X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fnew_dynarec.c;h=cbc289e7c7f9b0fc912f42b7dee7742f8de82bda;hp=c1fa775d27a8c60730906b04652ebe00b316f34c;hb=d3f3bf09b5b3f1d8b025cc9dbd902eb157aae0b7;hpb=311301dc26918fcb95577247257380960f2313ba

diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index c1fa775d..cbc289e7 100644
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -80,6 +80,10 @@ struct ll_entry
   u_char dep1[MAXBLOCK];
   u_char dep2[MAXBLOCK];
   u_char lt1[MAXBLOCK];
+  static uint64_t gte_rs[MAXBLOCK]; // gte: 32 data and 32 ctl regs
+  static uint64_t gte_rt[MAXBLOCK];
+  static uint64_t gte_unneeded[MAXBLOCK];
+  static int gte_reads_flags; // gte flag read encountered
   int imm[MAXBLOCK];
   u_int ba[MAXBLOCK];
   char likely[MAXBLOCK];
@@ -371,6 +375,7 @@ void *get_addr(u_int vaddr)
       if(verify_dirty(head->addr)) {
         //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]);
         invalid_code[vaddr>>12]=0;
+        inv_code_start=inv_code_end=~0;
         memory_map[vaddr>>12]|=0x40000000;
         if(vpage<2048) {
 #ifndef DISABLE_TLB
@@ -463,6 +468,7 @@ void *get_addr_32(u_int vaddr,u_int flags)
       if(verify_dirty(head->addr)) {
         //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]);
         invalid_code[vaddr>>12]=0;
+        inv_code_start=inv_code_end=~0;
         memory_map[vaddr>>12]|=0x40000000;
         if(vpage<2048) {
 #ifndef DISABLE_TLB
@@ -1127,39 +1133,10 @@ void invalidate_page(u_int page)
     head=next;
   }
 }
-void invalidate_block(u_int block)
+
+static void invalidate_block_range(u_int block, u_int first, u_int last)
 {
   u_int page=get_page(block<<12);
-  u_int vpage=get_vpage(block<<12);
-  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<0x80000000+RAM_SIZE) {
-        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;
-        }
-      }
-#ifndef DISABLE_TLB
-      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;
-        }
-      }
-#endif
-    }
-    head=head->next;
-  }
   //printf("first=%d last=%d\n",first,last);
   invalidate_page(page);
   assert(first+5>page); // NB: this assumes MAXBLOCK<=4096 (4 pages)
@@ -1178,9 +1155,6 @@ void invalidate_block(u_int block)
   
   // Don't trap writes
   invalid_code[block]=1;
-#ifdef PCSX
-  invalid_code[((u_int)0x80000000>>12)|page]=1;
-#endif
 #ifndef DISABLE_TLB
   // If there is a valid TLB entry for this page, remove write protect
   if(tlb_LUT_w[block]) {
@@ -1198,10 +1172,98 @@ void invalidate_block(u_int block)
   memset(mini_ht,-1,sizeof(mini_ht));
   #endif
 }
+
+void invalidate_block(u_int block)
+{
+  u_int page=get_page(block<<12);
+  u_int vpage=get_vpage(block<<12);
+  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<0x80000000+RAM_SIZE) {
+        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;
+        }
+      }
+#ifndef DISABLE_TLB
+      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;
+        }
+      }
+#endif
+    }
+    head=head->next;
+  }
+  invalidate_block_range(block,first,last);
+}
+
 void invalidate_addr(u_int addr)
 {
+#ifdef PCSX
+  //static int rhits;
+  // this check is done by the caller
+  //if (inv_code_start<=addr&&addr<=inv_code_end) { rhits++; return; }
+  u_int page=get_page(addr);
+  if(page<2048) { // RAM
+    struct ll_entry *head;
+    u_int addr_min=~0, addr_max=0;
+    int mask=RAM_SIZE-1;
+    int pg1;
+    inv_code_start=addr&~0xfff;
+    inv_code_end=addr|0xfff;
+    pg1=page;
+    if (pg1>0) {
+      // must check previous page too because of spans..
