X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm.c;h=21640f84ad0d2313a934c86e1e544eabea1d9209;hp=da32f5b782f5bb06124024269e028062c7af0fe4;hb=HEAD;hpb=9de8a0c3587effdbd19584cbca3baf566d1d21bd

diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c
index da32f5b7..70798eff 100644
--- a/libpcsxcore/new_dynarec/assem_arm.c
+++ b/libpcsxcore/new_dynarec/assem_arm.c
@@ -27,8 +27,6 @@
 #include "pcnt.h"
 #include "arm_features.h"
 
-#define unused __attribute__((unused))
-
 #ifdef DRC_DBG
 #pragma GCC diagnostic ignored "-Wunused-function"
 #pragma GCC diagnostic ignored "-Wunused-variable"
@@ -194,6 +192,7 @@ static void *find_extjump_insn(void *stub)
 // get address that insn one after stub loads (dyna_linker arg1),
 // treat it as a pointer to branch insn,
 // return addr where that branch jumps to
+#if 0
 static void *get_pointer(void *stub)
 {
   //printf("get_pointer(%x)\n",(int)stub);
@@ -201,107 +200,7 @@ static void *get_pointer(void *stub)
   assert((*i_ptr&0x0f000000)==0x0a000000); // b
   return (u_char *)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
-static void *get_clean_addr(void *addr)
-{
-  signed int *ptr = addr;
-  #ifndef HAVE_ARMV7
-  ptr+=4;
-  #else
-  ptr+=6;
-  #endif
-  if((*ptr&0xFF000000)!=0xeb000000) ptr++;
-  assert((*ptr&0xFF000000)==0xeb000000); // bl instruction
-  ptr++;
-  if((*ptr&0xFF000000)==0xea000000) {
-    return (char *)ptr+((*ptr<<8)>>6)+8; // follow jump
-  }
-  return ptr;
-}
-
-static int verify_dirty(const u_int *ptr)
-{
-  #ifndef HAVE_ARMV7
-  u_int offset;
-  // get from literal pool
-  assert((*ptr&0xFFFF0000)==0xe59f0000);
-  offset=*ptr&0xfff;
-  u_int source=*(u_int*)((void *)ptr+offset+8);
-  ptr++;
-  assert((*ptr&0xFFFF0000)==0xe59f0000);
-  offset=*ptr&0xfff;
-  u_int copy=*(u_int*)((void *)ptr+offset+8);
-  ptr++;
-  assert((*ptr&0xFFFF0000)==0xe59f0000);
-  offset=*ptr&0xfff;
-  u_int len=*(u_int*)((void *)ptr+offset+8);
-  ptr++;
-  ptr++;
-  #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
-  //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
-static int isclean(void *addr)
-{
-  #ifndef HAVE_ARMV7
-  u_int *ptr=((u_int *)addr)+4;
-  #else
-  u_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_ds) return 0;
-  return 1;
-}
-
-// get source that block at addr was compiled from (host pointers)
-static void get_bounds(void *addr, u_char **start, u_char **end)
-{
-  u_int *ptr = addr;
-  #ifndef HAVE_ARMV7
-  u_int offset;
-  // get from literal pool
-  assert((*ptr&0xFFFF0000)==0xe59f0000);
-  offset=*ptr&0xfff;
-  u_int source=*(u_int*)((void *)ptr+offset+8);
-  ptr++;
-  //assert((*ptr&0xFFFF0000)==0xe59f0000);
-  //offset=*ptr&0xfff;
-  //u_int copy=*(u_int*)((void *)ptr+offset+8);
-  ptr++;
-  assert((*ptr&0xFFFF0000)==0xe59f0000);
-  offset=*ptr&0xfff;
-  u_int len=*(u_int*)((void *)ptr+offset+8);
-  ptr++;
-  ptr++;
-  #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
-  *start=(u_char *)source;
-  *end=(u_char *)source+len;
-}
+#endif
 
 // Allocate a specific ARM register.
 static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
@@ -318,16 +217,26 @@ static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
     }
   }
 
