 1
+#define HOST_IMM_ADDR32 1
+#define INVERTED_CARRY 1
+#define DESTRUCTIVE_WRITEBACK 1
+#define DESTRUCTIVE_SHIFT 1
+
+#define USE_MINI_HT 1
+
+#define BASE_ADDR 0x70000000 // Code generator target address
+#define TARGET_SIZE_2 25 // 2^25 = 32 megabytes
+
+#define ROM_COPY ((void *)0x78000000) // For Goldeneye hack
+
+/* x86-64 calling convention:
+   func(rdi, rsi, rdx, rcx, r8, r9) {return rax;}
+   callee-save: %rbp %rbx %r12-%r15 */
+
+#define ARG1_REG 7 /* RDI */
+#define ARG2_REG 6 /* RSI */
diff --git a/libpcsxcore/new_dynarec/assem_x86.c b/libpcsxcore/new_dynarec/assem_x86.c
new file mode 100644 (file)
index 0000000..76ee0c2
--- /dev/null
@@ -0,0 +1,4363 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - assem_x86.c                                             *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+int cycle_count;
+int last_count;
+int pcaddr;
+int pending_exception;
+int branch_target;
+uint64_t readmem_dword;
+precomp_instr fake_pc;
+u_int memory_map[1048576];
+u_int mini_ht[32][2]  __attribute__((aligned(8)));
+u_char restore_candidate[512]  __attribute__((aligned(4)));
+
+void do_interrupt();
+void jump_vaddr_eax();
+void jump_vaddr_ecx();
+void jump_vaddr_edx();
+void jump_vaddr_ebx();
+void jump_vaddr_ebp();
+void jump_vaddr_edi();
+
+const u_int jump_vaddr_reg[8] = {
+  (int)jump_vaddr_eax,
+  (int)jump_vaddr_ecx,
+  (int)jump_vaddr_edx,
+  (int)jump_vaddr_ebx,
+  0,
+  (int)jump_vaddr_ebp,
+  0,
+  (int)jump_vaddr_edi };
+
+const u_short rounding_modes[4] = {
+  0x33F, // round
+  0xF3F, // trunc
+  0xB3F, // ceil
+  0x73F};// floor
+
+#include "fpu.h"
+
+// We need these for cmovcc instructions on x86
+u_int const_zero=0;
+u_int const_one=1;
+
+/* Linker */
+
+void set_jump_target(int addr,int target)
+{
+  u_char *ptr=(u_char *)addr;
+  if(*ptr==0x0f)
+  {
+    assert(ptr[1]>=0x80&&ptr[1]<=0x8f);
+    u_int *ptr2=(u_int *)(ptr+2);
+    *ptr2=target-(int)ptr2-4;
+  }
+  else if(*ptr==0xe8||*ptr==0xe9) {
+    u_int *ptr2=(u_int *)(ptr+1);
+    *ptr2=target-(int)ptr2-4;
+  }
+  else
+  {
+    assert(*ptr==0xc7); /* mov immediate (store address) */
+    u_int *ptr2=(u_int *)(ptr+6);
+    *ptr2=target;
+  }
+}
+
+void kill_pointer(void *stub)
+{
+  int *i_ptr=*((int **)(stub+6));
+  *i_ptr=(int)stub-(int)i_ptr-4;
+}
+int get_pointer(void *stub)
+{
+  int *i_ptr=*((int **)(stub+6));
+  return *i_ptr+(int)i_ptr+4;
+}
+
+// Find the "clean" entry point from a "dirty" entry point
+// by skipping past the call to verify_code
+u_int get_clean_addr(int addr)
+{
+  u_char *ptr=(u_char *)addr;
+  assert(ptr[20]==0xE8); // call instruction
+  assert(ptr[25]==0x83); // pop (add esp,4) instruction
+  if(ptr[28]==0xE9) return *(u_int *)(ptr+29)+addr+33; // follow jmp
+  else return(addr+28);
+}
+
+int verify_dirty(int addr)
+{
+  u_char *ptr=(u_char *)addr;
+  assert(ptr[5]==0xB8);
+  u_int source=*(u_int *)(ptr+6);
+  u_int copy=*(u_int *)(ptr+11);
+  u_int len=*(u_int *)(ptr+16);
+  assert(ptr[20]==0xE8); // call instruction
+  u_int verifier=*(u_int *)(ptr+21)+(u_int)ptr+25;
+  if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) {
+    unsigned int page=source>>12;
+    unsigned int map_value=memory_map[page];
+    if(map_value>=0x80000000) return 0;
+    while(page<((source+len-1)>>12)) {
+      if((memory_map[++page]<<2)!=(map_value<<2)) return 0;
+    }
+    source = source+(map_value<<2);
+  }
+  //printf("verify_dirty: %x %x %x\n",source,copy,len);
+  return !memcmp((void *)source,(void *)copy,len);
+}
+
+// This doesn't necessarily find all clean entry points, just
+// guarantees that it's not dirty
+int isclean(int addr)
+{
+  u_char *ptr=(u_char *)addr;
+  if(ptr[5]!=0xB8) return 1; // mov imm,%eax
+  if(ptr[10]!=0xBB) return 1; // mov imm,%ebx
+  if(ptr[15]!=0xB9) return 1; // mov imm,%ecx
+  if(ptr[20]!=0xE8) return 1; // call instruction
+  if(ptr[25]!=0x83) return 1; // pop (add esp,4) instruction
+  return 0;
+}
+
+void get_bounds(int addr,u_int *start,u_int *end)
+{
+  u_char *ptr=(u_char *)addr;
+  assert(ptr[5]==0xB8);
+  u_int source=*(u_int *)(ptr+6);
+  //u_int copy=*(u_int *)(ptr+11);
+  u_int len=*(u_int *)(ptr+16);
+  assert(ptr[20]==0xE8); // call instruction
+  u_int verifier=*(u_int *)(ptr+21)+(u_int)ptr+25;
+  if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) {
+    if(memory_map[source>>12]>=0x80000000) source = 0;
+    else source = source+(memory_map[source>>12]<<2);
+  }
+  if(start) *start=source;
+  if(end) *end=source+len;
+}
+
+/* Register allocation */
+
+// Note: registers are allocated clean (unmodified state)
+// if you intend to modify the register, you must call dirty_reg().
+void alloc_reg(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = (reg&3)+(reg>28)*4-(reg==32)+2*(reg==36)-(reg==40);
+  
+  // Don't allocate unused registers
+  if((cur->u>>reg)&1) return;
+  
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg) return;
+  }
+  
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+  
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  if(r<64&&((cur->u>>r)&1)) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  if(r>=64&&((cur->uu>>(r&63))&1)) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  
+  // Try to allocate EAX, EBX, ECX, or EDX
+  // We prefer these because they can do byte and halfword loads
+  for(hr=0;hr<4;hr++) {
+    if(cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1)
+          if(i==0||(unneeded_reg[i-1]>>r)&1) {cur->regmap[hr]=-1;break;}
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1)
+          if(i==0||(unneeded_reg_upper[i-1]>>(r&63))&1) {cur->regmap[hr]=-1;break;}
+      }
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen (alloc_reg)");exit(1);
+}
+
+void alloc_reg64(struct regstat *cur,int i,signed char reg)
+{
+  int preferred_reg = 5+reg%3;
+  int r,hr;
+  
+  // allocate the lower 32 bits
+  alloc_reg(cur,i,reg);
+  
+  // Don't allocate unused registers
+  if((cur->uu>>reg)&1) return;
+  
+  // see if the upper half is already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg+64) return;
+  }
+  
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+  
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  if(r<64&&((cur->u>>r)&1)) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  if(r>=64&&((cur->uu>>(r&63))&1)) {
+    cur->regmap[preferred_reg]=reg|64;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  
+  // Try to allocate EBP, ESI or EDI
+  for(hr=5;hr<8;hr++) {
+    if(cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg|64;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;}
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {cur->regmap[hr]=-1;break;}
+      }
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg|64;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg|64;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg|64;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg|64;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg|64;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg|64;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg|64;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen");exit(1);
+}
+
+// Allocate a temporary register.  This is done without regard to
+// dirty status or whether the register we request is on the unneeded list
+// Note: This will only allocate one register, even if called multiple times
+void alloc_reg_temp(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = -1;
+  
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==reg) return;
+  }
+  
+  // Try to allocate any available register, starting with EDI, ESI, EBP...
+  // We prefer EDI, ESI, EBP since the others are used for byte/halfword stores
+  for(hr=HOST_REGS-1;hr>=0;hr--) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+  
+  // Find an unneeded register
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      if(r<64) {
+        if((cur->u>>r)&1) {
+          if(i==0||((unneeded_reg[i-1]>>r)&1)) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+      else
+      {
+        if((cur->uu>>(r&63))&1) {
+          if(i==0||((unneeded_reg_upper[i-1]>>(r&63))&1)) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  // TODO: we might want to follow unconditional jumps here
+  // TODO: get rid of dupe code and make this into a function
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("hsn: %d %d %d %d %d %d %d\n",hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r+64) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r+64) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  printf("This shouldn't happen");exit(1);
+}
+// Allocate a specific x86 register.
+void alloc_x86_reg(struct regstat *cur,int i,signed char reg,char hr)
+{
+  int n;
+  
+  // see if it's already allocated (and dealloc it)
+  for(n=0;n<HOST_REGS;n++)
+  {
+    if(n!=ESP&&cur->regmap[n]==reg) {cur->regmap[n]=-1;}
+  }
+  
+  cur->regmap[hr]=reg;
+  cur->dirty&=~(1<<hr);
+  cur->isconst&=~(1<<hr);
+}
+
+// Alloc cycle count into dedicated register
+alloc_cc(struct regstat *cur,int i)
+{
+  alloc_x86_reg(cur,i,CCREG,ESI);
+}
+
+/* Special alloc */
+
+void multdiv_alloc_x86(struct regstat *current,int i)
+{
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  //  case 0x1C: DMULT
+  //  case 0x1D: DMULTU
+  //  case 0x1E: DDIV
+  //  case 0x1F: DDIVU
+  clear_const(current,rs1[i]);
+  clear_const(current,rs2[i]);
+  if(rs1[i]&&rs2[i])
+  {
+    if((opcode2[i]&4)==0) // 32-bit
+    {
+      current->u&=~(1LL<<HIREG);
+      current->u&=~(1LL<<LOREG);
+      alloc_x86_reg(current,i,HIREG,EDX);
+      alloc_x86_reg(current,i,LOREG,EAX);
+      alloc_reg(current,i,rs1[i]);
+      alloc_reg(current,i,rs2[i]);
+      current->is32|=1LL<<HIREG;
+      current->is32|=1LL<<LOREG;
+      dirty_reg(current,HIREG);
+      dirty_reg(current,LOREG);
+    }
+    else // 64-bit
+    {
+      alloc_x86_reg(current,i,HIREG|64,EDX);
+      alloc_x86_reg(current,i,HIREG,EAX);
+      alloc_reg64(current,i,rs1[i]);
+      alloc_reg64(current,i,rs2[i]);
+      alloc_all(current,i);
+      current->is32&=~(1LL<<HIREG);
+      current->is32&=~(1LL<<LOREG);
+      dirty_reg(current,HIREG);
+      dirty_reg(current,LOREG);
+    }
+  }
+  else
+  {
+    // Multiply by zero is zero.
+    // MIPS does not have a divide by zero exception.
+    // The result is undefined, we return zero.
+    alloc_reg(current,i,HIREG);
+    alloc_reg(current,i,LOREG);
+    current->is32|=1LL<<HIREG;
+    current->is32|=1LL<<LOREG;
+    dirty_reg(current,HIREG);
+    dirty_reg(current,LOREG);
+  }
+}
+#define multdiv_alloc multdiv_alloc_x86
+
+/* Assembler */
+
+char regname[8][4] = {
+ "eax",
+ "ecx",
+ "edx",
+ "ebx",
+ "esp",
+ "ebp",
+ "esi",
+ "edi"};
+
+void output_byte(u_char byte)
+{
+  *(out++)=byte;
+}
+void output_modrm(u_char mod,u_char rm,u_char ext)
+{
+  assert(mod<4);
+  assert(rm<8);
+  assert(ext<8);
+  u_char byte=(mod<<6)|(ext<<3)|rm;
+  *(out++)=byte;
+}
+void output_sib(u_char scale,u_char index,u_char base)
+{
+  assert(scale<4);
+  assert(index<8);
+  assert(base<8);
+  u_char byte=(scale<<6)|(index<<3)|base;
+  *(out++)=byte;
+}
+void output_w32(u_int word)
+{
+  *((u_int *)out)=word;
+  out+=4;
+}
+
+void emit_mov(int rs,int rt)
+{
+  assem_debug("mov %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x89);
+  output_modrm(3,rt,rs);
+}
+
+void emit_add(int rs1,int rs2,int rt)
+{
+  if(rs1==rt) {
+    assem_debug("add %%%s,%%%s\n",regname[rs2],regname[rs1]);
+    output_byte(0x01);
+    output_modrm(3,rs1,rs2);
+  }else if(rs2==rt) {
+    assem_debug("add %%%s,%%%s\n",regname[rs1],regname[rs2]);
+    output_byte(0x01);
+    output_modrm(3,rs2,rs1);
+  }else {
+    assem_debug("lea (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+    output_byte(0x8D);
+    if(rs1!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,rs2,rs1);
+    }else if(rs2!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,rs1,rs2);
+    }else /* lea 0(,%ebp,2) */{
+      output_modrm(0,4,rt);
+      output_sib(1,EBP,5);
+      output_w32(0);
+    }
+  }
+}
+
+void emit_adds(int rs1,int rs2,int rt)
+{
+  emit_add(rs1,rs2,rt);
+}
+
+void emit_lea8(int rs1,int rt)
+{
+  assem_debug("lea 0(%%%s,8),%%%s\n",regname[rs1],regname[rt]);
+  output_byte(0x8D);
+  output_modrm(0,4,rt);
+  output_sib(3,rs1,5);
+  output_w32(0);
+}
+void emit_leairrx1(int imm,int rs1,int rs2,int rt)
+{
+  assem_debug("lea %x(%%%s,%%%s,1),%%%s\n",imm,regname[rs1],regname[rs2],regname[rt]);
+  output_byte(0x8D);
+  if(imm!=0||rs1==EBP) {
+    output_modrm(2,4,rt);
+    output_sib(0,rs2,rs1);
+    output_w32(imm);
+  }else{
+    output_modrm(0,4,rt);
+    output_sib(0,rs2,rs1);
+  }
+}
+void emit_leairrx4(int imm,int rs1,int rs2,int rt)
+{
+  assem_debug("lea %x(%%%s,%%%s,4),%%%s\n",imm,regname[rs1],regname[rs2],regname[rt]);
+  output_byte(0x8D);
+  if(imm!=0||rs1==EBP) {
+    output_modrm(2,4,rt);
+    output_sib(2,rs2,rs1);
+    output_w32(imm);
+  }else{
+    output_modrm(0,4,rt);
+    output_sib(2,rs2,rs1);
+  }
+}
+
+void emit_neg(int rs, int rt)
+{
+  if(rs!=rt) emit_mov(rs,rt);
+  assem_debug("neg %%%s\n",regname[rt]);
+  output_byte(0xF7);
+  output_modrm(3,rt,3);
+}
+
+void emit_negs(int rs, int rt)
+{
+  emit_neg(rs,rt);
+}
+
+void emit_sub(int rs1,int rs2,int rt)
+{
+  if(rs1==rt) {
+    assem_debug("sub %%%s,%%%s\n",regname[rs2],regname[rs1]);
+    output_byte(0x29);
+    output_modrm(3,rs1,rs2);
+  } else if(rs2==rt) {
+    emit_neg(rs2,rs2);
+    emit_add(rs2,rs1,rs2);
+  } else {
+    emit_mov(rs1,rt);
+    emit_sub(rt,rs2,rt);
+  }
+}
+
+void emit_subs(int rs1,int rs2,int rt)
+{
+  emit_sub(rs1,rs2,rt);
+}
+
+void emit_zeroreg(int rt)
+{
+  output_byte(0x31);
+  output_modrm(3,rt,rt);
+  assem_debug("xor %%%s,%%%s\n",regname[rt],regname[rt]);
+}
+
+void emit_loadreg(int r, int hr)
+{
+  if((r&63)==0)
+    emit_zeroreg(hr);
+  else {
+    int addr=((int)reg)+((r&63)<<3)+((r&64)>>4);
+    if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
+    if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
+    if(r==CCREG) addr=(int)&cycle_count;
+    if(r==CSREG) addr=(int)&Status;
+    if(r==FSREG) addr=(int)&FCR31;
+    assem_debug("mov %x+%d,%%%s\n",addr,r,regname[hr]);
+    output_byte(0x8B);
+    output_modrm(0,5,hr);
+    output_w32(addr);
+  }
+}
+void emit_storereg(int r, int hr)
+{
+  int addr=((int)reg)+((r&63)<<3)+((r&64)>>4);
+  if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
+  if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
+  if(r==CCREG) addr=(int)&cycle_count;
+  if(r==FSREG) addr=(int)&FCR31;
+  assem_debug("mov %%%s,%x+%d\n",regname[hr],addr,r);
+  output_byte(0x89);
+  output_modrm(0,5,hr);
+  output_w32(addr);
+}
+
+void emit_test(int rs, int rt)
+{
+  assem_debug("test %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x85);
+  output_modrm(3,rs,rt);
+}
+
+void emit_testimm(int rs,int imm)
+{
+  assem_debug("test $0x%x,%%%s\n",imm,regname[rs]);
+  if(imm<128&&imm>=-128&&rs<4) {
+    output_byte(0xF6);
+    output_modrm(3,rs,0);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0xF7);
+    output_modrm(3,rs,0);
+    output_w32(imm);
+  }
+}
+
+void emit_not(int rs,int rt)
+{
+  if(rs!=rt) emit_mov(rs,rt);
+  assem_debug("not %%%s\n",regname[rt]);
+  output_byte(0xF7);
+  output_modrm(3,rt,2);
+}
+
+void emit_and(u_int rs1,u_int rs2,u_int rt)
+{
+  assert(rs1<8);
+  assert(rs2<8);
+  assert(rt<8);
+  if(rs1==rt) {
+    assem_debug("and %%%s,%%%s\n",regname[rs2],regname[rt]);
+    output_byte(0x21);
+    output_modrm(3,rs1,rs2);
+  }
+  else
+  if(rs2==rt) {
+    assem_debug("and %%%s,%%%s\n",regname[rs1],regname[rt]);
+    output_byte(0x21);
+    output_modrm(3,rs2,rs1);
+  }
+  else {
+    emit_mov(rs1,rt);
+    emit_and(rt,rs2,rt);
+  }
+}
+
+void emit_or(u_int rs1,u_int rs2,u_int rt)
+{
+  assert(rs1<8);
+  assert(rs2<8);
+  assert(rt<8);
+  if(rs1==rt) {
+    assem_debug("or %%%s,%%%s\n",regname[rs2],regname[rt]);
+    output_byte(0x09);
+    output_modrm(3,rs1,rs2);
+  }
+  else
+  if(rs2==rt) {
+    assem_debug("or %%%s,%%%s\n",regname[rs1],regname[rt]);
+    output_byte(0x09);
+    output_modrm(3,rs2,rs1);
+  }
+  else {
+    emit_mov(rs1,rt);
+    emit_or(rt,rs2,rt);
+  }
+}
+void emit_or_and_set_flags(int rs1,int rs2,int rt)
+{
+  emit_or(rs1,rs2,rt);
+}
+
+void emit_xor(u_int rs1,u_int rs2,u_int rt)
+{
+  assert(rs1<8);
+  assert(rs2<8);
+  assert(rt<8);
+  if(rs1==rt) {
+    assem_debug("xor %%%s,%%%s\n",regname[rs2],regname[rt]);
+    output_byte(0x31);
+    output_modrm(3,rs1,rs2);
+  }
+  else
+  if(rs2==rt) {
+    assem_debug("xor %%%s,%%%s\n",regname[rs1],regname[rt]);
+    output_byte(0x31);
+    output_modrm(3,rs2,rs1);
+  }
+  else {
+    emit_mov(rs1,rt);
+    emit_xor(rt,rs2,rt);
+  }
+}
+
+void emit_movimm(int imm,u_int rt)
+{
+  assem_debug("mov $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<8);
+  output_byte(0xB8+rt);
+  output_w32(imm);
+}
+
+void emit_addimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    if(imm!=0) {
+      assem_debug("add $%d,%%%s\n",imm,regname[rt]);
+      if(imm<128&&imm>=-128) {
+        output_byte(0x83);
+        output_modrm(3,rt,0);
+        output_byte(imm);
+      }
+      else
+      {
+        output_byte(0x81);
+        output_modrm(3,rt,0);
+        output_w32(imm);
+      }
+    }
+  }
+  else {
+    if(imm!=0) {
+      assem_debug("lea %d(%%%s),%%%s\n",imm,regname[rs],regname[rt]);
+      output_byte(0x8D);
+      if(imm<128&&imm>=-128) {
+        output_modrm(1,rs,rt);
+        output_byte(imm);
+      }else{
+        output_modrm(2,rs,rt);
+        output_w32(imm);
+      }
+    }else{
+      emit_mov(rs,rt);
+    }
+  }
+}
+
+void emit_addimm_and_set_flags(int imm,int rt)
+{
+  assem_debug("add $%d,%%%s\n",imm,regname[rt]);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,0);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,0);
+    output_w32(imm);
+  }
+}
+void emit_addimm_no_flags(int imm,int rt)
+{
+  if(imm!=0) {
+    assem_debug("lea %d(%%%s),%%%s\n",imm,regname[rt],regname[rt]);
+    output_byte(0x8D);
+    if(imm<128&&imm>=-128) {
+      output_modrm(1,rt,rt);
+      output_byte(imm);
+    }else{
+      output_modrm(2,rt,rt);
+      output_w32(imm);
+    }
+  }
+}
+
+void emit_adcimm(int imm,u_int rt)
+{
+  assem_debug("adc $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<8);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,2);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,2);
+    output_w32(imm);
+  }
+}
+void emit_sbbimm(int imm,u_int rt)
+{
+  assem_debug("sbb $%d,%%%s\n",imm,regname[rt]);
+  assert(rt<8);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rt,3);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rt,3);
+    output_w32(imm);
+  }
+}
+
+void emit_addimm64_32(int rsh,int rsl,int imm,int rth,int rtl)
+{
+  if(rsh==rth&&rsl==rtl) {
+    assem_debug("add $%d,%%%s\n",imm,regname[rtl]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rtl,0);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rtl,0);
+      output_w32(imm);
+    }
+    assem_debug("adc $%d,%%%s\n",imm>>31,regname[rth]);
+    output_byte(0x83);
+    output_modrm(3,rth,2);
+    output_byte(imm>>31);
+  }
+  else {
+    emit_mov(rsh,rth);
+    emit_mov(rsl,rtl);
+    emit_addimm64_32(rth,rtl,imm,rth,rtl);
+  }
+}
+
+void emit_sbb(int rs1,int rs2)
+{
+  assem_debug("sbb %%%s,%%%s\n",regname[rs2],regname[rs1]);
+  output_byte(0x19);
+  output_modrm(3,rs1,rs2);
+}
+
+void emit_andimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("and $%d,%%%s\n",imm,regname[rt]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rt,4);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rt,4);
+      output_w32(imm);
+    }
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_andimm(rt,imm,rt);
+  }
+}
+
+void emit_orimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("or $%d,%%%s\n",imm,regname[rt]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rt,1);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rt,1);
+      output_w32(imm);
+    }
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_orimm(rt,imm,rt);
+  }
+}
+
+void emit_xorimm(int rs,int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("xor $%d,%%%s\n",imm,regname[rt]);
+    if(imm<128&&imm>=-128) {
+      output_byte(0x83);
+      output_modrm(3,rt,6);
+      output_byte(imm);
+    }
+    else
+    {
+      output_byte(0x81);
+      output_modrm(3,rt,6);
+      output_w32(imm);
+    }
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_xorimm(rt,imm,rt);
+  }
+}
+
+void emit_shlimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shl %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,4);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shlimm(rt,imm,rt);
+  }
+}
+
+void emit_shrimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shr %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,5);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shrimm(rt,imm,rt);
+  }
+}
+
+void emit_sarimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("sar %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,7);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_sarimm(rt,imm,rt);
+  }
+}
+
+void emit_rorimm(int rs,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("ror %%%s,%d\n",regname[rt],imm);
+    assert(imm>0);
+    if(imm==1) output_byte(0xD1);
+    else output_byte(0xC1);
+    output_modrm(3,rt,1);
+    if(imm>1) output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_sarimm(rt,imm,rt);
+  }
+}
+
+void emit_shldimm(int rs,int rs2,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shld %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm);
+    assert(imm>0);
+    output_byte(0x0F);
+    output_byte(0xA4);
+    output_modrm(3,rt,rs2);
+    output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shldimm(rt,rs2,imm,rt);
+  }
+}
+
+void emit_shrdimm(int rs,int rs2,u_int imm,int rt)
+{
+  if(rs==rt) {
+    assem_debug("shrd %%%s,%%%s,%d\n",regname[rt],regname[rs2],imm);
+    assert(imm>0);
+    output_byte(0x0F);
+    output_byte(0xAC);
+    output_modrm(3,rt,rs2);
+    output_byte(imm);
+  }
+  else {
+    emit_mov(rs,rt);
+    emit_shrdimm(rt,rs2,imm,rt);
+  }
+}
+
+void emit_shlcl(int r)
+{
+  assem_debug("shl %%%s,%%cl\n",regname[r]);
+  output_byte(0xD3);
+  output_modrm(3,r,4);
+}
+void emit_shrcl(int r)
+{
+  assem_debug("shr %%%s,%%cl\n",regname[r]);
+  output_byte(0xD3);
+  output_modrm(3,r,5);
+}
+void emit_sarcl(int r)
+{
+  assem_debug("sar %%%s,%%cl\n",regname[r]);
+  output_byte(0xD3);
+  output_modrm(3,r,7);
+}
+
+void emit_shldcl(int r1,int r2)
+{
+  assem_debug("shld %%%s,%%%s,%%cl\n",regname[r1],regname[r2]);
+  output_byte(0x0F);
+  output_byte(0xA5);
+  output_modrm(3,r1,r2);
+}
+void emit_shrdcl(int r1,int r2)
+{
+  assem_debug("shrd %%%s,%%%s,%%cl\n",regname[r1],regname[r2]);
+  output_byte(0x0F);
+  output_byte(0xAD);
+  output_modrm(3,r1,r2);
+}
+
+void emit_cmpimm(int rs,int imm)
+{
+  assem_debug("cmp $%d,%%%s\n",imm,regname[rs]);
+  if(imm<128&&imm>=-128) {
+    output_byte(0x83);
+    output_modrm(3,rs,7);
+    output_byte(imm);
+  }
+  else
+  {
+    output_byte(0x81);
+    output_modrm(3,rs,7);
+    output_w32(imm);
+  }
+}
+
+void emit_cmovne(u_int *addr,int rt)
+{
+  assem_debug("cmovne %x,%%%s",(int)addr,regname[rt]);
+  if(addr==&const_zero) assem_debug(" [zero]\n");
+  else if(addr==&const_one) assem_debug(" [one]\n");
+  else assem_debug("\n");
+  output_byte(0x0F);
+  output_byte(0x45);
+  output_modrm(0,5,rt);
+  output_w32((int)addr);
+}
+void emit_cmovl(u_int *addr,int rt)
+{
+  assem_debug("cmovl %x,%%%s",(int)addr,regname[rt]);
+  if(addr==&const_zero) assem_debug(" [zero]\n");
+  else if(addr==&const_one) assem_debug(" [one]\n");
+  else assem_debug("\n");
+  output_byte(0x0F);
+  output_byte(0x4C);
+  output_modrm(0,5,rt);
+  output_w32((int)addr);
+}
+void emit_cmovs(u_int *addr,int rt)
+{
+  assem_debug("cmovs %x,%%%s",(int)addr,regname[rt]);
+  if(addr==&const_zero) assem_debug(" [zero]\n");
+  else if(addr==&const_one) assem_debug(" [one]\n");
+  else assem_debug("\n");
+  output_byte(0x0F);
+  output_byte(0x48);
+  output_modrm(0,5,rt);
+  output_w32((int)addr);
+}
+void emit_cmovne_reg(int rs,int rt)
+{
+  assem_debug("cmovne %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x45);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovl_reg(int rs,int rt)
+{
+  assem_debug("cmovl %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x4C);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovs_reg(int rs,int rt)
+{
+  assem_debug("cmovs %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x48);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovnc_reg(int rs,int rt)
+{
+  assem_debug("cmovae %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x43);
+  output_modrm(3,rs,rt);
+}
+void emit_cmova_reg(int rs,int rt)
+{
+  assem_debug("cmova %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x47);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovp_reg(int rs,int rt)
+{
+  assem_debug("cmovp %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x4A);
+  output_modrm(3,rs,rt);
+}
+void emit_cmovnp_reg(int rs,int rt)
+{
+  assem_debug("cmovnp %%%s,%%%s\n",regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x4B);
+  output_modrm(3,rs,rt);
+}
+void emit_setl(int rt)
+{
+  assem_debug("setl %%%s\n",regname[rt]);
+  output_byte(0x0F);
+  output_byte(0x9C);
+  output_modrm(3,rt,2);
+}
+void emit_movzbl_reg(int rs, int rt)
+{
+  assem_debug("movzbl %%%s,%%%s\n",regname[rs]+1,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB6);
+  output_modrm(3,rs,rt);
+}
+
+void emit_slti32(int rs,int imm,int rt)
+{
+  if(rs!=rt) emit_zeroreg(rt);
+  emit_cmpimm(rs,imm);
+  if(rt<4) {
+    emit_setl(rt);
+    if(rs==rt) emit_movzbl_reg(rt,rt);
+  }
+  else
+  {
+    if(rs==rt) emit_movimm(0,rt);
+    emit_cmovl(&const_one,rt);
+  }
+}
+void emit_sltiu32(int rs,int imm,int rt)
+{
+  if(rs!=rt) emit_zeroreg(rt);
+  emit_cmpimm(rs,imm);
+  if(rs==rt) emit_movimm(0,rt);
+  emit_adcimm(0,rt);
+}
+void emit_slti64_32(int rsh,int rsl,int imm,int rt)
+{
+  assert(rsh!=rt);
+  emit_slti32(rsl,imm,rt);
+  if(imm>=0)
+  {
+    emit_test(rsh,rsh);
+    emit_cmovne(&const_zero,rt);
+    emit_cmovs(&const_one,rt);
+  }
+  else
+  {
+    emit_cmpimm(rsh,-1);
+    emit_cmovne(&const_zero,rt);
+    emit_cmovl(&const_one,rt);
+  }
+}
+void emit_sltiu64_32(int rsh,int rsl,int imm,int rt)
+{
+  assert(rsh!=rt);
+  emit_sltiu32(rsl,imm,rt);
+  if(imm>=0)
+  {
+    emit_test(rsh,rsh);
+    emit_cmovne(&const_zero,rt);
+  }
+  else
+  {
+    emit_cmpimm(rsh,-1);
+    emit_cmovne(&const_one,rt);
+  }
+}
+
+void emit_cmp(int rs,int rt)
+{
+  assem_debug("cmp %%%s,%%%s\n",regname[rt],regname[rs]);
+  output_byte(0x39);
+  output_modrm(3,rs,rt);
+}
+void emit_set_gz32(int rs, int rt)
+{
+  //assem_debug("set_gz32\n");
+  emit_cmpimm(rs,1);
+  emit_movimm(1,rt);
+  emit_cmovl(&const_zero,rt);
+}
+void emit_set_nz32(int rs, int rt)
+{
+  //assem_debug("set_nz32\n");
+  emit_cmpimm(rs,1);
+  emit_movimm(1,rt);
+  emit_sbbimm(0,rt);
+}
+void emit_set_gz64_32(int rsh, int rsl, int rt)
+{
+  //assem_debug("set_gz64\n");
+  emit_set_gz32(rsl,rt);
+  emit_test(rsh,rsh);
+  emit_cmovne(&const_one,rt);
+  emit_cmovs(&const_zero,rt);
+}
+void emit_set_nz64_32(int rsh, int rsl, int rt)
+{
+  //assem_debug("set_nz64\n");
+  emit_or_and_set_flags(rsh,rsl,rt);
+  emit_cmovne(&const_one,rt);
+}
+void emit_set_if_less32(int rs1, int rs2, int rt)
+{
+  //assem_debug("set if less (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+  if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt);
+  emit_cmp(rs1,rs2);
+  if(rs1==rt||rs2==rt) emit_movimm(0,rt);
+  emit_cmovl(&const_one,rt);
+}
+void emit_set_if_carry32(int rs1, int rs2, int rt)
+{
+  //assem_debug("set if carry (%%%s,%%%s),%%%s\n",regname[rs1],regname[rs2],regname[rt]);
+  if(rs1!=rt&&rs2!=rt) emit_zeroreg(rt);
+  emit_cmp(rs1,rs2);
+  if(rs1==rt||rs2==rt) emit_movimm(0,rt);
+  emit_adcimm(0,rt);
+}
+void emit_set_if_less64_32(int u1, int l1, int u2, int l2, int rt)
+{
+  //assem_debug("set if less64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]);
+  assert(u1!=rt);
+  assert(u2!=rt);
+  emit_cmp(l1,l2);
+  emit_mov(u1,rt);
+  emit_sbb(rt,u2);
+  emit_movimm(0,rt);
+  emit_cmovl(&const_one,rt);
+}
+void emit_set_if_carry64_32(int u1, int l1, int u2, int l2, int rt)
+{
+  //assem_debug("set if carry64 (%%%s,%%%s,%%%s,%%%s),%%%s\n",regname[u1],regname[l1],regname[u2],regname[l2],regname[rt]);
+  assert(u1!=rt);
+  assert(u2!=rt);
+  emit_cmp(l1,l2);
+  emit_mov(u1,rt);
+  emit_sbb(rt,u2);
+  emit_movimm(0,rt);
+  emit_adcimm(0,rt);
+}
+
+void emit_call(int a)
+{
+  assem_debug("call %x (%x+%x)\n",a,(int)out+5,a-(int)out-5);
+  output_byte(0xe8);
+  output_w32(a-(int)out-4);
+}
+void emit_jmp(int a)
+{
+  assem_debug("jmp %x (%x+%x)\n",a,(int)out+5,a-(int)out-5);
+  output_byte(0xe9);
+  output_w32(a-(int)out-4);
+}
+void emit_jne(int a)
+{
+  assem_debug("jne %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x85);
+  output_w32(a-(int)out-4);
+}
+void emit_jeq(int a)
+{
+  assem_debug("jeq %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x84);
+  output_w32(a-(int)out-4);
+}
+void emit_js(int a)
+{
+  assem_debug("js %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x88);
+  output_w32(a-(int)out-4);
+}
+void emit_jns(int a)
+{
+  assem_debug("jns %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x89);
+  output_w32(a-(int)out-4);
+}
+void emit_jl(int a)
+{
+  assem_debug("jl %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x8c);
+  output_w32(a-(int)out-4);
+}
+void emit_jge(int a)
+{
+  assem_debug("jge %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x8d);
+  output_w32(a-(int)out-4);
+}
+void emit_jno(int a)
+{
+  assem_debug("jno %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x81);
+  output_w32(a-(int)out-4);
+}
+void emit_jc(int a)
+{
+  assem_debug("jc %x\n",a);
+  output_byte(0x0f);
+  output_byte(0x82);
+  output_w32(a-(int)out-4);
+}
+
+void emit_pushimm(int imm)
+{
+  assem_debug("push $%x\n",imm);
+  output_byte(0x68);
+  output_w32(imm);
+}
+void emit_pushmem(int addr)
+{
+  assem_debug("push *%x\n",addr);
+  output_byte(0xFF);
+  output_modrm(0,5,6);
+  output_w32(addr);
+}
+void emit_pusha()
+{
+  assem_debug("pusha\n");
+  output_byte(0x60);
+}
+void emit_popa()
+{
+  assem_debug("popa\n");
+  output_byte(0x61);
+}
+void emit_pushreg(u_int r)
+{
+  assem_debug("push %%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0x50+r);
+}
+void emit_popreg(u_int r)
+{
+  assem_debug("pop %%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0x58+r);
+}
+void emit_callreg(u_int r)
+{
+  assem_debug("call *%%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0xFF);
+  output_modrm(3,r,2);
+}
+void emit_jmpreg(u_int r)
+{
+  assem_debug("jmp *%%%s\n",regname[r]);
+  assert(r<8);
+  output_byte(0xFF);
+  output_modrm(3,r,4);
+}
+void emit_jmpmem_indexed(u_int addr,u_int r)
+{
+  assem_debug("jmp *%x(%%%s)\n",addr,regname[r]);
+  assert(r<8);
+  output_byte(0xFF);
+  output_modrm(2,r,4);
+  output_w32(addr);
+}
+
+void emit_readword(int addr, int rt)
+{
+  assem_debug("mov %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x8B);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_readword_indexed(int addr, int rs, int rt)
+{
+  assem_debug("mov %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x8B);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_w32(addr);
+  }
+}
+void emit_readword_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_readword(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("mov (%x,%%%s,4),%%%s\n",addr+(int)rdram-0x80000000,regname[map],regname[rt]);
+    output_byte(0x8B);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr+(int)rdram-0x80000000);
+  }
+}
+void emit_readword_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_readword_indexed(addr+(int)rdram-0x80000000, rs, rt);
+  else {
+    assem_debug("mov %x(%%%s,%%%s,4),%%%s\n",addr,regname[rs],regname[map],regname[rt]);
+    assert(rs!=ESP);
+    output_byte(0x8B);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(2,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(2,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(2,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_movmem_indexedx4(int addr, int rs, int rt)
+{
+  assem_debug("mov (%x,%%%s,4),%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x8B);
+  output_modrm(0,4,rt);
+  output_sib(2,rs,5);
+  output_w32(addr);
+}
+void emit_readdword_tlb(int addr, int map, int rh, int rl)
+{
+  if(map<0) {
+    if(rh>=0) emit_readword(addr+(int)rdram-0x80000000, rh);
+    emit_readword(addr+(int)rdram-0x7FFFFFFC, rl);
+  }
+  else {
+    if(rh>=0) emit_movmem_indexedx4(addr+(int)rdram-0x80000000, map, rh);
+    emit_movmem_indexedx4(addr+(int)rdram-0x7FFFFFFC, map, rl);
+  }
+}
+void emit_readdword_indexed_tlb(int addr, int rs, int map, int rh, int rl)
+{
+  assert(rh!=rs);
+  if(rh>=0) emit_readword_indexed_tlb(addr, rs, map, rh);
+  emit_readword_indexed_tlb(addr+4, rs, map, rl);
+}
+void emit_movsbl(int addr, int rt)
+{
+  assem_debug("movsbl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBE);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_movsbl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movsbl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBE);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movsbl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movsbl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("movsbl (%x,%%%s,4),%%%s\n",addr+(int)rdram-0x80000000,regname[map],regname[rt]);
+    output_byte(0x0F);
+    output_byte(0xBE);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr+(int)rdram-0x80000000);
+  }
+}
+void emit_movsbl_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_movsbl_indexed(addr+(int)rdram-0x80000000, rs, rt);
+  else {
+    assem_debug("movsbl %x(%%%s,%%%s,4),%%%s\n",addr,regname[rs],regname[map],regname[rt]);
+    assert(rs!=ESP);
+    output_byte(0x0F);
+    output_byte(0xBE);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(2,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(2,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(2,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_movswl(int addr, int rt)
+{
+  assem_debug("movswl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBF);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_movswl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movswl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xBF);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movswl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movswl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("movswl (%x,%%%s,4),%%%s\n",addr+(int)rdram-0x80000000,regname[map],regname[rt]);
+    output_byte(0x0F);
+    output_byte(0xBF);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr+(int)rdram-0x80000000);
+  }
+}
+void emit_movzbl(int addr, int rt)
+{
+  assem_debug("movzbl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB6);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_movzbl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movzbl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB6);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movzbl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movzbl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("movzbl (%x,%%%s,4),%%%s\n",addr+(int)rdram-0x80000000,regname[map],regname[rt]);
+    output_byte(0x0F);
+    output_byte(0xB6);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr+(int)rdram-0x80000000);
+  }
+}
+void emit_movzbl_indexed_tlb(int addr, int rs, int map, int rt)
+{
+  if(map<0) emit_movzbl_indexed(addr+(int)rdram-0x80000000, rs, rt);
+  else {
+    assem_debug("movzbl %x(%%%s,%%%s,4),%%%s\n",addr,regname[rs],regname[map],regname[rt]);
+    assert(rs!=ESP);
+    output_byte(0x0F);
+    output_byte(0xB6);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(2,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(2,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(2,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_movzwl(int addr, int rt)
+{
+  assem_debug("movzwl %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB7);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_movzwl_indexed(int addr, int rs, int rt)
+{
+  assem_debug("movzwl %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB7);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+void emit_movzwl_tlb(int addr, int map, int rt)
+{
+  if(map<0) emit_movzwl(addr+(int)rdram-0x80000000, rt);
+  else
+  {
+    assem_debug("movzwl (%x,%%%s,4),%%%s\n",addr+(int)rdram-0x80000000,regname[map],regname[rt]);
+    output_byte(0x0F);
+    output_byte(0xB7);
+    output_modrm(0,4,rt);
+    output_sib(2,map,5);
+    output_w32(addr+(int)rdram-0x80000000);
+  }
+}
+void emit_movzwl_reg(int rs, int rt)
+{
+  assem_debug("movzwl %%%s,%%%s\n",regname[rs]+1,regname[rt]);
+  output_byte(0x0F);
+  output_byte(0xB7);
+  output_modrm(3,rs,rt);
+}
+
+void emit_xchg(int rs, int rt)
+{
+  assem_debug("xchg %%%s,%%%s\n",regname[rs],regname[rt]);
+  if(rs==EAX) {
+    output_byte(0x90+rt);
+  }
+  else
+  {
+    output_byte(0x87);
+    output_modrm(3,rs,rt);
+  }
+}
+void emit_writeword(int rt, int addr)
+{
+  assem_debug("movl %%%s,%x\n",regname[rt],addr);
+  output_byte(0x89);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_writeword_indexed(int rt, int addr, int rs)
+{
+  assem_debug("mov %%%s,%x+%%%s\n",regname[rt],addr,regname[rs]);
+  output_byte(0x89);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs,rt);
+    if(rs==ESP) output_sib(0,4,4);
+    output_w32(addr);
+  }
+}
+void emit_writeword_tlb(int rt, int addr, int map)
+{
+  if(map<0) {
+    emit_writeword(rt, addr+(int)rdram-0x80000000);
+  } else {
+    emit_writeword_indexed(rt, addr+(int)rdram-0x80000000, map);
+  }
+}
+void emit_writeword_indexed_tlb(int rt, int addr, int rs, int map, int temp)
+{
+  if(map<0) emit_writeword_indexed(rt, addr+(int)rdram-0x80000000, rs);
+  else {
+    assem_debug("mov %%%s,%x(%%%s,%%%s,1)\n",regname[rt],addr,regname[rs],regname[map]);
+    assert(rs!=ESP);
+    output_byte(0x89);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(0,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(0,map,rs);
+      output_w32(addr);
+    }
+  }
+}
+void emit_writedword_tlb(int rh, int rl, int addr, int map)
+{
+  assert(rh>=0);
+  if(map<0) {
+    emit_writeword(rh, addr+(int)rdram-0x80000000);
+    emit_writeword(rl, addr+(int)rdram-0x7FFFFFFC);
+  }
+  else {
+    emit_writeword_indexed(rh, addr+(int)rdram-0x80000000, map);
+    emit_writeword_indexed(rl, addr+(int)rdram-0x7FFFFFFC, map);
+  }
+}
+void emit_writedword_indexed_tlb(int rh, int rl, int addr, int rs, int map, int temp)
+{
+  assert(rh>=0);
+  emit_writeword_indexed_tlb(rh, addr, rs, map, temp);
+  emit_writeword_indexed_tlb(rl, addr+4, rs, map, temp);
+}
+void emit_writehword(int rt, int addr)
+{
+  assem_debug("movw %%%s,%x\n",regname[rt]+1,addr);
+  output_byte(0x66);
+  output_byte(0x89);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+void emit_writehword_indexed(int rt, int addr, int rs)
+{
+  assem_debug("movw %%%s,%x+%%%s\n",regname[rt]+1,addr,regname[rs]);
+  output_byte(0x66);
+  output_byte(0x89);
+  if(addr<128&&addr>=-128) {
+    output_modrm(1,rs,rt);
+    output_byte(addr);
+  }
+  else
+  {
+    output_modrm(2,rs,rt);
+    output_w32(addr);
+  }
+}
+void emit_writehword_tlb(int rt, int addr, int map)
+{
+  if(map<0) {
+    emit_writehword(rt, addr+(int)rdram-0x80000000);
+  } else {
+    emit_writehword_indexed(rt, addr+(int)rdram-0x80000000, map);
+  }
+}
+void emit_writebyte(int rt, int addr)
+{
+  if(rt<4) {
+    assem_debug("movb %%%cl,%x\n",regname[rt][1],addr);
+    output_byte(0x88);
+    output_modrm(0,5,rt);
+    output_w32(addr);
+  }
+  else
+  {
+    emit_xchg(EAX,rt);
+    emit_writebyte(EAX,addr);
+    emit_xchg(EAX,rt);
+  }
+}
+void emit_writebyte_indexed(int rt, int addr, int rs)
+{
+  if(rt<4) {
+    assem_debug("movb %%%cl,%x+%%%s\n",regname[rt][1],addr,regname[rs]);
+    output_byte(0x88);
+    if(addr<128&&addr>=-128) {
+      output_modrm(1,rs,rt);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,rs,rt);
+      output_w32(addr);
+    }
+  }
+  else
+  {
+    emit_xchg(EAX,rt);
+    emit_writebyte_indexed(EAX,addr,rs==EAX?rt:rs);
+    emit_xchg(EAX,rt);
+  }
+}
+void emit_writebyte_tlb(int rt, int addr, int map)
+{
+  if(map<0) {
+    emit_writebyte(rt, addr+(int)rdram-0x80000000);
+  } else {
+    emit_writebyte_indexed(rt, addr+(int)rdram-0x80000000, map);
+  }
+}
+void emit_writebyte_indexed_tlb(int rt, int addr, int rs, int map, int temp)
+{
+  if(map<0) emit_writebyte_indexed(rt, addr+(int)rdram-0x80000000, rs);
+  else
+  if(rt<4) {
+    assem_debug("movb %%%cl,%x(%%%s,%%%s,1)\n",regname[rt][1],addr,regname[rs],regname[map]);
+    assert(rs!=ESP);
+    output_byte(0x88);
+    if(addr==0&&rs!=EBP) {
+      output_modrm(0,4,rt);
+      output_sib(0,map,rs);
+    }
+    else if(addr<128&&addr>=-128) {
+      output_modrm(1,4,rt);
+      output_sib(0,map,rs);
+      output_byte(addr);
+    }
+    else
+    {
+      output_modrm(2,4,rt);
+      output_sib(0,map,rs);
+      output_w32(addr);
+    }
+  }
+  else
+  {
+    emit_xchg(EAX,rt);
+    emit_writebyte_indexed_tlb(EAX,addr,rs==EAX?rt:rs,map==EAX?rt:map,temp);
+    emit_xchg(EAX,rt);
+  }
+}
+void emit_writeword_imm(int imm, int addr)
+{
+  assem_debug("movl $%x,%x\n",imm,addr);
+  output_byte(0xC7);
+  output_modrm(0,5,0);
+  output_w32(addr);
+  output_w32(imm);
+}
+void emit_writeword_imm_esp(int imm, int addr)
+{
+  assem_debug("mov $%x,%x(%%esp)\n",imm,addr);
+  assert(addr>=-128&&addr<128);
+  output_byte(0xC7);
+  output_modrm(1,4,0);
+  output_sib(0,4,4);
+  output_byte(addr);
+  output_w32(imm);
+}
+void emit_writebyte_imm(int imm, int addr)
+{
+  assem_debug("movb $%x,%x\n",imm,addr);
+  assert(imm>=-128&&imm<128);
+  output_byte(0xC6);
+  output_modrm(0,5,0);
+  output_w32(addr);
+  output_byte(imm);
+}
+void emit_writebyte_imm_esp(int imm, int addr)
+{
+  assem_debug("movb $%x,%x(%%esp)\n",imm,addr);
+  assert(addr>=-128&&addr<128);
+  output_byte(0xC6);
+  output_modrm(1,4,0);
+  output_sib(0,4,4);
+  output_byte(addr);
+  output_byte(imm);
+}
+
+void emit_mul(int rs)
+{
+  assem_debug("mul %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,4);
+}
+void emit_imul(int rs)
+{
+  assem_debug("imul %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,5);
+}
+void emit_div(int rs)
+{
+  assem_debug("div %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,6);
+}
+void emit_idiv(int rs)
+{
+  assem_debug("idiv %%%s\n",regname[rs]);
+  output_byte(0xF7);
+  output_modrm(3,rs,7);
+}
+void emit_cdq()
+{
+  assem_debug("cdq\n");
+  output_byte(0x99);
+}
+
+// Load 2 immediates optimizing for small code size
+void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
+{
+  emit_movimm(imm1,rt1);
+  if(imm2-imm1<128&&imm2-imm1>=-128) emit_addimm(rt1,imm2-imm1,rt2);
+  else emit_movimm(imm2,rt2);
+}
+
+// special case for checking pending_exception
+void emit_cmpmem_imm_byte(int addr,int imm)
+{
+  assert(imm<128&&imm>=-127);
+  assem_debug("cmpb $%d,%x\n",imm,addr);
+  output_byte(0x80);
+  output_modrm(0,5,7);
+  output_w32(addr);
+  output_byte(imm);
+}
+
+// special case for checking invalid_code
+void emit_cmpmem_indexedsr12_imm(int addr,int r,int imm)
+{
+  assert(imm<128&&imm>=-127);
+  assert(r>=0&&r<8);
+  emit_shrimm(r,12,r);
+  assem_debug("cmp $%d,%x+%%%s\n",imm,addr,regname[r]);
+  output_byte(0x80);
+  output_modrm(2,r,7);
+  output_w32(addr);
+  output_byte(imm);
+}
+
+// special case for checking hash_table
+void emit_cmpmem_indexed(int addr,int rs,int rt)
+{
+  assert(rs>=0&&rs<8);
+  assert(rt>=0&&rt<8);
+  assem_debug("cmp %x+%%%s,%%%s\n",addr,regname[rs],regname[rt]);
+  output_byte(0x39);
+  output_modrm(2,rs,rt);
+  output_w32(addr);
+}
+
+// special case for checking memory_map in verify_mapping
+void emit_cmpmem(int addr,int rt)
+{
+  assert(rt>=0&&rt<8);
+  assem_debug("cmp %x,%%%s\n",addr,regname[rt]);
+  output_byte(0x39);
+  output_modrm(0,5,rt);
+  output_w32(addr);
+}
+
+// Used to preload hash table entries
+void emit_prefetch(void *addr)
+{
+  assem_debug("prefetch %x\n",(int)addr);
+  output_byte(0x0F);
+  output_byte(0x18);
+  output_modrm(0,5,1);
+  output_w32((int)addr);
+}
+
+/*void emit_submem(int r,int addr)
+{
+  assert(r>=0&&r<8);
+  assem_debug("sub %x,%%%s\n",addr,regname[r]);
+  output_byte(0x2B);
+  output_modrm(0,5,r);
+  output_w32((int)addr);
+}*/
+void emit_subfrommem(int addr,int r)
+{
+  assert(r>=0&&r<8);
+  assem_debug("sub %%%s,%x\n",regname[r],addr);
+  output_byte(0x29);
+  output_modrm(0,5,r);
+  output_w32((int)addr);
+}
+
+void emit_flds(int r)
+{
+  assem_debug("flds (%%%s)\n",regname[r]);
+  output_byte(0xd9);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fldl(int r)
+{
+  assem_debug("fldl (%%%s)\n",regname[r]);
+  output_byte(0xdd);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fucomip(u_int r)
+{
+  assem_debug("fucomip %d\n",r);
+  assert(r<8);
+  output_byte(0xdf);
+  output_byte(0xe8+r);
+}
+void emit_fchs()
+{
+  assem_debug("fchs\n");
+  output_byte(0xd9);
+  output_byte(0xe0);
+}
+void emit_fabs()
+{
+  assem_debug("fabs\n");
+  output_byte(0xd9);
+  output_byte(0xe1);
+}
+void emit_fsqrt()
+{
+  assem_debug("fsqrt\n");
+  output_byte(0xd9);
+  output_byte(0xfa);
+}
+void emit_fadds(int r)
+{
+  assem_debug("fadds (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_faddl(int r)
+{
+  assem_debug("faddl (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fadd(int r)
+{
+  assem_debug("fadd st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xc0+r);
+}
+void emit_fsubs(int r)
+{
+  assem_debug("fsubs (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,4);
+  else {output_modrm(1,EBP,4);output_byte(0);}
+}
+void emit_fsubl(int r)
+{
+  assem_debug("fsubl (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,4);
+  else {output_modrm(1,EBP,4);output_byte(0);}
+}
+void emit_fsub(int r)
+{
+  assem_debug("fsub st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xe0+r);
+}
+void emit_fmuls(int r)
+{
+  assem_debug("fmuls (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,1);
+  else {output_modrm(1,EBP,1);output_byte(0);}
+}
+void emit_fmull(int r)
+{
+  assem_debug("fmull (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,1);
+  else {output_modrm(1,EBP,1);output_byte(0);}
+}
+void emit_fmul(int r)
+{
+  assem_debug("fmul st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xc8+r);
+}
+void emit_fdivs(int r)
+{
+  assem_debug("fdivs (%%%s)\n",regname[r]);
+  output_byte(0xd8);
+  if(r!=EBP) output_modrm(0,r,6);
+  else {output_modrm(1,EBP,6);output_byte(0);}
+}
+void emit_fdivl(int r)
+{
+  assem_debug("fdivl (%%%s)\n",regname[r]);
+  output_byte(0xdc);
+  if(r!=EBP) output_modrm(0,r,6);
+  else {output_modrm(1,EBP,6);output_byte(0);}
+}
+void emit_fdiv(int r)
+{
+  assem_debug("fdiv st%d\n",r);
+  output_byte(0xd8);
+  output_byte(0xf0+r);
+}
+void emit_fpop()
+{
+  // fstp st(0)
+  assem_debug("fpop\n");
+  output_byte(0xdd);
+  output_byte(0xd8);
+}
+void emit_fildl(int r)
+{
+  assem_debug("fildl (%%%s)\n",regname[r]);
+  output_byte(0xdb);
+  if(r!=EBP) output_modrm(0,r,0);
+  else {output_modrm(1,EBP,0);output_byte(0);}
+}
+void emit_fildll(int r)
+{
+  assem_debug("fildll (%%%s)\n",regname[r]);
+  output_byte(0xdf);
+  if(r!=EBP) output_modrm(0,r,5);
+  else {output_modrm(1,EBP,5);output_byte(0);}
+}
+void emit_fistpl(int r)
+{
+  assem_debug("fistpl (%%%s)\n",regname[r]);
+  output_byte(0xdb);
+  if(r!=EBP) output_modrm(0,r,3);
+  else {output_modrm(1,EBP,3);output_byte(0);}
+}
+void emit_fistpll(int r)
+{
+  assem_debug("fistpll (%%%s)\n",regname[r]);
+  output_byte(0xdf);
+  if(r!=EBP) output_modrm(0,r,7);
+  else {output_modrm(1,EBP,7);output_byte(0);}
+}
+void emit_fstps(int r)
+{
+  assem_debug("fstps (%%%s)\n",regname[r]);
+  output_byte(0xd9);
+  if(r!=EBP) output_modrm(0,r,3);
+  else {output_modrm(1,EBP,3);output_byte(0);}
+}
+void emit_fstpl(int r)
+{
+  assem_debug("fstpl (%%%s)\n",regname[r]);
+  output_byte(0xdd);
+  if(r!=EBP) output_modrm(0,r,3);
+  else {output_modrm(1,EBP,3);output_byte(0);}
+}
+void emit_fnstcw_stack()
+{
+  assem_debug("fnstcw (%%esp)\n");
+  output_byte(0xd9);
+  output_modrm(0,4,7);
+  output_sib(0,4,4);
+}
+void emit_fldcw_stack()
+{
+  assem_debug("fldcw (%%esp)\n");
+  output_byte(0xd9);
+  output_modrm(0,4,5);
+  output_sib(0,4,4);
+}
+void emit_fldcw_indexed(int addr,int r)
+{
+  assem_debug("fldcw %x(%%%s)\n",addr,regname[r]);
+  output_byte(0xd9);
+  output_modrm(0,4,5);
+  output_sib(1,r,5);
+  output_w32(addr);
+}
+void emit_fldcw(int addr)
+{
+  assem_debug("fldcw %x\n",addr);
+  output_byte(0xd9);
+  output_modrm(0,5,5);
+  output_w32(addr);
+}
+void emit_movss_load(u_int addr,u_int ssereg)
+{
+  assem_debug("movss (%%%s),xmm%d\n",regname[addr],ssereg);
+  assert(ssereg<8);
+  output_byte(0xf3);
+  output_byte(0x0f);
+  output_byte(0x10);
+  if(addr!=EBP) output_modrm(0,addr,ssereg);
+  else {output_modrm(1,EBP,ssereg);output_byte(0);}
+}
+void emit_movsd_load(u_int addr,u_int ssereg)
+{
+  assem_debug("movsd (%%%s),xmm%d\n",regname[addr],ssereg);
+  assert(ssereg<8);
+  output_byte(0xf2);
+  output_byte(0x0f);
+  output_byte(0x10);
+  if(addr!=EBP) output_modrm(0,addr,ssereg);
+  else {output_modrm(1,EBP,ssereg);output_byte(0);}
+}
+void emit_movd_store(u_int ssereg,u_int addr)
+{
+  assem_debug("movd xmm%d,(%%%s)\n",ssereg,regname[addr]);
+  assert(ssereg<8);
+  output_byte(0x66);
+  output_byte(0x0f);
+  output_byte(0x7e);
+  if(addr!=EBP) output_modrm(0,addr,ssereg);
+  else {output_modrm(1,EBP,ssereg);output_byte(0);}
+}
+void emit_cvttps2dq(u_int ssereg1,u_int ssereg2)
+{
+  assem_debug("cvttps2dq xmm%d,xmm%d\n",ssereg1,ssereg2);
+  assert(ssereg1<8);
+  assert(ssereg2<8);
+  output_byte(0xf3);
+  output_byte(0x0f);
+  output_byte(0x5b);
+  output_modrm(3,ssereg1,ssereg2);
+}
+void emit_cvttpd2dq(u_int ssereg1,u_int ssereg2)
+{
+  assem_debug("cvttpd2dq xmm%d,xmm%d\n",ssereg1,ssereg2);
+  assert(ssereg1<8);
+  assert(ssereg2<8);
+  output_byte(0x66);
+  output_byte(0x0f);
+  output_byte(0xe6);
+  output_modrm(3,ssereg1,ssereg2);
+}
+
+/* Stubs/epilogue */
+
+emit_extjump2(int addr, int target, int linker)
+{
+  u_char *ptr=(u_char *)addr;
+  if(*ptr==0x0f)
+  {
+    assert(ptr[1]>=0x80&&ptr[1]<=0x8f);
+    addr+=2;
+  }
+  else
+  {
+    assert(*ptr==0xe8||*ptr==0xe9);
+    addr++;
+  }
+  emit_movimm(target,EAX);
+  emit_movimm(addr,EBX);
+  //assert(addr>=0x7000000&&addr<0x7FFFFFF);
+  //assert((target>=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000));
+//DEBUG >
+#ifdef DEBUG_CYCLE_COUNT
+  emit_readword((int)&last_count,ECX);
+  emit_add(HOST_CCREG,ECX,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_writeword(HOST_CCREG,(int)&Count);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+#endif
+//DEBUG <
+  emit_jmp(linker);
+}
+
+emit_extjump(int addr, int target)
+{
+  emit_extjump2(addr, target, (int)dyna_linker);
+}
+emit_extjump_ds(int addr, int target)
+{
+  emit_extjump2(addr, target, (int)dyna_linker_ds);
+}
+
+do_readstub(int n)
+{
+  assem_debug("do_readstub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int type=stubs[n][0];
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  signed char *i_regmap=i_regs->regmap;
+  int addr=get_reg(i_regmap,AGEN1+(i&1));
+  int rth,rt;
+  int ds;
+  if(itype[i]==C1LS||itype[i]==LOADLR) {
+    rth=get_reg(i_regmap,FTEMP|64);
+    rt=get_reg(i_regmap,FTEMP);
+  }else{
+    rth=get_reg(i_regmap,rt1[i]|64);
+    rt=get_reg(i_regmap,rt1[i]);
+  }
+  assert(rs>=0);
+  assert(rt>=0);
+  if(addr<0) addr=rt;
+  assert(addr>=0);
+  int ftable=0;
+  if(type==LOADB_STUB||type==LOADBU_STUB)
+    ftable=(int)readmemb;
+  if(type==LOADH_STUB||type==LOADHU_STUB)
+    ftable=(int)readmemh;
+  if(type==LOADW_STUB)
+    ftable=(int)readmem;
+  if(type==LOADD_STUB)
+    ftable=(int)readmemd;
+  emit_writeword(rs,(int)&address);
+  emit_shrimm(rs,16,addr);
+  emit_movmem_indexedx4(ftable,addr,addr);
+  emit_pusha();
+  ds=i_regs!=&regs[i];
+  int real_rs=(itype[i]==LOADLR)?-1:get_reg(i_regmap,rs1[i]);
+  if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)),i);
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)));
+  
+  int temp;
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    if(addr==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!addr;
+    }
+  }
+  else
+  {
+    temp=!addr;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(stubs[n][6]+1),cc);
+  emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,32);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  emit_callreg(addr);
+  // We really shouldn't need to update the count here,
+  // but not doing so causes random crashes...
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(stubs[n][6]+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  emit_popa();
+  if((cc=get_reg(i_regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+  if(type==LOADB_STUB)
+    emit_movsbl((int)&readmem_dword,rt);
+  if(type==LOADBU_STUB)
+    emit_movzbl((int)&readmem_dword,rt);
+  if(type==LOADH_STUB)
+    emit_movswl((int)&readmem_dword,rt);
+  if(type==LOADHU_STUB)
+    emit_movzwl((int)&readmem_dword,rt);
+  if(type==LOADW_STUB)
+    emit_readword((int)&readmem_dword,rt);
+  if(type==LOADD_STUB) {
+    emit_readword((int)&readmem_dword,rt);
+    if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth);
+  }
+  emit_jmp(stubs[n][2]); // return address
+}
+
+inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+{
+  assem_debug("inline_readstub\n");
+  int rs=get_reg(regmap,target);
+  int rth=get_reg(regmap,target|64);
+  int rt=get_reg(regmap,target);
+  assert(rs>=0);
+  assert(rt>=0);
+  int ftable=0;
+  if(type==LOADB_STUB||type==LOADBU_STUB)
+    ftable=(int)readmemb;
+  if(type==LOADH_STUB||type==LOADHU_STUB)
+    ftable=(int)readmemh;
+  if(type==LOADW_STUB)
+    ftable=(int)readmem;
+  if(type==LOADD_STUB)
+    ftable=(int)readmemd;
+  #ifdef HOST_IMM_ADDR32
+  emit_writeword_imm(addr,(int)&address);
+  #else
+  emit_writeword(rs,(int)&address);
+  #endif
+  emit_pusha();
+  int cc=get_reg(regmap,CCREG);
+  int temp;
+  if(cc<0) {
+    if(rs==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!rs;
+    }
+  }
+  else
+  {
+    temp=!rs;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(adj+1),cc);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  if((signed int)addr>=(signed int)0xC0000000) {
+    // Pagefault address
+    int ds=regmap!=regs[i].regmap;
+    emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,32);
+  }
+  emit_call(((u_int *)ftable)[addr>>16]);
+  // We really shouldn't need to update the count here,
+  // but not doing so causes random crashes...
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(adj+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  emit_popa();
+  if((cc=get_reg(regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+  if(type==LOADB_STUB)
+    emit_movsbl((int)&readmem_dword,rt);
+  if(type==LOADBU_STUB)
+    emit_movzbl((int)&readmem_dword,rt);
+  if(type==LOADH_STUB)
+    emit_movswl((int)&readmem_dword,rt);
+  if(type==LOADHU_STUB)
+    emit_movzwl((int)&readmem_dword,rt);
+  if(type==LOADW_STUB)
+    emit_readword((int)&readmem_dword,rt);
+  if(type==LOADD_STUB) {
+    emit_readword((int)&readmem_dword,rt);
+    if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth);
+  }
+}
+
+do_writestub(int n)
+{
+  assem_debug("do_writestub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int type=stubs[n][0];
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  signed char *i_regmap=i_regs->regmap;
+  int addr=get_reg(i_regmap,AGEN1+(i&1));
+  int rth,rt,r;
+  int ds;
+  if(itype[i]==C1LS) {
+    rth=get_reg(i_regmap,FTEMP|64);
+    rt=get_reg(i_regmap,r=FTEMP);
+  }else{
+    rth=get_reg(i_regmap,rs2[i]|64);
+    rt=get_reg(i_regmap,r=rs2[i]);
+  }
+  assert(rs>=0);
+  assert(rt>=0);
+  if(addr<0) addr=get_reg(i_regmap,-1);
+  assert(addr>=0);
+  int ftable=0;
+  if(type==STOREB_STUB)
+    ftable=(int)writememb;
+  if(type==STOREH_STUB)
+    ftable=(int)writememh;
+  if(type==STOREW_STUB)
+    ftable=(int)writemem;
+  if(type==STORED_STUB)
+    ftable=(int)writememd;
+  emit_writeword(rs,(int)&address);
+  emit_shrimm(rs,16,addr);
+  emit_movmem_indexedx4(ftable,addr,addr);
+  if(type==STOREB_STUB)
+    emit_writebyte(rt,(int)&byte);
+  if(type==STOREH_STUB)
+    emit_writehword(rt,(int)&hword);
+  if(type==STOREW_STUB)
+    emit_writeword(rt,(int)&word);
+  if(type==STORED_STUB) {
+    emit_writeword(rt,(int)&dword);
+    emit_writeword(r?rth:rt,(int)&dword+4);
+  }
+  emit_pusha();
+  ds=i_regs!=&regs[i];
+  int real_rs=get_reg(i_regmap,rs1[i]);
+  if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)),i);
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty&~(1<<addr)&(real_rs<0?-1:~(1<<real_rs)));
+  
+  int temp;
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    if(addr==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!addr;
+    }
+  }
+  else
+  {
+    temp=!addr;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(stubs[n][6]+1),cc);
+  emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,32);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  emit_callreg(addr);
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(stubs[n][6]+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  emit_popa();
+  if((cc=get_reg(i_regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+  emit_jmp(stubs[n][2]); // return address
+}
+
+inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+{
+  assem_debug("inline_writestub\n");
+  int rs=get_reg(regmap,-1);
+  int rth=get_reg(regmap,target|64);
+  int rt=get_reg(regmap,target);
+  assert(rs>=0);
+  assert(rt>=0);
+  int ftable=0;
+  if(type==STOREB_STUB)
+    ftable=(int)writememb;
+  if(type==STOREH_STUB)
+    ftable=(int)writememh;
+  if(type==STOREW_STUB)
+    ftable=(int)writemem;
+  if(type==STORED_STUB)
+    ftable=(int)writememd;
+  emit_writeword(rs,(int)&address);
+  if(type==STOREB_STUB)
+    emit_writebyte(rt,(int)&byte);
+  if(type==STOREH_STUB)
+    emit_writehword(rt,(int)&hword);
+  if(type==STOREW_STUB)
+    emit_writeword(rt,(int)&word);
+  if(type==STORED_STUB) {
+    emit_writeword(rt,(int)&dword);
+    emit_writeword(target?rth:rt,(int)&dword+4);
+  }
+  emit_pusha();
+  int cc=get_reg(regmap,CCREG);
+  int temp;
+  if(cc<0) {
+    if(rs==HOST_CCREG)
+    {
+      cc=0;temp=1;
+      assert(cc!=HOST_CCREG);
+      assert(temp!=HOST_CCREG);
+      emit_loadreg(CCREG,cc);
+    }
+    else
+    {
+      cc=HOST_CCREG;
+      emit_loadreg(CCREG,cc);
+      temp=!rs;
+    }
+  }
+  else
+  {
+    temp=!rs;
+  }
+  emit_readword((int)&last_count,temp);
+  emit_addimm(cc,CLOCK_DIVIDER*(adj+1),cc);
+  emit_add(cc,temp,cc);
+  emit_writeword(cc,(int)&Count);
+  if((signed int)addr>=(signed int)0xC0000000) {
+    // Pagefault address
+    int ds=regmap!=regs[i].regmap;
+    emit_writeword_imm_esp(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,32);
+  }
+  emit_call(((u_int *)ftable)[addr>>16]);
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*(adj+1),HOST_CCREG);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+  emit_storereg(CCREG,HOST_CCREG);
+  emit_popa();
+  if((cc=get_reg(regmap,CCREG))>=0) {
+    emit_loadreg(CCREG,cc);
+  }
+}
+
+do_unalignedwritestub(int n)
+{
+  set_jump_target(stubs[n][1],(int)out);
+  output_byte(0xCC);
+  emit_jmp(stubs[n][2]); // return address
+}
+
+void printregs(int edi,int esi,int ebp,int esp,int b,int d,int c,int a)
+{
+  printf("regs: %x %x %x %x %x %x %x (%x)\n",a,b,c,d,ebp,esi,edi,(&edi)[-1]);
+}
+
+do_invstub(int n)
+{
+  set_jump_target(stubs[n][1],(int)out);
+  if(stubs[n][4]!=EDI) emit_xchg(stubs[n][4],EDI);
+  emit_pusha();
+  emit_call((int)&invalidate_block);
+  emit_popa();
+  if(stubs[n][4]!=EDI) emit_xchg(stubs[n][4],EDI);
+  emit_jmp(stubs[n][2]); // return address
+}
+
+int do_dirty_stub(int i)
+{
+  assem_debug("do_dirty_stub %x\n",start+i*4);
+  emit_pushimm(start+i*4);
+  emit_movimm((int)start<(int)0xC0000000?(int)source:(int)start,EAX);
+  emit_movimm((int)copy,EBX);
+  emit_movimm(slen*4,ECX);
+  emit_call((int)start<(int)0xC0000000?(int)&verify_code:(int)&verify_code_vm);
+  emit_addimm(ESP,4,ESP);
+  int entry=(int)out;
+  load_regs_entry(i);
+  if(entry==(int)out) entry=instr_addr[i];
+  emit_jmp(instr_addr[i]);
+  return entry;
+}
+
+void do_dirty_stub_ds()
+{
+  emit_pushimm(start+1);
+  emit_movimm((int)start<(int)0xC0000000?(int)source:(int)start,EAX);
+  emit_movimm((int)copy,EBX);
+  emit_movimm(slen*4,ECX);
+  emit_call((int)&verify_code_ds);
+  emit_addimm(ESP,4,ESP);
+}
+
+do_cop1stub(int n)
+{
+  assem_debug("do_cop1stub %x\n",start+stubs[n][3]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int i=stubs[n][3];
+  int rs=stubs[n][4];
+  struct regstat *i_regs=(struct regstat *)stubs[n][5];
+  int ds=stubs[n][6];
+  if(!ds) {
+    load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i);
+    //if(i_regs!=&regs[i]) printf("oops: regs[i]=%x i_regs=%x",(int)&regs[i],(int)i_regs);
+  }
+  //else {printf("fp exception in delay slot\n");}
+  wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty);
+  if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+  emit_movimm(start+(i-ds)*4,EAX); // Get PC
+  emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
+  emit_jmp(ds?(int)fp_exception_ds:(int)fp_exception);
+}
+
+/* TLB */
+
+int do_tlb_r(int s,int ar,int map,int x,int a,int shift,int c,u_int addr)
+{
+  if(c) {
+    if((signed int)addr>=(signed int)0xC0000000) {
+      emit_readword((int)(memory_map+(addr>>12)),map);
+    }
+    else
+      return -1; // No mapping
+  }
+  else {
+    if(s!=map) emit_mov(s,map);
+    emit_shrimm(map,12,map);
+    // Schedule this while we wait on the load
+    //if(x) emit_xorimm(addr,x,addr);
+    if(shift>=0) emit_lea8(s,shift);
+    if(~a) emit_andimm(s,a,ar);
+    emit_movmem_indexedx4((int)memory_map,map,map);
+  }
+  return map;
+}
+int do_tlb_r_branch(int map, int c, u_int addr, int *jaddr)
+{
+  if(!c||(signed int)addr>=(signed int)0xC0000000) {
+    emit_test(map,map);
+    *jaddr=(int)out;
+    emit_js(0);
+  }
+  return map;
+}
+
+int gen_tlb_addr_r(int ar, int map) {
+  if(map>=0) {
+    emit_leairrx4(0,ar,map,ar);
+  }
+}
+
+int do_tlb_w(int s,int ar,int map,int x,int c,u_int addr)
+{
+  if(c) {
+    if(addr<0x80800000||addr>=0xC0000000) {
+      emit_readword((int)(memory_map+(addr>>12)),map);
+    }
+    else
+      return -1; // No mapping
+  }
+  else {
+    if(s!=map) emit_mov(s,map);
+    //if(s!=ar) emit_mov(s,ar);
+    emit_shrimm(map,12,map);
+    // Schedule this while we wait on the load
+    //if(x) emit_xorimm(s,x,addr);
+    emit_movmem_indexedx4((int)memory_map,map,map);
+  }
+  emit_shlimm(map,2,map);
+  return map;
+}
+int do_tlb_w_branch(int map, int c, u_int addr, int *jaddr)
+{
+  if(!c||addr<0x80800000||addr>=0xC0000000) {
+    *jaddr=(int)out;
+    emit_jc(0);
+  }
+}
+
+int gen_tlb_addr_w(int ar, int map) {
+  if(map>=0) {
+    emit_leairrx1(0,ar,map,ar);
+  }
+}
+
+// We don't need this for x86
+generate_map_const(u_int addr,int reg) {
+  // void *mapaddr=memory_map+(addr>>12);
+}
+
+/* Special assem */
+
+void shift_assemble_x86(int i,struct regstat *i_regs)
+{
+  if(rt1[i]) {
+    if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV
+    {
+      char s,t,shift;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      s=get_reg(i_regs->regmap,rs1[i]);
+      shift=get_reg(i_regs->regmap,rs2[i]);
+      if(t>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(t);
+        }
+        else if(rs2[i]==0)
+        {
+          assert(s>=0);
+          if(s!=t) emit_mov(s,t);
+        }
+        else
+        {
+          char temp=get_reg(i_regs->regmap,-1);
+          assert(s>=0);
+          if(t==ECX&&s!=ECX) {
+            if(shift!=ECX) emit_mov(shift,ECX);
+            if(rt1[i]==rs2[i]) {shift=temp;}
+            if(s!=shift) emit_mov(s,shift);
+          }
+          else
+          {
+            if(rt1[i]==rs2[i]) {emit_mov(shift,temp);shift=temp;}
+            if(s!=t) emit_mov(s,t);
+            if(shift!=ECX) {
+              if(i_regs->regmap[ECX]<0)
+                emit_mov(shift,ECX);
+              else
+                emit_xchg(shift,ECX);
+            }
+          }
+          if(opcode2[i]==4) // SLLV
+          {
+            emit_shlcl(t==ECX?shift:t);
+          }
+          if(opcode2[i]==6) // SRLV
+          {
+            emit_shrcl(t==ECX?shift:t);
+          }
+          if(opcode2[i]==7) // SRAV
+          {
+            emit_sarcl(t==ECX?shift:t);
+          }
+          if(shift!=ECX&&i_regs->regmap[ECX]>=0) emit_xchg(shift,ECX);
+        }
+      }
+    } else { // DSLLV/DSRLV/DSRAV
+      char sh,sl,th,tl,shift;
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      shift=get_reg(i_regs->regmap,rs2[i]);
+      if(tl>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(tl);
+          if(th>=0) emit_zeroreg(th);
+        }
+        else if(rs2[i]==0)
+        {
+          assert(sl>=0);
+          if(sl!=tl) emit_mov(sl,tl);
+          if(th>=0&&sh!=th) emit_mov(sh,th);
+        }
+        else
+        {
+          // FIXME: What if shift==tl ?
