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=e29f6c3f9a9d770281f6b0c5cc9f60d86e412e32;hp=c95058532947a798e3a1e3139944a8cdd9e7ac2b;hb=ffb0b9e0f051789f97f5efdcfab0b261e62688f9;hpb=7cd19de2fd689b38596df1c5821ce02f845dc654 diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c index c9505853..e29f6c3f 100644 --- a/libpcsxcore/new_dynarec/assem_arm.c +++ b/libpcsxcore/new_dynarec/assem_arm.c @@ -1,6 +1,7 @@ /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * - * Mupen64plus - assem_arm.c * - * Copyright (C) 2009-2010 Ari64 * + * Mupen64plus/PCSX - assem_arm.c * + * Copyright (C) 2009-2011 Ari64 * + * Copyright (C) 2010-2011 Gražvydas "notaz" Ignotas * * * * This program is free software; you can redistribute it and/or modify * * it under the terms of the GNU General Public License as published by * @@ -66,8 +67,41 @@ 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) @@ -137,6 +171,7 @@ void set_jump_target_fillslot(int addr,u_int target,int copy) /* Literal pool */ add_literal(int addr,int val) { + assert(literalcountregmap[n]==reg) {cur->regmap[n]=-1;} + if(n!=EXCLUDE_REG&&cur->regmap[n]==reg) { + dirty=(cur->dirty>>n)&1; + cur->regmap[n]=-1; + } } cur->regmap[hr]=reg; cur->dirty&=~(1<dirty|=dirty<isconst&=~(1<0) { @@ -914,12 +959,63 @@ 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 if(r&64) { printf("64bit load in 32bit mode!\n"); - exit(1); + assert(0); + return; } #endif if((r&63)==0) @@ -943,7 +1039,8 @@ void emit_storereg(int r, int hr) #ifdef FORCE32 if(r&64) { printf("64bit store in 32bit mode!\n"); - exit(1); + assert(0); + return; } #endif int addr=((int)reg)+((r&63)<>4); @@ -966,7 +1063,7 @@ void emit_test(int rs, int rt) void emit_testimm(int rs,int imm) { u_int armval; - assem_debug("tst %s,$%d\n",regname[rs],imm); + assem_debug("tst %s,#%d\n",regname[rs],imm); genimm_checked(imm,&armval); output_w32(0xe3100000|rd_rn_rm(0,rs,0)|armval); } @@ -1032,62 +1129,11 @@ 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); assert(rt<16); if(imm!=0) { - assert(imm>-65536&&imm<65536); u_int armval; if(genimm(imm,&armval)) { assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],imm); @@ -1096,11 +1142,13 @@ void emit_addimm(u_int rs,int imm,u_int rt) assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],imm); output_w32(0xe2400000|rd_rn_rm(rt,rs,0)|armval); }else if(imm<0) { + assert(imm>-65536); assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],(-imm)&0xFF00); assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rt],(-imm)&0xFF); output_w32(0xe2400000|rd_rn_imm_shift(rt,rs,(-imm)>>8,8)); output_w32(0xe2400000|rd_rn_imm_shift(rt,rt,(-imm)&0xff,0)); }else{ + assert(imm<65536); assem_debug("add %s,%s,#%d\n",regname[rt],regname[rs],imm&0xFF00); assem_debug("add %s,%s,#%d\n",regname[rt],regname[rt],imm&0xFF); output_w32(0xe2800000|rd_rn_imm_shift(rt,rs,imm>>8,8)); @@ -1202,7 +1250,9 @@ 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 +1286,9 @@ 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 +1303,9 @@ 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{ @@ -1272,6 +1326,14 @@ void emit_shlimm(int rs,u_int imm,int rt) output_w32(0xe1a00000|rd_rn_rm(rt,0,rs)|(imm<<7)); } +void emit_lsls_imm(int rs,int imm,int rt) +{ + assert(imm>0); + assert(imm<32); + assem_debug("lsls %s,%s,#%d\n",regname[rt],regname[rs],imm); + output_w32(0xe1b00000|rd_rn_rm(rt,0,rs)|(imm<<7)); +} + void emit_shrimm(int rs,u_int imm,int rt) { assert(imm>0); @@ -1331,6 +1393,17 @@ void emit_signextend16(int rs,int rt) #endif } +void emit_signextend8(int rs,int rt) +{ + #ifdef