-  cur->regmap[hr]=reg;
-  cur->dirty&=~(1<<hr);
-  cur->dirty|=dirty<<hr;
-  cur->isconst&=~(1<<hr);
+  assert(n == hr || cur->regmap[hr] < 0 || !((cur->noevict >> hr) & 1));
+  cur->regmap[hr] = reg;
+  cur->dirty &= ~(1 << hr);
+  cur->dirty |= dirty << hr;
+  cur->isconst &= ~(1u << hr);
+  cur->noevict |= 1u << hr;
 }
 
 // Alloc cycle count into dedicated register
-static void alloc_cc(struct regstat *cur,int i)
+static void alloc_cc(struct regstat *cur, int i)
 {
-  alloc_arm_reg(cur,i,CCREG,HOST_CCREG);
+  alloc_arm_reg(cur, i, CCREG, HOST_CCREG);
+}
+
+static void alloc_cc_optional(struct regstat *cur, int i)
+{
+  if (cur->regmap[HOST_CCREG] < 0) {
+    alloc_arm_reg(cur, i, CCREG, HOST_CCREG);
+    cur->noevict &= ~(1u << HOST_CCREG);
+  }
 }
 
 /* Assembler */
@@ -452,12 +361,24 @@ static void emit_neg(int rs, int rt)
   output_w32(0xe2600000|rd_rn_rm(rt,rs,0));
 }
 
+static 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));
+}
+
 static 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));
 }
 
+static 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));
+}
+
 static void emit_zeroreg(int rt)
 {
   assem_debug("mov %s,#0\n",regname[rt]);
@@ -522,12 +443,8 @@ static void emit_pcreladdr(u_int rt)
 
 static void emit_loadreg(int r, int hr)
 {
-  if(r&64) {
-    SysPrintf("64bit load in 32bit mode!\n");
-    assert(0);
-    return;
-  }
-  if((r&63)==0)
+  assert(hr != EXCLUDE_REG);
+  if (r == 0)
     emit_zeroreg(hr);
   else {
     void *addr;
@@ -535,7 +452,6 @@ static void emit_loadreg(int r, int hr)
     //case HIREG: addr = &hi; break;
     //case LOREG: addr = &lo; break;
     case CCREG: addr = &cycle_count; break;
-    case CSREG: addr = &Status; break;
     case INVCP: addr = &invc_ptr; break;
     case ROREG: addr = &ram_offset; break;
     default:
@@ -552,11 +468,7 @@ static void emit_loadreg(int r, int hr)
 
 static void emit_storereg(int r, int hr)
 {
-  if(r&64) {
-    SysPrintf("64bit store in 32bit mode!\n");
-    assert(0);
-    return;
-  }
+  assert(hr != EXCLUDE_REG);
   int addr = (int)&psxRegs.GPR.r[r];
   switch (r) {
   //case HIREG: addr = &hi; break;
@@ -598,6 +510,12 @@ static void emit_not(int rs,int rt)
   output_w32(0xe1e00000|rd_rn_rm(rt,0,rs));
 }
 
+static void emit_mvneq(int rs,int rt)
+{
+  assem_debug("mvneq %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0x01e00000|rd_rn_rm(rt,0,rs));
+}
+
 static 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]);
@@ -681,29 +599,42 @@ static void emit_addimm(u_int rs,int imm,u_int rt)
   else if(rs!=rt) emit_mov(rs,rt);
 }
 
-static void emit_addimm_and_set_flags(int imm,int rt)
+static void emit_addimm_ptr(u_int rs, uintptr_t imm, u_int rt)
+{
+  emit_addimm(rs, imm, rt);
+}
+
+static void emit_addimm_and_set_flags3(u_int rs, int imm, u_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);
+  if (genimm(imm, &armval)) {
+    assem_debug("adds %s,%s,#%d\n",regname[rt],regname[rs],imm);
+    output_w32(0xe2900000|rd_rn_rm(rt,rs,0)|armval);
+  } else if (genimm(-imm, &armval)) {
+    assem_debug("subs %s,%s,#%d\n",regname[rt],regname[rs],imm);
+    output_w32(0xe2500000|rd_rn_rm(rt,rs,0)|armval);
+  } else if (rs != rt) {
+    emit_movimm(imm, rt);
+    emit_adds(rs, rt, rt);
+  } else if (imm < 0) {
+    assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],(-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(0xe2400000|rd_rn_imm_shift(rt,rs,(-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);
+  } else {
+    assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],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(0xe2800000|rd_rn_imm_shift(rt,rs,imm>>8,8));
     output_w32(0xe2900000|rd_rn_imm_shift(rt,rt,imm&0xff,0));
   }
 }
 
