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=23d3b7f7bdb1f5286f4226ae4b9555049d5de499;hp=c61145643efa3efb4f33f5a9e42ea26fe476b825;hb=b7ec323c2e42a9ff8df844e5a95665733abb4bc1;hpb=3968e69e7fa8f9cb0d44ac79477d5929b9649271 diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c index c6114564..23d3b7f7 100644 --- a/libpcsxcore/new_dynarec/assem_arm.c +++ b/libpcsxcore/new_dynarec/assem_arm.c @@ -1,7 +1,7 @@ /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Mupen64plus/PCSX - assem_arm.c * * Copyright (C) 2009-2011 Ari64 * - * Copyright (C) 2010-2011 Gražvydas "notaz" Ignotas * + * Copyright (C) 2010-2021 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 * @@ -19,7 +19,6 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ -#include "../gte.h" #define FLAGLESS #include "../gte.h" #undef FLAGLESS @@ -28,19 +27,6 @@ #include "pcnt.h" #include "arm_features.h" -#if defined(BASE_ADDR_FIXED) -#elif defined(BASE_ADDR_DYNAMIC) -u_char *translation_cache; -#else -u_char translation_cache[1 << TARGET_SIZE_2] __attribute__((aligned(4096))); -#endif - -#ifndef __MACH__ -#define CALLER_SAVE_REGS 0x100f -#else -#define CALLER_SAVE_REGS 0x120f -#endif - #define unused __attribute__((unused)) #ifdef DRC_DBG @@ -214,7 +200,7 @@ static void *get_pointer(void *stub) { //printf("get_pointer(%x)\n",(int)stub); int *i_ptr=find_extjump_insn(stub); - assert((*i_ptr&0x0f000000)==0x0a000000); + assert((*i_ptr&0x0f000000)==0x0a000000); // b return (u_char *)i_ptr+((*i_ptr<<8)>>6)+8; } @@ -449,6 +435,13 @@ static void emit_add(int rs1,int rs2,int rt) output_w32(0xe0800000|rd_rn_rm(rt,rs1,rs2)); } +static void emit_adds(int rs1,int rs2,int rt) +{ + assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe0900000|rd_rn_rm(rt,rs1,rs2)); +} +#define emit_adds_ptr emit_adds + static void emit_adcs(int rs1,int rs2,int rt) { assem_debug("adcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); @@ -544,6 +537,7 @@ static void emit_loadreg(int r, int hr) case CCREG: addr = (int)&cycle_count; break; case CSREG: addr = (int)&Status; break; case INVCP: addr = (int)&invc_ptr; break; + case ROREG: addr = (int)&ram_offset; break; default: assert(r < 34); break; } u_int offset = addr-(u_int)&dynarec_local; @@ -707,11 +701,6 @@ static void emit_addimm_and_set_flags(int imm,int rt) } } -static void emit_addimm_no_flags(u_int imm,u_int rt) -{ - emit_addimm(rt,imm,rt); -} - static void emit_addnop(u_int r) { assert(r<16); @@ -972,6 +961,12 @@ static void emit_cmovl_reg(int rs,int rt) output_w32(0xb1a00000|rd_rn_rm(rt,0,rs)); } +static void emit_cmovb_reg(int rs,int rt) +{ + assem_debug("movcc %s,%s\n",regname[rt],regname[rs]); + output_w32(0x31a00000|rd_rn_rm(rt,0,rs)); +} + static void emit_cmovs_reg(int rs,int rt) { assem_debug("movmi %s,%s\n",regname[rt],regname[rs]); @@ -1034,6 +1029,12 @@ static void emit_set_if_carry32(int rs1, int rs2, int rt) emit_cmovb_imm(1,rt); } +static int can_jump_or_call(const void *a) +{ + intptr_t offset = (u_char *)a - out - 8; + return (-33554432 <= offset && offset < 33554432); +} + static void emit_call(const void *a_) { int a = (int)a_; @@ -1156,6 +1157,13 @@ static 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); } +#define emit_readptr_dualindexedx_ptrlen emit_readword_dualindexedx4 + +static void emit_ldr_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldr %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe7900000|rd_rn_rm(rt,rs1,rs2)); +} static void emit_ldrcc_dualindexed(int rs1, int rs2, int rt) { @@ -1163,30 +1171,72 @@ static