+          assert(shift!=tl);
+          int temp=get_reg(i_regs->regmap,-1);
+          int real_th=th;
+          if(th<0&&opcode2[i]!=0x14) {th=temp;} // DSLLV doesn't need a temporary register
+          assert(sl>=0);
+          assert(sh>=0);
+          if(tl==ECX&&sl!=ECX) {
+            if(shift!=ECX) emit_mov(shift,ECX);
+            if(sl!=shift) emit_mov(sl,shift);
+            if(th>=0 && sh!=th) emit_mov(sh,th);
+          }
+          else if(th==ECX&&sh!=ECX) {
+            if(shift!=ECX) emit_mov(shift,ECX);
+            if(sh!=shift) emit_mov(sh,shift);
+            if(sl!=tl) emit_mov(sl,tl);
+          }
+          else
+          {
+            if(sl!=tl) emit_mov(sl,tl);
+            if(th>=0 && sh!=th) emit_mov(sh,th);
+            if(shift!=ECX) {
+              if(i_regs->regmap[ECX]<0)
+                emit_mov(shift,ECX);
+              else
+                emit_xchg(shift,ECX);
+            }
+          }
+          if(opcode2[i]==0x14) // DSLLV
+          {
+            if(th>=0) emit_shldcl(th==ECX?shift:th,tl==ECX?shift:tl);
+            emit_shlcl(tl==ECX?shift:tl);
+            emit_testimm(ECX,32);
+            if(th>=0) emit_cmovne_reg(tl==ECX?shift:tl,th==ECX?shift:th);
+            emit_cmovne(&const_zero,tl==ECX?shift:tl);
+          }
+          if(opcode2[i]==0x16) // DSRLV
+          {
+            assert(th>=0);
+            emit_shrdcl(tl==ECX?shift:tl,th==ECX?shift:th);
+            emit_shrcl(th==ECX?shift:th);
+            emit_testimm(ECX,32);
+            emit_cmovne_reg(th==ECX?shift:th,tl==ECX?shift:tl);
+            if(real_th>=0) emit_cmovne(&const_zero,th==ECX?shift:th);
+          }
+          if(opcode2[i]==0x17) // DSRAV
+          {
+            assert(th>=0);
+            emit_shrdcl(tl==ECX?shift:tl,th==ECX?shift:th);
+            if(real_th>=0) {
+              assert(temp>=0);
+              emit_mov(th==ECX?shift:th,temp==ECX?shift:temp);
+            }
+            emit_sarcl(th==ECX?shift:th);
+            if(real_th>=0) emit_sarimm(temp==ECX?shift:temp,31,temp==ECX?shift:temp);
+            emit_testimm(ECX,32);
+            emit_cmovne_reg(th==ECX?shift:th,tl==ECX?shift:tl);
+            if(real_th>=0) emit_cmovne_reg(temp==ECX?shift:temp,th==ECX?shift:th);
+          }
+          if(shift!=ECX&&(i_regs->regmap[ECX]>=0||temp==ECX)) emit_xchg(shift,ECX);
+        }
+      }
+    }
+  }
+}
+#define shift_assemble shift_assemble_x86
+
+void loadlr_assemble_x86(int i,struct regstat *i_regs)
+{
+  int s,th,tl,temp,temp2,addr,map=-1;
+  int offset;
+  int jaddr=0;
+  int memtarget,c=0;
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,rt1[i]|64);
+  tl=get_reg(i_regs->regmap,rt1[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,-1);
+  temp2=get_reg(i_regs->regmap,FTEMP);
+  addr=get_reg(i_regs->regmap,AGEN1+(i&1));
+  assert(addr<0);
+  offset=imm[i];
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  reglist|=1<<temp;
+  if(offset||s<0||c) addr=temp2;
+  else addr=s;
+  if(s>=0) {
+    c=(i_regs->wasconst>>s)&1;
+    memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+    if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
+  }
+  if(tl>=0) {
+    //assert(tl>=0);
+    //assert(rt1[i]);
+    if(!using_tlb) {
+      if(!c) {
+        emit_lea8(addr,temp);
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR
+        }else{
+          emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
+        }
+        emit_cmpimm(addr,0x800000);
+        jaddr=(int)out;
+        emit_jno(0);
+      }
+      else {
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+        }else{
+          emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+        }
+      }
+    }else{ // using tlb
+      int a;
+      if(c) {
+        a=-1;
+      }else if (opcode[i]==0x22||opcode[i]==0x26) {
+        a=0xFFFFFFFC; // LWL/LWR
+      }else{
+        a=0xFFFFFFF8; // LDL/LDR
+      }
+      map=get_reg(i_regs->regmap,TLREG);
+      assert(map>=0);
+      map=do_tlb_r(addr,temp2,map,0,a,c?-1:temp,c,constmap[i][s]+offset);
+      if(c) {
+        if (opcode[i]==0x22||opcode[i]==0x26) {
+          emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+        }else{
+          emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+        }
+      }
+      do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
+    }
+    if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR
+      if(!c||memtarget) {
+        //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2);
+        emit_readword_indexed_tlb(0,temp2,map,temp2);
+        if(jaddr) add_stub(LOADW_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist);
+      emit_andimm(temp,24,temp);
+      if (opcode[i]==0x26) emit_xorimm(temp,24,temp); // LWR
+      if(temp==ECX)
+      {
+        int temp3=EDX;
+        if(temp3==temp2) temp3++;
+        emit_pushreg(temp3);
+        emit_movimm(-1,temp3);
+        if (opcode[i]==0x26) {
+          emit_shrcl(temp3);
+          emit_shrcl(temp2);
+        }else{
+          emit_shlcl(temp3);
+          emit_shlcl(temp2);
+        }
+        emit_mov(temp3,ECX);
+        emit_not(ECX,ECX);
+        emit_popreg(temp3);
+      }
+      else
+      {
+        int temp3=EBP;
+        if(temp3==temp) temp3++;
+        if(temp3==temp2) temp3++;
+        if(temp3==temp) temp3++;
+        emit_xchg(ECX,temp);
+        emit_pushreg(temp3);
+        emit_movimm(-1,temp3);
+        if (opcode[i]==0x26) {
+          emit_shrcl(temp3);
+          emit_shrcl(temp2==ECX?temp:temp2);
+        }else{
+          emit_shlcl(temp3);
+          emit_shlcl(temp2==ECX?temp:temp2);
+        }
+        emit_not(temp3,temp3);
+        emit_mov(temp,ECX);
+        emit_mov(temp3,temp);
+        emit_popreg(temp3);
+      }
+      emit_and(temp,tl,tl);
+      emit_or(temp2,tl,tl);
+      //emit_storereg(rt1[i],tl); // DEBUG
+    /*emit_pusha();
+    //save_regs(0x100f);
+        emit_readword((int)&last_count,ECX);
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,HOST_CCREG);
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+    emit_call((int)memdebug);
+    emit_popa();
+    //restore_regs(0x100f);*/
+    }
+    if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR
+      if(s>=0) 
+        if((i_regs->wasdirty>>s)&1)
+          emit_storereg(rs1[i],s);
+      if(get_reg(i_regs->regmap,rs1[i]|64)>=0) 
+        if((i_regs->wasdirty>>get_reg(i_regs->regmap,rs1[i]|64))&1)
+          emit_storereg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64));
+      int temp2h=get_reg(i_regs->regmap,FTEMP|64);
+      if(!c||memtarget) {
+        //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,temp2,temp2h);
+        //emit_readword_indexed((int)rdram-0x7FFFFFFC,temp2,temp2);
+        emit_readdword_indexed_tlb(0,temp2,map,temp2h,temp2);
+        if(jaddr) add_stub(LOADD_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADD_STUB,i,(constmap[i][s]+offset)&0xFFFFFFF8,i_regs->regmap,FTEMP,ccadj[i],reglist);
+      emit_andimm(temp,56,temp);
+      emit_pushreg(temp);
+      emit_pushreg(temp2h);
+      emit_pushreg(temp2);
+      emit_pushreg(th);
+      emit_pushreg(tl);
+      if(opcode[i]==0x1A) emit_call((int)ldl_merge);
+      if(opcode[i]==0x1B) emit_call((int)ldr_merge);
+      emit_addimm(ESP,20,ESP);
+      if(tl!=EDX) {
+        if(tl!=EAX) emit_mov(EAX,tl);
+        if(th!=EDX) emit_mov(EDX,th);
+      } else
+      if(th!=EAX) {
+        if(th!=EDX) emit_mov(EDX,th);
+        if(tl!=EAX) emit_mov(EAX,tl);
+      } else {
+        emit_xchg(EAX,EDX);
+      }
+      if(s>=0) emit_loadreg(rs1[i],s);
+      if(get_reg(i_regs->regmap,rs1[i]|64)>=0)
+        emit_loadreg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64));
+    }
+  }
+}
+#define loadlr_assemble loadlr_assemble_x86
+
+void cop0_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]==0) // MFC0
+  {
+    signed char t=get_reg(i_regs->regmap,rt1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    //assert(t>=0); // Why does this happen?  OOT is weird
+    if(t>=0) {
+      emit_writeword_imm((int)&fake_pc,(int)&PC);
+      emit_writebyte_imm((source[i]>>11)&0x1f,(int)&(fake_pc.f.r.nrd));
+      if(copr==9) {
+        emit_readword((int)&last_count,ECX);
+        emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+      }
+      emit_call((int)MFC0);
+      emit_readword((int)&readmem_dword,t);
+    }
+  }
+  else if(opcode2[i]==4) // MTC0
+  {
+    signed char s=get_reg(i_regs->regmap,rs1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    assert(s>=0);
+    emit_writeword(s,(int)&readmem_dword);
+    emit_pusha();
+    emit_writeword_imm((int)&fake_pc,(int)&PC);
+    emit_writebyte_imm((source[i]>>11)&0x1f,(int)&(fake_pc.f.r.nrd));
+    if(copr==9||copr==11||copr==12) {
+      if(copr==12&&!is_delayslot) {
+        wb_register(rs1[i],i_regs->regmap,i_regs->dirty,i_regs->is32);
+      }
+      emit_readword((int)&last_count,ECX);
+      emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+      emit_add(HOST_CCREG,ECX,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_writeword(HOST_CCREG,(int)&Count);
+    }
+    // What a mess.  The status register (12) can enable interrupts,
+    // so needs a special case to handle a pending interrupt.
+    // The interrupt must be taken immediately, because a subsequent
+    // instruction might disable interrupts again.
+    if(copr==12&&!is_delayslot) {
+      emit_writeword_imm(start+i*4+4,(int)&pcaddr);
+      emit_writebyte_imm(0,(int)&pending_exception);
+    }
+    //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
+    //else
+    emit_call((int)MTC0);
+    if(copr==9||copr==11||copr==12) {
+      emit_readword((int)&Count,HOST_CCREG);
+      emit_readword((int)&next_interupt,ECX);
+      emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+      emit_writeword(ECX,(int)&last_count);
+      emit_storereg(CCREG,HOST_CCREG);
+    }
+    emit_popa();
+    if(copr==12) {
+      assert(!is_delayslot);
+      //if(is_delayslot) output_byte(0xcc);
+      emit_cmpmem_imm_byte((int)&pending_exception,0);
+      emit_jne((int)&do_interrupt);
+    }
+    cop1_usable=0;
+  }
+  else
+  {
+    assert(opcode2[i]==0x10);
+    if((source[i]&0x3f)==0x01) // TLBR
+      emit_call((int)TLBR);
+    if((source[i]&0x3f)==0x02) // TLBWI
+      emit_call((int)TLBWI_new);
+    if((source[i]&0x3f)==0x06) { // TLBWR
+      // The TLB entry written by TLBWR is dependent on the count,
+      // so update the cycle count
+      emit_readword((int)&last_count,ECX);
+      if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+      emit_add(HOST_CCREG,ECX,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG);
+      emit_writeword(HOST_CCREG,(int)&Count);
+      emit_call((int)TLBWR_new);
+    }
+    if((source[i]&0x3f)==0x08) // TLBP
+      emit_call((int)TLBP);
+    if((source[i]&0x3f)==0x18) // ERET
+    {
+      int count=ccadj[i];
+      if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+      emit_addimm_and_set_flags(CLOCK_DIVIDER*count,HOST_CCREG); // TODO: Should there be an extra cycle here?
+      emit_jmp((int)jump_eret);
+    }
+  }
+}
+
+void cop1_assemble(int i,struct regstat *i_regs)
+{
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  if (opcode2[i]==0) { // MFC1
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],tl);
+      emit_readword_indexed(0,tl,tl);
+    }
+  }
+  else if (opcode2[i]==1) { // DMFC1
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    signed char th=get_reg(i_regs->regmap,rt1[i]|64);
+    if(tl>=0) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],tl);
+      if(th>=0) emit_readword_indexed(4,tl,th);
+      emit_readword_indexed(0,tl,tl);
+    }
+  }
+  else if (opcode2[i]==4) { // MTC1
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    signed char temp=get_reg(i_regs->regmap,-1);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_writeword_indexed(sl,0,temp);
+  }
+  else if (opcode2[i]==5) { // DMTC1
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    signed char sh=rs1[i]>0?get_reg(i_regs->regmap,rs1[i]|64):sl;
+    signed char temp=get_reg(i_regs->regmap,-1);
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_writeword_indexed(sh,4,temp);
+    emit_writeword_indexed(sl,0,temp);
+  }
+  else if (opcode2[i]==2) // CFC1
+  {
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0) {
+      u_int copr=(source[i]>>11)&0x1f;
+      if(copr==0) emit_readword((int)&FCR0,tl);
+      if(copr==31) emit_readword((int)&FCR31,tl);
+    }
+  }
+  else if (opcode2[i]==6) // CTC1
+  {
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    u_int copr=(source[i]>>11)&0x1f;
+    assert(sl>=0);
+    if(copr==31)
+    {
+      emit_writeword(sl,(int)&FCR31);
+      // Set the rounding mode
+      char temp=get_reg(i_regs->regmap,-1);
+      emit_movimm(3,temp);
+      emit_and(sl,temp,temp);
+      emit_fldcw_indexed((int)&rounding_modes,temp);
+    }
+  }
+}
+
+void fconv_assemble_x86(int i,struct regstat *i_regs)
+{
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) { // trunc_w_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_movss_load(temp,0);
+    emit_cvttps2dq(0,0); // float->int, truncate
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_movd_store(0,temp);
+    return;
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) { // trunc_w_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_movsd_load(temp,0);
+    emit_cvttpd2dq(0,0); // double->int, truncate
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_movd_store(0,temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x20) { // cvt_s_w
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_fildl(temp);
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fstps(temp);
+    return;
+  }
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) { // cvt_d_w
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_fildl(temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fstpl(temp);
+    return;
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x20) { // cvt_s_l
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fildll(temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fstps(temp);
+    return;
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x21) { // cvt_d_l
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fildll(temp);
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fstpl(temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) { // cvt_d_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fstpl(temp);
+    return;
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) { // cvt_s_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fldl(temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fstps(temp);
+    return;
+  }
+  
+  if(opcode2[i]==0x10) { // cvt_*_s
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp);
+  }
+  if(opcode2[i]==0x11) { // cvt_*_d
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fldl(temp);
+  }
+  if((source[i]&0x3f)<0x10) {
+    emit_fnstcw_stack();
+    if((source[i]&3)==0) emit_fldcw((int)&round_mode); //printf("round\n");
+    if((source[i]&3)==1) emit_fldcw((int)&trunc_mode); //printf("trunc\n");
+    if((source[i]&3)==2) emit_fldcw((int)&ceil_mode); //printf("ceil\n");
+    if((source[i]&3)==3) emit_fldcw((int)&floor_mode); //printf("floor\n");
+  }
+  if((source[i]&0x3f)==0x24||(source[i]&0x3c)==0x0c) { // cvt_w_*
+    if(opcode2[i]!=0x10||((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+    emit_fistpl(temp);
+  }
+  if((source[i]&0x3f)==0x25||(source[i]&0x3c)==0x08) { // cvt_l_*
+    if(opcode2[i]!=0x11||((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f))
+      emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+    emit_fistpll(temp);
+  }
+  if((source[i]&0x3f)<0x10) {
+    emit_fldcw_stack();
+  }
+  return;
+  
+  // C emulation code for debugging
+  
+  emit_pusha();
+  
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x20) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_s_w);
+  }
+  if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_d_w);
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x20) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_s_l);
+  }
+  if(opcode2[i]==0x15&&(source[i]&0x3f)==0x21) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_d_l);
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_d_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x24) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x25) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_l_s);
+  }
+  
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_s_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x24) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x25) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)cvt_l_d);
+  }
+  
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x08) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)round_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x09) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)trunc_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0a) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)ceil_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0b) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)floor_l_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0c) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)round_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)trunc_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0e) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)ceil_w_s);
+  }
+  if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0f) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    emit_call((int)floor_w_s);
+  }
+  
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x08) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)round_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x09) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)trunc_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0a) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)ceil_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0b) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)floor_l_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0c) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)round_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)trunc_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0e) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)ceil_w_d);
+  }
+  if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0f) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    emit_call((int)floor_w_d);
+  }
+  
+  emit_addimm(ESP,8,ESP);
+  emit_popa();
+  //emit_loadreg(CSREG,rs);
+  return;
+}
+#define fconv_assemble fconv_assemble_x86
+
+void fcomp_assemble(int i,struct regstat *i_regs)
+{
+  signed char fs=get_reg(i_regs->regmap,FSREG);
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char cs=get_reg(i_regs->regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  if((source[i]&0x3f)==0x30) {
+    emit_andimm(fs,~0x800000,fs);
+    return;
+  }
+  
+  if((source[i]&0x3e)==0x38) {
+    // sf/ngle - these should throw exceptions for NaNs
+    emit_andimm(fs,~0x800000,fs);
+    return;
+  }
+  
+  if(opcode2[i]==0x10) {
+    emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],temp);
+    emit_flds(temp);
+    emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+    emit_flds(temp);
+    emit_movimm(0x800000,temp);
+    emit_or(fs,temp,fs);
+    emit_xor(temp,fs,temp);
+    emit_fucomip(1);
+    emit_fpop();
+    if((source[i]&0x3f)==0x31) emit_cmovnp_reg(temp,fs); // c_un_s
+    if((source[i]&0x3f)==0x32) {emit_cmovne_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_eq_s
+    if((source[i]&0x3f)==0x33) emit_cmovne_reg(temp,fs); // c_ueq_s
+    if((source[i]&0x3f)==0x34) {emit_cmovnc_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_olt_s
+    if((source[i]&0x3f)==0x35) emit_cmovnc_reg(temp,fs); // c_ult_s
+    if((source[i]&0x3f)==0x36) {emit_cmova_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_ole_s
+    if((source[i]&0x3f)==0x37) emit_cmova_reg(temp,fs); // c_ule_s
+    if((source[i]&0x3f)==0x3a) emit_cmovne_reg(temp,fs); // c_seq_s
+    if((source[i]&0x3f)==0x3b) emit_cmovne_reg(temp,fs); // c_ngl_s
+    if((source[i]&0x3f)==0x3c) emit_cmovnc_reg(temp,fs); // c_lt_s
+    if((source[i]&0x3f)==0x3d) emit_cmovnc_reg(temp,fs); // c_nge_s
+    if((source[i]&0x3f)==0x3e) emit_cmova_reg(temp,fs); // c_le_s
+    if((source[i]&0x3f)==0x3f) emit_cmova_reg(temp,fs); // c_ngt_s
+    return;
+  }
+  if(opcode2[i]==0x11) {
+    emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],temp);
+    emit_fldl(temp);
+    emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+    emit_fldl(temp);
+    emit_movimm(0x800000,temp);
+    emit_or(fs,temp,fs);
+    emit_xor(temp,fs,temp);
+    emit_fucomip(1);
+    emit_fpop();
+    if((source[i]&0x3f)==0x31) emit_cmovnp_reg(temp,fs); // c_un_d
+    if((source[i]&0x3f)==0x32) {emit_cmovne_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_eq_d
+    if((source[i]&0x3f)==0x33) emit_cmovne_reg(temp,fs); // c_ueq_d
+    if((source[i]&0x3f)==0x34) {emit_cmovnc_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_olt_d
+    if((source[i]&0x3f)==0x35) emit_cmovnc_reg(temp,fs); // c_ult_d
+    if((source[i]&0x3f)==0x36) {emit_cmova_reg(temp,fs);emit_cmovp_reg(temp,fs);} // c_ole_d
+    if((source[i]&0x3f)==0x37) emit_cmova_reg(temp,fs); // c_ule_d
+    if((source[i]&0x3f)==0x3a) emit_cmovne_reg(temp,fs); // c_seq_d
+    if((source[i]&0x3f)==0x3b) emit_cmovne_reg(temp,fs); // c_ngl_d
+    if((source[i]&0x3f)==0x3c) emit_cmovnc_reg(temp,fs); // c_lt_d
+    if((source[i]&0x3f)==0x3d) emit_cmovnc_reg(temp,fs); // c_nge_d
+    if((source[i]&0x3f)==0x3e) emit_cmova_reg(temp,fs); // c_le_d
+    if((source[i]&0x3f)==0x3f) emit_cmova_reg(temp,fs); // c_ngt_d
+    return;
+  }
+  
+  emit_pusha();
+  if(opcode2[i]==0x10) {
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>16)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    if((source[i]&0x3f)==0x30) emit_call((int)c_f_s);
+    if((source[i]&0x3f)==0x31) emit_call((int)c_un_s);
+    if((source[i]&0x3f)==0x32) emit_call((int)c_eq_s);
+    if((source[i]&0x3f)==0x33) emit_call((int)c_ueq_s);
+    if((source[i]&0x3f)==0x34) emit_call((int)c_olt_s);
+    if((source[i]&0x3f)==0x35) emit_call((int)c_ult_s);
+    if((source[i]&0x3f)==0x36) emit_call((int)c_ole_s);
+    if((source[i]&0x3f)==0x37) emit_call((int)c_ule_s);
+    if((source[i]&0x3f)==0x38) emit_call((int)c_sf_s);
+    if((source[i]&0x3f)==0x39) emit_call((int)c_ngle_s);
+    if((source[i]&0x3f)==0x3a) emit_call((int)c_seq_s);
+    if((source[i]&0x3f)==0x3b) emit_call((int)c_ngl_s);
+    if((source[i]&0x3f)==0x3c) emit_call((int)c_lt_s);
+    if((source[i]&0x3f)==0x3d) emit_call((int)c_nge_s);
+    if((source[i]&0x3f)==0x3e) emit_call((int)c_le_s);
+    if((source[i]&0x3f)==0x3f) emit_call((int)c_ngt_s);
+  }
+  if(opcode2[i]==0x11) {
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>16)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    if((source[i]&0x3f)==0x30) emit_call((int)c_f_d);
+    if((source[i]&0x3f)==0x31) emit_call((int)c_un_d);
+    if((source[i]&0x3f)==0x32) emit_call((int)c_eq_d);
+    if((source[i]&0x3f)==0x33) emit_call((int)c_ueq_d);
+    if((source[i]&0x3f)==0x34) emit_call((int)c_olt_d);
+    if((source[i]&0x3f)==0x35) emit_call((int)c_ult_d);
+    if((source[i]&0x3f)==0x36) emit_call((int)c_ole_d);
+    if((source[i]&0x3f)==0x37) emit_call((int)c_ule_d);
+    if((source[i]&0x3f)==0x38) emit_call((int)c_sf_d);
+    if((source[i]&0x3f)==0x39) emit_call((int)c_ngle_d);
+    if((source[i]&0x3f)==0x3a) emit_call((int)c_seq_d);
+    if((source[i]&0x3f)==0x3b) emit_call((int)c_ngl_d);
+    if((source[i]&0x3f)==0x3c) emit_call((int)c_lt_d);
+    if((source[i]&0x3f)==0x3d) emit_call((int)c_nge_d);
+    if((source[i]&0x3f)==0x3e) emit_call((int)c_le_d);
+    if((source[i]&0x3f)==0x3f) emit_call((int)c_ngt_d);
+  }
+  emit_addimm(ESP,8,ESP);
+  emit_popa();
+  emit_loadreg(FSREG,fs);
+  return;
+}
+
+void float_assemble(int i,struct regstat *i_regs)
+{
+  signed char temp=get_reg(i_regs->regmap,-1);
+  assert(temp>=0);
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char cs=get_reg(i_regs->regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    int jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  
+  if((source[i]&0x3f)==6) // mov
+  {
+    if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+      if(opcode2[i]==0x10) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+        emit_flds(temp);
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+        emit_fstps(temp);
+      }
+      if(opcode2[i]==0x11) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+        emit_fldl(temp);
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+        emit_fstpl(temp);
+      }
+    }
+    return;
+  }
+  
+  if((source[i]&0x3f)>3)
+  {
+    if(opcode2[i]==0x10) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+      emit_flds(temp);
+      if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+      }
+    }
+    if(opcode2[i]==0x11) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+      emit_fldl(temp);
+      if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+      }
+    }
+    if((source[i]&0x3f)==4) // sqrt
+      emit_fsqrt();
+    if((source[i]&0x3f)==5) // abs
+      emit_fabs();
+    if((source[i]&0x3f)==7) // neg
+      emit_fchs();
+    if(opcode2[i]==0x10) {
+      emit_fstps(temp);
+    }
+    if(opcode2[i]==0x11) {
+      emit_fstpl(temp);
+    }
+    return;
+  }
+  if((source[i]&0x3f)<4)
+  {
+    if(opcode2[i]==0x10) {
+      emit_readword((int)&reg_cop1_simple[(source[i]>>11)&0x1f],temp);
+      emit_flds(temp);
+    }
+    if(opcode2[i]==0x11) {
+      emit_readword((int)&reg_cop1_double[(source[i]>>11)&0x1f],temp);
+      emit_fldl(temp);
+    }
+    if(((source[i]>>11)&0x1f)!=((source[i]>>16)&0x1f)) {
+      if(opcode2[i]==0x10) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],temp);
+        if((source[i]&0x3f)==0) emit_fadds(temp);
+        if((source[i]&0x3f)==1) emit_fsubs(temp);
+        if((source[i]&0x3f)==2) emit_fmuls(temp);
+        if((source[i]&0x3f)==3) emit_fdivs(temp);
+      }
+      else if(opcode2[i]==0x11) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],temp);
+        if((source[i]&0x3f)==0) emit_faddl(temp);
+        if((source[i]&0x3f)==1) emit_fsubl(temp);
+        if((source[i]&0x3f)==2) emit_fmull(temp);
+        if((source[i]&0x3f)==3) emit_fdivl(temp);
+      }
+    }
+    else {
+      if((source[i]&0x3f)==0) emit_fadd(0);
+      if((source[i]&0x3f)==1) emit_fsub(0);
+      if((source[i]&0x3f)==2) emit_fmul(0);
+      if((source[i]&0x3f)==3) emit_fdiv(0);
+    }
+    if(opcode2[i]==0x10) {
+      if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_simple[(source[i]>>6)&0x1f],temp);
+      }
+      emit_fstps(temp);
+    }
+    if(opcode2[i]==0x11) {
+      if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) {
+        emit_readword((int)&reg_cop1_double[(source[i]>>6)&0x1f],temp);
+      }
+      emit_fstpl(temp);
+    }
+    return;
+  }
+  
+  if(opcode2[i]==0x10) { // Single precision
+    emit_pusha();
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>> 6)&0x1f]);
+    if((source[i]&0x3f)<4)
+      emit_pushmem((int)&reg_cop1_simple[(source[i]>>16)&0x1f]);
+    emit_pushmem((int)&reg_cop1_simple[(source[i]>>11)&0x1f]);
+    switch(source[i]&0x3f)
+    {
+      case 0x00: emit_call((int)add_s);break;
+      case 0x01: emit_call((int)sub_s);break;
+      case 0x02: emit_call((int)mul_s);break;
+      case 0x03: emit_call((int)div_s);break;
+      case 0x04: emit_call((int)sqrt_s);break;
+      case 0x05: emit_call((int)abs_s);break;
+      case 0x06: emit_call((int)mov_s);break;
+      case 0x07: emit_call((int)neg_s);break;
+    }
+    emit_addimm(ESP,(source[i]&0x3f)<4?12:8,ESP);
+    emit_popa();
+  }
+  if(opcode2[i]==0x11) { // Double precision
+    emit_pusha();
+    emit_pushmem((int)&reg_cop1_double[(source[i]>> 6)&0x1f]);
+    if((source[i]&0x3f)<4)
+      emit_pushmem((int)&reg_cop1_double[(source[i]>>16)&0x1f]);
+    emit_pushmem((int)&reg_cop1_double[(source[i]>>11)&0x1f]);
+    switch(source[i]&0x3f)
+    {
+      case 0x00: emit_call((int)add_d);break;
+      case 0x01: emit_call((int)sub_d);break;
+      case 0x02: emit_call((int)mul_d);break;
+      case 0x03: emit_call((int)div_d);break;
+      case 0x04: emit_call((int)sqrt_d);break;
+      case 0x05: emit_call((int)abs_d);break;
+      case 0x06: emit_call((int)mov_d);break;
+      case 0x07: emit_call((int)neg_d);break;
+    }
+    emit_addimm(ESP,(source[i]&0x3f)<4?12:8,ESP);
+    emit_popa();
+  }
+}
+
+void multdiv_assemble_x86(int i,struct regstat *i_regs)
+{
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  //  case 0x1C: DMULT
+  //  case 0x1D: DMULTU
+  //  case 0x1E: DDIV
+  //  case 0x1F: DDIVU
+  if(rs1[i]&&rs2[i])
+  {
+    if((opcode2[i]&4)==0) // 32-bit
+    {
+      if(opcode2[i]==0x18) // MULT
+      {
+        char m1=get_reg(i_regs->regmap,rs1[i]);
+        char m2=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1>=0);
+        assert(m2>=0);
+        emit_mov(m1,EAX);
+        emit_imul(m2);
+      }
+      if(opcode2[i]==0x19) // MULTU
+      {
+        char m1=get_reg(i_regs->regmap,rs1[i]);
+        char m2=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1>=0);
+        assert(m2>=0);
+        emit_mov(m1,EAX);
+        emit_mul(m2);
+      }
+      if(opcode2[i]==0x1A) // DIV
+      {
+        char d1=get_reg(i_regs->regmap,rs1[i]);
+        char d2=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1>=0);
+        assert(d2>=0);
+        emit_mov(d1,EAX);
+        emit_cdq();
+        emit_test(d2,d2);
+        emit_jeq((int)out+8);
+        emit_idiv(d2);
+      }
+      if(opcode2[i]==0x1B) // DIVU
+      {
+        char d1=get_reg(i_regs->regmap,rs1[i]);
+        char d2=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1>=0);
+        assert(d2>=0);
+        emit_mov(d1,EAX);
+        emit_zeroreg(EDX);
+        emit_test(d2,d2);
+        emit_jeq((int)out+8);
+        emit_div(d2);
+      }
+    }
+    else // 64-bit
+    {
+      if(opcode2[i]==0x1C) // DMULT
+      {
+        char m1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char m1l=get_reg(i_regs->regmap,rs1[i]);
+        char m2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char m2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        emit_pushreg(m2h);
+        emit_pushreg(m2l);
+        emit_pushreg(m1h);
+        emit_pushreg(m1l);
+        emit_call((int)&mult64);
+        emit_popreg(m1l);
+        emit_popreg(m1h);
+        emit_popreg(m2l);
+        emit_popreg(m2h);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        char loh=get_reg(i_regs->regmap,LOREG|64);
+        char lol=get_reg(i_regs->regmap,LOREG);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1D) // DMULTU
+      {
+        char m1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char m1l=get_reg(i_regs->regmap,rs1[i]);
+        char m2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char m2l=get_reg(i_regs->regmap,rs2[i]);
+        char temp=get_reg(i_regs->regmap,-1);
+        assert(m1h>=0);
+        assert(m2h>=0);
+        assert(m1l>=0);
+        assert(m2l>=0);
+        assert(temp>=0);
+        emit_mov(m1l,EAX);
+        emit_mul(m2l);
+        emit_storereg(LOREG,EAX);
+        emit_mov(EDX,temp);
+        emit_mov(m1h,EAX);
+        emit_mul(m2l);
+        emit_add(EAX,temp,temp);
+        emit_adcimm(0,EDX);
+        emit_storereg(HIREG,EDX);
+        emit_mov(m2h,EAX);
+        emit_mul(m1l);
+        emit_add(EAX,temp,temp);
+        emit_adcimm(0,EDX);
+        emit_storereg(LOREG|64,temp);
+        emit_mov(EDX,temp);
+        emit_mov(m2h,EAX);
+        emit_mul(m1h);
+        emit_add(EAX,temp,EAX);
+        emit_loadreg(HIREG,temp);
+        emit_adcimm(0,EDX);
+        emit_add(EAX,temp,EAX);
+        emit_adcimm(0,EDX);
+        // DEBUG
+        /*
+        emit_pushreg(m2h);
+        emit_pushreg(m2l);
+        emit_pushreg(m1h);
+        emit_pushreg(m1l);
+        emit_call((int)&multu64);
+        emit_popreg(m1l);
+        emit_popreg(m1h);
+        emit_popreg(m2l);
+        emit_popreg(m2h);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);  // DEBUG
+        if(hil>=0) emit_loadreg(HIREG,hil);  // DEBUG
+        */
+        // Shouldn't be necessary
+        //char loh=get_reg(i_regs->regmap,LOREG|64);
+        //char lol=get_reg(i_regs->regmap,LOREG);
+        //if(loh>=0) emit_loadreg(LOREG|64,loh);
+        //if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1E) // DDIV
+      {
+        char d1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char d1l=get_reg(i_regs->regmap,rs1[i]);
+        char d2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char d2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1h>=0);
+        assert(d2h>=0);
+        assert(d1l>=0);
+        assert(d2l>=0);
+        //emit_pushreg(d2h);
+        //emit_pushreg(d2l);
+        //emit_pushreg(d1h);
+        //emit_pushreg(d1l);
+        emit_addimm(ESP,-16,ESP);
+        emit_writeword_indexed(d2h,12,ESP);
+        emit_writeword_indexed(d2l,8,ESP);
+        emit_writeword_indexed(d1h,4,ESP);
+        emit_writeword_indexed(d1l,0,ESP);
+        emit_call((int)&div64);
+        //emit_popreg(d1l);
+        //emit_popreg(d1h);
+        //emit_popreg(d2l);
+        //emit_popreg(d2h);
+        emit_readword_indexed(0,ESP,d1l);
+        emit_readword_indexed(4,ESP,d1h);
+        emit_readword_indexed(8,ESP,d2l);
+        emit_readword_indexed(12,ESP,d2h);
+        emit_addimm(ESP,16,ESP);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        char loh=get_reg(i_regs->regmap,LOREG|64);
+        char lol=get_reg(i_regs->regmap,LOREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+      if(opcode2[i]==0x1F) // DDIVU
+      {
+        char d1h=get_reg(i_regs->regmap,rs1[i]|64);
+        char d1l=get_reg(i_regs->regmap,rs1[i]);
+        char d2h=get_reg(i_regs->regmap,rs2[i]|64);
+        char d2l=get_reg(i_regs->regmap,rs2[i]);
+        assert(d1h>=0);
+        assert(d2h>=0);
+        assert(d1l>=0);
+        assert(d2l>=0);
+        //emit_pushreg(d2h);
+        //emit_pushreg(d2l);
+        //emit_pushreg(d1h);
+        //emit_pushreg(d1l);
+        emit_addimm(ESP,-16,ESP);
+        emit_writeword_indexed(d2h,12,ESP);
+        emit_writeword_indexed(d2l,8,ESP);
+        emit_writeword_indexed(d1h,4,ESP);
+        emit_writeword_indexed(d1l,0,ESP);
+        emit_call((int)&divu64);
+        //emit_popreg(d1l);
+        //emit_popreg(d1h);
+        //emit_popreg(d2l);
+        //emit_popreg(d2h);
+        emit_readword_indexed(0,ESP,d1l);
+        emit_readword_indexed(4,ESP,d1h);
+        emit_readword_indexed(8,ESP,d2l);
+        emit_readword_indexed(12,ESP,d2h);
+        emit_addimm(ESP,16,ESP);
+        char hih=get_reg(i_regs->regmap,HIREG|64);
+        char hil=get_reg(i_regs->regmap,HIREG);
+        char loh=get_reg(i_regs->regmap,LOREG|64);
+        char lol=get_reg(i_regs->regmap,LOREG);
+        if(hih>=0) emit_loadreg(HIREG|64,hih);
+        if(hil>=0) emit_loadreg(HIREG,hil);
+        if(loh>=0) emit_loadreg(LOREG|64,loh);
+        if(lol>=0) emit_loadreg(LOREG,lol);
+      }
+    }
+  }
+  else
+  {
+    // Multiply by zero is zero.
+    // MIPS does not have a divide by zero exception.
+    // The result is undefined, we return zero.
+    char hr=get_reg(i_regs->regmap,HIREG);
+    char lr=get_reg(i_regs->regmap,LOREG);
+    if(hr>=0) emit_zeroreg(hr);
+    if(lr>=0) emit_zeroreg(lr);
+  }
+}
+#define multdiv_assemble multdiv_assemble_x86
+
+void do_preload_rhash(int r) {
+  emit_movimm(0xf8,r);
+}
+
+void do_preload_rhtbl(int r) {
+  // Don't need this for x86
+}
+
+void do_rhash(int rs,int rh) {
+  emit_and(rs,rh,rh);
+}
+
+void do_miniht_load(int ht,int rh) {
+  // Don't need this for x86.  The load and compare can be combined into
+  // a single instruction (below)
+}
+
+void do_miniht_jump(int rs,int rh,int ht) {
+  emit_cmpmem_indexed((int)mini_ht,rh,rs);
+  emit_jne(jump_vaddr_reg[rs]);
+  emit_jmpmem_indexed((int)mini_ht+4,rh);
+}
+
+void do_miniht_insert(int return_address,int rt,int temp) {
+  emit_movimm(return_address,rt); // PC into link register
+  //emit_writeword_imm(return_address,(int)&mini_ht[(return_address&0xFF)>>8][0]);
+  emit_writeword(rt,(int)&mini_ht[(return_address&0xFF)>>3][0]);
+  add_to_linker((int)out,return_address,1);
+  emit_writeword_imm(0,(int)&mini_ht[(return_address&0xFF)>>3][1]);
+}
+
+// We don't need this for x86
+void literal_pool(int n) {}
+void literal_pool_jumpover(int n) {}
+
+// CPU-architecture-specific initialization, not needed for x86
+void arch_init() {}
diff --git a/libpcsxcore/new_dynarec/assem_x86.h b/libpcsxcore/new_dynarec/assem_x86.h
new file mode 100644 (file)
index 0000000..dc34d7f
--- /dev/null
@@ -0,0 +1,19 @@
+#define HOST_REGS 8
+#define HOST_CCREG 6
+#define HOST_BTREG 5
+#define EXCLUDE_REG 4
+
+//#define IMM_PREFETCH 1
+#define HOST_IMM_ADDR32 1
+#define INVERTED_CARRY 1
+#define DESTRUCTIVE_WRITEBACK 1
+#define DESTRUCTIVE_SHIFT 1
+
+#define USE_MINI_HT 1
+
+#define BASE_ADDR 0x70000000 // Code generator target address
+#define TARGET_SIZE_2 25 // 2^25 = 32 megabytes
+
+/* x86 calling convention:
+   caller-save: %eax %ecx %edx
+   callee-save: %ebp %ebx %esi %edi */
diff --git a/libpcsxcore/new_dynarec/fpu.c b/libpcsxcore/new_dynarec/fpu.c
new file mode 100644 (file)
index 0000000..a189a53
--- /dev/null
@@ -0,0 +1,394 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - fpu.c                                                   *
+ *   Copyright (C) 2010 Ari64                                              *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+#include <math.h>
+
+extern int FCR0, FCR31;
+
+void cvt_s_w(int *source,float *dest)
+{
+  *dest = *source;
+}
+void cvt_d_w(int *source,double *dest)
+{
+  *dest = *source;
+}
+void cvt_s_l(long long *source,float *dest)
+{
+  *dest = *source;
+}
+void cvt_d_l(long long *source,double *dest)
+{
+  *dest = *source;
+}
+void cvt_d_s(float *source,double *dest)
+{
+  *dest = *source;
+}
+void cvt_s_d(double *source,float *dest)
+{
+  *dest = *source;
+}
+
+void round_l_s(float *source,long long *dest)
+{
+  *dest = roundf(*source);
+}
+void round_w_s(float *source,int *dest)
+{
+  *dest = roundf(*source);
+}
+void trunc_l_s(float *source,long long *dest)
+{
+  *dest = truncf(*source);
+}
+void trunc_w_s(float *source,int *dest)
+{
+  *dest = truncf(*source);
+}
+void ceil_l_s(float *source,long long *dest)
+{
+  *dest = ceilf(*source);
+}
+void ceil_w_s(float *source,int *dest)
+{
+  *dest = ceilf(*source);
+}
+void floor_l_s(float *source,long long *dest)
+{
+  *dest = floorf(*source);
+}
+void floor_w_s(float *source,int *dest)
+{
+  *dest = floorf(*source);
+}
+
+void round_l_d(double *source,long long *dest)
+{
+  *dest = round(*source);
+}
+void round_w_d(double *source,int *dest)
+{
+  *dest = round(*source);
+}
+void trunc_l_d(double *source,long long *dest)
+{
+  *dest = trunc(*source);
+}
+void trunc_w_d(double *source,int *dest)
+{
+  *dest = trunc(*source);
+}
+void ceil_l_d(double *source,long long *dest)
+{
+  *dest = ceil(*source);
+}
+void ceil_w_d(double *source,int *dest)
+{
+  *dest = ceil(*source);
+}
+void floor_l_d(double *source,long long *dest)
+{
+  *dest = floor(*source);
+}
+void floor_w_d(double *source,int *dest)
+{
+  *dest = floor(*source);
+}
+
+void cvt_w_s(float *source,int *dest)
+{
+  switch(FCR31&3)
+  {
+    case 0: round_w_s(source,dest);return;
+    case 1: trunc_w_s(source,dest);return;
+    case 2: ceil_w_s(source,dest);return;
+    case 3: floor_w_s(source,dest);return;
+  }
+}
+void cvt_w_d(double *source,int *dest)
+{
+  switch(FCR31&3)
+  {
+    case 0: round_w_d(source,dest);return;
+    case 1: trunc_w_d(source,dest);return;
+    case 2: ceil_w_d(source,dest);return;
+    case 3: floor_w_d(source,dest);return;
+  }
+}
+void cvt_l_s(float *source,long long *dest)
+{
+  switch(FCR31&3)
+  {
+    case 0: round_l_s(source,dest);return;
+    case 1: trunc_l_s(source,dest);return;
+    case 2: ceil_l_s(source,dest);return;
+    case 3: floor_l_s(source,dest);return;
+  }
+}
+void cvt_l_d(double *source,long long *dest)
+{
+  switch(FCR31&3)
+  {
+    case 0: round_l_d(source,dest);return;
+    case 1: trunc_l_d(source,dest);return;
+    case 2: ceil_l_d(source,dest);return;
+    case 3: floor_l_d(source,dest);return;
+  }
+}
+
+void c_f_s()
+{
+  FCR31 &= ~0x800000;
+}
+void c_un_s(float *source,float *target)
+{
+  FCR31=(isnan(*source) || isnan(*target)) ? FCR31|0x800000 : FCR31&~0x800000;
+}
+                          
+void c_eq_s(float *source,float *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31&=~0x800000;return;}
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ueq_s(float *source,float *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31|=0x800000;return;}
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_olt_s(float *source,float *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31&=~0x800000;return;}
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ult_s(float *source,float *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31|=0x800000;return;}
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_ole_s(float *source,float *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31&=~0x800000;return;}
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ule_s(float *source,float *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31|=0x800000;return;}
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_sf_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31&=~0x800000;
+}
+void c_ngle_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31&=~0x800000;
+}
+
+void c_seq_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ngl_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_lt_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_nge_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_le_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ngt_s(float *source,float *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_f_d()
+{
+  FCR31 &= ~0x800000;
+}
+void c_un_d(double *source,double *target)
+{
+  FCR31=(isnan(*source) || isnan(*target)) ? FCR31|0x800000 : FCR31&~0x800000;
+}
+                          
+void c_eq_d(double *source,double *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31&=~0x800000;return;}
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ueq_d(double *source,double *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31|=0x800000;return;}
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_olt_d(double *source,double *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31&=~0x800000;return;}
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ult_d(double *source,double *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31|=0x800000;return;}
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_ole_d(double *source,double *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31&=~0x800000;return;}
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ule_d(double *source,double *target)
+{
+  if (isnan(*source) || isnan(*target)) {FCR31|=0x800000;return;}
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_sf_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31&=~0x800000;
+}
+void c_ngle_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31&=~0x800000;
+}
+
+void c_seq_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ngl_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source==*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_lt_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_nge_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+void c_le_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+void c_ngt_d(double *source,double *target)
+{
+  //if (isnan(*source) || isnan(*target)) // FIXME - exception
+  FCR31 = *source<=*target ? FCR31|0x800000 : FCR31&~0x800000;
+}
+
+
+void add_s(float *source1,float *source2,float *target)
+{
+  *target=(*source1)+(*source2);
+}
+void sub_s(float *source1,float *source2,float *target)
+{
+  *target=(*source1)-(*source2);
+}
+void mul_s(float *source1,float *source2,float *target)
+{
+  *target=(*source1)*(*source2);
+}
+void div_s(float *source1,float *source2,float *target)
+{
+  *target=(*source1)/(*source2);
+}
+void sqrt_s(float *source,float *target)
+{
+  *target=sqrtf(*source);
+}
+void abs_s(float *source,float *target)
+{
+  *target=fabsf(*source);
+}
+void mov_s(float *source,float *target)
+{
+  *target=*source;
+}
+void neg_s(float *source,float *target)
+{
+  *target=-(*source);
+}
+void add_d(double *source1,double *source2,double *target)
+{
+  *target=(*source1)+(*source2);
+}
+void sub_d(double *source1,double *source2,double *target)
+{
+  *target=(*source1)-(*source2);
+}
+void mul_d(double *source1,double *source2,double *target)
+{
+  *target=(*source1)*(*source2);
+}
+void div_d(double *source1,double *source2,double *target)
+{
+  *target=(*source1)/(*source2);
+}
+void sqrt_d(double *source,double *target)
+{
+  *target=sqrt(*source);
+}
+void abs_d(double *source,double *target)
+{
+  *target=fabs(*source);
+}
+void mov_d(double *source,double *target)
+{
+  *target=*source;
+}
+void neg_d(double *source,double *target)
+{
+  *target=-(*source);
+}
+
diff --git a/libpcsxcore/new_dynarec/fpu.h b/libpcsxcore/new_dynarec/fpu.h
new file mode 100644 (file)
index 0000000..881ddbe
--- /dev/null
@@ -0,0 +1,74 @@
+void cvt_s_w(int *source,float *dest);
+void cvt_d_w(int *source,double *dest);
+void cvt_s_l(long long *source,float *dest);
+void cvt_d_l(long long *source,double *dest);
+void cvt_w_s(float *source,int *dest);
+void cvt_w_d(double *source,int *dest);
+void cvt_l_s(float *source,long long *dest);
+void cvt_l_d(double *source,long long *dest);
+void cvt_d_s(float *source,double *dest);
+void cvt_s_d(double *source,float *dest);
+void round_l_s(float *source,long long *dest);
+void round_w_s(float *source,int *dest);
+void trunc_l_s(float *source,long long *dest);
+void trunc_w_s(float *source,int *dest);
+void ceil_l_s(float *source,long long *dest);
+void ceil_w_s(float *source,int *dest);
+void floor_l_s(float *source,long long *dest);
+void floor_w_s(float *source,int *dest);
+void round_l_d(double *source,long long *dest);
+void round_w_d(double *source,int *dest);
+void trunc_l_d(double *source,long long *dest);
+void trunc_w_d(double *source,int *dest);
+void ceil_l_d(double *source,long long *dest);
+void ceil_w_d(double *source,int *dest);
+void floor_l_d(double *source,long long *dest);
+void floor_w_d(double *source,int *dest);
+void c_f_s();
+void c_un_s(float *source,float *target);
+void c_eq_s(float *source,float *target);
+void c_ueq_s(float *source,float *target);
+void c_olt_s(float *source,float *target);
+void c_ult_s(float *source,float *target);
+void c_ole_s(float *source,float *target);
+void c_ule_s(float *source,float *target);
+void c_sf_s(float *source,float *target);
+void c_ngle_s(float *source,float *target);
+void c_seq_s(float *source,float *target);
+void c_ngl_s(float *source,float *target);
+void c_lt_s(float *source,float *target);
+void c_nge_s(float *source,float *target);
+void c_le_s(float *source,float *target);
+void c_ngt_s(float *source,float *target);
+void c_f_d();
+void c_un_d(double *source,double *target);
+void c_eq_d(double *source,double *target);
+void c_ueq_d(double *source,double *target);
+void c_olt_d(double *source,double *target);
+void c_ult_d(double *source,double *target);
+void c_ole_d(double *source,double *target);
+void c_ule_d(double *source,double *target);
+void c_sf_d(double *source,double *target);
+void c_ngle_d(double *source,double *target);
+void c_seq_d(double *source,double *target);
+void c_ngl_d(double *source,double *target);
+void c_lt_d(double *source,double *target);
+void c_nge_d(double *source,double *target);
+void c_le_d(double *source,double *target);
+void c_ngt_d(double *source,double *target);
+void add_s(float *source1,float *source2,float *target);
+void sub_s(float *source1,float *source2,float *target);
+void mul_s(float *source1,float *source2,float *target);
+void div_s(float *source1,float *source2,float *target);
+void sqrt_s(float *source,float *target);
+void abs_s(float *source,float *target);
+void mov_s(float *source,float *target);
+void neg_s(float *source,float *target);
+void add_d(double *source1,double *source2,double *target);
+void sub_d(double *source1,double *source2,double *target);
+void mul_d(double *source1,double *source2,double *target);
+void div_d(double *source1,double *source2,double *target);
+void sqrt_d(double *source,double *target);
+void abs_d(double *source,double *target);
+void mov_d(double *source,double *target);
+void neg_d(double *source,double *target);
diff --git a/libpcsxcore/new_dynarec/linkage_arm.s b/libpcsxcore/new_dynarec/linkage_arm.s
new file mode 100644 (file)
index 0000000..f838fcb
--- /dev/null
@@ -0,0 +1,1002 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - linkage_arm.s                                           *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+       .cpu arm9tdmi
+       .fpu softvfp
+       .eabi_attribute 20, 1
+       .eabi_attribute 21, 1
+       .eabi_attribute 23, 3
+       .eabi_attribute 24, 1
+       .eabi_attribute 25, 1
+       .eabi_attribute 26, 2
+       .eabi_attribute 30, 6
+       .eabi_attribute 18, 4
+       .file   "linkage_arm.s"
+       .global rdram
+rdram = 0x80000000
+       .global dynarec_local
+       .global reg
+       .global hi
+       .global lo
+       .global reg_cop1_simple
+       .global reg_cop1_double
+       .global reg_cop0
+       .global FCR0
+       .global FCR31
+       .global rounding_modes
+       .global next_interupt
+       .global cycle_count
+       .global last_count
+       .global pending_exception
+       .global pcaddr
+       .global stop
+       .global invc_ptr
+       .global address
+       .global readmem_dword
+       .global dword
+       .global word
+       .global hword
+       .global byte
+       .global branch_target
+       .global PC
+       .global fake_pc
+       .global mini_ht
+       .global restore_candidate
+       .global memory_map
+       .bss
+       .align  4
+       .type   dynarec_local, %object
+       .size   dynarec_local, 64
+dynarec_local:
+       .space  64+16+16+8+8+8+8+256+8+8+128+128+128+16+8+132+4+256+512+4194304
+next_interupt = dynarec_local + 64
+       .type   next_interupt, %object
+       .size   next_interupt, 4
+cycle_count = next_interupt + 4
+       .type   cycle_count, %object
+       .size   cycle_count, 4
+last_count = cycle_count + 4
+       .type   last_count, %object
+       .size   last_count, 4
+pending_exception = last_count + 4
+       .type   pending_exception, %object
+       .size   pending_exception, 4
+pcaddr = pending_exception + 4
+       .type   pcaddr, %object
+       .size   pcaddr, 4
+stop = pcaddr + 4
+       .type   stop, %object
+       .size   stop, 4
+invc_ptr = stop + 4
+       .type   invc_ptr, %object
+       .size   invc_ptr, 4
+address = invc_ptr + 4
+       .type   address, %object
+       .size   address, 4
+readmem_dword = address + 4
+       .type   readmem_dword, %object
+       .size   readmem_dword, 8
+dword = readmem_dword + 8
+       .type   dword, %object
+       .size   dword, 8
+word = dword + 8
+       .type   word, %object
+       .size   word, 4
+hword = word + 4
+       .type   hword, %object
+       .size   hword, 2
+byte = hword + 2
+       .type   byte, %object
+       .size   byte, 1 /* 1 byte free */
+FCR0 = hword + 4
+       .type   FCR0, %object
+       .size   FCR0, 4
+FCR31 = FCR0 + 4
+       .type   FCR31, %object
+       .size   FCR31, 4
+reg = FCR31 + 4
+       .type   reg, %object
+       .size   reg, 256
+hi = reg + 256
+       .type   hi, %object
+       .size   hi, 8
+lo = hi + 8
+       .type   lo, %object
+       .size   lo, 8
+reg_cop0 = lo + 8
+       .type   reg_cop0, %object
+       .size   reg_cop0, 128
+reg_cop1_simple = reg_cop0 + 128
+       .type   reg_cop1_simple, %object
+       .size   reg_cop1_simple, 128
+reg_cop1_double = reg_cop1_simple + 128
+       .type   reg_cop1_double, %object
+       .size   reg_cop1_double, 128
+rounding_modes = reg_cop1_double + 128
+       .type   rounding_modes, %object
+       .size   rounding_modes, 16
+branch_target = rounding_modes + 16
+       .type   branch_target, %object
+       .size   branch_target, 4
+PC = branch_target + 4
+       .type   PC, %object
+       .size   PC, 4
+fake_pc = PC + 4
+       .type   fake_pc, %object
+       .size   fake_pc, 132
+/* 4 bytes free */
+mini_ht = fake_pc + 136
+       .type   mini_ht, %object
+       .size   mini_ht, 256
+restore_candidate = mini_ht + 256
+       .type   restore_candidate, %object
+       .size   restore_candidate, 512
+memory_map = restore_candidate + 512
+       .type   memory_map, %object
+       .size   memory_map, 4194304
+
+       .text
+       .align  2
+       .global dyna_linker
+       .type   dyna_linker, %function
+dyna_linker:
+       /* r0 = virtual target address */
+       /* r1 = instruction to patch */
+       ldr     r4, .tlbptr
+       lsr     r5, r0, #12
+       mov     r12, r0
+       cmp     r0, #0xC0000000
+       mov     r6, #4096
+       ldrge   r12, [r4, r5, lsl #2]
+       mov     r2, #0x80000
+       ldr     r3, .jiptr
+       tst     r12, r12
+       sub     r6, r6, #1
+       moveq   r12, r0
+       ldr     r7, [r1]
+       eor     r2, r2, r12, lsr #12
+       and     r6, r6, r12, lsr #12
+       cmp     r2, #2048
+       add     r12, r7, #2
+       orrcs   r2, r6, #2048
+       ldr     r5, [r3, r2, lsl #2]
+       lsl     r12, r12, #8
+       /* jump_in lookup */
+.A1:
+       movs    r4, r5
+       beq     .A3
+       ldr     r3, [r5]
+       ldr     r5, [r4, #12]
+       teq     r3, r0
+       bne     .A1
+       ldr     r3, [r4, #4]
+       ldr     r4, [r4, #8]
+       tst     r3, r3
+       bne     .A1
+.A2:
+       mov     r5, r1
+       add     r1, r1, r12, asr #6
+       teq     r1, r4
+       moveq   pc, r4 /* Stale i-cache */
+       bl      add_link
+       sub     r2, r4, r5
+       and     r1, r7, #0xff000000
+       lsl     r2, r2, #6
+       sub     r1, r1, #2
+       add     r1, r1, r2, lsr #8
+       str     r1, [r5]
+       mov     pc, r4
+.A3:
+       /* hash_table lookup */
+       cmp     r2, #2048
+       ldr     r3, .jdptr
+       eor     r4, r0, r0, lsl #16
+       lslcc   r2, r0, #9
+       ldr     r6, .htptr
+       lsr     r4, r4, #12
+       lsrcc   r2, r2, #21
+       bic     r4, r4, #15
+       ldr     r5, [r3, r2, lsl #2]
+       ldr     r7, [r6, r4]!
+       teq     r7, r0
+       ldreq   pc, [r6, #4]
+       ldr     r7, [r6, #8]
+       teq     r7, r0
+       ldreq   pc, [r6, #12]
+       /* jump_dirty lookup */
+.A6:
+       movs    r4, r5
+       beq     .A8
+       ldr     r3, [r5]
+       ldr     r5, [r4, #12]
+       teq     r3, r0
+       bne     .A6
+.A7:
+       ldr     r1, [r4, #8]
+       /* hash_table insert */
+       ldr     r2, [r6]
+       ldr     r3, [r6, #4]
+       str     r0, [r6]
+       str     r1, [r6, #4]
+       str     r2, [r6, #8]
+       str     r3, [r6, #12]
+       mov     pc, r1
+.A8:
+       mov     r4, r0
+       mov     r5, r1
+       bl      new_recompile_block
+       tst     r0, r0
+       mov     r0, r4
+       mov     r1, r5
+       beq     dyna_linker
+       /* pagefault */
+       mov     r1, r0
+       mov     r2, #8
+       .size   dyna_linker, .-dyna_linker
+       .global exec_pagefault
+       .type   exec_pagefault, %function
+exec_pagefault:
+       /* r0 = instruction pointer */
+       /* r1 = fault address */
+       /* r2 = cause */
+       ldr     r3, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       mvn     r6, #0xF000000F
+       ldr     r4, [fp, #reg_cop0+16-dynarec_local] /* Context */
+       bic     r6, r6, #0x0F800000
+       str     r0, [fp, #reg_cop0+56-dynarec_local] /* EPC */
+       orr     r3, r3, #2
+       str     r1, [fp, #reg_cop0+32-dynarec_local] /* BadVAddr */
+       bic     r4, r4, r6
+       str     r3, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       and     r5, r6, r1, lsr #9
+       str     r2, [fp, #reg_cop0+52-dynarec_local] /* Cause */
+       and     r1, r1, r6, lsl #9
+       str     r1, [fp, #reg_cop0+40-dynarec_local] /* EntryHi */
+       orr     r4, r4, r5
+       str     r4, [fp, #reg_cop0+16-dynarec_local] /* Context */
+       mov     r0, #0x80000000
+       bl      get_addr_ht
+       mov     pc, r0
+       .size   exec_pagefault, .-exec_pagefault
+/* Special dynamic linker for the case where a page fault
+   may occur in a branch delay slot */
+       .global dyna_linker_ds
+       .type   dyna_linker_ds, %function
+dyna_linker_ds:
+       /* r0 = virtual target address */
+       /* r1 = instruction to patch */
+       ldr     r4, .tlbptr
+       lsr     r5, r0, #12
+       mov     r12, r0
+       cmp     r0, #0xC0000000
+       mov     r6, #4096
+       ldrge   r12, [r4, r5, lsl #2]
+       mov     r2, #0x80000
+       ldr     r3, .jiptr
+       tst     r12, r12
+       sub     r6, r6, #1
+       moveq   r12, r0
+       ldr     r7, [r1]
+       eor     r2, r2, r12, lsr #12
+       and     r6, r6, r12, lsr #12
+       cmp     r2, #2048
+       add     r12, r7, #2
+       orrcs   r2, r6, #2048
+       ldr     r5, [r3, r2, lsl #2]
+       lsl     r12, r12, #8
+       /* jump_in lookup */
+.B1:
+       movs    r4, r5
+       beq     .B3
+       ldr     r3, [r5]
+       ldr     r5, [r4, #12]
+       teq     r3, r0
+       bne     .B1
+       ldr     r3, [r4, #4]
+       ldr     r4, [r4, #8]
+       tst     r3, r3
+       bne     .B1
+.B2:
+       mov     r5, r1
+       add     r1, r1, r12, asr #6
+       teq     r1, r4
+       moveq   pc, r4 /* Stale i-cache */
+       bl      add_link
+       sub     r2, r4, r5
+       and     r1, r7, #0xff000000
+       lsl     r2, r2, #6
+       sub     r1, r1, #2
+       add     r1, r1, r2, lsr #8
+       str     r1, [r5]
+       mov     pc, r4
+.B3:
+       /* hash_table lookup */
+       cmp     r2, #2048
+       ldr     r3, .jdptr
+       eor     r4, r0, r0, lsl #16
+       lslcc   r2, r0, #9
+       ldr     r6, .htptr
+       lsr     r4, r4, #12
+       lsrcc   r2, r2, #21
+       bic     r4, r4, #15
+       ldr     r5, [r3, r2, lsl #2]
+       ldr     r7, [r6, r4]!
+       teq     r7, r0
+       ldreq   pc, [r6, #4]
+       ldr     r7, [r6, #8]
+       teq     r7, r0
+       ldreq   pc, [r6, #12]
+       /* jump_dirty lookup */
+.B6:
+       movs    r4, r5
+       beq     .B8
+       ldr     r3, [r5]
+       ldr     r5, [r4, #12]
+       teq     r3, r0
+       bne     .B6
+.B7:
+       ldr     r1, [r4, #8]
+       /* hash_table insert */
+       ldr     r2, [r6]
+       ldr     r3, [r6, #4]
+       str     r0, [r6]
+       str     r1, [r6, #4]
+       str     r2, [r6, #8]
+       str     r3, [r6, #12]
+       mov     pc, r1
+.B8:
+       mov     r4, r0
+       bic     r0, r0, #7
+       mov     r5, r1
+       orr     r0, r0, #1
+       bl      new_recompile_block
+       tst     r0, r0
+       mov     r0, r4
+       mov     r1, r5
+       beq     dyna_linker_ds
+       /* pagefault */
+       bic     r1, r0, #7
+       mov     r2, #0x80000008 /* High bit set indicates pagefault in delay slot */
+       sub     r0, r1, #4
+       b       exec_pagefault
+       .size   dyna_linker_ds, .-dyna_linker_ds
+.jiptr:
+       .word   jump_in
+.jdptr:
+       .word   jump_dirty
+.tlbptr:
+       .word   tlb_LUT_r
+.htptr:
+       .word   hash_table
+       .align  2
+       .global jump_vaddr_r0
+       .type   jump_vaddr_r0, %function
+jump_vaddr_r0:
+       eor     r2, r0, r0, lsl #16
+       b       jump_vaddr
+       .size   jump_vaddr_r0, .-jump_vaddr_r0
+       .global jump_vaddr_r1
+       .type   jump_vaddr_r1, %function
+jump_vaddr_r1:
+       eor     r2, r1, r1, lsl #16
+       mov     r0, r1
+       b       jump_vaddr
+       .size   jump_vaddr_r1, .-jump_vaddr_r1
+       .global jump_vaddr_r2
+       .type   jump_vaddr_r2, %function
+jump_vaddr_r2:
+       mov     r0, r2
+       eor     r2, r2, r2, lsl #16
+       b       jump_vaddr
+       .size   jump_vaddr_r2, .-jump_vaddr_r2
+       .global jump_vaddr_r3
+       .type   jump_vaddr_r3, %function
+jump_vaddr_r3:
+       eor     r2, r3, r3, lsl #16
+       mov     r0, r3
+       b       jump_vaddr
+       .size   jump_vaddr_r3, .-jump_vaddr_r3
+       .global jump_vaddr_r4
+       .type   jump_vaddr_r4, %function
+jump_vaddr_r4:
+       eor     r2, r4, r4, lsl #16
+       mov     r0, r4
+       b       jump_vaddr
+       .size   jump_vaddr_r4, .-jump_vaddr_r4
+       .global jump_vaddr_r5
+       .type   jump_vaddr_r5, %function
+jump_vaddr_r5:
+       eor     r2, r5, r5, lsl #16
+       mov     r0, r5
+       b       jump_vaddr
+       .size   jump_vaddr_r5, .-jump_vaddr_r5
+       .global jump_vaddr_r6
+       .type   jump_vaddr_r6, %function
+jump_vaddr_r6:
+       eor     r2, r6, r6, lsl #16
+       mov     r0, r6
+       b       jump_vaddr
+       .size   jump_vaddr_r6, .-jump_vaddr_r6
+       .global jump_vaddr_r8
+       .type   jump_vaddr_r8, %function
+jump_vaddr_r8:
+       eor     r2, r8, r8, lsl #16
+       mov     r0, r8
+       b       jump_vaddr
+       .size   jump_vaddr_r8, .-jump_vaddr_r8
+       .global jump_vaddr_r9
+       .type   jump_vaddr_r9, %function
+jump_vaddr_r9:
+       eor     r2, r9, r9, lsl #16
+       mov     r0, r9
+       b       jump_vaddr
+       .size   jump_vaddr_r9, .-jump_vaddr_r9
+       .global jump_vaddr_r10
+       .type   jump_vaddr_r10, %function
+jump_vaddr_r10:
+       eor     r2, r10, r10, lsl #16
+       mov     r0, r10
+       b       jump_vaddr
+       .size   jump_vaddr_r10, .-jump_vaddr_r10
+       .global jump_vaddr_r12
+       .type   jump_vaddr_r12, %function
+jump_vaddr_r12:
+       eor     r2, r12, r12, lsl #16
+       mov     r0, r12
+       b       jump_vaddr
+       .size   jump_vaddr_r12, .-jump_vaddr_r12
+       .global jump_vaddr_r7
+       .type   jump_vaddr_r7, %function
+jump_vaddr_r7:
+       eor     r2, r7, r7, lsl #16
+       add     r0, r7, #0
+       .size   jump_vaddr_r7, .-jump_vaddr_r7
+       .global jump_vaddr
+       .type   jump_vaddr, %function
+jump_vaddr:
+       ldr     r1, .htptr
+       mvn     r3, #15
+       and     r2, r3, r2, lsr #12
+       ldr     r2, [r1, r2]!
+       teq     r2, r0
+       ldreq   pc, [r1, #4]
+       ldr     r2, [r1, #8]
+       teq     r2, r0
+       ldreq   pc, [r1, #12]
+       str     r10, [fp, #cycle_count-dynarec_local]
+       bl      get_addr
+       ldr     r10, [fp, #cycle_count-dynarec_local]
+       mov     pc, r0
+       .size   jump_vaddr, .-jump_vaddr
+       .align  2
+       .global verify_code_ds
+       .type   verify_code_ds, %function
+verify_code_ds:
+       str     r8, [fp, #branch_target-dynarec_local]
+       .size   verify_code_ds, .-verify_code_ds
+       .global verify_code_vm
+       .type   verify_code_vm, %function
+verify_code_vm:
+       /* r0 = instruction pointer (virtual address) */
+       /* r1 = source (virtual address) */
+       /* r2 = target */
+       /* r3 = length */
+       cmp     r1, #0xC0000000
+       blt     verify_code
+       add     r12, fp, #memory_map-dynarec_local
+       lsr     r4, r1, #12
+       add     r5, r1, r3
+       sub     r5, #1
+       ldr     r6, [r12, r4, lsl #2]
+       lsr     r5, r5, #12
+       movs    r7, r6
+       bmi     .D5
+       add     r1, r1, r6, lsl #2
+       lsl     r6, r6, #2
+.D1:
+       add     r4, r4, #1
+       teq     r6, r7, lsl #2
+       bne     .D5
+       ldr     r7, [r12, r4, lsl #2]
+       cmp     r4, r5
+       bls     .D1
+       .size   verify_code_vm, .-verify_code_vm
+       .global verify_code
+       .type   verify_code, %function
+verify_code:
+       /* r1 = source */
+       /* r2 = target */
+       /* r3 = length */
+       tst     r3, #4
+       mov     r4, #0
+       add     r3, r1, r3
+       mov     r5, #0
+       ldrne   r4, [r1], #4
+       mov     r12, #0
+       ldrne   r5, [r2], #4
+       teq     r1, r3
+       beq     .D3
+.D2:
+       ldr     r7, [r1], #4
+       eor     r9, r4, r5
+       ldr     r8, [r2], #4
+       orrs    r9, r9, r12
+       bne     .D4
+       ldr     r4, [r1], #4
+       eor     r12, r7, r8
+       ldr     r5, [r2], #4
+       cmp     r1, r3
+       bcc     .D2
+       teq     r7, r8
+.D3:
+       teqeq   r4, r5
+.D4:
+       ldr     r8, [fp, #branch_target-dynarec_local]
+       moveq   pc, lr
+.D5:
+       bl      get_addr
+       mov     pc, r0
+       .size   verify_code, .-verify_code
+       .align  2
+       .global cc_interrupt
+       .type   cc_interrupt, %function
+cc_interrupt:
+       ldr     r0, [fp, #last_count-dynarec_local]
+       mov     r1, #0
+       mov     r2, #0x1fc
+       add     r10, r0, r10
+       str     r1, [fp, #pending_exception-dynarec_local]
+       and     r2, r2, r10, lsr #17
+       add     r3, fp, #restore_candidate-dynarec_local
+       str     r10, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       ldr     r4, [r2, r3]
+       mov     r10, lr
+       tst     r4, r4
+       bne     .E4
+.E1:
+       bl      gen_interupt
+       mov     lr, r10
+       ldr     r10, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       ldr     r0, [fp, #next_interupt-dynarec_local]
+       ldr     r1, [fp, #pending_exception-dynarec_local]
+       ldr     r2, [fp, #stop-dynarec_local]
+       str     r0, [fp, #last_count-dynarec_local]
+       sub     r10, r10, r0
+       tst     r2, r2
+       bne     .E3
+       tst     r1, r1
+       moveq   pc, lr
+.E2:
+       ldr     r0, [fp, #pcaddr-dynarec_local]
+       bl      get_addr_ht
+       mov     pc, r0
+.E3:
+       add     r12, fp, #28
+       ldmia   r12, {r4, r5, r6, r7, r8, r9, sl, fp, pc}
+.E4:
+       /* Move 'dirty' blocks to the 'clean' list */
+       lsl     r5, r2, #3
+       str     r1, [r2, r3]
+.E5:
+       lsrs    r4, r4, #1
+       mov     r0, r5
+       add     r5, r5, #1
+       blcs    clean_blocks
+       tst     r5, #31
+       bne     .E5
+       b       .E1
+
+       .size   cc_interrupt, .-cc_interrupt
+       .align  2
+       .global do_interrupt
+       .type   do_interrupt, %function
+do_interrupt:
+       ldr     r0, [fp, #pcaddr-dynarec_local]
+       bl      get_addr_ht
+       ldr     r1, [fp, #next_interupt-dynarec_local]
+       ldr     r10, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       str     r1, [fp, #last_count-dynarec_local]
+       sub     r10, r10, r1
+       add     r10, r10, #2
+       mov     pc, r0
+       .size   do_interrupt, .-do_interrupt
+       .align  2
+       .global fp_exception
+       .type   fp_exception, %function
+fp_exception:
+       mov     r2, #0x10000000
+.E7:
+       ldr     r1, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       mov     r3, #0x80000000
+       str     r0, [fp, #reg_cop0+56-dynarec_local] /* EPC */
+       orr     r1, #2
+       add     r2, r2, #0x2c
+       str     r1, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       str     r2, [fp, #reg_cop0+52-dynarec_local] /* Cause */
+       add     r0, r3, #0x180
+       bl      get_addr_ht
+       mov     pc, r0
+       .size   fp_exception, .-fp_exception
+       .align  2
+       .global fp_exception_ds
+       .type   fp_exception_ds, %function
+fp_exception_ds:
+       mov     r2, #0x90000000 /* Set high bit if delay slot */
+       b       .E7
+       .size   fp_exception_ds, .-fp_exception_ds
+       .align  2
+       .global jump_syscall
+       .type   jump_syscall, %function
+jump_syscall:
+       ldr     r1, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       mov     r3, #0x80000000
+       str     r0, [fp, #reg_cop0+56-dynarec_local] /* EPC */
+       orr     r1, #2
+       mov     r2, #0x20
+       str     r1, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       str     r2, [fp, #reg_cop0+52-dynarec_local] /* Cause */
+       add     r0, r3, #0x180
+       bl      get_addr_ht
+       mov     pc, r0
+       .size   jump_syscall, .-jump_syscall
+       .align  2
+       .global indirect_jump_indexed
+       .type   indirect_jump_indexed, %function
+indirect_jump_indexed:
+       ldr     r0, [r0, r1, lsl #2]
+       .size   indirect_jump_indexed, .-indirect_jump_indexed
+       .align  2
+       .global indirect_jump
+       .type   indirect_jump, %function
+indirect_jump:
+       ldr     r12, [fp, #last_count-dynarec_local]
+       add     r2, r2, r12 
+       str     r2, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       mov     pc, r0
+       .size   indirect_jump, .-indirect_jump
+       .align  2
+       .global jump_eret
+       .type   jump_eret, %function
+jump_eret:
+       ldr     r1, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       ldr     r0, [fp, #last_count-dynarec_local]
+       bic     r1, r1, #2
+       add     r10, r0, r10
+       str     r1, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       str     r10, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       bl      check_interupt
+       ldr     r1, [fp, #next_interupt-dynarec_local]
+       ldr     r0, [fp, #reg_cop0+56-dynarec_local] /* EPC */
+       str     r1, [fp, #last_count-dynarec_local]
+       subs    r10, r10, r1
+       bpl     .E11
+.E8:
+       add     r6, fp, #reg+256-dynarec_local
+       mov     r5, #248
+       mov     r1, #0
+.E9:
+       ldr     r2, [r6, #-8]!