ARMv5_ONLY + emit_shlimm(rs,24,rt); + emit_sarimm(rt,24,rt); + #else + assem_debug("sxtb %s,%s\n",regname[rt],regname[rs]); + output_w32(0xe6af0070|rd_rn_rm(rt,0,rs)); + #endif +} + void emit_shl(u_int rs,u_int shift,u_int rt) { assert(rs<16); @@ -1403,10 +1476,10 @@ void emit_cmpimm(int rs,int imm) { u_int armval; if(genimm(imm,&armval)) { - assem_debug("cmp %s,$%d\n",regname[rs],imm); + assem_debug("cmp %s,#%d\n",regname[rs],imm); output_w32(0xe3500000|rd_rn_rm(0,rs,0)|armval); }else if(genimm(-imm,&armval)) { - assem_debug("cmn %s,$%d\n",regname[rs],imm); + assem_debug("cmn %s,#%d\n",regname[rs],imm); output_w32(0xe3700000|rd_rn_rm(0,rs,0)|armval); }else if(imm>0) { assert(imm<65536); @@ -1704,8 +1777,9 @@ void emit_popreg(u_int r) } void emit_callreg(u_int r) { - assem_debug("call *%%%s\n",regname[r]); - assert(0); + assert(r<15); + assem_debug("blx %s\n",regname[r]); + output_w32(0xe12fff30|r); } void emit_jmpreg(u_int r) { @@ -1728,6 +1802,31 @@ void emit_readword_dualindexedx4(int rs1, int rs2, int rt) assem_debug("ldr %s,%s,%s lsl #2\n",regname[rt],regname[rs1],regname[rs2]); output_w32(0xe7900000|rd_rn_rm(rt,rs1,rs2)|0x100); } +void emit_ldrcc_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrcc %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x37900000|rd_rn_rm(rt,rs1,rs2)); +} +void emit_ldrccb_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrccb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x37d00000|rd_rn_rm(rt,rs1,rs2)); +} +void emit_ldrccsb_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrccsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x319000d0|rd_rn_rm(rt,rs1,rs2)); +} +void emit_ldrcch_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrcch %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x319000b0|rd_rn_rm(rt,rs1,rs2)); +} +void emit_ldrccsh_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrccsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x319000f0|rd_rn_rm(rt,rs1,rs2)); +} void emit_readword_indexed_tlb(int addr, int rs, int map, int rt) { if(map<0) emit_readword_indexed(addr, rs, rt); @@ -1943,6 +2042,21 @@ void emit_writebyte_indexed_tlb(int rt, int addr, int rs, int map, int temp) } } } +void emit_strcc_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("strcc %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x37800000|rd_rn_rm(rt,rs1,rs2)); +} +void emit_strccb_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("strccb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x37c00000|rd_rn_rm(rt,rs1,rs2)); +} +void emit_strcch_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("strcch %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0x318000b0|rd_rn_rm(rt,rs1,rs2)); +} void emit_writeword(int rt, int addr) { u_int offset = addr-(u_int)&dynarec_local; @@ -2187,6 +2301,13 @@ 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) { @@ -2556,6 +2677,27 @@ emit_extjump_ds(int addr, int target) emit_extjump2(addr, target, (int)dyna_linker_ds); } +#ifdef PCSX +#include "pcsxmem_inline.c" +#endif + +// trashes r2 +static void pass_args(int a0, int a1) +{ + if(a0==1&&a1==0) { + // must swap + emit_mov(a0,2); emit_mov(a1,1); emit_mov(2,0); + } + else if(a0!=0&&a1==0) { + emit_mov(a1,1); + if (a0>=0) emit_mov(a0,0); + } + else { + if(a0>=0&&a0!=0) emit_mov(a0,0); + if(a1>=0&&a1!=1) emit_mov(a1,1); + } +} + do_readstub(int n) { assem_debug("do_readstub %x\n",start+stubs[n][3]*4); @@ -2578,10 +2720,75 @@ do_readstub(int n) rt=get_reg(i_regmap,rt1[i]); } assert(rs>=0); +#ifdef PCSX + int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0,restore_jump=0; + reglist|=(1<=0) + reglist&=~(1<=0&&rt1[i]!