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=c2f65ee6df827e63548d1b12d155eb02e54fd805;hp=1a4324dc98f9c7c6c9a1aa0680e3dbae784b8d7a;hb=71e490c5930e6e5f71d1f2d5165c3a801ac46be1;hpb=790ee18e679f9d6f69c5fa12c09a70720319df8d diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c index 1a4324dc..c2f65ee6 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 * @@ -18,15 +19,31 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ +#include "../gte.h" +#define FLAGLESS +#include "../gte.h" +#undef FLAGLESS +#include "../gte_arm.h" +#include "../gte_neon.h" +#include "pcnt.h" +#include "arm_features.h" + +#if !BASE_ADDR_FIXED +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 + extern int cycle_count; extern int last_count; extern int pcaddr; extern int pending_exception; extern int branch_target; extern uint64_t readmem_dword; -#ifdef MUPEN64 -extern precomp_instr fake_pc; -#endif extern void *dynarec_local; extern u_int memory_map[1048576]; extern u_int mini_ht[32][2]; @@ -66,7 +83,38 @@ const u_int jump_vaddr_reg[16] = { 0, 0}; -#include "fpu.h" +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}; + +unsigned int needs_clear_cache[1<<(TARGET_SIZE_2-17)]; /* Linker */ @@ -137,6 +185,7 @@ void set_jump_target_fillslot(int addr,u_int target,int copy) /* Literal pool */ add_literal(int addr,int val) { + assert(literalcount>6)+8; // get target of bl - if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) { - unsigned int page=source>>12; - unsigned int map_value=memory_map[page]; - if(map_value>=0x80000000) return 0; - while(page<((source+len-1)>>12)) { - if((memory_map[++page]<<2)!=(map_value<<2)) return 0; - } - source = source+(map_value<<2); - } //printf("verify_dirty: %x %x %x\n",source,copy,len); return !memcmp((void *)source,(void *)copy,len); } @@ -224,7 +267,7 @@ int verify_dirty(int addr) // guarantees that it's not dirty int isclean(int addr) { - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 int *ptr=((u_int *)addr)+4; #else int *ptr=((u_int *)addr)+6; @@ -237,12 +280,13 @@ int isclean(int addr) return 1; } +// get source that block at addr was compiled from (host pointers) void get_bounds(int addr,u_int *start,u_int *end) { u_int *ptr=(u_int *)addr; - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 // get from literal pool - assert((*ptr&0xFFF00000)==0xe5900000); + assert((*ptr&0xFFFF0000)==0xe59f0000); u_int offset=*ptr&0xfff; u_int *l_ptr=(void *)ptr+offset+8; u_int source=l_ptr[0]; @@ -259,11 +303,6 @@ void get_bounds(int addr,u_int *start,u_int *end) #endif if((*ptr&0xFF000000)!=0xeb000000) ptr++; assert((*ptr&0xFF000000)==0xeb000000); // bl instruction - u_int verifier=(int)ptr+((signed int)(*ptr<<8)>>6)+8; // get target of bl - if(verifier==(u_int)verify_code_vm||verifier==(u_int)verify_code_ds) { - if(memory_map[source>>12]>=0x80000000) source = 0; - else source = source+(memory_map[source>>12]<<2); - } *start=source; *end=source+len; } @@ -434,7 +473,7 @@ void alloc_reg(struct regstat *cur,int i,signed char reg) } } } - printf("This shouldn't happen (alloc_reg)");exit(1); + SysPrintf("This shouldn't happen (alloc_reg)");exit(1); } void alloc_reg64(struct regstat *cur,int i,signed char reg) @@ -600,7 +639,7 @@ void alloc_reg64(struct regstat *cur,int i,signed char reg) } } } - printf("This shouldn't happen");exit(1); + SysPrintf("This shouldn't happen");exit(1); } // Allocate a temporary register. This is done without regard to @@ -723,21 +762,26 @@ void alloc_reg_temp(struct regstat *cur,int i,signed char reg) } } } - printf("This shouldn't happen");exit(1); + SysPrintf("This shouldn't happen");exit(1); } // Allocate a specific ARM register. void alloc_arm_reg(struct regstat *cur,int i,signed char reg,char hr) { int n; + int dirty=0; // see if it's already allocated (and dealloc it) for(n=0;nregmap[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) { @@ -834,7 +879,7 @@ u_int genjmp(u_int addr) int offset=addr-(int)out-8; if(offset<-33554432||offset>=33554432) { if (addr>2) { - printf("genjmp: out of range: %08x\n", offset); + SysPrintf("genjmp: out of range: %08x\n", offset); exit(1); } return 0; @@ -941,7 +986,7 @@ void emit_movimm(u_int imm,u_int rt) 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 + #ifndef HAVE_ARMV7 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); @@ -950,7 +995,7 @@ void emit_movimm(u_int imm,u_int rt) emit_movw(imm,rt); #endif }else{ - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 emit_loadlp(imm,rt); #else emit_movw(imm&0x0000FFFF,rt); @@ -966,12 +1011,11 @@ void emit_pcreladdr(u_int rt) void emit_loadreg(int r, int hr) { -#ifdef FORCE32 if(r&64) { - printf("64bit load in 32bit mode!\n"); - exit(1); + SysPrintf("64bit load in 32bit mode!\n"); + assert(0); + return; } -#endif if((r&63)==0) emit_zeroreg(hr); else { @@ -990,12 +1034,11 @@ void emit_loadreg(int r, int hr) } void emit_storereg(int r, int hr) { -#ifdef FORCE32 if(r&64) { - printf("64bit store in 32bit mode!\n"); - exit(1); + SysPrintf("64bit store in 32bit mode!\n"); + assert(0); + return; } -#endif int addr=((int)reg)+((r&63)<>4); if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4); if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4); @@ -1016,7 +1059,7 @@ 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); } @@ -1087,24 +1130,37 @@ 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); output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval); }else if(genimm(-imm,&armval)) { - assem_debug("sub %s,%s,#%d\n",regname[rt],regname[rs],imm); + 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) { + #ifdef HAVE_ARMV7 + }else if(rt!=rs&&(u_int)imm<65536) { + emit_movw(imm&0x0000ffff,rt); + emit_add(rs,rt,rt); + }else if(rt!=rs&&(u_int)-imm<65536) { + emit_movw(-imm&0x0000ffff,rt); + emit_sub(rs,rt,rt); + #endif + }else if((u_int)-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{ - 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)); - output_w32(0xe2800000|rd_rn_imm_shift(rt,rt,imm&0xff,0)); + }else { + do { + int shift = (ffs(imm) - 1) & ~1; + int imm8 = imm & (0xff << shift); + genimm_checked(imm8,&armval); + assem_debug("add %s,%s,#0x%x\n",regname[rt],regname[rs],imm8); + output_w32(0xe2800000|rd_rn_rm(rt,rs,0)|armval); + rs = rt; + imm &= ~imm8; + } + while (imm != 0); } } else if(rs!=rt) emit_mov(rs,rt); @@ -1211,7 +1267,7 @@ void emit_andimm(int rs,int imm,int rt) assem_debug("bic %s,%s,#%d\n",regname[rt],regname[rs],imm); output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|armval); }else if(imm==65535) { - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV6 assem_debug("bic %s,%s,#FF000000\n",regname[rt],regname[rs]); output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|0x4FF); assem_debug("bic %s,%s,#00FF0000\n",regname[rt],regname[rt]); @@ -1222,7 +1278,7 @@ void emit_andimm(int rs,int imm,int rt) #endif }else{ assert(imm>0&&imm<65535); - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 assem_debug("mov r14,#%d\n",imm&0xFF00); output_w32(0xe3a00000|rd_rn_imm_shift(HOST_TEMPREG,0,imm>>8,8)); assem_debug("add r14,r14,#%d\n",imm&0xFF); @@ -1278,6 +1334,22 @@ 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_lslpls_imm(int rs,int imm,int rt) +{ + assert(imm>0); + assert(imm<32); + assem_debug("lslpls %s,%s,#%d\n",regname[rt],regname[rs],imm); + output_w32(0x51b00000|rd_rn_rm(rt,0,rs)|(imm<<7)); +} + void emit_shrimm(int rs,u_int imm,int rt) { assert(imm>0); @@ -1328,7 +1400,7 @@ void emit_shrdimm(int rs,int rs2,u_int imm,int rt) void emit_signextend16(int rs,int rt) { - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV6 emit_shlimm(rs,16,rt); emit_sarimm(rt,16,rt); #else @@ -1337,6 +1409,17 @@ void emit_signextend16(int rs,int rt) #endif } +void emit_signextend8(int rs,int rt) +{ + #ifndef HAVE_ARMV6 + 