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

emit_x64_vector: Use VBPROADCAST where applicable and available

Browse source 
Uses the instruction that does what it says in its name if available. Allows avoiding the use
of a scratch register in EmitVectorBroadcast8() and EmitVectorBroadcastLower8()'s SSSE3 path.
This commit is contained in:
Lioncash 2018-04-26 13:07:10 -04:00 committed by MerryMage
parent bebe7235ae
commit d70ee7c0d1
1 changed files with 23 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
src/backend_x64/emit_x64_vector.cpp
View file

@ -407,7 +407,10 @@ void EmitX64::EmitVectorBroadcastLower8(EmitContext& ctx, IR::Inst* inst) {
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX2)) {
code.vpbroadcastb(a, a);
code.movq(a, a);
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
code.pxor(tmp, tmp);
@ -446,7 +449,9 @@ void EmitX64::EmitVectorBroadcast8(EmitContext& ctx, IR::Inst* inst) {
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX2)) {
code.vpbroadcastb(a, a);
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
code.pxor(tmp, tmp);
@ -465,8 +470,12 @@ void EmitX64::EmitVectorBroadcast16(EmitContext& ctx, IR::Inst* inst) {
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
code.pshuflw(a, a, 0);
code.punpcklqdq(a, a);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX2)) {
code.vpbroadcastw(a, a);
} else {
code.pshuflw(a, a, 0);
code.punpcklqdq(a, a);
}
ctx.reg_alloc.DefineValue(inst, a);
}
@ -476,7 +485,11 @@ void EmitX64::EmitVectorBroadcast32(EmitContext& ctx, IR::Inst* inst) {
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
code.pshufd(a, a, 0);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX2)) {
code.vpbroadcastd(a, a);
} else {
code.pshufd(a, a, 0);
}
ctx.reg_alloc.DefineValue(inst, a);
}
@ -486,7 +499,11 @@ void EmitX64::EmitVectorBroadcast64(EmitContext& ctx, IR::Inst* inst) {
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
code.punpcklqdq(a, a);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX2)) {
code.vpbroadcastq(a, a);
} else {
code.punpcklqdq(a, a);
}
ctx.reg_alloc.DefineValue(inst, a);
}