+      pg1--;
+      inv_code_start-=0x1000;
+    }
+    for(;pg1<=page;pg1++) {
+      for(head=jump_dirty[pg1];head!=NULL;head=head->next) {
+        u_int start,end;
+        get_bounds((int)head->addr,&start,&end);
+        if((start&mask)<=(addr&mask)&&(addr&mask)<(end&mask)) {
+          if(start<addr_min) addr_min=start;
+          if(end>addr_max) addr_max=end;
+        }
+        else if(addr<start) {
+          if(start<inv_code_end)
+            inv_code_end=start-1;
+        }
+        else {
+          if(end>inv_code_start)
+            inv_code_start=end;
+        }
+      }
+    }
+    if (addr_min!=~0) {
+      inv_debug("INV ADDR: %08x hit %08x-%08x\n", addr, addr_min, addr_max);
+      inv_code_start=inv_code_end=~0;
+      invalidate_block_range(addr>>12,(addr_min&mask)>>12,(addr_max&mask)>>12);
+      return;
+    }
+    else {
+      inv_debug("INV ADDR: %08x miss, inv %08x-%08x, sk %d\n", addr, inv_code_start, inv_code_end, 0);//rhits);
+    }
+    //rhits=0;
+    if(page!=0) // FIXME: don't know what's up with page 0 (Klonoa)
+      return;
+  }
+#endif
   invalidate_block(addr>>12);
 }
+
 // This is called when loading a save state.
 // Anything could have changed, so invalidate everything.
 void invalidate_all_pages()
@@ -1240,6 +1302,8 @@ void add_link(u_int vaddr,void *src)
 {
   u_int page=get_page(vaddr);
   inv_debug("add_link: %x -> %x (%d)\n",(int)src,vaddr,page);
+  int *ptr=(int *)(src+4);
+  assert((*ptr&0x0fff0000)==0x059f0000);
   ll_add(jump_out+page,vaddr,src);
   //int ptr=get_pointer(src);
   //inv_debug("add_link: Pointer is to %x\n",(int)ptr);
@@ -2766,6 +2830,7 @@ void load_assemble(int i,struct regstat *i_regs)
   int offset;
   int jaddr=0;
   int memtarget=0,c=0;
+  int fastload_reg_override=0;
   u_int hr,reglist=0;
   th=get_reg(i_regs->regmap,rt1[i]|64);
   tl=get_reg(i_regs->regmap,rt1[i]);
@@ -2820,6 +2885,7 @@ void load_assemble(int i,struct regstat *i_regs)
         if(sp_in_mirror&&rs1[i]==29) {
           emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
           emit_cmpimm(HOST_TEMPREG,RAM_SIZE);
+          fastload_reg_override=HOST_TEMPREG;
         }
         else
         #endif
@@ -2864,9 +2930,8 @@ void load_assemble(int i,struct regstat *i_regs)
 #else
           if(!c) a=addr;
 #endif
-#ifdef PCSX
-          if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+          if(fastload_reg_override) a=fastload_reg_override;
+
           emit_movsbl_indexed_tlb(x,a,map,tl);
         }
       }
@@ -2892,9 +2957,7 @@ void load_assemble(int i,struct regstat *i_regs)
 #else
           if(!c) a=addr;
 #endif
-#ifdef PCSX
-          if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+          if(fastload_reg_override) a=fastload_reg_override;
           //#ifdef
           //emit_movswl_indexed_tlb(x,tl,map,tl);
           //else
@@ -2920,9 +2983,7 @@ void load_assemble(int i,struct regstat *i_regs)
     if(!c||memtarget) {
       if(!dummy) {
         int a=addr;
-#ifdef PCSX
-        if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+        if(fastload_reg_override) a=fastload_reg_override;
         //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
         #ifdef HOST_IMM_ADDR32
         if(c)
@@ -2956,9 +3017,8 @@ void load_assemble(int i,struct regstat *i_regs)
 #else
           if(!c) a=addr;
 #endif
-#ifdef PCSX
-          if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+          if(fastload_reg_override) a=fastload_reg_override;