=0)) { + switch(type) { + case LOADB_STUB: emit_ldrccsb_dualindexed(temp2,rs,rt); break; + case LOADBU_STUB: emit_ldrccb_dualindexed(temp2,rs,rt); break; + case LOADH_STUB: emit_ldrccsh_dualindexed(temp2,rs,rt); break; + case LOADHU_STUB: emit_ldrcch_dualindexed(temp2,rs,rt); break; + case LOADW_STUB: emit_ldrcc_dualindexed(temp2,rs,rt); break; + } + } + if(regs_saved) { + restore_jump=(int)out; + emit_jcc(0); // jump to reg restore + } + else + emit_jcc(stubs[n][2]); // return address + + if(!regs_saved) + save_regs(reglist); + int handler=0; + if(type==LOADB_STUB||type==LOADBU_STUB) + handler=(int)jump_handler_read8; + if(type==LOADH_STUB||type==LOADHU_STUB) + handler=(int)jump_handler_read16; + if(type==LOADW_STUB) + handler=(int)jump_handler_read32; + assert(handler!=0); + pass_args(rs,temp2); + int cc=get_reg(i_regmap,CCREG); + if(cc<0) + emit_loadreg(CCREG,2); + emit_addimm(cc<0?2:cc,CLOCK_DIVIDER*stubs[n][6]+2,2); + emit_call(handler); + if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) { + switch(type) { + case LOADB_STUB: emit_signextend8(0,rt); break; + case LOADBU_STUB: emit_andimm(0,0xff,rt); break; + case LOADH_STUB: emit_signextend16(0,rt); break; + case LOADHU_STUB: emit_andimm(0,0xffff,rt); break; + case LOADW_STUB: if(rt!=0) emit_mov(0,rt); break; + } + } + if(restore_jump) + set_jump_target(restore_jump,(int)out); + restore_regs(reglist); + emit_jmp(stubs[n][2]); // return address +#else // !PCSX 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) @@ -2598,12 +2805,14 @@ do_readstub(int n) emit_writeword(rs,(int)&address); //emit_pusha(); save_regs(reglist); +#ifndef PCSX ds=i_regs!=®s[i]; int real_rs=(itype[i]==LOADLR)?-1:get_reg(i_regmap,rs1[i]); u_int cmask=ds?-1:(0x100f|~i_regs->wasconst); if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<regmap_entry,i_regs->was32,i_regs->wasdirty&cmask&~(1<>rs1[i])&1)<<1)+ds,3); +#endif //emit_readword((int)&last_count,temp); //emit_add(cc,temp,cc); //emit_writeword(cc,(int)&Count); @@ -2619,6 +2830,7 @@ do_readstub(int n) emit_call((int)&indirect_jump_indexed); //emit_callreg(rs); //emit_readword_dualindexedx4(rs,HOST_TEMPREG,15); +#ifndef PCSX // We really shouldn't need to update the count here, // but not doing so causes random crashes... emit_readword((int)&Count,HOST_TEMPREG); @@ -2629,6 +2841,7 @@ do_readstub(int n) if(cc<0) { emit_storereg(CCREG,HOST_TEMPREG); } +#endif //emit_popa(); restore_regs(reglist); //if((cc=get_reg(regmap,CCREG))>=0) { @@ -2652,18 +2865,102 @@ do_readstub(int n) } } emit_jmp(stubs[n][2]); // return address +#endif // !PCSX +} + +#ifdef PCSX +// return memhandler, or get directly accessable address and return 0 +u_int get_direct_memhandler(void *table,u_int addr,int type,u_int *addr_host) +{ + u_int l1,l2=0; + l1=((u_int *)table)[addr>>12]; + if((l1&(1<<31))==0) { + u_int v=l1<<1; + *addr_host=v+addr; + return 0; + } + else { + l1<<=1; + if(type==LOADB_STUB||type==LOADBU_STUB||type==STOREB_STUB) + l2=((u_int *)l1)[0x1000/4 + 0x1000/2 + (addr&0xfff)]; + else if(type==LOADH_STUB||type==LOADHU_STUB||type==STOREH_STUB) + l2=((u_int *)l1)[0x1000/4 + (addr&0xfff)/2]; + else + l2=((u_int *)l1)[(addr&0xfff)/4]; + if((l2&(1<<31))==0) { + u_int v=l2<<1; + *addr_host=v+(addr&0xfff); + return 0; + } + return l2<<1; + } } +#endif inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) { 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); +#ifdef PCSX + u_int handler,host_addr=0; + if(pcsx_direct_read(type,addr,target?rs:-1,rt)) + return; + handler=get_direct_memhandler(mem_rtab,addr,type,&host_addr); + if (handler==0) { + if(rt<0) + return; + if(target==0||addr!