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); @@ -1409,27 +1492,19 @@ 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); - #ifdef ARMv5_ONLY emit_movimm(imm,HOST_TEMPREG); - #else - emit_movw(imm,HOST_TEMPREG); - #endif assem_debug("cmp %s,r14\n",regname[rs]); output_w32(0xe1500000|rd_rn_rm(0,rs,HOST_TEMPREG)); }else{ assert(imm>-65536); - #ifdef ARMv5_ONLY emit_movimm(-imm,HOST_TEMPREG); - #else - emit_movw(-imm,HOST_TEMPREG); - #endif assem_debug("cmn %s,r14\n",regname[rs]); output_w32(0xe1700000|rd_rn_rm(0,rs,HOST_TEMPREG)); } @@ -1710,8 +1785,9 @@ 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) { @@ -1734,6 +1810,31 @@ 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); @@ -1827,6 +1928,16 @@ void emit_movzwl_indexed(int offset, int rs, int rt) output_w32(0xe15000b0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); } } +static void emit_ldrd(int offset, int rs, int rt) +{ + assert(offset>-256&&offset<256); + assem_debug("ldrd %s,%s+%d\n",regname[rt],regname[rs],offset); + if(offset>=0) { + output_w32(0xe1c000d0|rd_rn_rm(rt,rs,0)|((offset<<4)&0xf00)|(offset&0xf)); + }else{ + output_w32(0xe14000d0|rd_rn_rm(rt,rs,0)|(((-offset)<<4)&0xf00)|((-offset)&0xf)); + } +} void emit_readword(int addr, int rt) { u_int offset = addr-(u_int)&dynarec_local; @@ -1949,6 +2060,21 @@ 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; @@ -2046,6 +2172,14 @@ void emit_shrcc_imm(int rs,u_int imm,int rt) output_w32(0x31a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7)); } +void emit_shrne_imm(int rs,u_int imm,int rt) +{ + assert(imm>0); + assert(imm<32); + assem_debug("lsrne %s,%s,#%d\n",regname[rt],regname[rs],imm); + output_w32(0x11a00000|rd_rn_rm(rt,0,rs)|0x20|(imm<<7)); +} + void emit_negmi(int rs, int rt) { assem_debug("rsbmi %s,%s,#0\n",regname[rt],regname[rs]); @@ -2150,7 +2284,7 @@ void emit_cmov2imm_e_ne_compact(int imm1,int imm2,u_int rt) output_w32(0x12400000|rd_rn_rm(rt,rt,0)|armval); } else { - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 emit_movimm(imm1,rt); add_literal((int)out,imm2); assem_debug("ldrne %s,pc+? [=%x]\n",regname[rt],imm2); @@ -2193,6 +2327,13 @@ 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) { @@ -2434,6 +2575,14 @@ void emit_andne_imm(int rs,int imm,int rt) output_w32(0x12000000|rd_rn_rm(rt,rs,0)|armval); } +void emit_addpl_imm(int rs,int imm,int rt) +{ + u_int armval; + genimm_checked(imm,&armval); + assem_debug("addpl %s,%s,#%d\n",regname[rt],regname[rs],imm); + output_w32(0x52800000|rd_rn_rm(rt,rs,0)|armval); +} + void emit_jno_unlikely(int a) { //emit_jno(a); @@ -2441,34 +2590,40 @@ void emit_jno_unlikely(int a) output_w32(0x72800000|rd_rn_rm(15,15,0)); } -// Save registers before function call -void save_regs(u_int reglist) +static void save_regs_all(u_int reglist) { - reglist&=0x100f; // only save the caller-save registers, r0-r3, r12 + int i; if(!reglist) return; assem_debug("stmia fp,{"); - if(reglist&1) assem_debug("r0, "); - if(reglist&2) assem_debug("r1, "); - if(reglist&4) assem_debug("r2, "); - if(reglist&8) assem_debug("r3, "); - if(reglist&0x1000) assem_debug("r12"); + for(i=0;i<16;i++) + if(reglist&(1<>i_regmap[hr])&1) { - if(value!=-1&&value!=0) emit_sarimm(HOST_TEMPREG,31,HOST_TEMPREG); - emit_storereg(i_regmap[hr]|64,HOST_TEMPREG); - } -#endif } } } @@ -2510,12 +2659,24 @@ void literal_pool(int n) int i; for(i=0;i>=2) + ; + if(xs<0x100) return 1; + for(xs=-diff;xs!=0&&(xs&3)==0;xs>>=2) + ; + if(xs<0x100) return 1; + return 0; +} + +// 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); + } +} + +static void mov_loadtype_adj(int type,int rs,int rt) +{ + switch(type) { + case LOADB_STUB: emit_signextend8(rs,rt); break; + case LOADBU_STUB: emit_andimm(rs,0xff,rt); break; + case LOADH_STUB: emit_signextend16(rs,rt); break; + case LOADHU_STUB: emit_andimm(rs,0xffff,rt); break; + case LOADW_STUB: if(rs!=rt) emit_mov(rs,rt); break; + default: assert(0); + } +} + +#include "pcsxmem.h" +#include "pcsxmem_inline.c" + do_readstub(int n) { assem_debug("do_readstub %x\n",start+stubs[n][3]*4); @@ -2584,80 +2821,92 @@ do_readstub(int n) rt=get_reg(i_regmap,rt1[i]); } assert(rs>=0); - if(addr<0) addr=rt; - if(addr<0) - // assume dummy read, no alloced reg - addr=get_reg(i_regmap,-1); - assert(addr>=0); - int ftable=0; + int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0,restore_jump=0; + reglist|=(1<=0&&rt1[i]!=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) - ftable=(int)readmemb; + handler=(int)jump_handler_read8; if(type==LOADH_STUB||type==LOADHU_STUB) - ftable=(int)readmemh; + handler=(int)jump_handler_read16; if(type==LOADW_STUB) - ftable=(int)readmem; -#ifndef FORCE32 - if(type==LOADD_STUB) - ftable=(int)readmemd; -#endif - assert(ftable!=0); - emit_writeword(rs,(int)&address); - //emit_pusha(); - save_regs(reglist); - 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<=0&&rt1[i]!=0)) { + mov_loadtype_adj(type,0,rt); } - emit_movimm(ftable,0); - emit_addimm(cc<0?2:cc,2*stubs[n][6]+2,2); - emit_movimm(start+stubs[n][3]*4+(((regs[i].was32>>rs1[i])&1)<<1)+ds,3); - //emit_readword((int)&last_count,temp); - //emit_add(cc,temp,cc); - //emit_writeword(cc,(int)&Count); - //emit_mov(15,14); - emit_call((int)&indirect_jump_indexed); - //emit_callreg(rs); - //emit_readword_dualindexedx4(rs,HOST_TEMPREG,15); - // We really shouldn't need to update the count here, - // but not doing so causes random crashes... - emit_readword((int)&Count,HOST_TEMPREG); - emit_readword((int)&next_interupt,2); - emit_addimm(HOST_TEMPREG,-2*stubs[n][6]-2,HOST_TEMPREG); - emit_writeword(2,(int)&last_count); - emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc); - if(cc<0) { - emit_storereg(CCREG,HOST_TEMPREG); - } - //emit_popa(); + if(restore_jump) + set_jump_target(restore_jump,(int)out); restore_regs(reglist); - //if((cc=get_reg(regmap,CCREG))>=0) { - // emit_loadreg(CCREG,cc); - //} - if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) { - assert(rt>=0); - if(type==LOADB_STUB) - emit_movsbl((int)&readmem_dword,rt); - if(type==LOADBU_STUB) - emit_movzbl((int)&readmem_dword,rt); - if(type==LOADH_STUB) - emit_movswl((int)&readmem_dword,rt); - if(type==LOADHU_STUB) - emit_movzwl((int)&readmem_dword,rt); - if(type==LOADW_STUB) - emit_readword((int)&readmem_dword,rt); - if(type==LOADD_STUB) { - emit_readword((int)&readmem_dword,rt); - if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth); + emit_jmp(stubs[n][2]); // return address +} + +// 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; } - emit_jmp(stubs[n][2]); // return address } inline_readstub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) @@ -2665,74 +2914,81 @@ 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) - ftable=(int)readmemb; - if(type==LOADH_STUB||type==LOADHU_STUB) - ftable=(int)readmemh; - if(type==LOADW_STUB) - ftable=(int)readmem; -#ifndef FORCE32 - if(type==LOADD_STUB) - ftable=(int)readmemd; -#endif - assert(ftable!=0); - if(target==0) - emit_movimm(addr,rs); - emit_writeword(rs,(int)&address); - //emit_pusha(); - save_regs(reglist); - //emit_shrimm(rs,16,1); + u_int handler,host_addr=0,is_dynamic,far_call=0; int cc=get_reg(regmap,CCREG); - if(cc<0) { + if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt)) + return; + handler=get_direct_memhandler(mem_rtab,addr,type,&host_addr); + if (handler==0) { + if(rt<0||rt1[i]==0) + return; + if(addr!=host_addr) + emit_movimm_from(addr,rs,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; + } + is_dynamic=pcsxmem_is_handler_dynamic(addr); + if(is_dynamic) { + 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; + } + + // call a memhandler + if(rt>=0&&rt1[i]!