+
           emit_movzbl_indexed_tlb(x,a,map,tl);
         }
       }
@@ -2984,9 +3044,7 @@ void load_assemble(int i,struct regstat *i_regs)
 #else
           if(!c) a=addr;
 #endif
-#ifdef PCSX
-          if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+          if(fastload_reg_override) a=fastload_reg_override;
           //#ifdef
           //emit_movzwl_indexed_tlb(x,tl,map,tl);
           //#else
@@ -3013,9 +3071,7 @@ void load_assemble(int i,struct regstat *i_regs)
     if(!c||memtarget) {
       if(!dummy) {
         int a=addr;
-#ifdef PCSX
-        if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+        if(fastload_reg_override) a=fastload_reg_override;
         //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
         #ifdef HOST_IMM_ADDR32
         if(c)
@@ -3036,9 +3092,7 @@ void load_assemble(int i,struct regstat *i_regs)
     if(!c||memtarget) {
       if(!dummy) {
         int a=addr;
-#ifdef PCSX
-        if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+        if(fastload_reg_override) a=fastload_reg_override;
         //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);
@@ -3103,6 +3157,7 @@ void store_assemble(int i,struct regstat *i_regs)
   int jaddr=0,jaddr2,type;
   int memtarget=0,c=0;
   int agr=AGEN1+(i&1);
+  int faststore_reg_override=0;
   u_int hr,reglist=0;
   th=get_reg(i_regs->regmap,rs2[i]|64);
   tl=get_reg(i_regs->regmap,rs2[i]);
@@ -3131,6 +3186,7 @@ void store_assemble(int i,struct regstat *i_regs)
       if(sp_in_mirror&&rs1[i]==29) {
         emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
         emit_cmpimm(HOST_TEMPREG,RAM_SIZE);
+        faststore_reg_override=HOST_TEMPREG;
       }
       else
       #endif
@@ -3177,9 +3233,7 @@ void store_assemble(int i,struct regstat *i_regs)
 #else
       if(!c) a=addr;
 #endif
-#ifdef PCSX
-      if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+      if(faststore_reg_override) a=faststore_reg_override;
       //gen_tlb_addr_w(temp,map);
       //emit_writebyte_indexed(tl,(int)rdram-0x80000000,temp);
       emit_writebyte_indexed_tlb(tl,x,a,map,a);
@@ -3195,9 +3249,7 @@ void store_assemble(int i,struct regstat *i_regs)
 #else
       if(!c) a=addr;
 #endif
-#ifdef PCSX
-      if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+      if(faststore_reg_override) a=faststore_reg_override;
       //#ifdef
       //emit_writehword_indexed_tlb(tl,x,temp,map,temp);
       //#else
@@ -3212,9 +3264,7 @@ void store_assemble(int i,struct regstat *i_regs)
   if (opcode[i]==0x2B) { // SW
     if(!c||memtarget) {
       int a=addr;
-#ifdef PCSX
-      if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+      if(faststore_reg_override) a=faststore_reg_override;
       //emit_writeword_indexed(tl,(int)rdram-0x80000000,addr);
       emit_writeword_indexed_tlb(tl,0,a,map,temp);
     }
@@ -3223,9 +3273,7 @@ void store_assemble(int i,struct regstat *i_regs)
   if (opcode[i]==0x3F) { // SD
     if(!c||memtarget) {
       int a=addr;
-#ifdef PCSX
-      if(sp_in_mirror&&rs1[i]==29) a=HOST_TEMPREG;
-#endif
+      if(faststore_reg_override) a=faststore_reg_override;
       if(rs2[i]) {
         assert(th>=0);
         //emit_writeword_indexed(th,(int)rdram-0x80000000,addr);
@@ -3274,8 +3322,12 @@ void store_assemble(int i,struct regstat *i_regs)
   //if(opcode[i]==0x2B)
   /*if(opcode[i]==0x2B || opcode[i]==0x28 || opcode[i]==0x29 || opcode[i]==0x3F)
   {
-    //emit_pusha();
+    #ifdef __i386__
