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=88c059aba139ab895a85e40508861db83802818e;hp=e4e6a9afb4030d0fe3df3e6618826d00649a60a8;hb=1fd1aceb4382827b39bda6fe6915e38a294534e6;hpb=f76eeef94001323d5019a134c54847ece2493ae9 diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c index e4e6a9af..88c059ab 100644 --- a/libpcsxcore/new_dynarec/new_dynarec.c +++ b/libpcsxcore/new_dynarec/new_dynarec.c @@ -680,8 +680,8 @@ void lsn(u_char hsn[], int i, int *preferred_reg) if(itype[i]==LOADLR) { hsn[FTEMP]=0; } - // Also 64-bit SDL/SDR - if(opcode[i]==0x2c||opcode[i]==0x2d) { + // Also SWL/SWR/SDL/SDR + if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { hsn[FTEMP]=0; } // Don't remove the TLB registers either @@ -1153,7 +1153,7 @@ void invalidate_block(u_int block) 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<0x80800000) { + 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)>12)&2047; if((((end-1-(u_int)rdram)>>12)&2047)>last) last=((end-1-(u_int)rdram)>>12)&2047; @@ -1266,7 +1266,7 @@ void clean_blocks(u_int page) u_int i; u_int inv=0; get_bounds((int)head->addr,&start,&end); - if(start-(u_int)rdram<0x800000) { + if(start-(u_int)rdram>12;i<=(end-1-(u_int)rdram+0x80000000)>>12;i++) { inv|=invalid_code[i]; } @@ -1276,7 +1276,7 @@ void clean_blocks(u_int page) //printf("addr=%x start=%x end=%x\n",addr,start,end); if(addr=end) inv=1; } - else if((signed int)head->vaddr>=(signed int)0x80800000) { + else if((signed int)head->vaddr>=(signed int)0x80000000+RAM_SIZE) { inv=1; } if(!inv) { @@ -1638,7 +1638,7 @@ void store_alloc(struct regstat *current,int i) // On CPUs without 32-bit immediates we need a pointer to invalid_code else alloc_reg(current,i,INVCP); #endif - if(opcode[i]==0x2c||opcode[i]==0x2d) { // 64-bit SDL/SDR + if(opcode[i]==0x2a||opcode[i]==0x2e||opcode[i]==0x2c||opcode[i]==0x2d) { // SWL/SWL/SDL/SDR alloc_reg(current,i,FTEMP); } // We need a temporary register for address generation @@ -2746,7 +2746,7 @@ void load_assemble(int i,struct regstat *i_regs) if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<=0) { c=(i_regs->wasconst>>s)&1; - memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000; + memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1; } //printf("load_assemble: c=%d\n",c); @@ -2774,10 +2774,10 @@ void load_assemble(int i,struct regstat *i_regs) //#define R29_HACK 1 #ifdef R29_HACK // Strmnnrmn's speed hack - if(rs1[i]!=29||start<0x80001000||start>=0x80800000) + if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE) #endif { - emit_cmpimm(addr,0x800000); + 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 @@ -3020,7 +3020,7 @@ void store_assemble(int i,struct regstat *i_regs) offset=imm[i]; if(s>=0) { c=(i_regs->wasconst>>s)&1; - memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000; + memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1; } assert(tl>=0); @@ -3036,14 +3036,14 @@ void store_assemble(int i,struct regstat *i_regs) #ifdef R29_HACK // Strmnnrmn's speed hack memtarget=1; - if(rs1[i]!=29||start<0x80001000||start>=0x80800000) + if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE) #endif - emit_cmpimm(addr,0x800000); + emit_cmpimm(addr,RAM_SIZE); #ifdef DESTRUCTIVE_SHIFT if(s==addr) emit_mov(s,temp); #endif #ifdef R29_HACK - if(rs1[i]!=29||start<0x80001000||start>=0x80800000) + if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE) #endif { jaddr=(int)out; @@ -3193,15 +3193,17 @@ void storelr_assemble(int i,struct regstat *i_regs) int case1,case2,case3; int done0,done1,done2; int memtarget,c=0; + int agr=AGEN1+(i&1); u_int hr,reglist=0; th=get_reg(i_regs->regmap,rs2[i]|64); tl=get_reg(i_regs->regmap,rs2[i]); s=get_reg(i_regs->regmap,rs1[i]); - temp=get_reg(i_regs->regmap,-1); + temp=get_reg(i_regs->regmap,agr); + if(temp<0) temp=get_reg(i_regs->regmap,-1); offset=imm[i]; if(s>=0) { c=(i_regs->isconst>>s)&1; - memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80800000; + memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; if(using_tlb&&((signed int)(constmap[i][s]+offset))>=(signed int)0xC0000000) memtarget=1; } assert(tl>=0); @@ -3212,7 +3214,7 @@ void storelr_assemble(int i,struct regstat *i_regs) assert(temp>=0); if(!using_tlb) { if(!c) { - emit_cmpimm(s<0||offset?temp:s,0x800000); + emit_cmpimm(s<0||offset?temp:s,RAM_SIZE); if(!offset&&s!