OpcodeDispatcher: Handle VINSERTF128

This commit is contained in:
lioncash 2022-12-14 04:17:34 +00:00
parent f3d0fa6f60
commit fe7c6da1e2
5 changed files with 62 additions and 1 deletions

View File

@ -5982,6 +5982,9 @@ void OpDispatchBuilder::InstallHostSpecificOpcodeHandlers() {
{OPD(3, 0b01, 0x00), 1, &OpDispatchBuilder::VPERMQOp},
{OPD(3, 0b01, 0x01), 1, &OpDispatchBuilder::VPERMQOp},
{OPD(3, 0b01, 0x06), 1, &OpDispatchBuilder::VPERM2Op},
{OPD(3, 0b01, 0x18), 1, &OpDispatchBuilder::VINSERTOp},
{OPD(3, 0b01, 0x46), 1, &OpDispatchBuilder::VPERM2Op},
};
#undef OPD

View File

@ -418,6 +418,8 @@ public:
template <size_t ElementSize>
void VHADDPOp(OpcodeArgs);
void VINSERTOp(OpcodeArgs);
void VMOVAPS_VMOVAPD_Op(OpcodeArgs);
void VMOVUPS_VMOVUPD_Op(OpcodeArgs);

View File

@ -2939,6 +2939,19 @@ void OpDispatchBuilder::MPSADBWOp(OpcodeArgs) {
StoreResult(FPRClass, Op, Result, -1);
}
void OpDispatchBuilder::VINSERTOp(OpcodeArgs) {
const auto DstSize = GetDstSize(Op);
OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
OrderedNode *Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], 16, Op->Flags, -1);
LOGMAN_THROW_A_FMT(Op->Src[2].IsLiteral(), "Src2 needs to be literal here");
const auto Selector = Op->Src[2].Data.Literal.Value & 1;
OrderedNode *Result = _VInsElement(DstSize, 16, Selector, 0, Src1, Src2);
StoreResult(FPRClass, Op, Result, -1);
}
void OpDispatchBuilder::VPERM2Op(OpcodeArgs) {
const auto DstSize = GetDstSize(Op);
OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);

View File

@ -427,7 +427,7 @@ void InitializeVEXTables() {
{OPD(3, 0b01, 0x16), 1, X86InstInfo{"VPEXTRD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
{OPD(3, 0b01, 0x17), 1, X86InstInfo{"VEXTRACTPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
{OPD(3, 0b01, 0x18), 1, X86InstInfo{"VINSERTF128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
{OPD(3, 0b01, 0x18), 1, X86InstInfo{"VINSERTF128", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
{OPD(3, 0b01, 0x19), 1, X86InstInfo{"VEXTRACTF128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
{OPD(3, 0b01, 0x1D), 1, X86InstInfo{"VCVTPS2PH", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},

View File

@ -0,0 +1,43 @@
%ifdef CONFIG
{
"HostFeatures": ["AVX"],
"RegData": {
"XMM0": ["0x3FF0000000000000", "0xEEEEEEEEEEEEEEEE", "0xFFFFFFFFFFFFFFFF", "0xAAAAAAAAAAAAAAAA"],
"XMM1": ["0xBBBBBBBBBBBBBBBB", "0xDDDDDDDDDDDDDDDD", "0xCCCCCCCCCCCCCCCC", "0x9999999999999999"],
"XMM2": ["0x3FF0000000000000", "0xEEEEEEEEEEEEEEEE", "0xBBBBBBBBBBBBBBBB", "0xDDDDDDDDDDDDDDDD"],
"XMM3": ["0x3FF0000000000000", "0xEEEEEEEEEEEEEEEE", "0xBBBBBBBBBBBBBBBB", "0xDDDDDDDDDDDDDDDD"],
"XMM4": ["0xBBBBBBBBBBBBBBBB", "0xDDDDDDDDDDDDDDDD", "0xFFFFFFFFFFFFFFFF", "0xAAAAAAAAAAAAAAAA"],
"XMM5": ["0xBBBBBBBBBBBBBBBB", "0xDDDDDDDDDDDDDDDD", "0xFFFFFFFFFFFFFFFF", "0xAAAAAAAAAAAAAAAA"]
},
"MemoryRegions": {
"0x100000000": "4096"
}
}
%endif
lea rdx, [rel .data]
vmovapd ymm0, [rdx]
vmovapd ymm1, [rdx + 32]
; Insert into upper lane
vinsertf128 ymm2, ymm0, xmm1, 1
vinsertf128 ymm3, ymm0, [rdx + 32], 1
; Insert into lower lane
vinsertf128 ymm4, ymm0, xmm1, 0
vinsertf128 ymm5, ymm0, [rdx + 32], 0
hlt
align 32
.data:
dq 0x3FF0000000000000
dq 0xEEEEEEEEEEEEEEEE
dq 0xFFFFFFFFFFFFFFFF
dq 0xAAAAAAAAAAAAAAAA
dq 0xBBBBBBBBBBBBBBBB
dq 0xDDDDDDDDDDDDDDDD
dq 0xCCCCCCCCCCCCCCCC
dq 0x9999999999999999