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

emit_x64_vector: Improve code emission of VectorGetElement* for index == 0

Browse source
This commit is contained in:
MerryMage 2018-07-31 22:04:11 +01:00
parent e9ab7f7664
commit 5ae045d67e
1 changed files with 24 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
src/backend_x64/emit_x64_vector.cpp
View file

@ -128,6 +128,11 @@ void EmitX64::EmitVectorGetElement8(EmitContext& ctx, IR::Inst* inst) {
ASSERT(args[1].IsImmediate()); ASSERT(args[1].IsImmediate());
const u8 index = args[1].GetImmediateU8(); const u8 index = args[1].GetImmediateU8();
if (index == 0) {
ctx.reg_alloc.DefineValue(inst, args[0]);
return;
}
const Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]); const Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32(); const Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32();
@ -146,7 +151,12 @@ void EmitX64::EmitVectorGetElement8(EmitContext& ctx, IR::Inst* inst) {
void EmitX64::EmitVectorGetElement16(EmitContext& ctx, IR::Inst* inst) { void EmitX64::EmitVectorGetElement16(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
ASSERT(args[1].IsImmediate()); ASSERT(args[1].IsImmediate());
u8 index = args[1].GetImmediateU8(); const u8 index = args[1].GetImmediateU8();
if (index == 0) {
ctx.reg_alloc.DefineValue(inst, args[0]);
return;
}
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]); Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32(); Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32();
@ -157,14 +167,16 @@ void EmitX64::EmitVectorGetElement16(EmitContext& ctx, IR::Inst* inst) {
void EmitX64::EmitVectorGetElement32(EmitContext& ctx, IR::Inst* inst) { void EmitX64::EmitVectorGetElement32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
ASSERT(args[1].IsImmediate()); ASSERT(args[1].IsImmediate());
u8 index = args[1].GetImmediateU8(); const u8 index = args[1].GetImmediateU8();
if (index == 0) {
ctx.reg_alloc.DefineValue(inst, args[0]);
return;
}
Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32(); Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32();
if (index == 0) { if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
code.movd(dest, source);
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]); Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
code.pextrd(dest, source, index); code.pextrd(dest, source, index);
} else { } else {
@ -181,12 +193,14 @@ void EmitX64::EmitVectorGetElement64(EmitContext& ctx, IR::Inst* inst) {
ASSERT(args[1].IsImmediate()); ASSERT(args[1].IsImmediate());
u8 index = args[1].GetImmediateU8(); u8 index = args[1].GetImmediateU8();
if (index == 0) {
ctx.reg_alloc.DefineValue(inst, args[0]);
return;
}
Xbyak::Reg64 dest = ctx.reg_alloc.ScratchGpr().cvt64(); Xbyak::Reg64 dest = ctx.reg_alloc.ScratchGpr().cvt64();
if (index == 0) { if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
code.movq(dest, source);
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]); Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
code.pextrq(dest, source, 1); code.pextrq(dest, source, 1);
} else { } else {