drc: fall back to interpreter on delay slot link reg dependencies

[pcsx_rearmed.git] / libpcsxcore / new_dynarec / new_dynarec.c

diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index 63e21e1..641c673 100644 (file)
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -309,7 +309,14 @@ static void tlb_hacks()
 
 static u_int get_page(u_int vaddr)
 {
+#ifndef PCSX
   u_int page=(vaddr^0x80000000)>>12;
+#else
+  u_int page=vaddr&~0xe0000000;
+  if (page < 0x1000000)
+    page &= ~0x0e00000; // RAM mirrors
+  page>>=12;
+#endif
 #ifndef DISABLE_TLB
   if(page>262143&&tlb_LUT_r[vaddr>>12]) page=(tlb_LUT_r[vaddr>>12]^0x80000000)>>12;
 #endif
@@ -1783,16 +1790,17 @@ void cop1_alloc(struct regstat *current,int i)
   alloc_reg(current,i,CSREG); // Load status
   if(opcode2[i]<3) // MFC1/DMFC1/CFC1
   {
-    assert(rt1[i]);
-    clear_const(current,rt1[i]);
-    if(opcode2[i]==1) {
-      alloc_reg64(current,i,rt1[i]); // DMFC1
-      current->is32&=~(1LL<<rt1[i]);
-    }else{
-      alloc_reg(current,i,rt1[i]); // MFC1/CFC1
-      current->is32|=1LL<<rt1[i];
+    if(rt1[i]){
+      clear_const(current,rt1[i]);
+      if(opcode2[i]==1) {
+        alloc_reg64(current,i,rt1[i]); // DMFC1
+        current->is32&=~(1LL<<rt1[i]);
+      }else{
+        alloc_reg(current,i,rt1[i]); // MFC1/CFC1
+        current->is32|=1LL<<rt1[i];
+      }
+      dirty_reg(current,rt1[i]);
     }
-    dirty_reg(current,rt1[i]);
     alloc_reg_temp(current,i,-1);
   }
   else if(opcode2[i]>3) // MTC1/DMTC1/CTC1
@@ -5643,14 +5651,17 @@ void sjump_assemble(int i,struct regstat *i_regs)
   //if(opcode2[i]>=0x10) return; // FIXME (BxxZAL)
   //assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL)
 
-  if(ooo)
+  if(ooo) {
     if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))
-  {
-    // Write-after-read dependency prevents out of order execution
-    // First test branch condition, then execute delay slot, then branch
-    ooo=0;
+    {
+      // Write-after-read dependency prevents out of order execution
+      // First test branch condition, then execute delay slot, then branch
+      ooo=0;
+    }
+    if(rt1[i]==31&&(rs1[i+1]==31||rs2[i+1]==31||rt1[i+1]==31||rt2[i+1]==31))
+      // BxxZAL $ra is available to delay insn, so do it in order
+      ooo=0;
   }
-  assert(opcode2[i]<0x10||ooo); // FIXME (BxxZALL)
 
