From: notaz <notasas@gmail.com>
Date: Tue, 2 Nov 2021 15:47:56 +0000 (+0200)
Subject: drc: move some stuff out of assem_arm for reuse
X-Git-Tag: r23~91
X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=8062d65a99a541d3092672655e11cb2babfe3857;p=pcsx_rearmed.git

drc: move some stuff out of assem_arm for reuse
---

diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c
index 0e0acdc7..4953faa0 100644
--- a/libpcsxcore/new_dynarec/assem_arm.c
+++ b/libpcsxcore/new_dynarec/assem_arm.c
@@ -329,276 +329,6 @@ static void get_bounds(void *addr, u_char **start, u_char **end)
   *end=(u_char *)source+len;
 }
 
-/* Register allocation */
-
-// Note: registers are allocated clean (unmodified state)
-// if you intend to modify the register, you must call dirty_reg().
-static 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];
-  assert(r < 64);
-  if((cur->u>>r)&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) {
-      assert(r < 64);
-      if((cur->u>>r)&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)) 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;
-          }
-        }
-      }
-    }
-  }
-  SysPrintf("This shouldn't happen (alloc_reg)");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
-static 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) {
-      assert(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;
-        }
-      }
-    }
-  }
-
-  // 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)) 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;
-          }
-        }
-      }
-    }
-  }
-  SysPrintf("This shouldn't happen");exit(1);
-}
-
 // Allocate a specific ARM register.
 static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
 {
@@ -626,9 +356,6 @@ static void alloc_cc(struct regstat *cur,int i)
   alloc_arm_reg(cur,i,CCREG,HOST_CCREG);
 }
 
-/* Special alloc */
-
-
 /* Assembler */
 
 static unused char regname[16][4] = {
@@ -1320,47 +1047,6 @@ static void emit_set_if_carry32(int rs1, int rs2, int rt)
   emit_cmovb_imm(1,rt);
 }
 
-#ifdef DRC_DBG
-extern void gen_interupt();
-extern void do_insn_cmp();
-#define FUNCNAME(f) { (intptr_t)f, " " #f }
-static const struct {
-  intptr_t addr;
-  const char *name;
-} function_names[] = {
-  FUNCNAME(cc_interrupt),
-  FUNCNAME(gen_interupt),
-  FUNCNAME(get_addr_ht),
-  FUNCNAME(get_addr),
-  FUNCNAME(jump_handler_read8),
-  FUNCNAME(jump_handler_read16),
-  FUNCNAME(jump_handler_read32),
-  FUNCNAME(jump_handler_write8),
-  FUNCNAME(jump_handler_write16),
-  FUNCNAME(jump_handler_write32),
-  FUNCNAME(invalidate_addr),
-  FUNCNAME(verify_code_vm),
-  FUNCNAME(verify_code),
-  FUNCNAME(jump_hlecall),
-  FUNCNAME(jump_syscall_hle),
-  FUNCNAME(new_dyna_leave),
-  FUNCNAME(pcsx_mtc0),
-  FUNCNAME(pcsx_mtc0_ds),
-  FUNCNAME(do_insn_cmp),
-};
-
-static const char *func_name(intptr_t a)
-{
-  int i;
-  for (i = 0; i < sizeof(function_names)/sizeof(function_names[0]); i++)
-    if (function_names[i].addr == a)
-      return function_names[i].name;
-  return "";
-}
-#else
-#define func_name(x) ""
-#endif
-
 static void emit_call(const void *a_)
 {
   int a = (int)a_;
@@ -2467,28 +2153,6 @@ static void do_dirty_stub_ds()
   emit_call(&verify_code_ds);
 }
 
-// FP_STUB
-static void do_cop1stub(int n)
-{
-  literal_pool(256);
-  assem_debug("do_cop1stub %x\n",start+stubs[n].a*4);
-  set_jump_target(stubs[n].addr, out);
-  int i=stubs[n].a;
-//  int rs=stubs[n].b;
-  struct regstat *i_regs=(struct regstat *)stubs[n].c;
-  int ds=stubs[n].d;
-  if(!ds) {
-    load_all_consts(regs[i].regmap_entry,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->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_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
-  emit_jmp(ds?fp_exception_ds:fp_exception);
-}
-
 /* Special assem */
 