+       ldr     r3, [r6, #4]
+       eor     r3, r3, r2, asr #31
+       subs    r3, r3, #1
+       adc     r1, r1, r1
+       subs    r5, r5, #8
+       bne     .E9
+       ldr     r2, [fp, #hi-dynarec_local]
+       ldr     r3, [fp, #hi+4-dynarec_local]
+       eors    r3, r3, r2, asr #31
+       ldr     r2, [fp, #lo-dynarec_local]
+       ldreq   r3, [fp, #lo+4-dynarec_local]
+       eoreq   r3, r3, r2, asr #31
+       subs    r3, r3, #1
+       adc     r1, r1, r1
+       bl      get_addr_32
+       mov     pc, r0
+.E11:
+       str     r0, [fp, #pcaddr-dynarec_local]
+       bl      cc_interrupt
+       ldr     r0, [fp, #pcaddr-dynarec_local]
+       b       .E8
+       .size   jump_eret, .-jump_eret
+       .align  2
+       .global new_dyna_start
+       .type   new_dyna_start, %function
+new_dyna_start:
+       ldr     r12, .dlptr
+       mov     r0, #0xa4000000
+       stmia   r12, {r4, r5, r6, r7, r8, r9, sl, fp, lr}
+       sub     fp, r12, #28
+       add     r0, r0, #0x40
+       bl      new_recompile_block
+       ldr     r0, [fp, #next_interupt-dynarec_local]
+       ldr     r10, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       str     r0, [fp, #last_count-dynarec_local]
+       sub     r10, r10, r0
+       mov     pc, #0x7000000
+.dlptr:
+       .word   dynarec_local+28
+       .size   new_dyna_start, .-new_dyna_start
+       .align  2
+       .global write_rdram_new
+       .type   write_rdram_new, %function
+write_rdram_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       ldr     r0, [fp, #word-dynarec_local]
+       str     r0, [r2]
+       b       .E12
+       .size   write_rdram_new, .-write_rdram_new
+       .align  2
+       .global write_rdramb_new
+       .type   write_rdramb_new, %function
+write_rdramb_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       ldrb    r0, [fp, #byte-dynarec_local]
+       eor     r2, r2, #3
+       strb    r0, [r2]
+       b       .E12
+       .size   write_rdramb_new, .-write_rdramb_new
+       .align  2
+       .global write_rdramh_new
+       .type   write_rdramh_new, %function
+write_rdramh_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       ldrh    r0, [fp, #hword-dynarec_local]
+       eor     r2, r2, #2
+       strh    r0, [r2]
+       b       .E12
+       .size   write_rdramh_new, .-write_rdramh_new
+       .align  2
+       .global write_rdramd_new
+       .type   write_rdramd_new, %function
+write_rdramd_new:
+       ldr     r2, [fp, #address-dynarec_local]
+/*     ldrd    r0, [fp, #dword-dynarec_local]*/
+       ldr     r0, [fp, #dword-dynarec_local]
+       ldr     r1, [fp, #dword+4-dynarec_local]
+       str     r0, [r2, #4]
+       str     r1, [r2]
+       b       .E12
+       .size   write_rdramd_new, .-write_rdramd_new
+       .align  2
+       .global do_invalidate
+       .type   do_invalidate, %function
+do_invalidate:
+       ldr     r2, [fp, #address-dynarec_local]
+.E12:
+       ldr     r1, [fp, #invc_ptr-dynarec_local]
+       lsr     r0, r2, #12
+       ldrb    r2, [r1, r0]
+       tst     r2, r2
+       beq     invalidate_block
+       mov     pc, lr
+       .size   do_invalidate, .-do_invalidate
+       .align  2
+       .global read_nomem_new
+       .type   read_nomem_new, %function
+read_nomem_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       lsr     r0, r2, #12
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #8
+       tst     r12, r12
+       bmi     tlb_exception
+       ldr     r0, [r2, r12, lsl #2]
+       str     r0, [fp, #readmem_dword-dynarec_local]
+       mov     pc, lr
+       .size   read_nomem_new, .-read_nomem_new
+       .align  2
+       .global read_nomemb_new
+       .type   read_nomemb_new, %function
+read_nomemb_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       lsr     r0, r2, #12
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #8
+       tst     r12, r12
+       bmi     tlb_exception
+       eor     r2, r2, #3
+       ldrb    r0, [r2, r12, lsl #2]
+       str     r0, [fp, #readmem_dword-dynarec_local]
+       mov     pc, lr
+       .size   read_nomemb_new, .-read_nomemb_new
+       .align  2
+       .global read_nomemh_new
+       .type   read_nomemh_new, %function
+read_nomemh_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       lsr     r0, r2, #12
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #8
+       tst     r12, r12
+       bmi     tlb_exception
+       lsl     r12, r12, #2
+       eor     r2, r2, #2
+       ldrh    r0, [r2, r12]
+       str     r0, [fp, #readmem_dword-dynarec_local]
+       mov     pc, lr
+       .size   read_nomemh_new, .-read_nomemh_new
+       .align  2
+       .global read_nomemd_new
+       .type   read_nomemd_new, %function
+read_nomemd_new:
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       lsr     r0, r2, #12
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #8
+       tst     r12, r12
+       bmi     tlb_exception
+       lsl     r12, r12, #2
+/*     ldrd    r0, [r2, r12]*/
+       add     r3, r2, #4
+       ldr     r0, [r2, r12]
+       ldr     r1, [r3, r12]
+       str     r0, [fp, #readmem_dword+4-dynarec_local]
+       str     r1, [fp, #readmem_dword-dynarec_local]
+       mov     pc, lr
+       .size   read_nomemd_new, .-read_nomemd_new
+       .align  2
+       .global write_nomem_new
+       .type   write_nomem_new, %function
+write_nomem_new:
+       str     r3, [fp, #24]
+       str     lr, [fp, #28]
+       bl      do_invalidate
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       ldr     lr, [fp, #28]
+       lsr     r0, r2, #12
+       ldr     r3, [fp, #24]
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #0xc
+       tst     r12, #0x40000000
+       bne     tlb_exception
+       ldr     r0, [fp, #word-dynarec_local]
+       str     r0, [r2, r12, lsl #2]
+       mov     pc, lr
+       .size   write_nomem_new, .-write_nomem_new
+       .align  2
+       .global write_nomemb_new
+       .type   write_nomemb_new, %function
+write_nomemb_new:
+       str     r3, [fp, #24]
+       str     lr, [fp, #28]
+       bl      do_invalidate
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       ldr     lr, [fp, #28]
+       lsr     r0, r2, #12
+       ldr     r3, [fp, #24]
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #0xc
+       tst     r12, #0x40000000
+       bne     tlb_exception
+       eor     r2, r2, #3
+       ldrb    r0, [fp, #byte-dynarec_local]
+       strb    r0, [r2, r12, lsl #2]
+       mov     pc, lr
+       .size   write_nomemb_new, .-write_nomemb_new
+       .align  2
+       .global write_nomemh_new
+       .type   write_nomemh_new, %function
+write_nomemh_new:
+       str     r3, [fp, #24]
+       str     lr, [fp, #28]
+       bl      do_invalidate
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       ldr     lr, [fp, #28]
+       lsr     r0, r2, #12
+       ldr     r3, [fp, #24]
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #0xc
+       lsls    r12, #2
+       bcs     tlb_exception
+       eor     r2, r2, #2
+       ldrh    r0, [fp, #hword-dynarec_local]
+       strh    r0, [r2, r12]
+       mov     pc, lr
+       .size   write_nomemh_new, .-write_nomemh_new
+       .align  2
+       .global write_nomemd_new
+       .type   write_nomemd_new, %function
+write_nomemd_new:
+       str     r3, [fp, #24]
+       str     lr, [fp, #28]
+       bl      do_invalidate
+       ldr     r2, [fp, #address-dynarec_local]
+       add     r12, fp, #memory_map-dynarec_local
+       ldr     lr, [fp, #28]
+       lsr     r0, r2, #12
+       ldr     r3, [fp, #24]
+       ldr     r12, [r12, r0, lsl #2]
+       mov     r1, #0xc
+       lsls    r12, #2
+       bcs     tlb_exception
+       add     r3, r2, #4
+       ldr     r0, [fp, #dword+4-dynarec_local]
+       ldr     r1, [fp, #dword-dynarec_local]
+/*     strd    r0, [r2, r12]*/
+       str     r0, [r2, r12]
+       str     r1, [r3, r12]
+       mov     pc, lr
+       .size   write_nomemd_new, .-write_nomemd_new
+       .align  2
+       .global tlb_exception
+       .type   tlb_exception, %function
+tlb_exception:
+       /* r1 = cause */
+       /* r2 = address */
+       /* r3 = instr addr/flags */
+       ldr     r4, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       add     r5, fp, #memory_map-dynarec_local
+       lsr     r6, r3, #12
+       orr     r1, r1, r3, lsl #31
+       orr     r4, r4, #2
+       ldr     r7, [r5, r6, lsl #2]
+       bic     r8, r3, #3
+       str     r4, [fp, #reg_cop0+48-dynarec_local] /* Status */
+       mov     r6, #0x6000000
+       str     r1, [fp, #reg_cop0+52-dynarec_local] /* Cause */
+       orr     r6, r6, #0x22
+       ldr     r0, [r8, r7, lsl #2]
+       add     r4, r8, r1, asr #29
+       add     r5, fp, #reg-dynarec_local
+       str     r4, [fp, #reg_cop0+56-dynarec_local] /* EPC */
+       mov     r7, #0xf8
+       ldr     r8, [fp, #reg_cop0+16-dynarec_local] /* Context */
+       lsl     r1, r0, #16
+       lsr     r4, r0, #26
+       and     r7, r7, r0, lsr #18
+       mvn     r9, #0xF000000F
+       sub     r2, r2, r1, asr #16
+       bic     r9, r9, #0x0F800000
+       rors    r6, r6, r4
+       mov     r0, #0x80000000
+       ldrcs   r2, [r5, r7]
+       bic     r8, r8, r9
+       tst     r3, #2
+       str     r2, [r5, r7]
+       add     r4, r2, r1, asr #16
+       add     r6, fp, #reg+4-dynarec_local
+       asr     r3, r2, #31
+       str     r4, [fp, #reg_cop0+32-dynarec_local] /* BadVAddr */
+       add     r0, r0, #0x180
+       and     r4, r9, r4, lsr #9
+       strne   r3, [r6, r7]
+       orr     r8, r8, r4
+       str     r8, [fp, #reg_cop0+16-dynarec_local] /* Context */
+       bl      get_addr_ht
+       ldr     r1, [fp, #next_interupt-dynarec_local]
+       ldr     r10, [fp, #reg_cop0+36-dynarec_local] /* Count */
+       str     r1, [fp, #last_count-dynarec_local]
+       sub     r10, r10, r1
+       mov     pc, r0  
+       .size   tlb_exception, .-tlb_exception
+       .align  2
+       .global breakpoint
+       .type   breakpoint, %function
+breakpoint:
+       /* Set breakpoint here for debugging */
+       mov     pc, lr
+       .size   breakpoint, .-breakpoint
+       .section        .note.GNU-stack,"",%progbits
diff --git a/libpcsxcore/new_dynarec/linkage_x86.s b/libpcsxcore/new_dynarec/linkage_x86.s
new file mode 100644 (file)
index 0000000..676c1fe
--- /dev/null
@@ -0,0 +1,819 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - linkage_x86.s                                           *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+       .file   "linkage_x86.s"
+       .bss
+       .align 4
+.globl rdram
+rdram = 0x80000000
+/*rdram:
+       .space  8388608
+       .type   rdram, %object
+       .size   rdram, 8388608
+*/
+       .section        .rodata
+       .text
+.globl dyna_linker
+       .type   dyna_linker, @function
+dyna_linker:
+       /* eax = virtual target address */
+       /* ebx = instruction to patch */
+       mov     %eax, %edi
+       mov     %eax, %ecx
+       shr     $12, %edi
+       cmp     $0xC0000000, %eax
+       cmovge  tlb_LUT_r(,%edi,4), %ecx
+       test    %ecx, %ecx
+       cmovz   %eax, %ecx
+       xor     $0x80000000, %ecx
+       mov     $2047, %edx
+       shr     $12, %ecx
+       and     %ecx, %edx
+       or      $2048, %edx
+       cmp     %edx, %ecx
+       cmova   %edx, %ecx
+       /* jump_in lookup */
+       mov     jump_in(,%ecx,4), %edx
+.A1:
+       test    %edx, %edx
+       je      .A3
+       mov     (%edx), %edi
+       xor     %eax, %edi
+       or      4(%edx), %edi
+       je      .A2
+       movl    12(%edx), %edx
+       jmp     .A1
+.A2:
+       mov     (%ebx), %edi
+       mov     %esi, %ebp
+       lea     4(%ebx,%edi,1), %esi
+       mov     %eax, %edi
+       pusha
+       call    add_link
+       popa
+       mov     8(%edx), %edi
+       mov     %ebp, %esi
+       lea     -4(%edi), %edx
+       subl    %ebx, %edx
+       movl    %edx, (%ebx)
+       jmp     *%edi
+.A3:
+       /* hash_table lookup */
+       mov     %eax, %edi
+       mov     %eax, %edx
+       shr     $16, %edi
+       shr     $12, %edx
+       xor     %eax, %edi
+       and     $2047, %edx
+       movzwl  %di, %edi
+       shl     $4, %edi
+       cmp     $2048, %ecx
+       cmovc   %edx, %ecx
+       cmp     hash_table(%edi), %eax
+       jne     .A5
+.A4:
+       mov     hash_table+4(%edi), %edx
+       jmp     *%edx
+.A5:
+       cmp     hash_table+8(%edi), %eax
+       lea     8(%edi), %edi
+       je      .A4
+       /* jump_dirty lookup */
+       mov     jump_dirty(,%ecx,4), %edx
+.A6:
+       testl   %edx, %edx
+       je      .A8
+       mov     (%edx), %ecx
+       xor     %eax, %ecx
+       or      4(%edx), %ecx
+       je      .A7
+       movl    12(%edx), %edx
+       jmp     .A6
+.A7:
+       mov     8(%edx), %edx
+       /* hash_table insert */
+       mov     hash_table-8(%edi), %ebx
+       mov     hash_table-4(%edi), %ecx
+       mov     %eax, hash_table-8(%edi)
+       mov     %edx, hash_table-4(%edi)
+       mov     %ebx, hash_table(%edi)
+       mov     %ecx, hash_table+4(%edi)
+       jmp     *%edx
+.A8:
+       mov     %eax, %edi
+       pusha
+       call    new_recompile_block
+       test    %eax, %eax
+       popa
+       je      dyna_linker
+       /* pagefault */
+       mov     %eax, %ebx
+       mov     $0x08, %ecx
+       .size   dyna_linker, .-dyna_linker
+
+.globl exec_pagefault
+       .type   exec_pagefault, @function
+exec_pagefault:
+       /* eax = instruction pointer */
+       /* ebx = fault address */
+       /* ecx = cause */
+       mov     reg_cop0+48, %edx
+       mov     reg_cop0+16, %edi
+       or      $2, %edx
+       mov     %ebx, reg_cop0+32 /* BadVAddr */
+       and     $0xFF80000F, %edi
+       mov     %edx, reg_cop0+48 /* Status */
+       mov     %ecx, reg_cop0+52 /* Cause */
+       mov     %eax, reg_cop0+56 /* EPC */
+       mov     %ebx, %ecx
+       shr     $9, %ebx
+       and     $0xFFFFE000, %ecx
+       and     $0x007FFFF0, %ebx
+       mov     %ecx, reg_cop0+40 /* EntryHI */
+       or      %ebx, %edi
+       mov     %edi, reg_cop0+16 /* Context */
+       push    %esi
+       push    $0x80000000
+       call    get_addr_ht
+       pop     %esi
+       pop     %esi
+       jmp     *%eax
+       .size   exec_pagefault, .-exec_pagefault
+
+/* Special dynamic linker for the case where a page fault
+   may occur in a branch delay slot */
+.globl dyna_linker_ds
+       .type   dyna_linker_ds, @function
+dyna_linker_ds:
+       mov     %eax, %edi
+       mov     %eax, %ecx
+       shr     $12, %edi
+       cmp     $0xC0000000, %eax
+       cmovge  tlb_LUT_r(,%edi,4), %ecx
+       test    %ecx, %ecx
+       cmovz   %eax, %ecx
+       xor     $0x80000000, %ecx
+       mov     $2047, %edx
+       shr     $12, %ecx
+       and     %ecx, %edx
+       or      $2048, %edx
+       cmp     %edx, %ecx
+       cmova   %edx, %ecx
+       /* jump_in lookup */
+       mov     jump_in(,%ecx,4), %edx
+.B1:
+       test    %edx, %edx
+       je      .B3
+       mov     (%edx), %edi
+       xor     %eax, %edi
+       or      4(%edx), %edi
+       je      .B2
+       movl    12(%edx), %edx
+       jmp     .B1
+.B2:
+       mov     (%ebx), %edi
+       mov     %esi, %ecx
+       lea     4(%ebx,%edi,1), %esi
+       mov     %eax, %edi
+       pusha
+       call    add_link
+       popa
+       mov     8(%edx), %edi
+       mov     %ecx, %esi
+       lea     -4(%edi), %edx
+       subl    %ebx, %edx
+       movl    %edx, (%ebx)
+       jmp     *%edi
+.B3:
+       /* hash_table lookup */
+       mov     %eax, %edi
+       mov     %eax, %edx
+       shr     $16, %edi
+       shr     $12, %edx
+       xor     %eax, %edi
+       and     $2047, %edx
+       movzwl  %di, %edi
+       shl     $4, %edi
+       cmp     $2048, %ecx
+       cmovc   %edx, %ecx
+       cmp     hash_table(%edi), %eax
+       jne     .B5
+.B4:
+       mov     hash_table+4(%edi), %edx
+       jmp     *%edx
+.B5:
+       cmp     hash_table+8(%edi), %eax
+       lea     8(%edi), %edi
+       je      .B4
+       /* jump_dirty lookup */
+       mov     jump_dirty(,%ecx,4), %edx
+.B6:
+       testl   %edx, %edx
+       je      .B8
+       mov     (%edx), %ecx
+       xor     %eax, %ecx
+       or      4(%edx), %ecx
+       je      .B7
+       movl    12(%edx), %edx
+       jmp     .B6
+.B7:
+       mov     8(%edx), %edx
+       /* hash_table insert */
+       mov     hash_table-8(%edi), %ebx
+       mov     hash_table-4(%edi), %ecx
+       mov     %eax, hash_table-8(%edi)
+       mov     %edx, hash_table-4(%edi)
+       mov     %ebx, hash_table(%edi)
+       mov     %ecx, hash_table+4(%edi)
+       jmp     *%edx
+.B8:
+       mov     %eax, %edi
+       and     $0xFFFFFFF8, %edi
+       inc     %edi
+       pusha
+       call    new_recompile_block
+       test    %eax, %eax
+       popa
+       je      dyna_linker_ds
+       /* pagefault */
+       and     $0xFFFFFFF8, %eax
+       mov     $0x80000008, %ecx /* High bit set indicates pagefault in delay slot */
+       mov     %eax, %ebx
+       sub     $4, %eax
+       jmp     exec_pagefault
+       .size   dyna_linker_ds, .-dyna_linker_ds
+
+.globl jump_vaddr_eax
+       .type   jump_vaddr_eax, @function
+jump_vaddr_eax:
+       mov     %eax, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_eax, .-jump_vaddr_eax
+.globl jump_vaddr_ecx
+       .type   jump_vaddr_ecx, @function
+jump_vaddr_ecx:
+       mov     %ecx, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_ecx, .-jump_vaddr_ecx
+.globl jump_vaddr_edx
+       .type   jump_vaddr_edx, @function
+jump_vaddr_edx:
+       mov     %edx, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_edx, .-jump_vaddr_edx
+.globl jump_vaddr_ebx
+       .type   jump_vaddr_ebx, @function
+jump_vaddr_ebx:
+       mov     %ebx, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_ebx, .-jump_vaddr_ebx
+.globl jump_vaddr_ebp
+       .type   jump_vaddr_ebp, @function
+jump_vaddr_ebp:
+       mov     %ebp, %edi
+       .size   jump_vaddr_ebp, .-jump_vaddr_ebp
+.globl jump_vaddr_edi
+       .type   jump_vaddr_edi, @function
+jump_vaddr_edi:
+       mov     %edi, %eax
+       .size   jump_vaddr_edi, .-jump_vaddr_edi
+
+.globl jump_vaddr
+       .type   jump_vaddr, @function
+jump_vaddr:
+  /* Check hash table */
+       shr     $16, %eax
+       xor     %edi, %eax
+       movzwl  %ax, %eax
+       shl     $4, %eax
+       cmp     hash_table(%eax), %edi
+       jne     .C2
+.C1:
+       mov     hash_table+4(%eax), %edi
+       jmp     *%edi
+.C2:
+       cmp     hash_table+8(%eax), %edi
+       lea     8(%eax), %eax
+       je      .C1
+  /* No hit on hash table, call compiler */
+       push    %edi
+       mov     %esi, cycle_count /* CCREG */
+       call    get_addr
+       mov     cycle_count, %esi
+       add     $4, %esp
+       jmp     *%eax
+       .size   jump_vaddr, .-jump_vaddr
+
+.globl verify_code_ds
+       .type   verify_code_ds, @function
+verify_code_ds:
+       mov     %ebp, branch_target
+       .size   verify_code_ds, .-verify_code_ds
+
+.globl verify_code_vm
+       .type   verify_code_vm, @function
+verify_code_vm:
+       /* eax = source (virtual address) */
+       /* ebx = target */
+       /* ecx = length */
+       cmp     $0xC0000000, %eax
+       jl      verify_code
+       mov     %eax, %edx
+       lea     -1(%eax,%ecx,1), %ebp
+       shr     $12, %edx
+       shr     $12, %ebp
+       mov     memory_map(,%edx,4), %edi
+       test    %edi, %edi
+       js      .D5
+       lea     (%eax,%edi,4), %eax
+.D1:
+       xor     memory_map(,%edx,4), %edi
+       shl     $2, %edi
+       jne     .D5
+       mov     memory_map(,%edx,4), %edi
+       inc     %edx
+       cmp     %ebp, %edx
+       jbe     .D1
+       .size   verify_code_vm, .-verify_code_vm
+
+.globl verify_code
+       .type   verify_code, @function
+verify_code:
+       /* eax = source */
+       /* ebx = target */
+       /* ecx = length */
+       mov     -4(%eax,%ecx,1), %edi
+       xor     -4(%ebx,%ecx,1), %edi
+       jne     .D5
+       mov     %ecx, %edx
+       add     $-4, %ecx
+       je      .D3
+       test    $4, %edx
+       cmove   %edx, %ecx
+       mov     %esi, cycle_count
+.D2:
+       mov     -4(%eax,%ecx,1), %edx
+       mov     -4(%ebx,%ecx,1), %ebp
+       mov     -8(%eax,%ecx,1), %esi
+       xor     %edx, %ebp
+       mov     -8(%ebx,%ecx,1), %edi
+       jne     .D4
+       xor     %esi, %edi
+       jne     .D4
+       add     $-8, %ecx
+       jne     .D2
+       mov     cycle_count, %esi
+       mov     branch_target, %ebp
+.D3:
+       ret
+.D4:
+       mov     cycle_count, %esi
+.D5:
+       mov     branch_target, %ebp
+       add     $4, %esp /* pop return address, we're not returning */
+       call    get_addr
+       add     $4, %esp /* pop virtual address */
+       jmp     *%eax
+       .size   verify_code, .-verify_code
+
+.globl cc_interrupt
+       .type   cc_interrupt, @function
+cc_interrupt:
+       add     last_count, %esi
+       add     $-28, %esp /* Align stack */
+       mov     %esi, reg_cop0+36 /* Count */
+       shr     $19, %esi
+       movl    $0, pending_exception
+       and     $0x7f, %esi
+       cmpl    $0, restore_candidate(,%esi,4)
+       jne     .E4
+.E1:
+       call    gen_interupt
+       mov     reg_cop0+36, %esi
+       mov     next_interupt, %eax
+       mov     pending_exception, %ebx
+       mov     stop, %ecx
+       add     $28, %esp
+       mov     %eax, last_count
+       sub     %eax, %esi
+       test    %ecx, %ecx
+       jne     .E3
+       test    %ebx, %ebx
+       jne     .E2
+       ret
+.E2:
+       mov     pcaddr, %edi
+       mov     %esi, cycle_count /* CCREG */
+       push    %edi
+       call    get_addr_ht
+       mov     cycle_count, %esi
+       add     $8, %esp
+       jmp     *%eax
+.E3:
+       add     $16, %esp /* pop stack */
+       pop     %edi /* restore edi */
+       pop     %esi /* restore esi */
+       pop     %ebx /* restore ebx */
+       pop     %ebp /* restore ebp */
+       ret          /* exit dynarec */
+.E4:
+       /* Move 'dirty' blocks to the 'clean' list */
+       mov     restore_candidate(,%esi,4), %ebx
+       mov     %esi, %ebp
+       movl    $0, restore_candidate(,%esi,4)
+       shl     $5, %ebp
+.E5:
+       shr     $1, %ebx
+       jnc     .E6
+       mov     %ebp, (%esp)
+       call    clean_blocks
+.E6:
+       inc     %ebp
+       test    $31, %ebp
+       jne     .E5
+       jmp     .E1
+       .size   cc_interrupt, .-cc_interrupt
+
+.globl do_interrupt
+       .type   do_interrupt, @function
+do_interrupt:
+       mov     pcaddr, %edi
+       push    %edi
+       call    get_addr_ht
+       add     $4, %esp
+       mov     reg_cop0+36, %esi
+       mov     next_interupt, %ebx
+       mov     %ebx, last_count
+       sub     %ebx, %esi
+       add     $2, %esi
+       jmp     *%eax
+       .size   do_interrupt, .-do_interrupt
+
+.globl fp_exception
+       .type   fp_exception, @function
+fp_exception:
+       mov     $0x1000002c, %edx
+.E7:
+       mov     reg_cop0+48, %ebx
+       or      $2, %ebx
+       mov     %ebx, reg_cop0+48 /* Status */
+       mov     %edx, reg_cop0+52 /* Cause */
+       mov     %eax, reg_cop0+56 /* EPC */
+       push    %esi
+       push    $0x80000180
+       call    get_addr_ht
+       pop     %esi
+       pop     %esi
+       jmp     *%eax
+       .size   fp_exception, .-fp_exception
+
+.globl fp_exception_ds
+       .type   fp_exception_ds, @function
+fp_exception_ds:
+       mov     $0x9000002c, %edx /* Set high bit if delay slot */
+       jmp     .E7
+       .size   fp_exception_ds, .-fp_exception_ds
+
+.globl jump_syscall
+       .type   jump_syscall, @function
+jump_syscall:
+       mov     $0x20, %edx
+       mov     reg_cop0+48, %ebx
+       or      $2, %ebx
+       mov     %ebx, reg_cop0+48 /* Status */
+       mov     %edx, reg_cop0+52 /* Cause */
+       mov     %eax, reg_cop0+56 /* EPC */
+       push    %esi
+       push    $0x80000180
+       call    get_addr_ht
+       pop     %esi
+       pop     %esi
+       jmp     *%eax
+       .size   jump_syscall, .-jump_syscall
+
+.globl jump_eret
+       .type   jump_eret, @function
+jump_eret:
+       mov     reg_cop0+48, %ebx /* Status */
+       add     last_count, %esi
+       and     $0xFFFFFFFD, %ebx
+       mov     %esi, reg_cop0+36 /* Count */
+       mov     %ebx, reg_cop0+48 /* Status */
+       call    check_interupt
+       mov     next_interupt, %eax
+       mov     reg_cop0+36, %esi
+       mov     %eax, last_count
+       sub     %eax, %esi
+       mov     reg_cop0+56, %eax /* EPC */
+       jns     .E11
+.E8:
+       mov     $248, %ebx
+       xor     %edi, %edi
+.E9:
+       mov     reg(%ebx), %ecx
+       mov     reg+4(%ebx), %edx
+       sar     $31, %ecx
+       xor     %ecx, %edx
+       neg     %edx
+       adc     %edi, %edi
+       sub     $8, %ebx
+       jne     .E9
+       mov     hi(%ebx), %ecx
+       mov     hi+4(%ebx), %edx
+       sar     $31, %ecx
+       xor     %ecx, %edx
+       jne     .E10
+       mov     lo(%ebx), %ecx
+       mov     lo+4(%ebx), %edx
+       sar     $31, %ecx
+       xor     %ecx, %edx
+.E10:
+       neg     %edx
+       adc     %edi, %edi
+       push    %edi
+       push    %eax
+       mov     %esi, cycle_count
+       call    get_addr_32
+       mov     cycle_count, %esi
+       add     $8, %esp
+       jmp     *%eax
+.E11:
+       mov     %eax, pcaddr
+       call    cc_interrupt
+       mov     pcaddr, %eax
+       jmp     .E8
+       .size   jump_eret, .-jump_eret
+
+.globl new_dyna_start
+       .type   new_dyna_start, @function
+new_dyna_start:
+       push    %ebp
+       push    %ebx
+       push    %esi
+       push    %edi
+       push    $0xa4000040
+       call    new_recompile_block
+       add     $-8, %esp /* align stack */
+       movl    next_interupt, %edi
+       movl    reg_cop0+36, %esi
+       movl    %edi, last_count
+       subl    %edi, %esi
+       jmp     0x70000000
+       .size   new_dyna_start, .-new_dyna_start
+
+.globl write_rdram_new
+       .type   write_rdram_new, @function
+write_rdram_new:
+       mov     address, %edi
+       mov     word, %ecx
+       mov     %ecx, rdram-0x80000000(%edi)
+       jmp     .E12
+       .size   write_rdram_new, .-write_rdram_new
+
+.globl write_rdramb_new
+       .type   write_rdramb_new, @function
+write_rdramb_new:
+       mov     address, %edi
+       xor     $3, %edi
+       movb    byte, %cl
+       movb    %cl, rdram-0x80000000(%edi)
+       jmp     .E12
+       .size   write_rdramb_new, .-write_rdramb_new
+
+.globl write_rdramh_new
+       .type   write_rdramh_new, @function
+write_rdramh_new:
+       mov     address, %edi
+       xor     $2, %edi
+       movw    hword, %cx
+       movw    %cx, rdram-0x80000000(%edi)
+       jmp     .E12
+       .size   write_rdramh_new, .-write_rdramh_new
+
+.globl write_rdramd_new
+       .type   write_rdramd_new, @function
+write_rdramd_new:
+       mov     address, %edi
+       mov     dword+4, %ecx
+       mov     dword, %edx
+       mov     %ecx, rdram-0x80000000(%edi)
+       mov     %edx, rdram-0x80000000+4(%edi)
+       jmp     .E12
+       .size   write_rdramd_new, .-write_rdramd_new
+
+.globl do_invalidate
+       .type   do_invalidate, @function
+do_invalidate:
+       mov     address, %edi
+       mov     %edi, %ebx /* Return ebx to caller */
+.E12:
+       shr     $12, %edi
+       cmpb    $1, invalid_code(%edi)
+       je      .E13
+       push    %edi
+       call    invalidate_block
+       pop     %edi
+.E13:
+       ret
+       .size   do_invalidate, .-do_invalidate
+
+.globl read_nomem_new
+       .type   read_nomem_new, @function
+read_nomem_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       mov     (%ebx,%edi,4), %ecx
+       mov     %ecx, readmem_dword
+       ret
+       .size   read_nomem_new, .-read_nomem_new
+
+.globl read_nomemb_new
+       .type   read_nomemb_new, @function
+read_nomemb_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       xor     $3, %ebx
+       movzbl  (%ebx,%edi,4), %ecx
+       mov     %ecx, readmem_dword
+       ret
+       .size   read_nomemb_new, .-read_nomemb_new
+
+.globl read_nomemh_new
+       .type   read_nomemh_new, @function
+read_nomemh_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       xor     $2, %ebx
+       movzwl  (%ebx,%edi,4), %ecx
+       mov     %ecx, readmem_dword
+       ret
+       .size   read_nomemh_new, .-read_nomemh_new
+
+.globl read_nomemd_new
+       .type   read_nomemd_new, @function
+read_nomemd_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       mov     4(%ebx,%edi,4), %ecx
+       mov     (%ebx,%edi,4), %edx
+       mov     %ecx, readmem_dword
+       mov     %edx, readmem_dword+4
+       ret
+       .size   read_nomemd_new, .-read_nomemd_new
+
+.globl write_nomem_new
+       .type   write_nomem_new, @function
+write_nomem_new:
+       call    do_invalidate
+       mov     memory_map(,%edi,4),%edi
+       mov     word, %ecx
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       mov     %ecx, (%ebx,%edi)
+       ret
+       .size   write_nomem_new, .-write_nomem_new
+
+.globl write_nomemb_new
+       .type   write_nomemb_new, @function
+write_nomemb_new:
+       call    do_invalidate
+       mov     memory_map(,%edi,4),%edi
+       movb    byte, %cl
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       xor     $3, %ebx
+       movb    %cl, (%ebx,%edi)
+       ret
+       .size   write_nomemb_new, .-write_nomemb_new
+
+.globl write_nomemh_new
+       .type   write_nomemh_new, @function
+write_nomemh_new:
+       call    do_invalidate
+       mov     memory_map(,%edi,4),%edi
+       movw    hword, %cx
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       xor     $2, %ebx
+       movw    %cx, (%ebx,%edi)
+       ret
+       .size   write_nomemh_new, .-write_nomemh_new
+
+.globl write_nomemd_new
+       .type   write_nomemd_new, @function
+write_nomemd_new:
+       call    do_invalidate
+       mov     memory_map(,%edi,4),%edi
+       mov     dword+4, %edx
+       mov     dword, %ecx
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       mov     %edx, (%ebx,%edi)
+       mov     %ecx, 4(%ebx,%edi)
+       ret
+       .size   write_nomemd_new, .-write_nomemd_new
+
+.globl tlb_exception
+       .type   tlb_exception, @function
+tlb_exception:
+       /* eax = cause */
+       /* ebx = address */
+       /* ebp = instr addr + flags */
+       mov     0x24(%esp), %ebp
+/* Debug: 
+       push    %ebp
+       push    %ebx
+       push    %eax
+       call    tlb_debug
+       pop     %eax
+       pop     %ebx
+       pop     %ebp
+/* end debug */
+       mov     reg_cop0+48, %esi
+       mov     %ebp, %ecx
+       mov     %ebp, %edx
+       mov     %ebp, %edi
+       shl     $31, %ebp
+       shr     $12, %ecx
+       or      %ebp, %eax
+       sar     $29, %ebp
+       and     $0xFFFFFFFC, %edx
+       mov     memory_map(,%ecx,4), %ecx
+       or      $2, %esi
+       mov     (%edx, %ecx, 4), %ecx
+       add     %ebp, %edx
+       mov     %esi, reg_cop0+48 /* Status */
+       mov     %eax, reg_cop0+52 /* Cause */
+       mov     %edx, reg_cop0+56 /* EPC */
+       add     $0x24, %esp
+       mov     $0x6000022, %edx
+       mov     %ecx, %ebp
+       movswl  %cx, %eax
+       shr     $26, %ecx
+       shr     $21, %ebp
+       sub     %eax, %ebx
+       and     $0x1f, %ebp
+       ror     %cl, %edx
+       mov     reg_cop0+16, %esi
+       cmovc   reg(,%ebp,8), %ebx
+       and     $0xFF80000F, %esi
+       mov     %ebx, reg(,%ebp,8)
+       add     %ebx, %eax
+       sar     $31, %ebx
+       mov     %eax, reg_cop0+32 /* BadVAddr */
+       shr     $9, %eax
+       test    $2, %edi
+       cmove   reg+4(,%ebp,8), %ebx
+       and     $0x007FFFF0, %eax
+       push    $0x80000180
+       mov     %ebx, reg+4(,%ebp,8)
+       or      %eax, %esi
+       mov     %esi, reg_cop0+16 /* Context */
+       call    get_addr_ht
+       pop     %esi
+       movl    next_interupt, %edi
+       movl    reg_cop0+36, %esi /* Count */
+       movl    %edi, last_count
+       subl    %edi, %esi
+       jmp     *%eax
+       .size   tlb_exception, .-tlb_exception
diff --git a/libpcsxcore/new_dynarec/linkage_x86_64.s b/libpcsxcore/new_dynarec/linkage_x86_64.s
new file mode 100644 (file)
index 0000000..8e35ea4
--- /dev/null
@@ -0,0 +1,794 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - linkage_x86_64.s                                        *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+       .file   "linkage_x86_64.s"
+       .bss
+       .align 4
+//.globl rdram
+//rdram = 0x80000000
+       .section        .rodata
+       .text
+.globl dyna_linker
+       .type   dyna_linker, @function
+dyna_linker:
+       /* eax = virtual target address */
+       /* ebx = instruction to patch */
+       mov     %eax, %edi
+       mov     %eax, %ecx
+       shr     $12, %edi
+       cmp     $0xC0000000, %eax
+       cmovge  tlb_LUT_r(,%edi,4), %ecx
+       test    %ecx, %ecx
+       cmovz   %eax, %ecx
+       xor     $0x80000000, %ecx
+       mov     $2047, %edx
+       shr     $12, %ecx
+       and     %ecx, %edx
+       or      $2048, %edx
+       cmp     %edx, %ecx
+       cmova   %edx, %ecx
+       /* jump_in lookup */
+       movq    jump_in(,%ecx,8), %r12
+.A1:
+       test    %r12, %r12
+       je      .A3
+       mov     (%r12), %edi
+       xor     %eax, %edi
+       or      4(%r12), %edi
+       je      .A2
+       movq    16(%r12), %r12
+       jmp     .A1
+.A2:
+       mov     (%ebx), %edi
+       mov     %esi, %ebp
+       lea     4(%ebx,%edi,1), %esi
+       mov     %eax, %edi
+       call    add_link
+       mov     8(%r12), %edi
+       mov     %ebp, %esi
+       lea     -4(%edi), %edx
+       subl    %ebx, %edx
+       movl    %edx, (%ebx)
+       jmp     *%rdi
+.A3:
+       /* hash_table lookup */
+       mov     %eax, %edi
+       mov     %eax, %edx
+       shr     $16, %edi
+       shr     $12, %edx
+       xor     %eax, %edi
+       and     $2047, %edx
+       movzwl  %di, %edi
+       shl     $4, %edi
+       cmp     $2048, %ecx
+       cmovc   %edx, %ecx
+       cmp     hash_table(%edi), %eax
+       jne     .A5
+.A4:
+       mov     hash_table+4(%edi), %edx
+       jmp     *%rdx
+.A5:
+       cmp     hash_table+8(%edi), %eax
+       lea     8(%edi), %edi
+       je      .A4
+       /* jump_dirty lookup */
+       movq    jump_dirty(,%ecx,8), %r12
+.A6:
+       test    %r12, %r12
+       je      .A8
+       mov     (%r12), %ecx
+       xor     %eax, %ecx
+       or      4(%r12), %ecx
+       je      .A7
+       movq    16(%r12), %r12
+       jmp     .A6
+.A7:
+       movl    8(%r12), %edx
+       /* hash_table insert */
+       mov     hash_table-8(%edi), %ebx
+       mov     hash_table-4(%edi), %ecx
+       mov     %eax, hash_table-8(%edi)
+       mov     %edx, hash_table-4(%edi)
+       mov     %ebx, hash_table(%edi)
+       mov     %ecx, hash_table+4(%edi)
+       jmp     *%rdx
+.A8:
+       mov     %eax, %edi
+       mov     %eax, %ebp /* Note: assumes %rbx and %rbp are callee-saved */
+       mov     %esi, %r12d
+       call    new_recompile_block
+       test    %eax, %eax
+       mov     %ebp, %eax
+       mov     %r12d, %esi
+       je      dyna_linker
+       /* pagefault */
+       mov     %eax, %ebx
+       mov     $0x08, %ecx
+       .size   dyna_linker, .-dyna_linker
+
+.globl exec_pagefault
+       .type   exec_pagefault, @function
+exec_pagefault:
+       /* eax = instruction pointer */
+       /* ebx = fault address */
+       /* ecx = cause */
+       mov     reg_cop0+48, %edx
+       mov     reg_cop0+16, %edi
+       or      $2, %edx
+       mov     %ebx, reg_cop0+32 /* BadVAddr */
+       and     $0xFF80000F, %edi
+       mov     %edx, reg_cop0+48 /* Status */
+       mov     %ecx, reg_cop0+52 /* Cause */
+       mov     %eax, reg_cop0+56 /* EPC */
+       mov     %ebx, %ecx
+       shr     $9, %ebx
+       and     $0xFFFFE000, %ecx
+       and     $0x007FFFF0, %ebx
+       mov     %ecx, reg_cop0+40 /* EntryHI */
+       or      %ebx, %edi
+       mov     %edi, reg_cop0+16 /* Context */
+       mov     %esi, %ebx
+       mov     $0x80000000, %edi
+       call    get_addr_ht
+       mov     %ebx, %esi
+       jmp     *%rax
+       .size   exec_pagefault, .-exec_pagefault
+
+/* Special dynamic linker for the case where a page fault
+   may occur in a branch delay slot */
+.globl dyna_linker_ds
+       .type   dyna_linker_ds, @function
+dyna_linker_ds:
+       mov     %eax, %edi
+       mov     %eax, %ecx
+       shr     $12, %edi
+       cmp     $0xC0000000, %eax
+       cmovge  tlb_LUT_r(,%edi,4), %ecx
+       test    %ecx, %ecx
+       cmovz   %eax, %ecx
+       xor     $0x80000000, %ecx
+       mov     $2047, %edx
+       shr     $12, %ecx
+       and     %ecx, %edx
+       or      $2048, %edx
+       cmp     %edx, %ecx
+       cmova   %edx, %ecx
+       /* jump_in lookup */
+       movq    jump_in(,%ecx,8), %r12
+.B1:
+       test    %r12, %r12
+       je      .B3
+       mov     (%r12), %edi
+       xor     %eax, %edi
+       or      4(%r12), %edi
+       je      .B2
+       movq    16(%r12), %r12
+       jmp     .B1
+.B2:
+       mov     (%ebx), %edi
+       mov     %esi, %r13d
+       lea     4(%ebx,%edi,1), %esi
+       mov     %eax, %edi
+       call    add_link
+       mov     8(%r12), %edi
+       mov     %r13d, %esi
+       lea     -4(%edi), %edx
+       subl    %ebx, %edx
+       movl    %edx, (%ebx)
+       jmp     *%rdi
+.B3:
+       /* hash_table lookup */
+       mov     %eax, %edi
+       mov     %eax, %edx
+       shr     $16, %edi
+       shr     $12, %edx
+       xor     %eax, %edi
+       and     $2047, %edx
+       movzwl  %di, %edi
+       shl     $4, %edi
+       cmp     $2048, %ecx
+       cmovc   %edx, %ecx
+       cmp     hash_table(%edi), %eax
+       jne     .B5
+.B4:
+       mov     hash_table+4(%edi), %edx
+       jmp     *%rdx
+.B5:
+       cmp     hash_table+8(%edi), %eax
+       lea     8(%edi), %edi
+       je      .B4
+       /* jump_dirty lookup */
+       movq    jump_dirty(,%ecx,8), %r12
+.B6:
+       test    %r12, %r12
+       je      .B8
+       mov     (%r12), %ecx
+       xor     %eax, %ecx
+       or      4(%r12), %ecx
+       je      .B7
+       movq    16(%r12), %r12
+       jmp     .B6
+.B7:
+       movl    8(%r12), %edx
+       /* hash_table insert */
+       mov     hash_table-8(%edi), %ebx
+       mov     hash_table-4(%edi), %ecx
+       mov     %eax, hash_table-8(%edi)
+       mov     %edx, hash_table-4(%edi)
+       mov     %ebx, hash_table(%edi)
+       mov     %ecx, hash_table+4(%edi)
+       jmp     *%rdx
+.B8:
+       mov     %eax, %edi
+       mov     %eax, %r12d /* Note: assumes %rbx and %rbp are callee-saved */
+       and     $0xFFFFFFF8, %edi
+       mov     %esi, %r13d
+       inc     %edi
+       call    new_recompile_block
+       test    %eax, %eax
+       mov     %r12d, %eax
+       mov     %r13d, %esi
+       je      dyna_linker_ds
+       /* pagefault */
+       and     $0xFFFFFFF8, %eax
+       mov     $0x80000008, %ecx /* High bit set indicates pagefault in delay slot */
+       mov     %eax, %ebx
+       sub     $4, %eax
+       jmp     exec_pagefault
+       .size   dyna_linker_ds, .-dyna_linker_ds
+
+.globl jump_vaddr_eax
+       .type   jump_vaddr_eax, @function
+jump_vaddr_eax:
+       mov     %eax, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_eax, .-jump_vaddr_eax
+.globl jump_vaddr_ecx
+       .type   jump_vaddr_ecx, @function
+jump_vaddr_ecx:
+       mov     %ecx, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_ecx, .-jump_vaddr_ecx
+.globl jump_vaddr_edx
+       .type   jump_vaddr_edx, @function
+jump_vaddr_edx:
+       mov     %edx, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_edx, .-jump_vaddr_edx
+.globl jump_vaddr_ebx
+       .type   jump_vaddr_ebx, @function
+jump_vaddr_ebx:
+       mov     %ebx, %edi
+       jmp     jump_vaddr_edi
+       .size   jump_vaddr_ebx, .-jump_vaddr_ebx
+.globl jump_vaddr_ebp
+       .type   jump_vaddr_ebp, @function
+jump_vaddr_ebp:
+       mov     %ebp, %edi
+       .size   jump_vaddr_ebp, .-jump_vaddr_ebp
+.globl jump_vaddr_edi
+       .type   jump_vaddr_edi, @function
+jump_vaddr_edi:
+       mov     %edi, %eax
+       .size   jump_vaddr_edi, .-jump_vaddr_edi
+
+.globl jump_vaddr
+       .type   jump_vaddr, @function
+jump_vaddr:
+  /* Check hash table */
+       shr     $16, %eax
+       xor     %edi, %eax
+       movzwl  %ax, %eax
+       shl     $4, %eax
+       cmp     hash_table(%eax), %edi
+       jne     .C2
+.C1:
+       mov     hash_table+4(%eax), %edi
+       jmp     *%rdi
+.C2:
+       cmp     hash_table+8(%eax), %edi
+       lea     8(%eax), %eax
+       je      .C1
+  /* No hit on hash table, call compiler */
+       mov     %esi, cycle_count /* CCREG */
+       call    get_addr
+       mov     cycle_count, %esi
+       jmp     *%rax
+       .size   jump_vaddr, .-jump_vaddr
+
+.globl verify_code_ds
+       .type   verify_code_ds, @function
+verify_code_ds:
+       nop
+       .size   verify_code_ds, .-verify_code_ds
+
+.globl verify_code_vm
+       .type   verify_code_vm, @function
+verify_code_vm:
+       /* eax = source (virtual address) */
+       /* ebx = target */
+       /* ecx = length */
+       cmp     $0xC0000000, %eax
+       jl      verify_code
+       mov     %eax, %edx
+       lea     -1(%eax,%ecx,1), %r9d
+       shr     $12, %edx
+       shr     $12, %r9d
+       mov     memory_map(,%edx,4), %edi
+       test    %edi, %edi
+       js      .D4
+       lea     (%eax,%edi,4), %eax
+       mov     %edi, %r8d
+.D1:
+       xor     memory_map(,%edx,4), %edi
+       shl     $2, %edi
+       jne     .D4
+       mov     %r8d, %edi
+       inc     %edx
+       cmp     %r9d, %edx
+       jbe     .D1
+       .size   verify_code_vm, .-verify_code_vm
+
+.globl verify_code
+       .type   verify_code, @function
+verify_code:
+       /* eax = source */
+       /* ebx = target */
+       /* ecx = length */
+       /* r12d = instruction pointer */
+       mov     -4(%eax,%ecx,1), %edi
+       xor     -4(%ebx,%ecx,1), %edi
+       jne     .D4
+       mov     %ecx, %edx
+       add     $-4, %ecx
+       je      .D3
+       test    $4, %edx
+       cmove   %edx, %ecx
+.D2:
+       mov     -8(%eax,%ecx,1), %rdi
+       cmp     -8(%ebx,%ecx,1), %rdi
+       jne     .D4
+       add     $-8, %ecx
+       jne     .D2
+.D3:
+       ret
+.D4:
+       add     $8, %rsp /* pop return address, we're not returning */
+       mov     %r12d, %edi
+       mov     %esi, %ebx
+       call    get_addr
+       mov     %ebx, %esi
+       jmp     *%rax
+       .size   verify_code, .-verify_code
+
+.globl cc_interrupt
+       .type   cc_interrupt, @function
+cc_interrupt:
+       add     last_count, %esi
+       add     $-8, %rsp /* Align stack */
+       mov     %esi, reg_cop0+36 /* Count */
+       shr     $19, %esi
+       movl    $0, pending_exception
+       and     $0x7f, %esi
+       cmpl    $0, restore_candidate(,%esi,4)
+       jne     .E4
+.E1:
+       call    gen_interupt
+       mov     reg_cop0+36, %esi
+       mov     next_interupt, %eax
+       mov     pending_exception, %ebx
+       mov     stop, %ecx
+       add     $8, %rsp
+       mov     %eax, last_count
+       sub     %eax, %esi
+       test    %ecx, %ecx
+       jne     .E3
+       test    %ebx, %ebx
+       jne     .E2
+       ret
+.E2:
+       mov     pcaddr, %edi
+       mov     %esi, cycle_count /* CCREG */
+       call    get_addr_ht
+       mov     cycle_count, %esi
+       add     $8, %rsp /* pop return address */
+       jmp     *%rax
+.E3:
+       pop     %rbp /* pop return address and discard it */
+       pop     %rbp /* pop junk */
+       pop     %r15 /* restore callee-save registers */
+       pop     %r14
+       pop     %r13
+       pop     %r12
+       pop     %rbx
+       pop     %rbp
+       ret          /* exit dynarec */
+.E4:
+       /* Move 'dirty' blocks to the 'clean' list */
+       mov     restore_candidate(,%esi,4), %ebx
+       mov     %esi, %ebp
+       movl    $0, restore_candidate(,%esi,4)
+       shl     $5, %ebp
+.E5:
+       shr     $1, %ebx
+       jnc     .E6
+       mov     %ebp, %edi
+       call    clean_blocks
+.E6:
+       inc     %ebp
+       test    $31, %ebp
+       jne     .E5
+       jmp     .E1
+       .size   cc_interrupt, .-cc_interrupt
+
+.globl do_interrupt
+       .type   do_interrupt, @function
+do_interrupt:
+       mov     pcaddr, %edi
+       call    get_addr_ht
+       mov     reg_cop0+36, %esi
+       mov     next_interupt, %ebx
+       mov     %ebx, last_count
+       sub     %ebx, %esi
+       add     $2, %esi
+       jmp     *%rax
+       .size   do_interrupt, .-do_interrupt
+
+.globl fp_exception
+       .type   fp_exception, @function
+fp_exception:
+       mov     $0x1000002c, %edx
+.E7:
+       mov     reg_cop0+48, %ebx
+       or      $2, %ebx
+       mov     %ebx, reg_cop0+48 /* Status */
+       mov     %edx, reg_cop0+52 /* Cause */
+       mov     %eax, reg_cop0+56 /* EPC */
+       mov     %esi, %ebx
+       mov     $0x80000180, %edi
+       call    get_addr_ht
+       mov     %ebx, %esi
+       jmp     *%rax
+       .size   fp_exception, .-fp_exception
+
+.globl fp_exception_ds
+       .type   fp_exception_ds, @function
+fp_exception_ds:
+       mov     $0x9000002c, %edx /* Set high bit if delay slot */
+       jmp     .E7
+       .size   fp_exception_ds, .-fp_exception_ds
+
+.globl jump_syscall
+       .type   jump_syscall, @function
+jump_syscall:
+       mov     $0x20, %edx
+       mov     reg_cop0+48, %ebx
+       or      $2, %ebx
+       mov     %ebx, reg_cop0+48 /* Status */
+       mov     %edx, reg_cop0+52 /* Cause */
+       mov     %eax, reg_cop0+56 /* EPC */
+       mov     %esi, %ebx
+       mov     $0x80000180, %edi
+       call    get_addr_ht
+       mov     %ebx, %esi
+       jmp     *%rax
+       .size   jump_syscall, .-jump_syscall
+
+.globl jump_eret
+       .type   jump_eret, @function
+jump_eret:
+       mov     reg_cop0+48, %ebx /* Status */
+       add     last_count, %esi
+       and     $0xFFFFFFFD, %ebx
+       mov     %esi, reg_cop0+36 /* Count */
+       mov     %ebx, reg_cop0+48 /* Status */
+       call    check_interupt
+       mov     next_interupt, %eax
+       mov     reg_cop0+36, %esi
+       mov     %eax, last_count
+       sub     %eax, %esi
+       mov     reg_cop0+56, %edi /* EPC */
+       jns     .E11
+.E8:
+       mov     %esi, %r12d
+       mov     $248, %ebx
+       xor     %esi, %esi
+.E9:
+       mov     reg(%ebx), %ecx
+       mov     reg+4(%ebx), %edx
+       sar     $31, %ecx
+       xor     %ecx, %edx
+       neg     %edx
+       adc     %esi, %esi
+       sub     $8, %ebx
+       jne     .E9
+       mov     hi(%ebx), %ecx
+       mov     hi+4(%ebx), %edx
+       sar     $31, %ecx
+       xor     %ecx, %edx
+       jne     .E10
+       mov     lo(%ebx), %ecx
+       mov     lo+4(%ebx), %edx
+       sar     $31, %ecx
+       xor     %ecx, %edx
+.E10:
+       neg     %edx
+       adc     %esi, %esi
+       call    get_addr_32
+       mov     %r12d, %esi
+       jmp     *%rax
+.E11:
+       mov     %edi, pcaddr
+       call    cc_interrupt
+       mov     pcaddr, %edi
+       jmp     .E8
+       .size   jump_eret, .-jump_eret
+
+.globl new_dyna_start
+       .type   new_dyna_start, @function
+new_dyna_start:
+       push    %rbp
+       push    %rbx
+       push    %r12
+       push    %r13
+       push    %r14
+       push    %r15
+       mov     $0xa4000040, %edi
+       call    new_recompile_block
+       add     $-8, %rsp /* align stack */
+       movl    next_interupt, %edi
+       movl    reg_cop0+36, %esi
+       movl    %edi, last_count
+       subl    %edi, %esi
+       jmp     0x70000000
+       .size   new_dyna_start, .-new_dyna_start
+
+.globl write_rdram_new
+       .type   write_rdram_new, @function
+write_rdram_new:
+       mov     address, %edi
+       mov     word, %ecx
+       and     $0x7FFFFFFF, %edi
+       mov     %ecx, rdram(%rdi)
+       jmp     .E12
+       .size   write_rdram_new, .-write_rdram_new
+
+.globl write_rdramb_new
+       .type   write_rdramb_new, @function
+write_rdramb_new:
+       mov     address, %edi
+       xor     $3, %edi
+       movb    byte, %cl
+       and     $0x7FFFFFFF, %edi
+       movb    %cl, rdram(%rdi)
+       jmp     .E12
+       .size   write_rdramb_new, .-write_rdramb_new
+
+.globl write_rdramh_new
+       .type   write_rdramh_new, @function
+write_rdramh_new:
+       mov     address, %edi
+       xor     $2, %edi
+       movw    hword, %cx
+       and     $0x7FFFFFFF, %edi
+       movw    %cx, rdram(%rdi)
+       jmp     .E12
+       .size   write_rdramh_new, .-write_rdramh_new
+
+.globl write_rdramd_new
+       .type   write_rdramd_new, @function
+write_rdramd_new:
+       mov     address, %edi
+       mov     dword+4, %ecx
+       mov     dword, %edx
+       and     $0x7FFFFFFF, %edi
+       mov     %ecx, rdram(%rdi)
+       mov     %edx, rdram+4(%rdi)
+       jmp     .E12
+       .size   write_rdramd_new, .-write_rdramd_new
+
+.globl do_invalidate
+       .type   do_invalidate, @function
+do_invalidate:
+       mov     address, %edi
+       mov     %edi, %ebx /* Return ebx to caller */
+.E12:
+       shr     $12, %edi
+       mov     %edi, %r12d /* Return r12 to caller */
+       cmpb    $1, invalid_code(%edi)
+       je      .E13
+       call    invalidate_block
+.E13:
+       ret
+       .size   do_invalidate, .-do_invalidate
+
+.globl read_nomem_new
+       .type   read_nomem_new, @function
+read_nomem_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       mov     (%ebx,%edi,4), %ecx
+       mov     %ecx, readmem_dword
+       ret
+       .size   read_nomem_new, .-read_nomem_new
+
+.globl read_nomemb_new
+       .type   read_nomemb_new, @function
+read_nomemb_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       xor     $3, %ebx
+       movzbl  (%ebx,%edi,4), %ecx
+       mov     %ecx, readmem_dword
+       ret
+       .size   read_nomemb_new, .-read_nomemb_new
+
+.globl read_nomemh_new
+       .type   read_nomemh_new, @function
+read_nomemh_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       xor     $2, %ebx
+       movzwl  (%ebx,%edi,4), %ecx
+       mov     %ecx, readmem_dword
+       ret
+       .size   read_nomemh_new, .-read_nomemh_new
+
+.globl read_nomemd_new
+       .type   read_nomemd_new, @function
+read_nomemd_new:
+       mov     address, %edi
+       mov     %edi, %ebx
+       shr     $12, %edi
+       mov     memory_map(,%edi,4),%edi
+       mov     $0x8, %eax
+       test    %edi, %edi
+       js      tlb_exception
+       mov     4(%ebx,%edi,4), %ecx
+       mov     (%ebx,%edi,4), %edx
+       mov     %ecx, readmem_dword
+       mov     %edx, readmem_dword+4
+       ret
+       .size   read_nomemd_new, .-read_nomemd_new
+
+.globl write_nomem_new
+       .type   write_nomem_new, @function
+write_nomem_new:
+       call    do_invalidate
+       mov     memory_map(,%r12d,4),%edi
+       mov     word, %ecx
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       mov     %ecx, (%ebx,%edi)
+       ret
+       .size   write_nomem_new, .-write_nomem_new
+
+.globl write_nomemb_new
+       .type   write_nomemb_new, @function
+write_nomemb_new:
+       call    do_invalidate
+       mov     memory_map(,%r12d,4),%edi
+       movb    byte, %cl
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       xor     $3, %ebx
+       movb    %cl, (%ebx,%edi)
+       ret
+       .size   write_nomemb_new, .-write_nomemb_new
+
+.globl write_nomemh_new
+       .type   write_nomemh_new, @function
+write_nomemh_new:
+       call    do_invalidate
+       mov     memory_map(,%r12d,4),%edi
+       movw    hword, %cx
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       xor     $2, %ebx
+       movw    %cx, (%ebx,%edi)
+       ret
+       .size   write_nomemh_new, .-write_nomemh_new
+
+.globl write_nomemd_new
+       .type   write_nomemd_new, @function
+write_nomemd_new:
+       call    do_invalidate
+       mov     memory_map(,%r12d,4),%edi
+       mov     dword+4, %edx
+       mov     dword, %ecx
+       mov     $0xc, %eax
+       shl     $2, %edi
+       jc      tlb_exception
+       mov     %edx, (%ebx,%edi)
+       mov     %ecx, 4(%ebx,%edi)
+       ret
+       .size   write_nomemd_new, .-write_nomemd_new
+
+.globl tlb_exception
+       .type   tlb_exception, @function
+tlb_exception:
+       /* eax = cause */
+       /* ebx = address */
+       /* ebp = instr addr + flags */
+       mov     8(%rsp), %ebp
+       mov     reg_cop0+48, %esi
+       mov     %ebp, %ecx
+       mov     %ebp, %edx
+       mov     %ebp, %edi
+       shl     $31, %ebp
+       shr     $12, %ecx
+       or      %ebp, %eax
+       sar     $29, %ebp
+       and     $0xFFFFFFFC, %edx
+       mov     memory_map(,%ecx,4), %ecx
+       or      $2, %esi
+       mov     (%edx, %ecx, 4), %ecx
+       add     %ebp, %edx
+       mov     %esi, reg_cop0+48 /* Status */
+       mov     %eax, reg_cop0+52 /* Cause */
+       mov     %edx, reg_cop0+56 /* EPC */
+       add     $0x48, %rsp
+       mov     $0x6000022, %edx
+       mov     %ecx, %ebp
+       movswl  %cx, %eax
+       shr     $26, %ecx
+       shr     $21, %ebp
+       sub     %eax, %ebx
+       and     $0x1f, %ebp
+       ror     %cl, %edx
+       mov     reg_cop0+16, %esi
+       cmovc   reg(,%ebp,8), %ebx
+       and     $0xFF80000F, %esi
+       mov     %ebx, reg(,%ebp,8)
+       add     %ebx, %eax
+       sar     $31, %ebx
+       mov     %eax, reg_cop0+32 /* BadVAddr */
+       shr     $9, %eax
+       test    $2, %edi
+       cmove   reg+4(,%ebp,8), %ebx
+       and     $0x007FFFF0, %eax
+       mov     $0x80000180, %edi
+       mov     %ebx, reg+4(,%ebp,8)
+       or      %eax, %esi
+       mov     %esi, reg_cop0+16 /* Context */
+       call    get_addr_ht
+       movl    next_interupt, %edi
+       movl    reg_cop0+36, %esi /* Count */
+       movl    %edi, last_count
+       subl    %edi, %esi
+       jmp     *%rax
+       .size   tlb_exception, .-tlb_exception
diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
new file mode 100644 (file)
index 0000000..9b8f153
--- /dev/null
@@ -0,0 +1,10487 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ *   Mupen64plus - new_dynarec.c                                           *
+ *   Copyright (C) 2009-2010 Ari64                                         *
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ *   This program is distributed in the hope that it will be useful,       *
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+ *   GNU General Public License for more details.                          *
+ *                                                                         *
+ *   You should have received a copy of the GNU General Public License     *
+ *   along with this program; if not, write to the                         *
+ *   Free Software Foundation, Inc.,                                       *
+ *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+#include <stdlib.h>
+#include <stdint.h> //include for uint64_t
+#include <assert.h>
+
+#include "../recomp.h"
+#include "../recomph.h" //include for function prototypes
+#include "../macros.h"
+#include "../r4300.h"
+#include "../ops.h"
+#include "../interupt.h"
+
+#include "../../memory/memory.h"
+
+#include <sys/mman.h>
+
+#ifdef __i386__
+#include "assem_x86.h"
+#endif
+#ifdef __x86_64__
+#include "assem_x64.h"
+#endif
+#ifdef __arm__
+#include "assem_arm.h"
+#endif
+
+#define MAXBLOCK 4096
+#define MAX_OUTPUT_BLOCK_SIZE 262144
+#define CLOCK_DIVIDER 2
+
+struct regstat
+{
+  signed char regmap_entry[HOST_REGS];
+  signed char regmap[HOST_REGS];
+  uint64_t was32;
+  uint64_t is32;
+  uint64_t wasdirty;
+  uint64_t dirty;
+  uint64_t u;
+  uint64_t uu;
+  u_int wasconst;
+  u_int isconst;
+  uint64_t constmap[HOST_REGS];
+};
+
+struct ll_entry
+{
+  u_int vaddr;
+  u_int reg32;
+  void *addr;
+  struct ll_entry *next;
+};
+
+  u_int start;
+  u_int *source;
+  u_int pagelimit;
+  char insn[MAXBLOCK][10];
+  u_char itype[MAXBLOCK];
+  u_char opcode[MAXBLOCK];
+  u_char opcode2[MAXBLOCK];
+  u_char bt[MAXBLOCK];
+  u_char rs1[MAXBLOCK];
+  u_char rs2[MAXBLOCK];
+  u_char rt1[MAXBLOCK];
+  u_char rt2[MAXBLOCK];
+  u_char us1[MAXBLOCK];
+  u_char us2[MAXBLOCK];
+  u_char dep1[MAXBLOCK];
+  u_char dep2[MAXBLOCK];
+  u_char lt1[MAXBLOCK];
+  int imm[MAXBLOCK];
+  u_int ba[MAXBLOCK];
+  char likely[MAXBLOCK];
+  char is_ds[MAXBLOCK];
+  uint64_t unneeded_reg[MAXBLOCK];
+  uint64_t unneeded_reg_upper[MAXBLOCK];
+  uint64_t branch_unneeded_reg[MAXBLOCK];
+  uint64_t branch_unneeded_reg_upper[MAXBLOCK];
+  uint64_t p32[MAXBLOCK];
+  uint64_t pr32[MAXBLOCK];
+  signed char regmap_pre[MAXBLOCK][HOST_REGS];
+  signed char regmap[MAXBLOCK][HOST_REGS];
+  signed char regmap_entry[MAXBLOCK][HOST_REGS];
+  uint64_t constmap[MAXBLOCK][HOST_REGS];
+  uint64_t known_value[HOST_REGS];
+  u_int known_reg;
+  struct regstat regs[MAXBLOCK];
+  struct regstat branch_regs[MAXBLOCK];
+  u_int needed_reg[MAXBLOCK];
+  uint64_t requires_32bit[MAXBLOCK];
+  u_int wont_dirty[MAXBLOCK];
+  u_int will_dirty[MAXBLOCK];
+  int ccadj[MAXBLOCK];
+  int slen;
+  u_int instr_addr[MAXBLOCK];
+  u_int link_addr[MAXBLOCK][3];
+  int linkcount;
+  u_int stubs[MAXBLOCK*3][8];
+  int stubcount;
+  u_int literals[1024][2];
+  int literalcount;
+  int is_delayslot;
+  int cop1_usable;
+  u_char *out;
+  struct ll_entry *jump_in[4096];
+  struct ll_entry *jump_out[4096];
+  struct ll_entry *jump_dirty[4096];
+  u_int hash_table[65536][4]  __attribute__((aligned(16)));
+  char shadow[1048576]  __attribute__((aligned(16)));
+  void *copy;
+  int expirep;
+  u_int using_tlb;
+  u_int stop_after_jal;
+  extern u_char restore_candidate[512];
+  extern int cycle_count;
+
+  /* registers that may be allocated */
+  /* 1-31 gpr */
+#define HIREG 32 // hi
+#define LOREG 33 // lo
+#define FSREG 34 // FPU status (FCSR)
+#define CSREG 35 // Coprocessor status
+#define CCREG 36 // Cycle count
+#define INVCP 37 // Pointer to invalid_code
+#define TEMPREG 38
+#define FTEMP 38 // FPU temporary register
+#define PTEMP 39 // Prefetch temporary register
+#define TLREG 40 // TLB mapping offset
+#define RHASH 41 // Return address hash
+#define RHTBL 42 // Return address hash table address
+#define RTEMP 43 // JR/JALR address register
+#define MAXREG 43
+#define AGEN1 44 // Address generation temporary register
+#define AGEN2 45 // Address generation temporary register
+#define MGEN1 46 // Maptable address generation temporary register
+#define MGEN2 47 // Maptable address generation temporary register
+#define BTREG 48 // Branch target temporary register
+
+  /* instruction types */
+#define NOP 0     // No operation
+#define LOAD 1    // Load
+#define STORE 2   // Store
+#define LOADLR 3  // Unaligned load
+#define STORELR 4 // Unaligned store
+#define MOV 5     // Move 
+#define ALU 6     // Arithmetic/logic
+#define MULTDIV 7 // Multiply/divide
+#define SHIFT 8   // Shift by register
+#define SHIFTIMM 9// Shift by immediate
+#define IMM16 10  // 16-bit immediate
+#define RJUMP 11  // Unconditional jump to register
+#define UJUMP 12  // Unconditional jump
+#define CJUMP 13  // Conditional branch (BEQ/BNE/BGTZ/BLEZ)
+#define SJUMP 14  // Conditional branch (regimm format)
+#define COP0 15   // Coprocessor 0
+#define COP1 16   // Coprocessor 1
+#define C1LS 17   // Coprocessor 1 load/store
+#define FJUMP 18  // Conditional branch (floating point)
+#define FLOAT 19  // Floating point unit
+#define FCONV 20  // Convert integer to float
+#define FCOMP 21  // Floating point compare (sets FSREG)
+#define SYSCALL 22// SYSCALL
+#define OTHER 23  // Other
+#define SPAN 24   // Branch/delay slot spans 2 pages
+#define NI 25     // Not implemented
+
+  /* stubs */
+#define CC_STUB 1
+#define FP_STUB 2
+#define LOADB_STUB 3
+#define LOADH_STUB 4
+#define LOADW_STUB 5
+#define LOADD_STUB 6
+#define LOADBU_STUB 7
+#define LOADHU_STUB 8
+#define STOREB_STUB 9
+#define STOREH_STUB 10
+#define STOREW_STUB 11
+#define STORED_STUB 12
+#define STORELR_STUB 13
+#define INVCODE_STUB 14
+
+  /* branch codes */
+#define TAKEN 1
+#define NOTTAKEN 2
+#define NULLDS 3
+
+// asm linkage
+int new_recompile_block(int addr);
+void *get_addr_ht(u_int vaddr);
+void invalidate_block(u_int block);
+void invalidate_addr(u_int addr);
+void remove_hash(int vaddr);
+void jump_vaddr();
+void dyna_linker();
+void dyna_linker_ds();
+void verify_code();
+void verify_code_vm();
+void verify_code_ds();
+void cc_interrupt();
+void fp_exception();
+void fp_exception_ds();
+void jump_syscall();
+void jump_eret();
+
+// TLB
+void TLBWI_new();
+void TLBWR_new();
+void read_nomem_new();
+void read_nomemb_new();
+void read_nomemh_new();
+void read_nomemd_new();
+void write_nomem_new();
+void write_nomemb_new();
+void write_nomemh_new();
+void write_nomemd_new();
+void write_rdram_new();
+void write_rdramb_new();
+void write_rdramh_new();
+void write_rdramd_new();
+extern u_int memory_map[1048576];
+
+// Needed by assembler
+void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32);
+void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty);
+void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr);
+void load_all_regs(signed char i_regmap[]);
+void load_needed_regs(signed char i_regmap[],signed char next_regmap[]);
+void load_regs_entry(int t);
+void load_all_consts(signed char regmap[],int is32,u_int dirty,int i);
+
+int tracedebug=0;
+
+//#define DEBUG_CYCLE_COUNT 1
+
+void nullf() {}
+//#define assem_debug printf
+//#define inv_debug printf
+#define assem_debug nullf
+#define inv_debug nullf
+
+void tlb_hacks()
+{
+  // Goldeneye hack
+  if (strncmp((char *) ROM_HEADER->nom, "GOLDENEYE",9) == 0)
+  {
+    u_int addr;
+    int n;
+    switch (ROM_HEADER->Country_code&0xFF) 
+    {
+      case 0x45: // U
+        addr=0x34b30;
+        break;                   
+      case 0x4A: // J 
+        addr=0x34b70;    
+        break;    
+      case 0x50: // E 
+        addr=0x329f0;
+        break;                        
+      default: 
+        // Unknown country code
+        addr=0;
+        break;
+    }
+    u_int rom_addr=(u_int)rom;
+    #ifdef ROM_COPY
+    // Since memory_map is 32-bit, on 64-bit systems the rom needs to be
+    // in the lower 4G of memory to use this hack.  Copy it if necessary.