=host_addr) + emit_movimm(host_addr,rs); + switch(type) { + case LOADB_STUB: emit_movsbl_indexed(0,rs,rt); break; + case LOADBU_STUB: emit_movzbl_indexed(0,rs,rt); break; + case LOADH_STUB: emit_movswl_indexed(0,rs,rt); break; + case LOADHU_STUB: emit_movzwl_indexed(0,rs,rt); break; + case LOADW_STUB: emit_readword_indexed(0,rs,rt); break; + default: assert(0); + } + return; + } + + // call a memhandler + if(rt>=0) + reglist&=~(1<=33554432) { + // unreachable memhandler, a plugin func perhaps + emit_movimm(handler,1); + emit_callreg(1); + } + else + emit_call(handler); + if(rt>=0) { + switch(type) { + case LOADB_STUB: emit_signextend8(0,rt); break; + case LOADBU_STUB: emit_andimm(0,0xff,rt); break; + case LOADH_STUB: emit_signextend16(0,rt); break; + case LOADHU_STUB: emit_andimm(0,0xffff,rt); break; + case LOADW_STUB: if(rt!=0) emit_mov(0,rt); break; + default: assert(0); + } + } + restore_regs(reglist); +#else // if !PCSX int ftable=0; if(type==LOADB_STUB||type==LOADBU_STUB) ftable=(int)readmemb; @@ -2681,6 +2978,18 @@ inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, i emit_writeword(rs,(int)&address); //emit_pusha(); save_regs(reglist); +#ifndef PCSX + if((signed int)addr>=(signed int)0xC0000000) { + // Theoretically we can have a pagefault here, if the TLB has never + // been enabled and the address is outside the range 80000000..BFFFFFFF + // Write out the registers so the pagefault can be handled. This is + // a very rare case and likely represents a bug. + int ds=regmap!=regs[i].regmap; + if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i); + if(!ds) wb_dirtys(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty); + else wb_dirtys(branch_regs[i-1].regmap_entry,branch_regs[i-1].was32,branch_regs[i-1].wasdirty); + } +#endif //emit_shrimm(rs,16,1); int cc=get_reg(regmap,CCREG); if(cc<0) { @@ -2690,15 +2999,18 @@ inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, i emit_movimm(((u_int *)ftable)[addr>>16],0); //emit_readword((int)&last_count,12); emit_addimm(cc<0?2:cc,CLOCK_DIVIDER*(adj+1),2); +#ifndef PCSX if((signed int)addr>=(signed int)0xC0000000) { // Pagefault address int ds=regmap!=regs[i].regmap; emit_movimm(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3); } +#endif //emit_add(12,2,2); //emit_writeword(2,(int)&Count); //emit_call(((u_int *)ftable)[addr>>16]); emit_call((int)&indirect_jump); +#ifndef PCSX // We really shouldn't need to update the count here, // but not doing so causes random crashes... emit_readword((int)&Count,HOST_TEMPREG); @@ -2709,6 +3021,7 @@ inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, i if(cc<0) { emit_storereg(CCREG,HOST_TEMPREG); } +#endif //emit_popa(); restore_regs(reglist); if(rt>=0) { @@ -2727,6 +3040,7 @@ inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, i if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth); } } +#endif // !PCSX } do_writestub(int n) @@ -2752,6 +3066,68 @@ do_writestub(int n) } assert(rs>=0); assert(rt>=0); +#ifdef PCSX + int rtmp,temp=-1,temp2=HOST_TEMPREG,regs_saved=0,restore_jump=0,ra; + int reglist2=reglist|(1<=0); int ftable=0; @@ -2785,12 +3161,14 @@ do_writestub(int n) } //emit_pusha(); save_regs(reglist); +#ifndef PCSX ds=i_regs!=®s[i]; int real_rs=get_reg(i_regmap,rs1[i]); u_int cmask=ds?