X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fnew_dynarec.c;h=2ca0f60448ecbadf04cc170d7c09f9456ffa5da7;hb=d1150cd66676ce43b8451c65818c2dc3e2f8a1d6;hp=e937b762e33d8f8e01df106e350ae3375644a788;hpb=9c67c98f2350736b964e616743daf7aa10d022ec;p=pcsx_rearmed.git diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c index e937b762..2ca0f604 100644 --- a/libpcsxcore/new_dynarec/new_dynarec.c +++ b/libpcsxcore/new_dynarec/new_dynarec.c @@ -49,6 +49,7 @@ //#define DISASM //#define ASSEM_PRINT +//#define REG_ALLOC_PRINT #ifdef ASSEM_PRINT #define assem_debug printf @@ -75,9 +76,17 @@ #define MAXBLOCK 4096 #define MAX_OUTPUT_BLOCK_SIZE 262144 +#ifdef VITA +// apparently Vita has a 16MB limit, so either we cut tc in half, +// or use this hack (it's a hack because tc size was designed to be power-of-2) +#define TC_REDUCE_BYTES 4096 +#else +#define TC_REDUCE_BYTES 0 +#endif + struct ndrc_mem { - u_char translation_cache[1 << TARGET_SIZE_2]; + u_char translation_cache[(1 << TARGET_SIZE_2) - TC_REDUCE_BYTES]; struct { struct tramp_insns ops[2048 / sizeof(struct tramp_insns)]; @@ -224,7 +233,7 @@ static struct decoded_insn static void *copy; static int expirep; static u_int stop_after_jal; - static u_int f1_hack; // 0 - off, ~0 - capture address, else addr + static u_int f1_hack; int new_dynarec_hacks; int new_dynarec_hacks_pergame; @@ -289,7 +298,7 @@ static struct decoded_insn //#define FLOAT 19 // Floating point unit //#define FCONV 20 // Convert integer to float //#define FCOMP 21 // Floating point compare (sets FSREG) -#define SYSCALL 22// SYSCALL +#define SYSCALL 22// SYSCALL,BREAK #define OTHER 23 // Other #define SPAN 24 // Branch/delay slot spans 2 pages #define NI 25 // Not implemented @@ -320,6 +329,10 @@ void verify_code_ds(); void cc_interrupt(); void fp_exception(); void fp_exception_ds(); +void jump_syscall (u_int u0, u_int u1, u_int pc); +void jump_syscall_ds(u_int u0, u_int u1, u_int pc); +void jump_break (u_int u0, u_int u1, u_int pc); +void jump_break_ds(u_int u0, u_int u1, u_int pc); void jump_to_new_pc(); void call_gteStall(); void new_dyna_leave(); @@ -918,6 +931,10 @@ static const struct { FUNCNAME(jump_handler_write32), FUNCNAME(invalidate_addr), FUNCNAME(jump_to_new_pc), + FUNCNAME(jump_break), + FUNCNAME(jump_break_ds), + FUNCNAME(jump_syscall), + FUNCNAME(jump_syscall_ds), FUNCNAME(call_gteStall), FUNCNAME(new_dyna_leave), FUNCNAME(pcsx_mtc0), @@ -3927,9 +3944,16 @@ static void call_c_cpu_handler(int i, const struct regstat *i_regs, int ccadj_, static void syscall_assemble(int i, const struct regstat *i_regs, int ccadj_) { - emit_movimm(0x20,0); // cause code - emit_movimm(0,1); // not in delay slot - call_c_cpu_handler(i, i_regs, ccadj_, start+i*4, psxException); + // 'break' tends to be littered around to catch things like + // division by 0 and is almost never executed, so don't emit much code here + void *func = (dops[i].opcode2 == 0x0C) + ? (is_delayslot ? jump_syscall_ds : jump_syscall) + : (is_delayslot ? jump_break_ds : jump_break); + signed char ccreg = get_reg(i_regs->regmap, CCREG); + assert(ccreg == HOST_CCREG); + emit_movimm(start + i*4, 2); // pc + emit_addimm(HOST_CCREG, ccadj_ + CLOCK_ADJUST(1), HOST_CCREG); + emit_far_jump(func); } static void hlecall_assemble(int i, const struct regstat *i_regs, int ccadj_) @@ -6831,6 +6855,7 @@ void new_dynarec_clear_full(void) literalcount=0; stop_after_jal=0; inv_code_start=inv_code_end=~0; + hack_addr=0; f1_hack=0; // TLB for(n=0;n<4096;n++) ll_clear(jump_in+n); @@ -6843,7 +6868,7 @@ void new_dynarec_clear_full(void) void new_dynarec_init(void) { - SysPrintf("Init new dynarec\n"); + SysPrintf("Init new dynarec, ndrc size %x\n", (int)sizeof(*ndrc)); #ifdef _3DS check_rosalina(); @@ -6851,11 +6876,11 @@ void new_dynarec_init(void) #ifdef