FEX-Emu/FEX
1
0
Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2024-12-15 09:59:28 +00:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

OpcodeDispatcher: drop ZeroMultipleFlags

Browse Source 
lot of complexity for only a single interesting case. we can massively simplify.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
This commit is contained in:
Alyssa Rosenzweig 2024-04-01 13:47:43 -04:00
parent e2a095372e
commit f8b68d8b5a
3 changed files with 11 additions and 78 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -1422,7 +1422,7 @@ private:
SetRFLAG<FEXCore::X86State::RFLAG_AF_RAW_LOC>(_Constant(Constant << 4));
}
void ZeroMultipleFlags(uint32_t BitMask);
void ZeroPF_AF();
CondClassType CondForNZCVBit(unsigned BitOffset, bool Invert) {
switch (BitOffset) {

72
FEXCore/Source/Interface/Core/OpcodeDispatcher/Flags.cpp
View File

@ -38,61 +38,10 @@ constexpr std::array<uint32_t, 17> FlagOffsets = {
FEXCore::X86State::RFLAG_ID_LOC,
};
void OpDispatchBuilder::ZeroMultipleFlags(uint32_t FlagsMask) {
auto ZeroConst = _Constant(0);
if (ContainsNZCV(FlagsMask)) {
// NZCV is stored packed together.
// It's more optimal to zero NZCV with move+bic instead of multiple bics.
auto NZCVFlagsMask = FlagsMask & FullNZCVMask;
if (NZCVFlagsMask == FullNZCVMask) {
ZeroNZCV();
}
else {
const auto IndexMask = NZCVIndexMask(FlagsMask);
if (std::popcount(NZCVFlagsMask) == 1) {
// It's more optimal to store only one here.
for (size_t i = 0; NZCVFlagsMask && i < FlagOffsets.size(); ++i) {
const auto FlagOffset = FlagOffsets[i];
const auto FlagMask = 1U << FlagOffset;
if (!(FlagMask & NZCVFlagsMask)) {
continue;
}
SetRFLAG(ZeroConst, FlagOffset);
NZCVFlagsMask &= ~(FlagMask);
}
}
else {
auto IndexMaskConstant = _Constant(IndexMask);
auto NewNZCV = _Andn(OpSize::i64Bit, GetNZCV(), IndexMaskConstant);
SetNZCV(NewNZCV);
}
// Unset the possibly set bits.
PossiblySetNZCVBits &= ~IndexMask;
}
// Handled NZCV, so remove it from the mask.
FlagsMask &= ~FullNZCVMask;
}
void OpDispatchBuilder::ZeroPF_AF() {
// PF is stored inverted, so invert it when we zero.
if (FlagsMask & (1u << X86State::RFLAG_PF_RAW_LOC)) {
SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(_Constant(1));
FlagsMask &= ~(1u << X86State::RFLAG_PF_RAW_LOC);
}
// Handle remaining masks.
for (size_t i = 0; FlagsMask && i < FlagOffsets.size(); ++i) {
const auto FlagOffset = FlagOffsets[i];
const auto FlagMask = 1U << FlagOffset;
if (!(FlagMask & FlagsMask)) {
continue;
}
SetRFLAG(ZeroConst, FlagOffset);
FlagsMask &= ~(FlagMask);
}
SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(_Constant(1));
SetAF(0);
}
void OpDispatchBuilder::SetPackedRFLAG(bool Lower8, OrderedNode *Src) {
@ -994,9 +943,7 @@ void OpDispatchBuilder::CalculateFlags_POPCOUNT(OrderedNode *Result) {
// is in the range [0, 63]. In particular, it is always positive. So a
// combined NZ test will correctly zero SF/CF/OF while setting ZF.
SetNZ_ZeroCV(OpSize::i32Bit, Result);
ZeroMultipleFlags((1U << X86State::RFLAG_AF_RAW_LOC) |
(1U << X86State::RFLAG_PF_RAW_LOC));
ZeroPF_AF();
}
void OpDispatchBuilder::CalculateFlags_BZHI(uint8_t SrcSize, OrderedNode *Result, OrderedNode *Src) {
@ -1022,15 +969,10 @@ void OpDispatchBuilder::CalculateFlags_ZCNT(uint8_t SrcSize, OrderedNode *Result
void OpDispatchBuilder::CalculateFlags_RDRAND(OrderedNode *Src) {
// OF, SF, ZF, AF, PF all zero
ZeroNZCV();
ZeroPF_AF();
// CF is set to the incoming source
uint32_t FlagsMaskToZero =
FullNZCVMask |
(1U << X86State::RFLAG_AF_RAW_LOC) |
(1U << X86State::RFLAG_PF_RAW_LOC);
ZeroMultipleFlags(FlagsMaskToZero);
SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Src);
}

15
FEXCore/Source/Interface/Core/OpcodeDispatcher/Vector.cpp
View File

@ -4594,11 +4594,7 @@ void OpDispatchBuilder::PTestOp(OpcodeArgs) {
SetNZ_ZeroCV(32, Test1);
SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Test2);
uint32_t FlagsMaskToZero =
(1U << X86State::RFLAG_PF_RAW_LOC) |
(1U << X86State::RFLAG_AF_RAW_LOC);
ZeroMultipleFlags(FlagsMaskToZero);
ZeroPF_AF();
}
void OpDispatchBuilder::VTESTOpImpl(OpcodeArgs, size_t ElementSize) {
@ -4635,8 +4631,7 @@ void OpDispatchBuilder::VTESTOpImpl(OpcodeArgs, size_t ElementSize) {
SetNZ_ZeroCV(32, AndGPR);
SetRFLAG<X86State::RFLAG_CF_RAW_LOC>(CFResult);
ZeroMultipleFlags((1U << X86State::RFLAG_PF_RAW_LOC) |
(1U << X86State::RFLAG_AF_RAW_LOC));
ZeroPF_AF();
}
template <size_t ElementSize>
@ -5568,11 +5563,7 @@ void OpDispatchBuilder::PCMPXSTRXOpImpl(OpcodeArgs, bool IsExplicit, bool IsMask
SetRFLAG<X86State::RFLAG_CF_RAW_LOC>(GetFlagBit(18));
SetRFLAG<X86State::RFLAG_OF_RAW_LOC>(GetFlagBit(19));
uint32_t FlagsMaskToZero =
(1U << X86State::RFLAG_PF_RAW_LOC) |
(1U << X86State::RFLAG_AF_RAW_LOC);
ZeroMultipleFlags(FlagsMaskToZero);
ZeroPF_AF();
}
void OpDispatchBuilder::VPCMPESTRIOp(OpcodeArgs) {