-1:(0x100f|~i_regs->wasconst); if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<regmap_entry,i_regs->was32,i_regs->wasdirty&cmask&~(1<>rs1[i])&1)<<1)+ds,3); +#endif //emit_readword((int)&last_count,temp); //emit_addimm(cc,2*stubs[n][5]+2,cc); //emit_add(cc,temp,cc); @@ -2819,6 +3199,7 @@ do_writestub(int n) // emit_loadreg(CCREG,cc); //} emit_jmp(stubs[n][2]); // return address +#endif // !PCSX } inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) @@ -2828,6 +3209,40 @@ inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int rt=get_reg(regmap,target); assert(rs>=0); assert(rt>=0); +#ifdef PCSX + u_int handler,host_addr=0; + if(pcsx_direct_write(type,addr,rs,rt,regmap)) + return; + handler=get_direct_memhandler(mem_wtab,addr,type,&host_addr); + if (handler==0) { + if(target==0||addr!=host_addr) + emit_movimm(host_addr,rs); + switch(type) { + case STOREB_STUB: emit_writebyte_indexed(rt,0,rs); break; + case STOREH_STUB: emit_writehword_indexed(rt,0,rs); break; + case STOREW_STUB: emit_writeword_indexed(rt,0,rs); break; + default: assert(0); + } + return; + } + + // call a memhandler + save_regs(reglist); + pass_args(target!=0?rs:-1,rt); + if(target==0) + emit_movimm(addr,0); + int cc=get_reg(regmap,CCREG); + if(cc<0) + emit_loadreg(CCREG,2); + emit_addimm(cc<0?2:cc,CLOCK_DIVIDER*(adj+1),2); + emit_movimm(handler,3); + // returns new cycle_count + emit_call((int)jump_handler_write_h); + emit_addimm(0,-CLOCK_DIVIDER*(adj+1),cc<0?2:cc); + if(cc<0) + emit_storereg(CCREG,2); + restore_regs(reglist); +#else // if !pcsx int ftable=0; if(type==STOREB_STUB) ftable=(int)writememb; @@ -2859,6 +3274,19 @@ inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, } //emit_pusha(); save_regs(reglist); +#ifndef PCSX + // rearmed note: load_all_consts prevents BIOS boot, some bug? + if((signed int)addr>=(signed int)0xC0000000) { + // Theoretically we can have a pagefault here, if the TLB has never + // been enabled and the address is outside the range 80000000..BFFFFFFF + // Write out the registers so the pagefault can be handled. This is + // a very rare case and likely represents a bug. + int ds=regmap!=regs[i].regmap; + if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i); + if(!ds) wb_dirtys(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty); + else wb_dirtys(branch_regs[i-1].regmap_entry,branch_regs[i-1].was32,branch_regs[i-1].wasdirty); + } +#endif //emit_shrimm(rs,16,1); int cc=get_reg(regmap,CCREG); if(cc<0) { @@ -2868,11 +3296,13 @@ inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, emit_movimm(((u_int *)ftable)[addr>>16],0); //emit_readword((int)&last_count,12); emit_addimm(cc<0?2:cc,CLOCK_DIVIDER*(adj+1),2); +#ifndef PCSX if((signed int)addr>=(signed int)0xC0000000) { // Pagefault address int ds=regmap!=regs[i].regmap; emit_movimm(start+i*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3); } +#endif //emit_add(12,2,2); //emit_writeword(2,(int)&Count); //emit_call(((u_int *)ftable)[addr>>16]); @@ -2887,6 +3317,7 @@ inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, } //emit_popa(); restore_regs(reglist); +#endif } do_unalignedwritestub(int n) @@ -2910,16 +3341,33 @@ do_unalignedwritestub(int n) reglist|=(1<wasconst); if(!ds) load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty&~(1<regmap_entry,i_regs->was32,i_regs->wasdirty&cmask&~(1<>rs1[i])&1)<<1)+ds,3); // XXX: can be rm'd? +#ifndef PCSX + // pagefault address + emit_movimm(start+stubs[n][3]*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3); +#endif emit_call((int)&indirect_jump_indexed); restore_regs(reglist); @@ -2967,6 +3418,7 @@ do_unalignedwritestub(int n) } restore_regs(reglist); emit_jmp(stubs[n][2]); // return address +#endif } void printregs(int edi,int esi,int ebp,int esp,int b,int d,int c,int a) @@ -3247,6 +3699,182 @@ void shift_assemble_arm(int i,struct regstat *i_regs) } } } + +#ifdef PCSX +static void speculate_mov(int rs,int rt) +{ + if(rt!