 static void shift_assemble_arm(int i,struct regstat *i_regs)
@@ -2600,188 +2264,6 @@ static void shift_assemble_arm(int i,struct regstat *i_regs)
     }
   }
 }
-
-static void speculate_mov(int rs,int rt)
-{
-  if(rt!=0) {
-    smrv_strong_next|=1<<rt;
-    smrv[rt]=smrv[rs];
-  }
-}
-
-static void speculate_mov_weak(int rs,int rt)
-{
-  if(rt!=0) {
-    smrv_weak_next|=1<<rt;
-    smrv[rt]=smrv[rs];
-  }
-}
-
-static void speculate_register_values(int i)
-{
-  if(i==0) {
-    memcpy(smrv,psxRegs.GPR.r,sizeof(smrv));
-    // gp,sp are likely to stay the same throughout the block
-    smrv_strong_next=(1<<28)|(1<<29)|(1<<30);
-    smrv_weak_next=~smrv_strong_next;
-    //printf(" llr %08x\n", smrv[4]);
-  }
-  smrv_strong=smrv_strong_next;
-  smrv_weak=smrv_weak_next;
-  switch(itype[i]) {
-    case ALU:
-      if     ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]);
-      else if((smrv_strong>>rs2[i])&1) speculate_mov(rs2[i],rt1[i]);
-      else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]);
-      else if((smrv_weak>>rs2[i])&1) speculate_mov_weak(rs2[i],rt1[i]);
-      else {
-        smrv_strong_next&=~(1<<rt1[i]);
-        smrv_weak_next&=~(1<<rt1[i]);
-      }
-      break;
-    case SHIFTIMM:
-      smrv_strong_next&=~(1<<rt1[i]);
-      smrv_weak_next&=~(1<<rt1[i]);
-      // fallthrough
-    case IMM16:
-      if(rt1[i]&&is_const(&regs[i],rt1[i])) {
-        int value,hr=get_reg(regs[i].regmap,rt1[i]);
-        if(hr>=0) {
-          if(get_final_value(hr,i,&value))
-               smrv[rt1[i]]=value;
-          else smrv[rt1[i]]=constmap[i][hr];
-          smrv_strong_next|=1<<rt1[i];
-        }
-      }
-      else {
-        if     ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]);
-        else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]);
-      }
-      break;
-    case LOAD:
-      if(start<0x2000&&(rt1[i]==26||(smrv[rt1[i]]>>24)==0xa0)) {
-        // special case for BIOS
-        smrv[rt1[i]]=0xa0000000;
-        smrv_strong_next|=1<<rt1[i];
-        break;
-      }
-      // fallthrough
-    case SHIFT:
-    case LOADLR:
-    case MOV:
-      smrv_strong_next&=~(1<<rt1[i]);
-      smrv_weak_next&=~(1<<rt1[i]);
-      break;
-    case COP0:
-    case COP2:
-      if(opcode2[i]==0||opcode2[i]==2) { // MFC/CFC
-        smrv_strong_next&=~(1<<rt1[i]);
-        smrv_weak_next&=~(1<<rt1[i]);
-      }
-      break;
-    case C2LS:
-      if (opcode[i]==0x32) { // LWC2
-        smrv_strong_next&=~(1<<rt1[i]);
-        smrv_weak_next&=~(1<<rt1[i]);
-      }
-      break;
-  }
-#if 0
-  int r=4;
-  printf("x %08x %08x %d %d c %08x %08x\n",smrv[r],start+i*4,
-    ((smrv_strong>>r)&1),(smrv_weak>>r)&1,regs[i].isconst,regs[i].wasconst);
-#endif
-}
-
-enum {
-  MTYPE_8000 = 0,
-  MTYPE_8020,
-  MTYPE_0000,
-  MTYPE_A000,
-  MTYPE_1F80,
-};
-
-static int get_ptr_mem_type(u_int a)
-{
-  if(a < 0x00200000) {
-    if(a<0x1000&&((start>>20)==0xbfc||(start>>24)==0xa0))
-      // return wrong, must use memhandler for BIOS self-test to pass
-      // 007 does similar stuff from a00 mirror, weird stuff
-      return MTYPE_8000;
-    return MTYPE_0000;
-  }
-  if(0x1f800000 <= a && a < 0x1f801000)
-    return MTYPE_1F80;
-  if(0x80200000 <= a && a < 0x80800000)
-    return MTYPE_8020;
-  if(0xa0000000 <= a && a < 0xa0200000)
-    return MTYPE_A000;
-  return MTYPE_8000;
-}
-
-static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override)
-{
-  void *jaddr = NULL;
-  int type=0;
-  int mr=rs1[i];
-  if(((smrv_strong|smrv_weak)>>mr)&1) {
-    type=get_ptr_mem_type(smrv[mr]);
-    //printf("set %08x @%08x r%d %d\n", smrv[mr], start+i*4, mr, type);
-  }
-  else {
-    // use the mirror we are running on
-    type=get_ptr_mem_type(start);
-    //printf("set nospec   @%08x r%d %d\n", start+i*4, mr, type);
-  }
-
-  if(type==MTYPE_8020) { // RAM 80200000+ mirror
-    emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
-    addr=*addr_reg_override=HOST_TEMPREG;
-    type=0;
-  }
-  else if(type==MTYPE_0000) { // RAM 0 mirror
-    emit_orimm(addr,0x80000000,HOST_TEMPREG);
-    addr=*addr_reg_override=HOST_TEMPREG;
-    type=0;
-  }
-  else if(type==MTYPE_A000) { // RAM A mirror
-    emit_andimm(addr,~0x20000000,HOST_TEMPREG);
-    addr=*addr_reg_override=HOST_TEMPREG;
-    type=0;
-  }
-  else if(type==MTYPE_1F80) { // scratchpad
-    if (psxH == (void *)0x1f800000) {
-      emit_addimm(addr,-0x1f800000,HOST_TEMPREG);
-      emit_cmpimm(HOST_TEMPREG,0x1000);
-      jaddr=out;
-      emit_jc(0);
-    }
-    else {
-      // do usual RAM check, jump will go to the right handler
-      type=0;
-    }
-  }
-
-  if(type==0)
-  {
-    emit_cmpimm(addr,RAM_SIZE);
-    jaddr=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);
-    if(ram_offset!=0) {
-      emit_addimm(addr,ram_offset,HOST_TEMPREG);
-      addr=*addr_reg_override=HOST_TEMPREG;
-    }
-  }
-
-  return jaddr;
-}
-
 #define shift_assemble shift_assemble_arm
 
 static void loadlr_assemble_arm(int i,struct regstat *i_regs)
@@ -2867,243 +2349,6 @@ static void loadlr_assemble_arm(int i,struct regstat *i_regs)
 }
 #define loadlr_assemble loadlr_assemble_arm
 