+    if((void *)rom>(void *)0xffffffff) {
+      munmap(ROM_COPY, 67108864);
+      if(mmap(ROM_COPY, 12582912,
+              PROT_READ | PROT_WRITE,
+              MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
+              -1, 0) <= 0) {printf("mmap() failed\n");}
+      memcpy(ROM_COPY,rom,12582912);
+      rom_addr=(u_int)ROM_COPY;
+    }
+    #endif
+    if(addr) {
+      for(n=0x7F000;n<0x80000;n++) {
+        memory_map[n]=(((u_int)(rom_addr+addr-0x7F000000))>>2)|0x40000000;
+      }
+    }
+  }
+}
+
+// Get address from virtual address
+// This is called from the recompiled JR/JALR instructions
+void *get_addr(u_int vaddr)
+{
+  u_int page=(vaddr^0x80000000)>>12;
+  u_int vpage=page;
+  if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
+  if(page>2048) page=2048+(page&2047);
+  if(vpage>262143&&tlb_LUT_r[vaddr>>12]) vpage&=2047; // jump_dirty uses a hash of the virtual address instead
+  if(vpage>2048) vpage=2048+(vpage&2047);
+  struct ll_entry *head;
+  //printf("TRACE: count=%d next=%d (get_addr %x,page %d)\n",Count,next_interupt,vaddr,page);
+  head=jump_in[page];
+  while(head!=NULL) {
+    if(head->vaddr==vaddr&&head->reg32==0) {
+  //printf("TRACE: count=%d next=%d (get_addr match %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
+      int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+      ht_bin[3]=ht_bin[1];
+      ht_bin[2]=ht_bin[0];
+      ht_bin[1]=(int)head->addr;
+      ht_bin[0]=vaddr;
+      return head->addr;
+    }
+    head=head->next;
+  }
+  head=jump_dirty[vpage];
+  while(head!=NULL) {
+    if(head->vaddr==vaddr&&head->reg32==0) {
+      //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
+      // Don't restore blocks which are about to expire from the cache
+      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
+      if(verify_dirty(head->addr)) {
+        //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]);
+        invalid_code[vaddr>>12]=0;
+        memory_map[vaddr>>12]|=0x40000000;
+        if(vpage<2048) {
+          if(tlb_LUT_r[vaddr>>12]) {
+            invalid_code[tlb_LUT_r[vaddr>>12]>>12]=0;
+            memory_map[tlb_LUT_r[vaddr>>12]>>12]|=0x40000000;
+          }
+          restore_candidate[vpage>>3]|=1<<(vpage&7);
+        }
+        else restore_candidate[page>>3]|=1<<(page&7);
+        int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+        if(ht_bin[0]==vaddr) {
+          ht_bin[1]=(int)head->addr; // Replace existing entry
+        }
+        else
+        {
+          ht_bin[3]=ht_bin[1];
+          ht_bin[2]=ht_bin[0];
+          ht_bin[1]=(int)head->addr;
+          ht_bin[0]=vaddr;
+        }
+        return head->addr;
+      }
+    }
+    head=head->next;
+  }
+  //printf("TRACE: count=%d next=%d (get_addr no-match %x)\n",Count,next_interupt,vaddr);
+  int r=new_recompile_block(vaddr);
+  if(r==0) return get_addr(vaddr);
+  // Execute in unmapped page, generate pagefault execption
+  Status|=2;
+  Cause=(vaddr<<31)|0x8;
+  EPC=(vaddr&1)?vaddr-5:vaddr;
+  BadVAddr=(vaddr&~1);
+  Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0);
+  EntryHi=BadVAddr&0xFFFFE000;
+  return get_addr_ht(0x80000000);
+}
+// Look up address in hash table first
+void *get_addr_ht(u_int vaddr)
+{
+  //printf("TRACE: count=%d next=%d (get_addr_ht %x)\n",Count,next_interupt,vaddr);
+  int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+  if(ht_bin[0]==vaddr) return (void *)ht_bin[1];
+  if(ht_bin[2]==vaddr) return (void *)ht_bin[3];
+  return get_addr(vaddr);
+}
+
+void *get_addr_32(u_int vaddr,u_int flags)
+{
+  //printf("TRACE: count=%d next=%d (get_addr_32 %x,flags %x)\n",Count,next_interupt,vaddr,flags);
+  int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+  if(ht_bin[0]==vaddr) return (void *)ht_bin[1];
+  if(ht_bin[2]==vaddr) return (void *)ht_bin[3];
+  u_int page=(vaddr^0x80000000)>>12;
+  u_int vpage=page;
+  if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
+  if(page>2048) page=2048+(page&2047);
+  if(vpage>262143&&tlb_LUT_r[vaddr>>12]) vpage&=2047; // jump_dirty uses a hash of the virtual address instead
+  if(vpage>2048) vpage=2048+(vpage&2047);
+  struct ll_entry *head;
+  head=jump_in[page];
+  while(head!=NULL) {
+    if(head->vaddr==vaddr&&(head->reg32&flags)==0) {
+      //printf("TRACE: count=%d next=%d (get_addr_32 match %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
+      if(head->reg32==0) {
+        int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+        if(ht_bin[0]==-1) {
+          ht_bin[1]=(int)head->addr;
+          ht_bin[0]=vaddr;
+        }else if(ht_bin[2]==-1) {
+          ht_bin[3]=(int)head->addr;
+          ht_bin[2]=vaddr;
+        }
+        //ht_bin[3]=ht_bin[1];
+        //ht_bin[2]=ht_bin[0];
+        //ht_bin[1]=(int)head->addr;
+        //ht_bin[0]=vaddr;
+      }
+      return head->addr;
+    }
+    head=head->next;
+  }
+  head=jump_dirty[vpage];
+  while(head!=NULL) {
+    if(head->vaddr==vaddr&&(head->reg32&flags)==0) {
+      //printf("TRACE: count=%d next=%d (get_addr_32 match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
+      // Don't restore blocks which are about to expire from the cache
+      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
+      if(verify_dirty(head->addr)) {
+        //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]);
+        invalid_code[vaddr>>12]=0;
+        memory_map[vaddr>>12]|=0x40000000;
+        if(vpage<2048) {
+          if(tlb_LUT_r[vaddr>>12]) {
+            invalid_code[tlb_LUT_r[vaddr>>12]>>12]=0;
+            memory_map[tlb_LUT_r[vaddr>>12]>>12]|=0x40000000;
+          }
+          restore_candidate[vpage>>3]|=1<<(vpage&7);
+        }
+        else restore_candidate[page>>3]|=1<<(page&7);
+        if(head->reg32==0) {
+          int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+          if(ht_bin[0]==-1) {
+            ht_bin[1]=(int)head->addr;
+            ht_bin[0]=vaddr;
+          }else if(ht_bin[2]==-1) {
+            ht_bin[3]=(int)head->addr;
+            ht_bin[2]=vaddr;
+          }
+          //ht_bin[3]=ht_bin[1];
+          //ht_bin[2]=ht_bin[0];
+          //ht_bin[1]=(int)head->addr;
+          //ht_bin[0]=vaddr;
+        }
+        return head->addr;
+      }
+    }
+    head=head->next;
+  }
+  //printf("TRACE: count=%d next=%d (get_addr_32 no-match %x,flags %x)\n",Count,next_interupt,vaddr,flags);
+  int r=new_recompile_block(vaddr);
+  if(r==0) return get_addr(vaddr);
+  // Execute in unmapped page, generate pagefault execption
+  Status|=2;
+  Cause=(vaddr<<31)|0x8;
+  EPC=(vaddr&1)?vaddr-5:vaddr;
+  BadVAddr=(vaddr&~1);
+  Context=(Context&0xFF80000F)|((BadVAddr>>9)&0x007FFFF0);
+  EntryHi=BadVAddr&0xFFFFE000;
+  return get_addr_ht(0x80000000);
+}
+
+void clear_all_regs(signed char regmap[])
+{
+  int hr;
+  for (hr=0;hr<HOST_REGS;hr++) regmap[hr]=-1;
+}
+
+signed char get_reg(signed char regmap[],int r)
+{
+  int hr;
+  for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&&regmap[hr]==r) return hr;
+  return -1;
+}
+
+// Find a register that is available for two consecutive cycles
+signed char get_reg2(signed char regmap1[],signed char regmap2[],int r)
+{
+  int hr;
+  for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&&regmap1[hr]==r&&regmap2[hr]==r) return hr;
+  return -1;
+}
+
+int count_free_regs(signed char regmap[])
+{
+  int count=0;
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(hr!=EXCLUDE_REG) {
+      if(regmap[hr]<0) count++;
+    }
+  }
+  return count;
+}
+
+void dirty_reg(struct regstat *cur,signed char reg)
+{
+  int hr;
+  if(!reg) return;
+  for (hr=0;hr<HOST_REGS;hr++) {
+    if((cur->regmap[hr]&63)==reg) {
+      cur->dirty|=1<<hr;
+    }
+  }
+}
+
+// If we dirty the lower half of a 64 bit register which is now being
+// sign-extended, we need to dump the upper half.
+// Note: Do this only after completion of the instruction, because
+// some instructions may need to read the full 64-bit value even if
+// overwriting it (eg SLTI, DSRA32).
+static void flush_dirty_uppers(struct regstat *cur)
+{
+  int hr,reg;
+  for (hr=0;hr<HOST_REGS;hr++) {
+    if((cur->dirty>>hr)&1) {
+      reg=cur->regmap[hr];
+      if(reg>=64) 
+        if((cur->is32>>(reg&63))&1) cur->regmap[hr]=-1;
+    }
+  }
+}
+
+void set_const(struct regstat *cur,signed char reg,uint64_t value)
+{
+  int hr;
+  if(!reg) return;
+  for (hr=0;hr<HOST_REGS;hr++) {
+    if(cur->regmap[hr]==reg) {
+      cur->isconst|=1<<hr;
+      cur->constmap[hr]=value;
+    }
+    else if((cur->regmap[hr]^64)==reg) {
+      cur->isconst|=1<<hr;
+      cur->constmap[hr]=value>>32;
+    }
+  }
+}
+
+void clear_const(struct regstat *cur,signed char reg)
+{
+  int hr;
+  if(!reg) return;
+  for (hr=0;hr<HOST_REGS;hr++) {
+    if((cur->regmap[hr]&63)==reg) {
+      cur->isconst&=~(1<<hr);
+    }
+  }
+}
+
+int is_const(struct regstat *cur,signed char reg)
+{
+  int hr;
+  if(!reg) return 1;
+  for (hr=0;hr<HOST_REGS;hr++) {
+    if((cur->regmap[hr]&63)==reg) {
+      return (cur->isconst>>hr)&1;
+    }
+  }
+  return 0;
+}
+uint64_t get_const(struct regstat *cur,signed char reg)
+{
+  int hr;
+  if(!reg) return 0;
+  for (hr=0;hr<HOST_REGS;hr++) {
+    if(cur->regmap[hr]==reg) {
+      return cur->constmap[hr];
+    }
+  }
+  printf("Unknown constant in r%d\n",reg);
+  exit(1);
+}
+
+// Least soon needed registers
+// Look at the next ten instructions and see which registers
+// will be used.  Try not to reallocate these.
+void lsn(u_char hsn[], int i, int *preferred_reg)
+{
+  int j;
+  int b=-1;
+  for(j=0;j<9;j++)
+  {
+    if(i+j>=slen) {
+      j=slen-i-1;
+      break;
+    }
+    if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+    {
+      // Don't go past an unconditonal jump
+      j++;
+      break;
+    }
+  }
+  for(;j>=0;j--)
+  {
+    if(rs1[i+j]) hsn[rs1[i+j]]=j;
+    if(rs2[i+j]) hsn[rs2[i+j]]=j;
+    if(rt1[i+j]) hsn[rt1[i+j]]=j;
+    if(rt2[i+j]) hsn[rt2[i+j]]=j;
+    if(itype[i+j]==STORE || itype[i+j]==STORELR) {
+      // Stores can allocate zero
+      hsn[rs1[i+j]]=j;
+      hsn[rs2[i+j]]=j;
+    }
+    // On some architectures stores need invc_ptr
+    #if defined(HOST_IMM8)
+    if(itype[i+j]==STORE || itype[i+j]==STORELR || (opcode[i+j]&0x3b)==0x39) {
+      hsn[INVCP]=j;
+    }
+    #endif
+    if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP||itype[i+j]==FJUMP))
+    {
+      hsn[CCREG]=j;
+      b=j;
+    }
+  }
+  if(b>=0)
+  {
+    if(ba[i+b]>=start && ba[i+b]<(start+slen*4))
+    {
+      // Follow first branch
+      int t=(ba[i+b]-start)>>2;
+      j=7-b;if(t+j>=slen) j=slen-t-1;
+      for(;j>=0;j--)
+      {
+        if(rs1[t+j]) if(hsn[rs1[t+j]]>j+b+2) hsn[rs1[t+j]]=j+b+2;
+        if(rs2[t+j]) if(hsn[rs2[t+j]]>j+b+2) hsn[rs2[t+j]]=j+b+2;
+        //if(rt1[t+j]) if(hsn[rt1[t+j]]>j+b+2) hsn[rt1[t+j]]=j+b+2;
+        //if(rt2[t+j]) if(hsn[rt2[t+j]]>j+b+2) hsn[rt2[t+j]]=j+b+2;
+      }
+    }
+    // TODO: preferred register based on backward branch
+  }
+  // Delay slot should preferably not overwrite branch conditions or cycle count
+  if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)) {
+    if(rs1[i-1]) if(hsn[rs1[i-1]]>1) hsn[rs1[i-1]]=1;
+    if(rs2[i-1]) if(hsn[rs2[i-1]]>1) hsn[rs2[i-1]]=1;
+    hsn[CCREG]=1;
+    // ...or hash tables
+    hsn[RHASH]=1;
+    hsn[RHTBL]=1;
+  }
+  // Coprocessor load/store needs FTEMP, even if not declared
+  if(itype[i]==C1LS) {
+    hsn[FTEMP]=0;
+  }
+  // Load L/R also uses FTEMP as a temporary register
+  if(itype[i]==LOADLR) {
+    hsn[FTEMP]=0;
+  }
+  // Also 64-bit SDL/SDR
+  if(opcode[i]==0x2c||opcode[i]==0x2d) {
+    hsn[FTEMP]=0;
+  }
+  // Don't remove the TLB registers either
+  if(itype[i]==LOAD || itype[i]==LOADLR || itype[i]==STORE || itype[i]==STORELR || itype[i]==C1LS ) {
+    hsn[TLREG]=0;
+  }
+  // Don't remove the miniht registers
+  if(itype[i]==UJUMP||itype[i]==RJUMP)
+  {
+    hsn[RHASH]=0;
+    hsn[RHTBL]=0;
+  }
+}
+
+// We only want to allocate registers if we're going to use them again soon
+int needed_again(int r, int i)
+{
+  int j;
+  int b=-1;
+  int rn=10;
+  int hr;
+  u_char hsn[MAXREG+1];
+  int preferred_reg;
+  
+  memset(hsn,10,sizeof(hsn));
+  lsn(hsn,i,&preferred_reg);
+  
+  if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000))
+  {
+    if(ba[i-1]<start || ba[i-1]>start+slen*4-4)
+      return 0; // Don't need any registers if exiting the block
+  }
+  for(j=0;j<9;j++)
+  {
+    if(i+j>=slen) {
+      j=slen-i-1;
+      break;
+    }
+    if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+    {
+      // Don't go past an unconditonal jump
+      j++;
+      break;
+    }
+    if(itype[i+j]==SYSCALL||((source[i+j]&0xfc00003f)==0x0d))
+    {
+      break;
+    }
+  }
+  for(;j>=1;j--)
+  {
+    if(rs1[i+j]==r) rn=j;
+    if(rs2[i+j]==r) rn=j;
+    if((unneeded_reg[i+j]>>r)&1) rn=10;
+    if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP||itype[i+j]==FJUMP))
+    {
+      b=j;
+    }
+  }
+  /*
+  if(b>=0)
+  {
+    if(ba[i+b]>=start && ba[i+b]<(start+slen*4))
+    {
+      // Follow first branch
+      int o=rn;
+      int t=(ba[i+b]-start)>>2;
+      j=7-b;if(t+j>=slen) j=slen-t-1;
+      for(;j>=0;j--)
+      {
+        if(!((unneeded_reg[t+j]>>r)&1)) {
+          if(rs1[t+j]==r) if(rn>j+b+2) rn=j+b+2;
+          if(rs2[t+j]==r) if(rn>j+b+2) rn=j+b+2;
+        }
+        else rn=o;
+      }
+    }
+  }*/
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(rn<hsn[hr]) return 1;
+    }
+  }
+  return 0;
+}
+
+// Try to match register allocations at the end of a loop with those
+// at the beginning
+int loop_reg(int i, int r, int hr)
+{
+  int j,k;
+  for(j=0;j<9;j++)
+  {
+    if(i+j>=slen) {
+      j=slen-i-1;
+      break;
+    }
+    if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+    {
+      // Don't go past an unconditonal jump
+      j++;
+      break;
+    }
+  }
+  k=0;
+  if(i>0){
+    if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)
+      k--;
+  }
+  for(;k<j;k++)
+  {
+    if(r<64&&((unneeded_reg[i+k]>>r)&1)) return hr;
+    if(r>64&&((unneeded_reg_upper[i+k]>>r)&1)) return hr;
+    if(i+k>=0&&(itype[i+k]==UJUMP||itype[i+k]==CJUMP||itype[i+k]==SJUMP||itype[i+k]==FJUMP))
+    {
+      if(ba[i+k]>=start && ba[i+k]<(start+i*4))
+      {
+        int t=(ba[i+k]-start)>>2;
+        int reg=get_reg(regs[t].regmap_entry,r);
+        if(reg>=0) return reg;
+        //reg=get_reg(regs[t+1].regmap_entry,r);
+        //if(reg>=0) return reg;
+      }
+    }
+  }
+  return hr;
+}
+
+
+// Allocate every register, preserving source/target regs
+void alloc_all(struct regstat *cur,int i)
+{
+  int hr;
+  
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(((cur->regmap[hr]&63)!=rs1[i])&&((cur->regmap[hr]&63)!=rs2[i])&&
+         ((cur->regmap[hr]&63)!=rt1[i])&&((cur->regmap[hr]&63)!=rt2[i]))
+      {
+        cur->regmap[hr]=-1;
+        cur->dirty&=~(1<<hr);
+      }
+      // Don't need zeros
+      if((cur->regmap[hr]&63)==0)
+      {
+        cur->regmap[hr]=-1;
+        cur->dirty&=~(1<<hr);
+      }
+    }
+  }
+}
+
+
+void div64(int64_t dividend,int64_t divisor)
+{
+  lo=dividend/divisor;
+  hi=dividend%divisor;
+  //printf("TRACE: ddiv %8x%8x %8x%8x\n" ,(int)reg[HIREG],(int)(reg[HIREG]>>32)
+  //                                     ,(int)reg[LOREG],(int)(reg[LOREG]>>32));
+}
+void divu64(uint64_t dividend,uint64_t divisor)
+{
+  lo=dividend/divisor;
+  hi=dividend%divisor;
+  //printf("TRACE: ddivu %8x%8x %8x%8x\n",(int)reg[HIREG],(int)(reg[HIREG]>>32)
+  //                                     ,(int)reg[LOREG],(int)(reg[LOREG]>>32));
+}
+
+void mult64(uint64_t m1,uint64_t m2)
+{
+   unsigned long long int op1, op2, op3, op4;
+   unsigned long long int result1, result2, result3, result4;
+   unsigned long long int temp1, temp2, temp3, temp4;
+   int sign = 0;
+   
+   if (m1 < 0)
+     {
+    op2 = -m1;
+    sign = 1 - sign;
+     }
+   else op2 = m1;
+   if (m2 < 0)
+     {
+    op4 = -m2;
+    sign = 1 - sign;
+     }
+   else op4 = m2;
+   
+   op1 = op2 & 0xFFFFFFFF;
+   op2 = (op2 >> 32) & 0xFFFFFFFF;
+   op3 = op4 & 0xFFFFFFFF;
+   op4 = (op4 >> 32) & 0xFFFFFFFF;
+   
+   temp1 = op1 * op3;
+   temp2 = (temp1 >> 32) + op1 * op4;
+   temp3 = op2 * op3;
+   temp4 = (temp3 >> 32) + op2 * op4;
+   
+   result1 = temp1 & 0xFFFFFFFF;
+   result2 = temp2 + (temp3 & 0xFFFFFFFF);
+   result3 = (result2 >> 32) + temp4;
+   result4 = (result3 >> 32);
+   
+   lo = result1 | (result2 << 32);
+   hi = (result3 & 0xFFFFFFFF) | (result4 << 32);
+   if (sign)
+     {
+    hi = ~hi;
+    if (!lo) hi++;
+    else lo = ~lo + 1;
+     }
+}
+
+void multu64(uint64_t m1,uint64_t m2)
+{
+   unsigned long long int op1, op2, op3, op4;
+   unsigned long long int result1, result2, result3, result4;
+   unsigned long long int temp1, temp2, temp3, temp4;
+   
+   op1 = m1 & 0xFFFFFFFF;
+   op2 = (m1 >> 32) & 0xFFFFFFFF;
+   op3 = m2 & 0xFFFFFFFF;
+   op4 = (m2 >> 32) & 0xFFFFFFFF;
+   
+   temp1 = op1 * op3;
+   temp2 = (temp1 >> 32) + op1 * op4;
+   temp3 = op2 * op3;
+   temp4 = (temp3 >> 32) + op2 * op4;
+   
+   result1 = temp1 & 0xFFFFFFFF;
+   result2 = temp2 + (temp3 & 0xFFFFFFFF);
+   result3 = (result2 >> 32) + temp4;
+   result4 = (result3 >> 32);
+   
+   lo = result1 | (result2 << 32);
+   hi = (result3 & 0xFFFFFFFF) | (result4 << 32);
+   
+  //printf("TRACE: dmultu %8x%8x %8x%8x\n",(int)reg[HIREG],(int)(reg[HIREG]>>32)
+  //                                      ,(int)reg[LOREG],(int)(reg[LOREG]>>32));
+}
+
+uint64_t ldl_merge(uint64_t original,uint64_t loaded,u_int bits)
+{
+  if(bits) {
+    original<<=64-bits;
+    original>>=64-bits;
+    loaded<<=bits;
+    original|=loaded;
+  }
+  else original=loaded;
+  return original;
+}
+uint64_t ldr_merge(uint64_t original,uint64_t loaded,u_int bits)
+{
+  if(bits^56) {
+    original>>=64-(bits^56);
+    original<<=64-(bits^56);
+    loaded>>=bits^56;
+    original|=loaded;
+  }
+  else original=loaded;
+  return original;
+}
+
+#ifdef __i386__
+#include "assem_x86.c"
+#endif
+#ifdef __x86_64__
+#include "assem_x64.c"
+#endif
+#ifdef __arm__
+#include "assem_arm.c"
+#endif
+
+// Add virtual address mapping to linked list
+void ll_add(struct ll_entry **head,int vaddr,void *addr)
+{
+  struct ll_entry *new_entry;
+  new_entry=malloc(sizeof(struct ll_entry));
+  assert(new_entry!=NULL);
+  new_entry->vaddr=vaddr;
+  new_entry->reg32=0;
+  new_entry->addr=addr;
+  new_entry->next=*head;
+  *head=new_entry;
+}
+
+// Add virtual address mapping for 32-bit compiled block
+void ll_add_32(struct ll_entry **head,int vaddr,u_int reg32,void *addr)
+{
+  struct ll_entry *new_entry;
+  new_entry=malloc(sizeof(struct ll_entry));
+  assert(new_entry!=NULL);
+  new_entry->vaddr=vaddr;
+  new_entry->reg32=reg32;
+  new_entry->addr=addr;
+  new_entry->next=*head;
+  *head=new_entry;
+}
+
+// Check if an address is already compiled
+// but don't return addresses which are about to expire from the cache
+void *check_addr(u_int vaddr)
+{
+  u_int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+  if(ht_bin[0]==vaddr) {
+    if(((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
+      if(isclean(ht_bin[1])) return (void *)ht_bin[1];
+  }
+  if(ht_bin[2]==vaddr) {
+    if(((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
+      if(isclean(ht_bin[3])) return (void *)ht_bin[3];
+  }
+  u_int page=(vaddr^0x80000000)>>12;
+  if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
+  if(page>2048) page=2048+(page&2047);
+  struct ll_entry *head;
+  head=jump_in[page];
+  while(head!=NULL) {
+    if(head->vaddr==vaddr&&head->reg32==0) {
+      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+        // Update existing entry with current address
+        if(ht_bin[0]==vaddr) {
+          ht_bin[1]=(int)head->addr;
+          return head->addr;
+        }
+        if(ht_bin[2]==vaddr) {
+          ht_bin[3]=(int)head->addr;
+          return head->addr;
+        }
+        // Insert into hash table with low priority.
+        // Don't evict existing entries, as they are probably
+        // addresses that are being accessed frequently.
+        if(ht_bin[0]==-1) {
+          ht_bin[1]=(int)head->addr;
+          ht_bin[0]=vaddr;
+        }else if(ht_bin[2]==-1) {
+          ht_bin[3]=(int)head->addr;
+          ht_bin[2]=vaddr;
+        }
+        return head->addr;
+      }
+    }
+    head=head->next;
+  }
+  return 0;
+}
+
+void remove_hash(int vaddr)
+{
+  //printf("remove hash: %x\n",vaddr);
+  int *ht_bin=hash_table[(((vaddr)>>16)^vaddr)&0xFFFF];
+  if(ht_bin[2]==vaddr) {
+    ht_bin[2]=ht_bin[3]=-1;
+  }
+  if(ht_bin[0]==vaddr) {
+    ht_bin[0]=ht_bin[2];
+    ht_bin[1]=ht_bin[3];
+    ht_bin[2]=ht_bin[3]=-1;
+  }
+}
+
+void ll_remove_matching_addrs(struct ll_entry **head,int addr,int shift)
+{
+  struct ll_entry *next;
+  while(*head) {
+    if(((u_int)((*head)->addr)>>shift)==(addr>>shift) || 
+       ((u_int)((*head)->addr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift))
+    {
+      inv_debug("EXP: Remove pointer to %x (%x)\n",(int)(*head)->addr,(*head)->vaddr);
+      remove_hash((*head)->vaddr);
+      next=(*head)->next;
+      free(*head);
+      *head=next;
+    }
+    else
+    {
+      head=&((*head)->next);
+    }
+  }
+}
+
+// Remove all entries from linked list
+void ll_clear(struct ll_entry **head)
+{
+  struct ll_entry *cur;
+  struct ll_entry *next;
+  if(cur=*head) {
+    *head=0;
+    while(cur) {
+      next=cur->next;
+      free(cur);
+      cur=next;
+    }
+  }
+}
+
+// Dereference the pointers and remove if it matches
+void ll_kill_pointers(struct ll_entry *head,int addr,int shift)
+{
+  while(head) {
+    int ptr=get_pointer(head->addr);
+    inv_debug("EXP: Lookup pointer to %x at %x (%x)\n",(int)ptr,(int)head->addr,head->vaddr);
+    if(((ptr>>shift)==(addr>>shift)) ||
+       (((ptr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)))
+    {
+      inv_debug("EXP: Kill pointer at %x (%x)\n",(int)head->addr,head->vaddr);
+      kill_pointer(head->addr);
+    }
+    head=head->next;
+  }
+}
+
+// This is called when we write to a compiled block (see do_invstub)
+int invalidate_page(u_int page)
+{
+  int modified=0;
+  struct ll_entry *head;
+  struct ll_entry *next;
+  head=jump_in[page];
+  jump_in[page]=0;
+  while(head!=NULL) {
+    inv_debug("INVALIDATE: %x\n",head->vaddr);
+    remove_hash(head->vaddr);
+    next=head->next;
+    free(head);
+    head=next;
+  }
+  head=jump_out[page];
+  jump_out[page]=0;
+  while(head!=NULL) {
+    inv_debug("INVALIDATE: kill pointer to %x (%x)\n",head->vaddr,(int)head->addr);
+    kill_pointer(head->addr);
+    modified=1;
+    next=head->next;
+    free(head);
+    head=next;
+  }
+  return modified;
+}
+void invalidate_block(u_int block)
+{
+  int modified;
+  u_int page,vpage;
+  page=vpage=block^0x80000;
+  if(page>262143&&tlb_LUT_r[block]) page=(tlb_LUT_r[block]^0x80000000)>>12;
+  if(page>2048) page=2048+(page&2047);
+  if(vpage>262143&&tlb_LUT_r[block]) vpage&=2047; // jump_dirty uses a hash of the virtual address instead
+  if(vpage>2048) vpage=2048+(vpage&2047);
+  inv_debug("INVALIDATE: %x (%d)\n",block<<12,page);
+  //inv_debug("invalid_code[block]=%d\n",invalid_code[block]);
+  u_int first,last;
+  first=last=page;
+  struct ll_entry *head;
+  head=jump_dirty[vpage];
+  //printf("page=%d vpage=%d\n",page,vpage);
+  while(head!=NULL) {
+    u_int start,end;
+    if(vpage>2047||(head->vaddr>>12)==block) { // Ignore vaddr hash collision
+      get_bounds((int)head->addr,&start,&end);
+      //printf("start: %x end: %x\n",start,end);
+      if(page<2048&&start>=0x80000000&&end<0x80800000) {
+        if(((start-(u_int)rdram)>>12)<=page&&((end-1-(u_int)rdram)>>12)>=page) {
+          if((((start-(u_int)rdram)>>12)&2047)<first) first=((start-(u_int)rdram)>>12)&2047;
+          if((((end-1-(u_int)rdram)>>12)&2047)>last) last=((end-1-(u_int)rdram)>>12)&2047;
+        }
+      }
+      if(page<2048&&(signed int)start>=(signed int)0xC0000000&&(signed int)end>=(signed int)0xC0000000) {
+        if(((start+memory_map[start>>12]-(u_int)rdram)>>12)<=page&&((end-1+memory_map[(end-1)>>12]-(u_int)rdram)>>12)>=page) {
+          if((((start+memory_map[start>>12]-(u_int)rdram)>>12)&2047)<first) first=((start+memory_map[start>>12]-(u_int)rdram)>>12)&2047;
+          if((((end-1+memory_map[(end-1)>>12]-(u_int)rdram)>>12)&2047)>last) last=((end-1+memory_map[(end-1)>>12]-(u_int)rdram)>>12)&2047;
+        }
+      }
+    }
+    head=head->next;
+  }
+  //printf("first=%d last=%d\n",first,last);
+  modified=invalidate_page(page);
+  assert(first+5>page); // NB: this assumes MAXBLOCK<=4096 (4 pages)
+  assert(last<page+5);
+  // Invalidate the adjacent pages if a block crosses a 4K boundary
+  while(first<page) {
+    invalidate_page(first);
+    first++;
+  }
+  for(first=page+1;first<last;first++) {
+    invalidate_page(first);
+  }
+  
+  // Don't trap writes
+  invalid_code[block]=1;
+  // If there is a valid TLB entry for this page, remove write protect
+  if(tlb_LUT_w[block]) {
+    assert(tlb_LUT_r[block]==tlb_LUT_w[block]);
+    // CHECK: Is this right?
+    memory_map[block]=((tlb_LUT_w[block]&0xFFFFF000)-(block<<12)+(unsigned int)rdram-0x80000000)>>2;
+    u_int real_block=tlb_LUT_w[block]>>12;
+    invalid_code[real_block]=1;
+    if(real_block>=0x80000&&real_block<0x80800) memory_map[real_block]=((u_int)rdram-0x80000000)>>2;
+  }
+  else if(block>=0x80000&&block<0x80800) memory_map[block]=((u_int)rdram-0x80000000)>>2;
+  #ifdef __arm__
+  if(modified)
+    __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
+  #endif
+  #ifdef USE_MINI_HT
+  memset(mini_ht,-1,sizeof(mini_ht));
+  #endif
+}
+void invalidate_addr(u_int addr)
+{
+  invalidate_block(addr>>12);
+}
+void invalidate_all_pages()
+{
+  u_int page,n;
+  for(page=0;page<4096;page++)
+    invalidate_page(page);
+  for(page=0;page<1048576;page++)
+    if(!invalid_code[page]) {
+      restore_candidate[(page&2047)>>3]|=1<<(page&7);
+      restore_candidate[((page&2047)>>3)+256]|=1<<(page&7);
+    }
+  #ifdef __arm__
+  __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
+  #endif
+  #ifdef USE_MINI_HT
+  memset(mini_ht,-1,sizeof(mini_ht));
+  #endif
+  // TLB
+  for(page=0;page<0x100000;page++) {
+    if(tlb_LUT_r[page]) {
+      memory_map[page]=((tlb_LUT_r[page]&0xFFFFF000)-(page<<12)+(unsigned int)rdram-0x80000000)>>2;
+      if(!tlb_LUT_w[page]||!invalid_code[page])
+        memory_map[page]|=0x40000000; // Write protect
+    }
+    else memory_map[page]=-1;
+    if(page==0x80000) page=0xC0000;
+  }
+  tlb_hacks();
+}
+
+// Add an entry to jump_out after making a link
+void add_link(u_int vaddr,void *src)
+{
+  u_int page=(vaddr^0x80000000)>>12;
+  if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
+  if(page>4095) page=2048+(page&2047);
+  inv_debug("add_link: %x -> %x (%d)\n",(int)src,vaddr,page);
+  ll_add(jump_out+page,vaddr,src);
+  //int ptr=get_pointer(src);
+  //inv_debug("add_link: Pointer is to %x\n",(int)ptr);
+}
+
+// If a code block was found to be unmodified (bit was set in
+// restore_candidate) and it remains unmodified (bit is clear
+// in invalid_code) then move the entries for that 4K page from
+// the dirty list to the clean list.
+void clean_blocks(u_int page)
+{
+  struct ll_entry *head;
+  inv_debug("INV: clean_blocks page=%d\n",page);
+  head=jump_dirty[page];
+  while(head!=NULL) {
+    if(!invalid_code[head->vaddr>>12]) {
+      // Don't restore blocks which are about to expire from the cache
+      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+        u_int start,end;
+        if(verify_dirty((int)head->addr)) {
+          //printf("Possibly Restore %x (%x)\n",head->vaddr, (int)head->addr);
+          u_int i;
+          u_int inv=0;
+          get_bounds((int)head->addr,&start,&end);
+          if(start-(u_int)rdram<0x800000) {
+            for(i=(start-(u_int)rdram+0x80000000)>>12;i<=(end-1-(u_int)rdram+0x80000000)>>12;i++) {
+              inv|=invalid_code[i];
+            }
+          }
+          if((signed int)head->vaddr>=(signed int)0xC0000000) {
+            u_int addr = (head->vaddr+(memory_map[head->vaddr>>12]<<2));
+            //printf("addr=%x start=%x end=%x\n",addr,start,end);
+            if(addr<start||addr>=end) inv=1;
+          }
+          else if((signed int)head->vaddr>=(signed int)0x80800000) {
+            inv=1;
+          }
+          if(!inv) {
+            void * clean_addr=(void *)get_clean_addr((int)head->addr);
+            if((((u_int)clean_addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+              u_int ppage=page;
+              if(page<2048&&tlb_LUT_r[head->vaddr>>12]) ppage=(tlb_LUT_r[head->vaddr>>12]^0x80000000)>>12;
+              inv_debug("INV: Restored %x (%x/%x)\n",head->vaddr, (int)head->addr, (int)clean_addr);
+              //printf("page=%x, addr=%x\n",page,head->vaddr);
+              //assert(head->vaddr>>12==(page|0x80000));
+              ll_add_32(jump_in+ppage,head->vaddr,head->reg32,clean_addr);
+              int *ht_bin=hash_table[((head->vaddr>>16)^head->vaddr)&0xFFFF];
+              if(!head->reg32) {
+                if(ht_bin[0]==head->vaddr) {
+                  ht_bin[1]=(int)clean_addr; // Replace existing entry
+                }
+                if(ht_bin[2]==head->vaddr) {
+                  ht_bin[3]=(int)clean_addr; // Replace existing entry
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+    head=head->next;
+  }
+}
+
+
+void mov_alloc(struct regstat *current,int i)
+{
+  // Note: Don't need to actually alloc the source registers
+  if((~current->is32>>rs1[i])&1) {
+    //alloc_reg64(current,i,rs1[i]);
+    alloc_reg64(current,i,rt1[i]);
+    current->is32&=~(1LL<<rt1[i]);
+  } else {
+    //alloc_reg(current,i,rs1[i]);
+    alloc_reg(current,i,rt1[i]);
+    current->is32|=(1LL<<rt1[i]);
+  }
+  clear_const(current,rs1[i]);
+  clear_const(current,rt1[i]);
+  dirty_reg(current,rt1[i]);
+}
+
+void shiftimm_alloc(struct regstat *current,int i)
+{
+  clear_const(current,rs1[i]);
+  clear_const(current,rt1[i]);
+  if(opcode2[i]<=0x3) // SLL/SRL/SRA
+  {
+    if(rt1[i]) {
+      if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+      else lt1[i]=rs1[i];
+      alloc_reg(current,i,rt1[i]);
+      current->is32|=1LL<<rt1[i];
+      dirty_reg(current,rt1[i]);
+    }
+  }
+  if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA
+  {
+    if(rt1[i]) {
+      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
+      alloc_reg64(current,i,rt1[i]);
+      current->is32&=~(1LL<<rt1[i]);
+      dirty_reg(current,rt1[i]);
+    }
+  }
+  if(opcode2[i]==0x3c) // DSLL32
+  {
+    if(rt1[i]) {
+      if(rs1[i]) alloc_reg(current,i,rs1[i]);
+      alloc_reg64(current,i,rt1[i]);
+      current->is32&=~(1LL<<rt1[i]);
+      dirty_reg(current,rt1[i]);
+    }
+  }
+  if(opcode2[i]==0x3e) // DSRL32
+  {
+    if(rt1[i]) {
+      alloc_reg64(current,i,rs1[i]);
+      if(imm[i]==32) {
+        alloc_reg64(current,i,rt1[i]);
+        current->is32&=~(1LL<<rt1[i]);
+      } else {
+        alloc_reg(current,i,rt1[i]);
+        current->is32|=1LL<<rt1[i];
+      }
+      dirty_reg(current,rt1[i]);
+    }
+  }
+  if(opcode2[i]==0x3f) // DSRA32
+  {
+    if(rt1[i]) {
+      alloc_reg64(current,i,rs1[i]);
+      alloc_reg(current,i,rt1[i]);
+      current->is32|=1LL<<rt1[i];
+      dirty_reg(current,rt1[i]);
+    }
+  }
+}
+
+void shift_alloc(struct regstat *current,int i)
+{
+  if(rt1[i]) {
+    if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV
+    {
+      if(rs1[i]) alloc_reg(current,i,rs1[i]);
+      if(rs2[i]) alloc_reg(current,i,rs2[i]);
+      alloc_reg(current,i,rt1[i]);
+      if(rt1[i]==rs2[i]) alloc_reg_temp(current,i,-1);
+      current->is32|=1LL<<rt1[i];
+    } else { // DSLLV/DSRLV/DSRAV
+      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
+      if(rs2[i]) alloc_reg(current,i,rs2[i]);
+      alloc_reg64(current,i,rt1[i]);
+      current->is32&=~(1LL<<rt1[i]);
+      if(opcode2[i]==0x16||opcode2[i]==0x17) // DSRLV and DSRAV need a temporary register
+        alloc_reg_temp(current,i,-1);
+    }
+    clear_const(current,rs1[i]);
+    clear_const(current,rs2[i]);
+    clear_const(current,rt1[i]);
+    dirty_reg(current,rt1[i]);
+  }
+}
+
+void alu_alloc(struct regstat *current,int i)
+{
+  if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU
+    if(rt1[i]) {
+      if(rs1[i]&&rs2[i]) {
+        alloc_reg(current,i,rs1[i]);
+        alloc_reg(current,i,rs2[i]);
+      }
+      else {
+        if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+        if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]);
+      }
+      alloc_reg(current,i,rt1[i]);
+    }
+    current->is32|=1LL<<rt1[i];
+  }
+  if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU
+    if(rt1[i]) {
+      if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
+      {
+        alloc_reg64(current,i,rs1[i]);
+        alloc_reg64(current,i,rs2[i]);
+        alloc_reg(current,i,rt1[i]);
+      } else {
+        alloc_reg(current,i,rs1[i]);
+        alloc_reg(current,i,rs2[i]);
+        alloc_reg(current,i,rt1[i]);
+      }
+    }
+    current->is32|=1LL<<rt1[i];
+  }
+  if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR
+    if(rt1[i]) {
+      if(rs1[i]&&rs2[i]) {
+        alloc_reg(current,i,rs1[i]);
+        alloc_reg(current,i,rs2[i]);
+      }
+      else
+      {
+        if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+        if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]);
+      }
+      alloc_reg(current,i,rt1[i]);
+      if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
+      {
+        if(!((current->uu>>rt1[i])&1)) {
+          alloc_reg64(current,i,rt1[i]);
+        }
+        if(get_reg(current->regmap,rt1[i]|64)>=0) {
+          if(rs1[i]&&rs2[i]) {
+            alloc_reg64(current,i,rs1[i]);
+            alloc_reg64(current,i,rs2[i]);
+          }
+          else
+          {
+            // Is is really worth it to keep 64-bit values in registers?
+            #ifdef NATIVE_64BIT
+            if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]);
+            if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg64(current,i,rs2[i]);
+            #endif
+          }
+        }
+        current->is32&=~(1LL<<rt1[i]);
+      } else {
+        current->is32|=1LL<<rt1[i];
+      }
+    }
+  }
+  if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU
+    if(rt1[i]) {
+      if(rs1[i]&&rs2[i]) {
+        if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
+          alloc_reg64(current,i,rs1[i]);
+          alloc_reg64(current,i,rs2[i]);
+          alloc_reg64(current,i,rt1[i]);
+        } else {
+          alloc_reg(current,i,rs1[i]);
+          alloc_reg(current,i,rs2[i]);
+          alloc_reg(current,i,rt1[i]);
+        }
+      }
+      else {
+        alloc_reg(current,i,rt1[i]);
+        if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
+          // DADD used as move, or zeroing
+          // If we have a 64-bit source, then make the target 64 bits too
+          if(rs1[i]&&!((current->is32>>rs1[i])&1)) {
+            if(get_reg(current->regmap,rs1[i])>=0) alloc_reg64(current,i,rs1[i]);
+            alloc_reg64(current,i,rt1[i]);
+          } else if(rs2[i]&&!((current->is32>>rs2[i])&1)) {
+            if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]);
+            alloc_reg64(current,i,rt1[i]);
+          }
+          if(opcode2[i]>=0x2e&&rs2[i]) {
+            // DSUB used as negation - 64-bit result
+            // If we have a 32-bit register, extend it to 64 bits
+            if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]);
+            alloc_reg64(current,i,rt1[i]);
+          }
+        }
+      }
+      if(rs1[i]&&rs2[i]) {
+        current->is32&=~(1LL<<rt1[i]);
+      } else if(rs1[i]) {
+        current->is32&=~(1LL<<rt1[i]);
+        if((current->is32>>rs1[i])&1)
+          current->is32|=1LL<<rt1[i];
+      } else if(rs2[i]) {
+        current->is32&=~(1LL<<rt1[i]);
+        if((current->is32>>rs2[i])&1)
+          current->is32|=1LL<<rt1[i];
+      } else {
+        current->is32|=1LL<<rt1[i];
+      }
+    }
+  }
+  clear_const(current,rs1[i]);
+  clear_const(current,rs2[i]);
+  clear_const(current,rt1[i]);
+  dirty_reg(current,rt1[i]);
+}
+
+void imm16_alloc(struct regstat *current,int i)
+{
+  if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+  else lt1[i]=rs1[i];
+  if(rt1[i]) alloc_reg(current,i,rt1[i]);
+  if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU
+    current->is32&=~(1LL<<rt1[i]);
+    if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
+      // TODO: Could preserve the 32-bit flag if the immediate is zero
+      alloc_reg64(current,i,rt1[i]);
+      alloc_reg64(current,i,rs1[i]);
+    }
+    clear_const(current,rs1[i]);
+    clear_const(current,rt1[i]);
+  }
+  else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU
+    if((~current->is32>>rs1[i])&1) alloc_reg64(current,i,rs1[i]);
+    current->is32|=1LL<<rt1[i];
+    clear_const(current,rs1[i]);
+    clear_const(current,rt1[i]);
+  }
+  else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI
+    if(((~current->is32>>rs1[i])&1)&&opcode[i]>0x0c) {
+      if(rs1[i]!=rt1[i]) {
+        if(needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]);
+        alloc_reg64(current,i,rt1[i]);
+        current->is32&=~(1LL<<rt1[i]);
+      }
+    }
+    else current->is32|=1LL<<rt1[i]; // ANDI clears upper bits
+    if(is_const(current,rs1[i])) {
+      int v=get_const(current,rs1[i]);
+      if(opcode[i]==0x0c) set_const(current,rt1[i],v&imm[i]);
+      if(opcode[i]==0x0d) set_const(current,rt1[i],v|imm[i]);
+      if(opcode[i]==0x0e) set_const(current,rt1[i],v^imm[i]);
+    }
+    else clear_const(current,rt1[i]);
+  }
+  else if(opcode[i]==0x08||opcode[i]==0x09) { // ADDI/ADDIU
+    if(is_const(current,rs1[i])) {
+      int v=get_const(current,rs1[i]);
+      set_const(current,rt1[i],v+imm[i]);
+    }
+    else clear_const(current,rt1[i]);
+    current->is32|=1LL<<rt1[i];
+  }
+  else {
+    set_const(current,rt1[i],((long long)((short)imm[i]))<<16); // LUI
+    current->is32|=1LL<<rt1[i];
+  }
+  dirty_reg(current,rt1[i]);
+}
+
+void load_alloc(struct regstat *current,int i)
+{
+  clear_const(current,rt1[i]);
+  //if(rs1[i]!=rt1[i]&&needed_again(rs1[i],i)) clear_const(current,rs1[i]); // Does this help or hurt?
+  if(!rs1[i]) current->u&=~1LL; // Allow allocating r0 if it's the source register
+  if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+  if(rt1[i]) {
+    alloc_reg(current,i,rt1[i]);
+    if(opcode[i]==0x27||opcode[i]==0x37) // LWU/LD
+    {
+      current->is32&=~(1LL<<rt1[i]);
+      alloc_reg64(current,i,rt1[i]);
+    }
+    else if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
+    {
+      current->is32&=~(1LL<<rt1[i]);
+      alloc_reg64(current,i,rt1[i]);
+      alloc_all(current,i);
+      alloc_reg64(current,i,FTEMP);
+    }
+    else current->is32|=1LL<<rt1[i];
+    dirty_reg(current,rt1[i]);
+    // If using TLB, need a register for pointer to the mapping table
+    if(using_tlb) alloc_reg(current,i,TLREG);
+    // LWL/LWR need a temporary register for the old value
+    if(opcode[i]==0x22||opcode[i]==0x26)
+    {
+      alloc_reg(current,i,FTEMP);
+      alloc_reg_temp(current,i,-1);
+    }
+  }
+  else
+  {
+    // Load to r0 (dummy load)
+    // but we still need a register to calculate the address
+    alloc_reg_temp(current,i,-1);
+  }
+}
+
+void store_alloc(struct regstat *current,int i)
+{
+  clear_const(current,rs2[i]);
+  if(!(rs2[i])) current->u&=~1LL; // Allow allocating r0 if necessary
+  if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+  alloc_reg(current,i,rs2[i]);
+  if(opcode[i]==0x2c||opcode[i]==0x2d||opcode[i]==0x3f) { // 64-bit SDL/SDR/SD
+    alloc_reg64(current,i,rs2[i]);
+    if(rs2[i]) alloc_reg(current,i,FTEMP);
+  }
+  // If using TLB, need a register for pointer to the mapping table
+  if(using_tlb) alloc_reg(current,i,TLREG);
+  #if defined(HOST_IMM8)
+  // On CPUs without 32-bit immediates we need a pointer to invalid_code
+  else alloc_reg(current,i,INVCP);
+  #endif
+  if(opcode[i]==0x2c||opcode[i]==0x2d) { // 64-bit SDL/SDR
+    alloc_reg(current,i,FTEMP);
+  }
+  // We need a temporary register for address generation
+  alloc_reg_temp(current,i,-1);
+}
+
+void c1ls_alloc(struct regstat *current,int i)
+{
+  //clear_const(current,rs1[i]); // FIXME
+  clear_const(current,rt1[i]);
+  if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
+  alloc_reg(current,i,CSREG); // Status
+  alloc_reg(current,i,FTEMP);
+  if(opcode[i]==0x35||opcode[i]==0x3d) { // 64-bit LDC1/SDC1
+    alloc_reg64(current,i,FTEMP);
+  }
+  // If using TLB, need a register for pointer to the mapping table
+  if(using_tlb) alloc_reg(current,i,TLREG);
+  #if defined(HOST_IMM8)
+  // On CPUs without 32-bit immediates we need a pointer to invalid_code
+  else if((opcode[i]&0x3b)==0x39) // SWC1/SDC1
+    alloc_reg(current,i,INVCP);
+  #endif
+  // We need a temporary register for address generation
+  alloc_reg_temp(current,i,-1);
+}
+
+#ifndef multdiv_alloc
+void multdiv_alloc(struct regstat *current,int i)
+{
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  //  case 0x1C: DMULT
+  //  case 0x1D: DMULTU
+  //  case 0x1E: DDIV
+  //  case 0x1F: DDIVU
+  clear_const(current,rs1[i]);
+  clear_const(current,rs2[i]);
+  if(rs1[i]&&rs2[i])
+  {
+    if((opcode2[i]&4)==0) // 32-bit
+    {
+      current->u&=~(1LL<<HIREG);
+      current->u&=~(1LL<<LOREG);
+      alloc_reg(current,i,HIREG);
+      alloc_reg(current,i,LOREG);
+      alloc_reg(current,i,rs1[i]);
+      alloc_reg(current,i,rs2[i]);
+      current->is32|=1LL<<HIREG;
+      current->is32|=1LL<<LOREG;
+      dirty_reg(current,HIREG);
+      dirty_reg(current,LOREG);
+    }
+    else // 64-bit
+    {
+      current->u&=~(1LL<<HIREG);
+      current->u&=~(1LL<<LOREG);
+      current->uu&=~(1LL<<HIREG);
+      current->uu&=~(1LL<<LOREG);
+      alloc_reg64(current,i,HIREG);
+      //if(HOST_REGS>10) alloc_reg64(current,i,LOREG);
+      alloc_reg64(current,i,rs1[i]);
+      alloc_reg64(current,i,rs2[i]);
+      alloc_all(current,i);
+      current->is32&=~(1LL<<HIREG);
+      current->is32&=~(1LL<<LOREG);
+      dirty_reg(current,HIREG);
+      dirty_reg(current,LOREG);
+    }
+  }
+  else
+  {
+    // Multiply by zero is zero.
+    // MIPS does not have a divide by zero exception.
+    // The result is undefined, we return zero.
+    alloc_reg(current,i,HIREG);
+    alloc_reg(current,i,LOREG);
+    current->is32|=1LL<<HIREG;
+    current->is32|=1LL<<LOREG;
+    dirty_reg(current,HIREG);
+    dirty_reg(current,LOREG);
+  }
+}
+#endif
+
+void cop0_alloc(struct regstat *current,int i)
+{
+  if(opcode2[i]==0) // MFC0
+  {
+    if(rt1[i]) {
+      clear_const(current,rt1[i]);
+      alloc_all(current,i);
+      alloc_reg(current,i,rt1[i]);
+      current->is32|=1LL<<rt1[i];
+      dirty_reg(current,rt1[i]);
+    }
+  }
+  else if(opcode2[i]==4) // MTC0
+  {
+    if(rs1[i]){
+      clear_const(current,rs1[i]);
+      alloc_reg(current,i,rs1[i]);
+      alloc_all(current,i);
+    }
+    else {
+      alloc_all(current,i); // FIXME: Keep r0
+      current->u&=~1LL;
+      alloc_reg(current,i,0);
+    }
+  }
+  else
+  {
+    // TLBR/TLBWI/TLBWR/TLBP/ERET
+    assert(opcode2[i]==0x10);
+    alloc_all(current,i);
+  }
+}
+
+void cop1_alloc(struct regstat *current,int i)
+{
+  alloc_reg(current,i,CSREG); // Load status
+  if(opcode2[i]<3) // MFC1/DMFC1/CFC1
+  {
+    assert(rt1[i]);
+    clear_const(current,rt1[i]);
+    if(opcode2[i]==1) {
+      alloc_reg64(current,i,rt1[i]); // DMFC1
+      current->is32&=~(1LL<<rt1[i]);
+    }else{
+      alloc_reg(current,i,rt1[i]); // MFC1/CFC1
+      current->is32|=1LL<<rt1[i];
+    }
+    dirty_reg(current,rt1[i]);
+    alloc_reg_temp(current,i,-1);
+  }
+  else if(opcode2[i]>3) // MTC1/DMTC1/CTC1
+  {
+    if(rs1[i]){
+      clear_const(current,rs1[i]);
+      if(opcode2[i]==5)
+        alloc_reg64(current,i,rs1[i]); // DMTC1
+      else
+        alloc_reg(current,i,rs1[i]); // MTC1/CTC1
+      alloc_reg_temp(current,i,-1);
+    }
+    else {
+      current->u&=~1LL;
+      alloc_reg(current,i,0);
+      alloc_reg_temp(current,i,-1);
+    }
+  }
+}
+void fconv_alloc(struct regstat *current,int i)
+{
+  alloc_reg(current,i,CSREG); // Load status
+  alloc_reg_temp(current,i,-1);
+}
+void float_alloc(struct regstat *current,int i)
+{
+  alloc_reg(current,i,CSREG); // Load status
+  alloc_reg_temp(current,i,-1);
+}
+void fcomp_alloc(struct regstat *current,int i)
+{
+  alloc_reg(current,i,CSREG); // Load status
+  alloc_reg(current,i,FSREG); // Load flags
+  dirty_reg(current,FSREG); // Flag will be modified
+  alloc_reg_temp(current,i,-1);
+}
+
+void syscall_alloc(struct regstat *current,int i)
+{
+  alloc_cc(current,i);
+  dirty_reg(current,CCREG);
+  alloc_all(current,i);
+  current->isconst=0;
+}
+
+void delayslot_alloc(struct regstat *current,int i)
+{
+  switch(itype[i]) {
+    case UJUMP:
+    case CJUMP:
+    case SJUMP:
+    case RJUMP:
+    case FJUMP:
+    case SYSCALL:
+    case SPAN:
+      assem_debug("jump in the delay slot.  this shouldn't happen.\n");//exit(1);
+      printf("Disabled speculative precompilation\n");
+      stop_after_jal=1;
+      break;
+    case IMM16:
+      imm16_alloc(current,i);
+      break;
+    case LOAD:
+    case LOADLR:
+      load_alloc(current,i);
+      break;
+    case STORE:
+    case STORELR:
+      store_alloc(current,i);
+      break;
+    case ALU:
+      alu_alloc(current,i);
+      break;
+    case SHIFT:
+      shift_alloc(current,i);
+      break;
+    case MULTDIV:
+      multdiv_alloc(current,i);
+      break;
+    case SHIFTIMM:
+      shiftimm_alloc(current,i);
+      break;
+    case MOV:
+      mov_alloc(current,i);
+      break;
+    case COP0:
+      cop0_alloc(current,i);
+      break;
+    case COP1:
+      cop1_alloc(current,i);
+      break;
+    case C1LS:
+      c1ls_alloc(current,i);
+      break;
+    case FCONV:
+      fconv_alloc(current,i);
+      break;
+    case FLOAT:
+      float_alloc(current,i);
+      break;
+    case FCOMP:
+      fcomp_alloc(current,i);
+      break;
+  }
+}
+
+// Special case where a branch and delay slot span two pages in virtual memory
+static void pagespan_alloc(struct regstat *current,int i)
+{
+  current->isconst=0;
+  current->wasconst=0;
+  regs[i].wasconst=0;
+  alloc_all(current,i);
+  alloc_cc(current,i);
+  dirty_reg(current,CCREG);
+  if(opcode[i]==3) // JAL
+  {
+    alloc_reg(current,i,31);
+    dirty_reg(current,31);
+  }
+  if(opcode[i]==0&&(opcode2[i]&0x3E)==8) // JR/JALR
+  {
+    alloc_reg(current,i,rs1[i]);
+    if (rt1[i]==31) {
+      alloc_reg(current,i,31);
+      dirty_reg(current,31);
+    }
+  }
+  if((opcode[i]&0x2E)==4) // BEQ/BNE/BEQL/BNEL
+  {
+    if(rs1[i]) alloc_reg(current,i,rs1[i]);
+    if(rs2[i]) alloc_reg(current,i,rs2[i]);
+    if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
+    {
+      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
+      if(rs2[i]) alloc_reg64(current,i,rs2[i]);
+    }
+  }
+  else
+  if((opcode[i]&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL
+  {
+    if(rs1[i]) alloc_reg(current,i,rs1[i]);
+    if(!((current->is32>>rs1[i])&1))
+    {
+      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
+    }
+  }
+  else
+  if(opcode[i]==0x11) // BC1
+  {
+    alloc_reg(current,i,FSREG);
+    alloc_reg(current,i,CSREG);
+  }
+  //else ...
+}
+
+add_stub(int type,int addr,int retaddr,int a,int b,int c,int d,int e)
+{
+  stubs[stubcount][0]=type;
+  stubs[stubcount][1]=addr;
+  stubs[stubcount][2]=retaddr;
+  stubs[stubcount][3]=a;
+  stubs[stubcount][4]=b;
+  stubs[stubcount][5]=c;
+  stubs[stubcount][6]=d;
+  stubs[stubcount][7]=e;
+  stubcount++;
+}
+
+// Write out a single register
+void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32)
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if((regmap[hr]&63)==r) {
+        if((dirty>>hr)&1) {
+          if(regmap[hr]<64) {
+            emit_storereg(r,hr);
+            if((is32>>regmap[hr])&1) {
+              emit_sarimm(hr,31,hr);
+              emit_storereg(r|64,hr);
+            }
+          }else{
+            emit_storereg(r|64,hr);
+          }
+        }
+      }
+    }
+  }
+}
+
+int mchecksum()
+{
+  //if(!tracedebug) return 0;
+  int i;
+  int sum=0;
+  for(i=0;i<2097152;i++) {
+    unsigned int temp=sum;
+    sum<<=1;
+    sum|=(~temp)>>31;
+    sum^=((u_int *)rdram)[i];
+  }
+  return sum;
+}
+int rchecksum()
+{
+  int i;
+  int sum=0;
+  for(i=0;i<64;i++)
+    sum^=((u_int *)reg)[i];
+  return sum;
+}
+int fchecksum()
+{
+  int i;
+  int sum=0;
+  for(i=0;i<64;i++)
+    sum^=((u_int *)reg_cop1_fgr_64)[i];
+  return sum;
+}
+void rlist()
+{
+  int i;
+  printf("TRACE: ");
+  for(i=0;i<32;i++)
+    printf("r%d:%8x%8x ",i,((int *)(reg+i))[1],((int *)(reg+i))[0]);
+  printf("\n");
+  printf("TRACE: ");
+  for(i=0;i<32;i++)
+    printf("f%d:%8x%8x ",i,((int*)reg_cop1_simple[i])[1],*((int*)reg_cop1_simple[i]));
+  printf("\n");
+}
+
+void enabletrace()
+{
+  tracedebug=1;
+}
+
+void memdebug(int i)
+{
+  //printf("TRACE: count=%d next=%d (checksum %x) lo=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[LOREG]>>32),(int)reg[LOREG]);
+  //printf("TRACE: count=%d next=%d (rchecksum %x)\n",Count,next_interupt,rchecksum());
+  //rlist();
+  //if(tracedebug) {
+  //if(Count>=-2084597794) {
+  if((signed int)Count>=-2084597794&&(signed int)Count<0) {
+  //if(0) {
+    printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum());
+    //printf("TRACE: count=%d next=%d (checksum %x) Status=%x\n",Count,next_interupt,mchecksum(),Status);
+    //printf("TRACE: count=%d next=%d (checksum %x) hi=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[HIREG]>>32),(int)reg[HIREG]);
+    rlist();
+    #ifdef __i386__
+    printf("TRACE: %x\n",(&i)[-1]);
+    #endif
+    #ifdef __arm__
+    int j;
+    printf("TRACE: %x \n",(&j)[10]);
+    printf("TRACE: %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x\n",(&j)[1],(&j)[2],(&j)[3],(&j)[4],(&j)[5],(&j)[6],(&j)[7],(&j)[8],(&j)[9],(&j)[10],(&j)[11],(&j)[12],(&j)[13],(&j)[14],(&j)[15],(&j)[16],(&j)[17],(&j)[18],(&j)[19],(&j)[20]);
+    #endif
+    //fflush(stdout);
+  }
+  //printf("TRACE: %x\n",(&i)[-1]);
+}
+
+void tlb_debug(u_int cause, u_int addr, u_int iaddr)
+{
+  printf("TLB Exception: instruction=%x addr=%x cause=%x\n",iaddr, addr, cause);
+}
+
+void alu_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU
+    if(rt1[i]) {
+      signed char s1,s2,t;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      if(t>=0) {
+        s1=get_reg(i_regs->regmap,rs1[i]);
+        s2=get_reg(i_regs->regmap,rs2[i]);
+        if(rs1[i]&&rs2[i]) {
+          assert(s1>=0);
+          assert(s2>=0);
+          if(opcode2[i]&2) emit_sub(s1,s2,t);
+          else emit_add(s1,s2,t);
+        }
+        else if(rs1[i]) {
+          if(s1>=0) emit_mov(s1,t);
+          else emit_loadreg(rs1[i],t);
+        }
+        else if(rs2[i]) {
+          if(s2>=0) {
+            if(opcode2[i]&2) emit_neg(s2,t);
+            else emit_mov(s2,t);
+          }
+          else {
+            emit_loadreg(rs2[i],t);
+            if(opcode2[i]&2) emit_neg(t,t);
+          }
+        }
+        else emit_zeroreg(t);
+      }
+    }
+  }
+  if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU
+    if(rt1[i]) {
+      signed char s1l,s2l,s1h,s2h,tl,th;
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      if(tl>=0) {
+        s1l=get_reg(i_regs->regmap,rs1[i]);
+        s2l=get_reg(i_regs->regmap,rs2[i]);
+        s1h=get_reg(i_regs->regmap,rs1[i]|64);
+        s2h=get_reg(i_regs->regmap,rs2[i]|64);
+        if(rs1[i]&&rs2[i]) {
+          assert(s1l>=0);
+          assert(s2l>=0);
+          if(opcode2[i]&2) emit_subs(s1l,s2l,tl);
+          else emit_adds(s1l,s2l,tl);
+          if(th>=0) {
+            #ifdef INVERTED_CARRY
+            if(opcode2[i]&2) {if(s1h!=th) emit_mov(s1h,th);emit_sbb(th,s2h);}
+            #else
+            if(opcode2[i]&2) emit_sbc(s1h,s2h,th);
+            #endif
+            else emit_add(s1h,s2h,th);
+          }
+        }
+        else if(rs1[i]) {
+          if(s1l>=0) emit_mov(s1l,tl);
+          else emit_loadreg(rs1[i],tl);
+          if(th>=0) {
+            if(s1h>=0) emit_mov(s1h,th);
+            else emit_loadreg(rs1[i]|64,th);
+          }
+        }
+        else if(rs2[i]) {
+          if(s2l>=0) {
+            if(opcode2[i]&2) emit_negs(s2l,tl);
+            else emit_mov(s2l,tl);
+          }
+          else {
+            emit_loadreg(rs2[i],tl);
+            if(opcode2[i]&2) emit_negs(tl,tl);
+          }
+          if(th>=0) {
+            #ifdef INVERTED_CARRY
+            if(s2h>=0) emit_mov(s2h,th);
+            else emit_loadreg(rs2[i]|64,th);
+            if(opcode2[i]&2) {
+              emit_adcimm(-1,th); // x86 has inverted carry flag
+              emit_not(th,th);
+            }
+            #else
+            if(opcode2[i]&2) {
+              if(s2h>=0) emit_rscimm(s2h,0,th);
+              else {
+                emit_loadreg(rs2[i]|64,th);
+                emit_rscimm(th,0,th);
+              }
+            }else{
+              if(s2h>=0) emit_mov(s2h,th);
+              else emit_loadreg(rs2[i]|64,th);
+            }
+            #endif
+          }
+        }
+        else {
+          emit_zeroreg(tl);
+          if(th>=0) emit_zeroreg(th);
+        }
+      }
+    }
+  }
+  if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU
+    if(rt1[i]) {
+      signed char s1l,s1h,s2l,s2h,t;
+      if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1))
+      {
+        t=get_reg(i_regs->regmap,rt1[i]);
+        //assert(t>=0);
+        if(t>=0) {
+          s1l=get_reg(i_regs->regmap,rs1[i]);
+          s1h=get_reg(i_regs->regmap,rs1[i]|64);
+          s2l=get_reg(i_regs->regmap,rs2[i]);
+          s2h=get_reg(i_regs->regmap,rs2[i]|64);
+          if(rs2[i]==0) // rx<r0
+          {
+            assert(s1h>=0);
+            if(opcode2[i]==0x2a) // SLT
+              emit_shrimm(s1h,31,t);
+            else // SLTU (unsigned can not be less than zero)
+              emit_zeroreg(t);
+          }
+          else if(rs1[i]==0) // r0<rx
+          {
+            assert(s2h>=0);
+            if(opcode2[i]==0x2a) // SLT
+              emit_set_gz64_32(s2h,s2l,t);
+            else // SLTU (set if not zero)
+              emit_set_nz64_32(s2h,s2l,t);
+          }
+          else {
+            assert(s1l>=0);assert(s1h>=0);
+            assert(s2l>=0);assert(s2h>=0);
+            if(opcode2[i]==0x2a) // SLT
+              emit_set_if_less64_32(s1h,s1l,s2h,s2l,t);
+            else // SLTU
+              emit_set_if_carry64_32(s1h,s1l,s2h,s2l,t);
+          }
+        }
+      } else {
+        t=get_reg(i_regs->regmap,rt1[i]);
+        //assert(t>=0);
+        if(t>=0) {
+          s1l=get_reg(i_regs->regmap,rs1[i]);
+          s2l=get_reg(i_regs->regmap,rs2[i]);
+          if(rs2[i]==0) // rx<r0
+          {
+            assert(s1l>=0);
+            if(opcode2[i]==0x2a) // SLT
+              emit_shrimm(s1l,31,t);
+            else // SLTU (unsigned can not be less than zero)
+              emit_zeroreg(t);
+          }
+          else if(rs1[i]==0) // r0<rx
+          {
+            assert(s2l>=0);
+            if(opcode2[i]==0x2a) // SLT
+              emit_set_gz32(s2l,t);
+            else // SLTU (set if not zero)
+              emit_set_nz32(s2l,t);
+          }
+          else{
+            assert(s1l>=0);assert(s2l>=0);
+            if(opcode2[i]==0x2a) // SLT
+              emit_set_if_less32(s1l,s2l,t);
+            else // SLTU
+              emit_set_if_carry32(s1l,s2l,t);
+          }
+        }
+      }
+    }
+  }
+  if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR
+    if(rt1[i]) {
+      signed char s1l,s1h,s2l,s2h,th,tl;
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1)&&th>=0)
+      {
+        assert(tl>=0);
+        if(tl>=0) {
+          s1l=get_reg(i_regs->regmap,rs1[i]);
+          s1h=get_reg(i_regs->regmap,rs1[i]|64);
+          s2l=get_reg(i_regs->regmap,rs2[i]);
+          s2h=get_reg(i_regs->regmap,rs2[i]|64);
+          if(rs1[i]&&rs2[i]) {
+            assert(s1l>=0);assert(s1h>=0);
+            assert(s2l>=0);assert(s2h>=0);
+            if(opcode2[i]==0x24) { // AND
+              emit_and(s1l,s2l,tl);
+              emit_and(s1h,s2h,th);
+            } else
+            if(opcode2[i]==0x25) { // OR
+              emit_or(s1l,s2l,tl);
+              emit_or(s1h,s2h,th);
+            } else
+            if(opcode2[i]==0x26) { // XOR
+              emit_xor(s1l,s2l,tl);
+              emit_xor(s1h,s2h,th);
+            } else
+            if(opcode2[i]==0x27) { // NOR
+              emit_or(s1l,s2l,tl);
+              emit_or(s1h,s2h,th);
+              emit_not(tl,tl);
+              emit_not(th,th);
+            }
+          }
+          else
+          {
+            if(opcode2[i]==0x24) { // AND
+              emit_zeroreg(tl);
+              emit_zeroreg(th);
+            } else
+            if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR
+              if(rs1[i]){
+                if(s1l>=0) emit_mov(s1l,tl);
+                else emit_loadreg(rs1[i],tl);
+                if(s1h>=0) emit_mov(s1h,th);
+                else emit_loadreg(rs1[i]|64,th);
+              }
+              else
+              if(rs2[i]){
+                if(s2l>=0) emit_mov(s2l,tl);
+                else emit_loadreg(rs2[i],tl);
+                if(s2h>=0) emit_mov(s2h,th);
+                else emit_loadreg(rs2[i]|64,th);
+              }
+              else{
+                emit_zeroreg(tl);
+                emit_zeroreg(th);
+              }
+            } else
+            if(opcode2[i]==0x27) { // NOR
+              if(rs1[i]){
+                if(s1l>=0) emit_not(s1l,tl);
+                else{
+                  emit_loadreg(rs1[i],tl);
+                  emit_not(tl,tl);
+                }
+                if(s1h>=0) emit_not(s1h,th);
+                else{
+                  emit_loadreg(rs1[i]|64,th);
+                  emit_not(th,th);
+                }
+              }
+              else
+              if(rs2[i]){
+                if(s2l>=0) emit_not(s2l,tl);
+                else{
+                  emit_loadreg(rs2[i],tl);
+                  emit_not(tl,tl);
+                }
+                if(s2h>=0) emit_not(s2h,th);
+                else{
+                  emit_loadreg(rs2[i]|64,th);
+                  emit_not(th,th);
+                }
+              }
+              else {
+                emit_movimm(-1,tl);
+                emit_movimm(-1,th);
+              }
+            }
+          }
+        }
+      }
+      else
+      {
+        // 32 bit
+        if(tl>=0) {
+          s1l=get_reg(i_regs->regmap,rs1[i]);
+          s2l=get_reg(i_regs->regmap,rs2[i]);
+          if(rs1[i]&&rs2[i]) {
+            assert(s1l>=0);
+            assert(s2l>=0);
+            if(opcode2[i]==0x24) { // AND
+              emit_and(s1l,s2l,tl);
+            } else
+            if(opcode2[i]==0x25) { // OR
+              emit_or(s1l,s2l,tl);
+            } else
+            if(opcode2[i]==0x26) { // XOR
+              emit_xor(s1l,s2l,tl);
+            } else
+            if(opcode2[i]==0x27) { // NOR
+              emit_or(s1l,s2l,tl);
+              emit_not(tl,tl);
+            }
+          }
+          else
+          {
+            if(opcode2[i]==0x24) { // AND
+              emit_zeroreg(tl);
+            } else
+            if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR
+              if(rs1[i]){
+                if(s1l>=0) emit_mov(s1l,tl);
+                else emit_loadreg(rs1[i],tl); // CHECK: regmap_entry?
+              }
+              else
+              if(rs2[i]){
+                if(s2l>=0) emit_mov(s2l,tl);
+                else emit_loadreg(rs2[i],tl); // CHECK: regmap_entry?
+              }
+              else emit_zeroreg(tl);
+            } else
+            if(opcode2[i]==0x27) { // NOR
+              if(rs1[i]){
+                if(s1l>=0) emit_not(s1l,tl);
+                else {
+                  emit_loadreg(rs1[i],tl);
+                  emit_not(tl,tl);
+                }
+              }
+              else
+              if(rs2[i]){
+                if(s2l>=0) emit_not(s2l,tl);
+                else {
+                  emit_loadreg(rs2[i],tl);
+                  emit_not(tl,tl);
+                }
+              }
+              else emit_movimm(-1,tl);
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+void imm16_assemble(int i,struct regstat *i_regs)
+{
+  if (opcode[i]==0x0f) { // LUI
+    if(rt1[i]) {
+      signed char t;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      //assert(t>=0);
+      if(t>=0) {
+        if(!((i_regs->isconst>>t)&1))
+          emit_movimm(imm[i]<<16,t);
+      }
+    }
+  }
+  if(opcode[i]==0x08||opcode[i]==0x09) { // ADDI/ADDIU
+    if(rt1[i]) {
+      signed char s,t;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      s=get_reg(i_regs->regmap,rs1[i]);
+      if(rs1[i]) {
+        //assert(t>=0);
+        //assert(s>=0);
+        if(t>=0) {
+          if(!((i_regs->isconst>>t)&1)) {
+            if(s<0) {
+              if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t);
+              emit_addimm(t,imm[i],t);
+            }else{
+              if(!((i_regs->wasconst>>s)&1))
+                emit_addimm(s,imm[i],t);
+              else
+                emit_movimm(constmap[i][s]+imm[i],t);
+            }
+          }
+        }
+      } else {
+        if(t>=0) {
+          if(!((i_regs->isconst>>t)&1))
+            emit_movimm(imm[i],t);
+        }
+      }
+    }
+  }
+  if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU
+    if(rt1[i]) {
+      signed char sh,sl,th,tl;
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      if(tl>=0) {
+        if(rs1[i]) {
+          assert(sh>=0);
+          assert(sl>=0);
+          if(th>=0) {
+            emit_addimm64_32(sh,sl,imm[i],th,tl);
+          }
+          else {
+            emit_addimm(sl,imm[i],tl);
+          }
+        } else {
+          emit_movimm(imm[i],tl);
+          if(th>=0) emit_movimm(((signed int)imm[i])>>31,th);
+        }
+      }
+    }
+  }
+  else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU
+    if(rt1[i]) {
+      //assert(rs1[i]!=0); // r0 might be valid, but it's probably a bug
+      signed char sh,sl,t;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      //assert(t>=0);
+      if(t>=0) {
+        if(rs1[i]>0) {
+          if(sh<0) assert((i_regs->was32>>rs1[i])&1);
+          if(sh<0||((i_regs->was32>>rs1[i])&1)) {
+            if(opcode[i]==0x0a) { // SLTI
+              if(sl<0) {
+                if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t);
+                emit_slti32(t,imm[i],t);
+              }else{
+                emit_slti32(sl,imm[i],t);
+              }
+            }
+            else { // SLTIU
+              if(sl<0) {
+                if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t);
+                emit_sltiu32(t,imm[i],t);
+              }else{
+                emit_sltiu32(sl,imm[i],t);
+              }
+            }
+          }else{ // 64-bit
+            assert(sl>=0);
+            if(opcode[i]==0x0a) // SLTI
+              emit_slti64_32(sh,sl,imm[i],t);
+            else // SLTIU
+              emit_sltiu64_32(sh,sl,imm[i],t);
+          }
+        }else{
+          // SLTI(U) with r0 is just stupid,
+          // nonetheless examples can be found
+          if(opcode[i]==0x0a) // SLTI
+            if(0<imm[i]) emit_movimm(1,t);
+            else emit_zeroreg(t);
+          else // SLTIU
+          {
+            if(imm[i]) emit_movimm(1,t);
+            else emit_zeroreg(t);
+          }
+        }
+      }
+    }
+  }
+  else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI
+    if(rt1[i]) {
+      signed char sh,sl,th,tl;
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      if(tl>=0 && !((i_regs->isconst>>tl)&1)) {
+        if(opcode[i]==0x0c) //ANDI
+        {
+          if(rs1[i]) {
+            if(sl<0) {
+              if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl);
+              emit_andimm(tl,imm[i],tl);
+            }else{
+              if(!((i_regs->wasconst>>sl)&1))
+                emit_andimm(sl,imm[i],tl);
+              else
+                emit_movimm(constmap[i][sl]&imm[i],tl);
+            }
+          }
+          else
+            emit_zeroreg(tl);
+          if(th>=0) emit_zeroreg(th);
+        }
+        else
+        {
+          if(rs1[i]) {
+            if(sl<0) {
+              if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl);
+            }
+            if(th>=0) {
+              if(sh<0) {
+                emit_loadreg(rs1[i]|64,th);
+              }else{
+                emit_mov(sh,th);
+              }
+            }
+            if(opcode[i]==0x0d) //ORI
+            if(sl<0) {
+              emit_orimm(tl,imm[i],tl);
+            }else{
+              if(!((i_regs->wasconst>>sl)&1))
+                emit_orimm(sl,imm[i],tl);
+              else
+                emit_movimm(constmap[i][sl]|imm[i],tl);
+            }
+            if(opcode[i]==0x0e) //XORI
+            if(sl<0) {
+              emit_xorimm(tl,imm[i],tl);
+            }else{
+              if(!((i_regs->wasconst>>sl)&1))
+                emit_xorimm(sl,imm[i],tl);
+              else
+                emit_movimm(constmap[i][sl]^imm[i],tl);
+            }
+          }
+          else {
+            emit_movimm(imm[i],tl);
+            if(th>=0) emit_zeroreg(th);
+          }
+        }
+      }
+    }
+  }
+}
+
+void shiftimm_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]<=0x3) // SLL/SRL/SRA
+  {
+    if(rt1[i]) {
+      signed char s,t;
+      t=get_reg(i_regs->regmap,rt1[i]);
+      s=get_reg(i_regs->regmap,rs1[i]);
+      //assert(t>=0);
+      if(t>=0){
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(t);
+        }
+        else
+        {
+          if(s<0&&i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t);
+          if(imm[i]) {
+            if(opcode2[i]==0) // SLL
+            {
+              emit_shlimm(s<0?t:s,imm[i],t);
+            }
+            if(opcode2[i]==2) // SRL
+            {
+              emit_shrimm(s<0?t:s,imm[i],t);
+            }
+            if(opcode2[i]==3) // SRA
+            {
+              emit_sarimm(s<0?t:s,imm[i],t);
+            }
+          }else{
+            // Shift by zero
+            if(s>=0 && s!=t) emit_mov(s,t);
+          }
+        }
+      }
+      //emit_storereg(rt1[i],t); //DEBUG
+    }
+  }
+  if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA
+  {
+    if(rt1[i]) {
+      signed char sh,sl,th,tl;
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      if(tl>=0) {
+        if(rs1[i]==0)
+        {
+          emit_zeroreg(tl);
+          if(th>=0) emit_zeroreg(th);
+        }
+        else
+        {
+          assert(sl>=0);
+          assert(sh>=0);
+          if(imm[i]) {
+            if(opcode2[i]==0x38) // DSLL
+            {
+              if(th>=0) emit_shldimm(sh,sl,imm[i],th);
+              emit_shlimm(sl,imm[i],tl);
+            }
+            if(opcode2[i]==0x3a) // DSRL
+            {
+              emit_shrdimm(sl,sh,imm[i],tl);
+              if(th>=0) emit_shrimm(sh,imm[i],th);
+            }
+            if(opcode2[i]==0x3b) // DSRA
+            {
+              emit_shrdimm(sl,sh,imm[i],tl);
+              if(th>=0) emit_sarimm(sh,imm[i],th);
+            }
+          }else{
+            // Shift by zero
+            if(sl!=tl) emit_mov(sl,tl);
+            if(th>=0&&sh!=th) emit_mov(sh,th);
+          }
+        }
+      }
+    }
+  }
+  if(opcode2[i]==0x3c) // DSLL32
+  {
+    if(rt1[i]) {
+      signed char sl,tl,th;
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      if(th>=0||tl>=0){
+        assert(tl>=0);
+        assert(th>=0);
+        assert(sl>=0);
+        emit_mov(sl,th);
+        emit_zeroreg(tl);
+        if(imm[i]>32)
+        {
+          emit_shlimm(th,imm[i]&31,th);
+        }
+      }
+    }
+  }
+  if(opcode2[i]==0x3e) // DSRL32
+  {
+    if(rt1[i]) {
+      signed char sh,tl,th;
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      th=get_reg(i_regs->regmap,rt1[i]|64);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      if(tl>=0){
+        assert(sh>=0);
+        emit_mov(sh,tl);
+        if(th>=0) emit_zeroreg(th);
+        if(imm[i]>32)
+        {
+          emit_shrimm(tl,imm[i]&31,tl);
+        }
+      }
+    }
+  }
+  if(opcode2[i]==0x3f) // DSRA32
+  {
+    if(rt1[i]) {
+      signed char sh,tl;
+      tl=get_reg(i_regs->regmap,rt1[i]);
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      if(tl>=0){
+        assert(sh>=0);
+        emit_mov(sh,tl);
+        if(imm[i]>32)
+        {
+          emit_sarimm(tl,imm[i]&31,tl);
+        }
+      }
+    }
+  }
+}
+
+#ifndef shift_assemble
+void shift_assemble(int i,struct regstat *i_regs)
+{
+  printf("Need shift_assemble for this architecture.\n");
+  exit(1);
+}
+#endif
+
+void load_assemble(int i,struct regstat *i_regs)
+{
+  int s,th,tl,addr,map=-1;
+  int offset;
+  int jaddr=0;
+  int memtarget,c=0;
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,rt1[i]|64);
+  tl=get_reg(i_regs->regmap,rt1[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  offset=imm[i];
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
+  if(s>=0) {
+    c=(i_regs->wasconst>>s)&1;
+    memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+    if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
+  }
+  if(offset||s<0||c) addr=tl;
+  else addr=s;
+  //printf("load_assemble: c=%d\n",c);
+  //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
+  // FIXME: Even if the load is a NOP, we should check for pagefaults...
