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=64ea3bc1f5172d7bb1eb055ea686a3f299ddce02;hp=0128758237383cb34bd6f1ad8acf2de2aebd792c;hb=63cb0298408cf47e8e765b048f4d9af50c2e5eb7;hpb=4600ba0381ea86bdfdcba8f4f762c89f5c51af5b

diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index 01287582..64ea3bc1 100644
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -89,6 +89,11 @@ struct ll_entry
   static uint64_t gte_rt[MAXBLOCK];
   static uint64_t gte_unneeded[MAXBLOCK];
   static int gte_reads_flags; // gte flag read encountered
+  static u_int smrv[32]; // speculated MIPS register values
+  static u_int smrv_strong; // mask or regs that are likely to have correct values
+  static u_int smrv_weak; // same, but somewhat less likely
+  static u_int smrv_strong_next; // same, but after current insn executes
+  static u_int smrv_weak_next;
   int imm[MAXBLOCK];
   u_int ba[MAXBLOCK];
   char likely[MAXBLOCK];
@@ -135,7 +140,7 @@ struct ll_entry
 #else
   static const u_int using_tlb=0;
 #endif
-  static u_int sp_in_mirror;
+  int new_dynarec_did_compile;
   u_int stop_after_jal;
   extern u_char restore_candidate[512];
   extern int cycle_count;
@@ -375,7 +380,9 @@ void *get_addr(u_int vaddr)
         //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;
+#ifndef DISABLE_TLB
         memory_map[vaddr>>12]|=0x40000000;
+#endif
         if(vpage<2048) {
 #ifndef DISABLE_TLB
           if(tlb_LUT_r[vaddr>>12]) {
@@ -1333,11 +1340,13 @@ void clean_blocks(u_int page)
               inv|=invalid_code[i];
             }
           }
+#ifndef DISABLE_TLB
           if((signed int)head->vaddr>=(signed int)0xC0000000) {
             u_int addr = (head->vaddr+(memory_map[head->vaddr>>12]<<2));
             //printf("addr=%x start=%x end=%x\n",addr,start,end);
             if(addr<start||addr>=end) inv=1;
           }
+#endif
           else if((signed int)head->vaddr>=(signed int)0x80000000+RAM_SIZE) {
             inv=1;
           }
@@ -1390,8 +1399,6 @@ void mov_alloc(struct regstat *current,int i)
 
 void shiftimm_alloc(struct regstat *current,int i)
 {
-  clear_const(current,rs1[i]);
-  clear_const(current,rt1[i]);
   if(opcode2[i]<=0x3) // SLL/SRL/SRA
   {
     if(rt1[i]) {
@@ -1400,8 +1407,21 @@ void shiftimm_alloc(struct regstat *current,int i)
       alloc_reg(current,i,rt1[i]);
       current->is32|=1LL<<rt1[i];
       dirty_reg(current,rt1[i]);
+      if(is_const(current,rs1[i])) {
+        int v=get_const(current,rs1[i]);
+        if(opcode2[i]==0x00) set_const(current,rt1[i],v<<imm[i]);
+        if(opcode2[i]==0x02) set_const(current,rt1[i],(u_int)v>>imm[i]);
+        if(opcode2[i]==0x03) set_const(current,rt1[i],v>>imm[i]);
+      }
+      else clear_const(current,rt1[i]);
     }
   }
+  else
+  {
+    clear_const(current,rs1[i]);
+    clear_const(current,rt1[i]);
+  }
+
   if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA
   {
     if(rt1[i]) {
@@ -2687,7 +2707,7 @@ void shiftimm_assemble(int i,struct regstat *i_regs)
       t=get_reg(i_regs->regmap,rt1[i]);
       s=get_reg(i_regs->regmap,rs1[i]);
       //assert(t>=0);
-      if(t>=0){
+      if(t>=0&&!((i_regs->isconst>>t)&1)){
         if(rs1[i]==0)
         {
           emit_zeroreg(t);
@@ -2881,23 +2901,7 @@ void load_assemble(int i,struct regstat *i_regs)
       if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
       #endif
       {
-        #ifdef PCSX
-        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
-        emit_cmpimm(addr,RAM_SIZE);
-        jaddr=(int)out;
-        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
-        // Hint to branch predictor that the branch is unlikely to be taken
-        if(rs1[i]>=28)
-          emit_jno_unlikely(0);
-        else
-        #endif
-        emit_jno(0);
+        jaddr=emit_fastpath_cmp_jump(i,addr,&fastload_reg_override);
       }
     }
   }else{ // using tlb
@@ -3182,14 +3186,7 @@ void store_assemble(int i,struct regstat *i_regs)
   else addr=s;
   if(!using_tlb) {
     if(!c) {
-      #ifdef PCSX
-      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
+      #ifndef PCSX
       #ifdef R29_HACK
       // Strmnnrmn's speed hack
       if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
@@ -3212,6 +3209,9 @@ void store_assemble(int i,struct regstat *i_regs)
         #endif
         emit_jno(0);
       }
+      #else
+        jaddr=emit_fastpath_cmp_jump(i,addr,&faststore_reg_override);
+      #endif
     }
   }else{ // using tlb
     int x=0;
@@ -3803,6 +3803,7 @@ void c2ls_assemble(int i,struct regstat *i_regs)
   int memtarget=0,c=0;
   int jaddr2=0,jaddr3,type;
   int agr=AGEN1+(i&1);
+  int fastio_reg_override=0;
   u_int hr,reglist=0;
   u_int copr=(source[i]>>16)&0x1f;
   s=get_reg(i_regs->regmap,rs1[i]);
@@ -3844,22 +3845,24 @@ void c2ls_assemble(int i,struct regstat *i_regs)
   }
   else {
     if(!c) {
-      emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE);
-      jaddr2=(int)out;
-      emit_jno(0);
+      jaddr2=emit_fastpath_cmp_jump(i,ar,&fastio_reg_override);
     }
     if (opcode[i]==0x32) { // LWC2
       #ifdef HOST_IMM_ADDR32
       if(c) emit_readword_tlb(constmap[i][s]+offset,-1,tl);
       else
       #endif
-      emit_readword_indexed(0,ar,tl);
+      int a=ar;
+      if(fastio_reg_override) a=fastio_reg_override;
+      emit_readword_indexed(0,a,tl);
     }
     if (opcode[i]==0x3a) { // SWC2
       #ifdef DESTRUCTIVE_SHIFT
       if(!offset&&!c&&s>=0) emit_mov(s,ar);
       #endif
-      emit_writeword_indexed(tl,0,ar);
+      int a=ar;
+      if(fastio_reg_override) a=fastio_reg_override;
+      emit_writeword_indexed(tl,0,a);
     }
   }
   if(jaddr2)
@@ -3964,6 +3967,7 @@ void intcall_assemble(int i,struct regstat *i_regs)
 
