X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fnew_dynarec.c;h=cd63d2bf218c52bd1a6544f81e6a0b2c9f43c061;hp=dcf940d3ad9af9eeeba5e11c032abde17e84aa6b;hb=HEAD;hpb=a22ccd6a80307ef5f711332f68de96949cdeee76 diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c index dcf940d3..b4295239 100644 --- a/libpcsxcore/new_dynarec/new_dynarec.c +++ b/libpcsxcore/new_dynarec/new_dynarec.c @@ -23,6 +23,7 @@ #include #include #include +#include #ifdef __MACH__ #include #endif @@ -37,6 +38,7 @@ static Jit g_jit; #include "new_dynarec_config.h" #include "../psxhle.h" #include "../psxinterpreter.h" +#include "../psxcounters.h" #include "../gte.h" #include "emu_if.h" // emulator interface #include "linkage_offsets.h" @@ -81,7 +83,7 @@ static Jit g_jit; #endif #define RAM_SIZE 0x200000 -#define MAXBLOCK 4096 +#define MAXBLOCK 2048 #define MAX_OUTPUT_BLOCK_SIZE 262144 #define EXPIRITY_OFFSET (MAX_OUTPUT_BLOCK_SIZE * 2) #define PAGE_COUNT 1024 @@ -110,18 +112,16 @@ struct ndrc_mem struct ndrc_tramp tramp; }; -#ifdef BASE_ADDR_DYNAMIC static struct ndrc_mem *ndrc; -#else -static struct ndrc_mem ndrc_ __attribute__((aligned(4096))); -static struct ndrc_mem *ndrc = &ndrc_; +#ifndef BASE_ADDR_DYNAMIC +// reserve .bss space with upto 64k page size in mind +static char ndrc_bss[((sizeof(*ndrc) + 65535) & ~65535) + 65536]; #endif #ifdef TC_WRITE_OFFSET # ifdef __GLIBC__ # include # include # include -# include # endif static long ndrc_write_ofs; #define NDRC_WRITE_OFFSET(x) (void *)((char *)(x) + ndrc_write_ofs) @@ -242,8 +242,13 @@ static struct decoded_insn u_char is_delay_load:1; // is_load + MFC/CFC u_char is_exception:1; // unconditional, also interp. fallback u_char may_except:1; // might generate an exception + u_char ls_type:2; // load/store type (ls_width_type) } dops[MAXBLOCK]; +enum ls_width_type { + LS_8 = 0, LS_16, LS_32, LS_LR +}; + static struct compile_info { int imm; @@ -642,6 +647,24 @@ static struct ht_entry *hash_table_get(u_int vaddr) return &hash_table[((vaddr>>16)^vaddr)&0xFFFF]; } +#define HASH_TABLE_BAD 0xbac + +static void hash_table_clear(void) +{ + struct ht_entry *ht_bin; + int i, j; + for (i = 0; i < ARRAY_SIZE(hash_table); i++) { + for (j = 0; j < ARRAY_SIZE(hash_table[i].vaddr); j++) { + hash_table[i].vaddr[j] = ~0; + hash_table[i].tcaddr[j] = (void *)(uintptr_t)HASH_TABLE_BAD; + } + } + // don't allow ~0 to hit + ht_bin = hash_table_get(~0); + for (j = 0; j < ARRAY_SIZE(ht_bin->vaddr); j++) + ht_bin->vaddr[j] = 1; +} + static void hash_table_add(u_int vaddr, void *tcaddr) { struct ht_entry *ht_bin = hash_table_get(vaddr); @@ -657,17 +680,30 @@ static void hash_table_remove(int vaddr) //printf("remove hash: %x\n",vaddr); struct ht_entry *ht_bin = hash_table_get(vaddr); if (ht_bin->vaddr[1] == vaddr) { - ht_bin->vaddr[1] = -1; - ht_bin->tcaddr[1] = NULL; + ht_bin->vaddr[1] = ~0; + ht_bin->tcaddr[1] = (void *)(uintptr_t)HASH_TABLE_BAD; } if (ht_bin->vaddr[0] == vaddr) { ht_bin->vaddr[0] = ht_bin->vaddr[1]; ht_bin->tcaddr[0] = ht_bin->tcaddr[1]; - ht_bin->vaddr[1] = -1; - ht_bin->tcaddr[1] = NULL; + ht_bin->vaddr[1] = ~0; + ht_bin->tcaddr[1] = (void *)(uintptr_t)HASH_TABLE_BAD; } } +static void mini_ht_clear(void) +{ +#ifdef USE_MINI_HT + int i; + for (i = 0; i < ARRAY_SIZE(mini_ht) - 1; i++) { + mini_ht[i][0] = ~0; + mini_ht[i][1] = HASH_TABLE_BAD; + } + mini_ht[i][0] = 1; + mini_ht[i][1] = HASH_TABLE_BAD; +#endif +} + static void mark_invalid_code(u_int vaddr, u_int len, char invalid) { u_int vaddr_m = vaddr & 0x1fffffff; @@ -1562,9 +1598,7 @@ static int invalidate_range(u_int start, u_int end, } if (hit) { do_clear_cache(); -#ifdef USE_MINI_HT - memset(mini_ht, -1, sizeof(mini_ht)); -#endif + mini_ht_clear(); } if (inv_start <= (start_m & ~0xfff) && inv_end >= (start_m | 0xfff)) @@ -1621,10 +1655,8 @@ void new_dynarec_invalidate_all_pages(void) } } - #ifdef USE_MINI_HT - memset(mini_ht, -1, sizeof(mini_ht)); - #endif do_clear_cache(); + mini_ht_clear(); } // Add an entry to jump_out after making a link @@ -2109,28 +2141,18 @@ static void multdiv_alloc(struct regstat *current,int i) clear_const(current,dops[i].rs2); alloc_cc(current,i); // for stalls dirty_reg(current,CCREG); - if(dops[i].rs1&&dops[i].rs2) + current->u &= ~(1ull << HIREG); + current->u &= ~(1ull << LOREG); + alloc_reg(current, i, HIREG); + alloc_reg(current, i, LOREG); + dirty_reg(current, HIREG); + dirty_reg(current, LOREG); + if ((dops[i].opcode2 & 0x3e) == 0x1a || (dops[i].rs1 && dops[i].rs2)) // div(u) { - current->u&=~(1LL<u&=~(1LL<regmap); tl=get_reg(i_regs->regmap,dops[i].rs2); s=get_reg(i_regs->regmap,dops[i].rs1); offset=cinfo[i].imm; if(s>=0) { c=(i_regs->wasconst>>s)&1; - if(c) { - memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; + if (c) { + addr_const = constmap[i][s] + offset; + memtarget = ((signed int)addr_const) < (signed int)(0x80000000 + RAM_SIZE); } } assert(tl>=0); assert(addr >= 0); if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<regmap,dops[i].rs2,ccadj_,reglist); } + if (!c || is_ram_addr(addr_const)) + do_store_smc_check(i, i_regs, reglist, addr); + if (c && !memtarget) + inline_writestub(type, i, addr_const, i_regs->regmap, dops[i].rs2, ccadj_, reglist); // basic current block modification detection.. // not looking back as that should be in mips cache already // (see Spyro2 title->attract mode) - if(c&&start+i*4regmap==regs[i].regmap); // not delay slot if(i_regs->regmap==regs[i].regmap) { load_all_consts(regs[i].regmap_entry,regs[i].wasdirty,i); @@ -3407,18 +3427,21 @@ static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_) void *done0, *done1, *done2; int memtarget=0,c=0; int offset_reg = -1; - u_int reglist=get_host_reglist(i_regs->regmap); + u_int addr_const = ~0; + u_int reglist = get_host_reglist(i_regs->regmap); tl=get_reg(i_regs->regmap,dops[i].rs2); s=get_reg(i_regs->regmap,dops[i].rs1); offset=cinfo[i].imm; if(s>=0) { - c=(i_regs->isconst>>s)&1; - if(c) { - memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE; + c = (i_regs->isconst >> s) & 1; + if (c) { + addr_const = constmap[i][s] + offset; + memtarget = ((signed int)addr_const) < (signed int)(0x80000000 + RAM_SIZE); } } assert(tl>=0); assert(addr >= 0); + reglist |= 1u << addr; if(!c) { emit_cmpimm(addr, RAM_SIZE); jaddr=out; @@ -3458,14 +3481,14 @@ static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_) if (dops[i].opcode == 0x2A) { // SWL // Write two msb into two least significant bytes if (dops[i].rs2) emit_rorimm(tl, 16, tl); - do_store_hword(addr, -1, tl, offset_reg, 0); + do_store_hword(addr, -1, tl, offset_reg, 1); if (dops[i].rs2) emit_rorimm(tl, 16, tl); } else if (dops[i].opcode == 0x2E) { // SWR // Write 3 lsb into three most significant bytes do_store_byte(addr, tl, offset_reg); if (dops[i].rs2) emit_rorimm(tl, 8, tl); - do_store_hword(addr, 1, tl, offset_reg, 0); + do_store_hword(addr, 1, tl, offset_reg, 1); if (dops[i].rs2) emit_rorimm(tl, 24, tl); } done1=out; @@ -3493,7 +3516,7 @@ static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_) // 3 set_jump_target(case3, out); if (dops[i].opcode == 0x2A) { // SWL - do_store_word(addr, -3, tl, offset_reg, 0); + do_store_word(addr, -3, tl, offset_reg, 1); } else if (dops[i].opcode == 0x2E) { // SWR do_store_byte(addr, tl, offset_reg); @@ -3503,9 +3526,10 @@ static void storelr_assemble(int i, const struct regstat *i_regs, int ccadj_) set_jump_target(done2, out); if (offset_reg == HOST_TEMPREG) host_tempreg_release(); - if(!c||!memtarget) + if (!c || !memtarget) add_stub_r(STORELR_STUB,jaddr,out,i,addr,i_regs,ccadj_,reglist); - do_store_smc_check(i, i_regs, reglist, addr); + if (!c || is_ram_addr(addr_const)) + do_store_smc_check(i, i_regs, reglist, addr); } static void cop0_assemble(int i, const struct regstat *i_regs, int ccadj_) @@ -3897,6 +3921,7 @@ static void c2ls_assemble(int i, const struct regstat *i_regs, int ccadj_) enum stub_type type; int offset_reg = -1; int fastio_reg_override = -1; + u_int addr_const = ~0; u_int reglist=get_host_reglist(i_regs->regmap); u_int copr=(source[i]>>16)&0x1f; s=get_reg(i_regs->regmap,dops[i].rs1); @@ -3913,8 +3938,13 @@ static void c2ls_assemble(int i, const struct regstat *i_regs, int ccadj_) if (dops[i].opcode==0x3a) { // SWC2 reglist |= 1<=0) c=(i_regs->wasconst>>s)&1; - memtarget=c&&(((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE); + if (s >= 0) { + c = (i_regs->isconst >> s) & 1; + if (c) { + addr_const = constmap[i][s] + offset; + memtarget = ((signed int)addr_const) < (signed int)(0x80000000 + RAM_SIZE); + } + } cop2_do_stall_check(0, i, i_regs, reglist); @@ -3963,9 +3993,9 @@ static void c2ls_assemble(int i, const struct regstat *i_regs, int ccadj_) host_tempreg_release(); if(jaddr2) add_stub_r(type,jaddr2,out,i,ar,i_regs,ccadj_,reglist); - if(dops[i].opcode==0x3a) // SWC2 + if (dops[i].opcode == 0x3a && (!c || is_ram_addr(addr_const))) // SWC2 do_store_smc_check(i, i_regs, reglist, ar); - if (dops[i].opcode==0x32) { // LWC2 + if (dops[i].opcode == 0x32) { // LWC2 host_tempreg_acquire(); cop2_put_dreg(copr,tl,HOST_TEMPREG); host_tempreg_release(); @@ -4529,12 +4559,6 @@ static void address_generation(int i, const struct regstat *i_regs, signed char cinfo[i].addr = rs; add_offset = 0; } - else if (dops[i].itype == STORELR) { // overwrites addr - assert(ra >= 0); - assert(rs != ra); - emit_mov(rs, ra); - cinfo[i].addr = ra; - } else cinfo[i].addr = rs; if (add_offset) { @@ -4631,11 +4655,13 @@ static void load_consts(signed char pre[],signed char regmap[],int i) if(i==0||dops[i].bt) regs[i].loadedconst=0; else { - for(hr=0;hr=0&&((regs[i-1].isconst>>hr)&1)&&pre[hr]==regmap[hr] - &®map[hr]==regs[i-1].regmap[hr]&&((regs[i-1].loadedconst>>hr)&1)) + for (hr = 0; hr < HOST_REGS; hr++) { + if (hr == EXCLUDE_REG || regmap[hr] < 0 || pre[hr] != regmap[hr]) + continue; + if ((((regs[i-1].isconst & regs[i-1].loadedconst) >> hr) & 1) + && regmap[hr] == regs[i-1].regmap[hr]) { - regs[i].loadedconst|=1<dirty & ~i_regs->loadedconst) { + assem_debug("/ drc_dbg_wb\n"); + wb_dirtys(i_regs->regmap, i_regs->dirty & ~i_regs->loadedconst); + assem_debug("\\ drc_dbg_wb\n"); + } +} #else #define drc_dbg_emit_do_cmp(x,y) +#define drc_dbg_emit_wb_dirtys(x,y) #endif // Used when a branch jumps into the delay slot of another branch @@ -5686,6 +5726,7 @@ static void cjump_assemble(int i, const struct regstat *i_regs) load_reg(regs[i].regmap,branch_regs[i].regmap,ROREG); load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,INVCP); ds_assemble(i+1,&branch_regs[i]); + drc_dbg_emit_wb_dirtys(i+1, &branch_regs[i]); cc=get_reg(branch_regs[i].regmap,CCREG); if(cc==-1) { emit_loadreg(CCREG,cc=HOST_CCREG); @@ -6177,6 +6218,12 @@ static noinline void new_dynarec_test(void) out = ndrc->translation_cache; } +static int get_cycle_multiplier(void) +{ + return Config.cycle_multiplier_override && Config.cycle_multiplier == CYCLE_MULT_DEFAULT + ? Config.cycle_multiplier_override : Config.cycle_multiplier; +} + // clear the state completely, instead of just marking // things invalid like invalidate_all_pages() does void new_dynarec_clear_full(void) @@ -6184,9 +6231,9 @@ void new_dynarec_clear_full(void) int n; out = ndrc->translation_cache; memset(invalid_code,1,sizeof(invalid_code)); - memset(hash_table,0xff,sizeof(hash_table)); - memset(mini_ht,-1,sizeof(mini_ht)); memset(shadow,0,sizeof(shadow)); + hash_table_clear(); + mini_ht_clear(); copy=shadow; expirep = EXPIRITY_OFFSET; pending_exception=0; @@ -6204,13 +6251,30 @@ void new_dynarec_clear_full(void) stat_clear(stat_blocks); stat_clear(stat_links); + if (cycle_multiplier_old != Config.cycle_multiplier + || new_dynarec_hacks_old != new_dynarec_hacks) + { + SysPrintf("ndrc config: mul=%d, ha=%x, pex=%d\n", + get_cycle_multiplier(), new_dynarec_hacks, Config.PreciseExceptions); + } cycle_multiplier_old = Config.cycle_multiplier; new_dynarec_hacks_old = new_dynarec_hacks; } +static