+    emit_pusha();
+    #endif
+    #ifdef __arm__
     save_regs(0x100f);
+    #endif
         emit_readword((int)&last_count,ECX);
         #ifdef __i386__
         if(get_reg(i_regs->regmap,CCREG)<0)
@@ -3294,8 +3346,12 @@ void store_assemble(int i,struct regstat *i_regs)
         emit_writeword(0,(int)&Count);
         #endif
     emit_call((int)memdebug);
-    //emit_popa();
+    #ifdef __i386__
+    emit_popa();
+    #endif
+    #ifdef __arm__
     restore_regs(0x100f);
+    #endif
   }/**/
 }
 
@@ -5144,9 +5200,45 @@ add_to_linker(int addr,int target,int ext)
   linkcount++;
 }
 
+static void ujump_assemble_write_ra(int i)
+{
+  int rt;
+  unsigned int return_address;
+  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)&&rt1[i+1]!=31) {
+      int temp=-1; // note: must be ds-safe
+      #ifdef HOST_TEMPREG
+      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
+    }
+  }
+}
+
 void ujump_assemble(int i,struct regstat *i_regs)
 {
   signed char *i_regmap=i_regs->regmap;
+  int ra_done=0;
   if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
   address_generation(i+1,i_regs,regs[i].regmap_entry);
   #ifdef REG_PREFETCH
@@ -5158,38 +5250,9 @@ void ujump_assemble(int i,struct regstat *i_regs)
     if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
   }
   #endif
-  if(rt1[i]==31) {
-    int rt;
-    unsigned int return_address;
-    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)&&rt1[i+1]!=31) {
-        int temp=-1; // note: must be ds-safe
-        #ifdef HOST_TEMPREG
-        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
-      }
-    }
+  if(rt1[i]==31&&(rt1[i]==rs1[i+1]||rt1[i]==rs2[i+1])) {
+    ujump_assemble_write_ra(i); // writeback ra for DS
+    ra_done=1;
   }
   ds_assemble(i+1,i_regs);
   uint64_t bc_unneeded=branch_regs[i].u;
@@ -5199,6 +5262,8 @@ void ujump_assemble(int i,struct regstat *i_regs)
   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(!ra_done&&rt1[i]==31)
+    ujump_assemble_write_ra(i);
   int cc,adj;
   cc=get_reg(branch_regs[i].regmap,CCREG);
   assert(cc==HOST_CCREG);
@@ -5222,11 +5287,33 @@ void ujump_assemble(int i,struct regstat *i_regs)
   }
 }
 
+static void rjump_assemble_write_ra(int i)
+{
+  int rt,return_address;
+  assert(rt1[i+1]!=rt1[i]);
+  assert(rt2[i+1]!=rt1[i]);
+  rt=get_reg(branch_regs[i].regmap,rt1[i]);
+  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
+}
+
 void rjump_assemble(int i,struct regstat *i_regs)
 {
   signed char *i_regmap=i_regs->regmap;
   int temp;
   int rs,cc,adj;
+  int ra_done=0;
   rs=get_reg(branch_regs[i].regmap,rs1[i]);
   assert(rs>=0);
   if(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]) {
@@ -5253,6 +5340,10 @@ void rjump_assemble(int i,struct regstat *i_regs)
     if(rh>=0) do_preload_rhash(rh);
   }
   #endif
+  if(rt1[i]!=0&&(rt1[i]==rs1[i+1]||rt1[i]==rs2[i+1])) {
+    rjump_assemble_write_ra(i);
+    ra_done=1;
+  }
   ds_assemble(i+1,i_regs);
   uint64_t bc_unneeded=branch_regs[i].u;
   uint64_t bc_unneeded_upper=branch_regs[i].uu;
@@ -5262,25 +5353,8 @@ void rjump_assemble(int i,struct regstat *i_regs)
   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]!=0) {
-    int rt,return_address;
-    assert(rt1[i+1]!=rt1[i]);
-    assert(rt2[i+1]!=rt1[i]);
-    rt=get_reg(branch_regs[i].regmap,rt1[i]);
-    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
-  }
+  if(!ra_done&&rt1[i]!=0)
+    rjump_assemble_write_ra(i);
   cc=get_reg(branch_regs[i].regmap,CCREG);
