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=a40acf1725543b466d5d7b686bec80c42d7144f4;hp=c95058532947a798e3a1e3139944a8cdd9e7ac2b;hb=0bbd14543fec5fd4f5664b676771812663235252;hpb=7cd19de2fd689b38596df1c5821ce02f845dc654 diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c index c9505853..a40acf17 100644 --- a/libpcsxcore/new_dynarec/assem_arm.c +++ b/libpcsxcore/new_dynarec/assem_arm.c @@ -66,8 +66,41 @@ const u_int jump_vaddr_reg[16] = { 0, 0}; +void invalidate_addr_r0(); +void invalidate_addr_r1(); +void invalidate_addr_r2(); +void invalidate_addr_r3(); +void invalidate_addr_r4(); +void invalidate_addr_r5(); +void invalidate_addr_r6(); +void invalidate_addr_r7(); +void invalidate_addr_r8(); +void invalidate_addr_r9(); +void invalidate_addr_r10(); +void invalidate_addr_r12(); + +const u_int invalidate_addr_reg[16] = { + (int)invalidate_addr_r0, + (int)invalidate_addr_r1, + (int)invalidate_addr_r2, + (int)invalidate_addr_r3, + (int)invalidate_addr_r4, + (int)invalidate_addr_r5, + (int)invalidate_addr_r6, + (int)invalidate_addr_r7, + (int)invalidate_addr_r8, + (int)invalidate_addr_r9, + (int)invalidate_addr_r10, + 0, + (int)invalidate_addr_r12, + 0, + 0, + 0}; + #include "fpu.h" +unsigned int needs_clear_cache[1<<(TARGET_SIZE_2-17)]; + /* Linker */ void set_jump_target(int addr,u_int target) @@ -914,6 +947,56 @@ void emit_zeroreg(int rt) output_w32(0xe3a00000|rd_rn_rm(rt,0,0)); } +void emit_loadlp(u_int imm,u_int rt) +{ + add_literal((int)out,imm); + assem_debug("ldr %s,pc+? [=%x]\n",regname[rt],imm); + output_w32(0xe5900000|rd_rn_rm(rt,15,0)); +} +void emit_movw(u_int imm,u_int rt) +{ + assert(imm<65536); + assem_debug("movw %s,#%d (0x%x)\n",regname[rt],imm,imm); + output_w32(0xe3000000|rd_rn_rm(rt,0,0)|(imm&0xfff)|((imm<<4)&0xf0000)); +} +void emit_movt(u_int imm,u_int rt) +{ + assem_debug("movt %s,#%d (0x%x)\n",regname[rt],imm&0xffff0000,imm&0xffff0000); + output_w32(0xe3400000|rd_rn_rm(rt,0,0)|((imm>>16)&0xfff)|((imm>>12)&0xf0000)); +} +void emit_movimm(u_int imm,u_int rt) +{ + u_int armval; + if(genimm(imm,&armval)) { + assem_debug("mov %s,#%d\n",regname[rt],imm); + output_w32(0xe3a00000|rd_rn_rm(rt,0,0)|armval); + }else if(genimm(~imm,&armval)) { + assem_debug("mvn %s,#%d\n",regname[rt],imm); + output_w32(0xe3e00000|rd_rn_rm(rt,0,0)|armval); + }else if(imm<65536) { + #ifdef ARMv5_ONLY + assem_debug("mov %s,#%d\n",regname[rt],imm&0xFF00); + output_w32(0xe3a00000|rd_rn_imm_shift(rt,0,imm>>8,8)); + assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF); + output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); + #else + emit_movw(imm,rt); + #endif + }else{ + #ifdef ARMv5_ONLY + emit_loadlp(imm,rt); + #else + emit_movw(imm&0x0000FFFF,rt); + emit_movt(imm&0xFFFF0000,rt); + #endif + } +} +void emit_pcreladdr(u_int rt) +{ + assem_debug("add %s,pc,#?\n",regname[rt]); + output_w32(0xe2800000|rd_rn_rm(rt,15,0)); +} + void emit_loadreg(int r, int hr) { #ifdef FORCE32 @@ -1032,56 +1115,6 @@ void emit_xor(u_int rs1,u_int rs2,u_int rt) output_w32(0xe0200000|rd_rn_rm(rt,rs1,rs2)); } -void emit_loadlp(u_int imm,u_int rt) -{ - add_literal((int)out,imm); - assem_debug("ldr %s,pc+? [=%x]\n",regname[rt],imm); - output_w32(0xe5900000|rd_rn_rm(rt,15,0)); -} -void emit_movw(u_int imm,u_int rt) -{ - assert(imm<65536); - assem_debug("movw %s,#%d (0x%x)\n",regname[rt],imm,imm); - output_w32(0xe3000000|rd_rn_rm(rt,0,0)|(imm&0xfff)|((imm<<4)&0xf0000)); -} -void emit_movt(u_int imm,u_int rt) -{ - assem_debug("movt %s,#%d (0x%x)\n",regname[rt],imm&0xffff0000,imm&0xffff0000); - output_w32(0xe3400000|rd_rn_rm(rt,0,0)|((imm>>16)&0xfff)|((imm>>12)&0xf0000)); -} -void emit_movimm(u_int imm,u_int rt) -{ - u_int armval; - if(genimm(imm,&armval)) { - assem_debug("mov %s,#%d\n",regname[rt],imm); - output_w32(0xe3a00000|rd_rn_rm(rt,0,0)|armval); - }else if(genimm(~imm,&armval)) { - assem_debug("mvn %s,#%d\n",regname[rt],imm); - output_w32(0xe3e00000|rd_rn_rm(rt,0,0)|armval); - }else if(imm<65536) { - #ifdef ARMv5_ONLY - assem_debug("mov %s,#%d\n",regname[rt],imm&0xFF00); - output_w32(0xe3a00000|rd_rn_imm_shift(rt,0,imm>>8,8)); - assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF); - output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); - #else - emit_movw(imm,rt); - #endif - }else{ - #ifdef ARMv5_ONLY - emit_loadlp(imm,rt); - #else - emit_movw(imm&0x0000FFFF,rt); - emit_movt(imm&0xFFFF0000,rt); - #endif - } -} -void emit_pcreladdr(u_int rt) -{ - assem_debug("add %s,pc,#?