+  if(tl>=0) {
+    //assert(tl>=0);
+    //assert(rt1[i]);
+    reglist&=~(1<<tl);
+    if(th>=0) reglist&=~(1<<th);
+    if(!using_tlb) {
+      if(!c) {
+//#define R29_HACK 1
+        #ifdef R29_HACK
+        // Strmnnrmn's speed hack
+        if(rs1[i]!=29||start<0x80001000||start>=0x80800000)
+        #endif
+        {
+          emit_cmpimm(addr,0x800000);
+          jaddr=(int)out;
+          #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+          // Hint to branch predictor that the branch is unlikely to be taken
+          if(rs1[i]>=28)
+            emit_jno_unlikely(0);
+          else
+          #endif
+          emit_jno(0);
+        }
+      }
+    }else{ // using tlb
+      int x=0;
+      if (opcode[i]==0x20||opcode[i]==0x24) x=3; // LB/LBU
+      if (opcode[i]==0x21||opcode[i]==0x25) x=2; // LH/LHU
+      map=get_reg(i_regs->regmap,TLREG);
+      assert(map>=0);
+      map=do_tlb_r(addr,tl,map,x,-1,-1,c,constmap[i][s]+offset);
+      do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr);
+    }
+    if (opcode[i]==0x20) { // LB
+      if(!c||memtarget) {
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_movsbl_tlb((constmap[i][s]+offset)^3,map,tl);
+        else
+        #endif
+        {
+          //emit_xorimm(addr,3,tl);
+          //gen_tlb_addr_r(tl,map);
+          //emit_movsbl_indexed((int)rdram-0x80000000,tl,tl);
+          int x=0;
+          if(!c) emit_xorimm(addr,3,tl);
+          else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
+          emit_movsbl_indexed_tlb(x,tl,map,tl);
+        }
+        if(jaddr)
+          add_stub(LOADB_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    }
+    if (opcode[i]==0x21) { // LH
+      if(!c||memtarget) {
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_movswl_tlb((constmap[i][s]+offset)^2,map,tl);
+        else
+        #endif
+        {
+          int x=0;
+          if(!c) emit_xorimm(addr,2,tl);
+          else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
+          //#ifdef
+          //emit_movswl_indexed_tlb(x,tl,map,tl);
+          //else
+          if(map>=0) {
+            gen_tlb_addr_r(tl,map);
+            emit_movswl_indexed(x,tl,tl);
+          }else
+            emit_movswl_indexed((int)rdram-0x80000000+x,tl,tl);
+        }
+        if(jaddr)
+          add_stub(LOADH_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    }
+    if (opcode[i]==0x23) { // LW
+      if(!c||memtarget) {
+        //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_readword_tlb(constmap[i][s]+offset,map,tl);
+        else
+        #endif
+        emit_readword_indexed_tlb(0,addr,map,tl);
+        if(jaddr)
+          add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    }
+    if (opcode[i]==0x24) { // LBU
+      if(!c||memtarget) {
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_movzbl_tlb((constmap[i][s]+offset)^3,map,tl);
+        else
+        #endif
+        {
+          //emit_xorimm(addr,3,tl);
+          //gen_tlb_addr_r(tl,map);
+          //emit_movzbl_indexed((int)rdram-0x80000000,tl,tl);
+          int x=0;
+          if(!c) emit_xorimm(addr,3,tl);
+          else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
+          emit_movzbl_indexed_tlb(x,tl,map,tl);
+        }
+        if(jaddr)
+          add_stub(LOADBU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    }
+    if (opcode[i]==0x25) { // LHU
+      if(!c||memtarget) {
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_movzwl_tlb((constmap[i][s]+offset)^2,map,tl);
+        else
+        #endif
+        {
+          int x=0;
+          if(!c) emit_xorimm(addr,2,tl);
+          else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
+          //#ifdef
+          //emit_movzwl_indexed_tlb(x,tl,map,tl);
+          //#else
+          if(map>=0) {
+            gen_tlb_addr_r(tl,map);
+            emit_movzwl_indexed(x,tl,tl);
+          }else
+            emit_movzwl_indexed((int)rdram-0x80000000+x,tl,tl);
+          if(jaddr)
+            add_stub(LOADHU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+        }
+      }
+      else
+        inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    }
+    if (opcode[i]==0x27) { // LWU
+      assert(th>=0);
+      if(!c||memtarget) {
+        //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_readword_tlb(constmap[i][s]+offset,map,tl);
+        else
+        #endif
+        emit_readword_indexed_tlb(0,addr,map,tl);
+        if(jaddr)
+          add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+      }
+      else {
+        inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+      }
+      emit_zeroreg(th);
+    }
+    if (opcode[i]==0x37) { // LD
+      if(!c||memtarget) {
+        //gen_tlb_addr_r(tl,map);
+        //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,addr,th);
+        //emit_readword_indexed((int)rdram-0x7FFFFFFC,addr,tl);
+        #ifdef HOST_IMM_ADDR32
+        if(c)
+          emit_readdword_tlb(constmap[i][s]+offset,map,th,tl);
+        else
+        #endif
+        emit_readdword_indexed_tlb(0,addr,map,th,tl);
+        if(jaddr)
+          add_stub(LOADD_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+      }
+      else
+        inline_readstub(LOADD_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    }
+    //emit_storereg(rt1[i],tl); // DEBUG
+  }
+  //if(opcode[i]==0x23)
+  //if(opcode[i]==0x24)
+  //if(opcode[i]==0x23||opcode[i]==0x24)
+  /*if(opcode[i]==0x21||opcode[i]==0x23||opcode[i]==0x24)
+  {
+    //emit_pusha();
+    save_regs(0x100f);
+        emit_readword((int)&last_count,ECX);
+        #ifdef __i386__
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,HOST_CCREG);
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+        #endif
+        #ifdef __arm__
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,0);
+        else
+          emit_mov(HOST_CCREG,0);
+        emit_add(0,ECX,0);
+        emit_addimm(0,2*ccadj[i],0);
+        emit_writeword(0,(int)&Count);
+        #endif
+    emit_call((int)memdebug);
+    //emit_popa();
+    restore_regs(0x100f);
+  }/**/
+}
+
+#ifndef loadlr_assemble
+void loadlr_assemble(int i,struct regstat *i_regs)
+{
+  printf("Need loadlr_assemble for this architecture.\n");
+  exit(1);
+}
+#endif
+
+void store_assemble(int i,struct regstat *i_regs)
+{
+  int s,th,tl,map=-1;
+  int addr,temp;
+  int offset;
+  int jaddr=0,jaddr2,type;
+  int memtarget,c=0;
+  int agr=AGEN1+(i&1);
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,rs2[i]|64);
+  tl=get_reg(i_regs->regmap,rs2[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,agr);
+  if(temp<0) temp=get_reg(i_regs->regmap,-1);
+  offset=imm[i];
+  if(s>=0) {
+    c=(i_regs->wasconst>>s)&1;
+    memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+    if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
+  }
+  assert(tl>=0);
+  assert(temp>=0);
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
+  if(offset||s<0||c) addr=temp;
+  else addr=s;
+  if(!using_tlb) {
+    if(!c) {
+      #ifdef R29_HACK
+      // Strmnnrmn's speed hack
+      memtarget=1;
+      if(rs1[i]!=29||start<0x80001000||start>=0x80800000)
+      #endif
+      emit_cmpimm(addr,0x800000);
+      #ifdef DESTRUCTIVE_SHIFT
+      if(s==addr) emit_mov(s,temp);
+      #endif
+      #ifdef R29_HACK
+      if(rs1[i]!=29||start<0x80001000||start>=0x80800000)
+      #endif
+      {
+        jaddr=(int)out;
+        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+        // Hint to branch predictor that the branch is unlikely to be taken
+        if(rs1[i]>=28)
+          emit_jno_unlikely(0);
+        else
+        #endif
+        emit_jno(0);
+      }
+    }
+  }else{ // using tlb
+    int x=0;
+    if (opcode[i]==0x28) x=3; // SB
+    if (opcode[i]==0x29) x=2; // SH
+    map=get_reg(i_regs->regmap,TLREG);
+    assert(map>=0);
+    map=do_tlb_w(addr,temp,map,x,c,constmap[i][s]+offset);
+    do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr);
+  }
+
+  if (opcode[i]==0x28) { // SB
+    if(!c||memtarget) {
+      int x=0;
+      if(!c) emit_xorimm(addr,3,temp);
+      else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
+      //gen_tlb_addr_w(temp,map);
+      //emit_writebyte_indexed(tl,(int)rdram-0x80000000,temp);
+      emit_writebyte_indexed_tlb(tl,x,temp,map,temp);
+    }
+    type=STOREB_STUB;
+  }
+  if (opcode[i]==0x29) { // SH
+    if(!c||memtarget) {
+      int x=0;
+      if(!c) emit_xorimm(addr,2,temp);
+      else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
+      //#ifdef
+      //emit_writehword_indexed_tlb(tl,x,temp,map,temp);
+      //#else
+      if(map>=0) {
+        gen_tlb_addr_w(temp,map);
+        emit_writehword_indexed(tl,x,temp);
+      }else
+        emit_writehword_indexed(tl,(int)rdram-0x80000000+x,temp);
+    }
+    type=STOREH_STUB;
+  }
+  if (opcode[i]==0x2B) { // SW
+    if(!c||memtarget)
+      //emit_writeword_indexed(tl,(int)rdram-0x80000000,addr);
+      emit_writeword_indexed_tlb(tl,0,addr,map,temp);
+    type=STOREW_STUB;
+  }
+  if (opcode[i]==0x3F) { // SD
+    if(!c||memtarget) {
+      if(rs2[i]) {
+        assert(th>=0);
+        //emit_writeword_indexed(th,(int)rdram-0x80000000,addr);
+        //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,addr);
+        emit_writedword_indexed_tlb(th,tl,0,addr,map,temp);
+      }else{
+        // Store zero
+        //emit_writeword_indexed(tl,(int)rdram-0x80000000,temp);
+        //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,temp);
+        emit_writedword_indexed_tlb(tl,tl,0,addr,map,temp);
+      }
+    }
+    type=STORED_STUB;
+  }
+  if(jaddr) {
+    add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+  } else if(!memtarget) {
+    inline_writestub(type,i,constmap[i][s]+offset,i_regs->regmap,rs2[i],ccadj[i],reglist);
+  }
+  if(!using_tlb) {
+    if(!c||memtarget) {
+      #ifdef DESTRUCTIVE_SHIFT
+      // The x86 shift operation is 'destructive'; it overwrites the
+      // source register, so we need to make a copy first and use that.
+      addr=temp;
+      #endif
+      #if defined(HOST_IMM8)
+      int ir=get_reg(i_regs->regmap,INVCP);
+      assert(ir>=0);
+      emit_cmpmem_indexedsr12_reg(ir,addr,1);
+      #else
+      emit_cmpmem_indexedsr12_imm((int)invalid_code,addr,1);
+      #endif
+      jaddr2=(int)out;
+      emit_jne(0);
+      add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),addr,0,0,0);
+    }
+  }
+  //if(opcode[i]==0x2B || opcode[i]==0x3F)
+  //if(opcode[i]==0x2B || opcode[i]==0x28)
+  //if(opcode[i]==0x2B || opcode[i]==0x29)
+  //if(opcode[i]==0x2B)
+  /*if(opcode[i]==0x2B || opcode[i]==0x28 || opcode[i]==0x29 || opcode[i]==0x3F)
+  {
+    //emit_pusha();
+    save_regs(0x100f);
+        emit_readword((int)&last_count,ECX);
+        #ifdef __i386__
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,HOST_CCREG);
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+        #endif
+        #ifdef __arm__
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,0);
+        else
+          emit_mov(HOST_CCREG,0);
+        emit_add(0,ECX,0);
+        emit_addimm(0,2*ccadj[i],0);
+        emit_writeword(0,(int)&Count);
+        #endif
+    emit_call((int)memdebug);
+    //emit_popa();
+    restore_regs(0x100f);
+  }/**/
+}
+
+void storelr_assemble(int i,struct regstat *i_regs)
+{
+  int s,th,tl;
+  int temp;
+  int temp2;
+  int offset;
+  int jaddr=0,jaddr2;
+  int case1,case2,case3;
+  int done0,done1,done2;
+  int memtarget,c=0;
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,rs2[i]|64);
+  tl=get_reg(i_regs->regmap,rs2[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,-1);
+  offset=imm[i];
+  if(s>=0) {
+    c=(i_regs->isconst>>s)&1;
+    memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000;
+    if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1;
+  }
+  assert(tl>=0);
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  if(tl>=0) {
+    assert(temp>=0);
+    if(!using_tlb) {
+      if(!c) {
+        emit_cmpimm(s<0||offset?temp:s,0x800000);
+        if(!offset&&s!=temp) emit_mov(s,temp);
+        jaddr=(int)out;
+        emit_jno(0);
+      }
+      else
+      {
+        if(!memtarget||!rs1[i]) {
+          jaddr=(int)out;
+          emit_jmp(0);
+        }
+      }
+      if((u_int)rdram!=0x80000000) 
+        emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp);
+    }else{ // using tlb
+      int map=get_reg(i_regs->regmap,TLREG);
+      assert(map>=0);
+      map=do_tlb_w(c||s<0||offset?temp:s,temp,map,0,c,constmap[i][s]+offset);
+      if(!c&&!offset&&s>=0) emit_mov(s,temp);
+      do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr);
+      if(!jaddr&&!memtarget) {
+        jaddr=(int)out;
+        emit_jmp(0);
+      }
+      gen_tlb_addr_w(temp,map);
+    }
+
+    if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR
+      temp2=get_reg(i_regs->regmap,FTEMP);
+      if(!rs2[i]) temp2=th=tl;
+    }
+
+    emit_testimm(temp,2);
+    case2=(int)out;
+    emit_jne(0);
+    emit_testimm(temp,1);
+    case1=(int)out;
+    emit_jne(0);
+    // 0
+    if (opcode[i]==0x2A) { // SWL
+      emit_writeword_indexed(tl,0,temp);
+    }
+    if (opcode[i]==0x2E) { // SWR
+      emit_writebyte_indexed(tl,3,temp);
+    }
+    if (opcode[i]==0x2C) { // SDL
+      emit_writeword_indexed(th,0,temp);
+      if(rs2[i]) emit_mov(tl,temp2);
+    }
+    if (opcode[i]==0x2D) { // SDR
+      emit_writebyte_indexed(tl,3,temp);
+      if(rs2[i]) emit_shldimm(th,tl,24,temp2);
+    }
+    done0=(int)out;
+    emit_jmp(0);
+    // 1
+    set_jump_target(case1,(int)out);
+    if (opcode[i]==0x2A) { // SWL
+      // Write 3 msb into three least significant bytes
+      if(rs2[i]) emit_rorimm(tl,8,tl);
+      emit_writehword_indexed(tl,-1,temp);
+      if(rs2[i]) emit_rorimm(tl,16,tl);
+      emit_writebyte_indexed(tl,1,temp);
+      if(rs2[i]) emit_rorimm(tl,8,tl);
+    }
+    if (opcode[i]==0x2E) { // SWR
+      // Write two lsb into two most significant bytes
+      emit_writehword_indexed(tl,1,temp);
+    }
+    if (opcode[i]==0x2C) { // SDL
+      if(rs2[i]) emit_shrdimm(tl,th,8,temp2);
+      // Write 3 msb into three least significant bytes
+      if(rs2[i]) emit_rorimm(th,8,th);
+      emit_writehword_indexed(th,-1,temp);
+      if(rs2[i]) emit_rorimm(th,16,th);
+      emit_writebyte_indexed(th,1,temp);
+      if(rs2[i]) emit_rorimm(th,8,th);
+    }
+    if (opcode[i]==0x2D) { // SDR
+      if(rs2[i]) emit_shldimm(th,tl,16,temp2);
+      // Write two lsb into two most significant bytes
+      emit_writehword_indexed(tl,1,temp);
+    }
+    done1=(int)out;
+    emit_jmp(0);
+    // 2
+    set_jump_target(case2,(int)out);
+    emit_testimm(temp,1);
+    case3=(int)out;
+    emit_jne(0);
+    if (opcode[i]==0x2A) { // SWL
+      // Write two msb into two least significant bytes
+      if(rs2[i]) emit_rorimm(tl,16,tl);
+      emit_writehword_indexed(tl,-2,temp);
+      if(rs2[i]) emit_rorimm(tl,16,tl);
+    }
+    if (opcode[i]==0x2E) { // SWR
+      // Write 3 lsb into three most significant bytes
+      emit_writebyte_indexed(tl,-1,temp);
+      if(rs2[i]) emit_rorimm(tl,8,tl);
+      emit_writehword_indexed(tl,0,temp);
+      if(rs2[i]) emit_rorimm(tl,24,tl);
+    }
+    if (opcode[i]==0x2C) { // SDL
+      if(rs2[i]) emit_shrdimm(tl,th,16,temp2);
+      // Write two msb into two least significant bytes
+      if(rs2[i]) emit_rorimm(th,16,th);
+      emit_writehword_indexed(th,-2,temp);
+      if(rs2[i]) emit_rorimm(th,16,th);
+    }
+    if (opcode[i]==0x2D) { // SDR
+      if(rs2[i]) emit_shldimm(th,tl,8,temp2);
+      // Write 3 lsb into three most significant bytes
+      emit_writebyte_indexed(tl,-1,temp);
+      if(rs2[i]) emit_rorimm(tl,8,tl);
+      emit_writehword_indexed(tl,0,temp);
+      if(rs2[i]) emit_rorimm(tl,24,tl);
+    }
+    done2=(int)out;
+    emit_jmp(0);
+    // 3
+    set_jump_target(case3,(int)out);
+    if (opcode[i]==0x2A) { // SWL
+      // Write msb into least significant byte
+      if(rs2[i]) emit_rorimm(tl,24,tl);
+      emit_writebyte_indexed(tl,-3,temp);
+      if(rs2[i]) emit_rorimm(tl,8,tl);
+    }
+    if (opcode[i]==0x2E) { // SWR
+      // Write entire word
+      emit_writeword_indexed(tl,-3,temp);
+    }
+    if (opcode[i]==0x2C) { // SDL
+      if(rs2[i]) emit_shrdimm(tl,th,24,temp2);
+      // Write msb into least significant byte
+      if(rs2[i]) emit_rorimm(th,24,th);
+      emit_writebyte_indexed(th,-3,temp);
+      if(rs2[i]) emit_rorimm(th,8,th);
+    }
+    if (opcode[i]==0x2D) { // SDR
+      if(rs2[i]) emit_mov(th,temp2);
+      // Write entire word
+      emit_writeword_indexed(tl,-3,temp);
+    }
+    set_jump_target(done0,(int)out);
+    set_jump_target(done1,(int)out);
+    set_jump_target(done2,(int)out);
+    if (opcode[i]==0x2C) { // SDL
+      emit_testimm(temp,4);
+      done0=(int)out;
+      emit_jne(0);
+      emit_andimm(temp,~3,temp);
+      emit_writeword_indexed(temp2,4,temp);
+      set_jump_target(done0,(int)out);
+    }
+    if (opcode[i]==0x2D) { // SDR
+      emit_testimm(temp,4);
+      done0=(int)out;
+      emit_jeq(0);
+      emit_andimm(temp,~3,temp);
+      emit_writeword_indexed(temp2,-4,temp);
+      set_jump_target(done0,(int)out);
+    }
+    if(!c||!memtarget)
+      add_stub(STORELR_STUB,jaddr,(int)out,0,(int)i_regs,rs2[i],ccadj[i],reglist);
+  }
+  if(!using_tlb) {
+    emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp);
+    #if defined(HOST_IMM8)
+    int ir=get_reg(i_regs->regmap,INVCP);
+    assert(ir>=0);
+    emit_cmpmem_indexedsr12_reg(ir,temp,1);
+    #else
+    emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
+    #endif
+    jaddr2=(int)out;
+    emit_jne(0);
+    add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
+  }
+  /*
+    emit_pusha();
+    //save_regs(0x100f);
+        emit_readword((int)&last_count,ECX);
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,HOST_CCREG);
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+    emit_call((int)memdebug);
+    emit_popa();
+    //restore_regs(0x100f);
+  /**/
+}
+
+void c1ls_assemble(int i,struct regstat *i_regs)
+{
+  int s,th,tl;
+  int temp,ar;
+  int map=-1;
+  int offset;
+  int c=0;
+  int jaddr,jaddr2=0,jaddr3,type;
+  int agr=AGEN1+(i&1);
+  u_int hr,reglist=0;
+  th=get_reg(i_regs->regmap,FTEMP|64);
+  tl=get_reg(i_regs->regmap,FTEMP);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,agr);
+  if(temp<0) temp=get_reg(i_regs->regmap,-1);
+  offset=imm[i];
+  assert(tl>=0);
+  assert(rs1[i]>0);
+  assert(temp>=0);
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
+  }
+  if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
+  if (opcode[i]==0x31||opcode[i]==0x35) // LWC1/LDC1
+  {
+    // Loads use a temporary register which we need to save
+    reglist|=1<<temp;
+  }
+  if (opcode[i]==0x39||opcode[i]==0x3D) // SWC1/SDC1
+    ar=temp;
+  else // LWC1/LDC1
+    ar=tl;
+  //if(s<0) emit_loadreg(rs1[i],ar); //address_generation does this now
+  //else c=(i_regs->wasconst>>s)&1;
+  if(s>=0) c=(i_regs->wasconst>>s)&1;
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    signed char rs=get_reg(i_regs->regmap,CSREG);
+    assert(rs>=0);
+    emit_testimm(rs,0x20000000);
+    jaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0);
+    cop1_usable=1;
+  }
+  if (opcode[i]==0x39) { // SWC1 (get float address)
+    emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],tl);
+  }
+  if (opcode[i]==0x3D) { // SDC1 (get double address)
+    emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],tl);
+  }
+  // Generate address + offset
+  if(!using_tlb) {
+    if(!c)
+      emit_cmpimm(offset||c||s<0?ar:s,0x800000);
+  }
+  else
+  {
+    map=get_reg(i_regs->regmap,TLREG);
+    assert(map>=0);
+    if (opcode[i]==0x31||opcode[i]==0x35) { // LWC1/LDC1
+      map=do_tlb_r(offset||c||s<0?ar:s,ar,map,0,-1,-1,c,constmap[i][s]+offset);
+    }
+    if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
+      map=do_tlb_w(offset||c||s<0?ar:s,ar,map,0,c,constmap[i][s]+offset);
+    }
+  }
+  if (opcode[i]==0x39) { // SWC1 (read float)
+    emit_readword_indexed(0,tl,tl);
+  }
+  if (opcode[i]==0x3D) { // SDC1 (read double)
+    emit_readword_indexed(4,tl,th);
+    emit_readword_indexed(0,tl,tl);
+  }
+  if (opcode[i]==0x31) { // LWC1 (get target address)
+    emit_readword((int)&reg_cop1_simple[(source[i]>>16)&0x1f],temp);
+  }
+  if (opcode[i]==0x35) { // LDC1 (get target address)
+    emit_readword((int)&reg_cop1_double[(source[i]>>16)&0x1f],temp);
+  }
+  if(!using_tlb) {
+    if(!c) {
+      jaddr2=(int)out;
+      emit_jno(0);
+    }
+    else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80800000) {
+      jaddr2=(int)out;
+      emit_jmp(0); // inline_readstub/inline_writestub?  Very rare case
+    }
+    #ifdef DESTRUCTIVE_SHIFT
+    if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
+      if(!offset&&!c&&s>=0) emit_mov(s,ar);
+    }
+    #endif
+  }else{
+    if (opcode[i]==0x31||opcode[i]==0x35) { // LWC1/LDC1
+      do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr2);
+    }
+    if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
+      do_tlb_w_branch(map,c,constmap[i][s]+offset,&jaddr2);
+    }
+  }
+  if (opcode[i]==0x31) { // LWC1
+    //if(s>=0&&!c&&!offset) emit_mov(s,tl);
+    //gen_tlb_addr_r(ar,map);
+    //emit_readword_indexed((int)rdram-0x80000000,tl,tl);
+    #ifdef HOST_IMM_ADDR32
+    if(c) emit_readword_tlb(constmap[i][s]+offset,map,tl);
+    else
+    #endif
+    emit_readword_indexed_tlb(0,offset||c||s<0?tl:s,map,tl);
+    type=LOADW_STUB;
+  }
+  if (opcode[i]==0x35) { // LDC1
+    assert(th>=0);
+    //if(s>=0&&!c&&!offset) emit_mov(s,tl);
+    //gen_tlb_addr_r(ar,map);
+    //emit_readword_indexed((int)rdram-0x80000000,tl,th);
+    //emit_readword_indexed((int)rdram-0x7FFFFFFC,tl,tl);
+    #ifdef HOST_IMM_ADDR32
+    if(c) emit_readdword_tlb(constmap[i][s]+offset,map,th,tl);
+    else
+    #endif
+    emit_readdword_indexed_tlb(0,offset||c||s<0?tl:s,map,th,tl);
+    type=LOADD_STUB;
+  }
+  if (opcode[i]==0x39) { // SWC1
+    //emit_writeword_indexed(tl,(int)rdram-0x80000000,temp);
+    emit_writeword_indexed_tlb(tl,0,offset||c||s<0?temp:s,map,temp);
+    type=STOREW_STUB;
+  }
+  if (opcode[i]==0x3D) { // SDC1
+    assert(th>=0);
+    //emit_writeword_indexed(th,(int)rdram-0x80000000,temp);
+    //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,temp);
+    emit_writedword_indexed_tlb(th,tl,0,offset||c||s<0?temp:s,map,temp);
+    type=STORED_STUB;
+  }
+  if(!using_tlb) {
+    if (opcode[i]==0x39||opcode[i]==0x3D) { // SWC1/SDC1
+      #ifndef DESTRUCTIVE_SHIFT
+      temp=offset||c||s<0?ar:s;
+      #endif
+      #if defined(HOST_IMM8)
+      int ir=get_reg(i_regs->regmap,INVCP);
+      assert(ir>=0);
+      emit_cmpmem_indexedsr12_reg(ir,temp,1);
+      #else
+      emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
+      #endif
+      jaddr3=(int)out;
+      emit_jne(0);
+      add_stub(INVCODE_STUB,jaddr3,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
+    }
+  }
+  if(jaddr2) add_stub(type,jaddr2,(int)out,i,offset||c||s<0?ar:s,(int)i_regs,ccadj[i],reglist);
+  if (opcode[i]==0x31) { // LWC1 (write float)
+    emit_writeword_indexed(tl,0,temp);
+  }
+  if (opcode[i]==0x35) { // LDC1 (write double)
+    emit_writeword_indexed(th,4,temp);
+    emit_writeword_indexed(tl,0,temp);
+  }
+  //if(opcode[i]==0x39)
+  /*if(opcode[i]==0x39||opcode[i]==0x31)
+  {
+    emit_pusha();
+        emit_readword((int)&last_count,ECX);
+        if(get_reg(i_regs->regmap,CCREG)<0)
+          emit_loadreg(CCREG,HOST_CCREG);
+        emit_add(HOST_CCREG,ECX,HOST_CCREG);
+        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
+        emit_writeword(HOST_CCREG,(int)&Count);
+    emit_call((int)memdebug);
+    emit_popa();
+  }/**/
+}
+
+#ifndef multdiv_assemble
+void multdiv_assemble(int i,struct regstat *i_regs)
+{
+  printf("Need multdiv_assemble for this architecture.\n");
+  exit(1);
+}
+#endif
+
+void mov_assemble(int i,struct regstat *i_regs)
+{
+  //if(opcode2[i]==0x10||opcode2[i]==0x12) { // MFHI/MFLO
+  //if(opcode2[i]==0x11||opcode2[i]==0x13) { // MTHI/MTLO
+  assert(rt1[i]>0);
+  if(rt1[i]) {
+    signed char sh,sl,th,tl;
+    th=get_reg(i_regs->regmap,rt1[i]|64);
+    tl=get_reg(i_regs->regmap,rt1[i]);
+    //assert(tl>=0);
+    if(tl>=0) {
+      sh=get_reg(i_regs->regmap,rs1[i]|64);
+      sl=get_reg(i_regs->regmap,rs1[i]);
+      if(sl>=0) emit_mov(sl,tl);
+      else emit_loadreg(rs1[i],tl);
+      if(th>=0) {
+        if(sh>=0) emit_mov(sh,th);
+        else emit_loadreg(rs1[i]|64,th);
+      }
+    }
+  }
+}
+
+#ifndef fconv_assemble
+void fconv_assemble(int i,struct regstat *i_regs)
+{
+  printf("Need fconv_assemble for this architecture.\n");
+  exit(1);
+}
+#endif
+
+#if 0
+void float_assemble(int i,struct regstat *i_regs)
+{
+  printf("Need float_assemble for this architecture.\n");
+  exit(1);
+}
+#endif
+
+void syscall_assemble(int i,struct regstat *i_regs)
+{
+  signed char ccreg=get_reg(i_regs->regmap,CCREG);
+  assert(ccreg==HOST_CCREG);
+  assert(!is_delayslot);
+  emit_movimm(start+i*4,EAX); // Get PC
+  emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
+  emit_jmp((int)jump_syscall);
+}
+
+void ds_assemble(int i,struct regstat *i_regs)
+{
+  is_delayslot=1;
+  switch(itype[i]) {
+    case ALU:
+      alu_assemble(i,i_regs);break;
+    case IMM16:
+      imm16_assemble(i,i_regs);break;
+    case SHIFT:
+      shift_assemble(i,i_regs);break;
+    case SHIFTIMM:
+      shiftimm_assemble(i,i_regs);break;
+    case LOAD:
+      load_assemble(i,i_regs);break;
+    case LOADLR:
+      loadlr_assemble(i,i_regs);break;
+    case STORE:
+      store_assemble(i,i_regs);break;
+    case STORELR:
+      storelr_assemble(i,i_regs);break;
+    case COP0:
+      cop0_assemble(i,i_regs);break;
+    case COP1:
+      cop1_assemble(i,i_regs);break;
+    case C1LS:
+      c1ls_assemble(i,i_regs);break;
+    case FCONV:
+      fconv_assemble(i,i_regs);break;
+    case FLOAT:
+      float_assemble(i,i_regs);break;
+    case FCOMP:
+      fcomp_assemble(i,i_regs);break;
+    case MULTDIV:
+      multdiv_assemble(i,i_regs);break;
+    case MOV:
+      mov_assemble(i,i_regs);break;
+    case SYSCALL:
+    case SPAN:
+    case UJUMP:
+    case RJUMP:
+    case CJUMP:
+    case SJUMP:
+    case FJUMP:
+      printf("Jump in the delay slot.  This is probably a bug.\n");
+  }
+  is_delayslot=0;
+}
+
+// Is the branch target a valid internal jump?
+int internal_branch(uint64_t i_is32,int addr)
+{
+  if(addr&1) return 0; // Indirect (register) jump
+  if(addr>=start && addr<start+slen*4-4)
+  {
+    int t=(addr-start)>>2;
+    // Delay slots are not valid branch targets
+    //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
+    // 64 -> 32 bit transition requires a recompile
+    /*if(is32[t]&~unneeded_reg_upper[t]&~i_is32)
+    {
+      if(requires_32bit[t]&~i_is32) printf("optimizable: no\n");
+      else printf("optimizable: yes\n");
+    }*/
+    //if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
+    if(requires_32bit[t]&~i_is32) return 0;
+    else return 1;
+  }
+  return 0;
+}
+
+#ifndef wb_invalidate
+void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32,
+  uint64_t u,uint64_t uu)
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(pre[hr]!=entry[hr]) {
+        if(pre[hr]>=0) {
+          if((dirty>>hr)&1) {
+            if(get_reg(entry,pre[hr])<0) {
+              if(pre[hr]<64) {
+                if(!((u>>pre[hr])&1)) {
+                  emit_storereg(pre[hr],hr);
+                  if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) {
+                    emit_sarimm(hr,31,hr);
+                    emit_storereg(pre[hr]|64,hr);
+                  }
+                }
+              }else{
+                if(!((uu>>(pre[hr]&63))&1) && !((is32>>(pre[hr]&63))&1)) {
+                  emit_storereg(pre[hr],hr);
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  // Move from one register to another (no writeback)
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(pre[hr]!=entry[hr]) {
+        if(pre[hr]>=0&&(pre[hr]&63)<TEMPREG) {
+          int nr;
+          if((nr=get_reg(entry,pre[hr]))>=0) {
+            emit_mov(hr,nr);
+          }
+        }
+      }
+    }
+  }
+}
+#endif
+
+// Load the specified registers
+// This only loads the registers given as arguments because
+// we don't want to load things that will be overwritten
+void load_regs(signed char entry[],signed char regmap[],int is32,int rs1,int rs2)
+{
+  int hr;
+  // Load 32-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
+      if(entry[hr]!=regmap[hr]) {
+        if(regmap[hr]==rs1||regmap[hr]==rs2)
+        {
+          if(regmap[hr]==0) {
+            emit_zeroreg(hr);
+          }
+          else
+          {
+            emit_loadreg(regmap[hr],hr);
+          }
+        }
+      }
+    }
+  }
+  //Load 64-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
+      if(entry[hr]!=regmap[hr]) {
+        if(regmap[hr]-64==rs1||regmap[hr]-64==rs2)
+        {
+          assert(regmap[hr]!=64);
+          if((is32>>(regmap[hr]&63))&1) {
+            int lr=get_reg(regmap,regmap[hr]-64);
+            if(lr>=0)
+              emit_sarimm(lr,31,hr);
+            else
+              emit_loadreg(regmap[hr],hr);
+          }
+          else
+          {
+            emit_loadreg(regmap[hr],hr);
+          }
+        }
+      }
+    }
+  }
+}
+
+// Load registers prior to the start of a loop
+// so that they are not loaded within the loop
+static void loop_preload(signed char pre[],signed char entry[])
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(pre[hr]!=entry[hr]) {
+        if(entry[hr]>=0) {
+          if(get_reg(pre,entry[hr])<0) {
+            assem_debug("loop preload:\n");
+            //printf("loop preload: %d\n",hr);
+            if(entry[hr]==0) {
+              emit_zeroreg(hr);
+            }
+            else if(entry[hr]<TEMPREG)
+            {
+              emit_loadreg(entry[hr],hr);
+            }
+            else if(entry[hr]-64<TEMPREG)
+            {
+              emit_loadreg(entry[hr],hr);
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+// Generate address for load/store instruction
+void address_generation(int i,struct regstat *i_regs,signed char entry[])
+{
+  if(itype[i]==LOAD||itype[i]==LOADLR||itype[i]==STORE||itype[i]==STORELR||itype[i]==C1LS) {
+    int ra;
+    int agr=AGEN1+(i&1);
+    int mgr=MGEN1+(i&1);
+    if(itype[i]==LOAD) {
+      ra=get_reg(i_regs->regmap,rt1[i]);
+      //if(rt1[i]) assert(ra>=0);
+    }
+    if(itype[i]==LOADLR) {
+      ra=get_reg(i_regs->regmap,FTEMP);
+    }
+    if(itype[i]==STORE||itype[i]==STORELR) {
+      ra=get_reg(i_regs->regmap,agr);
+      if(ra<0) ra=get_reg(i_regs->regmap,-1);
+    }
+    if(itype[i]==C1LS) {
+      if (opcode[i]==0x31||opcode[i]==0x35) // LWC1/LDC1
+        ra=get_reg(i_regs->regmap,FTEMP);
+      else { // SWC1/SDC1
+        ra=get_reg(i_regs->regmap,agr);
+        if(ra<0) ra=get_reg(i_regs->regmap,-1);
+      }
+    }
+    int rs=get_reg(i_regs->regmap,rs1[i]);
+    int rm=get_reg(i_regs->regmap,TLREG);
+    if(ra>=0) {
+      int offset=imm[i];
+      int c=(i_regs->wasconst>>rs)&1;
+      if(rs1[i]==0) {
+        // Using r0 as a base address
+        /*if(rm>=0) {
+          if(!entry||entry[rm]!=mgr) {
+            generate_map_const(offset,rm);
+          } // else did it in the previous cycle
+        }*/
+        if(!entry||entry[ra]!=agr) {
+          if (opcode[i]==0x22||opcode[i]==0x26) {
+            emit_movimm(offset&0xFFFFFFFC,ra); // LWL/LWR
+          }else if (opcode[i]==0x1a||opcode[i]==0x1b) {
+            emit_movimm(offset&0xFFFFFFF8,ra); // LDL/LDR
+          }else{
+            emit_movimm(offset,ra);
+          }
+        } // else did it in the previous cycle
+      }
+      else if(rs<0) {
+        if(!entry||entry[ra]!=rs1[i])
+          emit_loadreg(rs1[i],ra);
+        //if(!entry||entry[ra]!=rs1[i])
+        //  printf("poor load scheduling!\n");
+      }
+      else if(c) {
+        if(rm>=0) {
+          if(!entry||entry[rm]!=mgr) {
+            if(itype[i]==STORE||itype[i]==STORELR||opcode[i]==0x39||opcode[i]==0x3D) {
+              // Stores to memory go thru the mapper to detect self-modifying
+              // code, loads don't.
+              if((unsigned int)(constmap[i][rs]+offset)>=0xC0000000 ||
+                 (unsigned int)(constmap[i][rs]+offset)<0x80800000 )
+                generate_map_const(constmap[i][rs]+offset,rm);
+            }else{
+              if((signed int)(constmap[i][rs]+offset)>=(signed int)0xC0000000)
+                generate_map_const(constmap[i][rs]+offset,rm);
+            }
+          }
+        }
+        if(rs1[i]!=rt1[i]||itype[i]!=LOAD) {
+          if(!entry||entry[ra]!=agr) {
+            if (opcode[i]==0x22||opcode[i]==0x26) {
+              emit_movimm((constmap[i][rs]+offset)&0xFFFFFFFC,ra); // LWL/LWR
+            }else if (opcode[i]==0x1a||opcode[i]==0x1b) {
+              emit_movimm((constmap[i][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR
+            }else{
+              #ifdef HOST_IMM_ADDR32
+              if((itype[i]!=LOAD&&opcode[i]!=0x31&&opcode[i]!=0x35) ||
+                 (using_tlb&&((signed int)constmap[i][rs]+offset)>=(signed int)0xC0000000))
+              #endif
+              emit_movimm(constmap[i][rs]+offset,ra);
+            }
+          } // else did it in the previous cycle
+        } // else load_consts already did it
+      }
+      if(offset&&!c&&rs1[i]) {
+        if(rs>=0) {
+          emit_addimm(rs,offset,ra);
+        }else{
+          emit_addimm(ra,offset,ra);
+        }
+      }
+    }
+  }
+  // Preload constants for next instruction
+  if(itype[i+1]==LOAD||itype[i+1]==LOADLR||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS) {
+    int agr,ra;
+    #ifndef HOST_IMM_ADDR32
+    // Mapper entry
+    agr=MGEN1+((i+1)&1);
+    ra=get_reg(i_regs->regmap,agr);
+    if(ra>=0) {
+      int rs=get_reg(regs[i+1].regmap,rs1[i+1]);
+      int offset=imm[i+1];
+      int c=(regs[i+1].wasconst>>rs)&1;
+      if(c) {
+        if(itype[i+1]==STORE||itype[i+1]==STORELR||opcode[i+1]==0x39||opcode[i+1]==0x3D) {
+          // Stores to memory go thru the mapper to detect self-modifying
+          // code, loads don't.
+          if((unsigned int)(constmap[i+1][rs]+offset)>=0xC0000000 ||
+             (unsigned int)(constmap[i+1][rs]+offset)<0x80800000 )
+            generate_map_const(constmap[i+1][rs]+offset,ra);
+        }else{
+          if((signed int)(constmap[i+1][rs]+offset)>=(signed int)0xC0000000)
+            generate_map_const(constmap[i+1][rs]+offset,ra);
+        }
+      }
+      /*else if(rs1[i]==0) {
+        generate_map_const(offset,ra);
+      }*/
+    }
+    #endif
+    // Actual address
+    agr=AGEN1+((i+1)&1);
+    ra=get_reg(i_regs->regmap,agr);
+    if(ra>=0) {
+      int rs=get_reg(regs[i+1].regmap,rs1[i+1]);
+      int offset=imm[i+1];
+      int c=(regs[i+1].wasconst>>rs)&1;
+      if(c&&(rs1[i+1]!=rt1[i+1]||itype[i+1]!=LOAD)) {
+        if (opcode[i+1]==0x22||opcode[i+1]==0x26) {
+          emit_movimm((constmap[i+1][rs]+offset)&0xFFFFFFFC,ra); // LWL/LWR
+        }else if (opcode[i+1]==0x1a||opcode[i+1]==0x1b) {
+          emit_movimm((constmap[i+1][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR
+        }else{
+          #ifdef HOST_IMM_ADDR32
+          if((itype[i+1]!=LOAD&&opcode[i+1]!=0x31&&opcode[i+1]!=0x35) ||
+             (using_tlb&&((signed int)constmap[i+1][rs]+offset)>=(signed int)0xC0000000))
+          #endif
+          emit_movimm(constmap[i+1][rs]+offset,ra);
+        }
+      }
+      else if(rs1[i+1]==0) {
+        // Using r0 as a base address
+        if (opcode[i+1]==0x22||opcode[i+1]==0x26) {
+          emit_movimm(offset&0xFFFFFFFC,ra); // LWL/LWR
+        }else if (opcode[i+1]==0x1a||opcode[i+1]==0x1b) {
+          emit_movimm(offset&0xFFFFFFF8,ra); // LDL/LDR
+        }else{
+          emit_movimm(offset,ra);
+        }
+      }
+    }
+  }
+}
+
+int get_final_value(int hr, int i, int *value)
+{
+  int reg=regs[i].regmap[hr];
+  while(i<slen-1) {
+    if(regs[i+1].regmap[hr]!=reg) break;
+    if(!((regs[i+1].isconst>>hr)&1)) break;
+    if(bt[i+1]) break;
+    i++;
+  }
+  if(i<slen-1) {
+    if(itype[i]==UJUMP||itype[i]==RJUMP||itype[i]==CJUMP||itype[i]==SJUMP) {
+      *value=constmap[i][hr];
+      return 1;
+    }
+    if(!bt[i+1]) {
+      if(itype[i+1]==UJUMP||itype[i+1]==RJUMP||itype[i+1]==CJUMP||itype[i+1]==SJUMP) {
+        // Load in delay slot, out-of-order execution
+        if(itype[i+2]==LOAD&&rs1[i+2]==reg&&rt1[i+2]==reg&&((regs[i+1].wasconst>>hr)&1))
+        {
+          #ifdef HOST_IMM_ADDR32
+          if(!using_tlb||((signed int)constmap[i][hr]+imm[i+2])<(signed int)0xC0000000) return 0;
+          #endif
+          // Precompute load address
+          *value=constmap[i][hr]+imm[i+2];
+          return 1;
+        }
+      }
+      if(itype[i+1]==LOAD&&rs1[i+1]==reg&&rt1[i+1]==reg)
+      {
+        #ifdef HOST_IMM_ADDR32
+        if(!using_tlb||((signed int)constmap[i][hr]+imm[i+1])<(signed int)0xC0000000) return 0;
+        #endif
+        // Precompute load address
+        *value=constmap[i][hr]+imm[i+1];
+        //printf("c=%x imm=%x\n",(int)constmap[i][hr],imm[i+1]);
+        return 1;
+      }
+    }
+  }
+  *value=constmap[i][hr];
+  //printf("c=%x\n",(int)constmap[i][hr]);
+  if(i==slen-1) return 1;
+  if(reg<64) {
+    return !((unneeded_reg[i+1]>>reg)&1);
+  }else{
+    return !((unneeded_reg_upper[i+1]>>reg)&1);
+  }
+}
+
+// Load registers with known constants
+void load_consts(signed char pre[],signed char regmap[],int is32,int i)
+{
+  int hr;
+  // Load 32-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
+      //if(entry[hr]!=regmap[hr]) {
+      if(i==0||!((regs[i-1].isconst>>hr)&1)||pre[hr]!=regmap[hr]||bt[i]) {
+        if(((regs[i].isconst>>hr)&1)&&regmap[hr]<64&&regmap[hr]>0) {
+          int value;
+          if(get_final_value(hr,i,&value)) {
+            if(value==0) {
+              emit_zeroreg(hr);
+            }
+            else {
+              emit_movimm(value,hr);
+            }
+          }
+        }
+      }
+    }
+  }
+  // Load 64-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
+      //if(entry[hr]!=regmap[hr]) {
+      if(i==0||!((regs[i-1].isconst>>hr)&1)||pre[hr]!=regmap[hr]||bt[i]) {
+        if(((regs[i].isconst>>hr)&1)&&regmap[hr]>64) {
+          if((is32>>(regmap[hr]&63))&1) {
+            int lr=get_reg(regmap,regmap[hr]-64);
+            assert(lr>=0);
+            emit_sarimm(lr,31,hr);
+          }
+          else
+          {
+            int value;
+            if(get_final_value(hr,i,&value)) {
+              if(value==0) {
+                emit_zeroreg(hr);
+              }
+              else {
+                emit_movimm(value,hr);
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+void load_all_consts(signed char regmap[],int is32,u_int dirty,int i)
+{
+  int hr;
+  // Load 32-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&regmap[hr]>=0&&((dirty>>hr)&1)) {
+      if(((regs[i].isconst>>hr)&1)&&regmap[hr]<64&&regmap[hr]>0) {
+        int value=constmap[i][hr];
+        if(value==0) {
+          emit_zeroreg(hr);
+        }
+        else {
+          emit_movimm(value,hr);
+        }
+      }
+    }
+  }
+  // Load 64-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&regmap[hr]>=0&&((dirty>>hr)&1)) {
+      if(((regs[i].isconst>>hr)&1)&&regmap[hr]>64) {
+        if((is32>>(regmap[hr]&63))&1) {
+          int lr=get_reg(regmap,regmap[hr]-64);
+          assert(lr>=0);
+          emit_sarimm(lr,31,hr);
+        }
+        else
+        {
+          int value=constmap[i][hr];
+          if(value==0) {
+            emit_zeroreg(hr);
+          }
+          else {
+            emit_movimm(value,hr);
+          }
+        }
+      }
+    }
+  }
+}
+
+// Write out all dirty registers (except cycle count)
+void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty)
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(i_regmap[hr]>0) {
+        if(i_regmap[hr]!=CCREG) {
+          if((i_dirty>>hr)&1) {
+            if(i_regmap[hr]<64) {
+              emit_storereg(i_regmap[hr],hr);
+              if( ((i_is32>>i_regmap[hr])&1) ) {
+                #ifdef DESTRUCTIVE_WRITEBACK
+                emit_sarimm(hr,31,hr);
+                emit_storereg(i_regmap[hr]|64,hr);
+                #else
+                emit_sarimm(hr,31,HOST_TEMPREG);
+                emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
+                #endif
+              }
+            }else{
+              if( !((i_is32>>(i_regmap[hr]&63))&1) ) {
+                emit_storereg(i_regmap[hr],hr);
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+// Write out dirty registers that we need to reload (pair with load_needed_regs)
+// This writes the registers not written by store_regs_bt
+void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+{
+  int hr;
+  int t=(addr-start)>>2;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(i_regmap[hr]>0) {
+        if(i_regmap[hr]!=CCREG) {
+          if(i_regmap[hr]==regs[t].regmap_entry[hr] && ((regs[t].dirty>>hr)&1) && !(((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+            if((i_dirty>>hr)&1) {
+              if(i_regmap[hr]<64) {
+                emit_storereg(i_regmap[hr],hr);
+                if( ((i_is32>>i_regmap[hr])&1) ) {
+                  #ifdef DESTRUCTIVE_WRITEBACK
+                  emit_sarimm(hr,31,hr);
+                  emit_storereg(i_regmap[hr]|64,hr);
+                  #else
+                  emit_sarimm(hr,31,HOST_TEMPREG);
+                  emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
+                  #endif
+                }
+              }else{
+                if( !((i_is32>>(i_regmap[hr]&63))&1) ) {
+                  emit_storereg(i_regmap[hr],hr);
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+// Load all registers (except cycle count)
+void load_all_regs(signed char i_regmap[])
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(i_regmap[hr]==0) {
+        emit_zeroreg(hr);
+      }
+      else
+      if(i_regmap[hr]>0 && i_regmap[hr]!=CCREG)
+      {
+        emit_loadreg(i_regmap[hr],hr);
+      }
+    }
+  }
+}
+
+// Load all current registers also needed by next instruction
+void load_needed_regs(signed char i_regmap[],signed char next_regmap[])
+{
+  int hr;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      if(get_reg(next_regmap,i_regmap[hr])>=0) {
+        if(i_regmap[hr]==0) {
+          emit_zeroreg(hr);
+        }
+        else
+        if(i_regmap[hr]>0 && i_regmap[hr]!=CCREG)
+        {
+          emit_loadreg(i_regmap[hr],hr);
+        }
+      }
+    }
+  }
+}
+
+// Load all regs, storing cycle count if necessary
+void load_regs_entry(int t)
+{
+  int hr;
+  if(is_ds[t]) emit_addimm(HOST_CCREG,CLOCK_DIVIDER,HOST_CCREG);
+  else if(ccadj[t]) emit_addimm(HOST_CCREG,-ccadj[t]*CLOCK_DIVIDER,HOST_CCREG);
+  if(regs[t].regmap_entry[HOST_CCREG]!=CCREG) {
+    emit_storereg(CCREG,HOST_CCREG);
+  }
+  // Load 32-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(regs[t].regmap_entry[hr]>=0&&regs[t].regmap_entry[hr]<64) {
+      if(regs[t].regmap_entry[hr]==0) {
+        emit_zeroreg(hr);
+      }
+      else if(regs[t].regmap_entry[hr]!=CCREG)
+      {
+        emit_loadreg(regs[t].regmap_entry[hr],hr);
+      }
+    }
+  }
+  // Load 64-bit regs
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(regs[t].regmap_entry[hr]>=64) {
+      assert(regs[t].regmap_entry[hr]!=64);
+      if((regs[t].was32>>(regs[t].regmap_entry[hr]&63))&1) {
+        int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
+        if(lr<0) {
+          emit_loadreg(regs[t].regmap_entry[hr],hr);
+        }
+        else
+        {
+          emit_sarimm(lr,31,hr);
+        }
+      }
+      else
+      {
+        emit_loadreg(regs[t].regmap_entry[hr],hr);
+      }
+    }
+  }
+}
+
+// Store dirty registers prior to branch
+void store_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+{
+  if(internal_branch(i_is32,addr))
+  {
+    int t=(addr-start)>>2;
+    int hr;
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG) {
+        if(i_regmap[hr]>0 && i_regmap[hr]!=CCREG) {
+          if(i_regmap[hr]!=regs[t].regmap_entry[hr] || !((regs[t].dirty>>hr)&1) || (((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+            if((i_dirty>>hr)&1) {
+              if(i_regmap[hr]<64) {
+                if(!((unneeded_reg[t]>>i_regmap[hr])&1)) {
+                  emit_storereg(i_regmap[hr],hr);
+                  if( ((i_is32>>i_regmap[hr])&1) && !((unneeded_reg_upper[t]>>i_regmap[hr])&1) ) {
+                    #ifdef DESTRUCTIVE_WRITEBACK
+                    emit_sarimm(hr,31,hr);
+                    emit_storereg(i_regmap[hr]|64,hr);
+                    #else
+                    emit_sarimm(hr,31,HOST_TEMPREG);
+                    emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
+                    #endif
+                  }
+                }
+              }else{
+                if( !((i_is32>>(i_regmap[hr]&63))&1) && !((unneeded_reg_upper[t]>>(i_regmap[hr]&63))&1) ) {
+                  emit_storereg(i_regmap[hr],hr);
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  else
+  {
+    // Branch out of this block, write out all dirty regs
+    wb_dirtys(i_regmap,i_is32,i_dirty);
+  }
+}
+
+// Load all needed registers for branch target
+void load_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+{
+  //if(addr>=start && addr<(start+slen*4))
+  if(internal_branch(i_is32,addr))
+  {
+    int t=(addr-start)>>2;
+    int hr;
+    // Store the cycle count before loading something else
+    if(i_regmap[HOST_CCREG]!=CCREG) {
+      assert(i_regmap[HOST_CCREG]==-1);
+    }
+    if(regs[t].regmap_entry[HOST_CCREG]!=CCREG) {
+      emit_storereg(CCREG,HOST_CCREG);
+    }
+    // Load 32-bit regs
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&regs[t].regmap_entry[hr]>=0&&regs[t].regmap_entry[hr]<64) {
+        #ifdef DESTRUCTIVE_WRITEBACK
+        if(i_regmap[hr]!=regs[t].regmap_entry[hr] || ( !((regs[t].dirty>>hr)&1) && ((i_dirty>>hr)&1) && (((i_is32&~unneeded_reg_upper[t])>>i_regmap[hr])&1) ) || (((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+        #else
+        if(i_regmap[hr]!=regs[t].regmap_entry[hr] ) {
+        #endif
+          if(regs[t].regmap_entry[hr]==0) {
+            emit_zeroreg(hr);
+          }
+          else if(regs[t].regmap_entry[hr]!=CCREG)
+          {
+            emit_loadreg(regs[t].regmap_entry[hr],hr);
+          }
+        }
+      }
+    }
+    //Load 64-bit regs
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&regs[t].regmap_entry[hr]>=64) {
+        if(i_regmap[hr]!=regs[t].regmap_entry[hr]) {
+          assert(regs[t].regmap_entry[hr]!=64);
+          if((i_is32>>(regs[t].regmap_entry[hr]&63))&1) {
+            int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
+            if(lr<0) {
+              emit_loadreg(regs[t].regmap_entry[hr],hr);
+            }
+            else
+            {
+              emit_sarimm(lr,31,hr);
+            }
+          }
+          else
+          {
+            emit_loadreg(regs[t].regmap_entry[hr],hr);
+          }
+        }
+        else if((i_is32>>(regs[t].regmap_entry[hr]&63))&1) {
+          int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
+          assert(lr>=0);
+          emit_sarimm(lr,31,hr);
+        }
+      }
+    }
+  }
+}
+
+int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+{
+  if(addr>=start && addr<start+slen*4-4)
+  {
+    int t=(addr-start)>>2;
+    int hr;
+    if(regs[t].regmap_entry[HOST_CCREG]!=CCREG) return 0;
+    for(hr=0;hr<HOST_REGS;hr++)
+    {
+      if(hr!=EXCLUDE_REG)
+      {
+        if(i_regmap[hr]!=regs[t].regmap_entry[hr])
+        {
+          if(regs[t].regmap_entry[hr]!=-1)
+          {
+            return 0;
+          }
+          else 
+          if((i_dirty>>hr)&1)
+          {
+            if(i_regmap[hr]<64)
+            {
+              if(!((unneeded_reg[t]>>i_regmap[hr])&1))
+                return 0;
+            }
+            else
+            {
+              if(!((unneeded_reg_upper[t]>>(i_regmap[hr]&63))&1))
+                return 0;
+            }
+          }
+        }
+        else // Same register but is it 32-bit or dirty?
+        if(i_regmap[hr]>=0)
+        {
+          if(!((regs[t].dirty>>hr)&1))
+          {
+            if((i_dirty>>hr)&1)
+            {
+              if(!((unneeded_reg[t]>>i_regmap[hr])&1))
+              {
+                //printf("%x: dirty no match\n",addr);
+                return 0;
+              }
+            }
+          }
+          if((((regs[t].was32^i_is32)&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)
+          {
+            //printf("%x: is32 no match\n",addr);
+            return 0;
+          }
+        }
+      }
+    }
+    //if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
+    if(requires_32bit[t]&~i_is32) return 0;
+    // Delay slots are not valid branch targets
+    //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
+    // Delay slots require additional processing, so do not match
+    if(is_ds[t]) return 0;
+  }
+  else
+  {
+    int hr;
+    for(hr=0;hr<HOST_REGS;hr++)
+    {
+      if(hr!=EXCLUDE_REG)
+      {
+        if(i_regmap[hr]>=0)
+        {
+          if(hr!=HOST_CCREG||i_regmap[hr]!=CCREG)
+          {
+            if((i_dirty>>hr)&1)
+            {
+              return 0;
+            }
+          }
+        }
+      }
+    }
+  }
+  return 1;
+}
+
+// Used when a branch jumps into the delay slot of another branch
+void ds_assemble_entry(int i)
+{
+  int t=(ba[i]-start)>>2;
+  if(!instr_addr[t]) instr_addr[t]=(u_int)out;
+  assem_debug("Assemble delay slot at %x\n",ba[i]);
+  assem_debug("<->\n");
+  if(regs[t].regmap_entry[HOST_CCREG]==CCREG&&regs[t].regmap[HOST_CCREG]!=CCREG)
+    wb_register(CCREG,regs[t].regmap_entry,regs[t].wasdirty,regs[t].was32);
+  load_regs(regs[t].regmap_entry,regs[t].regmap,regs[t].was32,rs1[t],rs2[t]);
+  address_generation(t,&regs[t],regs[t].regmap_entry);
+  if(itype[t]==STORE||itype[t]==STORELR||(opcode[t]&0x3b)==0x39)
+    load_regs(regs[t].regmap_entry,regs[t].regmap,regs[t].was32,INVCP,INVCP);
+  cop1_usable=0;
+  is_delayslot=0;
+  switch(itype[t]) {
+    case ALU:
+      alu_assemble(t,&regs[t]);break;
+    case IMM16:
+      imm16_assemble(t,&regs[t]);break;
+    case SHIFT:
+      shift_assemble(t,&regs[t]);break;
+    case SHIFTIMM:
+      shiftimm_assemble(t,&regs[t]);break;
+    case LOAD:
+      load_assemble(t,&regs[t]);break;
+    case LOADLR:
+      loadlr_assemble(t,&regs[t]);break;
+    case STORE:
+      store_assemble(t,&regs[t]);break;
+    case STORELR:
+      storelr_assemble(t,&regs[t]);break;
+    case COP0:
+      cop0_assemble(t,&regs[t]);break;
+    case COP1:
+      cop1_assemble(t,&regs[t]);break;
+    case C1LS:
+      c1ls_assemble(t,&regs[t]);break;
+    case FCONV:
+      fconv_assemble(t,&regs[t]);break;
+    case FLOAT:
+      float_assemble(t,&regs[t]);break;
+    case FCOMP:
+      fcomp_assemble(t,&regs[t]);break;
+    case MULTDIV:
+      multdiv_assemble(t,&regs[t]);break;
+    case MOV:
+      mov_assemble(t,&regs[t]);break;
+    case SYSCALL:
+    case SPAN:
+    case UJUMP:
+    case RJUMP:
+    case CJUMP:
+    case SJUMP:
+    case FJUMP:
+      printf("Jump in the delay slot.  This is probably a bug.\n");
+  }
+  store_regs_bt(regs[t].regmap,regs[t].is32,regs[t].dirty,ba[i]+4);
+  load_regs_bt(regs[t].regmap,regs[t].is32,regs[t].dirty,ba[i]+4);
+  if(internal_branch(regs[t].is32,ba[i]+4))
+    assem_debug("branch: internal\n");
+  else
+    assem_debug("branch: external\n");
+  assert(internal_branch(regs[t].is32,ba[i]+4));
+  add_to_linker((int)out,ba[i]+4,internal_branch(regs[t].is32,ba[i]+4));
+  emit_jmp(0);
+}
+
+void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
+{
+  int count;
+  int jaddr;
+  int idle=0;
+  if(itype[i]==RJUMP)
+  {
+    *adj=0;
+  }
+  //if(ba[i]>=start && ba[i]<(start+slen*4))
+  if(internal_branch(branch_regs[i].is32,ba[i]))
+  {
+    int t=(ba[i]-start)>>2;
+    if(is_ds[t]) *adj=-1; // Branch into delay slot adds an extra cycle
+    else *adj=ccadj[t];
+  }
+  else
+  {
+    *adj=0;
+  }
+  count=ccadj[i];
+  if(taken==TAKEN && i==(ba[i]-start)>>2 && source[i+1]==0) {
+    // Idle loop
+    if(count&1) emit_addimm_and_set_flags(2*(count+2),HOST_CCREG);
+    idle=(int)out;
+    //emit_subfrommem(&idlecount,HOST_CCREG); // Count idle cycles
+    emit_andimm(HOST_CCREG,3,HOST_CCREG);
+    jaddr=(int)out;
+    emit_jmp(0);
+  }
+  else if(*adj==0||invert) {
+    emit_addimm_and_set_flags(CLOCK_DIVIDER*(count+2),HOST_CCREG);
+    jaddr=(int)out;
+    emit_jns(0);
+  }
+  else
+  {
+    emit_cmpimm(HOST_CCREG,-2*(count+2));
+    jaddr=(int)out;
+    emit_jns(0);
+  }
+  add_stub(CC_STUB,jaddr,idle?idle:(int)out,(*adj==0||invert||idle)?0:(count+2),i,addr,taken,0);
+}
+
+void do_ccstub(int n)
+{
+  literal_pool(256);
+  assem_debug("do_ccstub %x\n",start+stubs[n][4]*4);
+  set_jump_target(stubs[n][1],(int)out);
+  int i=stubs[n][4];
+  if(stubs[n][6]==NULLDS) {
+    // Delay slot instruction is nullified ("likely" branch)
+    wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+  }
+  else if(stubs[n][6]!=TAKEN) {
+    wb_dirtys(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty);
+  }
+  else {
+    if(internal_branch(branch_regs[i].is32,ba[i]))
+      wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  }
+  if(stubs[n][5]!=-1)
+  {
+    // Save PC as return address
+    emit_movimm(stubs[n][5],EAX);
+    emit_writeword(EAX,(int)&pcaddr);
+  }
+  else
+  {
+    // Return address depends on which way the branch goes
+    if(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      int s1l=get_reg(branch_regs[i].regmap,rs1[i]);
+      int s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
+      int s2l=get_reg(branch_regs[i].regmap,rs2[i]);
+      int s2h=get_reg(branch_regs[i].regmap,rs2[i]|64);
+      if(rs1[i]==0)
+      {
+        s1l=s2l;s1h=s2h;
+        s2l=s2h=-1;
+      }
+      else if(rs2[i]==0)
+      {
+        s2l=s2h=-1;
+      }
+      if((branch_regs[i].is32>>rs1[i])&(branch_regs[i].is32>>rs2[i])&1) {
+        s1h=s2h=-1;
+      }
+      assert(s1l>=0);
+      #ifdef DESTRUCTIVE_WRITEBACK
+      if(rs1[i]) {
+        if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs1[i])&1)
+          emit_loadreg(rs1[i],s1l);
+      } 
+      else {
+        if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs2[i])&1)
+          emit_loadreg(rs2[i],s1l);
+      }
+      if(s2l>=0)
+        if((branch_regs[i].dirty>>s2l)&(branch_regs[i].is32>>rs2[i])&1)
+          emit_loadreg(rs2[i],s2l);
+      #endif
+      int hr=0;
+      int addr,alt,ntaddr;
+      while(hr<HOST_REGS)
+      {
+        if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
+           (branch_regs[i].regmap[hr]&63)!=rs1[i] &&
+           (branch_regs[i].regmap[hr]&63)!=rs2[i] )
+        {
+          addr=hr++;break;
+        }
+        hr++;
+      }
+      while(hr<HOST_REGS)
+      {
+        if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
+           (branch_regs[i].regmap[hr]&63)!=rs1[i] &&
+           (branch_regs[i].regmap[hr]&63)!=rs2[i] )
+        {
+          alt=hr++;break;
+        }
+        hr++;
+      }
+      if((opcode[i]&0x2E)==6) // BLEZ/BGTZ needs another register
+      {
+        while(hr<HOST_REGS)
+        {
+          if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
+             (branch_regs[i].regmap[hr]&63)!=rs1[i] &&
+             (branch_regs[i].regmap[hr]&63)!=rs2[i] )
+          {
+            ntaddr=hr;break;
+          }
+          hr++;
+        }
+        assert(hr<HOST_REGS);
+      }
+      if((opcode[i]&0x2f)==4) // BEQ
+      {
+        #ifdef HAVE_CMOV_IMM
+        if(s1h<0) {
+          if(s2l>=0) emit_cmp(s1l,s2l);
+          else emit_test(s1l,s1l);
+          emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
+        }
+        else
+        #endif
+        {
+          emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+          if(s1h>=0) {
+            if(s2h>=0) emit_cmp(s1h,s2h);
+            else emit_test(s1h,s1h);
+            emit_cmovne_reg(alt,addr);
+          }
+          if(s2l>=0) emit_cmp(s1l,s2l);
+          else emit_test(s1l,s1l);
+          emit_cmovne_reg(alt,addr);
+        }
+      }
+      if((opcode[i]&0x2f)==5) // BNE
+      {
+        #ifdef HAVE_CMOV_IMM
+        if(s1h<0) {
+          if(s2l>=0) emit_cmp(s1l,s2l);
+          else emit_test(s1l,s1l);
+          emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+        }
+        else
+        #endif
+        {
+          emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+          if(s1h>=0) {
+            if(s2h>=0) emit_cmp(s1h,s2h);
+            else emit_test(s1h,s1h);
+            emit_cmovne_reg(alt,addr);
+          }
+          if(s2l>=0) emit_cmp(s1l,s2l);
+          else emit_test(s1l,s1l);
+          emit_cmovne_reg(alt,addr);
+        }
+      }
+      if((opcode[i]&0x2f)==6) // BLEZ
+      {
+        //emit_movimm(ba[i],alt);
+        //emit_movimm(start+i*4+8,addr);
+        emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+        emit_cmpimm(s1l,1);
+        if(s1h>=0) emit_mov(addr,ntaddr);
+        emit_cmovl_reg(alt,addr);
+        if(s1h>=0) {
+          emit_test(s1h,s1h);
+          emit_cmovne_reg(ntaddr,addr);
+          emit_cmovs_reg(alt,addr);
+        }
+      }
+      if((opcode[i]&0x2f)==7) // BGTZ
+      {
+        //emit_movimm(ba[i],addr);
+        //emit_movimm(start+i*4+8,ntaddr);
+        emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr);
+        emit_cmpimm(s1l,1);
+        if(s1h>=0) emit_mov(addr,alt);
+        emit_cmovl_reg(ntaddr,addr);
+        if(s1h>=0) {
+          emit_test(s1h,s1h);
+          emit_cmovne_reg(alt,addr);
+          emit_cmovs_reg(ntaddr,addr);
+        }
+      }
+      if((opcode[i]==1)&&(opcode2[i]&0x2D)==0) // BLTZ
+      {
+        //emit_movimm(ba[i],alt);
+        //emit_movimm(start+i*4+8,addr);
+        emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+        if(s1h>=0) emit_test(s1h,s1h);
+        else emit_test(s1l,s1l);
+        emit_cmovs_reg(alt,addr);
+      }
+      if((opcode[i]==1)&&(opcode2[i]&0x2D)==1) // BGEZ
+      {
+        //emit_movimm(ba[i],addr);
+        //emit_movimm(start+i*4+8,alt);
+        emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+        if(s1h>=0) emit_test(s1h,s1h);
+        else emit_test(s1l,s1l);
+        emit_cmovs_reg(alt,addr);
+      }
+      if(opcode[i]==0x11 && opcode2[i]==0x08 ) {
+        if(source[i]&0x10000) // BC1T
+        {
+          //emit_movimm(ba[i],alt);
+          //emit_movimm(start+i*4+8,addr);
+          emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+          emit_testimm(s1l,0x800000);
+          emit_cmovne_reg(alt,addr);
+        }
+        else // BC1F
+        {
+          //emit_movimm(ba[i],addr);
+          //emit_movimm(start+i*4+8,alt);
+          emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+          emit_testimm(s1l,0x800000);
+          emit_cmovne_reg(alt,addr);
+        }
+      }
+      emit_writeword(addr,(int)&pcaddr);
+    }
+    else
+    if(itype[i]==RJUMP)
+    {
+      int r=get_reg(branch_regs[i].regmap,rs1[i]);
+      if(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]) {
+        r=get_reg(branch_regs[i].regmap,RTEMP);
+      }
+      emit_writeword(r,(int)&pcaddr);
+    }
+    else {printf("Unknown branch type in do_ccstub\n");exit(1);}
+  }
+  // Update cycle count
+  assert(branch_regs[i].regmap[HOST_CCREG]==CCREG||branch_regs[i].regmap[HOST_CCREG]==-1);
+  if(stubs[n][3]) emit_addimm(HOST_CCREG,CLOCK_DIVIDER*stubs[n][3],HOST_CCREG);
+  emit_call((int)cc_interrupt);
+  if(stubs[n][3]) emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*stubs[n][3],HOST_CCREG);
+  if(stubs[n][6]==TAKEN) {
+    if(internal_branch(branch_regs[i].is32,ba[i]))
+      load_needed_regs(branch_regs[i].regmap,regs[(ba[i]-start)>>2].regmap_entry);
+    else if(itype[i]==RJUMP) {
+      if(get_reg(branch_regs[i].regmap,RTEMP)>=0)
+        emit_readword((int)&pcaddr,get_reg(branch_regs[i].regmap,RTEMP));
+      else
+        emit_loadreg(rs1[i],get_reg(branch_regs[i].regmap,rs1[i]));
+    }
+  }else if(stubs[n][6]==NOTTAKEN) {
+    if(i<slen-2) load_needed_regs(branch_regs[i].regmap,regmap_pre[i+2]);
+    else load_all_regs(branch_regs[i].regmap);
+  }else if(stubs[n][6]==NULLDS) {
+    // Delay slot instruction is nullified ("likely" branch)
+    if(i<slen-2) load_needed_regs(regs[i].regmap,regmap_pre[i+2]);
+    else load_all_regs(regs[i].regmap);
+  }else{
+    load_all_regs(branch_regs[i].regmap);
+  }
+  emit_jmp(stubs[n][2]); // return address
+  
+  /* This works but uses a lot of memory...