   assert(cc==HOST_CCREG);
   #ifdef USE_MINI_HT
@@ -6640,8 +6714,8 @@ static void pagespan_ds()
 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 u,uu,gte_u,b,bu,gte_bu;
+  uint64_t temp_u,temp_uu,temp_gte_u;
   uint64_t tdep;
   if(iend==slen-1) {
     u=1;uu=1;
@@ -6650,6 +6724,8 @@ void unneeded_registers(int istart,int iend,int r)
     uu=unneeded_reg_upper[iend+1];
     u=1;uu=1;
   }
+  gte_u=temp_gte_u=0;
+
   for (i=iend;i>=istart;i--)
   {
     //printf("unneeded registers i=%d (%d,%d) r=%d\n",i,istart,iend,r);
@@ -6663,6 +6739,7 @@ void unneeded_registers(int istart,int iend,int r)
         // Branch out of this block, flush all regs
         u=1;
         uu=1;
+        gte_u=0;
         /* Hexagon hack 
         if(itype[i]==UJUMP&&rt1[i]==31)
         {
@@ -6694,17 +6771,21 @@ void unneeded_registers(int istart,int iend,int r)
         uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
         uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
         u|=1;uu|=1;
+        gte_u|=gte_rt[i+1];
+        gte_u&=~gte_rs[i+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];
+            gte_u&=gte_unneeded[i+2];
           }
           else
           {
             u=1;
             uu=1;
+            gte_u=0;
           }
         }
       }
@@ -6718,10 +6799,12 @@ void unneeded_registers(int istart,int iend,int r)
           {
             // Unconditional branch
             temp_u=1;temp_uu=1;
+            temp_gte_u=0;
           } else {
             // Conditional branch (not taken case)
             temp_u=unneeded_reg[i+2];
             temp_uu=unneeded_reg_upper[i+2];
+            temp_gte_u&=gte_unneeded[i+2];
           }
           // Merge in delay slot
           tdep=(~temp_uu>>rt1[i+1])&1;
@@ -6731,17 +6814,21 @@ void unneeded_registers(int istart,int iend,int r)
           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;
+          temp_gte_u|=gte_rt[i+1];
+          temp_gte_u&=~gte_rs[i+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];
+              temp_gte_u&=gte_unneeded[i+2];
             }
             else
             {
               temp_u=1;
               temp_uu=1;
+              temp_gte_u=0;
             }
           }
           tdep=(~temp_uu>>rt1[i])&1;
@@ -6751,8 +6838,11 @@ void unneeded_registers(int istart,int iend,int r)
           temp_uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
           temp_uu&=~((tdep<<dep1[i])|(tdep<<dep2[i]));
           temp_u|=1;temp_uu|=1;
+          temp_gte_u|=gte_rt[i];
+          temp_gte_u&=~gte_rs[i];
           unneeded_reg[i]=temp_u;
           unneeded_reg_upper[i]=temp_uu;
+          gte_unneeded[i]=temp_gte_u;
           // 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) {
@@ -6760,6 +6850,7 @@ void unneeded_registers(int istart,int iend,int r)
           }else{
             unneeded_reg[(ba[i]-start)>>2]=1;
             unneeded_reg_upper[(ba[i]-start)>>2]=1;
+            gte_unneeded[(ba[i]-start)>>2]=0;
           }
         } /*else*/ if(1) {
           if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
@@ -6767,6 +6858,7 @@ void unneeded_registers(int istart,int iend,int r)
             // Unconditional branch
             u=unneeded_reg[(ba[i]-start)>>2];
             uu=unneeded_reg_upper[(ba[i]-start)>>2];
+            gte_u=gte_unneeded[(ba[i]-start)>>2];
             branch_unneeded_reg[i]=u;
             branch_unneeded_reg_upper[i]=uu;
         //u=1;