+static void emit_addimm_and_set_flags(int imm, u_int rt)
+{
+  emit_addimm_and_set_flags3(rt, imm, rt);
+}
+
 static void emit_addnop(u_int r)
 {
   assert(r<16);
@@ -960,7 +891,7 @@ static void emit_cmovs_imm(int imm,int rt)
   output_w32(0x43a00000|rd_rn_rm(rt,0,0)|armval);
 }
 
-static void emit_cmovne_reg(int rs,int rt)
+static unused 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));
@@ -1006,6 +937,12 @@ static void emit_cmp(int rs,int rt)
   output_w32(0xe1500000|rd_rn_rm(0,rs,rt));
 }
 
+static void emit_cmpcs(int rs,int rt)
+{
+  assem_debug("cmpcs %s,%s\n",regname[rs],regname[rt]);
+  output_w32(0x21500000|rd_rn_rm(0,rs,rt));
+}
+
 static void emit_set_gz32(int rs, int rt)
 {
   //assem_debug("set_gz32\n");
@@ -1110,6 +1047,14 @@ static void emit_jge(const void *a_)
   output_w32(0xaa000000|offset);
 }
 
+static void emit_jo(const void *a_)
+{
+  int a = (int)a_;
+  assem_debug("bvs %x\n",a);
+  u_int offset=genjmp(a);
+  output_w32(0x6a000000|offset);
+}
+
 static void emit_jno(const void *a_)
 {
   int a = (int)a_;
@@ -1134,6 +1079,15 @@ static void emit_jcc(const void *a_)
   output_w32(0x3a000000|offset);
 }
 
+static void *emit_cbz(int rs, const void *a)
+{
+  void *ret;
+  emit_test(rs, rs);
+  ret = out;
+  emit_jeq(a);
+  return ret;
+}
+
 static unused void emit_callreg(u_int r)
 {
   assert(r<15);
@@ -1307,7 +1261,7 @@ static void emit_readword(void *addr, int rt)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   assert(offset<4096);
-  assem_debug("ldr %s,fp+%d\n",regname[rt],offset);
+  assem_debug("ldr %s,fp+%#x%s\n", regname[rt], offset, fpofs_name(offset));
   output_w32(0xe5900000|rd_rn_rm(rt,FP,0)|offset);
 }
 #define emit_readptr emit_readword
@@ -1367,7 +1321,7 @@ static void emit_writeword(int rt, void *addr)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   assert(offset<4096);
-  assem_debug("str %s,fp+%d\n",regname[rt],offset);
+  assem_debug("str %s,fp+%#x%s\n", regname[rt], offset, fpofs_name(offset));
   output_w32(0xe5800000|rd_rn_rm(rt,FP,0)|offset);
 }
 
@@ -1493,13 +1447,10 @@ static void emit_cmov2imm_e_ne_compact(int imm1,int imm2,u_int rt)
 }
 
 // special case for checking invalid_code
-static void emit_cmpmem_indexedsr12_reg(int base,int r,int imm)
+static void emit_ldrb_indexedsr12_reg(int base, int r, int rt)
 {
-  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);
+  assem_debug("ldrb %s,%s,%s lsr #12\n",regname[rt],regname[base],regname[r]);
+  output_w32(0xe7d00000|rd_rn_rm(rt,base,r)|0x620);
 }
 
 static void emit_callne(int a)
@@ -1632,7 +1583,7 @@ static void literal_pool_jumpover(int n)
 }
 
