mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

a32_emit_x64: Shorten EmitA32GetCpsr

Browse source
This commit is contained in:
MerryMage 2018-07-27 11:15:52 +01:00
parent 9fe2bf8733
commit 0f85305933
2 changed files with 13 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
src/backend_x64/a32_emit_x64.cpp
View file

@ -280,27 +280,22 @@ static u32 GetCpsrImpl(A32JitState* jit_state) {
void A32EmitX64::EmitA32GetCpsr(A32EmitContext& ctx, IR::Inst* inst) { void A32EmitX64::EmitA32GetCpsr(A32EmitContext& ctx, IR::Inst* inst) {
if (code.DoesCpuSupport(Xbyak::util::Cpu::tBMI2)) { if (code.DoesCpuSupport(Xbyak::util::Cpu::tBMI2)) {
Xbyak::Reg32 result = ctx.reg_alloc.ScratchGpr().cvt32(); Xbyak::Reg32 result = ctx.reg_alloc.ScratchGpr().cvt32();
Xbyak::Reg32 b = ctx.reg_alloc.ScratchGpr().cvt32(); Xbyak::Reg32 tmp = ctx.reg_alloc.ScratchGpr().cvt32();
Xbyak::Reg32 c = ctx.reg_alloc.ScratchGpr().cvt32();
code.mov(c, dword[r15 + offsetof(A32JitState, CPSR_ge)]); // Here we observe that CPSR_et and CPSR_ge are right next to each other in memory,
// Here we observe that CPSR_q and CPSR_nzcv are right next to each other in memory,
// so we load them both at the same time with one 64-bit read. This allows us to // so we load them both at the same time with one 64-bit read. This allows us to
// extract all of their bits together at once with one pext. // extract all of their bits together at once with one pext.
static_assert(offsetof(A32JitState, CPSR_q) + 4 == offsetof(A32JitState, CPSR_nzcv)); static_assert(offsetof(A32JitState, CPSR_et) + 4 == offsetof(A32JitState, CPSR_ge));
code.mov(result.cvt64(), qword[r15 + offsetof(A32JitState, CPSR_q)]); code.mov(result.cvt64(), qword[r15 + offsetof(A32JitState, CPSR_et)]);
code.mov(b.cvt64(), 0xF000000000000001ull); code.mov(tmp.cvt64(), 0x8080808000000003ull);
code.pext(result.cvt64(), result.cvt64(), b.cvt64()); code.pext(result.cvt64(), result.cvt64(), tmp.cvt64());
code.mov(b, 0x80808080); code.mov(tmp, 0x000f0220);
code.pext(c.cvt64(), c.cvt64(), b.cvt64()); code.pdep(result, result, tmp);
code.shl(result, 27); code.mov(tmp, dword[r15 + offsetof(A32JitState, CPSR_q)]);
code.shl(c, 16); code.shl(tmp, 27);
code.or_(result, c); code.or_(result, tmp);
code.mov(b, 0x00000220); code.or_(result, dword[r15 + offsetof(A32JitState, CPSR_nzcv)]);
code.mov(c, dword[r15 + offsetof(A32JitState, CPSR_et)]);
code.pdep(c.cvt64(), c.cvt64(), b.cvt64());
code.or_(result, dword[r15 + offsetof(A32JitState, CPSR_jaifm)]); code.or_(result, dword[r15 + offsetof(A32JitState, CPSR_jaifm)]);
code.or_(result, c);
ctx.reg_alloc.DefineValue(inst, result); ctx.reg_alloc.DefineValue(inst, result);
} else { } else {

2
src/backend_x64/a32_jitstate.h
View file

@ -29,8 +29,8 @@ struct A32JitState {
std::array<u32, 16> Reg{}; // Current register file. std::array<u32, 16> Reg{}; // Current register file.
// TODO: Mode-specific register sets unimplemented. // TODO: Mode-specific register sets unimplemented.
u32 CPSR_ge = 0;
u32 CPSR_et = 0; u32 CPSR_et = 0;
u32 CPSR_ge = 0;
u32 CPSR_q = 0; u32 CPSR_q = 0;
u32 CPSR_nzcv = 0; u32 CPSR_nzcv = 0;
u32 CPSR_jaifm = 0; u32 CPSR_jaifm = 0;