+  emit_readword((int)&last_count,ECX);
+  emit_add(HOST_CCREG,ECX,EAX);
+  emit_writeword(EAX,(int)&Count);
+  emit_call((int)gen_interupt);
+  emit_readword((int)&Count,HOST_CCREG);
+  emit_readword((int)&next_interupt,EAX);
+  emit_readword((int)&pending_exception,EBX);
+  emit_writeword(EAX,(int)&last_count);
+  emit_sub(HOST_CCREG,EAX,HOST_CCREG);
+  emit_test(EBX,EBX);
+  int jne_instr=(int)out;
+  emit_jne(0);
+  if(stubs[n][3]) emit_addimm(HOST_CCREG,-2*stubs[n][3],HOST_CCREG);
+  load_all_regs(branch_regs[i].regmap);
+  emit_jmp(stubs[n][2]); // return address
+  set_jump_target(jne_instr,(int)out);
+  emit_readword((int)&pcaddr,EAX);
+  // Call get_addr_ht instead of doing the hash table here.
+  // This code is executed infrequently and takes up a lot of space
+  // so smaller is better.
+  emit_storereg(CCREG,HOST_CCREG);
+  emit_pushreg(EAX);
+  emit_call((int)get_addr_ht);
+  emit_loadreg(CCREG,HOST_CCREG);
+  emit_addimm(ESP,4,ESP);
+  emit_jmpreg(EAX);*/
+}
+
+add_to_linker(int addr,int target,int ext)
+{
+  link_addr[linkcount][0]=addr;
+  link_addr[linkcount][1]=target;
+  link_addr[linkcount][2]=ext;  
+  linkcount++;
+}
+
+void ujump_assemble(int i,struct regstat *i_regs)
+{
+  signed char *i_regmap=i_regs->regmap;
+  if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+  address_generation(i+1,i_regs,regs[i].regmap_entry);
+  #ifdef REG_PREFETCH
+  int temp=get_reg(branch_regs[i].regmap,PTEMP);
+  if(rt1[i]==31&&temp>=0) 
+  {
+    int return_address=start+i*4+8;
+    if(get_reg(branch_regs[i].regmap,31)>0) 
+    if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+  }
+  #endif
+  ds_assemble(i+1,i_regs);
+  uint64_t bc_unneeded=branch_regs[i].u;
+  uint64_t bc_unneeded_upper=branch_regs[i].uu;
+  bc_unneeded|=1|(1LL<<rt1[i]);
+  bc_unneeded_upper|=1|(1LL<<rt1[i]);
+  wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                bc_unneeded,bc_unneeded_upper);
+  load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+  if(rt1[i]==31) {
+    int rt;
+    unsigned int return_address;
+    assert(rt1[i+1]!=31);
+    assert(rt2[i+1]!=31);
+    rt=get_reg(branch_regs[i].regmap,31);
+    assem_debug("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+    //assert(rt>=0);
+    return_address=start+i*4+8;
+    if(rt>=0) {
+      #ifdef USE_MINI_HT
+      if(internal_branch(branch_regs[i].is32,return_address)) {
+        int temp=rt+1;
+        if(temp==EXCLUDE_REG||temp>=HOST_REGS||
+           branch_regs[i].regmap[temp]>=0)
+        {
+          temp=get_reg(branch_regs[i].regmap,-1);
+        }
+        #ifdef HOST_TEMPREG
+        if(temp<0) temp=HOST_TEMPREG;
+        #endif
+        if(temp>=0) do_miniht_insert(return_address,rt,temp);
+        else emit_movimm(return_address,rt);
+      }
+      else
+      #endif
+      {
+        #ifdef REG_PREFETCH
+        if(temp>=0) 
+        {
+          if(i_regmap[temp]!=PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+        }
+        #endif
+        emit_movimm(return_address,rt); // PC into link register
+        #ifdef IMM_PREFETCH
+        emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+        #endif
+      }
+    }
+  }
+  int cc,adj;
+  cc=get_reg(branch_regs[i].regmap,CCREG);
+  assert(cc==HOST_CCREG);
+  store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  #ifdef REG_PREFETCH
+  if(rt1[i]==31&&temp>=0) emit_prefetchreg(temp);
+  #endif
+  do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
+  if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+  load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  if(internal_branch(branch_regs[i].is32,ba[i]))
+    assem_debug("branch: internal\n");
+  else
+    assem_debug("branch: external\n");
+  if(internal_branch(branch_regs[i].is32,ba[i])&&is_ds[(ba[i]-start)>>2]) {
+    ds_assemble_entry(i);
+  }
+  else {
+    add_to_linker((int)out,ba[i],internal_branch(branch_regs[i].is32,ba[i]));
+    emit_jmp(0);
+  }
+}
+
+void rjump_assemble(int i,struct regstat *i_regs)
+{
+  signed char *i_regmap=i_regs->regmap;
+  int temp;
+  int rs,cc,adj;
+  rs=get_reg(branch_regs[i].regmap,rs1[i]);
+  assert(rs>=0);
+  if(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]) {
+    // Delay slot abuse, make a copy of the branch address register
+    temp=get_reg(branch_regs[i].regmap,RTEMP);
+    assert(temp>=0);
+    assert(regs[i].regmap[temp]==RTEMP);
+    emit_mov(rs,temp);
+    rs=temp;
+  }
+  address_generation(i+1,i_regs,regs[i].regmap_entry);
+  #ifdef REG_PREFETCH
+  if(rt1[i]==31) 
+  {
+    if((temp=get_reg(branch_regs[i].regmap,PTEMP))>=0) {
+      int return_address=start+i*4+8;
+      if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+    }
+  }
+  #endif
+  #ifdef USE_MINI_HT
+  if(rs1[i]==31) {
+    int rh=get_reg(regs[i].regmap,RHASH);
+    if(rh>=0) do_preload_rhash(rh);
+  }
+  #endif
+  ds_assemble(i+1,i_regs);
+  uint64_t bc_unneeded=branch_regs[i].u;
+  uint64_t bc_unneeded_upper=branch_regs[i].uu;
+  bc_unneeded|=1|(1LL<<rt1[i]);
+  bc_unneeded_upper|=1|(1LL<<rt1[i]);
+  bc_unneeded&=~(1LL<<rs1[i]);
+  wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                bc_unneeded,bc_unneeded_upper);
+  load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],CCREG);
+  if(rt1[i]==31) {
+    int rt,return_address;
+    assert(rt1[i+1]!=31);
+    assert(rt2[i+1]!=31);
+    rt=get_reg(branch_regs[i].regmap,31);
+    assem_debug("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+    assert(rt>=0);
+    return_address=start+i*4+8;
+    #ifdef REG_PREFETCH
+    if(temp>=0) 
+    {
+      if(i_regmap[temp]!=PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+    }
+    #endif
+    emit_movimm(return_address,rt); // PC into link register
+    #ifdef IMM_PREFETCH
+    emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+    #endif
+  }
+  cc=get_reg(branch_regs[i].regmap,CCREG);
+  assert(cc==HOST_CCREG);
+  #ifdef USE_MINI_HT
+  int rh=get_reg(branch_regs[i].regmap,RHASH);
+  int ht=get_reg(branch_regs[i].regmap,RHTBL);
+  if(rs1[i]==31) {
+    if(regs[i].regmap[rh]!=RHASH) do_preload_rhash(rh);
+    do_preload_rhtbl(ht);
+    do_rhash(rs,rh);
+  }
+  #endif
+  store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,-1);
+  #ifdef DESTRUCTIVE_WRITEBACK
+  if((branch_regs[i].dirty>>rs)&(branch_regs[i].is32>>rs1[i])&1) {
+    if(rs1[i]!=rt1[i+1]&&rs1[i]!=rt2[i+1]) {
+      emit_loadreg(rs1[i],rs);
+    }
+  }
+  #endif
+  #ifdef REG_PREFETCH
+  if(rt1[i]==31&&temp>=0) emit_prefetchreg(temp);
+  #endif
+  #ifdef USE_MINI_HT
+  if(rs1[i]==31) {
+    do_miniht_load(ht,rh);
+  }
+  #endif
+  //do_cc(i,branch_regs[i].regmap,&adj,-1,TAKEN);
+  //if(adj) emit_addimm(cc,2*(ccadj[i]+2-adj),cc); // ??? - Shouldn't happen
+  //assert(adj==0);
+  emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),HOST_CCREG);
+  add_stub(CC_STUB,(int)out,jump_vaddr_reg[rs],0,i,-1,TAKEN,0);
+  emit_jns(0);
+  //load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,-1);
+  #ifdef USE_MINI_HT
+  if(rs1[i]==31) {
+    do_miniht_jump(rs,rh,ht);
+  }
+  else
+  #endif
+  {
+    //if(rs!=EAX) emit_mov(rs,EAX);
+    //emit_jmp((int)jump_vaddr_eax);
+    emit_jmp(jump_vaddr_reg[rs]);
+  }
+  /* Check hash table
+  temp=!rs;
+  emit_mov(rs,temp);
+  emit_shrimm(rs,16,rs);
+  emit_xor(temp,rs,rs);
+  emit_movzwl_reg(rs,rs);
+  emit_shlimm(rs,4,rs);
+  emit_cmpmem_indexed((int)hash_table,rs,temp);
+  emit_jne((int)out+14);
+  emit_readword_indexed((int)hash_table+4,rs,rs);
+  emit_jmpreg(rs);
+  emit_cmpmem_indexed((int)hash_table+8,rs,temp);
+  emit_addimm_no_flags(8,rs);
+  emit_jeq((int)out-17);
+  // No hit on hash table, call compiler
+  emit_pushreg(temp);
+//DEBUG >
+#ifdef DEBUG_CYCLE_COUNT
+  emit_readword((int)&last_count,ECX);
+  emit_add(HOST_CCREG,ECX,HOST_CCREG);
+  emit_readword((int)&next_interupt,ECX);
+  emit_writeword(HOST_CCREG,(int)&Count);
+  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
+  emit_writeword(ECX,(int)&last_count);
+#endif
+//DEBUG <
+  emit_storereg(CCREG,HOST_CCREG);
+  emit_call((int)get_addr);
+  emit_loadreg(CCREG,HOST_CCREG);
+  emit_addimm(ESP,4,ESP);
+  emit_jmpreg(EAX);*/
+  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+  if(rt1[i]!=31&&i<slen-2&&(((u_int)out)&7)) emit_mov(13,13);
+  #endif
+}
+
+void cjump_assemble(int i,struct regstat *i_regs)
+{
+  signed char *i_regmap=i_regs->regmap;
+  int cc;
+  int match;
+  match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  assem_debug("match=%d\n",match);
+  int s1h,s1l,s2h,s2l;
+  int prev_cop1_usable=cop1_usable;
+  int unconditional=0,nop=0;
+  int only32=0;
+  int ooo=1;
+  int invert=0;
+  int internal=internal_branch(branch_regs[i].is32,ba[i]);
+  if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+  if(likely[i]) ooo=0;
+  if(!match) invert=1;
+  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+  if(i>(ba[i]-start)>>2) invert=1;
+  #endif
+    
+  if(ooo)
+    if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))||
+       (rs2[i]&&(rs2[i]==rt1[i+1]||rs2[i]==rt2[i+1])))
+  {
+    // Write-after-read dependency prevents out of order execution
+    // First test branch condition, then execute delay slot, then branch
+    ooo=0;
+  }
+
+  if(ooo) {
+    s1l=get_reg(branch_regs[i].regmap,rs1[i]);
+    s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
+    s2l=get_reg(branch_regs[i].regmap,rs2[i]);
+    s2h=get_reg(branch_regs[i].regmap,rs2[i]|64);
+  }
+  else {
+    s1l=get_reg(i_regmap,rs1[i]);
+    s1h=get_reg(i_regmap,rs1[i]|64);
+    s2l=get_reg(i_regmap,rs2[i]);
+    s2h=get_reg(i_regmap,rs2[i]|64);
+  }
+  if(rs1[i]==0&&rs2[i]==0)
+  {
+    if(opcode[i]&1) nop=1;
+    else unconditional=1;
+    //assert(opcode[i]!=5);
+    //assert(opcode[i]!=7);
+    //assert(opcode[i]!=0x15);
+    //assert(opcode[i]!=0x17);
+  }
+  else if(rs1[i]==0)
+  {
+    s1l=s2l;s1h=s2h;
+    s2l=s2h=-1;
+    only32=(regs[i].was32>>rs2[i])&1;
+  }
+  else if(rs2[i]==0)
+  {
+    s2l=s2h=-1;
+    only32=(regs[i].was32>>rs1[i])&1;
+  }
+  else {
+    only32=(regs[i].was32>>rs1[i])&(regs[i].was32>>rs2[i])&1;
+  }
+
+  if(ooo) {
+    // Out of order execution (delay slot first)
+    //printf("OOOE\n");
+    address_generation(i+1,i_regs,regs[i].regmap_entry);
+    ds_assemble(i+1,i_regs);
+    int adj;
+    uint64_t bc_unneeded=branch_regs[i].u;
+    uint64_t bc_unneeded_upper=branch_regs[i].uu;
+    bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
+    bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
+    bc_unneeded|=1;
+    bc_unneeded_upper|=1;
+    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                  bc_unneeded,bc_unneeded_upper);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs2[i]);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+    cc=get_reg(branch_regs[i].regmap,CCREG);
+    assert(cc==HOST_CCREG);
+    if(unconditional) 
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+    //do_cc(i,branch_regs[i].regmap,&adj,unconditional?ba[i]:-1,unconditional);
+    //assem_debug("cycle count (adj)\n");
+    if(unconditional) {
+      do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
+      if(i!=(ba[i]-start)>>2 || source[i+1]!=0) {
+        if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+        load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+        if(internal)
+          assem_debug("branch: internal\n");
+        else
+          assem_debug("branch: external\n");
+        if(internal&&is_ds[(ba[i]-start)>>2]) {
+          ds_assemble_entry(i);
+        }
+        else {
+          add_to_linker((int)out,ba[i],internal);
+          emit_jmp(0);
+        }
+        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+        if(((u_int)out)&7) emit_addnop(0);
+        #endif
+      }
+    }
+    else if(nop) {
+      emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
+      int jaddr=(int)out;
+      emit_jns(0);
+      add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+    }
+    else {
+      int taken=0,nottaken=0,nottaken1=0;
+      do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
+      if(adj&&!invert) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+      if(!only32)
+      {
+        assert(s1h>=0);
+        if(opcode[i]==4) // BEQ
+        {
+          if(s2h>=0) emit_cmp(s1h,s2h);
+          else emit_test(s1h,s1h);
+          nottaken1=(int)out;
+          emit_jne(1);
+        }
+        if(opcode[i]==5) // BNE
+        {
+          if(s2h>=0) emit_cmp(s1h,s2h);
+          else emit_test(s1h,s1h);
+          if(invert) taken=(int)out;
+          else add_to_linker((int)out,ba[i],internal);
+          emit_jne(0);
+        }
+        if(opcode[i]==6) // BLEZ
+        {
+          emit_test(s1h,s1h);
+          if(invert) taken=(int)out;
+          else add_to_linker((int)out,ba[i],internal);
+          emit_js(0);
+          nottaken1=(int)out;
+          emit_jne(1);
+        }
+        if(opcode[i]==7) // BGTZ
+        {
+          emit_test(s1h,s1h);
+          nottaken1=(int)out;
+          emit_js(1);
+          if(invert) taken=(int)out;
+          else add_to_linker((int)out,ba[i],internal);
+          emit_jne(0);
+        }
+      } // if(!only32)
+          
+      //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+      assert(s1l>=0);
+      if(opcode[i]==4) // BEQ
+      {
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        if(invert){
+          nottaken=(int)out;
+          emit_jne(1);
+        }else{
+          add_to_linker((int)out,ba[i],internal);
+          emit_jeq(0);
+        }
+      }
+      if(opcode[i]==5) // BNE
+      {
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        if(invert){
+          nottaken=(int)out;
+          emit_jeq(1);
+        }else{
+          add_to_linker((int)out,ba[i],internal);
+          emit_jne(0);
+        }
+      }
+      if(opcode[i]==6) // BLEZ
+      {
+        emit_cmpimm(s1l,1);
+        if(invert){
+          nottaken=(int)out;
+          emit_jge(1);
+        }else{
+          add_to_linker((int)out,ba[i],internal);
+          emit_jl(0);
+        }
+      }
+      if(opcode[i]==7) // BGTZ
+      {
+        emit_cmpimm(s1l,1);
+        if(invert){
+          nottaken=(int)out;
+          emit_jl(1);
+        }else{
+          add_to_linker((int)out,ba[i],internal);
+          emit_jge(0);
+        }
+      }
+      if(invert) {
+        if(taken) set_jump_target(taken,(int)out);
+        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+        if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) {
+          if(adj) {
+            emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
+            add_to_linker((int)out,ba[i],internal);
+          }else{
+            emit_addnop(13);
+            add_to_linker((int)out,ba[i],internal*2);
+          }
+          emit_jmp(0);
+        }else
+        #endif
+        {
+          if(adj) emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
+          store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+          load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+          if(internal)
+            assem_debug("branch: internal\n");
+          else
+            assem_debug("branch: external\n");
+          if(internal&&is_ds[(ba[i]-start)>>2]) {
+            ds_assemble_entry(i);
+          }
+          else {
+            add_to_linker((int)out,ba[i],internal);
+            emit_jmp(0);
+          }
+        }
+        set_jump_target(nottaken,(int)out);
+      }
+
+      if(nottaken1) set_jump_target(nottaken1,(int)out);
+      if(adj) {
+        if(!invert) emit_addimm(cc,CLOCK_DIVIDER*adj,cc);
+      }
+    } // (!unconditional)
+  } // if(ooo)
+  else
+  {
+    // In-order execution (branch first)
+    //if(likely[i]) printf("IOL\n");
+    //else
+    //printf("IOE\n");
+    int taken=0,nottaken=0,nottaken1=0;
+    if(!unconditional&&!nop) {
+      if(!only32)
+      {
+        assert(s1h>=0);
+        if((opcode[i]&0x2f)==4) // BEQ
+        {
+          if(s2h>=0) emit_cmp(s1h,s2h);
+          else emit_test(s1h,s1h);
+          nottaken1=(int)out;
+          emit_jne(2);
+        }
+        if((opcode[i]&0x2f)==5) // BNE
+        {
+          if(s2h>=0) emit_cmp(s1h,s2h);
+          else emit_test(s1h,s1h);
+          taken=(int)out;
+          emit_jne(1);
+        }
+        if((opcode[i]&0x2f)==6) // BLEZ
+        {
+          emit_test(s1h,s1h);
+          taken=(int)out;
+          emit_js(1);
+          nottaken1=(int)out;
+          emit_jne(2);
+        }
+        if((opcode[i]&0x2f)==7) // BGTZ
+        {
+          emit_test(s1h,s1h);
+          nottaken1=(int)out;
+          emit_js(2);
+          taken=(int)out;
+          emit_jne(1);
+        }
+      } // if(!only32)
+          
+      //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+      assert(s1l>=0);
+      if((opcode[i]&0x2f)==4) // BEQ
+      {
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        nottaken=(int)out;
+        emit_jne(2);
+      }
+      if((opcode[i]&0x2f)==5) // BNE
+      {
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        nottaken=(int)out;
+        emit_jeq(2);
+      }
+      if((opcode[i]&0x2f)==6) // BLEZ
+      {
+        emit_cmpimm(s1l,1);
+        nottaken=(int)out;
+        emit_jge(2);
+      }
+      if((opcode[i]&0x2f)==7) // BGTZ
+      {
+        emit_cmpimm(s1l,1);
+        nottaken=(int)out;
+        emit_jl(2);
+      }
+    } // if(!unconditional)
+    int adj;
+    uint64_t ds_unneeded=branch_regs[i].u;
+    uint64_t ds_unneeded_upper=branch_regs[i].uu;
+    ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+    ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+    if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+    ds_unneeded|=1;
+    ds_unneeded_upper|=1;
+    // branch taken
+    if(!nop) {
+      if(taken) set_jump_target(taken,(int)out);
+      assem_debug("1:\n");
+      wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                    ds_unneeded,ds_unneeded_upper);
+      // load regs
+      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+      address_generation(i+1,&branch_regs[i],0);
+      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
+      ds_assemble(i+1,&branch_regs[i]);
+      cc=get_reg(branch_regs[i].regmap,CCREG);
+      if(cc==-1) {
+        emit_loadreg(CCREG,cc=HOST_CCREG);
+        // CHECK: Is the following instruction (fall thru) allocated ok?
+      }
+      assert(cc==HOST_CCREG);
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
+      assem_debug("cycle count (adj)\n");
+      if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+      load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      if(internal)
+        assem_debug("branch: internal\n");
+      else
+        assem_debug("branch: external\n");
+      if(internal&&is_ds[(ba[i]-start)>>2]) {
+        ds_assemble_entry(i);
+      }
+      else {
+        add_to_linker((int)out,ba[i],internal);
+        emit_jmp(0);
+      }
+    }
+    // branch not taken
+    cop1_usable=prev_cop1_usable;
+    if(!unconditional) {
+      if(nottaken1) set_jump_target(nottaken1,(int)out);
+      set_jump_target(nottaken,(int)out);
+      assem_debug("2:\n");
+      if(!likely[i]) {
+        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                      ds_unneeded,ds_unneeded_upper);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+        address_generation(i+1,&branch_regs[i],0);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+        ds_assemble(i+1,&branch_regs[i]);
+      }
+      cc=get_reg(branch_regs[i].regmap,CCREG);
+      if(cc==-1&&!likely[i]) {
+        // Cycle count isn't in a register, temporarily load it then write it out
+        emit_loadreg(CCREG,HOST_CCREG);
+        emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),HOST_CCREG);
+        int jaddr=(int)out;
+        emit_jns(0);
+        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+        emit_storereg(CCREG,HOST_CCREG);
+      }
+      else{
+        cc=get_reg(i_regmap,CCREG);
+        assert(cc==HOST_CCREG);
+        emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
+        int jaddr=(int)out;
+        emit_jns(0);
+        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+      }
+    }
+  }
+}
+
+void sjump_assemble(int i,struct regstat *i_regs)
+{
+  signed char *i_regmap=i_regs->regmap;
+  int cc;
+  int match;
+  match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  assem_debug("smatch=%d\n",match);
+  int s1h,s1l;
+  int prev_cop1_usable=cop1_usable;
+  int unconditional=0,nevertaken=0;
+  int only32=0;
+  int ooo=1;
+  int invert=0;
+  int internal=internal_branch(branch_regs[i].is32,ba[i]);
+  if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+  if(likely[i]) ooo=0;
+  if(!match) invert=1;
+  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+  if(i>(ba[i]-start)>>2) invert=1;
+  #endif
+
+  //if(opcode2[i]>=0x10) return; // FIXME (BxxZAL)
+  assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL)
+
+  if(ooo)
+    if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))
+  {
+    // Write-after-read dependency prevents out of order execution
+    // First test branch condition, then execute delay slot, then branch
+    ooo=0;
+  }
+  // TODO: Conditional branches w/link must execute in-order so that
+  // condition test and write to r31 occur before cycle count test
+
+  if(ooo) {
+    s1l=get_reg(branch_regs[i].regmap,rs1[i]);
+    s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
+  }
+  else {
+    s1l=get_reg(i_regmap,rs1[i]);
+    s1h=get_reg(i_regmap,rs1[i]|64);
+  }
+  if(rs1[i]==0)
+  {
+    if(opcode2[i]&1) unconditional=1;
+    else nevertaken=1;
+    // These are never taken (r0 is never less than zero)
+    //assert(opcode2[i]!=0);
+    //assert(opcode2[i]!=2);
+    //assert(opcode2[i]!=0x10);
+    //assert(opcode2[i]!=0x12);
+  }
+  else {
+    only32=(regs[i].was32>>rs1[i])&1;
+  }
+
+  if(ooo) {
+    // Out of order execution (delay slot first)
+    //printf("OOOE\n");
+    address_generation(i+1,i_regs,regs[i].regmap_entry);
+    ds_assemble(i+1,i_regs);
+    int adj;
+    uint64_t bc_unneeded=branch_regs[i].u;
+    uint64_t bc_unneeded_upper=branch_regs[i].uu;
+    bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
+    bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
+    bc_unneeded|=1;
+    bc_unneeded_upper|=1;
+    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                  bc_unneeded,bc_unneeded_upper);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs1[i]);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+    if(rt1[i]==31) {
+      int rt,return_address;
+      assert(rt1[i+1]!=31);
+      assert(rt2[i+1]!=31);
+      rt=get_reg(branch_regs[i].regmap,31);
+      assem_debug("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+      if(rt>=0) {
+        // Save the PC even if the branch is not taken
+        return_address=start+i*4+8;
+        emit_movimm(return_address,rt); // PC into link register
+        #ifdef IMM_PREFETCH
+        if(!nevertaken) emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+        #endif
+      }
+    }
+    cc=get_reg(branch_regs[i].regmap,CCREG);
+    assert(cc==HOST_CCREG);
+    if(unconditional) 
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+    //do_cc(i,branch_regs[i].regmap,&adj,unconditional?ba[i]:-1,unconditional);
+    assem_debug("cycle count (adj)\n");
+    if(unconditional) {
+      do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
+      if(i!=(ba[i]-start)>>2 || source[i+1]!=0) {
+        if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+        load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+        if(internal)
+          assem_debug("branch: internal\n");
+        else
+          assem_debug("branch: external\n");
+        if(internal&&is_ds[(ba[i]-start)>>2]) {
+          ds_assemble_entry(i);
+        }
+        else {
+          add_to_linker((int)out,ba[i],internal);
+          emit_jmp(0);
+        }
+        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+        if(((u_int)out)&7) emit_addnop(0);
+        #endif
+      }
+    }
+    else if(nevertaken) {
+      emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
+      int jaddr=(int)out;
+      emit_jns(0);
+      add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+    }
+    else {
+      int nottaken=0;
+      do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
+      if(adj&&!invert) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+      if(!only32)
+      {
+        assert(s1h>=0);
+        if(opcode2[i]==0) // BLTZ
+        {
+          emit_test(s1h,s1h);
+          if(invert){
+            nottaken=(int)out;
+            emit_jns(1);
+          }else{
+            add_to_linker((int)out,ba[i],internal);
+            emit_js(0);
+          }
+        }
+        if(opcode2[i]==1) // BGEZ
+        {
+          emit_test(s1h,s1h);
+          if(invert){
+            nottaken=(int)out;
+            emit_js(1);
+          }else{
+            add_to_linker((int)out,ba[i],internal);
+            emit_jns(0);
+          }
+        }
+      } // if(!only32)
+      else
+      {
+        assert(s1l>=0);
+        if(opcode2[i]==0) // BLTZ
+        {
+          emit_test(s1l,s1l);
+          if(invert){
+            nottaken=(int)out;
+            emit_jns(1);
+          }else{
+            add_to_linker((int)out,ba[i],internal);
+            emit_js(0);
+          }
+        }
+        if(opcode2[i]==1) // BGEZ
+        {
+          emit_test(s1l,s1l);
+          if(invert){
+            nottaken=(int)out;
+            emit_js(1);
+          }else{
+            add_to_linker((int)out,ba[i],internal);
+            emit_jns(0);
+          }
+        }
+      } // if(!only32)
+          
+      if(invert) {
+        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+        if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) {
+          if(adj) {
+            emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
+            add_to_linker((int)out,ba[i],internal);
+          }else{
+            emit_addnop(13);
+            add_to_linker((int)out,ba[i],internal*2);
+          }
+          emit_jmp(0);
+        }else
+        #endif
+        {
+          if(adj) emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
+          store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+          load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+          if(internal)
+            assem_debug("branch: internal\n");
+          else
+            assem_debug("branch: external\n");
+          if(internal&&is_ds[(ba[i]-start)>>2]) {
+            ds_assemble_entry(i);
+          }
+          else {
+            add_to_linker((int)out,ba[i],internal);
+            emit_jmp(0);
+          }
+        }
+        set_jump_target(nottaken,(int)out);
+      }
+
+      if(adj) {
+        if(!invert) emit_addimm(cc,CLOCK_DIVIDER*adj,cc);
+      }
+    } // (!unconditional)
+  } // if(ooo)
+  else
+  {
+    // In-order execution (branch first)
+    //printf("IOE\n");
+    int nottaken=0;
+    if(!unconditional) {
+      //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+      if(!only32)
+      {
+        assert(s1h>=0);
+        if((opcode2[i]&0x1d)==0) // BLTZ/BLTZL
+        {
+          emit_test(s1h,s1h);
+          nottaken=(int)out;
+          emit_jns(1);
+        }
+        if((opcode2[i]&0x1d)==1) // BGEZ/BGEZL
+        {
+          emit_test(s1h,s1h);
+          nottaken=(int)out;
+          emit_js(1);
+        }
+      } // if(!only32)
+      else
+      {
+        assert(s1l>=0);
+        if((opcode2[i]&0x1d)==0) // BLTZ/BLTZL
+        {
+          emit_test(s1l,s1l);
+          nottaken=(int)out;
+          emit_jns(1);
+        }
+        if((opcode2[i]&0x1d)==1) // BGEZ/BGEZL
+        {
+          emit_test(s1l,s1l);
+          nottaken=(int)out;
+          emit_js(1);
+        }
+      }
+    } // if(!unconditional)
+    int adj;
+    uint64_t ds_unneeded=branch_regs[i].u;
+    uint64_t ds_unneeded_upper=branch_regs[i].uu;
+    ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+    ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+    if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+    ds_unneeded|=1;
+    ds_unneeded_upper|=1;
+    // branch taken
+    if(!nevertaken) {
+      //assem_debug("1:\n");
+      wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                    ds_unneeded,ds_unneeded_upper);
+      // load regs
+      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+      address_generation(i+1,&branch_regs[i],0);
+      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
+      ds_assemble(i+1,&branch_regs[i]);
+      cc=get_reg(branch_regs[i].regmap,CCREG);
+      if(cc==-1) {
+        emit_loadreg(CCREG,cc=HOST_CCREG);
+        // CHECK: Is the following instruction (fall thru) allocated ok?
+      }
+      assert(cc==HOST_CCREG);
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
+      assem_debug("cycle count (adj)\n");
+      if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+      load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      if(internal)
+        assem_debug("branch: internal\n");
+      else
+        assem_debug("branch: external\n");
+      if(internal&&is_ds[(ba[i]-start)>>2]) {
+        ds_assemble_entry(i);
+      }
+      else {
+        add_to_linker((int)out,ba[i],internal);
+        emit_jmp(0);
+      }
+    }
+    // branch not taken
+    cop1_usable=prev_cop1_usable;
+    if(!unconditional) {
+      set_jump_target(nottaken,(int)out);
+      assem_debug("1:\n");
+      if(!likely[i]) {
+        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                      ds_unneeded,ds_unneeded_upper);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+        address_generation(i+1,&branch_regs[i],0);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+        ds_assemble(i+1,&branch_regs[i]);
+      }
+      cc=get_reg(branch_regs[i].regmap,CCREG);
+      if(cc==-1&&!likely[i]) {
+        // Cycle count isn't in a register, temporarily load it then write it out
+        emit_loadreg(CCREG,HOST_CCREG);
+        emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),HOST_CCREG);
+        int jaddr=(int)out;
+        emit_jns(0);
+        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+        emit_storereg(CCREG,HOST_CCREG);
+      }
+      else{
+        cc=get_reg(i_regmap,CCREG);
+        assert(cc==HOST_CCREG);
+        emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
+        int jaddr=(int)out;
+        emit_jns(0);
+        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+      }
+    }
+  }
+}
+
+void fjump_assemble(int i,struct regstat *i_regs)
+{
+  signed char *i_regmap=i_regs->regmap;
+  int cc;
+  int match;
+  match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  assem_debug("fmatch=%d\n",match);
+  int fs,cs;
+  int eaddr;
+  int ooo=1;
+  int invert=0;
+  int internal=internal_branch(branch_regs[i].is32,ba[i]);
+  if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
+  if(likely[i]) ooo=0;
+  if(!match) invert=1;
+  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+  if(i>(ba[i]-start)>>2) invert=1;
+  #endif
+
+  if(ooo)
+    if(itype[i+1]==FCOMP)
+  {
+    // Write-after-read dependency prevents out of order execution
+    // First test branch condition, then execute delay slot, then branch
+    ooo=0;
+  }
+
+  if(ooo) {
+    fs=get_reg(branch_regs[i].regmap,FSREG);
+    address_generation(i+1,i_regs,regs[i].regmap_entry); // Is this okay?
+  }
+  else {
+    fs=get_reg(i_regmap,FSREG);
+  }
+
+  // Check cop1 unusable
+  if(!cop1_usable) {
+    cs=get_reg(i_regmap,CSREG);
+    assert(cs>=0);
+    emit_testimm(cs,0x20000000);
+    eaddr=(int)out;
+    emit_jeq(0);
+    add_stub(FP_STUB,eaddr,(int)out,i,cs,(int)i_regs,0,0);
+    cop1_usable=1;
+  }
+
+  if(ooo) {
+    // Out of order execution (delay slot first)
+    //printf("OOOE\n");
+    ds_assemble(i+1,i_regs);
+    int adj;
+    uint64_t bc_unneeded=branch_regs[i].u;
+    uint64_t bc_unneeded_upper=branch_regs[i].uu;
+    bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
+    bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
+    bc_unneeded|=1;
+    bc_unneeded_upper|=1;
+    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                  bc_unneeded,bc_unneeded_upper);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs1[i]);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+    cc=get_reg(branch_regs[i].regmap,CCREG);
+    assert(cc==HOST_CCREG);
+    do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
+    assem_debug("cycle count (adj)\n");
+    if(1) {
+      int nottaken=0;
+      if(adj&&!invert) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+      if(1) {
+        assert(fs>=0);
+        emit_testimm(fs,0x800000);
+        if(source[i]&0x10000) // BC1T
+        {
+          if(invert){
+            nottaken=(int)out;
+            emit_jeq(1);
+          }else{
+            add_to_linker((int)out,ba[i],internal);
+            emit_jne(0);
+          }
+        }
+        else // BC1F
+          if(invert){
+            nottaken=(int)out;
+            emit_jne(1);
+          }else{
+            add_to_linker((int)out,ba[i],internal);
+            emit_jeq(0);
+          }
+        {
+        }
+      } // if(!only32)
+          
+      if(invert) {
+        if(adj) emit_addimm(cc,-CLOCK_DIVIDER*adj,cc);
+        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+        else if(match) emit_addnop(13);
+        #endif
+        store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+        load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+        if(internal)
+          assem_debug("branch: internal\n");
+        else
+          assem_debug("branch: external\n");
+        if(internal&&is_ds[(ba[i]-start)>>2]) {
+          ds_assemble_entry(i);
+        }
+        else {
+          add_to_linker((int)out,ba[i],internal);
+          emit_jmp(0);
+        }
+        set_jump_target(nottaken,(int)out);
+      }
+
+      if(adj) {
+        if(!invert) emit_addimm(cc,CLOCK_DIVIDER*adj,cc);
+      }
+    } // (!unconditional)
+  } // if(ooo)
+  else
+  {
+    // In-order execution (branch first)
+    //printf("IOE\n");
+    int nottaken=0;
+    if(1) {
+      //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+      if(1) {
+        assert(fs>=0);
+        emit_testimm(fs,0x800000);
+        if(source[i]&0x10000) // BC1T
+        {
+          nottaken=(int)out;
+          emit_jeq(1);
+        }
+        else // BC1F
+        {
+          nottaken=(int)out;
+          emit_jne(1);
+        }
+      }
+    } // if(!unconditional)
+    int adj;
+    uint64_t ds_unneeded=branch_regs[i].u;
+    uint64_t ds_unneeded_upper=branch_regs[i].uu;
+    ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+    ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+    if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+    ds_unneeded|=1;
+    ds_unneeded_upper|=1;
+    // branch taken
+    //assem_debug("1:\n");
+    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                  ds_unneeded,ds_unneeded_upper);
+    // load regs
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+    address_generation(i+1,&branch_regs[i],0);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
+    ds_assemble(i+1,&branch_regs[i]);
+    cc=get_reg(branch_regs[i].regmap,CCREG);
+    if(cc==-1) {
+      emit_loadreg(CCREG,cc=HOST_CCREG);
+      // CHECK: Is the following instruction (fall thru) allocated ok?
+    }
+    assert(cc==HOST_CCREG);
+    store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+    do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
+    assem_debug("cycle count (adj)\n");
+    if(adj) emit_addimm(cc,CLOCK_DIVIDER*(ccadj[i]+2-adj),cc);
+    load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+    if(internal)
+      assem_debug("branch: internal\n");
+    else
+      assem_debug("branch: external\n");
+    if(internal&&is_ds[(ba[i]-start)>>2]) {
+      ds_assemble_entry(i);
+    }
+    else {
+      add_to_linker((int)out,ba[i],internal);
+      emit_jmp(0);
+    }
+
+    // branch not taken
+    if(1) { // <- FIXME (don't need this)
+      set_jump_target(nottaken,(int)out);
+      assem_debug("1:\n");
+      if(!likely[i]) {
+        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
+                      ds_unneeded,ds_unneeded_upper);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+        address_generation(i+1,&branch_regs[i],0);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+        ds_assemble(i+1,&branch_regs[i]);
+      }
+      cc=get_reg(branch_regs[i].regmap,CCREG);
+      if(cc==-1&&!likely[i]) {
+        // Cycle count isn't in a register, temporarily load it then write it out
+        emit_loadreg(CCREG,HOST_CCREG);
+        emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),HOST_CCREG);
+        int jaddr=(int)out;
+        emit_jns(0);
+        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+        emit_storereg(CCREG,HOST_CCREG);
+      }
+      else{
+        cc=get_reg(i_regmap,CCREG);
+        assert(cc==HOST_CCREG);
+        emit_addimm_and_set_flags(CLOCK_DIVIDER*(ccadj[i]+2),cc);
+        int jaddr=(int)out;
+        emit_jns(0);
+        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+      }
+    }
+  }
+}
+
+static void pagespan_assemble(int i,struct regstat *i_regs)
+{
+  int s1l=get_reg(i_regs->regmap,rs1[i]);
+  int s1h=get_reg(i_regs->regmap,rs1[i]|64);
+  int s2l=get_reg(i_regs->regmap,rs2[i]);
+  int s2h=get_reg(i_regs->regmap,rs2[i]|64);
+  void *nt_branch=NULL;
+  int taken=0;
+  int nottaken=0;
+  int unconditional=0;
+  if(rs1[i]==0)
+  {
+    s1l=s2l;s1h=s2h;
+    s2l=s2h=-1;
+  }
+  else if(rs2[i]==0)
+  {
+    s2l=s2h=-1;
+  }
+  if((i_regs->is32>>rs1[i])&(i_regs->is32>>rs2[i])&1) {
+    s1h=s2h=-1;
+  }
+  int hr=0;
+  int addr,alt,ntaddr;
+  if(i_regs->regmap[HOST_BTREG]<0) {addr=HOST_BTREG;}
+  else {
+    while(hr<HOST_REGS)
+    {
+      if(hr!=EXCLUDE_REG && hr!=HOST_CCREG &&
+         (i_regs->regmap[hr]&63)!=rs1[i] &&
+         (i_regs->regmap[hr]&63)!=rs2[i] )
+      {
+        addr=hr++;break;
+      }
+      hr++;
+    }
+  }
+  while(hr<HOST_REGS)
+  {
+    if(hr!=EXCLUDE_REG && hr!=HOST_CCREG && hr!=HOST_BTREG &&
+       (i_regs->regmap[hr]&63)!=rs1[i] &&
+       (i_regs->regmap[hr]&63)!=rs2[i] )
+    {
+      alt=hr++;break;
+    }
+    hr++;
+  }
+  if((opcode[i]&0x2E)==6) // BLEZ/BGTZ needs another register
+  {
+    while(hr<HOST_REGS)
+    {
+      if(hr!=EXCLUDE_REG && hr!=HOST_CCREG && hr!=HOST_BTREG &&
+         (i_regs->regmap[hr]&63)!=rs1[i] &&
+         (i_regs->regmap[hr]&63)!=rs2[i] )
+      {
+        ntaddr=hr;break;
+      }
+      hr++;
+    }
+  }
+  assert(hr<HOST_REGS);
+  if((opcode[i]&0x2e)==4||opcode[i]==0x11) { // BEQ/BNE/BEQL/BNEL/BC1
+    load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,CCREG,CCREG);
+  }
+  emit_addimm(HOST_CCREG,CLOCK_DIVIDER*(ccadj[i]+2),HOST_CCREG);
+  if(opcode[i]==2) // J
+  {
+    unconditional=1;
+  }
+  if(opcode[i]==3) // JAL
+  {
+    // TODO: mini_ht
+    int rt=get_reg(i_regs->regmap,31);
+    emit_movimm(start+i*4+8,rt);
+    unconditional=1;
+  }
+  if(opcode[i]==0&&(opcode2[i]&0x3E)==8) // JR/JALR
+  {
+    emit_mov(s1l,addr);
+    if(opcode2[i]==9) // JALR
+    {
+      int rt=get_reg(i_regs->regmap,31);
+      emit_movimm(start+i*4+8,rt);
+    }
+  }
+  if((opcode[i]&0x3f)==4) // BEQ
+  {
+    if(rs1[i]==rs2[i])
+    {
+      unconditional=1;
+    }
+    else
+    #ifdef HAVE_CMOV_IMM
+    if(s1h<0) {
+      if(s2l>=0) emit_cmp(s1l,s2l);
+      else emit_test(s1l,s1l);
+      emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
+    }
+    else
+    #endif
+    {
+      assert(s1l>=0);
+      emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+      if(s1h>=0) {
+        if(s2h>=0) emit_cmp(s1h,s2h);
+        else emit_test(s1h,s1h);
+        emit_cmovne_reg(alt,addr);
+      }
+      if(s2l>=0) emit_cmp(s1l,s2l);
+      else emit_test(s1l,s1l);
+      emit_cmovne_reg(alt,addr);
+    }
+  }
+  if((opcode[i]&0x3f)==5) // BNE
+  {
+    #ifdef HAVE_CMOV_IMM
+    if(s1h<0) {
+      if(s2l>=0) emit_cmp(s1l,s2l);
+      else emit_test(s1l,s1l);
+      emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+    }
+    else
+    #endif
+    {
+      assert(s1l>=0);
+      emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+      if(s1h>=0) {
+        if(s2h>=0) emit_cmp(s1h,s2h);
+        else emit_test(s1h,s1h);
+        emit_cmovne_reg(alt,addr);
+      }
+      if(s2l>=0) emit_cmp(s1l,s2l);
+      else emit_test(s1l,s1l);
+      emit_cmovne_reg(alt,addr);
+    }
+  }
+  if((opcode[i]&0x3f)==0x14) // BEQL
+  {
+    if(s1h>=0) {
+      if(s2h>=0) emit_cmp(s1h,s2h);
+      else emit_test(s1h,s1h);
+      nottaken=(int)out;
+      emit_jne(0);
+    }
+    if(s2l>=0) emit_cmp(s1l,s2l);
+    else emit_test(s1l,s1l);
+    if(nottaken) set_jump_target(nottaken,(int)out);
+    nottaken=(int)out;
+    emit_jne(0);
+  }
+  if((opcode[i]&0x3f)==0x15) // BNEL
+  {
+    if(s1h>=0) {
+      if(s2h>=0) emit_cmp(s1h,s2h);
+      else emit_test(s1h,s1h);
+      taken=(int)out;
+      emit_jne(0);
+    }
+    if(s2l>=0) emit_cmp(s1l,s2l);
+    else emit_test(s1l,s1l);
+    nottaken=(int)out;
+    emit_jeq(0);
+    if(taken) set_jump_target(taken,(int)out);
+  }
+  if((opcode[i]&0x3f)==6) // BLEZ
+  {
+    emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+    emit_cmpimm(s1l,1);
+    if(s1h>=0) emit_mov(addr,ntaddr);
+    emit_cmovl_reg(alt,addr);
+    if(s1h>=0) {
+      emit_test(s1h,s1h);
+      emit_cmovne_reg(ntaddr,addr);
+      emit_cmovs_reg(alt,addr);
+    }
+  }
+  if((opcode[i]&0x3f)==7) // BGTZ
+  {
+    emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr);
+    emit_cmpimm(s1l,1);
+    if(s1h>=0) emit_mov(addr,alt);
+    emit_cmovl_reg(ntaddr,addr);
+    if(s1h>=0) {
+      emit_test(s1h,s1h);
+      emit_cmovne_reg(alt,addr);
+      emit_cmovs_reg(ntaddr,addr);
+    }
+  }
+  if((opcode[i]&0x3f)==0x16) // BLEZL
+  {
+    assert((opcode[i]&0x3f)!=0x16);
+  }
+  if((opcode[i]&0x3f)==0x17) // BGTZL
+  {
+    assert((opcode[i]&0x3f)!=0x17);
+  }
+  assert(opcode[i]!=1); // BLTZ/BGEZ
+
+  //FIXME: Check CSREG
+  if(opcode[i]==0x11 && opcode2[i]==0x08 ) {
+    if((source[i]&0x30000)==0) // BC1F
+    {
+      emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+      emit_testimm(s1l,0x800000);
+      emit_cmovne_reg(alt,addr);
+    }
+    if((source[i]&0x30000)==0x10000) // BC1T
+    {
+      emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
+      emit_testimm(s1l,0x800000);
+      emit_cmovne_reg(alt,addr);
+    }
+    if((source[i]&0x30000)==0x20000) // BC1FL
+    {
+      emit_testimm(s1l,0x800000);
+      nottaken=(int)out;
+      emit_jne(0);
+    }
+    if((source[i]&0x30000)==0x30000) // BC1TL
+    {
+      emit_testimm(s1l,0x800000);
+      nottaken=(int)out;
+      emit_jeq(0);
+    }
+  }
+
+  assert(i_regs->regmap[HOST_CCREG]==CCREG);
+  wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+  if(likely[i]||unconditional)
+  {
+    emit_movimm(ba[i],HOST_BTREG);
+  }
+  else if(addr!=HOST_BTREG)
+  {
+    emit_mov(addr,HOST_BTREG);
+  }
+  void *branch_addr=out;
+  emit_jmp(0);
+  int target_addr=start+i*4+5;
+  void *stub=out;
+  void *compiled_target_addr=check_addr(target_addr);
+  emit_extjump_ds((int)branch_addr,target_addr);
+  if(compiled_target_addr) {
+    set_jump_target((int)branch_addr,(int)compiled_target_addr);
+    add_link(target_addr,stub);
+  }
+  else set_jump_target((int)branch_addr,(int)stub);
+  if(likely[i]) {
+    // Not-taken path
+    set_jump_target((int)nottaken,(int)out);
+    wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+    void *branch_addr=out;
+    emit_jmp(0);
+    int target_addr=start+i*4+8;
+    void *stub=out;
+    void *compiled_target_addr=check_addr(target_addr);
+    emit_extjump_ds((int)branch_addr,target_addr);
+    if(compiled_target_addr) {
+      set_jump_target((int)branch_addr,(int)compiled_target_addr);
+      add_link(target_addr,stub);
+    }
+    else set_jump_target((int)branch_addr,(int)stub);
+  }
+}
+
+// Assemble the delay slot for the above
+static void pagespan_ds()
+{
+  assem_debug("initial delay slot:\n");
+  u_int vaddr=start+1;
+  u_int page=(0x80000000^vaddr)>>12;
+  u_int vpage=page;
+  if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[page^0x80000]^0x80000000)>>12;
+  if(page>2048) page=2048+(page&2047);
+  if(vpage>262143&&tlb_LUT_r[vaddr>>12]) vpage&=2047; // jump_dirty uses a hash of the virtual address instead
+  if(vpage>2048) vpage=2048+(vpage&2047);
+  ll_add(jump_dirty+vpage,vaddr,(void *)out);
+  do_dirty_stub_ds();
+  ll_add(jump_in+page,vaddr,(void *)out);
+  assert(regs[0].regmap_entry[HOST_CCREG]==CCREG);
+  if(regs[0].regmap[HOST_CCREG]!=CCREG)
+    wb_register(CCREG,regs[0].regmap_entry,regs[0].wasdirty,regs[0].was32);
+  if(regs[0].regmap[HOST_BTREG]!=BTREG)
+    emit_writeword(HOST_BTREG,(int)&branch_target);
+  load_regs(regs[0].regmap_entry,regs[0].regmap,regs[0].was32,rs1[0],rs2[0]);
+  address_generation(0,&regs[0],regs[0].regmap_entry);
+  if(itype[0]==STORE||itype[0]==STORELR||(opcode[0]&0x3b)==0x39)
+    load_regs(regs[0].regmap_entry,regs[0].regmap,regs[0].was32,INVCP,INVCP);
+  cop1_usable=0;
+  is_delayslot=0;
+  switch(itype[0]) {
+    case ALU:
+      alu_assemble(0,&regs[0]);break;
+    case IMM16:
+      imm16_assemble(0,&regs[0]);break;
+    case SHIFT:
+      shift_assemble(0,&regs[0]);break;
+    case SHIFTIMM:
+      shiftimm_assemble(0,&regs[0]);break;
+    case LOAD:
+      load_assemble(0,&regs[0]);break;
+    case LOADLR:
+      loadlr_assemble(0,&regs[0]);break;
+    case STORE:
+      store_assemble(0,&regs[0]);break;
+    case STORELR:
+      storelr_assemble(0,&regs[0]);break;
+    case COP0:
+      cop0_assemble(0,&regs[0]);break;
+    case COP1:
+      cop1_assemble(0,&regs[0]);break;
+    case C1LS:
+      c1ls_assemble(0,&regs[0]);break;
+    case FCONV:
+      fconv_assemble(0,&regs[0]);break;
+    case FLOAT:
+      float_assemble(0,&regs[0]);break;
+    case FCOMP:
+      fcomp_assemble(0,&regs[0]);break;
+    case MULTDIV:
+      multdiv_assemble(0,&regs[0]);break;
+    case MOV:
+      mov_assemble(0,&regs[0]);break;
+    case SYSCALL:
+    case SPAN:
+    case UJUMP:
+    case RJUMP:
+    case CJUMP:
+    case SJUMP:
+    case FJUMP:
+      printf("Jump in the delay slot.  This is probably a bug.\n");
+  }
+  int btaddr=get_reg(regs[0].regmap,BTREG);
+  if(btaddr<0) {
+    btaddr=get_reg(regs[0].regmap,-1);
+    emit_readword((int)&branch_target,btaddr);
+  }
+  assert(btaddr!=HOST_CCREG);
+  if(regs[0].regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+#ifdef HOST_IMM8
+  emit_movimm(start+4,HOST_TEMPREG);
+  emit_cmp(btaddr,HOST_TEMPREG);
+#else
+  emit_cmpimm(btaddr,start+4);
+#endif
+  int branch=(int)out;
+  emit_jeq(0);
+  store_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,-1);
+  emit_jmp(jump_vaddr_reg[btaddr]);
+  set_jump_target(branch,(int)out);
+  store_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,start+4);
+  load_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,start+4);
+}
+
+// Basic liveness analysis for MIPS registers
+void unneeded_registers(int istart,int iend,int r)
+{
+  int i;
+  uint64_t u,uu,b,bu;
+  uint64_t temp_u,temp_uu;
+  uint64_t tdep;
+  if(iend==slen-1) {
+    u=1;uu=1;
+  }else{
+    u=unneeded_reg[iend+1];
+    uu=unneeded_reg_upper[iend+1];
+    u=1;uu=1;
+  }
+  for (i=iend;i>=istart;i--)
+  {
+    //printf("unneeded registers i=%d (%d,%d) r=%d\n",i,istart,iend,r);
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      // If subroutine call, flag return address as a possible branch target
+      if(rt1[i]==31 && i<slen-2) bt[i+2]=1;
+      
+      if(ba[i]<start || ba[i]>=(start+slen*4))
+      {
+        // Branch out of this block, flush all regs
+        u=1;
+        uu=1;
+        /* Hexagon hack 
+        if(itype[i]==UJUMP&&rt1[i]==31)
+        {
+          uu=u=0x300C00F; // Discard at, v0-v1, t6-t9
+        }
+        if(itype[i]==RJUMP&&rs1[i]==31)
+        {
+          uu=u=0x300C0F3; // Discard at, a0-a3, t6-t9
+        }
+        if(start>0x80000400&&start<0x80800000) {
+          if(itype[i]==UJUMP&&rt1[i]==31)
+          {
+            //uu=u=0x30300FF0FLL; // Discard at, v0-v1, t0-t9, lo, hi
+            uu=u=0x300FF0F; // Discard at, v0-v1, t0-t9
+          }
+          if(itype[i]==RJUMP&&rs1[i]==31)
+          {
+            //uu=u=0x30300FFF3LL; // Discard at, a0-a3, t0-t9, lo, hi
+            uu=u=0x300FFF3; // Discard at, a0-a3, t0-t9
+          }
+        }*/
+        branch_unneeded_reg[i]=u;
+        branch_unneeded_reg_upper[i]=uu;
+        // Merge in delay slot
+        tdep=(~uu>>rt1[i+1])&1;
+        u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+        uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+        u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+        uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+        uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
+        u|=1;uu|=1;
+        // If branch is "likely" (and conditional)
+        // then we skip the delay slot on the fall-thru path
+        if(likely[i]) {
+          if(i<slen-1) {
+            u&=unneeded_reg[i+2];
+            uu&=unneeded_reg_upper[i+2];
+          }
+          else
+          {
+            u=1;
+            uu=1;
+          }
+        }
+      }
+      else
+      {
+        // Internal branch, flag target
+        bt[(ba[i]-start)>>2]=1;
+        if(ba[i]<=start+i*4) {
+          // Backward branch
+          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          {
+            // Unconditional branch
+            temp_u=1;temp_uu=1;
+          } else {
+            // Conditional branch (not taken case)
+            temp_u=unneeded_reg[i+2];
+            temp_uu=unneeded_reg_upper[i+2];
+          }
+          // Merge in delay slot
+          tdep=(~temp_uu>>rt1[i+1])&1;
+          temp_u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+          temp_uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+          temp_u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+          temp_uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+          temp_uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
+          temp_u|=1;temp_uu|=1;
+          // If branch is "likely" (and conditional)
+          // then we skip the delay slot on the fall-thru path
+          if(likely[i]) {
+            if(i<slen-1) {
+              temp_u&=unneeded_reg[i+2];
+              temp_uu&=unneeded_reg_upper[i+2];
+            }
+            else
+            {
+              temp_u=1;
+              temp_uu=1;
+            }
+          }
+          tdep=(~temp_uu>>rt1[i])&1;
+          temp_u|=(1LL<<rt1[i])|(1LL<<rt2[i]);
+          temp_uu|=(1LL<<rt1[i])|(1LL<<rt2[i]);
+          temp_u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
+          temp_uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
+          temp_uu&=~((tdep<<dep1[i])|(tdep<<dep2[i]));
+          temp_u|=1;temp_uu|=1;
+          unneeded_reg[i]=temp_u;
+          unneeded_reg_upper[i]=temp_uu;
+          // Only go three levels deep.  This recursion can take an
+          // excessive amount of time if there are a lot of nested loops.
+          if(r<2) {
+            unneeded_registers((ba[i]-start)>>2,i-1,r+1);
+          }else{
+            unneeded_reg[(ba[i]-start)>>2]=1;
+            unneeded_reg_upper[(ba[i]-start)>>2]=1;
+          }
+        } /*else*/ if(1) {
+          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          {
+            // Unconditional branch
+            u=unneeded_reg[(ba[i]-start)>>2];
+            uu=unneeded_reg_upper[(ba[i]-start)>>2];
+            branch_unneeded_reg[i]=u;
+            branch_unneeded_reg_upper[i]=uu;
+        //u=1;
+        //uu=1;
+        //branch_unneeded_reg[i]=u;
+        //branch_unneeded_reg_upper[i]=uu;
+            // Merge in delay slot
+            tdep=(~uu>>rt1[i+1])&1;
+            u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+            uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+            u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+            uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+            uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
+            u|=1;uu|=1;
+          } else {
+            // Conditional branch
+            b=unneeded_reg[(ba[i]-start)>>2];
+            bu=unneeded_reg_upper[(ba[i]-start)>>2];
+            branch_unneeded_reg[i]=b;
+            branch_unneeded_reg_upper[i]=bu;
+        //b=1;
+        //bu=1;
+        //branch_unneeded_reg[i]=b;
+        //branch_unneeded_reg_upper[i]=bu;
+            // Branch delay slot
+            tdep=(~uu>>rt1[i+1])&1;
+            b|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+            bu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
+            b&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+            bu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+            bu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
+            b|=1;bu|=1;
+            // If branch is "likely" then we skip the
+            // delay slot on the fall-thru path
+            if(likely[i]) {
+              u=b;
+              uu=bu;
+              if(i<slen-1) {
+                u&=unneeded_reg[i+2];
+                uu&=unneeded_reg_upper[i+2];
+        //u=1;
+        //uu=1;
+              }
+            } else {
+              u&=b;
+              uu&=bu;
+        //u=1;
+        //uu=1;
+            }
+            if(i<slen-1) {
+              branch_unneeded_reg[i]&=unneeded_reg[i+2];
+              branch_unneeded_reg_upper[i]&=unneeded_reg_upper[i+2];
+        //branch_unneeded_reg[i]=1;
+        //branch_unneeded_reg_upper[i]=1;
+            } else {
+              branch_unneeded_reg[i]=1;
+              branch_unneeded_reg_upper[i]=1;
+            }
+          }
+        }
+      }
+    }
+    else if(itype[i]==SYSCALL)
+    {
+      // SYSCALL instruction (software interrupt)
+      u=1;
+      uu=1;
+    }
+    else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
+    {
+      // ERET instruction (return from interrupt)
+      u=1;
+      uu=1;
+    }
+    //u=uu=1; // DEBUG
+    tdep=(~uu>>rt1[i])&1;
+    // Written registers are unneeded
+    u|=1LL<<rt1[i];
+    u|=1LL<<rt2[i];
+    uu|=1LL<<rt1[i];
+    uu|=1LL<<rt2[i];
+    // Accessed registers are needed
+    u&=~(1LL<<rs1[i]);
+    u&=~(1LL<<rs2[i]);
+    uu&=~(1LL<<us1[i]);
+    uu&=~(1LL<<us2[i]);
+    // Source-target dependencies
+    uu&=~(tdep<<dep1[i]);
+    uu&=~(tdep<<dep2[i]);
+    // R0 is always unneeded
+    u|=1;uu|=1;
+    // Save it
+    unneeded_reg[i]=u;
+    unneeded_reg_upper[i]=uu;
+    /*
+    printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
+    printf("U:");
+    int r;
+    for(r=1;r<=CCREG;r++) {
+      if((unneeded_reg[i]>>r)&1) {
+        if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf(" UU:");
+    for(r=1;r<=CCREG;r++) {
+      if(((unneeded_reg_upper[i]&~unneeded_reg[i])>>r)&1) {
+        if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf("\n");*/
+  }
+}
+
+// Identify registers which are likely to contain 32-bit values
+// This is used to predict whether any branches will jump to a
+// location with 64-bit values in registers.
+static void provisional_32bit()
+{
+  int i,j;
+  uint64_t is32=1;
+  uint64_t lastbranch=1;
+  
+  for(i=0;i<slen;i++)
+  {
+    if(i>0) {
+      if(itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP) {
+        if(i>1) is32=lastbranch;
+        else is32=1;
+      }
+    }
+    if(i>1)
+    {
+      if(itype[i-2]==CJUMP||itype[i-2]==SJUMP||itype[i-2]==FJUMP) {
+        if(likely[i-2]) {
+          if(i>2) is32=lastbranch;
+          else is32=1;
+        }
+      }
+      if((opcode[i-2]&0x2f)==0x05) // BNE/BNEL
+      {
+        if(rs1[i-2]==0||rs2[i-2]==0)
+        {
+          if(rs1[i-2]) {
+            is32|=1LL<<rs1[i-2];
+          }
+          if(rs2[i-2]) {
+            is32|=1LL<<rs2[i-2];
+          }
+        }
+      }
+    }
+    // If something jumps here with 64-bit values
+    // then promote those registers to 64 bits
+    if(bt[i])
+    {
+      uint64_t temp_is32=is32;
+      for(j=i-1;j>=0;j--)
+      {
+        if(ba[j]==start+i*4) 
+          //temp_is32&=branch_regs[j].is32;
+          temp_is32&=p32[j];
+      }
+      for(j=i;j<slen;j++)
+      {
+        if(ba[j]==start+i*4) 
+          temp_is32=1;
+      }
+      is32=temp_is32;
+    }
+    int type=itype[i];
+    int op=opcode[i];
+    int op2=opcode2[i];
+    int rt=rt1[i];
+    int s1=rs1[i];
+    int s2=rs2[i];
+    if(type==UJUMP||type==RJUMP||type==CJUMP||type==SJUMP||type==FJUMP) {
+      // Branches don't write registers, consider the delay slot instead.
+      type=itype[i+1];
+      op=opcode[i+1];
+      op2=opcode2[i+1];
+      rt=rt1[i+1];
+      s1=rs1[i+1];
+      s2=rs2[i+1];
+      lastbranch=is32;
+    }
+    switch(type) {
+      case LOAD:
+        if(opcode[i]==0x27||opcode[i]==0x37|| // LWU/LD
+           opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
+          is32&=~(1LL<<rt);
+        else
+          is32|=1LL<<rt;
+        break;
+      case STORE:
+      case STORELR:
+        break;
+      case LOADLR:
+        if(op==0x1a||op==0x1b) is32&=~(1LL<<rt); // LDR/LDL
+        if(op==0x22) is32|=1LL<<rt; // LWL
+        break;
+      case IMM16:
+        if (op==0x08||op==0x09|| // ADDI/ADDIU
+            op==0x0a||op==0x0b|| // SLTI/SLTIU
+            op==0x0c|| // ANDI
+            op==0x0f)  // LUI
+        {
+          is32|=1LL<<rt;
+        }
+        if(op==0x18||op==0x19) { // DADDI/DADDIU
+          is32&=~(1LL<<rt);
+          //if(imm[i]==0)
+          //  is32|=((is32>>s1)&1LL)<<rt;
+        }
+        if(op==0x0d||op==0x0e) { // ORI/XORI
+          uint64_t sr=((is32>>s1)&1LL);
+          is32&=~(1LL<<rt);
+          is32|=sr<<rt;
+        }
+        break;
+      case UJUMP:
+        break;
+      case RJUMP:
+        break;
+      case CJUMP:
+        break;
+      case SJUMP:
+        break;
+      case FJUMP:
+        break;
+      case ALU:
+        if(op2>=0x20&&op2<=0x23) { // ADD/ADDU/SUB/SUBU
+          is32|=1LL<<rt;
+        }
+        if(op2==0x2a||op2==0x2b) { // SLT/SLTU
+          is32|=1LL<<rt;
+        }
+        else if(op2>=0x24&&op2<=0x27) { // AND/OR/XOR/NOR
+          uint64_t sr=((is32>>s1)&(is32>>s2)&1LL);
+          is32&=~(1LL<<rt);
+          is32|=sr<<rt;
+        }
+        else if(op2>=0x2c&&op2<=0x2d) { // DADD/DADDU
+          if(s1==0&&s2==0) {
+            is32|=1LL<<rt;
+          }
+          else if(s2==0) {
+            uint64_t sr=((is32>>s1)&1LL);
+            is32&=~(1LL<<rt);
+            is32|=sr<<rt;
+          }
+          else if(s1==0) {
+            uint64_t sr=((is32>>s2)&1LL);
+            is32&=~(1LL<<rt);
+            is32|=sr<<rt;
+          }
+          else {
+            is32&=~(1LL<<rt);
+          }
+        }
+        else if(op2>=0x2e&&op2<=0x2f) { // DSUB/DSUBU
+          if(s1==0&&s2==0) {
+            is32|=1LL<<rt;
+          }
+          else if(s2==0) {
+            uint64_t sr=((is32>>s1)&1LL);
+            is32&=~(1LL<<rt);
+            is32|=sr<<rt;
+          }
+          else {
+            is32&=~(1LL<<rt);
+          }
+        }
+        break;
+      case MULTDIV:
+        if (op2>=0x1c&&op2<=0x1f) { // DMULT/DMULTU/DDIV/DDIVU
+          is32&=~((1LL<<HIREG)|(1LL<<LOREG));
+        }
+        else {
+          is32|=(1LL<<HIREG)|(1LL<<LOREG);
+        }
+        break;
+      case MOV:
+        {
+          uint64_t sr=((is32>>s1)&1LL);
+          is32&=~(1LL<<rt);
+          is32|=sr<<rt;
+        }
+        break;
+      case SHIFT:
+        if(op2>=0x14&&op2<=0x17) is32&=~(1LL<<rt); // DSLLV/DSRLV/DSRAV
+        else is32|=1LL<<rt; // SLLV/SRLV/SRAV
+        break;
+      case SHIFTIMM:
+        is32|=1LL<<rt;
+        // DSLL/DSRL/DSRA/DSLL32/DSRL32 but not DSRA32 have 64-bit result
+        if(op2>=0x38&&op2<0x3f) is32&=~(1LL<<rt);
+        break;
+      case COP0:
+        if(op2==0) is32|=1LL<<rt; // MFC0
+        break;
+      case COP1:
+        if(op2==0) is32|=1LL<<rt; // MFC1
+        if(op2==1) is32&=~(1LL<<rt); // DMFC1
+        if(op2==2) is32|=1LL<<rt; // CFC1
+        break;
+      case C1LS:
+        break;
+      case FLOAT:
+      case FCONV:
+        break;
+      case FCOMP:
+        break;
+      case SYSCALL:
+        break;
+      default:
+        break;
+    }
+    is32|=1;
+    p32[i]=is32;
+
+    if(i>0)
+    {
+      if(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)
+      {
+        if(rt1[i-1]==31) // JAL/JALR
+        {
+          // Subroutine call will return here, don't alloc any registers
+          is32=1;
+        }
+        else if(i+1<slen)
+        {
+          // Internal branch will jump here, match registers to caller
+          is32=0x3FFFFFFFFLL;
+        }
+      }
+    }
+  }
+}
+
+// Identify registers which may be assumed to contain 32-bit values
+// and where optimizations will rely on this.
+// This is used to determine whether backward branches can safely
+// jump to a location with 64-bit values in registers.
+static void provisional_r32()
+{
+  u_int r32=0;
+  int i;
+  
+  for (i=slen-1;i>=0;i--)
+  {
+    int hr;
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      if(ba[i]<start || ba[i]>=(start+slen*4))
+      {
+        // Branch out of this block, don't need anything
+        r32=0;
+      }
+      else
+      {
+        // Internal branch
+        // Need whatever matches the target
+        // (and doesn't get overwritten by the delay slot instruction)
+        r32=0;
+        int t=(ba[i]-start)>>2;
+        if(ba[i]>start+i*4) {
+          // Forward branch
+          //if(!(requires_32bit[t]&~regs[i].was32))
+          //  r32|=requires_32bit[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+          if(!(pr32[t]&~regs[i].was32))
+            r32|=pr32[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+        }else{
+          // Backward branch
+          if(!(regs[t].was32&~unneeded_reg_upper[t]&~regs[i].was32))
+            r32|=regs[t].was32&~unneeded_reg_upper[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+        }
+      }
+      // Conditional branch may need registers for following instructions
+      if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+      {
+        if(i<slen-2) {
+          //r32|=requires_32bit[i+2];
+          r32|=pr32[i+2];
+          r32&=regs[i].was32;
+          // Mark this address as a branch target since it may be called
+          // upon return from interrupt
+          //bt[i+2]=1;
+        }
+      }
+      // Merge in delay slot
+      if(!likely[i]) {
+        // These are overwritten unless the branch is "likely"
+        // and the delay slot is nullified if not taken
+        r32&=~(1LL<<rt1[i+1]);
+        r32&=~(1LL<<rt2[i+1]);
+      }
+      // Assume these are needed (delay slot)
+      if(us1[i+1]>0)
+      {
+        if((regs[i].was32>>us1[i+1])&1) r32|=1LL<<us1[i+1];
+      }
+      if(us2[i+1]>0)
+      {
+        if((regs[i].was32>>us2[i+1])&1) r32|=1LL<<us2[i+1];
+      }
+      if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1))
+      {
+        if((regs[i].was32>>dep1[i+1])&1) r32|=1LL<<dep1[i+1];
+      }
+      if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1))
+      {
+        if((regs[i].was32>>dep2[i+1])&1) r32|=1LL<<dep2[i+1];
+      }
+    }
+    else if(itype[i]==SYSCALL)
+    {
+      // SYSCALL instruction (software interrupt)
+      r32=0;
+    }
+    else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
+    {
+      // ERET instruction (return from interrupt)
+      r32=0;
+    }
+    // Check 32 bits
+    r32&=~(1LL<<rt1[i]);
+    r32&=~(1LL<<rt2[i]);
+    if(us1[i]>0)
+    {
+      if((regs[i].was32>>us1[i])&1) r32|=1LL<<us1[i];
+    }
+    if(us2[i]>0)
+    {
+      if((regs[i].was32>>us2[i])&1) r32|=1LL<<us2[i];
+    }
+    if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1))
+    {
+      if((regs[i].was32>>dep1[i])&1) r32|=1LL<<dep1[i];
+    }
+    if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1))
+    {
+      if((regs[i].was32>>dep2[i])&1) r32|=1LL<<dep2[i];
+    }
+    //requires_32bit[i]=r32;
+    pr32[i]=r32;
+    
+    // Dirty registers which are 32-bit, require 32-bit input
+    // as they will be written as 32-bit values
+    for(hr=0;hr<HOST_REGS;hr++)
+    {
+      if(regs[i].regmap_entry[hr]>0&&regs[i].regmap_entry[hr]<64) {
+        if((regs[i].was32>>regs[i].regmap_entry[hr])&(regs[i].wasdirty>>hr)&1) {
+          if(!((unneeded_reg_upper[i]>>regs[i].regmap_entry[hr])&1))
+          pr32[i]|=1LL<<regs[i].regmap_entry[hr];
+          //requires_32bit[i]|=1LL<<regs[i].regmap_entry[hr];
+        }
+      }
+    }
+  }
+}
+
+// Write back dirty registers as soon as we will no longer modify them,
+// so that we don't end up with lots of writes at the branches.
+void clean_registers(int istart,int iend,int wr)
+{
+  int i;
+  int r;
+  u_int will_dirty_i,will_dirty_next,temp_will_dirty;
+  u_int wont_dirty_i,wont_dirty_next,temp_wont_dirty;
+  if(iend==slen-1) {
+    will_dirty_i=will_dirty_next=0;
+    wont_dirty_i=wont_dirty_next=0;
+  }else{
+    will_dirty_i=will_dirty_next=will_dirty[iend+1];
+    wont_dirty_i=wont_dirty_next=wont_dirty[iend+1];
+  }
+  for (i=iend;i>=istart;i--)
+  {
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      if(ba[i]<start || ba[i]>=(start+slen*4))
+      {
+        // Branch out of this block, flush all regs
+        if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+        {
+          // Unconditional branch
+          will_dirty_i=0;
+          wont_dirty_i=0;
+          // Merge in delay slot (will dirty)
+          for(r=0;r<HOST_REGS;r++) {
+            if(r!=EXCLUDE_REG) {
+              if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+              if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+              if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+              if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+              if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+            }
+          }
+        }
+        else
+        {
+          // Conditional branch
+          will_dirty_i=0;
+          wont_dirty_i=wont_dirty_next;
+          // Merge in delay slot (will dirty)
+          for(r=0;r<HOST_REGS;r++) {
+            if(r!=EXCLUDE_REG) {
+              if(!likely[i]) {
+                // Might not dirty if likely branch is not taken
+                if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+                if(branch_regs[i].regmap[r]==0) will_dirty_i&=~(1<<r);
+                if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+                //if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+                //if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+                if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+                if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+              }
+            }
+          }
+        }
+        // Merge in delay slot (wont dirty)
+        for(r=0;r<HOST_REGS;r++) {
+          if(r!=EXCLUDE_REG) {
+            if((regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
+            if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
+            if((regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
+            if((regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
+            if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
+            if((branch_regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
+            if((branch_regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
+            if((branch_regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
+            if((branch_regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
+            if(branch_regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
+          }
+        }
+        if(wr) {
+          #ifndef DESTRUCTIVE_WRITEBACK
+          branch_regs[i].dirty&=wont_dirty_i;
+          #endif
+          branch_regs[i].dirty|=will_dirty_i;
+        }
+      }
+      else
+      {
+        // Internal branch
+        if(ba[i]<=start+i*4) {
+          // Backward branch
+          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          {
+            // Unconditional branch
+            temp_will_dirty=0;
+            temp_wont_dirty=0;
+            // Merge in delay slot (will dirty)
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if((branch_regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
+                if((branch_regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
+                if(branch_regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
+                if(branch_regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
+                if((regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
+                if((regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
+                if((regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
+                if((regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
+                if((regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
+                if(regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
+                if(regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
+              }
+            }
+          } else {
+            // Conditional branch (not taken case)
+            temp_will_dirty=will_dirty_next;
+            temp_wont_dirty=wont_dirty_next;
+            // Merge in delay slot (will dirty)
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(!likely[i]) {
+                  // Will not dirty if likely branch is not taken
+                  if((branch_regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
+                  if(branch_regs[i].regmap[r]==0) temp_will_dirty&=~(1<<r);
+                  if(branch_regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
+                  //if((regs[i].regmap[r]&63)==rt1[i]) temp_will_dirty|=1<<r;
+                  //if((regs[i].regmap[r]&63)==rt2[i]) temp_will_dirty|=1<<r;
+                  if((regs[i].regmap[r]&63)==rt1[i+1]) temp_will_dirty|=1<<r;
+                  if((regs[i].regmap[r]&63)==rt2[i+1]) temp_will_dirty|=1<<r;
+                  if((regs[i].regmap[r]&63)>33) temp_will_dirty&=~(1<<r);
+                  if(regs[i].regmap[r]<=0) temp_will_dirty&=~(1<<r);
+                  if(regs[i].regmap[r]==CCREG) temp_will_dirty|=1<<r;
+                }
+              }
+            }
+          }
+          // Merge in delay slot (wont dirty)
+          for(r=0;r<HOST_REGS;r++) {
+            if(r!=EXCLUDE_REG) {
+              if((regs[i].regmap[r]&63)==rt1[i]) temp_wont_dirty|=1<<r;
+              if((regs[i].regmap[r]&63)==rt2[i]) temp_wont_dirty|=1<<r;
+              if((regs[i].regmap[r]&63)==rt1[i+1]) temp_wont_dirty|=1<<r;
+              if((regs[i].regmap[r]&63)==rt2[i+1]) temp_wont_dirty|=1<<r;
+              if(regs[i].regmap[r]==CCREG) temp_wont_dirty|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt1[i]) temp_wont_dirty|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt2[i]) temp_wont_dirty|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt1[i+1]) temp_wont_dirty|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt2[i+1]) temp_wont_dirty|=1<<r;
+              if(branch_regs[i].regmap[r]==CCREG) temp_wont_dirty|=1<<r;
+            }
+          }
+          // Deal with changed mappings
+          if(i<iend) {
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(regs[i].regmap[r]!=regmap_pre[i][r]) {
+                  temp_will_dirty&=~(1<<r);
+                  temp_wont_dirty&=~(1<<r);
+                  if((regmap_pre[i][r]&63)>0 && (regmap_pre[i][r]&63)<34) {
+                    temp_will_dirty|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
+                    temp_wont_dirty|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
+                  } else {
+                    temp_will_dirty|=1<<r;
+                    temp_wont_dirty|=1<<r;
+                  }
+                }
+              }
+            }
+          }
+          if(wr) {
+            will_dirty[i]=temp_will_dirty;
+            wont_dirty[i]=temp_wont_dirty;
+            clean_registers((ba[i]-start)>>2,i-1,0);
+          }else{
+            // Limit recursion.  It can take an excessive amount
+            // of time if there are a lot of nested loops.