=temp) emit_mov(s,temp); jaddr=(int)out; emit_jno(0); @@ -3379,7 +3381,7 @@ void storelr_assemble(int i,struct regstat *i_regs) set_jump_target(done0,(int)out); } if(!c||!memtarget) - add_stub(STORELR_STUB,jaddr,(int)out,0,(int)i_regs,rs2[i],ccadj[i],reglist); + add_stub(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist); } if(!using_tlb) { emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp); @@ -3464,7 +3466,7 @@ void c1ls_assemble(int i,struct regstat *i_regs) // Generate address + offset if(!using_tlb) { if(!c) - emit_cmpimm(offset||c||s<0?ar:s,0x800000); + emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE); } else { @@ -3495,7 +3497,7 @@ void c1ls_assemble(int i,struct regstat *i_regs) jaddr2=(int)out; emit_jno(0); } - else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80800000) { + else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80000000+RAM_SIZE) { jaddr2=(int)out; emit_jmp(0); // inline_readstub/inline_writestub? Very rare case } @@ -3596,7 +3598,7 @@ void c2ls_assemble(int i,struct regstat *i_regs) int s,tl; int ar; int offset; - int c=0; + int memtarget=0,c=0; int jaddr,jaddr2=0,jaddr3,type; int agr=AGEN1+(i&1); u_int hr,reglist=0; @@ -3622,36 +3624,41 @@ void c2ls_assemble(int i,struct regstat *i_regs) } else { // LWC2 ar=tl; } + if(s>=0) c=(i_regs->wasconst>>s)&1; + memtarget=c&&(((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE); if (!offset&&!c&&s>=0) ar=s; assert(ar>=0); if (opcode[i]==0x3a) { // SWC2 cop2_get_dreg(copr,tl,HOST_TEMPREG); + type=STOREW_STUB; } - if(s>=0) c=(i_regs->wasconst>>s)&1; - if(!c) { - emit_cmpimm(offset||c||s<0?ar:s,0x800000); - jaddr2=(int)out; - emit_jno(0); - } - else if(((signed int)(constmap[i][s]+offset))>=(signed int)0x80800000) { - jaddr2=(int)out; - emit_jmp(0); // inline_readstub/inline_writestub? Very rare case - } - 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); + else type=LOADW_STUB; + + if(c&&!memtarget) { + jaddr2=(int)out; + emit_jmp(0); // inline_readstub/inline_writestub? } - if (opcode[i]==0x3a) { // SWC2 -#ifdef DESTRUCTIVE_SHIFT - if(!offset&&!c&&s>=0) emit_mov(s,ar); -#endif - emit_writeword_indexed(tl,0,ar); - type=STOREW_STUB; + else { + if(!c) { + emit_cmpimm(offset||c||s<0?ar:s,RAM_SIZE); + jaddr2=(int)out; + emit_jno(0); + } + 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); + } + if (opcode[i]==0x3a) { // SWC2 + #ifdef DESTRUCTIVE_SHIFT + if(!offset&&!c&&s>=0) emit_mov(s,ar); + #endif + emit_writeword_indexed(tl,0,ar); + } } if(jaddr2) add_stub(type,jaddr2,(int)out,i,ar,(int)i_regs,ccadj[i],reglist); @@ -3962,7 +3969,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) if(itype[i]==C1LS||itype[i]==C2LS) { if ((opcode[i]&0x3b)==0x31||(opcode[i]&0x3b)==0x32) // LWC1/LDC1/LWC2/LDC2 ra=get_reg(i_regs->regmap,FTEMP); - else { // SWC1/SDC1 + else { // SWC1/SDC1/SWC2/SDC2 ra=get_reg(i_regs->regmap,agr); if(ra<0) ra=get_reg(i_regs->regmap,-1); } @@ -4002,7 +4009,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) // Stores to memory go thru the mapper to detect self-modifying // code, loads don't. if((unsigned int)(constmap[i][rs]+offset)>=0xC0000000 || - (unsigned int)(constmap[i][rs]+offset)<0x80800000 ) + (unsigned int)(constmap[i][rs]+offset)<0x80000000+RAM_SIZE ) generate_map_const(constmap[i][rs]+offset,rm); }else{ if((signed int)(constmap[i][rs]+offset)>=(signed int)0xC0000000) @@ -4052,7 +4059,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[]) // Stores to memory go thru the mapper to detect self-modifying // code, loads don't. if((unsigned int)(constmap[i+1][rs]+offset)>=0xC0000000 || - (unsigned int)(constmap[i+1][rs]+offset)<0x80800000 ) + (unsigned int)(constmap[i+1][rs]+offset)<0x80000000+RAM_SIZE ) generate_map_const(constmap[i+1][rs]+offset,ra); }else{ if((signed int)(constmap[i+1][rs]+offset)>=(signed int)0xC0000000) @@ -5619,7 +5626,7 @@ void sjump_assemble(int i,struct regstat *i_regs) #endif //if(opcode2[i]>=0x10) return; // FIXME (BxxZAL) - assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL) + //assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL) if(ooo) if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) @@ -5628,8 +5635,7 @@ void sjump_assemble(int i,struct regstat *i_regs) // First test branch condition, then execute delay slot, then branch ooo=0; } - // TODO: Conditional branches w/link must execute in-order so that - // condition test and write to r31 occur before cycle count test + assert(opcode2[i]<0x10||ooo); // FIXME (BxxZALL) if(ooo) { s1l=get_reg(branch_regs[i].regmap,rs1[i]); @@ -5724,7 +5730,7 @@ void sjump_assemble(int i,struct regstat *i_regs) if(!only32) { assert(s1h>=0); - if(opcode2[i]==0) // BLTZ + if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL { emit_test(s1h,s1h); if(invert){ @@ -5735,7 +5741,7 @@ void sjump_assemble(int i,struct regstat *i_regs) emit_js(0); } } - if(opcode2[i]==1) // BGEZ + if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL { emit_test(s1h,s1h); if(invert){ @@ -5750,7 +5756,7 @@ void sjump_assemble(int i,struct regstat *i_regs) else { assert(s1l>=0); - if(opcode2[i]==0) // BLTZ + if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL { emit_test(s1l,s1l); if(invert){ @@ -5761,7 +5767,7 @@ void sjump_assemble(int i,struct regstat *i_regs) emit_js(0); } } - if(opcode2[i]==1) // BGEZ + if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL { emit_test(s1l,s1l); if(invert){ @@ -6535,7 +6541,7 @@ void unneeded_registers(int istart,int iend,int r) { uu=u=0x300C0F3; // Discard at, a0-a3, t6-t9 } - if(start>0x80000400&&start<0x80800000) { + if(start>0x80000400&&start<0x80000000+RAM_SIZE) { if(itype[i]==UJUMP&&rt1[i]==31) { //uu=u=0x30300FF0FLL; // Discard at, v0-v1, t0-t9, lo, hi @@ -7764,9 +7770,9 @@ int new_recompile_block(int addr) } else #endif - if ((int)addr >= 0x80000000 && (int)addr < 0x80800000) { + if ((int)addr >= 0x80000000 && (int)addr < 0x80000000+RAM_SIZE) { source = (u_int *)((u_int)rdram+start-0x80000000); - pagelimit = 0x80800000; + pagelimit = 0x80000000+RAM_SIZE; } #ifndef DISABLE_TLB else if ((signed int)addr >= (signed int)0xC0000000) { @@ -8050,10 +8056,12 @@ int new_recompile_block(int addr) 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; case 0x1B: strcpy(insn[i],"LDR"); type=LOADLR; break; +#endif case 0x20: strcpy(insn[i],"LB"); type=LOAD; break; case 0x21: strcpy(insn[i],"LH"); type=LOAD; break; case 0x22: strcpy(insn[i],"LWL"); type=LOADLR; break; @@ -8066,20 +8074,26 @@ int new_recompile_block(int addr) case 0x29: strcpy(insn[i],"SH"); type=STORE; break; case 0x2A: strcpy(insn[i],"SWL"); type=STORELR; break; case 0x2B: strcpy(insn[i],"SW"); type=STORE; break; +#ifndef FORCE32 case 0x2C: strcpy(insn[i],"SDL"); type=STORELR; break; case 0x2D: strcpy(insn[i],"SDR"); type=STORELR; break; +#endif case 0x2E: strcpy(insn[i],"SWR"); type=STORELR; break; case 0x2F: strcpy(insn[i],"CACHE"); type=NOP; break; case 0x30: strcpy(insn[i],"LL"); type=NI; break; case 0x31: strcpy(insn[i],"LWC1"); type=C1LS; break; +#ifndef FORCE32 case 0x34: strcpy(insn[i],"LLD"); type=NI; break; case 0x35: strcpy(insn[i],"LDC1"); type=C1LS; break; case 0x37: strcpy(insn[i],"LD"); type=LOAD; break; +#endif case 0x38: strcpy(insn[i],"SC"); type=NI; break; case 0x39: strcpy(insn[i],"SWC1"); type=C1LS; break; +#ifndef FORCE32 case 0x3C: strcpy(insn[i],"SCD"); type=NI; break; case 0x3D: strcpy(insn[i],"SDC1"); type=C1LS; break; case 0x3F: strcpy(insn[i],"SD"); type=STORE; break; +#endif #ifdef PCSX case 0x12: strcpy(insn[i],"COP2"); type=NI; op2=(source[i]>>21)&0x1f; @@ -8102,7 +8116,7 @@ int new_recompile_block(int addr) case 0x3B: strcpy(insn[i],"HLECALL"); type=HLECALL; break; #endif default: strcpy(insn[i],"???"); type=NI; - printf("NI %08x @%08x\n", source[i], addr + i*4); + printf("NI %08x @%08x (%08x)\n", source[i], addr + i*4, addr); break; } itype[i]=type; @@ -8357,9 +8371,9 @@ int new_recompile_block(int addr) // Don't get too close to the limit if(i>MAXBLOCK/2) done=1; } - if(i>0&&itype[i-1]==SYSCALL&&stop_after_jal) done=1; - if(itype[i-1]==HLECALL) done=1; - assert(i