void emit_ldrcc_dualindexed(int rs1, int rs2, int rt) output_w32(0x37900000|rd_rn_rm(rt,rs1,rs2)); } +static void emit_ldrb_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe7d00000|rd_rn_rm(rt,rs1,rs2)); +} + static 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)); } +static void emit_ldrsb_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe19000d0|rd_rn_rm(rt,rs1,rs2)); +} + static 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)); } +static void emit_ldrh_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe19000b0|rd_rn_rm(rt,rs1,rs2)); +} + static 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)); } +static void emit_ldrsh_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("ldrsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe19000f0|rd_rn_rm(rt,rs1,rs2)); +} + static 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)); } +static void emit_str_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("str %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe7800000|rd_rn_rm(rt,rs1,rs2)); +} + +static void emit_strb_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("strb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe7c00000|rd_rn_rm(rt,rs1,rs2)); +} + +static void emit_strh_dualindexed(int rs1, int rs2, int rt) +{ + assem_debug("strh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]); + output_w32(0xe18000b0|rd_rn_rm(rt,rs1,rs2)); +} + static void emit_movsbl_indexed(int offset, int rs, int rt) { assert(offset>-256&&offset<256); @@ -1249,6 +1299,7 @@ static void emit_readword(void *addr, int rt) assem_debug("ldr %s,fp+%d\n",regname[rt],offset); output_w32(0xe5900000|rd_rn_rm(rt,FP,0)|offset); } +#define emit_readptr emit_readword static void emit_writeword_indexed(int rt, int offset, int rs) { @@ -1586,7 +1637,7 @@ static void emit_extjump2(u_char *addr, u_int target, void *linker) emit_loadlp(target,0); emit_loadlp((u_int)addr,1); - assert(addr>=translation_cache&&addr<(translation_cache+(1<=ndrc->translation_cache&&addr<(ndrc->translation_cache+(1<=0x80000000&&target<0x80800000)||(target>0xA4000000&&target<0xA4001000)); //DEBUG > #ifdef DEBUG_CYCLE_COUNT @@ -1598,7 +1649,7 @@ static void emit_extjump2(u_char *addr, u_int target, void *linker) emit_writeword(ECX,&last_count); #endif //DEBUG < - emit_jmp(linker); + emit_far_jump(linker); } static void check_extjump2(void *src) @@ -1675,14 +1726,14 @@ static void do_readstub(int n) enum stub_type type=stubs[n].type; int i=stubs[n].a; int rs=stubs[n].b; - struct regstat *i_regs=(struct regstat *)stubs[n].c; + const struct regstat *i_regs=(struct regstat *)stubs[n].c; u_int reglist=stubs[n].e; - signed char *i_regmap=i_regs->regmap; + const signed char *i_regmap=i_regs->regmap; int rt; - if(itype[i]==C1LS||itype[i]==C2LS||itype[i]==LOADLR) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS||dops[i].itype==LOADLR) { rt=get_reg(i_regmap,FTEMP); }else{ - rt=get_reg(i_regmap,rt1[i]); + rt=get_reg(i_regmap,dops[i].rt1); } assert(rs>=0); int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0; @@ -1693,7 +1744,7 @@ static void do_readstub(int n) temp=r; break; } } - if(rt>=0&&rt1[i]!=0) + if(rt>=0&&dops[i].rt1!=0) reglist&=~(1<=0&&rt1[i]!