=0) + reglist&=~(1<=33554432) { + // unreachable memhandler, a plugin func perhaps + emit_movimm(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_movimm(ftable,0); - 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); - 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); - } - //emit_add(12,2,2); - //emit_writeword(2,(int)&Count); - //emit_call(((u_int *)ftable)[addr>>16]); - emit_call((int)&indirect_jump); - // We really shouldn't need to update the count here, - // but not doing so causes random crashes... - emit_readword((int)&Count,HOST_TEMPREG); - emit_readword((int)&next_interupt,2); - emit_addimm(HOST_TEMPREG,-CLOCK_DIVIDER*(adj+1),HOST_TEMPREG); - emit_writeword(2,(int)&last_count); - emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc); - if(cc<0) { - emit_storereg(CCREG,HOST_TEMPREG); + else { + emit_readword((int)&last_count,3); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); + emit_add(2,3,2); + emit_writeword(2,(int)&Count); } - //emit_popa(); - restore_regs(reglist); - if(rt>=0) { - if(type==LOADB_STUB) - emit_movsbl((int)&readmem_dword,rt); - if(type==LOADBU_STUB) - emit_movzbl((int)&readmem_dword,rt); - if(type==LOADH_STUB) - emit_movswl((int)&readmem_dword,rt); - if(type==LOADHU_STUB) - emit_movzwl((int)&readmem_dword,rt); - if(type==LOADW_STUB) - emit_readword((int)&readmem_dword,rt); - if(type==LOADD_STUB) { - emit_readword((int)&readmem_dword,rt); - if(rth>=0) emit_readword(((int)&readmem_dword)+4,rth); + + if(far_call) + emit_callreg(12); + else + emit_call(handler); + + if(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; + default: assert(0); } } + restore_regs(reglist); } do_writestub(int n) @@ -2758,73 +3014,65 @@ do_writestub(int n) } assert(rs>=0); assert(rt>=0); - if(addr<0) addr=get_reg(i_regmap,-1); - assert(addr>=0); - int ftable=0; - if(type==STOREB_STUB) - ftable=(int)writememb; - if(type==STOREH_STUB) - ftable=(int)writememh; - if(type==STOREW_STUB) - ftable=(int)writemem; -#ifndef FORCE32 - if(type==STORED_STUB) - ftable=(int)writememd; -#endif - assert(ftable!=0); - emit_writeword(rs,(int)&address); - //emit_shrimm(rs,16,rs); - //emit_movmem_indexedx4(ftable,rs,rs); - if(type==STOREB_STUB) - emit_writebyte(rt,(int)&byte); - if(type==STOREH_STUB) - emit_writehword(rt,(int)&hword); - if(type==STOREW_STUB) - emit_writeword(rt,(int)&word); - if(type==STORED_STUB) { -#ifndef FORCE32 - emit_writeword(rt,(int)&dword); - emit_writeword(r?rth:rt,(int)&dword+4); -#else - printf("STORED_STUB\n"); -#endif + int rtmp,temp=-1,temp2=HOST_TEMPREG,regs_saved=0,restore_jump=0,ra; + int reglist2=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); - //emit_readword((int)&last_count,temp); - //emit_addimm(cc,2*stubs[n][5]+2,cc); - //emit_add(cc,temp,cc); - //emit_writeword(cc,(int)&Count); - emit_call((int)&indirect_jump_indexed); - //emit_callreg(rs); - emit_readword((int)&Count,HOST_TEMPREG); - emit_readword((int)&next_interupt,2); - emit_addimm(HOST_TEMPREG,-2*stubs[n][6]-2,HOST_TEMPREG); - emit_writeword(2,(int)&last_count); - emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc); - if(cc<0) { - emit_storereg(CCREG,HOST_TEMPREG); - } - //emit_popa(); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n][6]+1),2); + // returns new cycle_count + emit_call(handler); + emit_addimm(0,-CLOCK_ADJUST((int)stubs[n][6]+1),cc<0?2:cc); + if(cc<0) + emit_storereg(CCREG,2); + if(restore_jump) + set_jump_target(restore_jump,(int)out); restore_regs(reglist); - //if((cc=get_reg(regmap,CCREG))>=0) { - // emit_loadreg(CCREG,cc); - //} - emit_jmp(stubs[n][2]); // return address + ra=stubs[n][2]; + emit_jmp(ra); } inline_writestub(int type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) @@ -2834,64 +3082,33 @@ 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); - int ftable=0; - if(type==STOREB_STUB) - ftable=(int)writememb; - if(type==STOREH_STUB) - ftable=(int)writememh; - if(type==STOREW_STUB) - ftable=(int)writemem; -#ifndef FORCE32 - if(type==STORED_STUB) - ftable=(int)writememd; -#endif - assert(ftable!=0); - emit_writeword(rs,(int)&address); - //emit_shrimm(rs,16,rs); - //emit_movmem_indexedx4(ftable,rs,rs); - if(type==STOREB_STUB) - emit_writebyte(rt,(int)&byte); - if(type==STOREH_STUB) - emit_writehword(rt,(int)&hword); - if(type==STOREW_STUB) - emit_writeword(rt,(int)&word); - if(type==STORED_STUB) { -#ifndef FORCE32 - emit_writeword(rt,(int)&dword); - emit_writeword(target?rth:rt,(int)&dword+4); -#else - printf("STORED_STUB\n"); -#endif + u_int handler,host_addr=0; + handler=get_direct_memhandler(mem_wtab,addr,type,&host_addr); + if (handler==0) { + if(addr!=host_addr) + emit_movimm_from(addr,rs,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; } - //emit_pusha(); + + // call a memhandler save_regs(reglist); - //emit_shrimm(rs,16,1); + pass_args(rs,rt); int cc=get_reg(regmap,CCREG); - if(cc<0) { + if(cc<0) emit_loadreg(CCREG,2); - } - //emit_movimm(ftable,0); - 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); - 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); - } - //emit_add(12,2,2); - //emit_writeword(2,(int)&Count); - //emit_call(((u_int *)ftable)[addr>>16]); - emit_call((int)&indirect_jump); - emit_readword((int)&Count,HOST_TEMPREG); - emit_readword((int)&next_interupt,2); - emit_addimm(HOST_TEMPREG,-CLOCK_DIVIDER*(adj+1),HOST_TEMPREG); - emit_writeword(2,(int)&last_count); - emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc); - if(cc<0) { - emit_storereg(CCREG,HOST_TEMPREG); - } - //emit_popa(); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2); + emit_movimm(handler,3); + // returns new cycle_count + emit_call((int)jump_handler_write_h); + emit_addimm(0,-CLOCK_ADJUST(adj+1),cc<0?2:cc); + if(cc<0) + emit_storereg(CCREG,2); restore_regs(reglist); } @@ -2916,16 +3133,25 @@ 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? emit_call((int)&indirect_jump_indexed); restore_regs(reglist); @@ -2973,6 +3198,7 @@ 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) @@ -2995,12 +3221,9 @@ do_invstub(int n) int do_dirty_stub(int i) { assem_debug("do_dirty_stub %x\n",start+i*4); - u_int addr=(int)start<(int)0xC0000000?(u_int)source:(u_int)start; - #ifdef PCSX - addr=(u_int)source; - #endif + u_int addr=(u_int)source; // Careful about the code output here, verify_dirty needs to parse it. - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 emit_loadlp(addr,1); emit_loadlp((int)copy,2); emit_loadlp(slen*4,3); @@ -3023,7 +3246,7 @@ int do_dirty_stub(int i) void do_dirty_stub_ds() { // Careful about the code output here, verify_dirty needs to parse it. - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 emit_loadlp((int)start<(int)0xC0000000?(int)source:(int)start,1); emit_loadlp((int)copy,2); emit_loadlp(slen*4,3); @@ -3055,96 +3278,20 @@ do_cop1stub(int n) wb_dirtys(i_regs->regmap_entry,i_regs->was32,i_regs->wasdirty); if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); emit_movimm(start+(i-ds)*4,EAX); // Get PC - emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle... + emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right? There should probably be an extra cycle... emit_jmp(ds?(int)fp_exception_ds:(int)fp_exception); } -/* TLB */ - -int do_tlb_r(int s,int ar,int map,int x,int a,int shift,int c,u_int addr) -{ - if(c) { - if((signed int)addr>=(signed int)0xC0000000) { - // address_generation already loaded the const - emit_readword_dualindexedx4(FP,map,map); - } - else - return -1; // No mapping - } - else { - assert(s!