-static void cop0_assemble(int i,struct regstat *i_regs)
-{
-  if(opcode2[i]==0) // MFC0
-  {
-    signed char t=get_reg(i_regs->regmap,rt1[i]);
-    u_int copr=(source[i]>>11)&0x1f;
-    //assert(t>=0); // Why does this happen?  OOT is weird
-    if(t>=0&&rt1[i]!=0) {
-      emit_readword(&reg_cop0[copr],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);
-    wb_register(rs1[i],i_regs->regmap,i_regs->dirty);
-    if(copr==9||copr==11||copr==12||copr==13) {
-      emit_readword(&last_count,HOST_TEMPREG);
-      emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
-      emit_add(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
-      emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
-      emit_writeword(HOST_CCREG,&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||copr==13) {
-      if (is_delayslot) {
-        // burn cycles to cause cc_interrupt, which will
-        // reschedule next_interupt. Relies on CCREG from above.
-        assem_debug("MTC0 DS %d\n", copr);
-        emit_writeword(HOST_CCREG,&last_count);
-        emit_movimm(0,HOST_CCREG);
-        emit_storereg(CCREG,HOST_CCREG);
-        emit_loadreg(rs1[i],1);
-        emit_movimm(copr,0);
-        emit_call(pcsx_mtc0_ds);
-        emit_loadreg(rs1[i],s);
-        return;
-      }
-      emit_movimm(start+i*4+4,HOST_TEMPREG);
-      emit_writeword(HOST_TEMPREG,&pcaddr);
-      emit_movimm(0,HOST_TEMPREG);
-      emit_writeword(HOST_TEMPREG,&pending_exception);
-    }
-    //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
-    //else
-    if(s==HOST_CCREG)
-      emit_loadreg(rs1[i],1);
-    else if(s!=1)
-      emit_mov(s,1);
-    emit_movimm(copr,0);
-    emit_call(pcsx_mtc0);
-    if(copr==9||copr==11||copr==12||copr==13) {
-      emit_readword(&Count,HOST_CCREG);
-      emit_readword(&next_interupt,HOST_TEMPREG);
-      emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
-      emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
-      emit_writeword(HOST_TEMPREG,&last_count);
-      emit_storereg(CCREG,HOST_CCREG);
-    }
-    if(copr==12||copr==13) {
-      assert(!is_delayslot);
-      emit_readword(&pending_exception,14);
-      emit_test(14,14);
-      emit_jne(&do_interrupt);
-    }
-    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));
-  }
-  else
-  {
-    assert(opcode2[i]==0x10);
-    //if((source[i]&0x3f)==0x10) // RFE
-    {
-      emit_readword(&Status,0);
-      emit_andimm(0,0x3c,1);
-      emit_andimm(0,~0xf,0);
-      emit_orrshr_imm(1,2,0);
-      emit_writeword(0,&Status);
-    }
-  }
-}
-
-static void cop2_get_dreg(u_int copr,signed char tl,signed char temp)
-{
-  switch (copr) {
-    case 1:
-    case 3:
-    case 5:
-    case 8:
-    case 9:
-    case 10:
-    case 11:
-      emit_readword(&reg_cop2d[copr],tl);
-      emit_signextend16(tl,tl);
-      emit_writeword(tl,&reg_cop2d[copr]); // hmh
-      break;
-    case 7:
-    case 16:
-    case 17:
-    case 18:
-    case 19:
-      emit_readword(&reg_cop2d[copr],tl);
-      emit_andimm(tl,0xffff,tl);
-      emit_writeword(tl,&reg_cop2d[copr]);
-      break;
-    case 15:
-      emit_readword(&reg_cop2d[14],tl); // SXY2
-      emit_writeword(tl,&reg_cop2d[copr]);
-      break;
-    case 28:
-    case 29:
-      emit_readword(&reg_cop2d[9],temp);
-      emit_testimm(temp,0x8000); // do we need this?
-      emit_andimm(temp,0xf80,temp);
-      emit_andne_imm(temp,0,temp);
-      emit_shrimm(temp,7,tl);