=0) { + smrv_strong_next|=1<>rs1[i])&1) speculate_mov(rs1[i],rt1[i]); + else if((smrv_strong>>rs2[i])&1) speculate_mov(rs2[i],rt1[i]); + else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]); + else if((smrv_weak>>rs2[i])&1) speculate_mov_weak(rs2[i],rt1[i]); + else { + smrv_strong_next&=~(1<=0) { + if(get_final_value(hr,i,&value)) + smrv[rt1[i]]=value; + else smrv[rt1[i]]=constmap[i][hr]; + smrv_strong_next|=1<>rs1[i])&1) speculate_mov(rs1[i],rt1[i]); + else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]); + } + break; + case LOAD: + if(start<0x2000&&(rt1[i]==26||(smrv[rt1[i]]>>24)==0xa0)) { + // special case for BIOS + smrv[rt1[i]]=0xa0000000; + smrv_strong_next|=1<>r)&1),(smrv_weak>>r)&1,regs[i].isconst,regs[i].wasconst); +#endif +} + +enum { + MTYPE_8000 = 0, + MTYPE_8020, + MTYPE_0000, + MTYPE_A000, + MTYPE_1F80, +}; + +static int get_ptr_mem_type(u_int a) +{ + if(a < 0x00200000) { + if(a<0x1000&&((start>>20)==0xbfc||(start>>24)==0xa0)) + // return wrong, must use memhandler for BIOS self-test to pass + // 007 does similar stuff from a00 mirror, weird stuff + return MTYPE_8000; + return MTYPE_0000; + } + if(0x1f800000 <= a && a < 0x1f801000) + return MTYPE_1F80; + if(0x80200000 <= a && a < 0x80800000) + return MTYPE_8020; + if(0xa0000000 <= a && a < 0xa0200000) + return MTYPE_A000; + return MTYPE_8000; +} +#endif + +static int emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override) +{ + int jaddr,type=0; + +#ifdef PCSX + int mr=rs1[i]; + if(((smrv_strong|smrv_weak)>>mr)&1) { + type=get_ptr_mem_type(smrv[mr]); + //printf("set %08x @%08x r%d %d\n", smrv[mr], start+i*4, mr, type); + } + else { + // use the mirror we are running on + type=get_ptr_mem_type(start); + //printf("set nospec @%08x r%d %d\n", start+i*4, mr, type); + } + + if(type==MTYPE_8020) { // RAM 80200000+ mirror + emit_andimm(addr,~0x00e00000,HOST_TEMPREG); + addr=*addr_reg_override=HOST_TEMPREG; + type=0; + } + else if(type==MTYPE_0000) { // RAM 0 mirror + emit_orimm(addr,0x80000000,HOST_TEMPREG); + addr=*addr_reg_override=HOST_TEMPREG; + type=0; + } + else if(type==MTYPE_A000) { // RAM A mirror + emit_andimm(addr,~0x20000000,HOST_TEMPREG); + addr=*addr_reg_override=HOST_TEMPREG; + type=0; + } + else if(type==MTYPE_1F80) { // scratchpad + emit_addimm(addr,-0x1f800000,HOST_TEMPREG); + emit_cmpimm(HOST_TEMPREG,0x1000); + jaddr=(int)out; + emit_jc(0); + } +#endif + + if(type==0) + { + 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); + } + + return jaddr; +} + #define shift_assemble shift_assemble_arm void loadlr_assemble_arm(int i,struct regstat *i_regs) @@ -3254,7 +3882,8 @@ void loadlr_assemble_arm(int i,struct regstat *i_regs) int s,th,tl,temp,temp2,addr,map=-1; int offset; int jaddr=0; - int memtarget,c=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]); @@ -3272,60 +3901,66 @@ void loadlr_assemble_arm(int i,struct regstat *i_regs) else addr=s; if(s>=0) { c=(i_regs->wasconst>>s)&1; - 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(c) { + 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(!