=0)) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) { switch(type) { case LOADB_STUB: emit_ldrccsb_dualindexed(temp2,rs,rt); break; case LOADBU_STUB: emit_ldrccb_dualindexed(temp2,rs,rt); break; @@ -1737,9 +1788,9 @@ static void do_readstub(int n) int cc=get_reg(i_regmap,CCREG); if(cc<0) emit_loadreg(CCREG,2); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d+1),2); - emit_call(handler); - if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) { + emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2); + emit_far_call(handler); + if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) { mov_loadtype_adj(type,0,rt); } if(restore_jump) @@ -1748,21 +1799,22 @@ static void do_readstub(int n) emit_jmp(stubs[n].retaddr); // return address } -static void inline_readstub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) +static void inline_readstub(enum stub_type type, int i, u_int addr, + const signed char regmap[], int target, int adj, u_int reglist) { int rs=get_reg(regmap,target); int rt=get_reg(regmap,target); if(rs<0) rs=get_reg(regmap,-1); assert(rs>=0); - u_int is_dynamic,far_call=0; + u_int is_dynamic; uintptr_t host_addr = 0; void *handler; int cc=get_reg(regmap,CCREG); - if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt)) + if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj),cc,target?rs:-1,rt)) return; handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr); if (handler == NULL) { - if(rt<0||rt1[i]==0) + if(rt<0||dops[i].rt1==0) return; if(addr!=host_addr) emit_movimm_from(addr,rs,host_addr,rs); @@ -1787,38 +1839,29 @@ static void inline_readstub(enum stub_type type, int i, u_int addr, signed char } // call a memhandler - if(rt>=0&&rt1[i]!=0) + if(rt>=0&&dops[i].rt1!=0) reglist&=~(1<=33554432) { - // unreachable memhandler, a plugin func perhaps - emit_movimm((u_int)handler,12); - far_call=1; - } if(cc<0) emit_loadreg(CCREG,2); if(is_dynamic) { emit_movimm(((u_int *)mem_rtab)[addr>>12]<<1,1); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2); } else { emit_readword(&last_count,3); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2); emit_add(2,3,2); emit_writeword(2,&Count); } - if(far_call) - emit_callreg(12); - else - emit_call(handler); + emit_far_call(handler); - if(rt>=0&&rt1[i]!=0) { + if(rt>=0&&dops[i].rt1!=0) { switch(type) { case LOADB_STUB: emit_signextend8(0,rt); break; case LOADBU_STUB: emit_andimm(0,0xff,rt); break; @@ -1839,14 +1882,14 @@ static void do_writestub(int n) enum stub_type type=stubs[n].type; int i=stubs[n].a; int rs=stubs[n].b; - struct regstat *i_regs=(struct regstat *)stubs[n].c; + const struct regstat *i_regs=(struct regstat *)stubs[n].c; u_int reglist=stubs[n].e; - signed char *i_regmap=i_regs->regmap; + const signed char *i_regmap=i_regs->regmap; int rt,r; - if(itype[i]==C1LS||itype[i]==C2LS) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS) { rt=get_reg(i_regmap,r=FTEMP); }else{ - rt=get_reg(i_regmap,r=rs2[i]); + rt=get_reg(i_regmap,r=dops[i].rs2); } assert(rs>=0); assert(rt>=0); @@ -1900,10 +1943,10 @@ static void do_writestub(int n) int cc=get_reg(i_regmap,CCREG); if(cc<0) emit_loadreg(CCREG,2); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d+1),2); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2); // returns new cycle_count - emit_call(handler); - emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d+1),cc<0?2:cc); + emit_far_call(handler); + emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d),cc<0?2:cc); if(cc<0) emit_storereg(CCREG,2); if(restore_jump) @@ -1912,7 +1955,8 @@ static void do_writestub(int n) emit_jmp(stubs[n].retaddr); } -static void inline_writestub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) +static void inline_writestub(enum stub_type type, int i, u_int addr, + const signed char regmap[], int target, int adj, u_int reglist) { int