-      emit_readword(&reg_cop2d[10],temp);
-      emit_testimm(temp,0x8000);
-      emit_andimm(temp,0xf80,temp);
-      emit_andne_imm(temp,0,temp);
-      emit_orrshr_imm(temp,2,tl);
-      emit_readword(&reg_cop2d[11],temp);
-      emit_testimm(temp,0x8000);
-      emit_andimm(temp,0xf80,temp);
-      emit_andne_imm(temp,0,temp);
-      emit_orrshl_imm(temp,3,tl);
-      emit_writeword(tl,&reg_cop2d[copr]);
-      break;
-    default:
-      emit_readword(&reg_cop2d[copr],tl);
-      break;
-  }
-}
-
-static void cop2_put_dreg(u_int copr,signed char sl,signed char temp)
-{
-  switch (copr) {
-    case 15:
-      emit_readword(&reg_cop2d[13],temp);  // SXY1
-      emit_writeword(sl,&reg_cop2d[copr]);
-      emit_writeword(temp,&reg_cop2d[12]); // SXY0
-      emit_readword(&reg_cop2d[14],temp);  // SXY2
-      emit_writeword(sl,&reg_cop2d[14]);
-      emit_writeword(temp,&reg_cop2d[13]); // SXY1
-      break;
-    case 28:
-      emit_andimm(sl,0x001f,temp);
-      emit_shlimm(temp,7,temp);
-      emit_writeword(temp,&reg_cop2d[9]);
-      emit_andimm(sl,0x03e0,temp);
-      emit_shlimm(temp,2,temp);
-      emit_writeword(temp,&reg_cop2d[10]);
-      emit_andimm(sl,0x7c00,temp);
-      emit_shrimm(temp,3,temp);
-      emit_writeword(temp,&reg_cop2d[11]);
-      emit_writeword(sl,&reg_cop2d[28]);
-      break;
-    case 30:
-      emit_movs(sl,temp);
-      emit_mvnmi(temp,temp);
-#ifdef HAVE_ARMV5
-      emit_clz(temp,temp);
-#else
-      emit_movs(temp,HOST_TEMPREG);
-      emit_movimm(0,temp);
-      emit_jeq((int)out+4*4);
-      emit_addpl_imm(temp,1,temp);
-      emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG);
-      emit_jns((int)out-2*4);
-#endif
-      emit_writeword(sl,&reg_cop2d[30]);
-      emit_writeword(temp,&reg_cop2d[31]);
-      break;
-    case 31:
-      break;
-    default:
-      emit_writeword(sl,&reg_cop2d[copr]);
-      break;
-  }
-}
-
-static void cop2_assemble(int i,struct regstat *i_regs)
-{
-  u_int copr=(source[i]>>11)&0x1f;
-  signed char temp=get_reg(i_regs->regmap,-1);
-  if (opcode2[i]==0) { // MFC2
-    signed char tl=get_reg(i_regs->regmap,rt1[i]);
-    if(tl>=0&&rt1[i]!=0)
-      cop2_get_dreg(copr,tl,temp);
-  }
-  else if (opcode2[i]==4) { // MTC2
-    signed char sl=get_reg(i_regs->regmap,rs1[i]);
-    cop2_put_dreg(copr,sl,temp);
-  }
-  else if (opcode2[i]==2) // CFC2
-  {
-    signed char tl=get_reg(i_regs->regmap,rt1[i]);
-    if(tl>=0&&rt1[i]!=0)
-      emit_readword(&reg_cop2c[copr],tl);
-  }
-  else if (opcode2[i]==6) // CTC2
-  {
-    signed char sl=get_reg(i_regs->regmap,rs1[i]);
-    switch(copr) {
-      case 4:
-      case 12:
-      case 20:
-      case 26:
-      case 27:
-      case 29:
-      case 30:
-        emit_signextend16(sl,temp);
-        break;
-      case 31:
-        //value = value & 0x7ffff000;
-        //if (value & 0x7f87e000) value |= 0x80000000;
-        emit_shrimm(sl,12,temp);
-        emit_shlimm(temp,12,temp);
-        emit_testimm(temp,0x7f000000);
-        emit_testeqimm(temp,0x00870000);
-        emit_testeqimm(temp,0x0000e000);
-        emit_orrne_imm(temp,0x80000000,temp);
-        break;
-      default:
-        temp=sl;
-        break;
-    }
-    emit_writeword(temp,&reg_cop2c[copr]);
-    assert(sl>=0);
-  }
-}
-
 static void c2op_prologue(u_int op,u_int reglist)
 {
   save_regs_all(reglist);
@@ -3261,22 +2506,6 @@ static void c2op_assemble(int i,struct regstat *i_regs)
   }
 }
 
