FEX-Emu/FEX
1
0
Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2025-02-02 12:37:14 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Arm64: Implement support for non-temporal vector stores

Browse Source
This commit is contained in:
Ryan Houdek 2024-06-29 21:52:22 -07:00
parent 8ab4ab29f8
commit 2b4cec8385
No known key found for this signature in database
2 changed files with 66 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

42
FEXCore/Source/Interface/Core/JIT/Arm64/MemoryOps.cpp
View File

@ -2171,5 +2171,47 @@ DEF_OP(Prefetch) {
prfm(PrefetchType[LUT], MemSrc);
}
DEF_OP(VStoreNonTemporal) {
const auto Op = IROp->C<IR::IROp_VStoreNonTemporal>();
const auto OpSize = IROp->Size;
const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
const auto Is128Bit = OpSize == Core::CPUState::XMM_SSE_REG_SIZE;
const auto Value = GetVReg(Op->Value.ID());
const auto MemReg = GetReg(Op->Addr.ID());
const auto Offset = Op->Offset;
if (Is256Bit) {
LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use VStoreNonTemporal with 256-bit operation");
const auto GoverningPredicate = PRED_TMP_32B.Zeroing();
const auto OffsetScaled = Offset / 32;
stnt1b(Value.Z(), GoverningPredicate, MemReg, OffsetScaled);
} else if (Is128Bit && HostSupportsSVE128) {
const auto GoverningPredicate = PRED_TMP_16B.Zeroing();
const auto OffsetScaled = Offset / 16;
stnt1b(Value.Z(), GoverningPredicate, MemReg, OffsetScaled);
} else {
// Treat the non-temporal store as a regular vector store in this case for compatibility
str(Value.Q(), MemReg, Offset);
}
}
DEF_OP(VStoreNonTemporalPair) {
const auto Op = IROp->C<IR::IROp_VStoreNonTemporalPair>();
const auto OpSize = IROp->Size;
const auto Is128Bit = OpSize == Core::CPUState::XMM_SSE_REG_SIZE;
LOGMAN_THROW_A_FMT(Is128Bit, "This IR operation only operates at 128-bit wide");
const auto ValueLow = GetVReg(Op->ValueLow.ID());
const auto ValueHigh = GetVReg(Op->ValueHigh.ID());
const auto MemReg = GetReg(Op->Addr.ID());
const auto Offset = Op->Offset;
stnp(ValueLow.Q(), ValueHigh.Q(), MemReg, Offset);
}
#undef DEF_OP
} // namespace FEXCore::CPU

24
FEXCore/Source/Interface/IR/IR.json
View File

@ -653,6 +653,30 @@
],
"HasSideEffects": true,
"DestSize": "8"
},
"VStoreNonTemporal u8:#RegisterSize, FPR:$Value, GPR:$Addr, i8:$Offset": {
"Desc": ["Does a non-temporal memory store of a vector.",
"Matches arm64 SVE stnt1b semantics.",
"Specifically weak-memory model ordered to match x86 non-temporal stores."
],
"HasSideEffects": true,
"DestSize": "RegisterSize",
"EmitValidation": [
"_Offset % RegisterSize == 0",
"RegisterSize == FEXCore::IR::OpSize::i128Bit || RegisterSize == FEXCore::IR::OpSize::i256Bit"
]
},
"VStoreNonTemporalPair u8:#RegisterSize, FPR:$ValueLow, FPR:$ValueHigh, GPR:$Addr, i8:$Offset": {
"Desc": ["Does a non-temporal memory store of two vector registers.",
"Matches arm64 stnp semantics.",
"Specifically weak-memory model ordered to match x86 non-temporal stores."
],
"HasSideEffects": true,
"DestSize": "RegisterSize",
"EmitValidation": [
"_Offset % RegisterSize == 0",
"RegisterSize == FEXCore::IR::OpSize::i128Bit"
]
}
},
"Atomic": {