X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm64.c;h=ea1b8a318553175cc2a7fc91ef66566616ad7f23;hb=cf95b4f0e912b19298e38ae60dd66a9f4a773e9b;hp=ef87b293c572d3b23c7a7cebe4b516300f10a8eb;hpb=919981d0bca7a0898133362a91405395678612e3;p=pcsx_rearmed.git diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c index ef87b293..ea1b8a31 100644 --- a/libpcsxcore/new_dynarec/assem_arm64.c +++ b/libpcsxcore/new_dynarec/assem_arm64.c @@ -44,7 +44,7 @@ static void set_jump_target(void *addr, void *target) || (*ptr&0x7e000000) == 0x34000000) { // cbz/cbnz // Conditional branch are limited to +/- 1MB // block max size is 256k so branching beyond the +/- 1MB limit - // should only happen when jumping to an already compiled block (see add_link) + // should only happen when jumping to an already compiled block (see add_jump_out) // a workaround would be to do a trampoline jump via a stub at the end of the block assert(-1048576 <= offset && offset < 1048576); *ptr=(*ptr&0xFF00000F)|(((offset>>2)&0x7ffff)<<5); @@ -813,6 +813,12 @@ static void emit_cmovl_reg(u_int rs,u_int rt) output_w32(0x1a800000 | (COND_LT << 12) | rm_rn_rd(rt, rs, rt)); } +static void emit_cmovb_reg(u_int rs,u_int rt) +{ + assem_debug("csel %s,%s,%s,cc\n",regname[rt],regname[rs],regname[rt]); + output_w32(0x1a800000 | (COND_CC << 12) | rm_rn_rd(rt, rs, rt)); +} + static void emit_cmovs_reg(u_int rs,u_int rt) { assem_debug("csel %s,%s,%s,mi\n",regname[rt],regname[rs],regname[rt]); @@ -1374,10 +1380,10 @@ static void do_readstub(int n) u_int reglist = stubs[n].e; const signed char *i_regmap = i_regs->regmap; int rt; - if(itype[i]==C1LS||itype[i]==C2LS||itype[i]==LOADLR) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS||dops[i].itype==LOADLR) { rt=get_reg(i_regmap,FTEMP); }else{ - rt=get_reg(i_regmap,rt1[i]); + rt=get_reg(i_regmap,dops[i].rt1); } assert(rs>=0); int r,temp=-1,temp2=HOST_TEMPREG,regs_saved=0; @@ -1389,7 +1395,7 @@ static void do_readstub(int n) break; } } - if(rt>=0&&rt1[i]!=0) + if(rt>=0&&dops[i].rt1!=0) reglist&=~(1<=0&&rt1[i]!=0)) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) { switch(type) { case LOADB_STUB: emit_ldrsb_dualindexed(temp2,rs,rt); break; case LOADBU_STUB: emit_ldrb_dualindexed(temp2,rs,rt); break; @@ -1436,10 +1442,10 @@ static void do_readstub(int n) int cc=get_reg(i_regmap,CCREG); if(cc<0) emit_loadreg(CCREG,2); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d+1),2); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2); emit_far_call(handler); // (no cycle reload after read) - if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) { loadstore_extend(type,0,rt); } if(restore_jump) @@ -1448,7 +1454,8 @@ static void do_readstub(int n) emit_jmp(stubs[n].retaddr); } -static void inline_readstub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) +static void inline_readstub(enum stub_type type, int i, u_int addr, + const signed char regmap[], int target, int adj, u_int reglist) { int rs=get_reg(regmap,target); int rt=get_reg(regmap,target); @@ -1458,11 +1465,11 @@ static void inline_readstub(enum stub_type type, int i, u_int addr, signed char uintptr_t host_addr = 0; void *handler; int cc=get_reg(regmap,CCREG); - //if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt)) + //if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj),cc,target?rs:-1,rt)) // return; handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr); if (handler == NULL) { - if(rt<0||rt1[i]==0) + if(rt<0||dops[i].rt1==0) return; if (addr != host_addr) { if (host_addr >= 0x100000000ull) @@ -1490,7 +1497,7 @@ static void inline_readstub(enum stub_type type, int i, u_int addr, signed char } // call a memhandler - if(rt>=0&&rt1[i]!=0) + if(rt>=0&&dops[i].rt1!=0) reglist&=~(1<>12] << 1; emit_adrp((void *)l1, 1); @@ -1511,7 +1518,7 @@ static void inline_readstub(enum stub_type type, int i, u_int addr, signed char emit_far_call(handler); // (no cycle reload after read) - if(rt>=0&&rt1[i]!=0) + if(rt>=0&&dops[i].rt1!