+            will_dirty[(ba[i]-start)>>2]=0;
+            wont_dirty[(ba[i]-start)>>2]=-1;
+          }
+        }
+        /*else*/ if(1)
+        {
+          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          {
+            // Unconditional branch
+            will_dirty_i=0;
+            wont_dirty_i=0;
+          //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(branch_regs[i].regmap[r]==regs[(ba[i]-start)>>2].regmap_entry[r]) {
+                  will_dirty_i|=will_dirty[(ba[i]-start)>>2]&(1<<r);
+                  wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
+                }
+              }
+            }
+          //}
+            // Merge in delay slot
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+                if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+                if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+                if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+                if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+                if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+                if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+              }
+            }
+          } else {
+            // Conditional branch
+            will_dirty_i=will_dirty_next;
+            wont_dirty_i=wont_dirty_next;
+          //if(ba[i]>start+i*4) { // Disable recursion (for debugging)
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(branch_regs[i].regmap[r]==regs[(ba[i]-start)>>2].regmap_entry[r]) {
+                  will_dirty_i&=will_dirty[(ba[i]-start)>>2]&(1<<r);
+                  wont_dirty_i|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
+                }
+                else
+                {
+                  will_dirty_i&=~(1<<r);
+                }
+                // Treat delay slot as part of branch too
+                /*if(regs[i+1].regmap[r]==regs[(ba[i]-start)>>2].regmap_entry[r]) {
+                  will_dirty[i+1]&=will_dirty[(ba[i]-start)>>2]&(1<<r);
+                  wont_dirty[i+1]|=wont_dirty[(ba[i]-start)>>2]&(1<<r);
+                }
+                else
+                {
+                  will_dirty[i+1]&=~(1<<r);
+                }*/
+              }
+            }
+          //}
+            // Merge in delay slot
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(!likely[i]) {
+                  // Might not dirty if likely branch is not taken
+                  if((branch_regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+                  if((branch_regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+                  if(branch_regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+                  if(branch_regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+                  //if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+                  //if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+                  if((regs[i].regmap[r]&63)==rt1[i+1]) will_dirty_i|=1<<r;
+                  if((regs[i].regmap[r]&63)==rt2[i+1]) will_dirty_i|=1<<r;
+                  if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+                  if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+                  if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+                }
+              }
+            }
+          }
+          // Merge in delay slot
+          for(r=0;r<HOST_REGS;r++) {
+            if(r!=EXCLUDE_REG) {
+              if((regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
+              if((regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
+              if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt1[i+1]) wont_dirty_i|=1<<r;
+              if((branch_regs[i].regmap[r]&63)==rt2[i+1]) wont_dirty_i|=1<<r;
+              if(branch_regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
+            }
+          }
+          if(wr) {
+            #ifndef DESTRUCTIVE_WRITEBACK
+            branch_regs[i].dirty&=wont_dirty_i;
+            #endif
+            branch_regs[i].dirty|=will_dirty_i;
+          }
+        }
+      }
+    }
+    else if(itype[i]==SYSCALL)
+    {
+      // SYSCALL instruction (software interrupt)
+      will_dirty_i=0;
+      wont_dirty_i=0;
+    }
+    else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
+    {
+      // ERET instruction (return from interrupt)
+      will_dirty_i=0;
+      wont_dirty_i=0;
+    }
+    will_dirty_next=will_dirty_i;
+    wont_dirty_next=wont_dirty_i;
+    for(r=0;r<HOST_REGS;r++) {
+      if(r!=EXCLUDE_REG) {
+        if((regs[i].regmap[r]&63)==rt1[i]) will_dirty_i|=1<<r;
+        if((regs[i].regmap[r]&63)==rt2[i]) will_dirty_i|=1<<r;
+        if((regs[i].regmap[r]&63)>33) will_dirty_i&=~(1<<r);
+        if(regs[i].regmap[r]<=0) will_dirty_i&=~(1<<r);
+        if(regs[i].regmap[r]==CCREG) will_dirty_i|=1<<r;
+        if((regs[i].regmap[r]&63)==rt1[i]) wont_dirty_i|=1<<r;
+        if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
+        if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
+        if(i>istart) {
+          if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=FJUMP) 
+          {
+            // Don't store a register immediately after writing it,
+            // may prevent dual-issue.
+            if((regs[i].regmap[r]&63)==rt1[i-1]) wont_dirty_i|=1<<r;
+            if((regs[i].regmap[r]&63)==rt2[i-1]) wont_dirty_i|=1<<r;
+          }
+        }
+      }
+    }
+    // Save it
+    will_dirty[i]=will_dirty_i;
+    wont_dirty[i]=wont_dirty_i;
+    // Mark registers that won't be dirtied as not dirty
+    if(wr) {
+      /*printf("wr (%d,%d) %x will:",istart,iend,start+i*4);
+      for(r=0;r<HOST_REGS;r++) {
+        if((will_dirty_i>>r)&1) {
+          printf(" r%d",r);
+        }
+      }
+      printf("\n");*/
+
+      //if(i==istart||(itype[i-1]!=RJUMP&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=FJUMP)) {
+        regs[i].dirty|=will_dirty_i;
+        #ifndef DESTRUCTIVE_WRITEBACK
+        regs[i].dirty&=wont_dirty_i;
+        if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+        {
+          if(i<iend-1&&itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000) {
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(regs[i].regmap[r]==regmap_pre[i+2][r]) {
+                  regs[i+2].wasdirty&=wont_dirty_i|~(1<<r);
+                }else {/*printf("i: %x (%d) mismatch(+2): %d\n",start+i*4,i,r);/*assert(!((wont_dirty_i>>r)&1));*/}
+              }
+            }
+          }
+        }
+        else
+        {
+          if(i<iend) {
+            for(r=0;r<HOST_REGS;r++) {
+              if(r!=EXCLUDE_REG) {
+                if(regs[i].regmap[r]==regmap_pre[i+1][r]) {
+                  regs[i+1].wasdirty&=wont_dirty_i|~(1<<r);
+                }else {/*printf("i: %x (%d) mismatch(+1): %d\n",start+i*4,i,r);/*assert(!((wont_dirty_i>>r)&1));*/}
+              }
+            }
+          }
+        }
+        #endif
+      //}
+    }
+    // Deal with changed mappings
+    temp_will_dirty=will_dirty_i;
+    temp_wont_dirty=wont_dirty_i;
+    for(r=0;r<HOST_REGS;r++) {
+      if(r!=EXCLUDE_REG) {
+        int nr;
+        if(regs[i].regmap[r]==regmap_pre[i][r]) {
+          if(wr) {
+            #ifndef DESTRUCTIVE_WRITEBACK
+            regs[i].wasdirty&=wont_dirty_i|~(1<<r);
+            #endif
+            regs[i].wasdirty|=will_dirty_i&(1<<r);
+          }
+        }
+        else if((nr=get_reg(regs[i].regmap,regmap_pre[i][r]))>=0) {
+          // Register moved to a different register
+          will_dirty_i&=~(1<<r);
+          wont_dirty_i&=~(1<<r);
+          will_dirty_i|=((temp_will_dirty>>nr)&1)<<r;
+          wont_dirty_i|=((temp_wont_dirty>>nr)&1)<<r;
+          if(wr) {
+            #ifndef DESTRUCTIVE_WRITEBACK
+            regs[i].wasdirty&=wont_dirty_i|~(1<<r);
+            #endif
+            regs[i].wasdirty|=will_dirty_i&(1<<r);
+          }
+        }
+        else {
+          will_dirty_i&=~(1<<r);
+          wont_dirty_i&=~(1<<r);
+          if((regmap_pre[i][r]&63)>0 && (regmap_pre[i][r]&63)<34) {
+            will_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
+            wont_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
+          } else {
+            wont_dirty_i|=1<<r;
+            /*printf("i: %x (%d) mismatch: %d\n",start+i*4,i,r);/*assert(!((will_dirty>>r)&1));*/
+          }
+        }
+      }
+    }
+  }
+}
+
+  /* disassembly */
+void disassemble_inst(int i)
+{
+    if (bt[i]) printf("*"); else printf(" ");
+    switch(itype[i]) {
+      case UJUMP:
+        printf (" %x: %s %8x\n",start+i*4,insn[i],ba[i]);break;
+      case CJUMP:
+        printf (" %x: %s r%d,r%d,%8x\n",start+i*4,insn[i],rs1[i],rs2[i],i?start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14):*ba);break;
+      case SJUMP:
+        printf (" %x: %s r%d,%8x\n",start+i*4,insn[i],rs1[i],start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14));break;
+      case FJUMP:
+        printf (" %x: %s %8x\n",start+i*4,insn[i],ba[i]);break;
+      case RJUMP:
+        printf (" %x: %s r%d\n",start+i*4,insn[i],rs1[i]);break;
+      case SPAN:
+        printf (" %x: %s (pagespan) r%d,r%d,%8x\n",start+i*4,insn[i],rs1[i],rs2[i],ba[i]);break;
+      case IMM16:
+        if(opcode[i]==0xf) //LUI
+          printf (" %x: %s r%d,%4x0000\n",start+i*4,insn[i],rt1[i],imm[i]&0xffff);
+        else
+          printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],rt1[i],rs1[i],imm[i]);
+        break;
+      case LOAD:
+      case LOADLR:
+        printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],rt1[i],rs1[i],imm[i]);
+        break;
+      case STORE:
+      case STORELR:
+        printf (" %x: %s r%d,r%d+%x\n",start+i*4,insn[i],rs2[i],rs1[i],imm[i]);
+        break;
+      case ALU:
+      case SHIFT:
+        printf (" %x: %s r%d,r%d,r%d\n",start+i*4,insn[i],rt1[i],rs1[i],rs2[i]);
+        break;
+      case MULTDIV:
+        printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],rs1[i],rs2[i]);
+        break;
+      case SHIFTIMM:
+        printf (" %x: %s r%d,r%d,%d\n",start+i*4,insn[i],rt1[i],rs1[i],imm[i]);
+        break;
+      case MOV:
+        if((opcode2[i]&0x1d)==0x10)
+          printf (" %x: %s r%d\n",start+i*4,insn[i],rt1[i]);
+        else if((opcode2[i]&0x1d)==0x11)
+          printf (" %x: %s r%d\n",start+i*4,insn[i],rs1[i]);
+        else
+          printf (" %x: %s\n",start+i*4,insn[i]);
+        break;
+      case COP0:
+        if(opcode2[i]==0)
+          printf (" %x: %s r%d,cpr0[%d]\n",start+i*4,insn[i],rt1[i],(source[i]>>11)&0x1f); // MFC0
+        else if(opcode2[i]==4)
+          printf (" %x: %s r%d,cpr0[%d]\n",start+i*4,insn[i],rs1[i],(source[i]>>11)&0x1f); // MTC0
+        else printf (" %x: %s\n",start+i*4,insn[i]);
+        break;
+      case COP1:
+        if(opcode2[i]<3)
+          printf (" %x: %s r%d,cpr1[%d]\n",start+i*4,insn[i],rt1[i],(source[i]>>11)&0x1f); // MFC1
+        else if(opcode2[i]>3)
+          printf (" %x: %s r%d,cpr1[%d]\n",start+i*4,insn[i],rs1[i],(source[i]>>11)&0x1f); // MTC1
+        else printf (" %x: %s\n",start+i*4,insn[i]);
+        break;
+      case C1LS:
+        printf (" %x: %s cpr1[%d],r%d+%x\n",start+i*4,insn[i],(source[i]>>16)&0x1f,rs1[i],imm[i]);
+        break;
+      default:
+        //printf (" %s %8x\n",insn[i],source[i]);
+        printf (" %x: %s\n",start+i*4,insn[i]);
+    }
+}
+
+void new_dynarec_init()
+{
+  printf("Init new dynarec\n");
+  out=(u_char *)BASE_ADDR;
+  if (mmap (out, 1<<TARGET_SIZE_2,
+            PROT_READ | PROT_WRITE | PROT_EXEC,
+            MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
+            -1, 0) <= 0) {printf("mmap() failed\n");}
+  rdword=&readmem_dword;
+  fake_pc.f.r.rs=&readmem_dword;
+  fake_pc.f.r.rt=&readmem_dword;
+  fake_pc.f.r.rd=&readmem_dword;
+  int n;
+  for(n=0x80000;n<0x80800;n++)
+    invalid_code[n]=1;
+  for(n=0;n<65536;n++)
+    hash_table[n][0]=hash_table[n][2]=-1;
+  memset(mini_ht,-1,sizeof(mini_ht));
+  memset(restore_candidate,0,sizeof(restore_candidate));
+  copy=shadow;
+  expirep=16384; // Expiry pointer, +2 blocks
+  pending_exception=0;
+  literalcount=0;
+#ifdef HOST_IMM8
+  // Copy this into local area so we don't have to put it in every literal pool
+  invc_ptr=invalid_code;
+#endif
+  stop_after_jal=0;
+  // TLB
+  using_tlb=0;
+  for(n=0;n<524288;n++) // 0 .. 0x7FFFFFFF
+    memory_map[n]=-1;
+  for(n=524288;n<526336;n++) // 0x80000000 .. 0x807FFFFF
+    memory_map[n]=((u_int)rdram-0x80000000)>>2;
+  for(n=526336;n<1048576;n++) // 0x80800000 .. 0xFFFFFFFF
+    memory_map[n]=-1;
+  for(n=0;n<0x8000;n++) { // 0 .. 0x7FFFFFFF
+    writemem[n] = write_nomem_new;
+    writememb[n] = write_nomemb_new;
+    writememh[n] = write_nomemh_new;
+    writememd[n] = write_nomemd_new;
+    readmem[n] = read_nomem_new;
+    readmemb[n] = read_nomemb_new;
+    readmemh[n] = read_nomemh_new;
+    readmemd[n] = read_nomemd_new;
+  }
+  for(n=0x8000;n<0x8080;n++) { // 0x80000000 .. 0x807FFFFF
+    writemem[n] = write_rdram_new;
+    writememb[n] = write_rdramb_new;
+    writememh[n] = write_rdramh_new;
+    writememd[n] = write_rdramd_new;
+  }
+  for(n=0xC000;n<0x10000;n++) { // 0xC0000000 .. 0xFFFFFFFF
+    writemem[n] = write_nomem_new;
+    writememb[n] = write_nomemb_new;
+    writememh[n] = write_nomemh_new;
+    writememd[n] = write_nomemd_new;
+    readmem[n] = read_nomem_new;
+    readmemb[n] = read_nomemb_new;
+    readmemh[n] = read_nomemh_new;
+    readmemd[n] = read_nomemd_new;
+  }
+  tlb_hacks();
+  arch_init();
+}
+
+void new_dynarec_cleanup()
+{
+  int n;
+  if (munmap ((void *)BASE_ADDR, 1<<TARGET_SIZE_2) < 0) {printf("munmap() failed\n");}
+  for(n=0;n<4096;n++) ll_clear(jump_in+n);
+  for(n=0;n<4096;n++) ll_clear(jump_out+n);
+  for(n=0;n<4096;n++) ll_clear(jump_dirty+n);
+  #ifdef ROM_COPY
+  if (munmap (ROM_COPY, 67108864) < 0) {printf("munmap() failed\n");}
+  #endif
+}
+
+int new_recompile_block(int addr)
+{
+/*
+  if(addr==0x800cd050) {
+    int block;
+    for(block=0x80000;block<0x80800;block++) invalidate_block(block);
+    int n;
+    for(n=0;n<=2048;n++) ll_clear(jump_dirty+n);
+  }
+*/
+  //if(Count==365117028) tracedebug=1;
+  assem_debug("NOTCOMPILED: addr = %x -> %x\n", (int)addr, (int)out);
+  //printf("NOTCOMPILED: addr = %x -> %x\n", (int)addr, (int)out);
+  //printf("TRACE: count=%d next=%d (compile %x)\n",Count,next_interupt,addr);
+  //if(debug) 
+  //printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum());
+  //printf("fpu mapping=%x enabled=%x\n",(Status & 0x04000000)>>26,(Status & 0x20000000)>>29);
+  /*if(Count>=312978186) {
+    rlist();
+  }*/
+  //rlist();
+  start = (u_int)addr&~3;
+  //assert(((u_int)addr&1)==0);
+  if ((int)addr >= 0xa4000000 && (int)addr < 0xa4001000) {
+    source = (u_int *)((u_int)SP_DMEM+start-0xa4000000);
+    pagelimit = 0xa4001000;
+  }
+  else if ((int)addr >= 0x80000000 && (int)addr < 0x80800000) {
+    source = (u_int *)((u_int)rdram+start-0x80000000);
+    pagelimit = 0x80800000;
+  }
+  else if ((signed int)addr >= (signed int)0xC0000000) {
+    //printf("addr=%x mm=%x\n",(u_int)addr,(memory_map[start>>12]<<2));
+    //if(tlb_LUT_r[start>>12])
+      //source = (u_int *)(((int)rdram)+(tlb_LUT_r[start>>12]&0xFFFFF000)+(((int)addr)&0xFFF)-0x80000000);
+    if((signed int)memory_map[start>>12]>=0) {
+      source = (u_int *)((u_int)(start+(memory_map[start>>12]<<2)));
+      pagelimit=(start+4096)&0xFFFFF000;
+      int map=memory_map[start>>12];
+      int i;
+      for(i=0;i<5;i++) {
+        //printf("start: %x next: %x\n",map,memory_map[pagelimit>>12]);
+        if((map&0xBFFFFFFF)==(memory_map[pagelimit>>12]&0xBFFFFFFF)) pagelimit+=4096;
+      }
+      assem_debug("pagelimit=%x\n",pagelimit);
+      assem_debug("mapping=%x (%x)\n",memory_map[start>>12],(memory_map[start>>12]<<2)+start);
+    }
+    else {
+      assem_debug("Compile at unmapped memory address: %x \n", (int)addr);
+      //assem_debug("start: %x next: %x\n",memory_map[start>>12],memory_map[(start+4096)>>12]);
+      return 1; // Caller will invoke exception handler
+    }
+    //printf("source= %x\n",(int)source);
+  }
+  else {
+    printf("Compile at bogus memory address: %x \n", (int)addr);
+    exit(1);
+  }
+
+  /* Pass 1: disassemble */
+  /* Pass 2: register dependencies, branch targets */
+  /* Pass 3: register allocation */
+  /* Pass 4: branch dependencies */
+  /* Pass 5: pre-alloc */
+  /* Pass 6: optimize clean/dirty state */
+  /* Pass 7: flag 32-bit registers */
+  /* Pass 8: assembly */
+  /* Pass 9: linker */
+  /* Pass 10: garbage collection / free memory */
+
+  int i,j;
+  int done=0;
+  unsigned int type,op,op2;
+
+  //printf("addr = %x source = %x %x\n", addr,source,source[0]);
+  
+  /* Pass 1 disassembly */
+
+  for(i=0;!done;i++) {
+    bt[i]=0;likely[i]=0;op2=0;
+    opcode[i]=op=source[i]>>26;
+    switch(op)
+    {
+      case 0x00: strcpy(insn[i],"special"); type=NI;
+        op2=source[i]&0x3f;
+        switch(op2)
+        {
+          case 0x00: strcpy(insn[i],"SLL"); type=SHIFTIMM; break;
+          case 0x02: strcpy(insn[i],"SRL"); type=SHIFTIMM; break;
+          case 0x03: strcpy(insn[i],"SRA"); type=SHIFTIMM; break;
+          case 0x04: strcpy(insn[i],"SLLV"); type=SHIFT; break;
+          case 0x06: strcpy(insn[i],"SRLV"); type=SHIFT; break;
+          case 0x07: strcpy(insn[i],"SRAV"); type=SHIFT; break;
+          case 0x08: strcpy(insn[i],"JR"); type=RJUMP; break;
+          case 0x09: strcpy(insn[i],"JALR"); type=RJUMP; break;
+          case 0x0C: strcpy(insn[i],"SYSCALL"); type=SYSCALL; break;
+          case 0x0D: strcpy(insn[i],"BREAK"); type=OTHER; break;
+          case 0x0F: strcpy(insn[i],"SYNC"); type=OTHER; break;
+          case 0x10: strcpy(insn[i],"MFHI"); type=MOV; break;
+          case 0x11: strcpy(insn[i],"MTHI"); type=MOV; break;
+          case 0x12: strcpy(insn[i],"MFLO"); type=MOV; break;
+          case 0x13: strcpy(insn[i],"MTLO"); type=MOV; break;
+          case 0x14: strcpy(insn[i],"DSLLV"); type=SHIFT; break;
+          case 0x16: strcpy(insn[i],"DSRLV"); type=SHIFT; break;
+          case 0x17: strcpy(insn[i],"DSRAV"); type=SHIFT; break;
+          case 0x18: strcpy(insn[i],"MULT"); type=MULTDIV; break;
+          case 0x19: strcpy(insn[i],"MULTU"); type=MULTDIV; break;
+          case 0x1A: strcpy(insn[i],"DIV"); type=MULTDIV; break;
+          case 0x1B: strcpy(insn[i],"DIVU"); type=MULTDIV; break;
+          case 0x1C: strcpy(insn[i],"DMULT"); type=MULTDIV; break;
+          case 0x1D: strcpy(insn[i],"DMULTU"); type=MULTDIV; break;
+          case 0x1E: strcpy(insn[i],"DDIV"); type=MULTDIV; break;
+          case 0x1F: strcpy(insn[i],"DDIVU"); type=MULTDIV; break;
+          case 0x20: strcpy(insn[i],"ADD"); type=ALU; break;
+          case 0x21: strcpy(insn[i],"ADDU"); type=ALU; break;
+          case 0x22: strcpy(insn[i],"SUB"); type=ALU; break;
+          case 0x23: strcpy(insn[i],"SUBU"); type=ALU; break;
+          case 0x24: strcpy(insn[i],"AND"); type=ALU; break;
+          case 0x25: strcpy(insn[i],"OR"); type=ALU; break;
+          case 0x26: strcpy(insn[i],"XOR"); type=ALU; break;
+          case 0x27: strcpy(insn[i],"NOR"); type=ALU; break;
+          case 0x2A: strcpy(insn[i],"SLT"); type=ALU; break;
+          case 0x2B: strcpy(insn[i],"SLTU"); type=ALU; break;
+          case 0x2C: strcpy(insn[i],"DADD"); type=ALU; break;
+          case 0x2D: strcpy(insn[i],"DADDU"); type=ALU; break;
+          case 0x2E: strcpy(insn[i],"DSUB"); type=ALU; break;
+          case 0x2F: strcpy(insn[i],"DSUBU"); type=ALU; break;
+          case 0x30: strcpy(insn[i],"TGE"); type=NI; break;
+          case 0x31: strcpy(insn[i],"TGEU"); type=NI; break;
+          case 0x32: strcpy(insn[i],"TLT"); type=NI; break;
+          case 0x33: strcpy(insn[i],"TLTU"); type=NI; break;
+          case 0x34: strcpy(insn[i],"TEQ"); type=NI; break;
+          case 0x36: strcpy(insn[i],"TNE"); type=NI; break;
+          case 0x38: strcpy(insn[i],"DSLL"); type=SHIFTIMM; break;
+          case 0x3A: strcpy(insn[i],"DSRL"); type=SHIFTIMM; break;
+          case 0x3B: strcpy(insn[i],"DSRA"); type=SHIFTIMM; break;
+          case 0x3C: strcpy(insn[i],"DSLL32"); type=SHIFTIMM; break;
+          case 0x3E: strcpy(insn[i],"DSRL32"); type=SHIFTIMM; break;
+          case 0x3F: strcpy(insn[i],"DSRA32"); type=SHIFTIMM; break;
+        }
+        break;
+      case 0x01: strcpy(insn[i],"regimm"); type=NI;
+        op2=(source[i]>>16)&0x1f;
+        switch(op2)
+        {
+          case 0x00: strcpy(insn[i],"BLTZ"); type=SJUMP; break;
+          case 0x01: strcpy(insn[i],"BGEZ"); type=SJUMP; break;
+          case 0x02: strcpy(insn[i],"BLTZL"); type=SJUMP; break;
+          case 0x03: strcpy(insn[i],"BGEZL"); type=SJUMP; break;
+          case 0x08: strcpy(insn[i],"TGEI"); type=NI; break;
+          case 0x09: strcpy(insn[i],"TGEIU"); type=NI; break;
+          case 0x0A: strcpy(insn[i],"TLTI"); type=NI; break;
+          case 0x0B: strcpy(insn[i],"TLTIU"); type=NI; break;
+          case 0x0C: strcpy(insn[i],"TEQI"); type=NI; break;
+          case 0x0E: strcpy(insn[i],"TNEI"); type=NI; break;
+          case 0x10: strcpy(insn[i],"BLTZAL"); type=SJUMP; break;
+          case 0x11: strcpy(insn[i],"BGEZAL"); type=SJUMP; break;
+          case 0x12: strcpy(insn[i],"BLTZALL"); type=SJUMP; break;
+          case 0x13: strcpy(insn[i],"BGEZALL"); type=SJUMP; break;
+        }
+        break;
+      case 0x02: strcpy(insn[i],"J"); type=UJUMP; break;
+      case 0x03: strcpy(insn[i],"JAL"); type=UJUMP; break;
+      case 0x04: strcpy(insn[i],"BEQ"); type=CJUMP; break;
+      case 0x05: strcpy(insn[i],"BNE"); type=CJUMP; break;
+      case 0x06: strcpy(insn[i],"BLEZ"); type=CJUMP; break;
+      case 0x07: strcpy(insn[i],"BGTZ"); type=CJUMP; break;
+      case 0x08: strcpy(insn[i],"ADDI"); type=IMM16; break;
+      case 0x09: strcpy(insn[i],"ADDIU"); type=IMM16; break;
+      case 0x0A: strcpy(insn[i],"SLTI"); type=IMM16; break;
+      case 0x0B: strcpy(insn[i],"SLTIU"); type=IMM16; break;
+      case 0x0C: strcpy(insn[i],"ANDI"); type=IMM16; break;
+      case 0x0D: strcpy(insn[i],"ORI"); type=IMM16; break;
+      case 0x0E: strcpy(insn[i],"XORI"); type=IMM16; break;
+      case 0x0F: strcpy(insn[i],"LUI"); type=IMM16; break;
+      case 0x10: strcpy(insn[i],"cop0"); type=NI;
+        op2=(source[i]>>21)&0x1f;
+        switch(op2)
+        {
+          case 0x00: strcpy(insn[i],"MFC0"); type=COP0; break;
+          case 0x04: strcpy(insn[i],"MTC0"); type=COP0; break;
+          case 0x10: strcpy(insn[i],"tlb"); type=NI;
+          switch(source[i]&0x3f)
+          {
+            case 0x01: strcpy(insn[i],"TLBR"); type=COP0; break;
+            case 0x02: strcpy(insn[i],"TLBWI"); type=COP0; break;
+            case 0x06: strcpy(insn[i],"TLBWR"); type=COP0; break;
+            case 0x08: strcpy(insn[i],"TLBP"); type=COP0; break;
+            case 0x18: strcpy(insn[i],"ERET"); type=COP0; break;
+          }
+        }
+        break;
+      case 0x11: strcpy(insn[i],"cop1"); type=NI;
+        op2=(source[i]>>21)&0x1f;
+        switch(op2)
+        {
+          case 0x00: strcpy(insn[i],"MFC1"); type=COP1; break;
+          case 0x01: strcpy(insn[i],"DMFC1"); type=COP1; break;
+          case 0x02: strcpy(insn[i],"CFC1"); type=COP1; break;
+          case 0x04: strcpy(insn[i],"MTC1"); type=COP1; break;
+          case 0x05: strcpy(insn[i],"DMTC1"); type=COP1; break;
+          case 0x06: strcpy(insn[i],"CTC1"); type=COP1; break;
+          case 0x08: strcpy(insn[i],"BC1"); type=FJUMP;
+          switch((source[i]>>16)&0x3)
+          {
+            case 0x00: strcpy(insn[i],"BC1F"); break;
+            case 0x01: strcpy(insn[i],"BC1T"); break;
+            case 0x02: strcpy(insn[i],"BC1FL"); break;
+            case 0x03: strcpy(insn[i],"BC1TL"); break;
+          }
+          break;
+          case 0x10: strcpy(insn[i],"C1.S"); type=NI;
+          switch(source[i]&0x3f)
+          {
+            case 0x00: strcpy(insn[i],"ADD.S"); type=FLOAT; break;
+            case 0x01: strcpy(insn[i],"SUB.S"); type=FLOAT; break;
+            case 0x02: strcpy(insn[i],"MUL.S"); type=FLOAT; break;
+            case 0x03: strcpy(insn[i],"DIV.S"); type=FLOAT; break;
+            case 0x04: strcpy(insn[i],"SQRT.S"); type=FLOAT; break;
+            case 0x05: strcpy(insn[i],"ABS.S"); type=FLOAT; break;
+            case 0x06: strcpy(insn[i],"MOV.S"); type=FLOAT; break;
+            case 0x07: strcpy(insn[i],"NEG.S"); type=FLOAT; break;
+            case 0x08: strcpy(insn[i],"ROUND.L.S"); type=FCONV; break;
+            case 0x09: strcpy(insn[i],"TRUNC.L.S"); type=FCONV; break;
+            case 0x0A: strcpy(insn[i],"CEIL.L.S"); type=FCONV; break;
+            case 0x0B: strcpy(insn[i],"FLOOR.L.S"); type=FCONV; break;
+            case 0x0C: strcpy(insn[i],"ROUND.W.S"); type=FCONV; break;
+            case 0x0D: strcpy(insn[i],"TRUNC.W.S"); type=FCONV; break;
+            case 0x0E: strcpy(insn[i],"CEIL.W.S"); type=FCONV; break;
+            case 0x0F: strcpy(insn[i],"FLOOR.W.S"); type=FCONV; break;
+            case 0x21: strcpy(insn[i],"CVT.D.S"); type=FCONV; break;
+            case 0x24: strcpy(insn[i],"CVT.W.S"); type=FCONV; break;
+            case 0x25: strcpy(insn[i],"CVT.L.S"); type=FCONV; break;
+            case 0x30: strcpy(insn[i],"C.F.S"); type=FCOMP; break;
+            case 0x31: strcpy(insn[i],"C.UN.S"); type=FCOMP; break;
+            case 0x32: strcpy(insn[i],"C.EQ.S"); type=FCOMP; break;
+            case 0x33: strcpy(insn[i],"C.UEQ.S"); type=FCOMP; break;
+            case 0x34: strcpy(insn[i],"C.OLT.S"); type=FCOMP; break;
+            case 0x35: strcpy(insn[i],"C.ULT.S"); type=FCOMP; break;
+            case 0x36: strcpy(insn[i],"C.OLE.S"); type=FCOMP; break;
+            case 0x37: strcpy(insn[i],"C.ULE.S"); type=FCOMP; break;
+            case 0x38: strcpy(insn[i],"C.SF.S"); type=FCOMP; break;
+            case 0x39: strcpy(insn[i],"C.NGLE.S"); type=FCOMP; break;
+            case 0x3A: strcpy(insn[i],"C.SEQ.S"); type=FCOMP; break;
+            case 0x3B: strcpy(insn[i],"C.NGL.S"); type=FCOMP; break;
+            case 0x3C: strcpy(insn[i],"C.LT.S"); type=FCOMP; break;
+            case 0x3D: strcpy(insn[i],"C.NGE.S"); type=FCOMP; break;
+            case 0x3E: strcpy(insn[i],"C.LE.S"); type=FCOMP; break;
+            case 0x3F: strcpy(insn[i],"C.NGT.S"); type=FCOMP; break;
+          }
+          break;
+          case 0x11: strcpy(insn[i],"C1.D"); type=NI;
+          switch(source[i]&0x3f)
+          {
+            case 0x00: strcpy(insn[i],"ADD.D"); type=FLOAT; break;
+            case 0x01: strcpy(insn[i],"SUB.D"); type=FLOAT; break;
+            case 0x02: strcpy(insn[i],"MUL.D"); type=FLOAT; break;
+            case 0x03: strcpy(insn[i],"DIV.D"); type=FLOAT; break;
+            case 0x04: strcpy(insn[i],"SQRT.D"); type=FLOAT; break;
+            case 0x05: strcpy(insn[i],"ABS.D"); type=FLOAT; break;
+            case 0x06: strcpy(insn[i],"MOV.D"); type=FLOAT; break;
+            case 0x07: strcpy(insn[i],"NEG.D"); type=FLOAT; break;
+            case 0x08: strcpy(insn[i],"ROUND.L.D"); type=FCONV; break;
+            case 0x09: strcpy(insn[i],"TRUNC.L.D"); type=FCONV; break;
+            case 0x0A: strcpy(insn[i],"CEIL.L.D"); type=FCONV; break;
+            case 0x0B: strcpy(insn[i],"FLOOR.L.D"); type=FCONV; break;
+            case 0x0C: strcpy(insn[i],"ROUND.W.D"); type=FCONV; break;
+            case 0x0D: strcpy(insn[i],"TRUNC.W.D"); type=FCONV; break;
+            case 0x0E: strcpy(insn[i],"CEIL.W.D"); type=FCONV; break;
+            case 0x0F: strcpy(insn[i],"FLOOR.W.D"); type=FCONV; break;
+            case 0x20: strcpy(insn[i],"CVT.S.D"); type=FCONV; break;
+            case 0x24: strcpy(insn[i],"CVT.W.D"); type=FCONV; break;
+            case 0x25: strcpy(insn[i],"CVT.L.D"); type=FCONV; break;
+            case 0x30: strcpy(insn[i],"C.F.D"); type=FCOMP; break;
+            case 0x31: strcpy(insn[i],"C.UN.D"); type=FCOMP; break;
+            case 0x32: strcpy(insn[i],"C.EQ.D"); type=FCOMP; break;
+            case 0x33: strcpy(insn[i],"C.UEQ.D"); type=FCOMP; break;
+            case 0x34: strcpy(insn[i],"C.OLT.D"); type=FCOMP; break;
+            case 0x35: strcpy(insn[i],"C.ULT.D"); type=FCOMP; break;
+            case 0x36: strcpy(insn[i],"C.OLE.D"); type=FCOMP; break;
+            case 0x37: strcpy(insn[i],"C.ULE.D"); type=FCOMP; break;
+            case 0x38: strcpy(insn[i],"C.SF.D"); type=FCOMP; break;
+            case 0x39: strcpy(insn[i],"C.NGLE.D"); type=FCOMP; break;
+            case 0x3A: strcpy(insn[i],"C.SEQ.D"); type=FCOMP; break;
+            case 0x3B: strcpy(insn[i],"C.NGL.D"); type=FCOMP; break;
+            case 0x3C: strcpy(insn[i],"C.LT.D"); type=FCOMP; break;
+            case 0x3D: strcpy(insn[i],"C.NGE.D"); type=FCOMP; break;
+            case 0x3E: strcpy(insn[i],"C.LE.D"); type=FCOMP; break;
+            case 0x3F: strcpy(insn[i],"C.NGT.D"); type=FCOMP; break;
+          }
+          break;
+          case 0x14: strcpy(insn[i],"C1.W"); type=NI;
+          switch(source[i]&0x3f)
+          {
+            case 0x20: strcpy(insn[i],"CVT.S.W"); type=FCONV; break;
+            case 0x21: strcpy(insn[i],"CVT.D.W"); type=FCONV; break;
+          }
+          break;
+          case 0x15: strcpy(insn[i],"C1.L"); type=NI;
+          switch(source[i]&0x3f)
+          {
+            case 0x20: strcpy(insn[i],"CVT.S.L"); type=FCONV; break;
+            case 0x21: strcpy(insn[i],"CVT.D.L"); type=FCONV; break;
+          }
+          break;
+        }
+        break;
+      case 0x14: strcpy(insn[i],"BEQL"); type=CJUMP; break;
+      case 0x15: strcpy(insn[i],"BNEL"); type=CJUMP; break;
+      case 0x16: strcpy(insn[i],"BLEZL"); type=CJUMP; break;
+      case 0x17: strcpy(insn[i],"BGTZL"); type=CJUMP; break;
+      case 0x18: strcpy(insn[i],"DADDI"); type=IMM16; break;
+      case 0x19: strcpy(insn[i],"DADDIU"); type=IMM16; break;
+      case 0x1A: strcpy(insn[i],"LDL"); type=LOADLR; break;
+      case 0x1B: strcpy(insn[i],"LDR"); type=LOADLR; break;
+      case 0x20: strcpy(insn[i],"LB"); type=LOAD; break;
+      case 0x21: strcpy(insn[i],"LH"); type=LOAD; break;
+      case 0x22: strcpy(insn[i],"LWL"); type=LOADLR; break;
+      case 0x23: strcpy(insn[i],"LW"); type=LOAD; break;
+      case 0x24: strcpy(insn[i],"LBU"); type=LOAD; break;
+      case 0x25: strcpy(insn[i],"LHU"); type=LOAD; break;
+      case 0x26: strcpy(insn[i],"LWR"); type=LOADLR; break;
+      case 0x27: strcpy(insn[i],"LWU"); type=LOAD; break;
+      case 0x28: strcpy(insn[i],"SB"); type=STORE; break;
+      case 0x29: strcpy(insn[i],"SH"); type=STORE; break;
+      case 0x2A: strcpy(insn[i],"SWL"); type=STORELR; break;
+      case 0x2B: strcpy(insn[i],"SW"); type=STORE; break;
+      case 0x2C: strcpy(insn[i],"SDL"); type=STORELR; break;
+      case 0x2D: strcpy(insn[i],"SDR"); type=STORELR; break;
+      case 0x2E: strcpy(insn[i],"SWR"); type=STORELR; break;
+      case 0x2F: strcpy(insn[i],"CACHE"); type=NOP; break;
+      case 0x30: strcpy(insn[i],"LL"); type=NI; break;
+      case 0x31: strcpy(insn[i],"LWC1"); type=C1LS; break;
+      case 0x34: strcpy(insn[i],"LLD"); type=NI; break;
+      case 0x35: strcpy(insn[i],"LDC1"); type=C1LS; break;
+      case 0x37: strcpy(insn[i],"LD"); type=LOAD; break;
+      case 0x38: strcpy(insn[i],"SC"); type=NI; break;
+      case 0x39: strcpy(insn[i],"SWC1"); type=C1LS; break;
+      case 0x3C: strcpy(insn[i],"SCD"); type=NI; break;
+      case 0x3D: strcpy(insn[i],"SDC1"); type=C1LS; break;
+      case 0x3F: strcpy(insn[i],"SD"); type=STORE; break;
+      default: strcpy(insn[i],"???"); type=NI; break;
+    }
+    itype[i]=type;
+    opcode2[i]=op2;
+    /* Get registers/immediates */
+    lt1[i]=0;
+    us1[i]=0;
+    us2[i]=0;
+    dep1[i]=0;
+    dep2[i]=0;
+    switch(type) {
+      case LOAD:
+        rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=0;
+        rt1[i]=(source[i]>>16)&0x1f;
+        rt2[i]=0;
+        imm[i]=(short)source[i];
+        break;
+      case STORE:
+      case STORELR:
+        rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=(source[i]>>16)&0x1f;
+        rt1[i]=0;
+        rt2[i]=0;
+        imm[i]=(short)source[i];
+        if(op==0x2c||op==0x2d||op==0x3f) us1[i]=rs2[i]; // 64-bit SDL/SDR/SD
+        break;
+      case LOADLR:
+        // LWL/LWR only load part of the register,
+        // therefore the target register must be treated as a source too
+        rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=(source[i]>>16)&0x1f;
+        rt1[i]=(source[i]>>16)&0x1f;
+        rt2[i]=0;
+        imm[i]=(short)source[i];
+        if(op==0x1a||op==0x1b) us1[i]=rs2[i]; // LDR/LDL
+        if(op==0x26) dep1[i]=rt1[i]; // LWR
+        break;
+      case IMM16:
+        if (op==0x0f) rs1[i]=0; // LUI instruction has no source register
+        else rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=0;
+        rt1[i]=(source[i]>>16)&0x1f;
+        rt2[i]=0;
+        if(op>=0x0c&&op<=0x0e) { // ANDI/ORI/XORI
+          imm[i]=(unsigned short)source[i];
+        }else{
+          imm[i]=(short)source[i];
+        }
+        if(op==0x18||op==0x19) us1[i]=rs1[i]; // DADDI/DADDIU
+        if(op==0x0a||op==0x0b) us1[i]=rs1[i]; // SLTI/SLTIU
+        if(op==0x0d||op==0x0e) dep1[i]=rs1[i]; // ORI/XORI
+        break;
+      case UJUMP:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        // The JAL instruction writes to r31.
+        if (op&1) {
+          rt1[i]=31;
+        }
+        rs2[i]=CCREG;
+        break;
+      case RJUMP:
+        rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        // The JALR instruction writes to r31.
+        if (op2&1) {
+          rt1[i]=31;   
+        }
+        rs2[i]=CCREG;
+        break;
+      case CJUMP:
+        rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=(source[i]>>16)&0x1f;
+        rt1[i]=0;
+        rt2[i]=0;
+        if(op&2) { // BGTZ/BLEZ
+          rs2[i]=0;
+        }
+        us1[i]=rs1[i];
+        us2[i]=rs2[i];
+        likely[i]=op>>4;
+        break;
+      case SJUMP:
+        rs1[i]=(source[i]>>21)&0x1f;
+        rs2[i]=CCREG;
+        rt1[i]=0;
+        rt2[i]=0;
+        us1[i]=rs1[i];
+        if(op2&0x10) { // BxxAL
+          rt1[i]=31;
+          // NOTE: If the branch is not taken, r31 is still overwritten
+        }
+        likely[i]=(op2&2)>>1;
+        break;
+      case FJUMP:
+        rs1[i]=FSREG;
+        rs2[i]=CSREG;
+        rt1[i]=0;
+        rt2[i]=0;
+        likely[i]=((source[i])>>17)&1;
+        break;
+      case ALU:
+        rs1[i]=(source[i]>>21)&0x1f; // source
+        rs2[i]=(source[i]>>16)&0x1f; // subtract amount
+        rt1[i]=(source[i]>>11)&0x1f; // destination
+        rt2[i]=0;
+        if(op2==0x2a||op2==0x2b) { // SLT/SLTU
+          us1[i]=rs1[i];us2[i]=rs2[i];
+        }
+        else if(op2>=0x24&&op2<=0x27) { // AND/OR/XOR/NOR
+          dep1[i]=rs1[i];dep2[i]=rs2[i];
+        }
+        else if(op2>=0x2c&&op2<=0x2f) { // DADD/DSUB
+          dep1[i]=rs1[i];dep2[i]=rs2[i];
+        }
+        break;
+      case MULTDIV:
+        rs1[i]=(source[i]>>21)&0x1f; // source
+        rs2[i]=(source[i]>>16)&0x1f; // divisor
+        rt1[i]=HIREG;
+        rt2[i]=LOREG;
+        if (op2>=0x1c&&op2<=0x1f) { // DMULT/DMULTU/DDIV/DDIVU
+          us1[i]=rs1[i];us2[i]=rs2[i];
+        }
+        break;
+      case MOV:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        if(op2==0x10) rs1[i]=HIREG; // MFHI
+        if(op2==0x11) rt1[i]=HIREG; // MTHI
+        if(op2==0x12) rs1[i]=LOREG; // MFLO
+        if(op2==0x13) rt1[i]=LOREG; // MTLO
+        if((op2&0x1d)==0x10) rt1[i]=(source[i]>>11)&0x1f; // MFxx
+        if((op2&0x1d)==0x11) rs1[i]=(source[i]>>21)&0x1f; // MTxx
+        dep1[i]=rs1[i];
+        break;
+      case SHIFT:
+        rs1[i]=(source[i]>>16)&0x1f; // target of shift
+        rs2[i]=(source[i]>>21)&0x1f; // shift amount
+        rt1[i]=(source[i]>>11)&0x1f; // destination
+        rt2[i]=0;
+        // DSLLV/DSRLV/DSRAV are 64-bit
+        if(op2>=0x14&&op2<=0x17) us1[i]=rs1[i];
+        break;
+      case SHIFTIMM:
+        rs1[i]=(source[i]>>16)&0x1f;
+        rs2[i]=0;
+        rt1[i]=(source[i]>>11)&0x1f;
+        rt2[i]=0;
+        imm[i]=(source[i]>>6)&0x1f;
+        // DSxx32 instructions
+        if(op2>=0x3c) imm[i]|=0x20;
+        // DSLL/DSRL/DSRA/DSRA32/DSRL32 but not DSLL32 require 64-bit source
+        if(op2>=0x38&&op2!=0x3c) us1[i]=rs1[i];
+        break;
+      case COP0:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        if(op2==0) rt1[i]=(source[i]>>16)&0x1F; // MFC0
+        if(op2==4) rs1[i]=(source[i]>>16)&0x1F; // MTC0
+        if(op2==4&&((source[i]>>11)&0x1f)==12) rt2[i]=CSREG; // Status
+        if(op2==16) if((source[i]&0x3f)==0x18) rs2[i]=CCREG; // ERET
+        break;
+      case COP1:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        if(op2<3) rt1[i]=(source[i]>>16)&0x1F; // MFC1/DMFC1/CFC1
+        if(op2>3) rs1[i]=(source[i]>>16)&0x1F; // MTC1/DMTC1/CTC1
+        if(op2==5) us1[i]=rs1[i]; // DMTC1
+        rs2[i]=CSREG;
+        break;
+      case C1LS:
+        rs1[i]=(source[i]>>21)&0x1F;
+        rs2[i]=CSREG;
+        rt1[i]=0;
+        rt2[i]=0;
+        imm[i]=(short)source[i];
+        break;
+      case FLOAT:
+      case FCONV:
+        rs1[i]=0;
+        rs2[i]=CSREG;
+        rt1[i]=0;
+        rt2[i]=0;
+        break;
+      case FCOMP:
+        rs1[i]=FSREG;
+        rs2[i]=CSREG;
+        rt1[i]=FSREG;
+        rt2[i]=0;
+        break;
+      case SYSCALL:
+        rs1[i]=CCREG;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        break;
+      default:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+    }
+    /* Calculate branch target addresses */
+    if(type==UJUMP)
+      ba[i]=((start+i*4+4)&0xF0000000)|(((unsigned int)source[i]<<6)>>4);
+    else if(type==CJUMP&&rs1[i]==rs2[i]&&(op&1))
+      ba[i]=start+i*4+8; // Ignore never taken branch
+    else if(type==SJUMP&&rs1[i]==0&&!(op2&1))
+      ba[i]=start+i*4+8; // Ignore never taken branch
+    else if(type==CJUMP||type==SJUMP||type==FJUMP)
+      ba[i]=start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14);
+    else ba[i]=-1;
+    /* Is this the end of the block? */
+    if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) {
+      if(rt1[i-1]!=31) { // Continue past subroutine call (JAL)
+        done=1;
+        // Does the block continue due to a branch?
+        for(j=i-1;j>=0;j--)
+        {
+          if(ba[j]==start+i*4+4) done=j=0;
+          if(ba[j]==start+i*4+8) done=j=0;
+        }
+      }
+      else {
+        if(stop_after_jal) done=1;
+        // Stop on BREAK
+        if((source[i+1]&0xfc00003f)==0x0d) done=1;
+      }
+      // Don't recompile stuff that's already compiled
+      if(check_addr(start+i*4+4)) done=1;
+      // Don't get too close to the limit
+      if(i>MAXBLOCK/2) done=1;
+    }
+    if(i>0&&itype[i-1]==SYSCALL&&stop_after_jal) done=1;
+    assert(i<MAXBLOCK-1);
+    if(start+i*4==pagelimit-4) done=1;
+    assert(start+i*4<pagelimit);
+    if (i==MAXBLOCK-1) done=1;
+    // Stop if we're compiling junk
+    if(itype[i]==NI&&opcode[i]==0x11) {
+      done=stop_after_jal=1;
+      printf("Disabled speculative precompilation\n");
+    }
+  }
+  slen=i;
+  if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==RJUMP||itype[i-1]==FJUMP) {
+    if(start+i*4==pagelimit) {
+      itype[i-1]=SPAN;
+    }
+  }
+  assert(slen>0);
+
+  /* Pass 2 - Register dependencies and branch targets */
+
+  unneeded_registers(0,slen-1,0);
+  
+  /* Pass 3 - Register allocation */
+
+  struct regstat current; // Current register allocations/status
+  current.is32=1;
+  current.dirty=0;
+  current.u=unneeded_reg[0];
+  current.uu=unneeded_reg_upper[0];
+  clear_all_regs(current.regmap);
+  alloc_reg(&current,0,CCREG);
+  dirty_reg(&current,CCREG);
+  current.isconst=0;
+  current.wasconst=0;
+  int ds=0;
+  int cc=0;
+  int hr;
+  
+  provisional_32bit();
+  
+  if((u_int)addr&1) {
+    // First instruction is delay slot
+    cc=-1;
+    bt[1]=1;
+    ds=1;
+    unneeded_reg[0]=1;
+    unneeded_reg_upper[0]=1;
+    current.regmap[HOST_BTREG]=BTREG;
+  }
+  
+  for(i=0;i<slen;i++)
+  {
+    if(bt[i])
+    {
+      int hr;
+      for(hr=0;hr<HOST_REGS;hr++)
+      {
+        // Is this really necessary?
+        if(current.regmap[hr]==0) current.regmap[hr]=-1;
+      }
+      current.isconst=0;
+    }
+    if(i>1)
+    {
+      if((opcode[i-2]&0x2f)==0x05) // BNE/BNEL
+      {
+        if(rs1[i-2]==0||rs2[i-2]==0)
+        {
+          if(rs1[i-2]) {
+            current.is32|=1LL<<rs1[i-2];
+            int hr=get_reg(current.regmap,rs1[i-2]|64);
+            if(hr>=0) current.regmap[hr]=-1;
+          }
+          if(rs2[i-2]) {
+            current.is32|=1LL<<rs2[i-2];
+            int hr=get_reg(current.regmap,rs2[i-2]|64);
+            if(hr>=0) current.regmap[hr]=-1;
+          }
+        }
+      }
+    }
+    // If something jumps here with 64-bit values
+    // then promote those registers to 64 bits
+    if(bt[i])
+    {
+      uint64_t temp_is32=current.is32;
+      for(j=i-1;j>=0;j--)
+      {
+        if(ba[j]==start+i*4) 
+          temp_is32&=branch_regs[j].is32;
+      }
+      for(j=i;j<slen;j++)
+      {
+        if(ba[j]==start+i*4) 
+          //temp_is32=1;
+          temp_is32&=p32[j];
+      }
+      if(temp_is32!=current.is32) {
+        //printf("dumping 32-bit regs (%x)\n",start+i*4);
+        #ifdef DESTRUCTIVE_WRITEBACK
+        for(hr=0;hr<HOST_REGS;hr++)
+        {
+          int r=current.regmap[hr];
+          if(r>0&&r<64)
+          {
+            if((current.dirty>>hr)&((current.is32&~temp_is32)>>r)&1) {
+              temp_is32|=1LL<<r;
+              //printf("restore %d\n",r);
+            }
+          }
+        }
+        #endif
+        current.is32=temp_is32;
+      }
+    }
+    memcpy(regmap_pre[i],current.regmap,sizeof(current.regmap));
+    regs[i].wasconst=current.isconst;
+    regs[i].was32=current.is32;
+    regs[i].wasdirty=current.dirty;
+    #ifdef DESTRUCTIVE_WRITEBACK
+    // To change a dirty register from 32 to 64 bits, we must write
+    // it out during the previous cycle (for branches, 2 cycles)
+    if(i<slen-1&&bt[i+1]&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP)
+    {
+      uint64_t temp_is32=current.is32;
+      for(j=i-1;j>=0;j--)
+      {
+        if(ba[j]==start+i*4+4) 
+          temp_is32&=branch_regs[j].is32;
+      }
+      for(j=i;j<slen;j++)
+      {
+        if(ba[j]==start+i*4+4) 
+          //temp_is32=1;
+          temp_is32&=p32[j];
+      }
+      if(temp_is32!=current.is32) {
+        //printf("pre-dumping 32-bit regs (%x)\n",start+i*4);
+        for(hr=0;hr<HOST_REGS;hr++)
+        {
+          int r=current.regmap[hr];
+          if(r>0)
+          {
+            if((current.dirty>>hr)&((current.is32&~temp_is32)>>(r&63))&1) {
+              if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP)
+              {
+                if(rs1[i]!=(r&63)&&rs2[i]!=(r&63))
+                {
+                  //printf("dump %d/r%d\n",hr,r);
+                  current.regmap[hr]=-1;
+                  if(get_reg(current.regmap,r|64)>=0) 
+                    current.regmap[get_reg(current.regmap,r|64)]=-1;
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+    else if(i<slen-2&&bt[i+2]&&(source[i-1]>>16)!=0x1000&&(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP))
+    {
+      uint64_t temp_is32=current.is32;
+      for(j=i-1;j>=0;j--)
+      {
+        if(ba[j]==start+i*4+8) 
+          temp_is32&=branch_regs[j].is32;
+      }
+      for(j=i;j<slen;j++)
+      {
+        if(ba[j]==start+i*4+8) 
+          //temp_is32=1;
+          temp_is32&=p32[j];
+      }
+      if(temp_is32!=current.is32) {
+        //printf("pre-dumping 32-bit regs (%x)\n",start+i*4);
+        for(hr=0;hr<HOST_REGS;hr++)
+        {
+          int r=current.regmap[hr];
+          if(r>0)
+          {
+            if((current.dirty>>hr)&((current.is32&~temp_is32)>>(r&63))&1) {
+              if(rs1[i]!=(r&63)&&rs2[i]!=(r&63)&&rs1[i+1]!=(r&63)&&rs2[i+1]!=(r&63))
+              {
+                //printf("dump %d/r%d\n",hr,r);
+                current.regmap[hr]=-1;
+                if(get_reg(current.regmap,r|64)>=0) 
+                  current.regmap[get_reg(current.regmap,r|64)]=-1;
+              }
+            }
+          }
+        }
+      }
+    }
+    #endif
+    if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP) {
+      if(i+1<slen) {
+        current.u=unneeded_reg[i+1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
+        current.uu=unneeded_reg_upper[i+1]&~((1LL<<us1[i])|(1LL<<us2[i]));
+        if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+        current.u|=1;
+        current.uu|=1;
+      } else {
+        current.u=1;
+        current.uu=1;
+      }
+    } else {
+      if(i+1<slen) {
+        current.u=branch_unneeded_reg[i]&~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
+        current.uu=branch_unneeded_reg_upper[i]&~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+        if((~current.uu>>rt1[i+1])&1) current.uu&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+        current.u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
+        current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
+        current.u|=1;
+        current.uu|=1;
+      } else { printf("oops, branch at end of block with no delay slot\n");exit(1); }
+    }
+    is_ds[i]=ds;
+    if(ds) {
+      ds=0; // Skip delay slot, already allocated as part of branch
+      // ...but we need to alloc it in case something jumps here
+      if(i+1<slen) {
+        current.u=branch_unneeded_reg[i-1]&unneeded_reg[i+1];
+        current.uu=branch_unneeded_reg_upper[i-1]&unneeded_reg_upper[i+1];
+      }else{
+        current.u=branch_unneeded_reg[i-1];
+        current.uu=branch_unneeded_reg_upper[i-1];
+      }
+      current.u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
+      current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
+      if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+      current.u|=1;
+      current.uu|=1;
+      struct regstat temp;
+      memcpy(&temp,&current,sizeof(current));
+      temp.wasdirty=temp.dirty;
+      temp.was32=temp.is32;
+      // TODO: Take into account unconditional branches, as below
+      delayslot_alloc(&temp,i);
+      memcpy(regs[i].regmap,temp.regmap,sizeof(temp.regmap));
+      regs[i].wasdirty=temp.wasdirty;
+      regs[i].was32=temp.was32;
+      regs[i].dirty=temp.dirty;
+      regs[i].is32=temp.is32;
+      regs[i].isconst=0;
+      regs[i].wasconst=0;
+      current.isconst=0;
+      // Create entry (branch target) regmap
+      for(hr=0;hr<HOST_REGS;hr++)
+      {
+        int r=temp.regmap[hr];
+        if(r>=0) {
+          if(r!=regmap_pre[i][hr]) {
+            regs[i].regmap_entry[hr]=-1;
+          }
+          else
+          {
+            if(r<64){
+              if((current.u>>r)&1) {
+                regs[i].regmap_entry[hr]=-1;
+                regs[i].regmap[hr]=-1;
+                //Don't clear regs in the delay slot as the branch might need them
+                //current.regmap[hr]=-1;
+              }else
+                regs[i].regmap_entry[hr]=r;
+            }
+            else {
+              if((current.uu>>(r&63))&1) {
+                regs[i].regmap_entry[hr]=-1;
+                regs[i].regmap[hr]=-1;
+                //Don't clear regs in the delay slot as the branch might need them
+                //current.regmap[hr]=-1;
+              }else
+                regs[i].regmap_entry[hr]=r;
+            }
+          }
+        } else {
+          // First instruction expects CCREG to be allocated
+          if(i==0&&hr==HOST_CCREG) 
+            regs[i].regmap_entry[hr]=CCREG;
+          else
+            regs[i].regmap_entry[hr]=-1;
+        }
+      }
+    }
+    else { // Not delay slot
+      switch(itype[i]) {
+        case UJUMP:
+          //current.isconst=0; // DEBUG
+          //current.wasconst=0; // DEBUG
+          //regs[i].wasconst=0; // DEBUG
+          clear_const(&current,rt1[i]);
+          alloc_cc(&current,i);
+          dirty_reg(&current,CCREG);
+          if (rt1[i]==31) {
+            alloc_reg(&current,i,31);
+            dirty_reg(&current,31);
+            assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+            #ifdef REG_PREFETCH
+            alloc_reg(&current,i,PTEMP);
+            #endif
+            //current.is32|=1LL<<rt1[i];
+          }
+          delayslot_alloc(&current,i+1);
+          //current.isconst=0; // DEBUG
+          ds=1;
+          //printf("i=%d, isconst=%x\n",i,current.isconst);
+          break;
+        case RJUMP:
+          //current.isconst=0;
+          //current.wasconst=0;
+          //regs[i].wasconst=0;
+          clear_const(&current,rs1[i]);
+          clear_const(&current,rt1[i]);
+          alloc_cc(&current,i);
+          dirty_reg(&current,CCREG);
+          if(rs1[i]!=rt1[i+1]&&rs1[i]!=rt2[i+1]) {
+            alloc_reg(&current,i,rs1[i]);
+            if (rt1[i]==31) {
+              alloc_reg(&current,i,31);
+              dirty_reg(&current,31);
+              assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+              #ifdef REG_PREFETCH
+              alloc_reg(&current,i,PTEMP);
+              #endif
+            }
+            #ifdef USE_MINI_HT
+            if(rs1[i]==31) { // JALR
+              alloc_reg(&current,i,RHASH);
+              #ifndef HOST_IMM_ADDR32
+              alloc_reg(&current,i,RHTBL);
+              #endif
+            }
+            #endif
+            delayslot_alloc(&current,i+1);
+          } else {
+            // The delay slot overwrites our source register,
+            // allocate a temporary register to hold the old value.
+            current.isconst=0;
+            current.wasconst=0;
+            regs[i].wasconst=0;
+            delayslot_alloc(&current,i+1);
+            current.isconst=0;
+            alloc_reg(&current,i,RTEMP);
+          }
+          //current.isconst=0; // DEBUG
+          ds=1;
+          break;
+        case CJUMP:
+          //current.isconst=0;
+          //current.wasconst=0;
+          //regs[i].wasconst=0;
+          clear_const(&current,rs1[i]);
+          clear_const(&current,rs2[i]);
+          if((opcode[i]&0x3E)==4) // BEQ/BNE
+          {
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            if(rs1[i]) alloc_reg(&current,i,rs1[i]);
+            if(rs2[i]) alloc_reg(&current,i,rs2[i]);
+            if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
+            {
+              if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
+              if(rs2[i]) alloc_reg64(&current,i,rs2[i]);
+            }
+            if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))||
+               (rs2[i]&&(rs2[i]==rt1[i+1]||rs2[i]==rt2[i+1]))) {
+              // The delay slot overwrites one of our conditions.
+              // Allocate the branch condition registers instead.
+              // Note that such a sequence of instructions could
+              // be considered a bug since the branch can not be
+              // re-executed if an exception occurs.
+              current.isconst=0;
+              current.wasconst=0;
+              regs[i].wasconst=0;
+              if(rs1[i]) alloc_reg(&current,i,rs1[i]);
+              if(rs2[i]) alloc_reg(&current,i,rs2[i]);
+              if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
+              {
+                if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
+                if(rs2[i]) alloc_reg64(&current,i,rs2[i]);
+              }
+            }
+            else delayslot_alloc(&current,i+1);
+          }
+          else
+          if((opcode[i]&0x3E)==6) // BLEZ/BGTZ
+          {
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,rs1[i]);
+            if(!(current.is32>>rs1[i]&1))
+            {
+              alloc_reg64(&current,i,rs1[i]);
+            }
+            if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) {
+              // The delay slot overwrites one of our conditions.
+              // Allocate the branch condition registers instead.
+              // Note that such a sequence of instructions could
+              // be considered a bug since the branch can not be
+              // re-executed if an exception occurs.
+              current.isconst=0;
+              current.wasconst=0;
+              regs[i].wasconst=0;
+              if(rs1[i]) alloc_reg(&current,i,rs1[i]);
+              if(!((current.is32>>rs1[i])&1))
+              {
+                if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
+              }
+            }
+            else delayslot_alloc(&current,i+1);
+          }
+          else
+          // Don't alloc the delay slot yet because we might not execute it
+          if((opcode[i]&0x3E)==0x14) // BEQL/BNEL
+          {
+            current.isconst=0;
+            current.wasconst=0;
+            regs[i].wasconst=0;
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,rs1[i]);
+            alloc_reg(&current,i,rs2[i]);
+            if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
+            {
+              alloc_reg64(&current,i,rs1[i]);
+              alloc_reg64(&current,i,rs2[i]);
+            }
+          }
+          else
+          if((opcode[i]&0x3E)==0x16) // BLEZL/BGTZL
+          {
+            current.isconst=0;
+            current.wasconst=0;
+            regs[i].wasconst=0;
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,rs1[i]);
+            if(!(current.is32>>rs1[i]&1))
+            {
+              alloc_reg64(&current,i,rs1[i]);
+            }
+          }
+          ds=1;
+          //current.isconst=0;
+          break;
+        case SJUMP:
+          //current.isconst=0;
+          //current.wasconst=0;
+          //regs[i].wasconst=0;
+          clear_const(&current,rs1[i]);
+          clear_const(&current,rt1[i]);
+          //if((opcode2[i]&0x1E)==0x0) // BLTZ/BGEZ
+          if((opcode2[i]&0x0E)==0x0) // BLTZ/BGEZ
+          {
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,rs1[i]);
+            if(!(current.is32>>rs1[i]&1))
+            {
+              alloc_reg64(&current,i,rs1[i]);
+            }
+            if (rt1[i]==31) { // BLTZAL/BGEZAL
+              alloc_reg(&current,i,31);
+              dirty_reg(&current,31);
+              assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+              //#ifdef REG_PREFETCH
+              //alloc_reg(&current,i,PTEMP);
+              //#endif
+              //current.is32|=1LL<<rt1[i];
+            }
+            if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) {
+              // The delay slot overwrites the branch condition.
+              // Allocate the branch condition registers instead.
+              // Note that such a sequence of instructions could
+              // be considered a bug since the branch can not be
+              // re-executed if an exception occurs.
+              current.isconst=0;
+              current.wasconst=0;
+              regs[i].wasconst=0;
+              if(rs1[i]) alloc_reg(&current,i,rs1[i]);
+              if(!((current.is32>>rs1[i])&1))
+              {
+                if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
+              }
+            }
+            else delayslot_alloc(&current,i+1);
+          }
+          else
+          // Don't alloc the delay slot yet because we might not execute it
+          if((opcode2[i]&0x1E)==0x2) // BLTZL/BGEZL
+          {
+            current.isconst=0;
+            current.wasconst=0;
+            regs[i].wasconst=0;
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,rs1[i]);
+            if(!(current.is32>>rs1[i]&1))
+            {
+              alloc_reg64(&current,i,rs1[i]);
+            }
+          }
+          ds=1;
+          //current.isconst=0;
+          break;
+        case FJUMP:
+          current.isconst=0;
+          current.wasconst=0;
+          regs[i].wasconst=0;
+          if(likely[i]==0) // BC1F/BC1T
+          {
+            // TODO: Theoretically we can run out of registers here on x86.
+            // The delay slot can allocate up to six, and we need to check
+            // CSREG before executing the delay slot.  Possibly we can drop
+            // the cycle count and then reload it after checking that the
+            // FPU is in a usable state, or don't do out-of-order execution.
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,FSREG);
+            alloc_reg(&current,i,CSREG);
+            if(itype[i+1]==FCOMP) {
+              // The delay slot overwrites the branch condition.
+              // Allocate the branch condition registers instead.
+              // Note that such a sequence of instructions could
+              // be considered a bug since the branch can not be
+              // re-executed if an exception occurs.
+              alloc_cc(&current,i);
+              dirty_reg(&current,CCREG);
+              alloc_reg(&current,i,CSREG);
+              alloc_reg(&current,i,FSREG);
+            }
+            else {
+              delayslot_alloc(&current,i+1);
+              alloc_reg(&current,i+1,CSREG);
+            }
+          }
+          else
+          // Don't alloc the delay slot yet because we might not execute it
+          if(likely[i]) // BC1FL/BC1TL
+          {
+            alloc_cc(&current,i);
+            dirty_reg(&current,CCREG);
+            alloc_reg(&current,i,CSREG);
+            alloc_reg(&current,i,FSREG);
+          }
+          ds=1;
+          current.isconst=0;
+          break;
+        case IMM16:
+          imm16_alloc(&current,i);
+          break;
+        case LOAD:
+        case LOADLR:
+          load_alloc(&current,i);
+          break;
+        case STORE:
+        case STORELR:
+          store_alloc(&current,i);
+          break;
+        case ALU:
+          alu_alloc(&current,i);
+          break;
+        case SHIFT:
+          shift_alloc(&current,i);
+          break;
+        case MULTDIV:
+          multdiv_alloc(&current,i);
+          break;
+        case SHIFTIMM:
+          shiftimm_alloc(&current,i);
+          break;
+        case MOV:
+          mov_alloc(&current,i);
+          break;
+        case COP0:
+          cop0_alloc(&current,i);
+          break;
+        case COP1:
+          cop1_alloc(&current,i);
+          break;
+        case C1LS:
+          c1ls_alloc(&current,i);
+          break;
+        case FCONV:
+          fconv_alloc(&current,i);
+          break;
+        case FLOAT:
+          float_alloc(&current,i);
+          break;
+        case FCOMP:
+          fcomp_alloc(&current,i);
+          break;
+        case SYSCALL:
+          syscall_alloc(&current,i);
+          break;
+        case SPAN:
+          pagespan_alloc(&current,i);
+          break;
+      }
+      
+      // Drop the upper half of registers that have become 32-bit
+      current.uu|=current.is32&((1LL<<rt1[i])|(1LL<<rt2[i]));
+      if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP) {
+        current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
+        if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+        current.uu|=1;
+      } else {
+        current.uu|=current.is32&((1LL<<rt1[i+1])|(1LL<<rt2[i+1]));
+        current.uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
+        if((~current.uu>>rt1[i+1])&1) current.uu&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
+        current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
+        current.uu|=1;
+      }
+
+      // Create entry (branch target) regmap
+      for(hr=0;hr<HOST_REGS;hr++)
+      {
+        int r,or,er;
+        r=current.regmap[hr];
+        if(r>=0) {
+          if(r!=regmap_pre[i][hr]) {
+            // TODO: delay slot (?)
