FEX-Emu/FEX
1
0
Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2025-01-08 22:52:51 +00:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

OpcodeDispatcher: Handle VUNPCKHPS

Browse Source
This commit is contained in:
lioncash 2022-12-30 12:57:15 +00:00
parent 6bc1c3fc30
commit 6341ac6814
5 changed files with 69 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
External/FEXCore/Source/Interface/Core/OpcodeDispatcher.cpp vendored
View File

@ -5819,6 +5819,8 @@ void OpDispatchBuilder::InstallHostSpecificOpcodeHandlers() {
{OPD(1, 0b00, 0x14), 1, &OpDispatchBuilder::VPUNPCKLOp<4>},
{OPD(1, 0b01, 0x14), 1, &OpDispatchBuilder::VPUNPCKLOp<8>},
{OPD(1, 0b00, 0x15), 1, &OpDispatchBuilder::VPUNPCKHOp<4>},
{OPD(1, 0b00, 0x16), 1, &OpDispatchBuilder::VMOVHPOp},
{OPD(1, 0b01, 0x16), 1, &OpDispatchBuilder::VMOVHPOp},
{OPD(1, 0b10, 0x16), 1, &OpDispatchBuilder::VMOVSHDUPOp},

3
External/FEXCore/Source/Interface/Core/OpcodeDispatcher.h vendored
View File

@ -492,6 +492,9 @@ public:
void VPSRLDOp(OpcodeArgs);
void VPSRLDQOp(OpcodeArgs);
template <size_t ElementSize>
void VPUNPCKHOp(OpcodeArgs);
template <size_t ElementSize>
void VPUNPCKLOp(OpcodeArgs);

24
External/FEXCore/Source/Interface/Core/OpcodeDispatcher/Vector.cpp vendored
View File

@ -933,6 +933,30 @@ void OpDispatchBuilder::PUNPCKHOp<4>(OpcodeArgs);
template
void OpDispatchBuilder::PUNPCKHOp<8>(OpcodeArgs);
template <size_t ElementSize>
void OpDispatchBuilder::VPUNPCKHOp(OpcodeArgs) {
const auto SrcSize = GetSrcSize(Op);
const auto Is128Bit = SrcSize == Core::CPUState::XMM_SSE_REG_SIZE;
OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
OrderedNode *Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
OrderedNode *Result{};
if (Is128Bit) {
Result = _VZip2(SrcSize, ElementSize, Src1, Src2);
} else {
OrderedNode *ZipLo = _VZip(SrcSize, ElementSize, Src1, Src2);
OrderedNode *ZipHi = _VZip2(SrcSize, ElementSize, Src1, Src2);
Result = _VInsElement(SrcSize, 16, 0, 1, ZipHi, ZipLo);
}
StoreResult(FPRClass, Op, Result, -1);
}
template
void OpDispatchBuilder::VPUNPCKHOp<4>(OpcodeArgs);
void OpDispatchBuilder::PSHUFBOp(OpcodeArgs) {
auto Size = GetSrcSize(Op);
OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);

2
External/FEXCore/Source/Interface/Core/X86Tables/VEXTables.cpp vendored
View File

@ -38,7 +38,7 @@ void InitializeVEXTables() {
{OPD(1, 0b00, 0x14), 1, X86InstInfo{"VUNPCKLPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
{OPD(1, 0b01, 0x14), 1, X86InstInfo{"VUNPCKLPD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
{OPD(1, 0b00, 0x15), 1, X86InstInfo{"VUNPCKHPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
{OPD(1, 0b00, 0x15), 1, X86InstInfo{"VUNPCKHPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
{OPD(1, 0b01, 0x15), 1, X86InstInfo{"VUNPCKHPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
{OPD(1, 0b00, 0x16), 1, X86InstInfo{"VMOVHPS", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS | FLAGS_VEX_1ST_SRC, 0, nullptr}},

39
unittests/ASM/VEX/vunpckhps.asm Normal file
View File

@ -0,0 +1,39 @@
%ifdef CONFIG
{
"HostFeatures": ["AVX"],
"RegData": {
"XMM2": ["0x7576777855565758", "0x7172737451525354", "0x0000000000000000", "0x0000000000000000"],
"XMM3": ["0x7576777855565758", "0x7172737451525354", "0x0000000000000000", "0x0000000000000000"],
"XMM4": ["0x7576777855565758", "0x7172737451525354", "0x88888888CCCCCCCC", "0x99999999DDDDDDDD"],
"XMM5": ["0x7576777855565758", "0x7172737451525354", "0x88888888CCCCCCCC", "0x99999999DDDDDDDD"]
},
"MemoryRegions": {
"0x100000000": "4096"
}
}
%endif
lea rdx, [rel .data]
vmovaps ymm0, [rdx]
vmovaps ymm1, [rdx + 32]
vunpckhps xmm2, xmm0, xmm1
vunpckhps xmm3, xmm0, [rdx + 32]
vunpckhps ymm4, ymm0, ymm1
vunpckhps ymm5, ymm0, [rdx + 32]
hlt
align 32
.data:
dq 0x4142434445464748
dq 0x5152535455565758
dq 0xFFFFFFFFEEEEEEEE
dq 0xDDDDDDDDCCCCCCCC
dq 0x6162636465666768
dq 0x7172737475767778
dq 0xBBBBBBBBAAAAAAAA
dq 0x9999999988888888