-static void cop1_unusable(int i,struct regstat *i_regs)
-{
-  // XXX: should just just do the exception instead
-  //if(!cop1_usable)
-  {
-    void *jaddr=out;
-    emit_jmp(0);
-    add_stub_r(FP_STUB,jaddr,out,i,0,i_regs,is_delayslot,0);
-  }
-}
-
-static void cop1_assemble(int i,struct regstat *i_regs)
-{
-  cop1_unusable(i, i_regs);
-}
-
 static void multdiv_assemble_arm(int i,struct regstat *i_regs)
 {
   //  case 0x18: MULT
@@ -3448,29 +2677,6 @@ static void do_miniht_insert(u_int return_address,int rt,int temp) {
   #endif
 }
 
-static void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t u)
-{
-  //if(dirty_pre==dirty) return;
-  int hr,reg;
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(hr!=EXCLUDE_REG) {
-      reg=pre[hr];
-      if(((~u)>>(reg&63))&1) {
-        if(reg>0) {
-          if(((dirty_pre&~dirty)>>hr)&1) {
-            if(reg>0&&reg<34) {
-              emit_storereg(reg,hr);
-            }
-            else if(reg>=64) {
-              assert(0);
-            }
-          }
-        }
-      }
-    }
-  }
-}
-
 static void mark_clear_cache(void *target)
 {
   u_long offset = (u_char *)target - translation_cache;
diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index 78a342d2..987892f0 100644
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -295,6 +295,7 @@ static void add_stub(enum stub_type type, void *addr, void *retaddr,
 static void add_stub_r(enum stub_type type, void *addr, void *retaddr,
   int i, int addr_reg, struct regstat *i_regs, int ccadj, u_int reglist);
 static void add_to_linker(void *addr, u_int target, int ext);
+static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override);
 
 static void mprotect_w_x(void *start, void *end, int is_x)
 {
@@ -783,6 +784,47 @@ void alloc_all(struct regstat *cur,int i)
   }
 }
 
+#ifdef DRC_DBG
+extern void gen_interupt();
+extern void do_insn_cmp();
+#define FUNCNAME(f) { (intptr_t)f, " " #f }
+static const struct {
+  intptr_t addr;
+  const char *name;
+} function_names[] = {
+  FUNCNAME(cc_interrupt),
+  FUNCNAME(gen_interupt),
+  FUNCNAME(get_addr_ht),
+  FUNCNAME(get_addr),
+  FUNCNAME(jump_handler_read8),
+  FUNCNAME(jump_handler_read16),
+  FUNCNAME(jump_handler_read32),
+  FUNCNAME(jump_handler_write8),
+  FUNCNAME(jump_handler_write16),
+  FUNCNAME(jump_handler_write32),
+  FUNCNAME(invalidate_addr),
+  FUNCNAME(verify_code_vm),
+  FUNCNAME(verify_code),
+  FUNCNAME(jump_hlecall),
+  FUNCNAME(jump_syscall_hle),
+  FUNCNAME(new_dyna_leave),
+  FUNCNAME(pcsx_mtc0),
+  FUNCNAME(pcsx_mtc0_ds),
+  FUNCNAME(do_insn_cmp),
+};
+
+static const char *func_name(intptr_t a)
+{
+  int i;
+  for (i = 0; i < sizeof(function_names)/sizeof(function_names[0]); i++)
+    if (function_names[i].addr == a)
+      return function_names[i].name;
+  return "";
+}
+#else
+#define func_name(x) ""
+#endif
+
 #ifdef __i386__
 #include "assem_x86.c"
 #endif
@@ -1149,6 +1191,276 @@ void clean_blocks(u_int page)
   }
 }
 
+/* Register allocation */
+
+// Note: registers are allocated clean (unmodified state)
+// if you intend to modify the register, you must call dirty_reg().
+static 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];
+  assert(r < 64);
+  if((cur->u>>r)&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) {
+      assert(r < 64);
+      if((cur->u>>r)&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)) 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;
+          }
+        }
+      }
+    }
+  }
+  SysPrintf("This shouldn't happen (alloc_reg)");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
+static 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) {
+      assert(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;
+        }
+      }
+    }
+  }
+
+  // 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)) 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;
+          }
+        }
+      }
+    }
+  }
+  SysPrintf("This shouldn't happen");exit(1);
+}
+
 static void mov_alloc(struct regstat *current,int i)
 {
   // Note: Don't need to actually alloc the source registers
@@ -1651,6 +1963,29 @@ static void wb_register(signed char r,signed char regmap[],uint64_t dirty)
   }
 }
 
+static void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t u)
+{
+  //if(dirty_pre==dirty) return;
+  int hr,reg;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      reg=pre[hr];
+      if(((~u)>>(reg&63))&1) {
+        if(reg>0) {
+          if(((dirty_pre&~dirty)>>hr)&1) {
+            if(reg>0&&reg<34) {
+              emit_storereg(reg,hr);
+            }
+            else if(reg>=64) {
+              assert(0);
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
 void rlist()
 {
   int i;
@@ -2039,7 +2374,96 @@ void shift_assemble(int i,struct regstat *i_regs)
 }
 #endif
 