 void ds_assemble(int i,struct regstat *i_regs)
 {
+  speculate_register_values(i);
   is_delayslot=1;
   switch(itype[i]) {
     case ALU:
@@ -4223,6 +4227,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
         //  printf("poor load scheduling!\n");
       }
       else if(c) {
+#ifndef DISABLE_TLB
         if(rm>=0) {
           if(!entry||entry[rm]!=mgr) {
             if(itype[i]==STORE||itype[i]==STORELR||(opcode[i]&0x3b)==0x39||(opcode[i]&0x3b)==0x3a) {
@@ -4237,6 +4242,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
             }
           }
         }
+#endif
         if(rs1[i]!=rt1[i]||itype[i]!=LOAD) {
           if(!entry||entry[ra]!=agr) {
             if (opcode[i]==0x22||opcode[i]==0x26) {
@@ -4265,7 +4271,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
   // Preload constants for next instruction
   if(itype[i+1]==LOAD||itype[i+1]==LOADLR||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS||itype[i+1]==C2LS) {
     int agr,ra;
-    #ifndef HOST_IMM_ADDR32
+    #if !defined(HOST_IMM_ADDR32) && !defined(DISABLE_TLB)
     // Mapper entry
     agr=MGEN1+((i+1)&1);
     ra=get_reg(i_regs->regmap,agr);
@@ -7901,14 +7907,13 @@ void new_dynarec_clear_full()
   // TLB
 #ifndef DISABLE_TLB
   using_tlb=0;
-#endif
-  sp_in_mirror=0;
   for(n=0;n<524288;n++) // 0 .. 0x7FFFFFFF
     memory_map[n]=-1;
   for(n=524288;n<526336;n++) // 0x80000000 .. 0x807FFFFF
     memory_map[n]=((u_int)rdram-0x80000000)>>2;
   for(n=526336;n<1048576;n++) // 0x80800000 .. 0xFFFFFFFF
     memory_map[n]=-1;
+#endif
   for(n=0;n<4096;n++) ll_clear(jump_in+n);
   for(n=0;n<4096;n++) ll_clear(jump_out+n);
   for(n=0;n<4096;n++) ll_clear(jump_dirty+n);
@@ -8011,12 +8016,8 @@ int new_recompile_block(int addr)
   //rlist();
   start = (u_int)addr&~3;
   //assert(((u_int)addr&1)==0);
+  new_dynarec_did_compile=1;
 #ifdef PCSX
-  if(!sp_in_mirror&&(signed int)(psxRegs.GPR.n.sp&0xffe00000)>0x80200000&&
-     0x10000<=psxRegs.GPR.n.sp&&(psxRegs.GPR.n.sp&~0xe0e00000)<RAM_SIZE) {
-    printf("SP hack enabled (%08x), @%08x\n", psxRegs.GPR.n.sp, psxRegs.pc);
-    sp_in_mirror=1;
-  }
   if (Config.HLE && start == 0x80001000) // hlecall
   {
     // XXX: is this enough? Maybe check hleSoftCall?
@@ -11155,6 +11156,7 @@ int new_recompile_block(int addr)
       if(bt[i]) assem_debug("OOPS - branch into delay slot\n");
       instr_addr[i]=0;
     } else {
+      speculate_register_values(i);
       #ifndef DESTRUCTIVE_WRITEBACK
       if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
       {