rs=get_reg(regmap,-1); int rt=get_reg(regmap,target); @@ -1938,38 +1982,38 @@ static void inline_writestub(enum stub_type type, int i, u_int addr, signed char int cc=get_reg(regmap,CCREG); if(cc<0) emit_loadreg(CCREG,2); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2); emit_movimm((u_int)handler,3); // returns new cycle_count - emit_call(jump_handler_write_h); - emit_addimm(0,-CLOCK_ADJUST(adj+1),cc<0?2:cc); + emit_far_call(jump_handler_write_h); + emit_addimm(0,-CLOCK_ADJUST(adj),cc<0?2:cc); if(cc<0) emit_storereg(CCREG,2); restore_regs(reglist); } // this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr -static void do_dirty_stub_emit_args(u_int arg0) +static void do_dirty_stub_emit_args(u_int arg0, u_int source_len) { #ifndef HAVE_ARMV7 emit_loadlp((int)source, 1); emit_loadlp((int)copy, 2); - emit_loadlp(slen*4, 3); + emit_loadlp(source_len, 3); #else emit_movw(((u_int)source)&0x0000FFFF, 1); emit_movw(((u_int)copy)&0x0000FFFF, 2); emit_movt(((u_int)source)&0xFFFF0000, 1); emit_movt(((u_int)copy)&0xFFFF0000, 2); - emit_movw(slen*4, 3); + emit_movw(source_len, 3); #endif emit_movimm(arg0, 0); } -static void *do_dirty_stub(int i) +static void *do_dirty_stub(int i, u_int source_len) { assem_debug("do_dirty_stub %x\n",start+i*4); - do_dirty_stub_emit_args(start + i*4); - emit_call(verify_code); + do_dirty_stub_emit_args(start + i*4, source_len); + emit_far_call(verify_code); void *entry = out; load_regs_entry(i); if (entry == out) @@ -1978,29 +2022,30 @@ static void *do_dirty_stub(int i) return entry; } -static void do_dirty_stub_ds() +static void do_dirty_stub_ds(u_int source_len) { - do_dirty_stub_emit_args(start + 1); - emit_call(verify_code_ds); + do_dirty_stub_emit_args(start + 1, source_len); + emit_far_call(verify_code_ds); } /* Special assem */ -static void c2op_prologue(u_int op,u_int reglist) +static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist) { save_regs_all(reglist); + cop2_do_stall_check(op, i, i_regs, 0); #ifdef PCNT - emit_movimm(op,0); - emit_call(pcnt_gte_start); + emit_movimm(op, 0); + emit_far_call(pcnt_gte_start); #endif - emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); // cop2 regs + emit_addimm(FP, (u_char *)&psxRegs.CP2D.r[0] - (u_char *)&dynarec_local, 0); // cop2 regs } static void c2op_epilogue(u_int op,u_int reglist) { #ifdef PCNT emit_movimm(op,0); - emit_call(pcnt_gte_end); + emit_far_call(pcnt_gte_end); #endif restore_regs_all(reglist); } @@ -2008,37 +2053,34 @@ static void c2op_epilogue(u_int op,u_int reglist) static void c2op_call_MACtoIR(int lm,int need_flags) { if(need_flags) - emit_call(lm?gteMACtoIR_lm1:gteMACtoIR_lm0); + emit_far_call(lm?gteMACtoIR_lm1:gteMACtoIR_lm0); else - emit_call(lm?gteMACtoIR_lm1_nf:gteMACtoIR_lm0_nf); + emit_far_call(lm?gteMACtoIR_lm1_nf:gteMACtoIR_lm0_nf); } static void c2op_call_rgb_func(void *func,int lm,int need_ir,int need_flags) { - emit_call(func); + emit_far_call(func); // func is C code and trashes r0 emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); if(need_flags||need_ir) c2op_call_MACtoIR(lm,need_flags); - emit_call(need_flags?gteMACtoRGB:gteMACtoRGB_nf); + emit_far_call(need_flags?gteMACtoRGB:gteMACtoRGB_nf); } -static void c2op_assemble(int i,struct regstat *i_regs) +static void c2op_assemble(int i, const struct regstat *i_regs) { - u_int c2op=source[i]&0x3f; - u_int hr,reglist_full=0,reglist; - int need_flags,need_ir; - for(hr=0;hrregmap[hr]>=0) reglist_full|=1<regmap); + u_int reglist = reglist_full & CALLER_SAVE_REGS; + int need_flags, need_ir; if (gte_handlers[c2op]!