BASE_ADDR_DYNAMIC #ifdef VITA sceBlock = getVMBlock(); //sceKernelAllocMemBlockForVM("code", sizeof(*ndrc)); - if (sceBlock < 0) - SysPrintf("sceKernelAllocMemBlockForVM failed\n"); + if (sceBlock <= 0) + SysPrintf("sceKernelAllocMemBlockForVM failed: %x\n", sceBlock); int ret = sceKernelGetMemBlockBase(sceBlock, (void **)&ndrc); if (ret < 0) - SysPrintf("sceKernelGetMemBlockBase failed\n"); + SysPrintf("sceKernelGetMemBlockBase failed: %x\n", ret); sceKernelOpenVMDomain(); sceClibPrintf("translation_cache = 0x%08lx\n ", (long)ndrc->translation_cache); #elif defined(_MSC_VER) @@ -6903,6 +6928,7 @@ void new_dynarec_cleanup(void) int n; #ifdef BASE_ADDR_DYNAMIC #ifdef VITA + // sceBlock is managed by retroarch's bootstrap code //sceKernelFreeMemBlock(sceBlock); //sceBlock = -1; #else @@ -7046,11 +7072,11 @@ void new_dynarec_load_blocks(const void *save, int size) memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save)); } -static void apply_hacks(void) +static int apply_hacks(void) { int i; if (HACK_ENABLED(NDHACK_NO_COMPAT_HACKS)) - return; + return 0; /* special hack(s) */ for (i = 0; i < slen - 4; i++) { @@ -7075,11 +7101,12 @@ static void apply_hacks(void) if (dops[i].itype == STORELR && dops[i].rs1 == 6 && dops[i-1].itype == STORELR && dops[i-1].rs1 == 6) { - SysPrintf("F1 hack from %08x\n", start); - if (f1_hack == 0) - f1_hack = ~0u; + SysPrintf("F1 hack from %08x, old dst %08x\n", start, hack_addr); + f1_hack = 1; + return 1; } } + return 0; } int new_recompile_block(u_int addr) @@ -7117,9 +7144,11 @@ int new_recompile_block(u_int addr) ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning); return 0; } - else if (f1_hack == ~0u || (f1_hack != 0 && start == f1_hack)) { + else if (f1_hack && hack_addr == 0) { void *beginning = start_block(); u_int page = get_page(start); + emit_movimm(start, 0); + emit_writeword(0, &hack_addr); emit_readword(&psxRegs.GPR.n.sp, 0); emit_readptr(&mem_rtab, 1); emit_shrimm(0, 12, 2); @@ -7135,7 +7164,6 @@ int new_recompile_block(u_int addr) ll_add_flags(jump_in + page, start, state_rflags, beginning); SysPrintf("F1 hack to %08x\n", start); - f1_hack = start; return 0; } @@ -7188,7 +7216,7 @@ int new_recompile_block(u_int addr) case 0x08: strcpy(insn[i],"JR"); type=RJUMP; break; case 0x09: strcpy(insn[i],"JALR"); type=RJUMP; break; case 0x0C: strcpy(insn[i],"SYSCALL"); type=SYSCALL; break; - case 0x0D: strcpy(insn[i],"BREAK"); type=OTHER; break; + case 0x0D: strcpy(insn[i],"BREAK"); type=SYSCALL; break; case 0x0F: strcpy(insn[i],"SYNC"); type=OTHER; break; case 0x10: strcpy(insn[i],"MFHI"); type=MOV; break; case 0x11: strcpy(insn[i],"MTHI"); type=MOV; break; @@ -7634,9 +7662,9 @@ int new_recompile_block(u_int addr) // Don't get too close to the limit if(i>MAXBLOCK/2) done=1; } - if(dops[i].itype==SYSCALL&&stop_after_jal) done=1; - if(dops[i].itype==HLECALL||dops[i].itype==INTCALL) done=2; - if(done==2) { + if (dops[i].itype == SYSCALL || dops[i].itype == HLECALL || dops[i].itype == INTCALL) + done = stop_after_jal ? 1 : 2; + if (done == 2) { // Does the block continue due to a branch? for(j=i-1;j>=0;j--) { @@ -7663,7 +7691,7 @@ int new_recompile_block(u_int addr) } assert(slen>0); - apply_hacks(); + int clear_hack_addr = apply_hacks(); /* Pass 2 - Register dependencies and branch targets */ @@ -9009,7 +9037,7 @@ int new_recompile_block(u_int addr) dops[slen-1].bt=1; // Mark as a branch target so instruction can restart after exception } -#ifdef DISASM +#ifdef REG_ALLOC_PRINT /* Debug/disassembly */ for(i=0;i\n"); drc_dbg_emit_do_cmp(i, ccadj[i]); + if (clear_hack_addr) { + emit_movimm(0, 0); + emit_writeword(0, &hack_addr); + clear_hack_addr = 0; + } // load regs if(regs[i].regmap_entry[HOST_CCREG]==CCREG&®s[i].regmap[HOST_CCREG]!=CCREG)