-void load_assemble(int i,struct regstat *i_regs)
+enum {
+  MTYPE_8000 = 0,
+  MTYPE_8020,
+  MTYPE_0000,
+  MTYPE_A000,
+  MTYPE_1F80,
+};
+
+static int get_ptr_mem_type(u_int a)
+{
+  if(a < 0x00200000) {
+    if(a<0x1000&&((start>>20)==0xbfc||(start>>24)==0xa0))
+      // return wrong, must use memhandler for BIOS self-test to pass
+      // 007 does similar stuff from a00 mirror, weird stuff
+      return MTYPE_8000;
+    return MTYPE_0000;
+  }
+  if(0x1f800000 <= a && a < 0x1f801000)
+    return MTYPE_1F80;
+  if(0x80200000 <= a && a < 0x80800000)
+    return MTYPE_8020;
+  if(0xa0000000 <= a && a < 0xa0200000)
+    return MTYPE_A000;
+  return MTYPE_8000;
+}
+
+static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override)
+{
+  void *jaddr = NULL;
+  int type=0;
+  int mr=rs1[i];
+  if(((smrv_strong|smrv_weak)>>mr)&1) {
+    type=get_ptr_mem_type(smrv[mr]);
+    //printf("set %08x @%08x r%d %d\n", smrv[mr], start+i*4, mr, type);
+  }
+  else {
+    // use the mirror we are running on
+    type=get_ptr_mem_type(start);
+    //printf("set nospec   @%08x r%d %d\n", start+i*4, mr, type);
+  }
+
+  if(type==MTYPE_8020) { // RAM 80200000+ mirror
+    emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
+    addr=*addr_reg_override=HOST_TEMPREG;
+    type=0;
+  }
+  else if(type==MTYPE_0000) { // RAM 0 mirror
+    emit_orimm(addr,0x80000000,HOST_TEMPREG);
+    addr=*addr_reg_override=HOST_TEMPREG;
+    type=0;
+  }
+  else if(type==MTYPE_A000) { // RAM A mirror
+    emit_andimm(addr,~0x20000000,HOST_TEMPREG);
+    addr=*addr_reg_override=HOST_TEMPREG;
+    type=0;
+  }
+  else if(type==MTYPE_1F80) { // scratchpad
+    if (psxH == (void *)0x1f800000) {
+      emit_addimm(addr,-0x1f800000,HOST_TEMPREG);
+      emit_cmpimm(HOST_TEMPREG,0x1000);
+      jaddr=out;
+      emit_jc(0);
+    }
+    else {
+      // do the usual RAM check, jump will go to the right handler
+      type=0;
+    }
+  }
+
+  if(type==0)
+  {
+    emit_cmpimm(addr,RAM_SIZE);
+    jaddr=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);
+    if(ram_offset!=0) {
+      emit_addimm(addr,ram_offset,HOST_TEMPREG);
+      addr=*addr_reg_override=HOST_TEMPREG;
+    }
+  }
+
+  return jaddr;
+}
+
+static void load_assemble(int i,struct regstat *i_regs)
 {
   int s,th,tl,addr;
   int offset;
@@ -2480,12 +2904,236 @@ void storelr_assemble(int i,struct regstat *i_regs)
   }
 }
 
-void c1ls_assemble(int i,struct regstat *i_regs)
+static void cop0_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]==0) // MFC0
+  {
+    signed char t=get_reg(i_regs->regmap,rt1[i]);
+    u_int copr=(source[i]>>11)&0x1f;
+    //assert(t>=0); // Why does this happen?  OOT is weird
+    if(t>=0&&rt1[i]!=0) {
+      emit_readword(&reg_cop0[copr],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);
+    wb_register(rs1[i],i_regs->regmap,i_regs->dirty);
+    if(copr==9||copr==11||copr==12||copr==13) {
+      emit_readword(&last_count,HOST_TEMPREG);
+      emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+      emit_add(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
+      emit_writeword(HOST_CCREG,&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||copr==13) {
+      if (is_delayslot) {
+        // burn cycles to cause cc_interrupt, which will
+        // reschedule next_interupt. Relies on CCREG from above.
+        assem_debug("MTC0 DS %d\n", copr);
+        emit_writeword(HOST_CCREG,&last_count);
+        emit_movimm(0,HOST_CCREG);
+        emit_storereg(CCREG,HOST_CCREG);
+        emit_loadreg(rs1[i],1);
+        emit_movimm(copr,0);
+        emit_call(pcsx_mtc0_ds);
+        emit_loadreg(rs1[i],s);
+        return;
+      }
+      emit_movimm(start+i*4+4,HOST_TEMPREG);
+      emit_writeword(HOST_TEMPREG,&pcaddr);
+      emit_movimm(0,HOST_TEMPREG);
+      emit_writeword(HOST_TEMPREG,&pending_exception);
+    }
+    //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12);
+    //else
+    if(s==HOST_CCREG)
+      emit_loadreg(rs1[i],1);
+    else if(s!=1)
+      emit_mov(s,1);
+    emit_movimm(copr,0);
+    emit_call(pcsx_mtc0);
+    if(copr==9||copr==11||copr==12||copr==13) {
+      emit_readword(&Count,HOST_CCREG);
+      emit_readword(&next_interupt,HOST_TEMPREG);
+      emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
+      emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
+      emit_writeword(HOST_TEMPREG,&last_count);
+      emit_storereg(CCREG,HOST_CCREG);
+    }
+    if(copr==12||copr==13) {
+      assert(!is_delayslot);
+      emit_readword(&pending_exception,14);
+      emit_test(14,14);
+      emit_jne(&do_interrupt);
+    }
+    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));
+  }
+  else
+  {
+    assert(opcode2[i]==0x10);
+    //if((source[i]&0x3f)==0x10) // RFE
+    {
+      emit_readword(&Status,0);
+      emit_andimm(0,0x3c,1);
+      emit_andimm(0,~0xf,0);
+      emit_orrshr_imm(1,2,0);
+      emit_writeword(0,&Status);
+    }
+  }
+}
+
+static void cop1_unusable(int i,struct regstat *i_regs)
+{
+  // XXX: should just just do the exception instead
+  //if(!cop1_usable)
+  {
+    void *jaddr=out;
+    emit_jmp(0);
+    add_stub_r(FP_STUB,jaddr,out,i,0,i_regs,is_delayslot,0);
+  }
+}
+
+static void cop1_assemble(int i,struct regstat *i_regs)
+{
+  cop1_unusable(i, i_regs);
+}
+
+static void c1ls_assemble(int i,struct regstat *i_regs)
 {
   cop1_unusable(i, i_regs);
 }
 