@@ -6781,10 +6873,13 @@ void unneeded_registers(int istart,int iend,int r)
             uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
             uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
             u|=1;uu|=1;
+            gte_u|=gte_rt[i+1];
+            gte_u&=~gte_rs[i+1];
           } else {
             // Conditional branch
             b=unneeded_reg[(ba[i]-start)>>2];
             bu=unneeded_reg_upper[(ba[i]-start)>>2];
+            gte_bu=gte_unneeded[(ba[i]-start)>>2];
             branch_unneeded_reg[i]=b;
             branch_unneeded_reg_upper[i]=bu;
         //b=1;
@@ -6799,20 +6894,25 @@ void unneeded_registers(int istart,int iend,int r)
             bu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
             bu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
             b|=1;bu|=1;
+            gte_bu|=gte_rt[i+1];
+            gte_bu&=~gte_rs[i+1];
             // If branch is "likely" then we skip the
             // delay slot on the fall-thru path
             if(likely[i]) {
               u=b;
               uu=bu;
+              gte_u=gte_bu;
               if(i<slen-1) {
                 u&=unneeded_reg[i+2];
                 uu&=unneeded_reg_upper[i+2];
+                gte_u&=gte_unneeded[i+2];
         //u=1;
         //uu=1;
               }
             } else {
               u&=b;
               uu&=bu;
+              gte_u&=gte_bu;
         //u=1;
         //uu=1;
             }
@@ -6848,11 +6948,13 @@ void unneeded_registers(int istart,int iend,int r)
     u|=1LL<<rt2[i];
     uu|=1LL<<rt1[i];
     uu|=1LL<<rt2[i];
+    gte_u|=gte_rt[i];
     // Accessed registers are needed
     u&=~(1LL<<rs1[i]);
     u&=~(1LL<<rs2[i]);
     uu&=~(1LL<<us1[i]);
     uu&=~(1LL<<us2[i]);
+    gte_u&=~gte_rs[i];
     // Source-target dependencies
     uu&=~(tdep<<dep1[i]);
     uu&=~(tdep<<dep2[i]);
@@ -6861,6 +6963,7 @@ void unneeded_registers(int istart,int iend,int r)
     // Save it
     unneeded_reg[i]=u;
     unneeded_reg_upper[i]=uu;
+    gte_unneeded[i]=gte_u;
     /*
     printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
     printf("U:");
@@ -7781,6 +7884,8 @@ void new_dynarec_clear_full()
   pending_exception=0;
   literalcount=0;
   stop_after_jal=0;
+  inv_code_start=inv_code_end=~0;
+  gte_reads_flags=0;
   // TLB
 #ifndef DISABLE_TLB
   using_tlb=0;
@@ -7909,6 +8014,7 @@ int new_recompile_block(int addr)
     emit_movimm(start,0);
     emit_writeword(0,(int)&pcaddr);
     emit_jmp((int)new_dyna_leave);
+    literal_pool(0);
 #ifdef __arm__
     __clear_cache((void *)beginning,out);
 #endif
@@ -8243,7 +8349,9 @@ int new_recompile_block(int addr)
       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;
+#ifndef FORCE32
       case 0x27: strcpy(insn[i],"LWU"); type=LOAD; break;
+#endif
       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;
@@ -8270,11 +8378,14 @@ int new_recompile_block(int addr)
 #endif
 #ifdef PCSX
       case 0x12: strcpy(insn[i],"COP2"); type=NI;
-        // note: COP MIPS-1 encoding differs from MIPS32
         op2=(source[i]>>21)&0x1f;
-        if (source[i]&0x3f) {
+        //if (op2 & 0x10) {
+        if (source[i]&0x3f) { // use this hack to support old savestates with patched gte insns
           if (gte_handlers[source[i]&0x3f]!=NULL) {
-            snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
+            if (gte_regnames[source[i]&0x3f]!=NULL)
+              strcpy(insn[i],gte_regnames[source[i]&0x3f]);