+            or=get_reg(regmap_pre[i],r); // Get old mapping for this register
+            if(or<0||(r&63)>=TEMPREG){
+              regs[i].regmap_entry[hr]=-1;
+            }
+            else
+            {
+              // Just move it to a different register
+              regs[i].regmap_entry[hr]=r;
+              // If it was dirty before, it's still dirty
+              if((regs[i].wasdirty>>or)&1) dirty_reg(&current,r&63);
+            }
+          }
+          else
+          {
+            // Unneeded
+            if(r==0){
+              regs[i].regmap_entry[hr]=0;
+            }
+            else
+            if(r<64){
+              if((current.u>>r)&1) {
+                regs[i].regmap_entry[hr]=-1;
+                //regs[i].regmap[hr]=-1;
+                current.regmap[hr]=-1;
+              }else
+                regs[i].regmap_entry[hr]=r;
+            }
+            else {
+              if((current.uu>>(r&63))&1) {
+                regs[i].regmap_entry[hr]=-1;
+                //regs[i].regmap[hr]=-1;
+                current.regmap[hr]=-1;
+              }else
+                regs[i].regmap_entry[hr]=r;
+            }
+          }
+        } else {
+          // Branches expect CCREG to be allocated at the target
+          if(regmap_pre[i][hr]==CCREG) 
+            regs[i].regmap_entry[hr]=CCREG;
+          else
+            regs[i].regmap_entry[hr]=-1;
+        }
+      }
+      memcpy(regs[i].regmap,current.regmap,sizeof(current.regmap));
+    }
+    /* Branch post-alloc */
+    if(i>0)
+    {
+      current.was32=current.is32;
+      current.wasdirty=current.dirty;
+      switch(itype[i-1]) {
+        case UJUMP:
+          memcpy(&branch_regs[i-1],&current,sizeof(current));
+          branch_regs[i-1].isconst=0;
+          branch_regs[i-1].wasconst=0;
+          branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
+          branch_regs[i-1].uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
+          alloc_cc(&branch_regs[i-1],i-1);
+          dirty_reg(&branch_regs[i-1],CCREG);
+          if(rt1[i-1]==31) { // JAL
+            alloc_reg(&branch_regs[i-1],i-1,31);
+            dirty_reg(&branch_regs[i-1],31);
+            branch_regs[i-1].is32|=1LL<<31;
+          }
+          memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
+          memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+          break;
+        case RJUMP:
+          memcpy(&branch_regs[i-1],&current,sizeof(current));
+          branch_regs[i-1].isconst=0;
+          branch_regs[i-1].wasconst=0;
+          branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
+          branch_regs[i-1].uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
+          alloc_cc(&branch_regs[i-1],i-1);
+          dirty_reg(&branch_regs[i-1],CCREG);
+          alloc_reg(&branch_regs[i-1],i-1,rs1[i-1]);
+          if(rt1[i-1]==31) { // JALR
+            alloc_reg(&branch_regs[i-1],i-1,31);
+            dirty_reg(&branch_regs[i-1],31);
+            branch_regs[i-1].is32|=1LL<<31;
+          }
+          #ifdef USE_MINI_HT
+          if(rs1[i-1]==31) { // JALR
+            alloc_reg(&branch_regs[i-1],i-1,RHASH);
+            #ifndef HOST_IMM_ADDR32
+            alloc_reg(&branch_regs[i-1],i-1,RHTBL);
+            #endif
+          }
+          #endif
+          memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
+          memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+          break;
+        case CJUMP:
+          if((opcode[i-1]&0x3E)==4) // BEQ/BNE
+          {
+            alloc_cc(&current,i-1);
+            dirty_reg(&current,CCREG);
+            if((rs1[i-1]&&(rs1[i-1]==rt1[i]||rs1[i-1]==rt2[i]))||
+               (rs2[i-1]&&(rs2[i-1]==rt1[i]||rs2[i-1]==rt2[i]))) {
+              // The delay slot overwrote one of our conditions
+              // Delay slot goes after the test (in order)
+              current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
+              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
+              if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+              current.u|=1;
+              current.uu|=1;
+              delayslot_alloc(&current,i);
+              current.isconst=0;
+            }
+            else
+            {
+              current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
+              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
+              // Alloc the branch condition registers
+              if(rs1[i-1]) alloc_reg(&current,i-1,rs1[i-1]);
+              if(rs2[i-1]) alloc_reg(&current,i-1,rs2[i-1]);
+              if(!((current.is32>>rs1[i-1])&(current.is32>>rs2[i-1])&1))
+              {
+                if(rs1[i-1]) alloc_reg64(&current,i-1,rs1[i-1]);
+                if(rs2[i-1]) alloc_reg64(&current,i-1,rs2[i-1]);
+              }
+            }
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].isconst=0;
+            branch_regs[i-1].wasconst=0;
+            memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
+            memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+          }
+          else
+          if((opcode[i-1]&0x3E)==6) // BLEZ/BGTZ
+          {
+            alloc_cc(&current,i-1);
+            dirty_reg(&current,CCREG);
+            if(rs1[i-1]==rt1[i]||rs1[i-1]==rt2[i]) {
+              // The delay slot overwrote the branch condition
+              // Delay slot goes after the test (in order)
+              current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
+              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
+              if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+              current.u|=1;
+              current.uu|=1;
+              delayslot_alloc(&current,i);
+              current.isconst=0;
+            }
+            else
+            {
+              current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
+              current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
+              // Alloc the branch condition register
+              alloc_reg(&current,i-1,rs1[i-1]);
+              if(!(current.is32>>rs1[i-1]&1))
+              {
+                alloc_reg64(&current,i-1,rs1[i-1]);
+              }
+            }
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].isconst=0;
+            branch_regs[i-1].wasconst=0;
+            memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
+            memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+          }
+          else
+          // Alloc the delay slot in case the branch is taken
+          if((opcode[i-1]&0x3E)==0x14) // BEQL/BNEL
+          {
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+            alloc_cc(&branch_regs[i-1],i);
+            dirty_reg(&branch_regs[i-1],CCREG);
+            delayslot_alloc(&branch_regs[i-1],i);
+            branch_regs[i-1].isconst=0;
+            alloc_reg(&current,i,CCREG); // Not taken path
+            dirty_reg(&current,CCREG);
+            memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
+          }
+          else
+          if((opcode[i-1]&0x3E)==0x16) // BLEZL/BGTZL
+          {
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+            alloc_cc(&branch_regs[i-1],i);
+            dirty_reg(&branch_regs[i-1],CCREG);
+            delayslot_alloc(&branch_regs[i-1],i);
+            branch_regs[i-1].isconst=0;
+            alloc_reg(&current,i,CCREG); // Not taken path
+            dirty_reg(&current,CCREG);
+            memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
+          }
+          break;
+        case SJUMP:
+          //if((opcode2[i-1]&0x1E)==0) // BLTZ/BGEZ
+          if((opcode2[i-1]&0x0E)==0) // BLTZ/BGEZ
+          {
+            alloc_cc(&current,i-1);
+            dirty_reg(&current,CCREG);
+            if(rs1[i-1]==rt1[i]||rs1[i-1]==rt2[i]) {
+              // The delay slot overwrote the branch condition
+              // Delay slot goes after the test (in order)
+              current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
+              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
+              if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
+              current.u|=1;
+              current.uu|=1;
+              delayslot_alloc(&current,i);
+              current.isconst=0;
+            }
+            else
+            {
+              current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
+              current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
+              // Alloc the branch condition register
+              alloc_reg(&current,i-1,rs1[i-1]);
+              if(!(current.is32>>rs1[i-1]&1))
+              {
+                alloc_reg64(&current,i-1,rs1[i-1]);
+              }
+            }
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].isconst=0;
+            branch_regs[i-1].wasconst=0;
+            memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
+            memcpy(constmap[i],constmap[i-1],sizeof(current.constmap));
+          }
+          else
+          // Alloc the delay slot in case the branch is taken
+          if((opcode2[i-1]&0x1E)==2) // BLTZL/BGEZL
+          {
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+            alloc_cc(&branch_regs[i-1],i);
+            dirty_reg(&branch_regs[i-1],CCREG);
+            delayslot_alloc(&branch_regs[i-1],i);
+            branch_regs[i-1].isconst=0;
+            alloc_reg(&current,i,CCREG); // Not taken path
+            dirty_reg(&current,CCREG);
+            memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
+          }
+          // FIXME: BLTZAL/BGEZAL
+          if(opcode2[i-1]&0x10) { // BxxZAL
+            alloc_reg(&branch_regs[i-1],i-1,31);
+            dirty_reg(&branch_regs[i-1],31);
+            branch_regs[i-1].is32|=1LL<<31;
+          }
+          break;
+        case FJUMP:
+          if(likely[i-1]==0) // BC1F/BC1T
+          {
+            alloc_cc(&current,i-1);
+            dirty_reg(&current,CCREG);
+            if(itype[i]==FCOMP) {
+              // The delay slot overwrote the branch condition
+              // Delay slot goes after the test (in order)
+              delayslot_alloc(&current,i);
+              current.isconst=0;
+            }
+            else
+            {
+              current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
+              current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
+              // Alloc the branch condition register
+              alloc_reg(&current,i-1,FSREG);
+            }
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
+          }
+          else // BC1FL/BC1TL
+          {
+            // Alloc the delay slot in case the branch is taken
+            memcpy(&branch_regs[i-1],&current,sizeof(current));
+            branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
+            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
+            alloc_cc(&branch_regs[i-1],i);
+            dirty_reg(&branch_regs[i-1],CCREG);
+            delayslot_alloc(&branch_regs[i-1],i);
+            branch_regs[i-1].isconst=0;
+            alloc_reg(&current,i,CCREG); // Not taken path
+            dirty_reg(&current,CCREG);
+            memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
+          }
+          break;
+      }
+
+      if(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)
+      {
+        if(rt1[i-1]==31) // JAL/JALR
+        {
+          // Subroutine call will return here, don't alloc any registers
+          current.is32=1;
+          current.dirty=0;
+          clear_all_regs(current.regmap);
+          alloc_reg(&current,i,CCREG);
+          dirty_reg(&current,CCREG);
+        }
+        else if(i+1<slen)
+        {
+          // Internal branch will jump here, match registers to caller
+          current.is32=0x3FFFFFFFFLL;
+          current.dirty=0;
+          clear_all_regs(current.regmap);
+          alloc_reg(&current,i,CCREG);
+          dirty_reg(&current,CCREG);
+          for(j=i-1;j>=0;j--)
+          {
+            if(ba[j]==start+i*4+4) {
+              memcpy(current.regmap,branch_regs[j].regmap,sizeof(current.regmap));
+              current.is32=branch_regs[j].is32;
+              current.dirty=branch_regs[j].dirty;
+              break;
+            }
+          }
+          while(j>=0) {
+            if(ba[j]==start+i*4+4) {
+              for(hr=0;hr<HOST_REGS;hr++) {
+                if(current.regmap[hr]!=branch_regs[j].regmap[hr]) {
+                  current.regmap[hr]=-1;
+                }
+                current.is32&=branch_regs[j].is32;
+                current.dirty&=branch_regs[j].dirty;
+              }
+            }
+            j--;
+          }
+        }
+      }
+    }
+
+    // Count cycles in between branches
+    ccadj[i]=cc;
+    if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP||itype[i]==SYSCALL))
+    {
+      cc=0;
+    }
+    else
+    {
+      cc++;
+    }
+
+    flush_dirty_uppers(&current);
+    if(!is_ds[i]) {
+      regs[i].is32=current.is32;
+      regs[i].dirty=current.dirty;
+      regs[i].isconst=current.isconst;
+      memcpy(constmap[i],current.constmap,sizeof(current.constmap));
+    }
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&regs[i].regmap[hr]>=0) {
+        if(regmap_pre[i][hr]!=regs[i].regmap[hr]) {
+          regs[i].wasconst&=~(1<<hr);
+        }
+      }
+    }
+    if(current.regmap[HOST_BTREG]==BTREG) current.regmap[HOST_BTREG]=-1;
+  }
+  
+  /* Pass 4 - Cull unused host registers */
+  
+  uint64_t nr=0;
+  
+  for (i=slen-1;i>=0;i--)
+  {
+    int hr;
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      if(ba[i]<start || ba[i]>=(start+slen*4))
+      {
+        // Branch out of this block, don't need anything
+        nr=0;
+      }
+      else
+      {
+        // Internal branch
+        // Need whatever matches the target
+        nr=0;
+        int t=(ba[i]-start)>>2;
+        for(hr=0;hr<HOST_REGS;hr++)
+        {
+          if(regs[i].regmap_entry[hr]>=0) {
+            if(regs[i].regmap_entry[hr]==regs[t].regmap_entry[hr]) nr|=1<<hr;
+          }
+        }
+      }
+      // Conditional branch may need registers for following instructions
+      if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+      {
+        if(i<slen-2) {
+          nr|=needed_reg[i+2];
+          for(hr=0;hr<HOST_REGS;hr++)
+          {
+            if(regmap_pre[i+2][hr]>=0&&get_reg(regs[i+2].regmap_entry,regmap_pre[i+2][hr])<0) nr&=~(1<<hr);
+            //if((regmap_entry[i+2][hr])>=0) if(!((nr>>hr)&1)) printf("%x-bogus(%d=%d)\n",start+i*4,hr,regmap_entry[i+2][hr]);
+          }
+        }
+      }
+      // Don't need stuff which is overwritten
+      if(regs[i].regmap[hr]!=regmap_pre[i][hr]) nr&=~(1<<hr);
+      if(regs[i].regmap[hr]<0) nr&=~(1<<hr);
+      // Merge in delay slot
+      for(hr=0;hr<HOST_REGS;hr++)
+      {
+        if(!likely[i]) {
+          // These are overwritten unless the branch is "likely"
+          // and the delay slot is nullified if not taken
+          if(rt1[i+1]&&rt1[i+1]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+          if(rt2[i+1]&&rt2[i+1]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+        }
+        if(us1[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+        if(us2[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+        if(rs1[i+1]==regmap_pre[i][hr]) nr|=1<<hr;
+        if(rs2[i+1]==regmap_pre[i][hr]) nr|=1<<hr;
+        if(us1[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+        if(us2[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+        if(rs1[i+1]==regs[i].regmap_entry[hr]) nr|=1<<hr;
+        if(rs2[i+1]==regs[i].regmap_entry[hr]) nr|=1<<hr;
+        if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1)) {
+          if(dep1[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+          if(dep2[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+        }
+        if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1)) {
+          if(dep1[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+          if(dep2[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+        }
+        if(itype[i+1]==STORE || itype[i+1]==STORELR || (opcode[i+1]&0x3b)==0x39) {
+          if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
+          if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
+        }
+      }
+    }
+    else if(itype[i]==SYSCALL)
+    {
+      // SYSCALL instruction (software interrupt)
+      nr=0;
+    }
+    else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
+    {
+      // ERET instruction (return from interrupt)
+      nr=0;
+    }
+    else // Non-branch
+    {
+      if(i<slen-1) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(regmap_pre[i+1][hr]>=0&&get_reg(regs[i+1].regmap_entry,regmap_pre[i+1][hr])<0) nr&=~(1<<hr);
+          if(regs[i].regmap[hr]!=regmap_pre[i+1][hr]) nr&=~(1<<hr);
+          if(regs[i].regmap[hr]!=regmap_pre[i][hr]) nr&=~(1<<hr);
+          if(regs[i].regmap[hr]<0) nr&=~(1<<hr);
+        }
+      }
+    }
+    for(hr=0;hr<HOST_REGS;hr++)
+    {
+      // Overwritten registers are not needed
+      if(rt1[i]&&rt1[i]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+      if(rt2[i]&&rt2[i]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+      if(FTEMP==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
+      // Source registers are needed
+      if(us1[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+      if(us2[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+      if(rs1[i]==regmap_pre[i][hr]) nr|=1<<hr;
+      if(rs2[i]==regmap_pre[i][hr]) nr|=1<<hr;
+      if(us1[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+      if(us2[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+      if(rs1[i]==regs[i].regmap_entry[hr]) nr|=1<<hr;
+      if(rs2[i]==regs[i].regmap_entry[hr]) nr|=1<<hr;
+      if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1)) {
+        if(dep1[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+        if(dep1[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+      }
+      if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1)) {
+        if(dep2[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+        if(dep2[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+      }
+      if(itype[i]==STORE || itype[i]==STORELR || (opcode[i]&0x3b)==0x39) {
+        if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
+        if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
+      }
+      // Don't store a register immediately after writing it,
+      // may prevent dual-issue.
+      // But do so if this is a branch target, otherwise we
+      // might have to load the register before the branch.
+      if(i>0&&!bt[i]&&((regs[i].wasdirty>>hr)&1)) {
+        if((regmap_pre[i][hr]>0&&regmap_pre[i][hr]<64&&!((unneeded_reg[i]>>regmap_pre[i][hr])&1)) ||
+           (regmap_pre[i][hr]>64&&!((unneeded_reg_upper[i]>>(regmap_pre[i][hr]&63))&1)) ) {
+          if(rt1[i-1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+          if(rt2[i-1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
+        }
+        if((regs[i].regmap_entry[hr]>0&&regs[i].regmap_entry[hr]<64&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1)) ||
+           (regs[i].regmap_entry[hr]>64&&!((unneeded_reg_upper[i]>>(regs[i].regmap_entry[hr]&63))&1)) ) {
+          if(rt1[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+          if(rt2[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
+        }
+      }
+    }
+    // Cycle count is needed at branches.  Assume it is needed at the target too.
+    if(i==0||bt[i]||itype[i]==CJUMP||itype[i]==FJUMP||itype[i]==SPAN) {
+      if(regmap_pre[i][HOST_CCREG]==CCREG) nr|=1<<HOST_CCREG;
+      if(regs[i].regmap_entry[HOST_CCREG]==CCREG) nr|=1<<HOST_CCREG;
+    }
+    // Save it
+    needed_reg[i]=nr;
+    
+    // Deallocate unneeded registers
+    for(hr=0;hr<HOST_REGS;hr++)
+    {
+      if(!((nr>>hr)&1)) {
+        if(regs[i].regmap_entry[hr]!=CCREG) regs[i].regmap_entry[hr]=-1;
+        if((regs[i].regmap[hr]&63)!=rs1[i] && (regs[i].regmap[hr]&63)!=rs2[i] &&
+           (regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] &&
+           (regs[i].regmap[hr]&63)!=PTEMP && (regs[i].regmap[hr]&63)!=CCREG)
+        {
+          if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+          {
+            if(likely[i]) {
+              regs[i].regmap[hr]=-1;
+              regs[i].isconst&=~(1<<hr);
+              if(i<slen-2) regmap_pre[i+2][hr]=-1;
+            }
+          }
+        }
+        if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+        {
+          int d1=0,d2=0,map=0,temp=0;
+          if(get_reg(regs[i].regmap,rt1[i+1]|64)>=0||get_reg(branch_regs[i].regmap,rt1[i+1]|64)>=0)
+          {
+            d1=dep1[i+1];
+            d2=dep2[i+1];
+          }
+          if(using_tlb) {
+            if(itype[i+1]==LOAD || itype[i+1]==LOADLR ||
+               itype[i+1]==STORE || itype[i+1]==STORELR ||
+               itype[i+1]==C1LS )
+            map=TLREG;
+          } else
+          if(itype[i+1]==STORE || itype[i+1]==STORELR || (opcode[i+1]&0x3b)==0x39) {
+            map=INVCP;
+          }
+          if(itype[i+1]==LOADLR || itype[i+1]==STORELR ||
+             itype[i+1]==C1LS )
+            temp=FTEMP;
+          if((regs[i].regmap[hr]&63)!=rs1[i] && (regs[i].regmap[hr]&63)!=rs2[i] &&
+             (regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] &&
+             (regs[i].regmap[hr]&63)!=rt1[i+1] && (regs[i].regmap[hr]&63)!=rt2[i+1] &&
+             (regs[i].regmap[hr]^64)!=us1[i+1] && (regs[i].regmap[hr]^64)!=us2[i+1] &&
+             (regs[i].regmap[hr]^64)!=d1 && (regs[i].regmap[hr]^64)!=d2 &&
+             regs[i].regmap[hr]!=rs1[i+1] && regs[i].regmap[hr]!=rs2[i+1] &&
+             (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=PTEMP &&
+             regs[i].regmap[hr]!=RHASH && regs[i].regmap[hr]!=RHTBL &&
+             regs[i].regmap[hr]!=RTEMP && regs[i].regmap[hr]!=CCREG &&
+             regs[i].regmap[hr]!=map )
+          {
+            regs[i].regmap[hr]=-1;
+            regs[i].isconst&=~(1<<hr);
+            if((branch_regs[i].regmap[hr]&63)!=rs1[i] && (branch_regs[i].regmap[hr]&63)!=rs2[i] &&
+               (branch_regs[i].regmap[hr]&63)!=rt1[i] && (branch_regs[i].regmap[hr]&63)!=rt2[i] &&
+               (branch_regs[i].regmap[hr]&63)!=rt1[i+1] && (branch_regs[i].regmap[hr]&63)!=rt2[i+1] &&
+               (branch_regs[i].regmap[hr]^64)!=us1[i+1] && (branch_regs[i].regmap[hr]^64)!=us2[i+1] &&
+               (branch_regs[i].regmap[hr]^64)!=d1 && (branch_regs[i].regmap[hr]^64)!=d2 &&
+               branch_regs[i].regmap[hr]!=rs1[i+1] && branch_regs[i].regmap[hr]!=rs2[i+1] &&
+               (branch_regs[i].regmap[hr]&63)!=temp && branch_regs[i].regmap[hr]!=PTEMP &&
+               branch_regs[i].regmap[hr]!=RHASH && branch_regs[i].regmap[hr]!=RHTBL &&
+               branch_regs[i].regmap[hr]!=RTEMP && branch_regs[i].regmap[hr]!=CCREG &&
+               branch_regs[i].regmap[hr]!=map)
+            {
+              branch_regs[i].regmap[hr]=-1;
+              branch_regs[i].regmap_entry[hr]=-1;
+              if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+              {
+                if(!likely[i]&&i<slen-2) {
+                  regmap_pre[i+2][hr]=-1;
+                }
+              }
+            }
+          }
+        }
+        else
+        {
+          // Non-branch
+          if(i>0)
+          {
+            int d1=0,d2=0,map=-1,temp=-1;
+            if(get_reg(regs[i].regmap,rt1[i]|64)>=0)
+            {
+              d1=dep1[i];
+              d2=dep2[i];
+            }
+            if(using_tlb) {
+              if(itype[i]==LOAD || itype[i]==LOADLR ||
+                 itype[i]==STORE || itype[i]==STORELR ||
+                 itype[i]==C1LS )
+              map=TLREG;
+            } else if(itype[i]==STORE || itype[i]==STORELR || (opcode[i]&0x3b)==0x39) {
+              map=INVCP;
+            }
+            if(itype[i]==LOADLR || itype[i]==STORELR ||
+               itype[i]==C1LS )
+              temp=FTEMP;
+            if((regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] &&
+               (regs[i].regmap[hr]^64)!=us1[i] && (regs[i].regmap[hr]^64)!=us2[i] &&
+               (regs[i].regmap[hr]^64)!=d1 && (regs[i].regmap[hr]^64)!=d2 &&
+               regs[i].regmap[hr]!=rs1[i] && regs[i].regmap[hr]!=rs2[i] &&
+               (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=map &&
+               (itype[i]!=SPAN||regs[i].regmap[hr]!=CCREG))
+            {
+              if(i<slen-1&&!is_ds[i]) {
+                if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]!=-1)
+                if(regmap_pre[i+1][hr]!=regs[i].regmap[hr])
+                if(regs[i].regmap[hr]<64||!((regs[i].was32>>(regs[i].regmap[hr]&63))&1))
+                {
+                  printf("fail: %x (%d %d!=%d)\n",start+i*4,hr,regmap_pre[i+1][hr],regs[i].regmap[hr]);
+                  assert(regmap_pre[i+1][hr]==regs[i].regmap[hr]);
+                }
+                regmap_pre[i+1][hr]=-1;
+                if(regs[i+1].regmap_entry[hr]==CCREG) regs[i+1].regmap_entry[hr]=-1;
+              }
+              regs[i].regmap[hr]=-1;
+              regs[i].isconst&=~(1<<hr);
+            }
+          }
+        }
+      }
+    }
+  }
+  
+  /* Pass 5 - Pre-allocate registers */
+  
+  // If a register is allocated during a loop, try to allocate it for the
+  // entire loop, if possible.  This avoids loading/storing registers
+  // inside of the loop.
+
+  signed char f_regmap[HOST_REGS];
+  clear_all_regs(f_regmap);
+  for(i=0;i<slen-1;i++)
+  {
+    if(itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      if(ba[i]>=start && ba[i]<(start+i*4)) 
+      if(itype[i+1]==NOP||itype[i+1]==MOV||itype[i+1]==ALU
+      ||itype[i+1]==SHIFTIMM||itype[i+1]==IMM16||itype[i+1]==LOAD
+      ||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS
+      ||itype[i+1]==SHIFT||itype[i+1]==COP1||itype[i+1]==FLOAT
+      ||itype[i+1]==FCOMP||itype[i+1]==FCONV)
+      {
+        int t=(ba[i]-start)>>2;
+        if(t>0&&(itype[t-1]!=UJUMP&&itype[t-1]!=RJUMP&&itype[t-1]!=CJUMP&&itype[t-1]!=SJUMP&&itype[t-1]!=FJUMP)) // loop_preload can't handle jumps into delay slots
+        if(t<2||(itype[t-2]!=UJUMP)) // call/ret assumes no registers allocated
+        for(hr=0;hr<HOST_REGS;hr++)
+        {
+          if(regs[i].regmap[hr]>64) {
+            if(!((regs[i].dirty>>hr)&1))
+              f_regmap[hr]=regs[i].regmap[hr];
+            else f_regmap[hr]=-1;
+          }
+          else if(regs[i].regmap[hr]>=0) f_regmap[hr]=regs[i].regmap[hr];
+          if(branch_regs[i].regmap[hr]>64) {
+            if(!((branch_regs[i].dirty>>hr)&1))
+              f_regmap[hr]=branch_regs[i].regmap[hr];
+            else f_regmap[hr]=-1;
+          }
+          else if(branch_regs[i].regmap[hr]>=0) f_regmap[hr]=branch_regs[i].regmap[hr];
+          if(itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS
+          ||itype[i+1]==SHIFT||itype[i+1]==COP1||itype[i+1]==FLOAT
+          ||itype[i+1]==FCOMP||itype[i+1]==FCONV)
+          {
+            // Test both in case the delay slot is ooo,
+            // could be done better...
+            if(count_free_regs(branch_regs[i].regmap)<2
+             ||count_free_regs(regs[i].regmap)<2) 
+              f_regmap[hr]=branch_regs[i].regmap[hr];
+          }
+          // Avoid dirty->clean transition
+          // #ifdef DESTRUCTIVE_WRITEBACK here?
+          if(t>0) if(get_reg(regmap_pre[t],f_regmap[hr])>=0) if((regs[t].wasdirty>>get_reg(regmap_pre[t],f_regmap[hr]))&1) f_regmap[hr]=-1;
+          if(f_regmap[hr]>0) {
+            if(regs[t].regmap_entry[hr]<0) {
+              int r=f_regmap[hr];
+              for(j=t;j<=i;j++)
+              {
+                //printf("Test %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
+                if(r<34&&((unneeded_reg[j]>>r)&1)) break;
+                if(r>63&&((unneeded_reg_upper[j]>>(r&63))&1)) break;
+                if(r>63) {
+                  // NB This can exclude the case where the upper-half
+                  // register is lower numbered than the lower-half
+                  // register.  Not sure if it's worth fixing...
+                  if(get_reg(regs[j].regmap,r&63)<0) break;
+                  if(regs[j].is32&(1LL<<(r&63))) break;
+                }
+                if(regs[j].regmap[hr]==f_regmap[hr]&&(f_regmap[hr]&63)<TEMPREG) {
+                  //printf("Hit %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
+                  int k;
+                  if(regs[i].regmap[hr]==-1&&branch_regs[i].regmap[hr]==-1) {
+                    if(get_reg(regs[i+2].regmap,f_regmap[hr])>=0) break;
+                    if(r>63) {
+                      if(get_reg(regs[i].regmap,r&63)<0) break;
+                      if(get_reg(branch_regs[i].regmap,r&63)<0) break;
+                    }
+                    k=i;
+                    while(k>1&&regs[k-1].regmap[hr]==-1) {
+                      if(itype[k-1]==STORE||itype[k-1]==STORELR
+                      ||itype[k-1]==C1LS||itype[k-1]==SHIFT||itype[k-1]==COP1
+                      ||itype[k-1]==FLOAT||itype[k-1]==FCONV
+                      ||itype[k-1]==FCOMP) {
+                        if(count_free_regs(regs[k-1].regmap)<2) {
+                          //printf("no free regs for store %x\n",start+(k-1)*4);
+                          break;
+                        }
+                      }
+                      else
+                      if(itype[k-1]!=NOP&&itype[k-1]!=MOV&&itype[k-1]!=ALU&&itype[k-1]!=SHIFTIMM&&itype[k-1]!=IMM16&&itype[k-1]!=LOAD) break;
+                      if(get_reg(regs[k-1].regmap,f_regmap[hr])>=0) {
+                        //printf("no-match due to different register\n");
+                        break;
+                      }
+                      if(itype[k-2]==UJUMP||itype[k-2]==RJUMP||itype[k-2]==CJUMP||itype[k-2]==SJUMP||itype[k-2]==FJUMP) {
+                        //printf("no-match due to branch\n");
+                        break;
+                      }
+                      // call/ret fast path assumes no registers allocated
+                      if(k>2&&(itype[k-3]==UJUMP||itype[k-3]==RJUMP)) {
+                        break;
+                      }
+                      if(r>63) {
+                        // NB This can exclude the case where the upper-half
+                        // register is lower numbered than the lower-half
+                        // register.  Not sure if it's worth fixing...
+                        if(get_reg(regs[k-1].regmap,r&63)<0) break;
+                        if(regs[k-1].is32&(1LL<<(r&63))) break;
+                      }
+                      k--;
+                    }
+                    if(i<slen-1) {
+                      if((regs[k].is32&(1LL<<f_regmap[hr]))!=
+                        (regs[i+2].was32&(1LL<<f_regmap[hr]))) {
+                        //printf("bad match after branch\n");
+                        break;
+                      }
+                    }
+                    if(regs[k-1].regmap[hr]==f_regmap[hr]&&regmap_pre[k][hr]==f_regmap[hr]) {
+                      //printf("Extend r%d, %x ->\n",hr,start+k*4);
+                      while(k<i) {
+                        regs[k].regmap_entry[hr]=f_regmap[hr];
+                        regs[k].regmap[hr]=f_regmap[hr];
+                        regmap_pre[k+1][hr]=f_regmap[hr];
+                        regs[k].wasdirty&=~(1<<hr);
+                        regs[k].dirty&=~(1<<hr);
+                        regs[k].wasdirty|=(1<<hr)&regs[k-1].dirty;
+                        regs[k].dirty|=(1<<hr)&regs[k].wasdirty;
+                        regs[k].wasconst&=~(1<<hr);
+                        regs[k].isconst&=~(1<<hr);
+                        k++;
+                      }
+                    }
+                    else {
+                      //printf("Fail Extend r%d, %x ->\n",hr,start+k*4);
+                      break;
+                    }
+                    assert(regs[i-1].regmap[hr]==f_regmap[hr]);
+                    if(regs[i-1].regmap[hr]==f_regmap[hr]&&regmap_pre[i][hr]==f_regmap[hr]) {
+                      //printf("OK fill %x (r%d)\n",start+i*4,hr);
+                      regs[i].regmap_entry[hr]=f_regmap[hr];
+                      regs[i].regmap[hr]=f_regmap[hr];
+                      regs[i].wasdirty&=~(1<<hr);
+                      regs[i].dirty&=~(1<<hr);
+                      regs[i].wasdirty|=(1<<hr)&regs[i-1].dirty;
+                      regs[i].dirty|=(1<<hr)&regs[i-1].dirty;
+                      regs[i].wasconst&=~(1<<hr);
+                      regs[i].isconst&=~(1<<hr);
+                      branch_regs[i].regmap_entry[hr]=f_regmap[hr];
+                      branch_regs[i].wasdirty&=~(1<<hr);
+                      branch_regs[i].wasdirty|=(1<<hr)&regs[i].dirty;
+                      branch_regs[i].regmap[hr]=f_regmap[hr];
+                      branch_regs[i].dirty&=~(1<<hr);
+                      branch_regs[i].dirty|=(1<<hr)&regs[i].dirty;
+                      branch_regs[i].wasconst&=~(1<<hr);
+                      branch_regs[i].isconst&=~(1<<hr);
+                      if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000) {
+                        regmap_pre[i+2][hr]=f_regmap[hr];
+                        regs[i+2].wasdirty&=~(1<<hr);
+                        regs[i+2].wasdirty|=(1<<hr)&regs[i].dirty;
+                        assert((branch_regs[i].is32&(1LL<<f_regmap[hr]))==
+                          (regs[i+2].was32&(1LL<<f_regmap[hr])));
+                      }
+                    }
+                  }
+                  for(k=t;k<j;k++) {
+                    regs[k].regmap_entry[hr]=f_regmap[hr];
+                    regs[k].regmap[hr]=f_regmap[hr];
+                    regmap_pre[k+1][hr]=f_regmap[hr];
+                    regs[k+1].wasdirty&=~(1<<hr);
+                    regs[k].dirty&=~(1<<hr);
+                    regs[k].wasconst&=~(1<<hr);
+                    regs[k].isconst&=~(1<<hr);
+                  }
+                  if(regs[j].regmap[hr]==f_regmap[hr])
+                    regs[j].regmap_entry[hr]=f_regmap[hr];
+                  break;
+                }
+                if(j==i) break;
+                if(regs[j].regmap[hr]>=0)
+                  break;
+                if(get_reg(regs[j].regmap,f_regmap[hr])>=0) {
+                  //printf("no-match due to different register\n");
+                  break;
+                }
+                if((regs[j+1].is32&(1LL<<f_regmap[hr]))!=(regs[j].is32&(1LL<<f_regmap[hr]))) {
+                  //printf("32/64 mismatch %x %d\n",start+j*4,hr);
+                  break;
+                }
+                if(itype[j]==STORE||itype[j]==STORELR||itype[j]==C1LS
+                ||itype[j]==SHIFT||itype[j]==COP1||itype[j]==FLOAT
+                ||itype[j]==FCOMP||itype[j]==FCONV) {
+                  if(count_free_regs(regs[j].regmap)<2) {
+                    //printf("No free regs for store %x\n",start+j*4);
+                    break;
+                  }
+                }
+                else if(itype[j]!=NOP&&itype[j]!=MOV&&itype[j]!=ALU&&itype[j]!=SHIFTIMM&&itype[j]!=IMM16&&itype[j]!=LOAD) break;
+                if(f_regmap[hr]>=64) {
+                  if(regs[j].is32&(1LL<<(f_regmap[hr]&63))) {
+                    break;
+                  }
+                  else
+                  {
+                    if(get_reg(regs[j].regmap,f_regmap[hr]&63)<0) {
+                      break;
+                    }
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }else{
+      int count=0;
+      for(hr=0;hr<HOST_REGS;hr++)
+      {
+        if(hr!=EXCLUDE_REG) {
+          if(regs[i].regmap[hr]>64) {
+            if(!((regs[i].dirty>>hr)&1))
+              f_regmap[hr]=regs[i].regmap[hr];
+          }
+          else if(regs[i].regmap[hr]>=0) f_regmap[hr]=regs[i].regmap[hr];
+          else if(regs[i].regmap[hr]<0) count++;
+        }
+      }
+      // Try to restore cycle count at branch targets
+      if(bt[i]) {
+        for(j=i;j<slen-1;j++) {
+          if(regs[j].regmap[HOST_CCREG]!=-1) break;
+          if(itype[j]==STORE||itype[j]==STORELR||itype[j]==C1LS
+          ||itype[j]==SHIFT||itype[j]==COP1||itype[j]==FLOAT
+          ||itype[j]==FCOMP||itype[j]==FCONV) {
+            if(count_free_regs(regs[j].regmap)<2) {
+              //printf("no free regs for store %x\n",start+j*4);
+              break;
+            }
+          }
+          else
+          if(itype[j]!=NOP&&itype[j]!=MOV&&itype[j]!=ALU&&itype[j]!=SHIFTIMM&&itype[j]!=IMM16&&itype[j]!=LOAD) break;
+        }
+        if(regs[j].regmap[HOST_CCREG]==CCREG) {
+          int k=i;
+          //printf("Extend CC, %x -> %x\n",start+k*4,start+j*4);
+          while(k<j) {
+            regs[k].regmap_entry[HOST_CCREG]=CCREG;
+            regs[k].regmap[HOST_CCREG]=CCREG;
+            regmap_pre[k+1][HOST_CCREG]=CCREG;
+            regs[k+1].wasdirty|=1<<HOST_CCREG;
+            regs[k].dirty|=1<<HOST_CCREG;
+            regs[k].wasconst&=~(1<<HOST_CCREG);
+            regs[k].isconst&=~(1<<HOST_CCREG);
+            k++;
+          }
+          regs[j].regmap_entry[HOST_CCREG]=CCREG;          
+        }
+        // Work backwards from the branch target
+        if(j>i&&f_regmap[HOST_CCREG]==CCREG)
+        {
+          //printf("Extend backwards\n");
+          int k;
+          k=i;
+          while(regs[k-1].regmap[HOST_CCREG]==-1) {
+            if(itype[k-1]==STORE||itype[k-1]==STORELR||itype[k-1]==C1LS
+            ||itype[k-1]==SHIFT||itype[k-1]==COP1||itype[k-1]==FLOAT
+            ||itype[k-1]==FCONV||itype[k-1]==FCOMP) {
+              if(count_free_regs(regs[k-1].regmap)<2) {
+                //printf("no free regs for store %x\n",start+(k-1)*4);
+                break;
+              }
+            }
+            else
+            if(itype[k-1]!=NOP&&itype[k-1]!=MOV&&itype[k-1]!=ALU&&itype[k-1]!=SHIFTIMM&&itype[k-1]!=IMM16&&itype[k-1]!=LOAD) break;
+            k--;
+          }
+          if(regs[k-1].regmap[HOST_CCREG]==CCREG) {
+            //printf("Extend CC, %x ->\n",start+k*4);
+            while(k<=i) {
+              regs[k].regmap_entry[HOST_CCREG]=CCREG;
+              regs[k].regmap[HOST_CCREG]=CCREG;
+              regmap_pre[k+1][HOST_CCREG]=CCREG;
+              regs[k+1].wasdirty|=1<<HOST_CCREG;
+              regs[k].dirty|=1<<HOST_CCREG;
+              regs[k].wasconst&=~(1<<HOST_CCREG);
+              regs[k].isconst&=~(1<<HOST_CCREG);
+              k++;
+            }
+          }
+          else {
+            //printf("Fail Extend CC, %x ->\n",start+k*4);
+          }
+        }
+      }
+      if(itype[i]!=STORE&&itype[i]!=STORELR&&itype[i]!=C1LS&&itype[i]!=SHIFT&&
+         itype[i]!=NOP&&itype[i]!=MOV&&itype[i]!=ALU&&itype[i]!=SHIFTIMM&&
+         itype[i]!=IMM16&&itype[i]!=LOAD&&itype[i]!=COP1&&itype[i]!=FLOAT&&
+         itype[i]!=FCONV&&itype[i]!=FCOMP)
+      {
+        memcpy(f_regmap,regs[i].regmap,sizeof(f_regmap));
+      }
+    }
+  }
+  
+  // This allocates registers (if possible) one instruction prior
+  // to use, which can avoid a load-use penalty on certain CPUs.
+  for(i=0;i<slen-1;i++)
+  {
+    if(!i||(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP))
+    {
+      if(!bt[i+1])
+      {
+        if(itype[i]==ALU||itype[i]==MOV||itype[i]==LOAD||itype[i]==SHIFTIMM||itype[i]==IMM16||(itype[i]==COP1&&opcode2[i]<3))
+        {
+          if(rs1[i+1]) {
+            if((hr=get_reg(regs[i+1].regmap,rs1[i+1]))>=0)
+            {
+              if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+              {
+                regs[i].regmap[hr]=regs[i+1].regmap[hr];
+                regmap_pre[i+1][hr]=regs[i+1].regmap[hr];
+                regs[i+1].regmap_entry[hr]=regs[i+1].regmap[hr];
+                regs[i].isconst&=~(1<<hr);
+                regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                constmap[i][hr]=constmap[i+1][hr];
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+            }
+          }
+          if(rs2[i+1]) {
+            if((hr=get_reg(regs[i+1].regmap,rs2[i+1]))>=0)
+            {
+              if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+              {
+                regs[i].regmap[hr]=regs[i+1].regmap[hr];
+                regmap_pre[i+1][hr]=regs[i+1].regmap[hr];
+                regs[i+1].regmap_entry[hr]=regs[i+1].regmap[hr];
+                regs[i].isconst&=~(1<<hr);
+                regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                constmap[i][hr]=constmap[i+1][hr];
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+            }
+          }
+          if(itype[i+1]==LOAD&&rs1[i+1]&&get_reg(regs[i+1].regmap,rs1[i+1])<0) {
+            if((hr=get_reg(regs[i+1].regmap,rt1[i+1]))>=0)
+            {
+              if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+              {
+                regs[i].regmap[hr]=rs1[i+1];
+                regmap_pre[i+1][hr]=rs1[i+1];
+                regs[i+1].regmap_entry[hr]=rs1[i+1];
+                regs[i].isconst&=~(1<<hr);
+                regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                constmap[i][hr]=constmap[i+1][hr];
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+            }
+          }
+          if(lt1[i+1]&&get_reg(regs[i+1].regmap,rs1[i+1])<0) {
+            if((hr=get_reg(regs[i+1].regmap,rt1[i+1]))>=0)
+            {
+              if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+              {
+                regs[i].regmap[hr]=rs1[i+1];
+                regmap_pre[i+1][hr]=rs1[i+1];
+                regs[i+1].regmap_entry[hr]=rs1[i+1];
+                regs[i].isconst&=~(1<<hr);
+                regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                constmap[i][hr]=constmap[i+1][hr];
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+            }
+          }
+          #ifndef HOST_IMM_ADDR32
+          if(itype[i+1]==LOAD||itype[i+1]==LOADLR||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS) {
+            hr=get_reg(regs[i+1].regmap,TLREG);
+            if(hr>=0) {
+              int sr=get_reg(regs[i+1].regmap,rs1[i+1]);
+              if(sr>=0&&((regs[i+1].wasconst>>sr)&1)) {
+                int nr;
+                if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+                {
+                  regs[i].regmap[hr]=MGEN1+((i+1)&1);
+                  regmap_pre[i+1][hr]=MGEN1+((i+1)&1);
+                  regs[i+1].regmap_entry[hr]=MGEN1+((i+1)&1);
+                  regs[i].isconst&=~(1<<hr);
+                  regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                  constmap[i][hr]=constmap[i+1][hr];
+                  regs[i+1].wasdirty&=~(1<<hr);
+                  regs[i].dirty&=~(1<<hr);
+                }
+                else if((nr=get_reg2(regs[i].regmap,regs[i+1].regmap,-1))>=0)
+                {
+                  // move it to another register
+                  regs[i+1].regmap[hr]=-1;
+                  regmap_pre[i+2][hr]=-1;
+                  regs[i+1].regmap[nr]=TLREG;
+                  regmap_pre[i+2][nr]=TLREG;
+                  regs[i].regmap[nr]=MGEN1+((i+1)&1);
+                  regmap_pre[i+1][nr]=MGEN1+((i+1)&1);
+                  regs[i+1].regmap_entry[nr]=MGEN1+((i+1)&1);
+                  regs[i].isconst&=~(1<<nr);
+                  regs[i+1].isconst&=~(1<<nr);
+                  regs[i].dirty&=~(1<<nr);
+                  regs[i+1].wasdirty&=~(1<<nr);
+                  regs[i+1].dirty&=~(1<<nr);
+                  regs[i+2].wasdirty&=~(1<<nr);
+                }
+              }
+            }
+          }
+          #endif
+          if(itype[i+1]==STORE||itype[i+1]==STORELR||opcode[i+1]==0x39||opcode[i+1]==0x3D) { // SB/SH/SW/SD/SWC1/SDC1
+            if(get_reg(regs[i+1].regmap,rs1[i+1])<0) {
+              hr=get_reg2(regs[i].regmap,regs[i+1].regmap,-1);
+              if(hr<0) hr=get_reg(regs[i+1].regmap,-1);
+              else {regs[i+1].regmap[hr]=AGEN1+((i+1)&1);regs[i+1].isconst&=~(1<<hr);}
+              assert(hr>=0);
+              if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+              {
+                regs[i].regmap[hr]=rs1[i+1];
+                regmap_pre[i+1][hr]=rs1[i+1];
+                regs[i+1].regmap_entry[hr]=rs1[i+1];
+                regs[i].isconst&=~(1<<hr);
+                regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                constmap[i][hr]=constmap[i+1][hr];
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+            }
+          }
+          if(itype[i+1]==LOADLR||opcode[i+1]==0x31||opcode[i+1]==0x35) { // LWC1/LDC1
+            if(get_reg(regs[i+1].regmap,rs1[i+1])<0) {
+              int nr;
+              hr=get_reg(regs[i+1].regmap,FTEMP);
+              assert(hr>=0);
+              if(regs[i].regmap[hr]<0&&regs[i+1].regmap_entry[hr]<0)
+              {
+                regs[i].regmap[hr]=rs1[i+1];
+                regmap_pre[i+1][hr]=rs1[i+1];
+                regs[i+1].regmap_entry[hr]=rs1[i+1];
+                regs[i].isconst&=~(1<<hr);
+                regs[i].isconst|=regs[i+1].isconst&(1<<hr);
+                constmap[i][hr]=constmap[i+1][hr];
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+              else if((nr=get_reg2(regs[i].regmap,regs[i+1].regmap,-1))>=0)
+              {
+                // move it to another register
+                regs[i+1].regmap[hr]=-1;
+                regmap_pre[i+2][hr]=-1;
+                regs[i+1].regmap[nr]=FTEMP;
+                regmap_pre[i+2][nr]=FTEMP;
+                regs[i].regmap[nr]=rs1[i+1];
+                regmap_pre[i+1][nr]=rs1[i+1];
+                regs[i+1].regmap_entry[nr]=rs1[i+1];
+                regs[i].isconst&=~(1<<nr);
+                regs[i+1].isconst&=~(1<<nr);
+                regs[i].dirty&=~(1<<nr);
+                regs[i+1].wasdirty&=~(1<<nr);
+                regs[i+1].dirty&=~(1<<nr);
+                regs[i+2].wasdirty&=~(1<<nr);
+              }
+            }
+          }
+          if(itype[i+1]==LOAD||itype[i+1]==LOADLR||itype[i+1]==STORE||itype[i+1]==STORELR/*||itype[i+1]==C1LS*/) {
+            if(itype[i+1]==LOAD) 
+              hr=get_reg(regs[i+1].regmap,rt1[i+1]);
+            if(itype[i+1]==LOADLR||opcode[i+1]==0x31||opcode[i+1]==0x35) // LWC1/LDC1
+              hr=get_reg(regs[i+1].regmap,FTEMP);
+            if(itype[i+1]==STORE||itype[i+1]==STORELR||opcode[i+1]==0x39||opcode[i+1]==0x3D) { // SWC1/SDC1
+              hr=get_reg(regs[i+1].regmap,AGEN1+((i+1)&1));
+              if(hr<0) hr=get_reg(regs[i+1].regmap,-1);
+            }
+            if(hr>=0&&regs[i].regmap[hr]<0) {
+              int rs=get_reg(regs[i+1].regmap,rs1[i+1]);
+              if(rs>=0&&((regs[i+1].wasconst>>rs)&1)) {
+                regs[i].regmap[hr]=AGEN1+((i+1)&1);
+                regmap_pre[i+1][hr]=AGEN1+((i+1)&1);
+                regs[i+1].regmap_entry[hr]=AGEN1+((i+1)&1);
+                regs[i].isconst&=~(1<<hr);
+                regs[i+1].wasdirty&=~(1<<hr);
+                regs[i].dirty&=~(1<<hr);
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  
+  /* Pass 6 - Optimize clean/dirty state */
+  clean_registers(0,slen-1,1);
+  
+  /* Pass 7 - Identify 32-bit registers */
+  
+  provisional_r32();
+
+  u_int r32=0;
+  
+  for (i=slen-1;i>=0;i--)
+  {
+    int hr;
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    {
+      if(ba[i]<start || ba[i]>=(start+slen*4))
+      {
+        // Branch out of this block, don't need anything
+        r32=0;
+      }
+      else
+      {
+        // Internal branch
+        // Need whatever matches the target
+        // (and doesn't get overwritten by the delay slot instruction)
+        r32=0;
+        int t=(ba[i]-start)>>2;
+        if(ba[i]>start+i*4) {
+          // Forward branch
+          if(!(requires_32bit[t]&~regs[i].was32))
+            r32|=requires_32bit[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+        }else{
+          // Backward branch
+          //if(!(regs[t].was32&~unneeded_reg_upper[t]&~regs[i].was32))
+          //  r32|=regs[t].was32&~unneeded_reg_upper[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+          if(!(pr32[t]&~regs[i].was32))
+            r32|=pr32[t]&(~(1LL<<rt1[i+1]))&(~(1LL<<rt2[i+1]));
+        }
+      }
+      // Conditional branch may need registers for following instructions
+      if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+      {
+        if(i<slen-2) {
+          r32|=requires_32bit[i+2];
+          r32&=regs[i].was32;
+          // Mark this address as a branch target since it may be called
+          // upon return from interrupt
+          bt[i+2]=1;
+        }
+      }
+      // Merge in delay slot
+      if(!likely[i]) {
+        // These are overwritten unless the branch is "likely"
+        // and the delay slot is nullified if not taken
+        r32&=~(1LL<<rt1[i+1]);
+        r32&=~(1LL<<rt2[i+1]);
+      }
+      // Assume these are needed (delay slot)
+      if(us1[i+1]>0)
+      {
+        if((regs[i].was32>>us1[i+1])&1) r32|=1LL<<us1[i+1];
+      }
+      if(us2[i+1]>0)
+      {
+        if((regs[i].was32>>us2[i+1])&1) r32|=1LL<<us2[i+1];
+      }
+      if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1))
+      {
+        if((regs[i].was32>>dep1[i+1])&1) r32|=1LL<<dep1[i+1];
+      }
+      if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1))
+      {
+        if((regs[i].was32>>dep2[i+1])&1) r32|=1LL<<dep2[i+1];
+      }
+    }
+    else if(itype[i]==SYSCALL)
+    {
+      // SYSCALL instruction (software interrupt)
+      r32=0;
+    }
+    else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
+    {
+      // ERET instruction (return from interrupt)
+      r32=0;
+    }
+    // Check 32 bits
+    r32&=~(1LL<<rt1[i]);
+    r32&=~(1LL<<rt2[i]);
+    if(us1[i]>0)
+    {
+      if((regs[i].was32>>us1[i])&1) r32|=1LL<<us1[i];
+    }
+    if(us2[i]>0)
+    {
+      if((regs[i].was32>>us2[i])&1) r32|=1LL<<us2[i];
+    }
+    if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1))
+    {
+      if((regs[i].was32>>dep1[i])&1) r32|=1LL<<dep1[i];
+    }
+    if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1))
+    {
+      if((regs[i].was32>>dep2[i])&1) r32|=1LL<<dep2[i];
+    }
+    requires_32bit[i]=r32;
+    
+    // Dirty registers which are 32-bit, require 32-bit input
+    // as they will be written as 32-bit values
+    for(hr=0;hr<HOST_REGS;hr++)
+    {
+      if(regs[i].regmap_entry[hr]>0&&regs[i].regmap_entry[hr]<64) {
+        if((regs[i].was32>>regs[i].regmap_entry[hr])&(regs[i].wasdirty>>hr)&1) {
+          if(!((unneeded_reg_upper[i]>>regs[i].regmap_entry[hr])&1))
+          requires_32bit[i]|=1LL<<regs[i].regmap_entry[hr];
+        }
+      }
+    }
+    //requires_32bit[i]=is32[i]&~unneeded_reg_upper[i]; // DEBUG
+  }
+
+  if(itype[slen-1]==SPAN) {
+    bt[slen-1]=1; // Mark as a branch target so instruction can restart after exception
+  }
+  
+  /* Debug/disassembly */
+  if((void*)assem_debug==(void*)printf) 
+  for(i=0;i<slen;i++)
+  {
+    printf("U:");
+    int r;
+    for(r=1;r<=CCREG;r++) {
+      if((unneeded_reg[i]>>r)&1) {
+        if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf(" UU:");
+    for(r=1;r<=CCREG;r++) {
+      if(((unneeded_reg_upper[i]&~unneeded_reg[i])>>r)&1) {
+        if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf(" 32:");
+    for(r=0;r<=CCREG;r++) {
+      //if(((is32[i]>>r)&(~unneeded_reg[i]>>r))&1) {
+      if((regs[i].was32>>r)&1) {
+        if(r==CCREG) printf(" CC");
+        else if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf("\n");
+    #if defined(__i386__) || defined(__x86_64__)
+    printf("pre: eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",regmap_pre[i][0],regmap_pre[i][1],regmap_pre[i][2],regmap_pre[i][3],regmap_pre[i][5],regmap_pre[i][6],regmap_pre[i][7]);
+    #endif
+    #ifdef __arm__
+    printf("pre: r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d\n",regmap_pre[i][0],regmap_pre[i][1],regmap_pre[i][2],regmap_pre[i][3],regmap_pre[i][4],regmap_pre[i][5],regmap_pre[i][6],regmap_pre[i][7],regmap_pre[i][8],regmap_pre[i][9],regmap_pre[i][10],regmap_pre[i][12]);
+    #endif
+    printf("needs: ");
+    if(needed_reg[i]&1) printf("eax ");
+    if((needed_reg[i]>>1)&1) printf("ecx ");
+    if((needed_reg[i]>>2)&1) printf("edx ");
+    if((needed_reg[i]>>3)&1) printf("ebx ");
+    if((needed_reg[i]>>5)&1) printf("ebp ");
+    if((needed_reg[i]>>6)&1) printf("esi ");
+    if((needed_reg[i]>>7)&1) printf("edi ");
+    printf("r:");
+    for(r=0;r<=CCREG;r++) {
+      //if(((requires_32bit[i]>>r)&(~unneeded_reg[i]>>r))&1) {
+      if((requires_32bit[i]>>r)&1) {
+        if(r==CCREG) printf(" CC");
+        else if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf("\n");
+    /*printf("pr:");
+    for(r=0;r<=CCREG;r++) {
+      //if(((requires_32bit[i]>>r)&(~unneeded_reg[i]>>r))&1) {
+      if((pr32[i]>>r)&1) {
+        if(r==CCREG) printf(" CC");
+        else if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    if(pr32[i]!=requires_32bit[i]) printf(" OOPS");
+    printf("\n");*/
+    #if defined(__i386__) || defined(__x86_64__)
+    printf("entry: eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",regs[i].regmap_entry[0],regs[i].regmap_entry[1],regs[i].regmap_entry[2],regs[i].regmap_entry[3],regs[i].regmap_entry[5],regs[i].regmap_entry[6],regs[i].regmap_entry[7]);
+    printf("dirty: ");
+    if(regs[i].wasdirty&1) printf("eax ");
+    if((regs[i].wasdirty>>1)&1) printf("ecx ");
+    if((regs[i].wasdirty>>2)&1) printf("edx ");
+    if((regs[i].wasdirty>>3)&1) printf("ebx ");
+    if((regs[i].wasdirty>>5)&1) printf("ebp ");
+    if((regs[i].wasdirty>>6)&1) printf("esi ");
+    if((regs[i].wasdirty>>7)&1) printf("edi ");
+    #endif
+    #ifdef __arm__
+    printf("entry: r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d\n",regs[i].regmap_entry[0],regs[i].regmap_entry[1],regs[i].regmap_entry[2],regs[i].regmap_entry[3],regs[i].regmap_entry[4],regs[i].regmap_entry[5],regs[i].regmap_entry[6],regs[i].regmap_entry[7],regs[i].regmap_entry[8],regs[i].regmap_entry[9],regs[i].regmap_entry[10],regs[i].regmap_entry[12]);
+    printf("dirty: ");
+    if(regs[i].wasdirty&1) printf("r0 ");
+    if((regs[i].wasdirty>>1)&1) printf("r1 ");
+    if((regs[i].wasdirty>>2)&1) printf("r2 ");
+    if((regs[i].wasdirty>>3)&1) printf("r3 ");
+    if((regs[i].wasdirty>>4)&1) printf("r4 ");
+    if((regs[i].wasdirty>>5)&1) printf("r5 ");
+    if((regs[i].wasdirty>>6)&1) printf("r6 ");
+    if((regs[i].wasdirty>>7)&1) printf("r7 ");
+    if((regs[i].wasdirty>>8)&1) printf("r8 ");
+    if((regs[i].wasdirty>>9)&1) printf("r9 ");
+    if((regs[i].wasdirty>>10)&1) printf("r10 ");
+    if((regs[i].wasdirty>>12)&1) printf("r12 ");
+    #endif
+    printf("\n");
+    disassemble_inst(i);
+    //printf ("ccadj[%d] = %d\n",i,ccadj[i]);
+    #if defined(__i386__) || defined(__x86_64__)
+    printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d dirty: ",regs[i].regmap[0],regs[i].regmap[1],regs[i].regmap[2],regs[i].regmap[3],regs[i].regmap[5],regs[i].regmap[6],regs[i].regmap[7]);
+    if(regs[i].dirty&1) printf("eax ");
+    if((regs[i].dirty>>1)&1) printf("ecx ");
+    if((regs[i].dirty>>2)&1) printf("edx ");
+    if((regs[i].dirty>>3)&1) printf("ebx ");
+    if((regs[i].dirty>>5)&1) printf("ebp ");
+    if((regs[i].dirty>>6)&1) printf("esi ");
+    if((regs[i].dirty>>7)&1) printf("edi ");
+    #endif
+    #ifdef __arm__
+    printf("r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d dirty: ",regs[i].regmap[0],regs[i].regmap[1],regs[i].regmap[2],regs[i].regmap[3],regs[i].regmap[4],regs[i].regmap[5],regs[i].regmap[6],regs[i].regmap[7],regs[i].regmap[8],regs[i].regmap[9],regs[i].regmap[10],regs[i].regmap[12]);
+    if(regs[i].dirty&1) printf("r0 ");
+    if((regs[i].dirty>>1)&1) printf("r1 ");
+    if((regs[i].dirty>>2)&1) printf("r2 ");
+    if((regs[i].dirty>>3)&1) printf("r3 ");
+    if((regs[i].dirty>>4)&1) printf("r4 ");
+    if((regs[i].dirty>>5)&1) printf("r5 ");
+    if((regs[i].dirty>>6)&1) printf("r6 ");
+    if((regs[i].dirty>>7)&1) printf("r7 ");
+    if((regs[i].dirty>>8)&1) printf("r8 ");
+    if((regs[i].dirty>>9)&1) printf("r9 ");
+    if((regs[i].dirty>>10)&1) printf("r10 ");
+    if((regs[i].dirty>>12)&1) printf("r12 ");
+    #endif
+    printf("\n");
+    if(regs[i].isconst) {
+      printf("constants: ");
+      #if defined(__i386__) || defined(__x86_64__)
+      if(regs[i].isconst&1) printf("eax=%x ",(int)constmap[i][0]);
+      if((regs[i].isconst>>1)&1) printf("ecx=%x ",(int)constmap[i][1]);
+      if((regs[i].isconst>>2)&1) printf("edx=%x ",(int)constmap[i][2]);
+      if((regs[i].isconst>>3)&1) printf("ebx=%x ",(int)constmap[i][3]);
+      if((regs[i].isconst>>5)&1) printf("ebp=%x ",(int)constmap[i][5]);
+      if((regs[i].isconst>>6)&1) printf("esi=%x ",(int)constmap[i][6]);
+      if((regs[i].isconst>>7)&1) printf("edi=%x ",(int)constmap[i][7]);
+      #endif
+      #ifdef __arm__
+      if(regs[i].isconst&1) printf("r0=%x ",(int)constmap[i][0]);
+      if((regs[i].isconst>>1)&1) printf("r1=%x ",(int)constmap[i][1]);
+      if((regs[i].isconst>>2)&1) printf("r2=%x ",(int)constmap[i][2]);
+      if((regs[i].isconst>>3)&1) printf("r3=%x ",(int)constmap[i][3]);
+      if((regs[i].isconst>>4)&1) printf("r4=%x ",(int)constmap[i][4]);
+      if((regs[i].isconst>>5)&1) printf("r5=%x ",(int)constmap[i][5]);
+      if((regs[i].isconst>>6)&1) printf("r6=%x ",(int)constmap[i][6]);
+      if((regs[i].isconst>>7)&1) printf("r7=%x ",(int)constmap[i][7]);
+      if((regs[i].isconst>>8)&1) printf("r8=%x ",(int)constmap[i][8]);
+      if((regs[i].isconst>>9)&1) printf("r9=%x ",(int)constmap[i][9]);
+      if((regs[i].isconst>>10)&1) printf("r10=%x ",(int)constmap[i][10]);
+      if((regs[i].isconst>>12)&1) printf("r12=%x ",(int)constmap[i][12]);
+      #endif
+      printf("\n");
+    }
+    printf(" 32:");
+    for(r=0;r<=CCREG;r++) {
+      if((regs[i].is32>>r)&1) {
+        if(r==CCREG) printf(" CC");
+        else if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    printf("\n");
+    /*printf(" p32:");
+    for(r=0;r<=CCREG;r++) {
+      if((p32[i]>>r)&1) {
+        if(r==CCREG) printf(" CC");
+        else if(r==HIREG) printf(" HI");
+        else if(r==LOREG) printf(" LO");
+        else printf(" r%d",r);
+      }
+    }
+    if(p32[i]!=regs[i].is32) printf(" NO MATCH\n");
+    else printf("\n");*/
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
+      #if defined(__i386__) || defined(__x86_64__)
+      printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d dirty: ",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
+      if(branch_regs[i].dirty&1) printf("eax ");
+      if((branch_regs[i].dirty>>1)&1) printf("ecx ");
+      if((branch_regs[i].dirty>>2)&1) printf("edx ");
+      if((branch_regs[i].dirty>>3)&1) printf("ebx ");
+      if((branch_regs[i].dirty>>5)&1) printf("ebp ");
+      if((branch_regs[i].dirty>>6)&1) printf("esi ");
+      if((branch_regs[i].dirty>>7)&1) printf("edi ");
+      #endif
+      #ifdef __arm__
+      printf("branch(%d): r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d dirty: ",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[4],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7],branch_regs[i].regmap[8],branch_regs[i].regmap[9],branch_regs[i].regmap[10],branch_regs[i].regmap[12]);
+      if(branch_regs[i].dirty&1) printf("r0 ");
+      if((branch_regs[i].dirty>>1)&1) printf("r1 ");
+      if((branch_regs[i].dirty>>2)&1) printf("r2 ");
+      if((branch_regs[i].dirty>>3)&1) printf("r3 ");
+      if((branch_regs[i].dirty>>4)&1) printf("r4 ");
+      if((branch_regs[i].dirty>>5)&1) printf("r5 ");
+      if((branch_regs[i].dirty>>6)&1) printf("r6 ");
+      if((branch_regs[i].dirty>>7)&1) printf("r7 ");
+      if((branch_regs[i].dirty>>8)&1) printf("r8 ");
+      if((branch_regs[i].dirty>>9)&1) printf("r9 ");
+      if((branch_regs[i].dirty>>10)&1) printf("r10 ");
+      if((branch_regs[i].dirty>>12)&1) printf("r12 ");
+      #endif
+      printf(" 32:");
+      for(r=0;r<=CCREG;r++) {
+        if((branch_regs[i].is32>>r)&1) {
+          if(r==CCREG) printf(" CC");
+          else if(r==HIREG) printf(" HI");
+          else if(r==LOREG) printf(" LO");
+          else printf(" r%d",r);
+        }
+      }
+      printf("\n");
+    }
+  }
+
+  /* Pass 8 - Assembly */
+  linkcount=0;stubcount=0;
+  ds=0;is_delayslot=0;
+  cop1_usable=0;
+  uint64_t is32_pre=0;
+  u_int dirty_pre=0;
+  u_int beginning=(u_int)out;
+  if((u_int)addr&1) {
+    ds=1;
+    pagespan_ds();
+  }
+  for(i=0;i<slen;i++)
+  {
+    //if(ds) printf("ds: ");
+    if((void*)assem_debug==(void*)printf) disassemble_inst(i);
+    if(ds) {
+      ds=0; // Skip delay slot
+      if(bt[i]) assem_debug("OOPS - branch into delay slot\n");
+      instr_addr[i]=0;
+    } else {
+      #ifndef DESTRUCTIVE_WRITEBACK
+      if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
+      {
+        wb_sx(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,is32_pre,regs[i].was32,
+              unneeded_reg[i],unneeded_reg_upper[i]);
+        wb_valid(regmap_pre[i],regs[i].regmap_entry,dirty_pre,regs[i].wasdirty,is32_pre,
+              unneeded_reg[i],unneeded_reg_upper[i]);
+      }
+      is32_pre=regs[i].is32;
+      dirty_pre=regs[i].dirty;
+      #endif
+      // write back
+      if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
+      {
+        wb_invalidate(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,regs[i].was32,
+                      unneeded_reg[i],unneeded_reg_upper[i]);
+        loop_preload(regmap_pre[i],regs[i].regmap_entry);
+      }
+      // branch target entry point
+      instr_addr[i]=(u_int)out;
+      assem_debug("<->\n");
+      // load regs
+      if(regs[i].regmap_entry[HOST_CCREG]==CCREG&&regs[i].regmap[HOST_CCREG]!=CCREG)
+        wb_register(CCREG,regs[i].regmap_entry,regs[i].wasdirty,regs[i].was32);
+      load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i],rs2[i]);
+      address_generation(i,&regs[i],regs[i].regmap_entry);
+      load_consts(regmap_pre[i],regs[i].regmap,regs[i].was32,i);
+      if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+      {
+        // Load the delay slot registers if necessary
+        if(rs1[i+1]!=rs1[i]&&rs1[i+1]!=rs2[i])
+          load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i+1],rs1[i+1]);
+        if(rs2[i+1]!=rs1[i+1]&&rs2[i+1]!=rs1[i]&&rs2[i+1]!=rs2[i])
+          load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs2[i+1],rs2[i+1]);
+        if(itype[i+1]==STORE||itype[i+1]==STORELR||(opcode[i+1]&0x3b)==0x39)
+          load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,INVCP,INVCP);
+      }
+      else if(i+1<slen)
+      {
+        // Preload registers for following instruction
+        if(rs1[i+1]!=rs1[i]&&rs1[i+1]!=rs2[i])
+          if(rs1[i+1]!=rt1[i]&&rs1[i+1]!=rt2[i])
+            load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i+1],rs1[i+1]);
+        if(rs2[i+1]!=rs1[i+1]&&rs2[i+1]!=rs1[i]&&rs2[i+1]!=rs2[i])
+          if(rs2[i+1]!=rt1[i]&&rs2[i+1]!=rt2[i])
+            load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs2[i+1],rs2[i+1]);
+      }
+      // TODO: if(is_ooo(i)) address_generation(i+1);
+      if(itype[i]==CJUMP||itype[i]==FJUMP)
+        load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,CCREG,CCREG);
+      if(itype[i]==STORE||itype[i]==STORELR||(opcode[i]&0x3b)==0x39)
+        load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,INVCP,INVCP);
+      if(bt[i]) cop1_usable=0;
+      // assemble
+      switch(itype[i]) {
+        case ALU:
+          alu_assemble(i,&regs[i]);break;
+        case IMM16:
+          imm16_assemble(i,&regs[i]);break;
+        case SHIFT:
+          shift_assemble(i,&regs[i]);break;
+        case SHIFTIMM:
+          shiftimm_assemble(i,&regs[i]);break;
+        case LOAD:
+          load_assemble(i,&regs[i]);break;
+        case LOADLR:
+          loadlr_assemble(i,&regs[i]);break;
+        case STORE:
+          store_assemble(i,&regs[i]);break;
+        case STORELR:
+          storelr_assemble(i,&regs[i]);break;
+        case COP0:
+          cop0_assemble(i,&regs[i]);break;
+        case COP1:
+          cop1_assemble(i,&regs[i]);break;
+        case C1LS:
+          c1ls_assemble(i,&regs[i]);break;
+        case FCONV:
+          fconv_assemble(i,&regs[i]);break;
+        case FLOAT:
+          float_assemble(i,&regs[i]);break;
+        case FCOMP:
+          fcomp_assemble(i,&regs[i]);break;
+        case MULTDIV:
+          multdiv_assemble(i,&regs[i]);break;
+        case MOV:
+          mov_assemble(i,&regs[i]);break;
+        case SYSCALL:
+          syscall_assemble(i,&regs[i]);break;
+        case UJUMP:
+          ujump_assemble(i,&regs[i]);ds=1;break;
+        case RJUMP:
+          rjump_assemble(i,&regs[i]);ds=1;break;
+        case CJUMP:
+          cjump_assemble(i,&regs[i]);ds=1;break;
+        case SJUMP:
+          sjump_assemble(i,&regs[i]);ds=1;break;
+        case FJUMP:
+          fjump_assemble(i,&regs[i]);ds=1;break;
+        case SPAN:
+          pagespan_assemble(i,&regs[i]);break;
+      }
+      if(itype[i]==UJUMP||itype[i]==RJUMP||(source[i]>>16)==0x1000)
+        literal_pool(1024);
+      else
+        literal_pool_jumpover(256);
+    }
+  }
+  //assert(itype[i-2]==UJUMP||itype[i-2]==RJUMP||(source[i-2]>>16)==0x1000);
+  // If the block did not end with an unconditional branch,
+  // add a jump to the next instruction.
+  if(i>1) {
+    if(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000&&itype[i-1]!=SPAN) {
+      assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP);
+      assert(i==slen);
+      if(itype[i-2]!=CJUMP&&itype[i-2]!=SJUMP&&itype[i-2]!=FJUMP) {
+        store_regs_bt(regs[i-1].regmap,regs[i-1].is32,regs[i-1].dirty,start+i*4);
+        if(regs[i-1].regmap[HOST_CCREG]!=CCREG)
+          emit_loadreg(CCREG,HOST_CCREG);
+        emit_addimm(HOST_CCREG,CLOCK_DIVIDER*(ccadj[i-1]+1),HOST_CCREG);
+      }
+      else if(!likely[i-2])
+      {
+        store_regs_bt(branch_regs[i-2].regmap,branch_regs[i-2].is32,branch_regs[i-2].dirty,start+i*4);
+        assert(branch_regs[i-2].regmap[HOST_CCREG]==CCREG);
+      }
+      else
+      {
+        store_regs_bt(regs[i-2].regmap,regs[i-2].is32,regs[i-2].dirty,start+i*4);
+        assert(regs[i-2].regmap[HOST_CCREG]==CCREG);
+      }
+      add_to_linker((int)out,start+i*4,0);
+      emit_jmp(0);
+    }
+  }
+  else
+  {
+    assert(i>0);
+    assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP);
+    store_regs_bt(regs[i-1].regmap,regs[i-1].is32,regs[i-1].dirty,start+i*4);
+    if(regs[i-1].regmap[HOST_CCREG]!=CCREG)
+      emit_loadreg(CCREG,HOST_CCREG);
+    emit_addimm(HOST_CCREG,CLOCK_DIVIDER*(ccadj[i-1]+1),HOST_CCREG);
+    add_to_linker((int)out,start+i*4,0);
+    emit_jmp(0);
+  }
+
+  // TODO: delay slot stubs?
+  // Stubs
+  for(i=0;i<stubcount;i++)
+  {
+    switch(stubs[i][0])
+    {
+      case LOADB_STUB:
+      case LOADH_STUB:
+      case LOADW_STUB:
+      case LOADD_STUB:
+      case LOADBU_STUB:
+      case LOADHU_STUB:
+        do_readstub(i);break;
+      case STOREB_STUB:
+      case STOREH_STUB:
+      case STOREW_STUB:
+      case STORED_STUB:
+        do_writestub(i);break;
+      case CC_STUB:
+        do_ccstub(i);break;
+      case INVCODE_STUB:
+        do_invstub(i);break;
+      case FP_STUB:
+        do_cop1stub(i);break;
+      case STORELR_STUB:
+        do_unalignedwritestub(i);break;
+    }
+  }
+
+  /* Pass 9 - Linker */
+  for(i=0;i<linkcount;i++)
+  {
+    assem_debug("%8x -> %8x\n",link_addr[i][0],link_addr[i][1]);
+    literal_pool(64);
+    if(!link_addr[i][2])
+    {
+      void *stub=out;
+      void *addr=check_addr(link_addr[i][1]);
+      emit_extjump(link_addr[i][0],link_addr[i][1]);
+      if(addr) {
+        set_jump_target(link_addr[i][0],(int)addr);
+        add_link(link_addr[i][1],stub);
+      }
+      else set_jump_target(link_addr[i][0],(int)stub);
+    }
+    else
+    {
+      // Internal branch
+      int target=(link_addr[i][1]-start)>>2;
+      assert(target>=0&&target<slen);
+      assert(instr_addr[target]);
+      //#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+      //set_jump_target_fillslot(link_addr[i][0],instr_addr[target],link_addr[i][2]>>1);
+      //#else
+      set_jump_target(link_addr[i][0],instr_addr[target]);
+      //#endif
+    }
+  }
+  // External Branch Targets (jump_in)
+  if(copy+slen*4>(void *)shadow+sizeof(shadow)) copy=shadow;
+  for(i=0;i<slen;i++)
+  {
+    if(bt[i]||i==0)
+    {
+      if(instr_addr[i]) // TODO - delay slots (=null)
+      {
+        u_int vaddr=start+i*4;
+        u_int page=(0x80000000^vaddr)>>12;
+        u_int vpage=page;
+        if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[page^0x80000]^0x80000000)>>12;
+        if(page>2048) page=2048+(page&2047);
+        if(vpage>262143&&tlb_LUT_r[vaddr>>12]) vpage&=2047; // jump_dirty uses a hash of the virtual address instead
+        if(vpage>2048) vpage=2048+(vpage&2047);
+        literal_pool(256);
+        //if(!(is32[i]&(~unneeded_reg_upper[i])&~(1LL<<CCREG)))
+        if(!requires_32bit[i])
+        {
+          assem_debug("%8x (%d) <- %8x\n",instr_addr[i],i,start+i*4);
+          assem_debug("jump_in: %x\n",start+i*4);
+          ll_add(jump_dirty+vpage,vaddr,(void *)out);
+          int entry_point=do_dirty_stub(i);
+          ll_add(jump_in+page,vaddr,(void *)entry_point);
+          // If there was an existing entry in the hash table,
+          // replace it with the new address.
+          // Don't add new entries.  We'll insert the
+          // ones that actually get used in check_addr().
+          int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+          if(ht_bin[0]==vaddr) {
+            ht_bin[1]=entry_point;
+          }
+          if(ht_bin[2]==vaddr) {
+            ht_bin[3]=entry_point;
+          }
+        }
+        else
+        {
+          u_int r=requires_32bit[i]|!!(requires_32bit[i]>>32);
+          assem_debug("%8x (%d) <- %8x\n",instr_addr[i],i,start+i*4);
+          assem_debug("jump_in: %x (restricted - %x)\n",start+i*4,r);
+          //int entry_point=(int)out;
+          ////assem_debug("entry_point: %x\n",entry_point);
+          //load_regs_entry(i);
+          //if(entry_point==(int)out)
+          //  entry_point=instr_addr[i];
+          //else
+          //  emit_jmp(instr_addr[i]);
+          //ll_add_32(jump_in+page,vaddr,r,(void *)entry_point);
+          ll_add_32(jump_dirty+vpage,vaddr,r,(void *)out);
+          int entry_point=do_dirty_stub(i);
+          ll_add_32(jump_in+page,vaddr,r,(void *)entry_point);
+        }
+      }
+    }
+  }
+  // Write out the literal pool if necessary
+  literal_pool(0);
+  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+  // Align code
+  if(((u_int)out)&7) emit_addnop(13);
+  #endif
+  assert((u_int)out-beginning<MAX_OUTPUT_BLOCK_SIZE);
+  //printf("shadow buffer: %x-%x\n",(int)copy,(int)copy+slen*4);
+  memcpy(copy,source,slen*4);
+  copy+=slen*4;
+  
+  #ifdef __arm__
+  __clear_cache((void *)beginning,out);
+  #endif
+  
+  // If we're within 256K of the end of the buffer,
+  // start over from the beginning. (Is 256K enough?)
+  if((int)out>BASE_ADDR+(1<<TARGET_SIZE_2)-MAX_OUTPUT_BLOCK_SIZE) out=(u_char *)BASE_ADDR;
+  
+  // Trap writes to any of the pages we compiled
+  for(i=start>>12;i<=(start+slen*4)>>12;i++) {
+    invalid_code[i]=0;
+    memory_map[i]|=0x40000000;
+    if((signed int)start>=(signed int)0xC0000000) {
+      assert(using_tlb);
+      j=(((u_int)i<<12)+(memory_map[i]<<2)-(u_int)rdram+(u_int)0x80000000)>>12;
+      invalid_code[j]=0;
+      memory_map[j]|=0x40000000;
+      //printf("write protect physical page: %x (virtual %x)\n",j<<12,start);
+    }
+  }
+  
+  /* Pass 10 - Free memory by expiring oldest blocks */
+  
+  int end=((((int)out-BASE_ADDR)>>(TARGET_SIZE_2-16))+16384)&65535;
+  while(expirep!=end)
+  {
+    int shift=TARGET_SIZE_2-3; // Divide into 8 blocks
+    int base=BASE_ADDR+((expirep>>13)<<shift); // Base address of this block
+    inv_debug("EXP: Phase %d\n",expirep);
+    switch((expirep>>11)&3)
+    {
+      case 0:
+        // Clear jump_in and jump_dirty
+        ll_remove_matching_addrs(jump_in+(expirep&2047),base,shift);
+        ll_remove_matching_addrs(jump_dirty+(expirep&2047),base,shift);
+        ll_remove_matching_addrs(jump_in+2048+(expirep&2047),base,shift);
+        ll_remove_matching_addrs(jump_dirty+2048+(expirep&2047),base,shift);
+        break;
+      case 1:
+        // Clear pointers
+        ll_kill_pointers(jump_out[expirep&2047],base,shift);
+        ll_kill_pointers(jump_out[(expirep&2047)+2048],base,shift);
+        break;
+      case 2:
+        // Clear hash table
+        for(i=0;i<32;i++) {
+          int *ht_bin=hash_table[((expirep&2047)<<5)+i];
+          if((ht_bin[3]>>shift)==(base>>shift) ||
+             ((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
+            inv_debug("EXP: Remove hash %x -> %x\n",ht_bin[2],ht_bin[3]);
+            ht_bin[2]=ht_bin[3]=-1;
+          }
+          if((ht_bin[1]>>shift)==(base>>shift) ||
+             ((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
+            inv_debug("EXP: Remove hash %x -> %x\n",ht_bin[0],ht_bin[1]);
+            ht_bin[0]=ht_bin[2];
+            ht_bin[1]=ht_bin[3];
+            ht_bin[2]=ht_bin[3]=-1;
+          }
+        }
+        break;
+      case 3:
+        // Clear jump_out
+        #ifdef __arm__
+        if((expirep&2047)==0)
+          __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
+        #endif
+        ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift);
+        ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base,shift);
+        break;
+    }
+    expirep=(expirep+1)&65535;
+  }
+  return 0;
+}
diff --git a/libpcsxcore/new_dynarec/new_dynarec.h b/libpcsxcore/new_dynarec/new_dynarec.h
new file mode 100644 (file)
index 0000000..8bb0dca
--- /dev/null
@@ -0,0 +1,4 @@
+#define NEW_DYNAREC 1
+
+extern int pcaddr;
+extern int pending_exception;