-void c2ls_assemble(int i,struct regstat *i_regs)
+// FP_STUB
+static void do_cop1stub(int n)
+{
+  literal_pool(256);
+  assem_debug("do_cop1stub %x\n",start+stubs[n].a*4);
+  set_jump_target(stubs[n].addr, out);
+  int i=stubs[n].a;
+//  int rs=stubs[n].b;
+  struct regstat *i_regs=(struct regstat *)stubs[n].c;
+  int ds=stubs[n].d;
+  if(!ds) {
+    load_all_consts(regs[i].regmap_entry,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->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_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
+  emit_jmp(ds?fp_exception_ds:fp_exception);
+}
+
+static void cop2_get_dreg(u_int copr,signed char tl,signed char temp)
+{
+  switch (copr) {
+    case 1:
+    case 3:
+    case 5:
+    case 8:
+    case 9:
+    case 10:
+    case 11:
+      emit_readword(&reg_cop2d[copr],tl);
+      emit_signextend16(tl,tl);
+      emit_writeword(tl,&reg_cop2d[copr]); // hmh
+      break;
+    case 7:
+    case 16:
+    case 17:
+    case 18:
+    case 19:
+      emit_readword(&reg_cop2d[copr],tl);
+      emit_andimm(tl,0xffff,tl);
+      emit_writeword(tl,&reg_cop2d[copr]);
+      break;
+    case 15:
+      emit_readword(&reg_cop2d[14],tl); // SXY2
+      emit_writeword(tl,&reg_cop2d[copr]);
+      break;
+    case 28:
+    case 29:
+      emit_readword(&reg_cop2d[9],temp);
+      emit_testimm(temp,0x8000); // do we need this?
+      emit_andimm(temp,0xf80,temp);
+      emit_andne_imm(temp,0,temp);
+      emit_shrimm(temp,7,tl);
+      emit_readword(&reg_cop2d[10],temp);
+      emit_testimm(temp,0x8000);
+      emit_andimm(temp,0xf80,temp);
+      emit_andne_imm(temp,0,temp);
+      emit_orrshr_imm(temp,2,tl);
+      emit_readword(&reg_cop2d[11],temp);
+      emit_testimm(temp,0x8000);
+      emit_andimm(temp,0xf80,temp);
+      emit_andne_imm(temp,0,temp);
+      emit_orrshl_imm(temp,3,tl);
+      emit_writeword(tl,&reg_cop2d[copr]);
+      break;
+    default:
+      emit_readword(&reg_cop2d[copr],tl);
+      break;
+  }
+}
+
+static void cop2_put_dreg(u_int copr,signed char sl,signed char temp)
+{
+  switch (copr) {
+    case 15:
+      emit_readword(&reg_cop2d[13],temp);  // SXY1
+      emit_writeword(sl,&reg_cop2d[copr]);
+      emit_writeword(temp,&reg_cop2d[12]); // SXY0
+      emit_readword(&reg_cop2d[14],temp);  // SXY2
+      emit_writeword(sl,&reg_cop2d[14]);
+      emit_writeword(temp,&reg_cop2d[13]); // SXY1
+      break;
+    case 28:
+      emit_andimm(sl,0x001f,temp);
+      emit_shlimm(temp,7,temp);
+      emit_writeword(temp,&reg_cop2d[9]);
+      emit_andimm(sl,0x03e0,temp);
+      emit_shlimm(temp,2,temp);
+      emit_writeword(temp,&reg_cop2d[10]);
+      emit_andimm(sl,0x7c00,temp);
+      emit_shrimm(temp,3,temp);
+      emit_writeword(temp,&reg_cop2d[11]);
+      emit_writeword(sl,&reg_cop2d[28]);
+      break;
+    case 30:
+      emit_movs(sl,temp);
+      emit_mvnmi(temp,temp);
+#ifdef HAVE_ARMV5
+      emit_clz(temp,temp);
+#else
+      emit_movs(temp,HOST_TEMPREG);
+      emit_movimm(0,temp);
+      emit_jeq((int)out+4*4);
+      emit_addpl_imm(temp,1,temp);
+      emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG);
+      emit_jns((int)out-2*4);
+#endif
+      emit_writeword(sl,&reg_cop2d[30]);
+      emit_writeword(temp,&reg_cop2d[31]);
+      break;
+    case 31:
+      break;
+    default:
+      emit_writeword(sl,&reg_cop2d[copr]);
+      break;
+  }
+}
+
+static void c2ls_assemble(int i,struct regstat *i_regs)
 {
   int s,tl;
   int ar;
@@ -2579,6 +3227,57 @@ void c2ls_assemble(int i,struct regstat *i_regs)
   }
 }
 