=NULL) { need_flags=!(gte_unneeded[i+1]>>63); // +1 because of how liveness detection works need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00; assem_debug("gte op %08x, unneeded %016llx, need_flags %d, need_ir %d\n", source[i],gte_unneeded[i+1],need_flags,need_ir); - if(new_dynarec_hacks&NDHACK_GTE_NO_FLAGS) + if(HACK_ENABLED(NDHACK_GTE_NO_FLAGS)) need_flags=0; int shift = (source[i] >> 19) & 1; int lm = (source[i] >> 10) & 1; @@ -2050,7 +2092,7 @@ static void c2op_assemble(int i,struct regstat *i_regs) int cv = (source[i] >> 13) & 3; int mx = (source[i] >> 17) & 3; reglist=reglist_full&(CALLER_SAVE_REGS|0xf0); // +{r4-r7} - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); /* r4,r5 = VXYZ(v) packed; r6 = &MX11(mx); r7 = &CV1(cv) */ if(v<3) emit_ldrd(v*8,0,4); @@ -2070,73 +2112,73 @@ static void c2op_assemble(int i,struct regstat *i_regs) emit_readword(&zeromem_ptr,7); #ifdef __ARM_NEON__ emit_movimm(source[i],1); // opcode - emit_call(gteMVMVA_part_neon); + emit_far_call(gteMVMVA_part_neon); if(need_flags) { emit_movimm(lm,1); - emit_call(gteMACtoIR_flags_neon); + emit_far_call(gteMACtoIR_flags_neon); } #else if(cv==3&&shift) - emit_call((int)gteMVMVA_part_cv3sh12_arm); + emit_far_call((int)gteMVMVA_part_cv3sh12_arm); else { emit_movimm(shift,1); - emit_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm)); + emit_far_call((int)(need_flags?gteMVMVA_part_arm:gteMVMVA_part_nf_arm)); } if(need_flags||need_ir) c2op_call_MACtoIR(lm,need_flags); #endif #else /* if not HAVE_ARMV5 */ - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); emit_movimm(source[i],1); // opcode emit_writeword(1,&psxRegs.code); - emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op])); + emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]); #endif break; } case GTE_OP: - c2op_prologue(c2op,reglist); - emit_call(shift?gteOP_part_shift:gteOP_part_noshift); + c2op_prologue(c2op,i,i_regs,reglist); + emit_far_call(shift?gteOP_part_shift:gteOP_part_noshift); if(need_flags||need_ir) { emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); c2op_call_MACtoIR(lm,need_flags); } break; case GTE_DPCS: - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); c2op_call_rgb_func(shift?gteDPCS_part_shift:gteDPCS_part_noshift,lm,need_ir,need_flags); break; case GTE_INTPL: - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); c2op_call_rgb_func(shift?gteINTPL_part_shift:gteINTPL_part_noshift,lm,need_ir,need_flags); break; case GTE_SQR: - c2op_prologue(c2op,reglist); - emit_call(shift?gteSQR_part_shift:gteSQR_part_noshift); + c2op_prologue(c2op,i,i_regs,reglist); + emit_far_call(shift?gteSQR_part_shift:gteSQR_part_noshift); if(need_flags||need_ir) { emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); c2op_call_MACtoIR(lm,need_flags); } break; case GTE_DCPL: - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); c2op_call_rgb_func(gteDCPL_part,lm,need_ir,need_flags); break; case GTE_GPF: - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); c2op_call_rgb_func(shift?gteGPF_part_shift:gteGPF_part_noshift,lm,need_ir,need_flags); break; case GTE_GPL: - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); c2op_call_rgb_func(shift?gteGPL_part_shift:gteGPL_part_noshift,lm,need_ir,need_flags); break; #endif default: - c2op_prologue(c2op,reglist); + c2op_prologue(c2op,i,i_regs,reglist); #ifdef DRC_DBG emit_movimm(source[i],1); // opcode emit_writeword(1,&psxRegs.code); #endif - emit_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]); + emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]); break; } c2op_epilogue(c2op,reglist); @@ -2192,14 +2234,14 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs) // case 0x1D: DMULTU // case 0x1E: DDIV // case 0x1F: DDIVU - if(rs1[i]&&rs2[i]) + if(dops[i].rs1&&dops[i].rs2) { - if((opcode2[i]&4)==0) // 32-bit + if((dops[i].opcode2&4)==0) // 32-bit { - if(opcode2[i]==0x18) // MULT + if(dops[i].opcode2==0x18) // MULT { - signed char m1=get_reg(i_regs->regmap,rs1[i]); - signed char m2=get_reg(i_regs->regmap,rs2[i]); + signed char m1=get_reg(i_regs->regmap,dops[i].rs1); + signed char m2=get_reg(i_regs->regmap,dops[i].rs2); signed char hi=get_reg(i_regs->regmap,HIREG); signed char lo=get_reg(i_regs->regmap,LOREG); assert(m1>=0); @@ -2208,10 +2250,10 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs) assert(lo>=0); emit_smull(m1,m2,hi,lo); } - if(opcode2[i]==0x19) // MULTU + if(dops[i].opcode2==0x19) // MULTU { - signed char m1=get_reg(i_regs->regmap,rs1[i]); - signed char m2=get_reg(i_regs->regmap,rs2[i]); + signed char m1=get_reg(i_regs->regmap,dops[i].rs1); + signed char m2=get_reg(i_regs->regmap,dops[i].rs2); signed char hi=get_reg(i_regs->regmap,HIREG); signed char lo=get_reg(i_regs->regmap,LOREG); assert(m1>=0); @@ -2220,10 +2262,10 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs) assert(lo>=0); emit_umull(m1,m2,hi,lo); } - if(opcode2[i]==0x1A) // DIV + if(dops[i].opcode2==0x1A) // DIV { - signed char d1=get_reg(i_regs->regmap,rs1[i]); - signed char d2=get_reg(i_regs->regmap,rs2[i]); + signed char d1=get_reg(i_regs->regmap,dops[i].rs1); + signed char d2=get_reg(i_regs->regmap,dops[i].rs2); assert(d1>=0); assert(d2>=0); signed char quotient=get_reg(i_regs->regmap,LOREG); @@ -2258,10 +2300,10 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs) emit_test(d1,d1); emit_negmi(remainder,remainder); } - if(opcode2[i]==0x1B) // DIVU + if(dops[i].opcode2==0x1B) // DIVU { - signed char d1=get_reg(i_regs->regmap,rs1[i]); // dividend - signed char d2=get_reg(i_regs->regmap,rs2[i]); // divisor + signed char d1=get_reg(i_regs->regmap,dops[i].rs1); // dividend + signed char d2=get_reg(i_regs->regmap,dops[i].rs2); // divisor assert(d1>=0); assert(d2>=0); signed char quotient=get_reg(i_regs->regmap,LOREG); @@ -2309,7 +2351,7 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs) static void do_jump_vaddr(int rs) { - emit_jmp(jump_vaddr_reg[rs]); + emit_far_jump(jump_vaddr_reg[rs]); } static void do_preload_rhash(int r) { @@ -2358,51 +2400,18 @@ static void do_miniht_insert(u_int return_address,int rt,int temp) { #endif } -static void mark_clear_cache(void *target) -{ - u_long offset = (u_char *)target - translation_cache; - u_int mask = 1u << ((offset >> 12) & 31); - if (!(needs_clear_cache[offset >> 17] & mask)) { - char *start = (char *)((u_long)target & ~4095ul); - start_tcache_write(start, start + 4096); - needs_clear_cache[offset >> 17] |= mask; - } -} - -// 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. -static 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_char *start, *end; - for(j=0;j<32;j++) - { - if(bitmap&(1<tramp.f - (u_char *)&ndrc->tramp.ops - 8; + struct tramp_insns *ops = ndrc->tramp.ops; + size_t i; + assert(!(diff & 3)); + assert(diff < 0x1000); + start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops)); + for (i = 0; i < ARRAY_SIZE(ndrc->tramp.ops); i++) + ops[i].ldrpc = 0xe5900000 | rd_rn_rm(15,15,0) | diff; // ldr pc, [=val] + end_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops)); } // vim:shiftwidth=2:expandtab