\n",regname[rt]); - output_w32(0xe2800000|rd_rn_rm(rt,15,0)); -} - void emit_addimm(u_int rs,int imm,u_int rt) { assert(rs<16); @@ -1202,7 +1235,9 @@ void emit_sbb(int rs1,int rs2) void emit_andimm(int rs,int imm,int rt) { u_int armval; - if(genimm(imm,&armval)) { + if(imm==0) { + emit_zeroreg(rt); + }else if(genimm(imm,&armval)) { assem_debug("and %s,%s,#%d\n",regname[rt],regname[rs],imm); output_w32(0xe2000000|rd_rn_rm(rt,rs,0)|armval); }else if(genimm(~imm,&armval)) { @@ -1236,7 +1271,9 @@ void emit_andimm(int rs,int imm,int rt) void emit_orimm(int rs,int imm,int rt) { u_int armval; - if(genimm(imm,&armval)) { + if(imm==0) { + if(rs!=rt) emit_mov(rs,rt); + }else if(genimm(imm,&armval)) { assem_debug("orr %s,%s,#%d\n",regname[rt],regname[rs],imm); output_w32(0xe3800000|rd_rn_rm(rt,rs,0)|armval); }else{ @@ -1251,7 +1288,9 @@ void emit_orimm(int rs,int imm,int rt) void emit_xorimm(int rs,int imm,int rt) { u_int armval; - if(genimm(imm,&armval)) { + if(imm==0) { + if(rs!=rt) emit_mov(rs,rt); + }else if(genimm(imm,&armval)) { assem_debug("eor %s,%s,#%d\n",regname[rt],regname[rs],imm); output_w32(0xe2200000|rd_rn_rm(rt,rs,0)|armval); }else{ @@ -2187,6 +2226,13 @@ void emit_addsr12(int rs1,int rs2,int rt) output_w32(0xe0800620|rd_rn_rm(rt,rs1,rs2)); } +void emit_callne(int a) +{ + assem_debug("blne %x\n",a); + u_int offset=genjmp(a); + output_w32(0x1b000000|offset); +} + // Used to preload hash table entries void emit_prefetch(void *addr) { @@ -2579,9 +2625,7 @@ do_readstub(int n) } assert(rs>=0); if(addr<0) addr=rt; - if(addr<0) - // assume dummy read, no alloced reg - addr=get_reg(i_regmap,-1); + if(addr<0&&itype[i]!=C1LS&&itype[i]!=C2LS&&itype[i]!=LOADLR) addr=get_reg(i_regmap,-1); assert(addr>=0); int ftable=0; if(type==LOADB_STUB||type==LOADBU_STUB) @@ -2659,10 +2703,7 @@ inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, i int rs=get_reg(regmap,target); int rth=get_reg(regmap,target|64); int rt=get_reg(regmap,target); - // allow for PCSX dummy reads - //assert(rt>=0); - if(rs<0) - rs=get_reg(regmap,-1); + if(rs<0) rs=get_reg(regmap,-1); assert(rs>=0); int ftable=0; if(type==LOADB_STUB||type==LOADBU_STUB) @@ -3275,57 +3316,60 @@ void loadlr_assemble_arm(int i,struct regstat *i_regs) 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; } - if(tl>=0) { - //assert(tl>=0); - //assert(rt1[i]); - if(!using_tlb) { - if(!c) { - emit_shlimm(addr,3,temp); - if (opcode[i]==0x22||opcode[i]==0x26) { - emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR - }else{ - emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR - } - emit_cmpimm(addr,RAM_SIZE); - jaddr=(int)out; - emit_jno(0); - } - else { - if (opcode[i]==0x22||opcode[i]==0x26) { - emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR - }else{ - emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR - } - } - }else{ // using tlb - int a; - if(c) { - a=-1; - }else if (opcode[i]==0x22||opcode[i]==0x26) { - a=0xFFFFFFFC; // LWL/LWR + if(!using_tlb) { + if(!c) { + #ifdef RAM_OFFSET + map=get_reg(i_regs->regmap,ROREG); + if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG); + #endif + emit_shlimm(addr,3,temp); + if (opcode[i]==0x22||opcode[i]==0x26) { + emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR }else{ - a=0xFFFFFFF8; // LDL/LDR + emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR } - map=get_reg(i_regs->regmap,TLREG); - assert(map>=0); - map=do_tlb_r(addr,temp2,map,0,a,c?