=map); - emit_movimm(((int)memory_map-(int)&dynarec_local)>>2,map); - emit_addsr12(map,s,map); - // Schedule this while we wait on the load - //if(x) emit_xorimm(s,x,ar); - if(shift>=0) emit_shlimm(s,3,shift); - if(~a) emit_andimm(s,a,ar); - emit_readword_dualindexedx4(FP,map,map); - } - return map; -} -int do_tlb_r_branch(int map, int c, u_int addr, int *jaddr) -{ - if(!c||(signed int)addr>=(signed int)0xC0000000) { - emit_test(map,map); - *jaddr=(int)out; - emit_js(0); - } - return map; -} +static int do_tlb_r(int a, ...) { return 0; } +static int do_tlb_r_branch(int a, ...) { return 0; } +static int gen_tlb_addr_r(int a, ...) { return 0; } +static int do_tlb_w(int a, ...) { return 0; } +static int do_tlb_w_branch(int a, ...) { return 0; } +static int gen_tlb_addr_w(int a, ...) { return 0; } -int gen_tlb_addr_r(int ar, int map) { - if(map>=0) { - assem_debug("add %s,%s,%s lsl #2\n",regname[ar],regname[ar],regname[map]); - output_w32(0xe0800100|rd_rn_rm(ar,ar,map)); - } -} +/* Special assem */ -int do_tlb_w(int s,int ar,int map,int x,int c,u_int addr) -{ - if(c) { - if(addr<0x80800000||addr>=0xC0000000) { - // address_generation already loaded the const - emit_readword_dualindexedx4(FP,map,map); - } - else - return -1; // No mapping - } - else { - assert(s!=map); - emit_movimm(((int)memory_map-(int)&dynarec_local)>>2,map); - emit_addsr12(map,s,map); - // Schedule this while we wait on the load - //if(x) emit_xorimm(s,x,ar); - emit_readword_dualindexedx4(FP,map,map); - } - return map; -} -int do_tlb_w_branch(int map, int c, u_int addr, int *jaddr) -{ - if(!c||addr<0x80800000||addr>=0xC0000000) { - emit_testimm(map,0x40000000); - *jaddr=(int)out; - emit_jne(0); - } -} - -int gen_tlb_addr_w(int ar, int map) { - if(map>=0) { - assem_debug("add %s,%s,%s lsl #2\n",regname[ar],regname[ar],regname[map]); - output_w32(0xe0800100|rd_rn_rm(ar,ar,map)); - } -} - -// Generate the address of the memory_map entry, relative to dynarec_local -generate_map_const(u_int addr,int reg) { - //printf("generate_map_const(%x,%s)\n",addr,regname[reg]); - emit_movimm((addr>>12)+(((u_int)memory_map-(u_int)&dynarec_local)>>2),reg); -} - -/* Special assem */ - -void shift_assemble_arm(int i,struct regstat *i_regs) +void shift_assemble_arm(int i,struct regstat *i_regs) { if(rt1[i]) { if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV @@ -3253,6 +3400,187 @@ void shift_assemble_arm(int i,struct regstat *i_regs) } } } + +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; +} + +static int emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override) +{ + int jaddr,type=0; + 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 + if (psxH == (void *)0x1f800000) { + emit_addimm(addr,-0x1f800000,HOST_TEMPREG); + emit_cmpimm(HOST_TEMPREG,0x1000); + jaddr=(int)out; + emit_jc(0); + } + else { + // do usual RAM check, jump will go to the right handler + type=0; + } + } + + 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); + if(ram_offset!=0) { + emit_addimm(addr,ram_offset,HOST_TEMPREG); + addr=*addr_reg_override=HOST_TEMPREG; + } + } + + return jaddr; +} + #define shift_assemble shift_assemble_arm void loadlr_assemble_arm(int i,struct regstat *i_regs) @@ -3260,7 +3588,8 @@ 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]); @@ -3278,60 +3607,68 @@ 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); + 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{ + emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR } - 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 - } + jaddr=emit_fastpath_cmp_jump(i,temp2,&fastload_reg_override); + } + else { + if(ram_offset&&memtarget) { + emit_addimm(temp2,ram_offset,HOST_TEMPREG); + fastload_reg_override=HOST_TEMPREG; } - }else{ // using tlb - int a; - if(c) { - a=-1; - }else if (opcode[i]==0x22||opcode[i]==0x26) { - a=0xFFFFFFFC; // LWL/LWR + if (opcode[i]==0x22||opcode[i]==0x26) { + emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR }else{ - a=0xFFFFFFF8; // LDL/LDR + emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // 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 - } - } - 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); + } + } + 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 @@ -3348,19 +3685,23 @@ 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 @@ -3401,23 +3742,7 @@ void cop0_assemble(int i,struct regstat *i_regs) char copr=(source[i]>>11)&0x1f; //assert(t>=0); // Why does this happen? OOT is weird if(t>=0&&rt1[i]!=0) { -#ifdef MUPEN64 - emit_addimm(FP,(int)&fake_pc-(int)&dynarec_local,0); - emit_movimm((source[i]>>11)&0x1f,1); - emit_writeword(0,(int)&PC); - emit_writebyte(1,(int)&(fake_pc.f.r.nrd)); - if(copr==9) { - emit_readword((int)&last_count,ECX); - emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc - emit_add(HOST_CCREG,ECX,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); - emit_writeword(HOST_CCREG,(int)&Count); - } - emit_call((int)MFC0); - emit_readword((int)&readmem_dword,t); -#else emit_readword((int)®_cop0+copr*4,t); -#endif } } else if(opcode2[i]==4) // MTC0 @@ -3425,19 +3750,12 @@ void cop0_assemble(int i,struct regstat *i_regs) signed char s=get_reg(i_regs->regmap,rs1[i]); char copr=(source[i]>>11)&0x1f; assert(s>=0); - emit_writeword(s,(int)&readmem_dword); wb_register(rs1[i],i_regs->regmap,i_regs->dirty,i_regs->is32); -#ifdef MUPEN64 - emit_addimm(FP,(int)&fake_pc-(int)&dynarec_local,0); - emit_movimm((source[i]>>11)&0x1f,1); - emit_writeword(0,(int)&PC); - emit_writebyte(1,(int)&(fake_pc.f.r.nrd)); -#endif if(copr==9||copr==11||copr==12||copr==13) { - emit_readword((int)&last_count,ECX); + emit_readword((int)&last_count,HOST_TEMPREG); emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc - emit_add(HOST_CCREG,ECX,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); + emit_add(HOST_CCREG,HOST_TEMPREG,HOST_CCREG); + emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); emit_writeword(HOST_CCREG,(int)&Count); } // What a mess. The status register (12) can enable interrupts, @@ -3445,7 +3763,6 @@ void cop0_assemble(int i,struct regstat *i_regs) // The interrupt must be taken immediately, because a subsequent // instruction might disable interrupts again. if(copr==12||copr==13) { -#ifdef PCSX if (is_delayslot) { // burn cycles to cause cc_interrupt, which will // reschedule next_interupt. Relies on CCREG from above. @@ -3453,67 +3770,47 @@ void cop0_assemble(int i,struct regstat *i_regs) emit_writeword(HOST_CCREG,(int)&last_count); emit_movimm(0,HOST_CCREG); emit_storereg(CCREG,HOST_CCREG); + emit_loadreg(rs1[i],1); emit_movimm(copr,0); emit_call((int)pcsx_mtc0_ds); + emit_loadreg(rs1[i],s); return; } -#endif - emit_movimm(start+i*4+4,0); - emit_movimm(0,1); - emit_writeword(0,(int)&pcaddr); - emit_writeword(1,(int)&pending_exception); + emit_movimm(start+i*4+4,HOST_TEMPREG); + emit_writeword(HOST_TEMPREG,(int)&pcaddr); + emit_movimm(0,HOST_TEMPREG); + emit_writeword(HOST_TEMPREG,(int)&pending_exception); } //else if(copr==12&&is_delayslot) emit_call((int)MTC0_R12); //else -#ifdef PCSX + if(s==HOST_CCREG) + emit_loadreg(rs1[i],1); + else if(s!=1) + emit_mov(s,1); emit_movimm(copr,0); emit_call((int)pcsx_mtc0); -#else - emit_call((int)MTC0); -#endif if(copr==9||copr==11||copr==12||copr==13) { emit_readword((int)&Count,HOST_CCREG); - emit_readword((int)&next_interupt,ECX); - emit_addimm(HOST_CCREG,-CLOCK_DIVIDER*ccadj[i],HOST_CCREG); - emit_sub(HOST_CCREG,ECX,HOST_CCREG); - emit_writeword(ECX,(int)&last_count); + emit_readword((int)&next_interupt,HOST_TEMPREG); + emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG); + emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG); + emit_writeword(HOST_TEMPREG,(int)&last_count); emit_storereg(CCREG,HOST_CCREG); } if(copr==12||copr==13) { assert(!is_delayslot); emit_readword((int)&pending_exception,14); + emit_test(14,14); + emit_jne((int)&do_interrupt); } emit_loadreg(rs1[i],s); if(get_reg(i_regs->regmap,rs1[i]|64)>=0) emit_loadreg(rs1[i]|64,get_reg(i_regs->regmap,rs1[i]|64)); - if(copr==12||copr==13) { - emit_test(14,14); - emit_jne((int)&do_interrupt); - } cop1_usable=0; } else { assert(opcode2[i]==0x10); -#ifndef DISABLE_TLB - if((source[i]&0x3f)==0x01) // TLBR - emit_call((int)TLBR); - if((source[i]&0x3f)==0x02) // TLBWI - emit_call((int)TLBWI_new); - if((source[i]&0x3f)==0x06) { // TLBWR - // The TLB entry written by TLBWR is dependent on the count, - // so update the cycle count - emit_readword((int)&last_count,ECX); - if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); - emit_add(HOST_CCREG,ECX,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_DIVIDER*ccadj[i],HOST_CCREG); - emit_writeword(HOST_CCREG,(int)&Count); - emit_call((int)TLBWR_new); - } - if((source[i]&0x3f)==0x08) // TLBP - emit_call((int)TLBP); -#endif -#ifdef PCSX if((source[i]&0x3f)==0x10) // RFE { emit_readword((int)&Status,0); @@ -3522,15 +3819,6 @@ void cop0_assemble(int i,struct regstat *i_regs) emit_orrshr_imm(1,2,0); emit_writeword(0,(int)&Status); } -#else - if((source[i]&0x3f)==0x18) // ERET - { - int count=ccadj[i]; - if(i_regs->regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG); - emit_addimm(HOST_CCREG,CLOCK_DIVIDER*count,HOST_CCREG); // TODO: Should there be an extra cycle here? - emit_jmp((int)jump_eret); - } -#endif } } @@ -3612,7 +3900,16 @@ static void cop2_put_dreg(u_int copr,signed char sl,signed char temp) case 30: emit_movs(sl,temp); emit_mvnmi(temp,temp); +#ifdef HAVE_ARMV5 emit_clz(temp,temp); +#else + emit_movs(temp,HOST_TEMPREG); + emit_movimm(0,temp); + emit_jeq((int)out+4*4); + emit_addpl_imm(temp,1,temp); + emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG); + emit_jns((int)out-2*4); +#endif emit_writeword(sl,(int)®_cop2d[30]); emit_writeword(temp,(int)®_cop2d[31]); break; @@ -3675,28 +3972,162 @@ void cop2_assemble(int i,struct regstat *i_regs) } } -void c2op_assemble(int i,struct regstat *i_regs) +static void c2op_prologue(u_int op,u_int reglist) +{ + save_regs_all(reglist); +#ifdef PCNT + emit_movimm(op,0); + emit_call((int)pcnt_gte_start); +#endif + emit_addimm(FP,(int)&psxRegs.CP2D.r[0]-(int)&dynarec_local,0); // cop2 regs +} + +static void c2op_epilogue(u_int op,u_int reglist) +{ +#ifdef PCNT + emit_movimm(op,0); + emit_call((int)pcnt_gte_end); +#endif + restore_regs_all(reglist); +} + +static void c2op_call_MACtoIR(int lm,int need_flags) +{ + if(need_flags) + emit_call((int)(lm?gteMACtoIR_lm1:gteMACtoIR_lm0)); + else + emit_call((int)(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((int)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((int)(need_flags?gteMACtoRGB:gteMACtoRGB_nf)); +} + +static 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; + u_int hr,reglist_full=0,reglist; + int need_flags,need_ir; for(hr=0;hrregmap[hr]>=0) reglist|=1<regmap[hr]>=0) reglist_full|=1<regmap,CCREG); - emit_movimm(source[i],temp); // 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]); + 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) + need_flags=0; + int shift = (source[i] >> 19) & 1; + int lm = (source[i] >> 10) & 1; + switch(c2op) { +#ifndef DRC_DBG + case GTE_MVMVA: { +#ifdef HAVE_ARMV5 + int v = (source[i] >> 15) & 3; + 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); + /* r4,r5 = VXYZ(v) packed; r6 = &MX11(mx); r7 = &CV1(cv) */ + if(v<3) + emit_ldrd(v*8,0,4); + else { + emit_movzwl_indexed(9*4,0,4); // gteIR + emit_movzwl_indexed(10*4,0,6); + emit_movzwl_indexed(11*4,0,5); + emit_orrshl_imm(6,16,4); + } + if(mx<3) + emit_addimm(0,32*4+mx*8*4,6); + else + emit_readword((int)&zeromem_ptr,6); + if(cv<3) + emit_addimm(0,32*4+(cv*8+5)*4,7); + else + emit_readword((int)&zeromem_ptr,7); +#ifdef __ARM_NEON__ + emit_movimm(source[i],1); // opcode + emit_call((int)gteMVMVA_part_neon); + if(need_flags) { + emit_movimm(lm,1); + emit_call((int)gteMACtoIR_flags_neon); + } +#else + if(cv==3&&shift) + emit_call((int)gteMVMVA_part_cv3sh12_arm); + else { + emit_movimm(shift,1); + emit_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); + emit_movimm(source[i],1); // opcode + emit_writeword(1,(int)&psxRegs.code); + emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op])); +#endif + break; + } + case GTE_OP: + c2op_prologue(c2op,reglist); + emit_call((int)(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_call_rgb_func(shift?gteDPCS_part_shift:gteDPCS_part_noshift,lm,need_ir,need_flags); + break; + case GTE_INTPL: + c2op_prologue(c2op,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((int)(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_call_rgb_func(gteDCPL_part,lm,need_ir,need_flags); + break; + case GTE_GPF: + c2op_prologue(c2op,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_call_rgb_func(shift?gteGPL_part_shift:gteGPL_part_noshift,lm,need_ir,need_flags); + break; +#endif + default: + c2op_prologue(c2op,reglist); +#ifdef DRC_DBG + emit_movimm(source[i],1); // opcode + emit_writeword(1,(int)&psxRegs.code); +#endif + emit_call((int)(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op])); + break; + } + c2op_epilogue(c2op,reglist); } - - if(i>=slen-1||itype[i+1]!=C2OP) - restore_regs(reglist); } void cop1_unusable(int i,struct regstat *i_regs) @@ -3712,624 +4143,23 @@ void cop1_unusable(int i,struct regstat *i_regs) void cop1_assemble(int i,struct regstat *i_regs) { -#ifndef DISABLE_COP1 - // Check cop1 unusable - if(!cop1_usable) { - signed char rs=get_reg(i_regs->regmap,CSREG); - assert(rs>=0); - emit_testimm(rs,0x20000000); - int jaddr=(int)out; - emit_jeq(0); - add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0); - cop1_usable=1; - } - if (opcode2[i]==0) { // MFC1 - signed char tl=get_reg(i_regs->regmap,rt1[i]); - if(tl>=0) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],tl); - emit_readword_indexed(0,tl,tl); - } - } - else if (opcode2[i]==1) { // DMFC1 - signed char tl=get_reg(i_regs->regmap,rt1[i]); - signed char th=get_reg(i_regs->regmap,rt1[i]|64); - if(tl>=0) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],tl); - if(th>=0) emit_readword_indexed(4,tl,th); - emit_readword_indexed(0,tl,tl); - } - } - else if (opcode2[i]==4) { // MTC1 - signed char sl=get_reg(i_regs->regmap,rs1[i]); - signed char temp=get_reg(i_regs->regmap,-1); - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_writeword_indexed(sl,0,temp); - } - else if (opcode2[i]==5) { // DMTC1 - signed char sl=get_reg(i_regs->regmap,rs1[i]); - signed char sh=rs1[i]>0?get_reg(i_regs->regmap,rs1[i]|64):sl; - signed char temp=get_reg(i_regs->regmap,-1); - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - emit_writeword_indexed(sh,4,temp); - emit_writeword_indexed(sl,0,temp); - } - else if (opcode2[i]==2) // CFC1 - { - signed char tl=get_reg(i_regs->regmap,rt1[i]); - if(tl>=0) { - u_int copr=(source[i]>>11)&0x1f; - if(copr==0) emit_readword((int)&FCR0,tl); - if(copr==31) emit_readword((int)&FCR31,tl); - } - } - else if (opcode2[i]==6) // CTC1 - { - signed char sl=get_reg(i_regs->regmap,rs1[i]); - u_int copr=(source[i]>>11)&0x1f; - assert(sl>=0); - if(copr==31) - { - emit_writeword(sl,(int)&FCR31); - // Set the rounding mode - //FIXME - //char temp=get_reg(i_regs->regmap,-1); - //emit_andimm(sl,3,temp); - //emit_fldcw_indexed((int)&rounding_modes,temp); - } - } -#else cop1_unusable(i, i_regs); -#endif } void fconv_assemble_arm(int i,struct regstat *i_regs) { -#ifndef DISABLE_COP1 - signed char temp=get_reg(i_regs->regmap,-1); - assert(temp>=0); - // Check cop1 unusable - if(!cop1_usable) { - signed char rs=get_reg(i_regs->regmap,CSREG); - assert(rs>=0); - emit_testimm(rs,0x20000000); - int jaddr=(int)out; - emit_jeq(0); - add_stub(FP_STUB,jaddr,(int)out,i,rs,(int)i_regs,is_delayslot,0); - cop1_usable=1; - } - - #if(defined(__VFP_FP__) && !defined(__SOFTFP__)) - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) { // trunc_w_s - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_flds(temp,15); - emit_ftosizs(15,15); // float->int, truncate - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - emit_fsts(15,temp); - return; - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) { // trunc_w_d - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - emit_vldr(temp,7); - emit_ftosizd(7,13); // double->int, truncate - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - emit_fsts(13,temp); - return; - } - - if(opcode2[i]==0x14&&(source[i]&0x3f)==0x20) { // cvt_s_w - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_flds(temp,13); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - emit_fsitos(13,15); - emit_fsts(15,temp); - return; - } - if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) { // cvt_d_w - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_flds(temp,13); - emit_readword((int)®_cop1_double[(source[i]>>6)&0x1f],temp); - emit_fsitod(13,7); - emit_vstr(7,temp); - return; - } - - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) { // cvt_d_s - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_flds(temp,13); - emit_readword((int)®_cop1_double[(source[i]>>6)&0x1f],temp); - emit_fcvtds(13,7); - emit_vstr(7,temp); - return; - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) { // cvt_s_d - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - emit_vldr(temp,7); - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - emit_fcvtsd(7,13); - emit_fsts(13,temp); - return; - } - #endif - - // C emulation code - - u_int hr,reglist=0; - for(hr=0;hrregmap[hr]>=0) reglist|=1<>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_s_w); - } - if(opcode2[i]==0x14&&(source[i]&0x3f)==0x21) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_d_w); - } - if(opcode2[i]==0x15&&(source[i]&0x3f)==0x20) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_s_l); - } - if(opcode2[i]==0x15&&(source[i]&0x3f)==0x21) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_d_l); - } - - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x21) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_d_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x24) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_w_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x25) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_l_s); - } - - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x20) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_s_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x24) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_w_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x25) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)cvt_l_d); - } - - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x08) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)round_l_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x09) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)trunc_l_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0a) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)ceil_l_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0b) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)floor_l_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0c) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)round_w_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0d) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)trunc_w_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0e) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)ceil_w_s); - } - if(opcode2[i]==0x10&&(source[i]&0x3f)==0x0f) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)floor_w_s); - } - - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x08) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)round_l_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x09) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)trunc_l_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0a) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)ceil_l_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0b) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)floor_l_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0c) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)round_w_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0d) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)trunc_w_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0e) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)ceil_w_d); - } - if(opcode2[i]==0x11&&(source[i]&0x3f)==0x0f) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - emit_call((int)floor_w_d); - } - - restore_regs(reglist); -#else cop1_unusable(i, i_regs); -#endif } #define fconv_assemble fconv_assemble_arm void fcomp_assemble(int i,struct regstat *i_regs) { -#ifndef DISABLE_COP1 - signed char fs=get_reg(i_regs->regmap,FSREG); - signed char temp=get_reg(i_regs->regmap,-1); - assert(temp>=0); - // Check cop1 unusable - if(!cop1_usable) { - signed char cs=get_reg(i_regs->regmap,CSREG); - assert(cs>=0); - emit_testimm(cs,0x20000000); - int jaddr=(int)out; - emit_jeq(0); - add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0); - cop1_usable=1; - } - - if((source[i]&0x3f)==0x30) { - emit_andimm(fs,~0x800000,fs); - return; - } - - if((source[i]&0x3e)==0x38) { - // sf/ngle - these should throw exceptions for NaNs - emit_andimm(fs,~0x800000,fs); - return; - } - - #if(defined(__VFP_FP__) && !defined(__SOFTFP__)) - if(opcode2[i]==0x10) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_readword((int)®_cop1_simple[(source[i]>>16)&0x1f],HOST_TEMPREG); - emit_orimm(fs,0x800000,fs); - emit_flds(temp,14); - emit_flds(HOST_TEMPREG,15); - emit_fcmps(14,15); - emit_fmstat(); - if((source[i]&0x3f)==0x31) emit_bicvc_imm(fs,0x800000,fs); // c_un_s - if((source[i]&0x3f)==0x32) emit_bicne_imm(fs,0x800000,fs); // c_eq_s - if((source[i]&0x3f)==0x33) {emit_bicne_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ueq_s - if((source[i]&0x3f)==0x34) emit_biccs_imm(fs,0x800000,fs); // c_olt_s - if((source[i]&0x3f)==0x35) {emit_biccs_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ult_s - if((source[i]&0x3f)==0x36) emit_bichi_imm(fs,0x800000,fs); // c_ole_s - if((source[i]&0x3f)==0x37) {emit_bichi_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ule_s - if((source[i]&0x3f)==0x3a) emit_bicne_imm(fs,0x800000,fs); // c_seq_s - if((source[i]&0x3f)==0x3b) emit_bicne_imm(fs,0x800000,fs); // c_ngl_s - if((source[i]&0x3f)==0x3c) emit_biccs_imm(fs,0x800000,fs); // c_lt_s - if((source[i]&0x3f)==0x3d) emit_biccs_imm(fs,0x800000,fs); // c_nge_s - if((source[i]&0x3f)==0x3e) emit_bichi_imm(fs,0x800000,fs); // c_le_s - if((source[i]&0x3f)==0x3f) emit_bichi_imm(fs,0x800000,fs); // c_ngt_s - return; - } - if(opcode2[i]==0x11) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - emit_readword((int)®_cop1_double[(source[i]>>16)&0x1f],HOST_TEMPREG); - emit_orimm(fs,0x800000,fs); - emit_vldr(temp,6); - emit_vldr(HOST_TEMPREG,7); - emit_fcmpd(6,7); - emit_fmstat(); - if((source[i]&0x3f)==0x31) emit_bicvc_imm(fs,0x800000,fs); // c_un_d - if((source[i]&0x3f)==0x32) emit_bicne_imm(fs,0x800000,fs); // c_eq_d - if((source[i]&0x3f)==0x33) {emit_bicne_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ueq_d - if((source[i]&0x3f)==0x34) emit_biccs_imm(fs,0x800000,fs); // c_olt_d - if((source[i]&0x3f)==0x35) {emit_biccs_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ult_d - if((source[i]&0x3f)==0x36) emit_bichi_imm(fs,0x800000,fs); // c_ole_d - if((source[i]&0x3f)==0x37) {emit_bichi_imm(fs,0x800000,fs);emit_orrvs_imm(fs,0x800000,fs);} // c_ule_d - if((source[i]&0x3f)==0x3a) emit_bicne_imm(fs,0x800000,fs); // c_seq_d - if((source[i]&0x3f)==0x3b) emit_bicne_imm(fs,0x800000,fs); // c_ngl_d - if((source[i]&0x3f)==0x3c) emit_biccs_imm(fs,0x800000,fs); // c_lt_d - if((source[i]&0x3f)==0x3d) emit_biccs_imm(fs,0x800000,fs); // c_nge_d - if((source[i]&0x3f)==0x3e) emit_bichi_imm(fs,0x800000,fs); // c_le_d - if((source[i]&0x3f)==0x3f) emit_bichi_imm(fs,0x800000,fs); // c_ngt_d - return; - } - #endif - - // C only - - u_int