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 - } + jaddr=emit_fastpath_cmp_jump(i,temp2,&fastload_reg_override); + } + 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); + reglist&=~(1<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) { + int a=temp2; + if(fastload_reg_override) a=fastload_reg_override; + //emit_readword_indexed((int)rdram-0x80000000,temp2,temp2); + emit_readword_indexed_tlb(0,a,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 +3977,23 @@ 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, fastload_reg_override + 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 @@ -3674,6 +4313,7 @@ void c2op_assemble(int i,struct regstat *i_regs) signed char temp=get_reg(i_regs->regmap,-1); u_int c2op=source[i]&0x3f; u_int hr,reglist=0; + int need_flags; for(hr=0;hrregmap[hr]>=0) reglist|=1<regmap,CCREG); - emit_movimm(source[i],temp); // opcode + emit_movimm(source[i],1); // opcode if (cc>=0&>e_cycletab[c2op]) - emit_addimm(cc,gte_cycletab[c2op]/2,cc); // XXX: cound just adjust ccadj? - emit_writeword(temp,(int)&psxRegs.code); - emit_call((int)gte_handlers[c2op]); + emit_addimm(cc,gte_cycletab[c2op]/2,cc); // XXX: could just adjust ccadj? + emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); // cop2 regs + emit_writeword(1,(int)&psxRegs.code); + need_flags=!(gte_unneeded[i+1]>>63); // +1 because of how liveness detection works + assem_debug("gte unneeded %016llx, need_flags %d\n",gte_unneeded[i+1],need_flags); +#ifdef ARMv5_ONLY + // let's take more risk here + need_flags=need_flags&>e_reads_flags; +#endif + emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op])); } if(i>=slen-1||itype[i+1]!=C2OP) @@ -4375,7 +5022,9 @@ void multdiv_assemble_arm(int i,struct regstat *i_regs) assert(quotient>=0); assert(remainder>=0); emit_movs(d1,remainder); - emit_negmi(remainder,remainder); + emit_movimm(0xffffffff,quotient); + emit_negmi(quotient,quotient); // .. quotient and .. + emit_negmi(remainder,remainder); // .. remainder for div0 case (will be negated back after jump) emit_movs(d2,HOST_TEMPREG); emit_jeq((int)out+52); // Division by zero emit_negmi(HOST_TEMPREG,HOST_TEMPREG); @@ -4403,12 +5052,13 @@ void multdiv_assemble_arm(int i,struct regstat *i_regs) signed char remainder=get_reg(i_regs->regmap,HIREG); assert(quotient>=0); assert(remainder>=0); + emit_mov(d1,remainder); + emit_movimm(0xffffffff,quotient); // div0 case emit_test(d2,d2); - emit_jeq((int)out+44); // Division by zero + emit_jeq((int)out+40); // Division by zero emit_clz(d2,HOST_TEMPREG); emit_movimm(1<<31,quotient); emit_shl(d2,HOST_TEMPREG,d2); - emit_mov(d1,remainder); emit_shr(quotient,HOST_TEMPREG,quotient); emit_cmp(remainder,d2); emit_subcs(remainder,d2,remainder); @@ -4418,6 +5068,7 @@ void multdiv_assemble_arm(int i,struct regstat *i_regs) } } else // 64-bit +#ifndef FORCE32 { if(opcode2[i]==0x1C) // DMULT { @@ -4591,6 +5242,9 @@ void multdiv_assemble_arm(int i,struct regstat *i_regs) if(lol>=0) emit_loadreg(LOREG,lol); } } +#else + assert(0); +#endif } else { @@ -4685,7 +5339,7 @@ void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty, if(hr!=EXCLUDE_REG) { reg=pre[hr]; if(((~u)>>(reg&63))&1) { - if(reg==entry[hr]||(reg>0&&entry[hr]<0)) { + if(reg>0) { if(((dirty_pre&~dirty)>>hr)&1) { if(reg>0&®<34) { emit_storereg(reg,hr); @@ -4699,21 +5353,6 @@ void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty, } } } - else // Check if register moved to a different register - if((new_hr=get_reg(entry,reg))>=0) { - if((dirty_pre>>hr)&(~dirty>>new_hr)&1) { - if(reg>0&®<34) { - emit_storereg(reg,hr); - if( ((is32_pre&~uu)>>reg)&1 ) { - emit_sarimm(hr,31,HOST_TEMPREG); - emit_storereg(reg|64,HOST_TEMPREG); - } - } - else if(reg>=64) { - emit_storereg(reg,hr); - } - } - } } } } @@ -4776,6 +5415,38 @@ 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<