=0) loadstore_extend(type, 0, rt); restore_regs(reglist); } @@ -1527,10 +1534,10 @@ static void do_writestub(int n) u_int reglist=stubs[n].e; signed char *i_regmap=i_regs->regmap; int rt,r; - if(itype[i]==C1LS||itype[i]==C2LS) { + if(dops[i].itype==C1LS||dops[i].itype==C2LS) { rt=get_reg(i_regmap,r=FTEMP); }else{ - rt=get_reg(i_regmap,r=rs2[i]); + rt=get_reg(i_regmap,r=dops[i].rs2); } assert(rs>=0); assert(rt>=0); @@ -1595,10 +1602,10 @@ static void do_writestub(int n) int cc=get_reg(i_regmap,CCREG); if(cc<0) emit_loadreg(CCREG,2); - emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d+1),2); + emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2); // returns new cycle_count emit_far_call(handler); - emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d+1),cc<0?2:cc); + emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d),cc<0?2:cc); if(cc<0) emit_storereg(CCREG,2); if(restore_jump) @@ -1607,7 +1614,8 @@ static void do_writestub(int n) emit_jmp(stubs[n].retaddr); } -static void inline_writestub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist) +static void inline_writestub(enum stub_type type, int i, u_int addr, + const signed char regmap[], int target, int adj, u_int reglist) { int rs = get_reg(regmap,-1); int rt = get_reg(regmap,target); @@ -1638,12 +1646,12 @@ static void inline_writestub(enum stub_type type, int i, u_int addr, signed char cc = cc_use = get_reg(regmap, CCREG); if (cc < 0) emit_loadreg(CCREG, (cc_use = 2)); - emit_addimm(cc_use, CLOCK_ADJUST(adj+1), 2); + emit_addimm(cc_use, CLOCK_ADJUST(adj), 2); emit_far_call(do_memhandler_pre); emit_far_call(handler); emit_far_call(do_memhandler_post); - emit_addimm(0, -CLOCK_ADJUST(adj+1), cc_use); + emit_addimm(0, -CLOCK_ADJUST(adj), cc_use); if (cc < 0) emit_storereg(CCREG, cc_use); restore_regs(reglist); @@ -1657,12 +1665,12 @@ static int verify_code_arm64(const void *source, const void *copy, u_int size) } // this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr -static void do_dirty_stub_base(u_int vaddr) +static void do_dirty_stub_base(u_int vaddr, u_int source_len) { - assert(slen <= MAXBLOCK); + assert(source_len <= MAXBLOCK*4); emit_loadlp_ofs(0, 0); // ldr x1, source emit_loadlp_ofs(0, 1); // ldr x2, copy - emit_movz(slen*4, 2); + emit_movz(source_len, 2); emit_far_call(verify_code_arm64); void *jmp = out; emit_cbz(0, 0); @@ -1677,7 +1685,7 @@ static void assert_dirty_stub(const u_int *ptr) { assert((ptr[0] & 0xff00001f) == 0x58000000); // ldr x0, source assert((ptr[1] & 0xff00001f) == 0x58000001); // ldr x1, copy - assert((ptr[2] & 0xffe0001f) == 0x52800002); // movz w2, #slen*4 + assert((ptr[2] & 0xffe0001f) == 0x52800002); // movz w2, #source_len assert( ptr[8] == 0xd61f0000); // br x0 } @@ -1698,11 +1706,11 @@ static void do_dirty_stub_emit_literals(u_int *loadlps) output_w64((uintptr_t)copy); } -static void *do_dirty_stub(int i) +static void *do_dirty_stub(int i, u_int source_len) { assem_debug("do_dirty_stub %x\n",start+i*4); u_int *loadlps = (void *)out; - do_dirty_stub_base(start + i*4); + do_dirty_stub_base(start + i*4, source_len); void *entry = out; load_regs_entry(i); if (entry == out) @@ -1712,10 +1720,10 @@ static void *do_dirty_stub(int i) return entry; } -static void do_dirty_stub_ds(void) +static void do_dirty_stub_ds(u_int source_len) { u_int *loadlps = (void *)out; - do_dirty_stub_base(start + 1); + do_dirty_stub_base(start + 1, source_len); void *lit_jumpover = out; emit_jmp(out + 8*2); do_dirty_stub_emit_literals(loadlps); @@ -1752,7 +1760,7 @@ static int verify_dirty(const u_int *ptr) assert_dirty_stub(ptr); source = (void *)get_from_ldr_literal(&ptr[0]); // ldr x1, source copy = (void *)get_from_ldr_literal(&ptr[1]); // ldr x1, copy - len = get_from_movz(&ptr[2]); // movz w3, #slen*4 + len = get_from_movz(&ptr[2]); // movz w3, #source_len return !memcmp(source, copy, len); } @@ -1772,14 +1780,15 @@ static void get_bounds(void *addr, u_char **start, u_char **end) const u_int *ptr = addr; assert_dirty_stub(ptr); *start = (u_char *)get_from_ldr_literal(&ptr[0]); // ldr x1, source - *end = *start + get_from_movz(&ptr[2]); // movz w3, #slen*4 + *end = *start + get_from_movz(&ptr[2]); // movz w3, #source_len } /* Special assem */ -static void c2op_prologue(u_int op,u_int reglist) +static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist) { save_load_regs_all(1, reglist); + cop2_do_stall_check(op, i, i_regs, 0); #ifdef PCNT emit_movimm(op, 0); emit_far_call(pcnt_gte_start); @@ -1797,7 +1806,7 @@ static void c2op_epilogue(u_int op,u_int reglist) save_load_regs_all(0, reglist); } -static void c2op_assemble(int i,struct regstat *i_regs) +static void c2op_assemble(int i, const struct regstat *i_regs) { u_int c2op=source[i]&0x3f; u_int hr,reglist_full=0,reglist; @@ -1812,14 +1821,14 @@ static void c2op_assemble(int i,struct regstat *i_regs) need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00; assem_debug("gte op %08x, unneeded %016lx, need_flags %d, need_ir %d\n", source[i],gte_unneeded[i+1],need_flags,need_ir); - if(new_dynarec_hacks&NDHACK_GTE_NO_FLAGS) + if(HACK_ENABLED(NDHACK_GTE_NO_FLAGS)) need_flags=0; //int shift = (source[i] >> 19) & 1; //int lm = (source[i] >> 10) & 1; switch(c2op) { default: (void)need_ir; - c2op_prologue(c2op,reglist); + c2op_prologue(c2op, i, i_regs, reglist); emit_movimm(source[i],1); // opcode emit_writeword(1,&psxRegs.code); emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]); @@ -1876,15 +1885,15 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs) // case 0x19: MULTU // case 0x1A: DIV // case 0x1B: DIVU - if(rs1[i]&&rs2[i]) + if(dops[i].rs1&&dops[i].rs2) { - switch(opcode2[i]) + switch(dops[i].opcode2) { case 0x18: // MULT case 0x19: // MULTU { - signed char m1=get_reg(i_regs->regmap,rs1[i]); - signed char m2=get_reg(i_regs->regmap,rs2[i]); + signed char m1=get_reg(i_regs->regmap,dops[i].rs1); + signed char m2=get_reg(i_regs->regmap,dops[i].rs2); signed char hi=get_reg(i_regs->regmap,HIREG); signed char lo=get_reg(i_regs->regmap,LOREG); assert(m1>=0); @@ -1892,7 +1901,7 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs) assert(hi>=0); assert(lo>=0); - if(opcode2[i]==0x18) // MULT + if(dops[i].opcode2==0x18) // MULT emit_smull(m1,m2,hi); else // MULTU emit_umull(m1,m2,hi); @@ -1904,8 +1913,8 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs) case 0x1A: // DIV case 0x1B: // DIVU { - signed char numerator=get_reg(i_regs->regmap,rs1[i]); - signed char denominator=get_reg(i_regs->regmap,rs2[i]); + signed char numerator=get_reg(i_regs->regmap,dops[i].rs1); + signed char denominator=get_reg(i_regs->regmap,dops[i].rs2); signed char quotient=get_reg(i_regs->regmap,LOREG); signed char remainder=get_reg(i_regs->regmap,HIREG); assert(numerator>=0); @@ -1913,7 +1922,7 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs) assert(quotient>=0); assert(remainder>=0); - if (opcode2[i] == 0x1A) // DIV + if (dops[i].opcode2 == 0x1A) // DIV emit_sdiv(numerator,denominator,quotient); else // DIVU emit_udiv(numerator,denominator,quotient); @@ -1921,7 +1930,7 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs) // div 0 quotient (remainder is already correct) host_tempreg_acquire(); - if (opcode2[i] == 0x1A) // DIV + if (dops[i].opcode2 == 0x1A) // DIV emit_sub_asrimm(0,numerator,31,HOST_TEMPREG); else emit_movimm(~0,HOST_TEMPREG); @@ -1938,15 +1947,15 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs) { signed char hr=get_reg(i_regs->regmap,HIREG); signed char lr=get_reg(i_regs->regmap,LOREG); - if ((opcode2[i]==0x1A || opcode2[i]==0x1B) && rs2[i]==0) // div 0 + if ((dops[i].opcode2==0x1A || dops[i].opcode2==0x1B) && dops[i].rs2==0) // div 0 { - if (rs1[i]) { - signed char numerator = get_reg(i_regs->regmap, rs1[i]); + if (dops[i].rs1) { + signed char numerator = get_reg(i_regs->regmap, dops[i].rs1); assert(numerator >= 0); if (hr >= 0) emit_mov(numerator,hr); if (lr >= 0) { - if (opcode2[i] == 0x1A) // DIV + if (dops[i].opcode2 == 0x1A) // DIV emit_sub_asrimm(0,numerator,31,lr); else emit_movimm(~0,lr);