 // parsed by get_pointer, find_extjump_insn
-static void emit_extjump2(u_char *addr, u_int target, void *linker)
+static void emit_extjump(u_char *addr, u_int target)
 {
   u_char *ptr=(u_char *)addr;
   assert((ptr[3]&0x0e)==0xa);
@@ -1642,18 +1593,7 @@ static void emit_extjump2(u_char *addr, u_int target, void *linker)
   emit_loadlp((u_int)addr,1);
   assert(ndrc->translation_cache <= addr &&
          addr < ndrc->translation_cache + sizeof(ndrc->translation_cache));
-  //assert((target>=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000));
-//DEBUG >
-#ifdef DEBUG_CYCLE_COUNT
-  emit_readword(&last_count,ECX);
-  emit_add(HOST_CCREG,ECX,HOST_CCREG);
-  emit_readword(&next_interupt,ECX);
-  emit_writeword(HOST_CCREG,&Count);
-  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
-  emit_writeword(ECX,&last_count);
-#endif
-//DEBUG <
-  emit_far_jump(linker);
+  emit_far_jump(dyna_linker);
 }
 
 static void check_extjump2(void *src)
@@ -1734,7 +1674,7 @@ static void do_readstub(int n)
   u_int reglist=stubs[n].e;
   const signed char *i_regmap=i_regs->regmap;
   int rt;
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||dops[i].itype==LOADLR) {
+  if(dops[i].itype==C2LS||dops[i].itype==LOADLR) {
     rt=get_reg(i_regmap,FTEMP);
   }else{
     rt=get_reg(i_regmap,dops[i].rt1);
@@ -1761,7 +1701,7 @@ static void do_readstub(int n)
   emit_shrimm(rs,12,temp2);
   emit_readword_dualindexedx4(temp,temp2,temp2);
   emit_lsls_imm(temp2,1,temp2);
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
+  if(dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
     switch(type) {
       case LOADB_STUB:  emit_ldrccsb_dualindexed(temp2,rs,rt); break;
       case LOADBU_STUB: emit_ldrccb_dualindexed(temp2,rs,rt); break;
@@ -1794,7 +1734,15 @@ static void do_readstub(int n)
     emit_loadreg(CCREG,2);
   emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
   emit_far_call(handler);
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
+#if 0
+  if (type == LOADW_STUB) {
+    // new cycle_count returned in r2
+    emit_addimm(2, -(int)stubs[n].d, cc<0?2:cc);
+    if (cc < 0)
+      emit_storereg(CCREG, 2);
+  }
+#endif
+  if(dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
     mov_loadtype_adj(type,0,rt);
   }
   if(restore_jump)
@@ -1806,28 +1754,27 @@ static void do_readstub(int n)
 static void inline_readstub(enum stub_type type, int i, u_int addr,
   const signed char regmap[], int target, int adj, u_int reglist)
 {
-  int rs=get_reg(regmap,target);
-  int rt=get_reg(regmap,target);
-  if(rs<0) rs=get_reg_temp(regmap);
-  assert(rs>=0);
+  int ra = cinfo[i].addr;
+  int rt = get_reg(regmap,target);
+  assert(ra >= 0);
   u_int is_dynamic;
   uintptr_t host_addr = 0;
   void *handler;
   int cc=get_reg(regmap,CCREG);
-  if(pcsx_direct_read(type,addr,adj,cc,target?rs:-1,rt))
+  if(pcsx_direct_read(type,addr,adj,cc,target?ra:-1,rt))
     return;
   handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
   if (handler == NULL) {
     if(rt<0||dops[i].rt1==0)
       return;
     if(addr!=host_addr)
-      emit_movimm_from(addr,rs,host_addr,rs);
+      emit_movimm_from(addr,ra,host_addr,ra);
     switch(type) {
-      case LOADB_STUB:  emit_movsbl_indexed(0,rs,rt); break;
-      case LOADBU_STUB: emit_movzbl_indexed(0,rs,rt); break;
-      case LOADH_STUB:  emit_movswl_indexed(0,rs,rt); break;
-      case LOADHU_STUB: emit_movzwl_indexed(0,rs,rt); break;
-      case LOADW_STUB:  emit_readword_indexed(0,rs,rt); break;
+      case LOADB_STUB:  emit_movsbl_indexed(0,ra,rt); break;
+      case LOADBU_STUB: emit_movzbl_indexed(0,ra,rt); break;
+      case LOADH_STUB:  emit_movswl_indexed(0,ra,rt); break;
+      case LOADHU_STUB: emit_movzwl_indexed(0,ra,rt); break;
+      case LOADW_STUB:  emit_readword_indexed(0,ra,rt); break;
       default:          assert(0);
     }
     return;
@@ -1848,8 +1795,8 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   save_regs(reglist);
   if(target==0)
     emit_movimm(addr,0);
-  else if(rs!=0)
-    emit_mov(rs,0);
+  else if(ra!=0)
+    emit_mov(ra,0);
   if(cc<0)
     emit_loadreg(CCREG,2);
   if(is_dynamic) {
@@ -1860,11 +1807,19 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
     emit_readword(&last_count,3);
     emit_addimm(cc<0?2:cc,adj,2);
     emit_add(2,3,2);
-    emit_writeword(2,&Count);
+    emit_writeword(2,&psxRegs.cycle);
   }
 