hr,reglist=0; - for(hr=0;hrregmap[hr]>=0) reglist|=1<>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_simple[(source[i]>>16)&0x1f],ARG2_REG); - if((source[i]&0x3f)==0x30) emit_call((int)c_f_s); - if((source[i]&0x3f)==0x31) emit_call((int)c_un_s); - if((source[i]&0x3f)==0x32) emit_call((int)c_eq_s); - if((source[i]&0x3f)==0x33) emit_call((int)c_ueq_s); - if((source[i]&0x3f)==0x34) emit_call((int)c_olt_s); - if((source[i]&0x3f)==0x35) emit_call((int)c_ult_s); - if((source[i]&0x3f)==0x36) emit_call((int)c_ole_s); - if((source[i]&0x3f)==0x37) emit_call((int)c_ule_s); - if((source[i]&0x3f)==0x38) emit_call((int)c_sf_s); - if((source[i]&0x3f)==0x39) emit_call((int)c_ngle_s); - if((source[i]&0x3f)==0x3a) emit_call((int)c_seq_s); - if((source[i]&0x3f)==0x3b) emit_call((int)c_ngl_s); - if((source[i]&0x3f)==0x3c) emit_call((int)c_lt_s); - if((source[i]&0x3f)==0x3d) emit_call((int)c_nge_s); - if((source[i]&0x3f)==0x3e) emit_call((int)c_le_s); - if((source[i]&0x3f)==0x3f) emit_call((int)c_ngt_s); - } - if(opcode2[i]==0x11) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - emit_readword((int)®_cop1_double[(source[i]>>16)&0x1f],ARG2_REG); - if((source[i]&0x3f)==0x30) emit_call((int)c_f_d); - if((source[i]&0x3f)==0x31) emit_call((int)c_un_d); - if((source[i]&0x3f)==0x32) emit_call((int)c_eq_d); - if((source[i]&0x3f)==0x33) emit_call((int)c_ueq_d); - if((source[i]&0x3f)==0x34) emit_call((int)c_olt_d); - if((source[i]&0x3f)==0x35) emit_call((int)c_ult_d); - if((source[i]&0x3f)==0x36) emit_call((int)c_ole_d); - if((source[i]&0x3f)==0x37) emit_call((int)c_ule_d); - if((source[i]&0x3f)==0x38) emit_call((int)c_sf_d); - if((source[i]&0x3f)==0x39) emit_call((int)c_ngle_d); - if((source[i]&0x3f)==0x3a) emit_call((int)c_seq_d); - if((source[i]&0x3f)==0x3b) emit_call((int)c_ngl_d); - if((source[i]&0x3f)==0x3c) emit_call((int)c_lt_d); - if((source[i]&0x3f)==0x3d) emit_call((int)c_nge_d); - if((source[i]&0x3f)==0x3e) emit_call((int)c_le_d); - if((source[i]&0x3f)==0x3f) emit_call((int)c_ngt_d); - } - restore_regs(reglist); - emit_loadreg(FSREG,fs); -#else cop1_unusable(i, i_regs); -#endif } void float_assemble(int i,struct regstat *i_regs) { -#ifndef DISABLE_COP1 - signed char temp=get_reg(i_regs->regmap,-1); - assert(temp>=0); - // Check cop1 unusable - if(!cop1_usable) { - signed char cs=get_reg(i_regs->regmap,CSREG); - assert(cs>=0); - emit_testimm(cs,0x20000000); - int jaddr=(int)out; - emit_jeq(0); - add_stub(FP_STUB,jaddr,(int)out,i,cs,(int)i_regs,is_delayslot,0); - cop1_usable=1; - } - - #if(defined(__VFP_FP__) && !defined(__SOFTFP__)) - if((source[i]&0x3f)==6) // mov - { - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - if(opcode2[i]==0x10) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],HOST_TEMPREG); - emit_readword_indexed(0,temp,temp); - emit_writeword_indexed(temp,0,HOST_TEMPREG); - } - if(opcode2[i]==0x11) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - emit_readword((int)®_cop1_double[(source[i]>>6)&0x1f],HOST_TEMPREG); - emit_vldr(temp,7); - emit_vstr(7,HOST_TEMPREG); - } - } - return; - } - - if((source[i]&0x3f)>3) - { - if(opcode2[i]==0x10) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - emit_flds(temp,15); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - } - if((source[i]&0x3f)==4) // sqrt - emit_fsqrts(15,15); - if((source[i]&0x3f)==5) // abs - emit_fabss(15,15); - if((source[i]&0x3f)==7) // neg - emit_fnegs(15,15); - emit_fsts(15,temp); - } - if(opcode2[i]==0x11) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - emit_vldr(temp,7); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - emit_readword((int)®_cop1_double[(source[i]>>6)&0x1f],temp); - } - if((source[i]&0x3f)==4) // sqrt - emit_fsqrtd(7,7); - if((source[i]&0x3f)==5) // abs - emit_fabsd(7,7); - if((source[i]&0x3f)==7) // neg - emit_fnegd(7,7); - emit_vstr(7,temp); - } - return; - } - if((source[i]&0x3f)<4) - { - if(opcode2[i]==0x10) { - emit_readword((int)®_cop1_simple[(source[i]>>11)&0x1f],temp); - } - if(opcode2[i]==0x11) { - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],temp); - } - if(((source[i]>>11)&0x1f)!=((source[i]>>16)&0x1f)) { - if(opcode2[i]==0x10) { - emit_readword((int)®_cop1_simple[(source[i]>>16)&0x1f],HOST_TEMPREG); - emit_flds(temp,15); - emit_flds(HOST_TEMPREG,13); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) { - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - } - } - if((source[i]&0x3f)==0) emit_fadds(15,13,15); - if((source[i]&0x3f)==1) emit_fsubs(15,13,15); - if((source[i]&0x3f)==2) emit_fmuls(15,13,15); - if((source[i]&0x3f)==3) emit_fdivs(15,13,15); - if(((source[i]>>16)&0x1f)==((source[i]>>6)&0x1f)) { - emit_fsts(15,HOST_TEMPREG); - }else{ - emit_fsts(15,temp); - } - } - else if(opcode2[i]==0x11) { - emit_readword((int)®_cop1_double[(source[i]>>16)&0x1f],HOST_TEMPREG); - emit_vldr(temp,7); - emit_vldr(HOST_TEMPREG,6); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - if(((source[i]>>16)&0x1f)!=((source[i]>>6)&0x1f)) { - emit_readword((int)®_cop1_double[(source[i]>>6)&0x1f],temp); - } - } - if((source[i]&0x3f)==0) emit_faddd(7,6,7); - if((source[i]&0x3f)==1) emit_fsubd(7,6,7); - if((source[i]&0x3f)==2) emit_fmuld(7,6,7); - if((source[i]&0x3f)==3) emit_fdivd(7,6,7); - if(((source[i]>>16)&0x1f)==((source[i]>>6)&0x1f)) { - emit_vstr(7,HOST_TEMPREG); - }else{ - emit_vstr(7,temp); - } - } - } - else { - if(opcode2[i]==0x10) { - emit_flds(temp,15); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - emit_readword((int)®_cop1_simple[(source[i]>>6)&0x1f],temp); - } - if((source[i]&0x3f)==0) emit_fadds(15,15,15); - if((source[i]&0x3f)==1) emit_fsubs(15,15,15); - if((source[i]&0x3f)==2) emit_fmuls(15,15,15); - if((source[i]&0x3f)==3) emit_fdivs(15,15,15); - emit_fsts(15,temp); - } - else if(opcode2[i]==0x11) { - emit_vldr(temp,7); - if(((source[i]>>11)&0x1f)!=((source[i]>>6)&0x1f)) { - emit_readword((int)®_cop1_double[(source[i]>>6)&0x1f],temp); - } - if((source[i]&0x3f)==0) emit_faddd(7,7,7); - if((source[i]&0x3f)==1) emit_fsubd(7,7,7); - if((source[i]&0x3f)==2) emit_fmuld(7,7,7); - if((source[i]&0x3f)==3) emit_fdivd(7,7,7); - emit_vstr(7,temp); - } - } - return; - } - #endif - - u_int hr,reglist=0; - for(hr=0;hrregmap[hr]>=0) reglist|=1<>11)&0x1f],ARG1_REG); - if((source[i]&0x3f)<4) { - emit_readword((int)®_cop1_simple[(source[i]>>16)&0x1f],ARG2_REG); - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG3_REG); - }else{ - emit_readword((int)®_cop1_simple[(source[i]>> 6)&0x1f],ARG2_REG); - } - switch(source[i]&0x3f) - { - case 0x00: emit_call((int)add_s);break; - case 0x01: emit_call((int)sub_s);break; - case 0x02: emit_call((int)mul_s);break; - case 0x03: emit_call((int)div_s);break; - case 0x04: emit_call((int)sqrt_s);break; - case 0x05: emit_call((int)abs_s);break; - case 0x06: emit_call((int)mov_s);break; - case 0x07: emit_call((int)neg_s);break; - } - restore_regs(reglist); - } - if(opcode2[i]==0x11) { // Double precision - save_regs(reglist); - emit_readword((int)®_cop1_double[(source[i]>>11)&0x1f],ARG1_REG); - if((source[i]&0x3f)<4) { - emit_readword((int)®_cop1_double[(source[i]>>16)&0x1f],ARG2_REG); - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG3_REG); - }else{ - emit_readword((int)®_cop1_double[(source[i]>> 6)&0x1f],ARG2_REG); - } - switch(source[i]&0x3f) - { - case 0x00: emit_call((int)add_d);break; - case 0x01: emit_call((int)sub_d);break; - case 0x02: emit_call((int)mul_d);break; - case 0x03: emit_call((int)div_d);break; - case 0x04: emit_call((int)sqrt_d);break; - case 0x05: emit_call((int)abs_d);break; - case 0x06: emit_call((int)mov_d);break; - case 0x07: emit_call((int)neg_d);break; - } - restore_regs(reglist); - } -#else cop1_unusable(i, i_regs); -#endif } void multdiv_assemble_arm(int i,struct regstat *i_regs) @@ -4381,12 +4211,21 @@ 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); + emit_negsmi(HOST_TEMPREG,HOST_TEMPREG); +#ifdef HAVE_ARMV5 emit_clz(HOST_TEMPREG,quotient); emit_shl(HOST_TEMPREG,quotient,HOST_TEMPREG); +#else + emit_movimm(0,quotient); + emit_addpl_imm(quotient,1,quotient); + emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG); + emit_jns((int)out-2*4); +#endif emit_orimm(quotient,1<<31,quotient); emit_shr(quotient,quotient,quotient); emit_cmp(remainder,HOST_TEMPREG); @@ -4409,12 +4248,21 @@ 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 +#ifdef HAVE_ARMV5 emit_clz(d2,HOST_TEMPREG); emit_movimm(1<<31,quotient); emit_shl(d2,HOST_TEMPREG,d2); - emit_mov(d1,remainder); +#else + emit_movimm(0,HOST_TEMPREG); + emit_addpl_imm(HOST_TEMPREG,1,HOST_TEMPREG); + emit_lslpls_imm(d2,1,d2); + emit_jns((int)out-2*4); + emit_movimm(1<<31,quotient); +#endif emit_shr(quotient,HOST_TEMPREG,quotient); emit_cmp(remainder,d2); emit_subcs(remainder,d2,remainder); @@ -4424,179 +4272,7 @@ void multdiv_assemble_arm(int i,struct regstat *i_regs) } } else // 64-bit - { - if(opcode2[i]==0x1C) // DMULT - { - assert(opcode2[i]!