-1:temp,c,constmap[i][s]+offset); - if(c) { - if (opcode[i]==0x22||opcode[i]==0x26) { - emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR - }else{ - emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR - } + emit_cmpimm(addr,RAM_SIZE); + jaddr=(int)out; + emit_jno(0); + } + else { + if (opcode[i]==0x22||opcode[i]==0x26) { + emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR + }else{ + emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR } - do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr); } - if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR - if(!c||memtarget) { - //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2); - emit_readword_indexed_tlb((int)rdram-0x80000000,temp2,map,temp2); - if(jaddr) add_stub(LOADW_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist); + }else{ // using tlb + int a; + if(c) { + a=-1; + }else if (opcode[i]==0x22||opcode[i]==0x26) { + a=0xFFFFFFFC; // LWL/LWR + }else{ + a=0xFFFFFFF8; // LDL/LDR + } + map=get_reg(i_regs->regmap,TLREG); + assert(map>=0); + map=do_tlb_r(addr,temp2,map,0,a,c?-1:temp,c,constmap[i][s]+offset); + if(c) { + if (opcode[i]==0x22||opcode[i]==0x26) { + emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR + }else{ + emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR } - else - inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist); + } + do_tlb_r_branch(map,c,constmap[i][s]+offset,&jaddr); + } + if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR + if(!c||memtarget) { + //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2); + emit_readword_indexed_tlb(0,temp2,map,temp2); + if(jaddr) add_stub(LOADW_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist); + } + else + inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist); + if(rt1[i]) { + assert(tl>=0); emit_andimm(temp,24,temp); #ifdef BIG_ENDIAN_MIPS if (opcode[i]==0x26) // LWR @@ -3342,19 +3386,23 @@ void loadlr_assemble_arm(int i,struct regstat *i_regs) emit_bic_lsl(tl,HOST_TEMPREG,temp,tl); } emit_or(temp2,tl,tl); - //emit_storereg(rt1[i],tl); // DEBUG } - if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR - // FIXME: little endian - int temp2h=get_reg(i_regs->regmap,FTEMP|64); - if(!c||memtarget) { - //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,temp2,temp2h); - //emit_readword_indexed((int)rdram-0x7FFFFFFC,temp2,temp2); - emit_readdword_indexed_tlb((int)rdram-0x80000000,temp2,map,temp2h,temp2); - if(jaddr) add_stub(LOADD_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist); - } - else - inline_readstub(LOADD_STUB,i,(constmap[i][s]+offset)&0xFFFFFFF8,i_regs->regmap,FTEMP,ccadj[i],reglist); + //emit_storereg(rt1[i],tl); // DEBUG + } + if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR + // FIXME: little endian + int temp2h=get_reg(i_regs->regmap,FTEMP|64); + if(!c||memtarget) { + //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,temp2,temp2h); + //emit_readword_indexed((int)rdram-0x7FFFFFFC,temp2,temp2); + emit_readdword_indexed_tlb(0,temp2,map,temp2h,temp2); + if(jaddr) add_stub(LOADD_STUB,jaddr,(int)out,i,temp2,(int)i_regs,ccadj[i],reglist); + } + else + inline_readstub(LOADD_STUB,i,(constmap[i][s]+offset)&0xFFFFFFF8,i_regs->regmap,FTEMP,ccadj[i],reglist); + if(rt1[i]) { + assert(th>=0); + assert(tl>=0); emit_testimm(temp,32); emit_andimm(temp,24,temp); if (opcode[i]==0x1A) { // LDL @@ -4776,6 +4824,38 @@ void wb_invalidate_arm(signed char pre[],signed char entry[],uint64_t dirty,uint #define wb_invalidate wb_invalidate_arm */ +// Clearing the cache is rather slow on ARM Linux, so mark the areas +// that need to be cleared, and then only clear these areas once. +void do_clear_cache() +{ + int i,j; + for (i=0;i<(1<<(TARGET_SIZE_2-17));i++) + { + u_int bitmap=needs_clear_cache[i]; + if(bitmap) { + u_int start,end; + for(j=0;j<32;j++) + { + if(bitmap&(1<