+static void cop2_assemble(int i,struct regstat *i_regs)
+{
+  u_int copr=(source[i]>>11)&0x1f;
+  signed char temp=get_reg(i_regs->regmap,-1);
+  if (opcode2[i]==0) { // MFC2
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0&&rt1[i]!=0)
+      cop2_get_dreg(copr,tl,temp);
+  }
+  else if (opcode2[i]==4) { // MTC2
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    cop2_put_dreg(copr,sl,temp);
+  }
+  else if (opcode2[i]==2) // CFC2
+  {
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0&&rt1[i]!=0)
+      emit_readword(&reg_cop2c[copr],tl);
+  }
+  else if (opcode2[i]==6) // CTC2
+  {
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    switch(copr) {
+      case 4:
+      case 12:
+      case 20:
+      case 26:
+      case 27:
+      case 29:
+      case 30:
+        emit_signextend16(sl,temp);
+        break;
+      case 31:
+        //value = value & 0x7ffff000;
+        //if (value & 0x7f87e000) value |= 0x80000000;
+        emit_shrimm(sl,12,temp);
+        emit_shlimm(temp,12,temp);
+        emit_testimm(temp,0x7f000000);
+        emit_testeqimm(temp,0x00870000);
+        emit_testeqimm(temp,0x0000e000);
+        emit_orrne_imm(temp,0x80000000,temp);
+        break;
+      default:
+        temp=sl;
+        break;
+    }
+    emit_writeword(temp,&reg_cop2c[copr]);
+    assert(sl>=0);
+  }
+}
+
 #ifndef multdiv_assemble
 void multdiv_assemble(int i,struct regstat *i_regs)
 {
@@ -2648,6 +3347,98 @@ void intcall_assemble(int i,struct regstat *i_regs)
   emit_jmp(jump_intcall);
 }
 
+static void speculate_mov(int rs,int rt)
+{
+  if(rt!=0) {
+    smrv_strong_next|=1<<rt;
+    smrv[rt]=smrv[rs];
+  }
+}
+
+static void speculate_mov_weak(int rs,int rt)
+{
+  if(rt!=0) {
+    smrv_weak_next|=1<<rt;
+    smrv[rt]=smrv[rs];
+  }
+}
+
+static void speculate_register_values(int i)
+{
+  if(i==0) {
+    memcpy(smrv,psxRegs.GPR.r,sizeof(smrv));
+    // gp,sp are likely to stay the same throughout the block
+    smrv_strong_next=(1<<28)|(1<<29)|(1<<30);
+    smrv_weak_next=~smrv_strong_next;
+    //printf(" llr %08x\n", smrv[4]);
+  }
+  smrv_strong=smrv_strong_next;
+  smrv_weak=smrv_weak_next;
+  switch(itype[i]) {
+    case ALU:
+      if     ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]);
+      else if((smrv_strong>>rs2[i])&1) speculate_mov(rs2[i],rt1[i]);
+      else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]);
+      else if((smrv_weak>>rs2[i])&1) speculate_mov_weak(rs2[i],rt1[i]);
+      else {
+        smrv_strong_next&=~(1<<rt1[i]);
+        smrv_weak_next&=~(1<<rt1[i]);
+      }
+      break;
+    case SHIFTIMM:
+      smrv_strong_next&=~(1<<rt1[i]);
+      smrv_weak_next&=~(1<<rt1[i]);
+      // fallthrough
+    case IMM16:
+      if(rt1[i]&&is_const(&regs[i],rt1[i])) {
+        int value,hr=get_reg(regs[i].regmap,rt1[i]);
+        if(hr>=0) {
+          if(get_final_value(hr,i,&value))
+               smrv[rt1[i]]=value;
+          else smrv[rt1[i]]=constmap[i][hr];
+          smrv_strong_next|=1<<rt1[i];
+        }
+      }
+      else {
+        if     ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]);
+        else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]);
+      }
+      break;
+    case LOAD:
+      if(start<0x2000&&(rt1[i]==26||(smrv[rt1[i]]>>24)==0xa0)) {
+        // special case for BIOS
+        smrv[rt1[i]]=0xa0000000;
+        smrv_strong_next|=1<<rt1[i];
+        break;
+      }
+      // fallthrough
+    case SHIFT:
+    case LOADLR:
+    case MOV:
+      smrv_strong_next&=~(1<<rt1[i]);
+      smrv_weak_next&=~(1<<rt1[i]);
+      break;
+    case COP0:
+    case COP2:
+      if(opcode2[i]==0||opcode2[i]==2) { // MFC/CFC
+        smrv_strong_next&=~(1<<rt1[i]);
+        smrv_weak_next&=~(1<<rt1[i]);
+      }
+      break;
+    case C2LS:
+      if (opcode[i]==0x32) { // LWC2
+        smrv_strong_next&=~(1<<rt1[i]);
+        smrv_weak_next&=~(1<<rt1[i]);
+      }
+      break;
+  }
+#if 0
+  int r=4;
+  printf("x %08x %08x %d %d c %08x %08x\n",smrv[r],start+i*4,
+    ((smrv_strong>>r)&1),(smrv_weak>>r)&1,regs[i].isconst,regs[i].wasconst);
+#endif
+}
+
 void ds_assemble(int i,struct regstat *i_regs)
 {
   speculate_register_values(i);