FEX-Emu/FEX
1
0
Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2024-12-14 17:38:47 +00:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

AVX128: Implement support for vperm2{f128,i128}

Browse Source
This commit is contained in:
Ryan Houdek 2024-06-19 21:20:03 -07:00 committed by Alyssa Rosenzweig
parent 5ce6c88a88
commit c1da525110
2 changed files with 23 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
FEXCore/Source/Interface/Core/OpcodeDispatcher.h
View File

@ -1198,6 +1198,8 @@ public:
void AVX128_RestoreAVXState(Ref MemBase);
void AVX128_DefaultAVXState();
void AVX128_VPERM2(OpcodeArgs);
// End of AVX 128-bit implementation
void InvalidOp(OpcodeArgs);

23
FEXCore/Source/Interface/Core/OpcodeDispatcher/AVX_128.cpp
View File

@ -338,7 +338,7 @@ void OpDispatchBuilder::InstallAVX128Handlers() {
{OPD(3, 0b01, 0x02), 1, &OpDispatchBuilder::AVX128_VBLEND<OpSize::i32Bit>},
{OPD(3, 0b01, 0x04), 1, &OpDispatchBuilder::AVX128_VPERMILImm<4>},
{OPD(3, 0b01, 0x05), 1, &OpDispatchBuilder::AVX128_VPERMILImm<8>},
// TODO: {OPD(3, 0b01, 0x06), 1, &OpDispatchBuilder::VPERM2Op},
{OPD(3, 0b01, 0x06), 1, &OpDispatchBuilder::AVX128_VPERM2},
{OPD(3, 0b01, 0x08), 1, &OpDispatchBuilder::AVX128_VectorRound<4>},
{OPD(3, 0b01, 0x09), 1, &OpDispatchBuilder::AVX128_VectorRound<8>},
{OPD(3, 0b01, 0x0A), 1, &OpDispatchBuilder::AVX128_InsertScalarRound<4>},
@ -366,7 +366,7 @@ void OpDispatchBuilder::InstallAVX128Handlers() {
{OPD(3, 0b01, 0x41), 1, &OpDispatchBuilder::AVX128_VDPP<8>},
{OPD(3, 0b01, 0x42), 1, &OpDispatchBuilder::AVX128_VMPSADBW},
// TODO: {OPD(3, 0b01, 0x46), 1, &OpDispatchBuilder::VPERM2Op},
{OPD(3, 0b01, 0x46), 1, &OpDispatchBuilder::AVX128_VPERM2},
{OPD(3, 0b01, 0x4A), 1, &OpDispatchBuilder::AVX128_VectorVariableBlend<4>},
{OPD(3, 0b01, 0x4B), 1, &OpDispatchBuilder::AVX128_VectorVariableBlend<8>},
@ -2070,4 +2070,23 @@ void OpDispatchBuilder::AVX128_DefaultAVXState() {
}
}
void OpDispatchBuilder::AVX128_VPERM2(OpcodeArgs) {
auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, true);
auto Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, true);
const auto Selector = Op->Src[2].Literal();
RefPair Result = AVX128_Zext(LoadZeroVector(OpSize::i128Bit));
Ref Elements[4] = {Src1.Low, Src1.High, Src2.Low, Src2.High};
if ((Selector & 0b00001000) == 0) {
Result.Low = Elements[Selector & 0b11];
}
if ((Selector & 0b10000000) == 0) {
Result.High = Elements[(Selector >> 4) & 0b11];
}
AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
}
} // namespace FEXCore::IR