int pgsize(void) +{ +#ifdef _SC_PAGESIZE + return sysconf(_SC_PAGESIZE); +#else + return 4096; +#endif +} + void new_dynarec_init(void) { - SysPrintf("Init new dynarec, ndrc size %x\n", (int)sizeof(*ndrc)); + int align = pgsize() - 1; + SysPrintf("Init new dynarec, ndrc size %x, pgsize %d\n", + (int)sizeof(*ndrc), align + 1); #ifdef _3DS check_rosalina(); @@ -6266,11 +6330,12 @@ void new_dynarec_init(void) #endif #else #ifndef NO_WRITE_EXEC + ndrc = (struct ndrc_mem *)((size_t)(ndrc_bss + align) & ~align); // not all systems allow execute in data segment by default // size must be 4K aligned for 3DS? if (mprotect(ndrc, sizeof(*ndrc), PROT_READ | PROT_WRITE | PROT_EXEC) != 0) - SysPrintf("mprotect() failed: %s\n", strerror(errno)); + SysPrintf("mprotect(%p) failed: %s\n", ndrc, strerror(errno)); #endif #endif out = ndrc->translation_cache; @@ -6326,9 +6391,9 @@ static u_int *get_source_start(u_int addr, u_int *limit) *limit = (addr & 0xa0600000) + 0x00200000; return (u_int *)(psxM + (addr & 0x1fffff)); } - else if (!Config.HLE && ( + else if ( /* (0x9fc00000 <= addr && addr < 0x9fc80000) ||*/ - (0xbfc00000 <= addr && addr < 0xbfc80000))) + (0xbfc00000 <= addr && addr < 0xbfc80000)) { // BIOS. The multiplier should be much higher as it's uncached 8bit mem, // but timings in PCSX are too tied to the interpreter's 2-per-insn assumption @@ -6474,6 +6539,15 @@ void new_dynarec_print_stats(void) #endif } +static void force_intcall(int i) +{ + memset(&dops[i], 0, sizeof(dops[i])); + dops[i].itype = INTCALL; + dops[i].rs1 = CCREG; + dops[i].is_exception = 1; + cinfo[i].ba = -1; +} + static int apply_hacks(void) { int i; @@ -6508,22 +6582,29 @@ static int apply_hacks(void) return 1; } } + if (Config.HLE) + { + if (start <= psxRegs.biosBranchCheck && psxRegs.biosBranchCheck < start + i*4) + { + i = (psxRegs.biosBranchCheck - start) / 4u + 23; + if (dops[i].is_jump && !dops[i+1].bt) + { + force_intcall(i); + dops[i+1].is_ds = 0; + } + } + } return 0; } -static int is_ld_use_hazard(int ld_rt, const struct decoded_insn *op) -{ - return ld_rt != 0 && (ld_rt == op->rs1 || ld_rt == op->rs2) - && op->itype != LOADLR && op->itype != CJUMP && op->itype != SJUMP; -} - -static void force_intcall(int i) +static int is_ld_use_hazard(const struct decoded_insn *op_ld, + const struct decoded_insn *op) { - memset(&dops[i], 0, sizeof(dops[i])); - dops[i].itype = INTCALL; - dops[i].rs1 = CCREG; - dops[i].is_exception = 1; - cinfo[i].ba = -1; + if (op_ld->rt1 == 0 || (op_ld->rt1 != op->rs1 && op_ld->rt1 != op->rs2)) + return 0; + if (op_ld->itype == LOADLR && op->itype == LOADLR) + return op_ld->rt1 == op_ld->rs1; + return op->itype != CJUMP && op->itype != SJUMP; } static void disassemble_one(int i, u_int src) @@ -6657,20 +6738,20 @@ static void disassemble_one(int i, u_int src) case 0x13: set_mnemonic(i, "COP3"); op2 = (src >> 21) & 0x1f; break; - case 0x20: set_mnemonic(i, "LB"); type=LOAD; break; - case 0x21: set_mnemonic(i, "LH"); type=LOAD; break; - case 0x22: set_mnemonic(i, "LWL"); type=LOADLR; break; - case 0x23: set_mnemonic(i, "LW"); type=LOAD; break; - case 0x24: set_mnemonic(i, "LBU"); type=LOAD; break; - case 0x25: set_mnemonic(i, "LHU"); type=LOAD; break; - case 0x26: set_mnemonic(i, "LWR"); type=LOADLR; break; - case 0x28: set_mnemonic(i, "SB"); type=STORE; break; - case 0x29: set_mnemonic(i, "SH"); type=STORE; break; - case 0x2A: set_mnemonic(i, "SWL"); type=STORELR; break; - case 0x2B: set_mnemonic(i, "SW"); type=STORE; break; - case 0x2E: set_mnemonic(i, "SWR"); type=STORELR; break; - case 0x32: set_mnemonic(i, "LWC2"); type=C2LS; break; - case 0x3A: set_mnemonic(i, "SWC2"); type=C2LS; break; + case 0x20: set_mnemonic(i, "LB"); type=LOAD; ls_type = LS_8; break; + case 0x21: set_mnemonic(i, "LH"); type=LOAD; ls_type = LS_16; break; + case 0x22: set_mnemonic(i, "LWL"); type=LOADLR; ls_type = LS_LR; break; + case 0x23: set_mnemonic(i, "LW"); type=LOAD; ls_type = LS_32; break; + case 0x24: set_mnemonic(i, "LBU"); type=LOAD; ls_type = LS_8; break; + case 0x25: set_mnemonic(i, "LHU"); type=LOAD; ls_type = LS_16; break; + case 0x26: set_mnemonic(i, "LWR"); type=LOADLR; ls_type = LS_LR; break; + case 0x28: set_mnemonic(i, "SB"); type=STORE; ls_type = LS_8; break; + case 0x29: set_mnemonic(i, "SH"); type=STORE; ls_type = LS_16; break; + case 0x2A: set_mnemonic(i, "SWL"); type=STORELR; ls_type = LS_LR; break; + case 0x2B: set_mnemonic(i, "SW"); type=STORE; ls_type = LS_32; break; + case 0x2E: set_mnemonic(i, "SWR"); type=STORELR; ls_type = LS_LR; break; + case 0x32: set_mnemonic(i, "LWC2"); type=C2LS; ls_type = LS_32; break; + case 0x3A: set_mnemonic(i, "SWC2"); type=C2LS; ls_type = LS_32; break; case 0x3B: if (Config.HLE && (src & 0x03ffffff) < ARRAY_SIZE(psxHLEt)) { set_mnemonic(i, "HLECALL"); @@ -6682,8 +6763,9 @@ static void disassemble_one(int i, u_int src) } if (type == INTCALL) SysPrintf("NI %08x @%08x (%08x)\n", src, start + i*4, start); - dops[i].itype=type; - dops[i].opcode2=op2; + dops[i].itype = type; + dops[i].opcode2 = op2; + dops[i].ls_type = ls_type; /* Get registers/immediates */ dops[i].use_lt1=0; gte_rs[i]=gte_rt[i]=0; @@ -6829,6 +6911,7 @@ static void disassemble_one(int i, u_int src) static noinline void pass1_disassemble(u_int pagelimit) { int i, j, done = 0, ni_count = 0; + int ds_next = 0; for (i = 0; !done; i++) { @@ -6836,6 +6919,7 @@ static noinline void pass1_disassemble(u_int pagelimit) unsigned int type, op, op2; disassemble_one(i, source[i]); + dops[i].is_ds = ds_next; ds_next = 0; type = dops[i].itype; op = dops[i].opcode; op2 = dops[i].opcode2; @@ -6868,6 +6952,7 @@ static noinline void pass1_disassemble(u_int pagelimit) dops[i].is_store = type == STORE || type == STORELR || op == 0x3a; // SWC2 dops[i].is_exception = type == SYSCALL || type == HLECALL || type == INTCALL; dops[i].may_except = dops[i].is_exception || (type == ALU && (op2 == 0x20 || op2 == 0x22)) || op == 8; + ds_next = dops[i].is_jump; if (((op & 0x37) == 0x21 || op == 0x25) // LH/SH/LHU && ((cinfo[i].imm & 1) || Config.PreciseExceptions)) @@ -6902,7 +6987,7 @@ static noinline void pass1_disassemble(u_int pagelimit) else dop = &dops[t]; } - if ((dop && is_ld_use_hazard(dops[i].rt1, dop)) + if ((dop && is_ld_use_hazard(&dops[i], dop)) || (!dop && Config.PreciseExceptions)) { // jump target wants DS result - potential load delay effect SysPrintf("load delay in DS @%08x (%08x)\n", start + i*4, start); @@ -6919,7 +7004,7 @@ static noinline void pass1_disassemble(u_int pagelimit) } } else if (i > 0 && dops[i-1].is_delay_load - && is_ld_use_hazard(dops[i-1].rt1, &dops[i]) + && is_ld_use_hazard(&dops[i-1], &dops[i]) && (i < 2 || !dops[i-2].is_ujump)) { SysPrintf("load delay @%08x (%08x)\n", start + i*4, start); for (j = i - 1; j > 0 && dops[j-1].is_delay_load; j--) @@ -6972,12 +7057,13 @@ static noinline void pass1_disassemble(u_int pagelimit) // Don't recompile stuff that's already compiled if(check_addr(start+i*4+4)) done=1; // Don't get too close to the limit - if(i>MAXBLOCK/2) done=1; + if (i > MAXBLOCK - 64) + done = 1; } if (dops[i].itype == HLECALL) - stop = 1; + done = 1; else if (dops[i].itype == INTCALL) - stop = 2; + done = 2; else if (dops[i].is_exception) done = stop_after_jal ? 1 : 2; if (done == 2) { @@ -6992,7 +7078,8 @@ static noinline void pass1_disassemble(u_int pagelimit) //assert(i 8 || dops[i].opcode == 0x11)) { done=stop_after_jal=1; @@ -7155,6 +7242,31 @@ static noinline void pass2_unneeded_regs(int istart,int iend,int r) } } +static noinline void pass2a_unneeded_other(void) +{ + int i, j; + for (i = 0; i < slen; i++) + { + // remove redundant alignment checks + if (dops[i].may_except && (dops[i].is_load || dops[i].is_store) + && dops[i].rt1 != dops[i].rs1 && !dops[i].is_ds) + { + int base = dops[i].rs1, lsb = cinfo[i].imm, ls_type = dops[i].ls_type; + int mask = ls_type == LS_32 ? 3 : 1; + lsb &= mask; + for (j = i + 1; j < slen; j++) { + if (dops[j].bt || dops[j].is_jump) + break; + if ((dops[j].is_load || dops[j].is_store) && dops[j].rs1 == base + && dops[j].ls_type == ls_type && (cinfo[j].imm & mask) == lsb) + dops[j].may_except = 0; + if (dops[j].rt1 == base) + break; + } + } + } +} + static noinline void pass3_register_alloc(u_int addr) { struct regstat current; // Current register allocations/status @@ -7219,7 +7331,7 @@ static noinline void pass3_register_alloc(u_int addr) abort(); } } - dops[i].is_ds=ds; + assert(dops[i].is_ds == ds); if(ds) { ds=0; // Skip delay slot, already allocated as part of branch // ...but we need to alloc it in case something jumps here @@ -8356,6 +8468,7 @@ static noinline void pass5b_preallocate2(void) regs[i+2].wasdirty&=~(1<=0); + #if 0 // what is this for? double allocs $0 in ps1_rom.bin if(regs[i].regmap[hr]<0&®s[i+1].regmap_entry[hr]<0) { regs[i].regmap[hr]=dops[i+1].rs1; @@ -8367,6 +8480,7 @@ static noinline void pass5b_preallocate2(void) regs[i+1].wasdirty&=~(1<