   emit_far_call(handler);
 
+#if 0
+  if (type == LOADW_STUB) {
+    // new cycle_count returned in r2
+    emit_addimm(2, -adj, cc<0?2:cc);
+    if (cc < 0)
+      emit_storereg(CCREG, 2);
+  }
+#endif
   if(rt>=0&&dops[i].rt1!=0) {
     switch(type) {
       case LOADB_STUB:  emit_signextend8(0,rt); break;
@@ -1890,7 +1845,7 @@ static void do_writestub(int n)
   u_int reglist=stubs[n].e;
   const signed char *i_regmap=i_regs->regmap;
   int rt,r;
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS) {
+  if(dops[i].itype==C2LS) {
     rt=get_reg(i_regmap,r=FTEMP);
   }else{
     rt=get_reg(i_regmap,r=dops[i].rs2);
@@ -1948,9 +1903,9 @@ static void do_writestub(int n)
   if(cc<0)
     emit_loadreg(CCREG,2);
   emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
-  // returns new cycle_count
   emit_far_call(handler);
-  emit_addimm(0,-(int)stubs[n].d,cc<0?2:cc);
+  // new cycle_count returned in r2
+  emit_addimm(2,-(int)stubs[n].d,cc<0?2:cc);
   if(cc<0)
     emit_storereg(CCREG,2);
   if(restore_jump)
@@ -1962,19 +1917,19 @@ static void do_writestub(int n)
 static void inline_writestub(enum stub_type type, int i, u_int addr,
   const signed char regmap[], int target, int adj, u_int reglist)
 {
-  int rs=get_reg_temp(regmap);
-  int rt=get_reg(regmap,target);
-  assert(rs>=0);
+  int ra = cinfo[i].addr;
+  int rt = get_reg(regmap, target);
+  assert(ra>=0);
   assert(rt>=0);
   uintptr_t host_addr = 0;
   void *handler = get_direct_memhandler(mem_wtab, addr, type, &host_addr);
   if (handler == NULL) {
     if(addr!=host_addr)
-      emit_movimm_from(addr,rs,host_addr,rs);
+      emit_movimm_from(addr,ra,host_addr,ra);
     switch(type) {
-      case STOREB_STUB: emit_writebyte_indexed(rt,0,rs); break;
-      case STOREH_STUB: emit_writehword_indexed(rt,0,rs); break;
-      case STOREW_STUB: emit_writeword_indexed(rt,0,rs); break;
+      case STOREB_STUB: emit_writebyte_indexed(rt,0,ra); break;
+      case STOREH_STUB: emit_writehword_indexed(rt,0,ra); break;
+      case STOREW_STUB: emit_writeword_indexed(rt,0,ra); break;
       default:          assert(0);
     }
     return;
@@ -1982,56 +1937,20 @@ static void inline_writestub(enum stub_type type, int i, u_int addr,
 