=0x1C); - signed char m1h=get_reg(i_regs->regmap,rs1[i]|64); - signed char m1l=get_reg(i_regs->regmap,rs1[i]); - signed char m2h=get_reg(i_regs->regmap,rs2[i]|64); - signed char m2l=get_reg(i_regs->regmap,rs2[i]); - assert(m1h>=0); - assert(m2h>=0); - assert(m1l>=0); - assert(m2l>=0); - emit_pushreg(m2h); - emit_pushreg(m2l); - emit_pushreg(m1h); - emit_pushreg(m1l); - emit_call((int)&mult64); - emit_popreg(m1l); - emit_popreg(m1h); - emit_popreg(m2l); - emit_popreg(m2h); - signed char hih=get_reg(i_regs->regmap,HIREG|64); - signed char hil=get_reg(i_regs->regmap,HIREG); - if(hih>=0) emit_loadreg(HIREG|64,hih); - if(hil>=0) emit_loadreg(HIREG,hil); - signed char loh=get_reg(i_regs->regmap,LOREG|64); - signed char lol=get_reg(i_regs->regmap,LOREG); - if(loh>=0) emit_loadreg(LOREG|64,loh); - if(lol>=0) emit_loadreg(LOREG,lol); - } - if(opcode2[i]==0x1D) // DMULTU - { - signed char m1h=get_reg(i_regs->regmap,rs1[i]|64); - signed char m1l=get_reg(i_regs->regmap,rs1[i]); - signed char m2h=get_reg(i_regs->regmap,rs2[i]|64); - signed char m2l=get_reg(i_regs->regmap,rs2[i]); - assert(m1h>=0); - assert(m2h>=0); - assert(m1l>=0); - assert(m2l>=0); - save_regs(0x100f); - if(m1l!=0) emit_mov(m1l,0); - if(m1h==0) emit_readword((int)&dynarec_local,1); - else if(m1h>1) emit_mov(m1h,1); - if(m2l<2) emit_readword((int)&dynarec_local+m2l*4,2); - else if(m2l>2) emit_mov(m2l,2); - if(m2h<3) emit_readword((int)&dynarec_local+m2h*4,3); - else if(m2h>3) emit_mov(m2h,3); - emit_call((int)&multu64); - restore_regs(0x100f); - signed char hih=get_reg(i_regs->regmap,HIREG|64); - signed char hil=get_reg(i_regs->regmap,HIREG); - signed char loh=get_reg(i_regs->regmap,LOREG|64); - signed char lol=get_reg(i_regs->regmap,LOREG); - /*signed char temp=get_reg(i_regs->regmap,-1); - signed char rh=get_reg(i_regs->regmap,HIREG|64); - signed char rl=get_reg(i_regs->regmap,HIREG); - assert(m1h>=0); - assert(m2h>=0); - assert(m1l>=0); - assert(m2l>=0); - assert(temp>=0); - //emit_mov(m1l,EAX); - //emit_mul(m2l); - emit_umull(rl,rh,m1l,m2l); - emit_storereg(LOREG,rl); - emit_mov(rh,temp); - //emit_mov(m1h,EAX); - //emit_mul(m2l); - emit_umull(rl,rh,m1h,m2l); - emit_adds(rl,temp,temp); - emit_adcimm(rh,0,rh); - emit_storereg(HIREG,rh); - //emit_mov(m2h,EAX); - //emit_mul(m1l); - emit_umull(rl,rh,m1l,m2h); - emit_adds(rl,temp,temp); - emit_adcimm(rh,0,rh); - emit_storereg(LOREG|64,temp); - emit_mov(rh,temp); - //emit_mov(m2h,EAX); - //emit_mul(m1h); - emit_umull(rl,rh,m1h,m2h); - emit_adds(rl,temp,rl); - emit_loadreg(HIREG,temp); - emit_adcimm(rh,0,rh); - emit_adds(rl,temp,rl); - emit_adcimm(rh,0,rh); - // DEBUG - /* - emit_pushreg(m2h); - emit_pushreg(m2l); - emit_pushreg(m1h); - emit_pushreg(m1l); - emit_call((int)&multu64); - emit_popreg(m1l); - emit_popreg(m1h); - emit_popreg(m2l); - emit_popreg(m2h); - signed char hih=get_reg(i_regs->regmap,HIREG|64); - signed char hil=get_reg(i_regs->regmap,HIREG); - if(hih>=0) emit_loadreg(HIREG|64,hih); // DEBUG - if(hil>=0) emit_loadreg(HIREG,hil); // DEBUG - */ - // Shouldn't be necessary - //char loh=get_reg(i_regs->regmap,LOREG|64); - //char lol=get_reg(i_regs->regmap,LOREG); - //if(loh>=0) emit_loadreg(LOREG|64,loh); - //if(lol>=0) emit_loadreg(LOREG,lol); - } - if(opcode2[i]==0x1E) // DDIV - { - signed char d1h=get_reg(i_regs->regmap,rs1[i]|64); - signed char d1l=get_reg(i_regs->regmap,rs1[i]); - signed char d2h=get_reg(i_regs->regmap,rs2[i]|64); - signed char d2l=get_reg(i_regs->regmap,rs2[i]); - assert(d1h>=0); - assert(d2h>=0); - assert(d1l>=0); - assert(d2l>=0); - save_regs(0x100f); - if(d1l!=0) emit_mov(d1l,0); - if(d1h==0) emit_readword((int)&dynarec_local,1); - else if(d1h>1) emit_mov(d1h,1); - if(d2l<2) emit_readword((int)&dynarec_local+d2l*4,2); - else if(d2l>2) emit_mov(d2l,2); - if(d2h<3) emit_readword((int)&dynarec_local+d2h*4,3); - else if(d2h>3) emit_mov(d2h,3); - emit_call((int)&div64); - restore_regs(0x100f); - signed char hih=get_reg(i_regs->regmap,HIREG|64); - signed char hil=get_reg(i_regs->regmap,HIREG); - signed char loh=get_reg(i_regs->regmap,LOREG|64); - signed char lol=get_reg(i_regs->regmap,LOREG); - if(hih>=0) emit_loadreg(HIREG|64,hih); - if(hil>=0) emit_loadreg(HIREG,hil); - if(loh>=0) emit_loadreg(LOREG|64,loh); - if(lol>=0) emit_loadreg(LOREG,lol); - } - if(opcode2[i]==0x1F) // DDIVU - { - //u_int hr,reglist=0; - //for(hr=0;hrregmap[hr]>=0 && (i_regs->regmap[hr]&62)!=HIREG) reglist|=1<regmap,rs1[i]|64); - signed char d1l=get_reg(i_regs->regmap,rs1[i]); - signed char d2h=get_reg(i_regs->regmap,rs2[i]|64); - signed char d2l=get_reg(i_regs->regmap,rs2[i]); - assert(d1h>=0); - assert(d2h>=0); - assert(d1l>=0); - assert(d2l>=0); - save_regs(0x100f); - if(d1l!=0) emit_mov(d1l,0); - if(d1h==0) emit_readword((int)&dynarec_local,1); - else if(d1h>1) emit_mov(d1h,1); - if(d2l<2) emit_readword((int)&dynarec_local+d2l*4,2); - else if(d2l>2) emit_mov(d2l,2); - if(d2h<3) emit_readword((int)&dynarec_local+d2h*4,3); - else if(d2h>3) emit_mov(d2h,3); - emit_call((int)&divu64); - restore_regs(0x100f); - signed char hih=get_reg(i_regs->regmap,HIREG|64); - signed char hil=get_reg(i_regs->regmap,HIREG); - signed char loh=get_reg(i_regs->regmap,LOREG|64); - signed char lol=get_reg(i_regs->regmap,LOREG); - if(hih>=0) emit_loadreg(HIREG|64,hih); - if(hil>=0) emit_loadreg(HIREG,hil); - if(loh>=0) emit_loadreg(LOREG|64,loh); - if(lol>=0) emit_loadreg(LOREG,lol); - } - } + assert(0); } else { @@ -4641,7 +4317,7 @@ void do_miniht_jump(int rs,int rh,int ht) { } void do_miniht_insert(u_int return_address,int rt,int temp) { - #ifdef ARMv5_ONLY + #ifndef HAVE_ARMV7 emit_movimm(return_address,rt); // PC into link register add_to_linker((int)out,return_address,1); emit_pcreladdr(temp); @@ -4657,32 +4333,6 @@ void do_miniht_insert(u_int return_address,int rt,int temp) { #endif } -// Sign-extend to 64 bits and write out upper half of a register -// This is useful where we have a 32-bit value in a register, and want to -// keep it in a 32-bit register, but can't guarantee that it won't be read -// as a 64-bit value later. -void wb_sx(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32_pre,uint64_t is32,uint64_t u,uint64_t uu) -{ -#ifndef FORCE32 - if(is32_pre==is32) return; - int hr,reg; - for(hr=0;hr=0) { - if((dirty>>hr)&1) { - if( ((is32_pre&~is32&~uu)>>reg)&1 ) { - emit_sarimm(hr,31,HOST_TEMPREG); - emit_storereg(reg|64,HOST_TEMPREG); - } - } - } - //} - } - } -#endif -} - void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t is32_pre,uint64_t u,uint64_t uu) { //if(dirty_pre==dirty) return; @@ -4691,7 +4341,7 @@ 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); @@ -4705,21 +4355,6 @@ 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); - } - } - } } } } @@ -4782,14 +4417,40 @@ 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<