   if(ooo) {
     s1l=get_reg(branch_regs[i].regmap,rs1[i]);
@@ -5692,8 +5703,6 @@ void sjump_assemble(int i,struct regstat *i_regs)
     load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
     if(rt1[i]==31) {
       int rt,return_address;
-      assert(rt1[i+1]!=31);
-      assert(rt2[i+1]!=31);
       rt=get_reg(branch_regs[i].regmap,31);
       assem_debug("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
       if(rt>=0) {
@@ -5837,18 +5846,30 @@ void sjump_assemble(int i,struct regstat *i_regs)
     // In-order execution (branch first)
     //printf("IOE\n");
     int nottaken=0;
+    if(rt1[i]==31) {
+      int rt,return_address;
+      rt=get_reg(branch_regs[i].regmap,31);
+      if(rt>=0) {
+        // Save the PC even if the branch is not taken
+        return_address=start+i*4+8;
+        emit_movimm(return_address,rt); // PC into link register
+        #ifdef IMM_PREFETCH
+        emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+        #endif
+      }
+    }
     if(!unconditional) {
       //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
       if(!only32)
       {
         assert(s1h>=0);
-        if((opcode2[i]&0x1d)==0) // BLTZ/BLTZL
+        if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
         {
           emit_test(s1h,s1h);
           nottaken=(int)out;
           emit_jns(1);
         }
-        if((opcode2[i]&0x1d)==1) // BGEZ/BGEZL
+        if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
         {
           emit_test(s1h,s1h);
           nottaken=(int)out;
@@ -5858,13 +5879,13 @@ void sjump_assemble(int i,struct regstat *i_regs)
       else
       {
         assert(s1l>=0);
-        if((opcode2[i]&0x1d)==0) // BLTZ/BLTZL
+        if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
         {
           emit_test(s1l,s1l);
           nottaken=(int)out;
           emit_jns(1);
         }
-        if((opcode2[i]&0x1d)==1) // BGEZ/BGEZL
+        if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
         {
           emit_test(s1l,s1l);
           nottaken=(int)out;
@@ -8084,11 +8105,11 @@ int new_recompile_block(int addr)
           break;
         }
         break;
+#ifndef FORCE32
       case 0x14: strcpy(insn[i],"BEQL"); type=CJUMP; break;
       case 0x15: strcpy(insn[i],"BNEL"); type=CJUMP; break;
       case 0x16: strcpy(insn[i],"BLEZL"); type=CJUMP; break;
       case 0x17: strcpy(insn[i],"BGTZL"); type=CJUMP; break;
-#ifndef FORCE32
       case 0x18: strcpy(insn[i],"DADDI"); type=IMM16; break;
       case 0x19: strcpy(insn[i],"DADDIU"); type=IMM16; break;
       case 0x1A: strcpy(insn[i],"LDL"); type=LOADLR; break;
@@ -8128,19 +8149,20 @@ 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;
-        switch(op2)
+        if (source[i]&0x3f) {
+          if (gte_handlers[source[i]&0x3f]!=NULL) {
+            snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
+            type=C2OP;
+          }
+        }
+        else switch(op2)
         {
           case 0x00: strcpy(insn[i],"MFC2"); type=COP2; break;
           case 0x02: strcpy(insn[i],"CFC2"); type=COP2; break;
           case 0x04: strcpy(insn[i],"MTC2"); type=COP2; break;
           case 0x06: strcpy(insn[i],"CTC2"); type=COP2; break;
-          default:
-            if (gte_handlers[source[i]&0x3f]!=NULL) {
-              snprintf(insn[i], sizeof(insn[i]), "COP2 %x", source[i]&0x3f);
-              type=C2OP;
-            }
-            break;
         }
         break;
       case 0x32: strcpy(insn[i],"LWC2"); type=C2LS; break;
@@ -8397,6 +8419,17 @@ int new_recompile_block(int addr)
     else ba[i]=-1;
     /* Is this the end of the block? */
     if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) {
+#ifdef PCSX
+      // check for link register access in delay slot
+      int rt1_=rt1[i-1];
+      if(rt1_!=0&&(rs1[i]==rt1_||rs2[i]==rt1_||rt1[i]==rt1_||rt2[i]==rt1_)) {
+        printf("link access in delay slot @%08x (%08x)\n", addr + i*4, addr);
+        ba[i-1]=-1;
+        itype[i-1]=INTCALL;
+        done=2;
+      }
+      else
+#endif
       if(rt1[i-1]==0) { // Continue past subroutine call (JAL)
         done=2;
       }
@@ -8721,7 +8754,7 @@ int new_recompile_block(int addr)
           if (rt1[i]==31) {
             alloc_reg(&current,i,31);
             dirty_reg(&current,31);
-            //assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+            assert(rs1[i+1]!=31&&rs2[i+1]!=31);
             assert(rt1[i+1]!=rt1[i]);
             #ifdef REG_PREFETCH
             alloc_reg(&current,i,PTEMP);
@@ -8746,7 +8779,7 @@ int new_recompile_block(int addr)
             if (rt1[i]!=0) {
               alloc_reg(&current,i,rt1[i]);
               dirty_reg(&current,rt1[i]);
-              //assert(rs1[i+1]!=31&&rs2[i+1]!=31);
+              assert(rs1[i+1]!=rt1[i]&&rs2[i+1]!=rt1[i]);
               assert(rt1[i+1]!=rt1[i]);
               #ifdef REG_PREFETCH
               alloc_reg(&current,i,PTEMP);
@@ -8891,7 +8924,6 @@ int new_recompile_block(int addr)
             if (rt1[i]==31) { // BLTZAL/BGEZAL
               alloc_reg(&current,i,31);
               dirty_reg(&current,31);
-              assert(rs1[i+1]!=31&&rs2[i+1]!=31);
               //#ifdef REG_PREFETCH
               //alloc_reg(&current,i,PTEMP);
               //#endif