   // call a memhandler
   save_regs(reglist);
-  pass_args(rs,rt);
+  pass_args(ra,rt);
   int cc=get_reg(regmap,CCREG);
   if(cc<0)
     emit_loadreg(CCREG,2);
   emit_addimm(cc<0?2:cc,adj,2);
   emit_movimm((u_int)handler,3);
-  // returns new cycle_count
   emit_far_call(jump_handler_write_h);
-  emit_addimm(0,-adj,cc<0?2:cc);
+  // new cycle_count returned in r2
+  emit_addimm(2,-adj,cc<0?2:cc);
   if(cc<0)
     emit_storereg(CCREG,2);
   restore_regs(reglist);
 }
 
-// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
-static void do_dirty_stub_emit_args(u_int arg0, u_int source_len)
-{
-  #ifndef HAVE_ARMV7
-  emit_loadlp((int)source, 1);
-  emit_loadlp((int)copy, 2);
-  emit_loadlp(source_len, 3);
-  #else
-  emit_movw(((u_int)source)&0x0000FFFF, 1);
-  emit_movw(((u_int)copy)&0x0000FFFF, 2);
-  emit_movt(((u_int)source)&0xFFFF0000, 1);
-  emit_movt(((u_int)copy)&0xFFFF0000, 2);
-  emit_movw(source_len, 3);
-  #endif
-  emit_movimm(arg0, 0);
-}
-
-static void *do_dirty_stub(int i, u_int source_len)
-{
-  assem_debug("do_dirty_stub %x\n",start+i*4);
-  do_dirty_stub_emit_args(start + i*4, source_len);
-  emit_far_call(verify_code);
-  void *entry = out;
-  load_regs_entry(i);
-  if (entry == out)
-    entry = instr_addr[i];
-  emit_jmp(instr_addr[i]);
-  return entry;
-}
-
-static void do_dirty_stub_ds(u_int source_len)
-{
-  do_dirty_stub_emit_args(start + 1, source_len);
-  emit_far_call(verify_code_ds);
-}
-
 /* Special assem */
 