+            else
+              snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
             type=C2OP;
           }
         }
@@ -8302,6 +8413,7 @@ int new_recompile_block(int addr)
     us2[i]=0;
     dep1[i]=0;
     dep2[i]=0;
+    gte_rs[i]=gte_rt[i]=0;
     switch(type) {
       case LOAD:
         rs1[i]=(source[i]>>21)&0x1f;
@@ -8465,7 +8577,6 @@ int new_recompile_block(int addr)
         if(op2==16) if((source[i]&0x3f)==0x18) rs2[i]=CCREG; // ERET
         break;
       case COP1:
-      case COP2:
         rs1[i]=0;
         rs2[i]=0;
         rt1[i]=0;
@@ -8475,6 +8586,28 @@ int new_recompile_block(int addr)
         if(op2==5) us1[i]=rs1[i]; // DMTC1
         rs2[i]=CSREG;
         break;
+      case COP2:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        if(op2<3) rt1[i]=(source[i]>>16)&0x1F; // MFC2/CFC2
+        if(op2>3) rs1[i]=(source[i]>>16)&0x1F; // MTC2/CTC2
+        rs2[i]=CSREG;
+        int gr=(source[i]>>11)&0x1F;
+        switch(op2)
+        {
+          case 0x00: gte_rs[i]=1ll<<gr; break; // MFC2
+          case 0x04: gte_rt[i]=1ll<<gr; break; // MTC2
+          case 0x02: gte_rs[i]=1ll<<(gr+32); // CFC2
+            if(gr==31&&!gte_reads_flags) {
+              assem_debug("gte flag read encountered @%08x\n",addr + i*4);
+              gte_reads_flags=1;
+            }
+            break;
+          case 0x06: gte_rt[i]=1ll<<(gr+32); break; // CTC2
+        }
+        break;
       case C1LS:
         rs1[i]=(source[i]>>21)&0x1F;
         rs2[i]=CSREG;
@@ -8488,6 +8621,16 @@ int new_recompile_block(int addr)
         rt1[i]=0;
         rt2[i]=0;
         imm[i]=(short)source[i];
+        if(op==0x32) gte_rt[i]=1ll<<((source[i]>>16)&0x1F); // LWC2
+        else gte_rs[i]=1ll<<((source[i]>>16)&0x1F); // SWC2
+        break;
+      case C2OP:
+        rs1[i]=0;
+        rs2[i]=0;
+        rt1[i]=0;
+        rt2[i]=0;
+        gte_rt[i]=1ll<<63; // every op changes flags
+        // TODO: other regs?
         break;
       case FLOAT:
       case FCONV:
@@ -8887,8 +9030,6 @@ int new_recompile_block(int addr)
           clear_const(&current,rt1[i]);
           alloc_cc(&current,i);
           dirty_reg(&current,CCREG);
-          ooo[i]=1;
-          delayslot_alloc(&current,i+1);
           if (rt1[i]==31) {
             alloc_reg(&current,i,31);
             dirty_reg(&current,31);
@@ -8899,6 +9040,8 @@ int new_recompile_block(int addr)
             #endif
             //current.is32|=1LL<<rt1[i];
           }
+          ooo[i]=1;
+          delayslot_alloc(&current,i+1);
           //current.isconst=0; // DEBUG
           ds=1;
           //printf("i=%d, isconst=%x\n",i,current.isconst);
@@ -10978,10 +11121,14 @@ int new_recompile_block(int addr)
 #ifdef PCSX
   if (start == 0x80030000) {
     // nasty hack for fastbios thing
+    // override block entry to this code
     instr_addr0_override=(u_int)out;
     emit_movimm(start,0);
-    emit_readword((int)&pcaddr,1);
+    // abuse io address var as a flag that we
+    // have already returned here once
+    emit_readword((int)&address,1);
     emit_writeword(0,(int)&pcaddr);
+    emit_writeword(0,(int)&address);
     emit_cmp(0,1);
     emit_jne((int)new_dyna_leave);
   }
@@ -11307,6 +11454,7 @@ int new_recompile_block(int addr)
     }
 #endif
   }
+  inv_code_start=inv_code_end=~0;
 #ifdef PCSX
   // PCSX maps all RAM mirror invalid_code tests to 0x80000000..0x80000000+RAM_SIZE
   if(get_page(start)<(RAM_SIZE>>12))