 static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist)
@@ -2234,15 +2153,11 @@ static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
   //  case 0x19: MULTU
   //  case 0x1A: DIV
   //  case 0x1B: DIVU
-  //  case 0x1C: DMULT
-  //  case 0x1D: DMULTU
-  //  case 0x1E: DDIV
-  //  case 0x1F: DDIVU
   if(dops[i].rs1&&dops[i].rs2)
   {
-    if((dops[i].opcode2&4)==0) // 32-bit
+    switch (dops[i].opcode2)
     {
-      if(dops[i].opcode2==0x18) // MULT
+    case 0x18: // MULT
       {
         signed char m1=get_reg(i_regs->regmap,dops[i].rs1);
         signed char m2=get_reg(i_regs->regmap,dops[i].rs2);
@@ -2254,7 +2169,8 @@ static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
         assert(lo>=0);
         emit_smull(m1,m2,hi,lo);
       }
-      if(dops[i].opcode2==0x19) // MULTU
+      break;
+    case 0x19: // MULTU
       {
         signed char m1=get_reg(i_regs->regmap,dops[i].rs1);
         signed char m2=get_reg(i_regs->regmap,dops[i].rs2);
@@ -2266,14 +2182,16 @@ static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
         assert(lo>=0);
         emit_umull(m1,m2,hi,lo);
       }
-      if(dops[i].opcode2==0x1A) // DIV
+      break;
+    case 0x1A: // DIV
       {
         signed char d1=get_reg(i_regs->regmap,dops[i].rs1);
         signed char d2=get_reg(i_regs->regmap,dops[i].rs2);
-        assert(d1>=0);
-        assert(d2>=0);
         signed char quotient=get_reg(i_regs->regmap,LOREG);
         signed char remainder=get_reg(i_regs->regmap,HIREG);
+        void *jaddr_div0;
+        assert(d1>=0);
+        assert(d2>=0);
         assert(quotient>=0);
         assert(remainder>=0);
         emit_movs(d1,remainder);
@@ -2281,11 +2199,12 @@ static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
         emit_negmi(quotient,quotient); // .. quotient and ..
         emit_negmi(remainder,remainder); // .. remainder for div0 case (will be negated back after jump)
         emit_movs(d2,HOST_TEMPREG);
-        emit_jeq(out+52); // Division by zero
+        jaddr_div0 = out;
+        emit_jeq(0); // Division by zero
         emit_negsmi(HOST_TEMPREG,HOST_TEMPREG);
 #ifdef HAVE_ARMV5
         emit_clz(HOST_TEMPREG,quotient);
-        emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG);
+        emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG);  // shifted divisor
 #else
         emit_movimm(0,quotient);
         emit_addpl_imm(quotient,1,quotient);
@@ -2301,23 +2220,27 @@ static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
         emit_jcc(out-16); // -4
         emit_teq(d1,d2);
         emit_negmi(quotient,quotient);
+        set_jump_target(jaddr_div0, out);
         emit_test(d1,d1);
         emit_negmi(remainder,remainder);
       }
-      if(dops[i].opcode2==0x1B) // DIVU
+      break;
+    case 0x1B: // DIVU
       {
         signed char d1=get_reg(i_regs->regmap,dops[i].rs1); // dividend
         signed char d2=get_reg(i_regs->regmap,dops[i].rs2); // 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);
+        void *jaddr_div0;
+        assert(d1>=0);
+        assert(d2>=0);
         assert(quotient>=0);
         assert(remainder>=0);
         emit_mov(d1,remainder);
         emit_movimm(0xffffffff,quotient); // div0 case
         emit_test(d2,d2);
-        emit_jeq(out+40); // Division by zero
+        jaddr_div0 = out;
+        emit_jeq(0); // Division by zero
 #ifdef HAVE_ARMV5
         emit_clz(d2,HOST_TEMPREG);
         emit_movimm(1<<31,quotient);
@@ -2335,20 +2258,54 @@ static void multdiv_assemble_arm(int i, const struct regstat *i_regs)
         emit_adcs(quotient,quotient,quotient);
         emit_shrcc_imm(d2,1,d2);
         emit_jcc(out-16); // -4
+        set_jump_target(jaddr_div0, out);
       }
+      break;
     }
-    else // 64-bit
-      assert(0);
   }
   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);
+    if ((dops[i].opcode2==0x1A || dops[i].opcode2==0x1B) && dops[i].rs2==0) // div 0
+    {
+      if (dops[i].rs1) {
+        signed char numerator = get_reg(i_regs->regmap, dops[i].rs1);
+        assert(numerator >= 0);
+        if (hr < 0)
+          hr = HOST_TEMPREG;
+        emit_movs(numerator, hr);
+        if (lr >= 0) {
+          if (dops[i].opcode2 == 0x1A) { // DIV
+            emit_movimm(0xffffffff, lr);
+            emit_negmi(lr, lr);
+          }
+          else
+            emit_movimm(~0, lr);
+        }
+      }
+      else {
+        if (hr >= 0) emit_zeroreg(hr);
+        if (lr >= 0) emit_movimm(~0,lr);
+      }
+    }
+    else if ((dops[i].opcode2==0x1A || dops[i].opcode2==0x1B) && dops[i].rs1==0)
+    {
+      signed char denominator = get_reg(i_regs->regmap, dops[i].rs2);
+      assert(denominator >= 0);
+      if (hr >= 0) emit_zeroreg(hr);
+      if (lr >= 0) {
+        emit_zeroreg(lr);
+        emit_test(denominator, denominator);
+        emit_mvneq(lr, lr);
+      }
+    }
+    else
+    {
+      // Multiply by zero is zero.
+      if (hr >= 0) emit_zeroreg(hr);
+      if (lr >= 0) emit_zeroreg(lr);
+    }
   }
 }
 #define multdiv_assemble multdiv_assemble_arm