Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Eliminate xblock liveness for shifts #3548

Merged
merged 14 commits into from
Apr 8, 2024
Merged
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
OpcodeDispatcher: rm deferred variable shift flag calcs
Signed-off-by: Alyssa Rosenzweig <[email protected]>
alyssarosenzweig committed Apr 6, 2024
commit 95589f6172a1e94afe618b4f0621f9ecdd0d9987
64 changes: 1 addition & 63 deletions FEXCore/Source/Interface/Core/OpcodeDispatcher.h
Original file line number Diff line number Diff line change
@@ -81,12 +81,9 @@ friend class FEXCore::IR::PassManager;
TYPE_MUL,
TYPE_UMUL,
TYPE_LOGICAL,
TYPE_LSHL,
TYPE_LSHLI,
TYPE_LSHR,
TYPE_LSHRI,
TYPE_LSHRDI,
TYPE_ASHR,
TYPE_ASHRI,
TYPE_BEXTR,
TYPE_BLSI,
@@ -1688,7 +1685,7 @@ friend class FEXCore::IR::PassManager;
OrderedNode *Src1;
} OneSource;

// Logical, LSHL, LSHR, ASHR
// Logical
struct {
OrderedNode *Src1;
OrderedNode *Src2;
@@ -1774,13 +1771,6 @@ friend class FEXCore::IR::PassManager;
PossiblySetNZCVBits |= OldSetNZCVBits;
}

template <typename F>
void CalculateFlags_ShiftVariable(OrderedNode *Shift, F&& CalculateFlags) {
// We are the ones calculating the deferred flags. Don't recurse!
InvalidateDeferredFlags();
Calculate_ShiftVariable(Shift, CalculateFlags);
}

/**
* @name These functions are used by the deferred flag handling while it is calculating and storing flags in to RFLAGs.
* @{ */
@@ -1806,7 +1796,6 @@ friend class FEXCore::IR::PassManager;
void CalculateFlags_ShiftRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
void CalculateFlags_ShiftRightDoubleImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
void CalculateFlags_ShiftRightImmediateCommon(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
void CalculateFlags_SignShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
void CalculateFlags_SignShiftRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
void CalculateFlags_BEXTR(OrderedNode *Src);
void CalculateFlags_BLSI(uint8_t SrcSize, OrderedNode *Src);
@@ -1876,57 +1865,6 @@ friend class FEXCore::IR::PassManager;
};
}

void GenerateFlags_ShiftLeft(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
// Flags need to be used, generate incoming flags first.
CalculateDeferredFlags();

CurrentDeferredFlags = DeferredFlagData {
.Type = FlagsGenerationType::TYPE_LSHL,
.SrcSize = GetSrcSize(Op),
.Res = Res,
.Sources = {
.TwoSource = {
.Src1 = Src1,
.Src2 = Src2,
},
},
};
}

void GenerateFlags_ShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
// Flags need to be used, generate incoming flags first.
CalculateDeferredFlags();

CurrentDeferredFlags = DeferredFlagData {
.Type = FlagsGenerationType::TYPE_LSHR,
.SrcSize = GetSrcSize(Op),
.Res = Res,
.Sources = {
.TwoSource = {
.Src1 = Src1,
.Src2 = Src2,
},
},
};
}

void GenerateFlags_SignShiftRight(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
// Flags need to be used, generate incoming flags first.
CalculateDeferredFlags();

CurrentDeferredFlags = DeferredFlagData {
.Type = FlagsGenerationType::TYPE_ASHR,
.SrcSize = GetSrcSize(Op),
.Res = Res,
.Sources = {
.TwoSource = {
.Src1 = Src1,
.Src2 = Src2,
},
},
};
}

void GenerateFlags_ShiftLeftImmediate(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
// No flags changed if shift is zero.
if (Shift == 0) return;
88 changes: 0 additions & 88 deletions FEXCore/Source/Interface/Core/OpcodeDispatcher/Flags.cpp
Original file line number Diff line number Diff line change
@@ -303,27 +303,13 @@ void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
CurrentDeferredFlags.Sources.TwoSource.Src1,
CurrentDeferredFlags.Sources.TwoSource.Src2);
break;
case FlagsGenerationType::TYPE_LSHL:
CalculateFlags_ShiftLeft(
CurrentDeferredFlags.SrcSize,
CurrentDeferredFlags.Res,
CurrentDeferredFlags.Sources.TwoSource.Src1,
CurrentDeferredFlags.Sources.TwoSource.Src2);
break;
case FlagsGenerationType::TYPE_LSHLI:
CalculateFlags_ShiftLeftImmediate(
CurrentDeferredFlags.SrcSize,
CurrentDeferredFlags.Res,
CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
break;
case FlagsGenerationType::TYPE_LSHR:
CalculateFlags_ShiftRight(
CurrentDeferredFlags.SrcSize,
CurrentDeferredFlags.Res,
CurrentDeferredFlags.Sources.TwoSource.Src1,
CurrentDeferredFlags.Sources.TwoSource.Src2);
break;
case FlagsGenerationType::TYPE_LSHRI:
CalculateFlags_ShiftRightImmediate(
CurrentDeferredFlags.SrcSize,
@@ -338,13 +324,6 @@ void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
break;
case FlagsGenerationType::TYPE_ASHR:
CalculateFlags_SignShiftRight(
CurrentDeferredFlags.SrcSize,
CurrentDeferredFlags.Res,
CurrentDeferredFlags.Sources.TwoSource.Src1,
CurrentDeferredFlags.Sources.TwoSource.Src2);
break;
case FlagsGenerationType::TYPE_ASHRI:
CalculateFlags_SignShiftRightImmediate(
CurrentDeferredFlags.SrcSize,
@@ -580,73 +559,6 @@ void OpDispatchBuilder::CalculateFlags_Logical(uint8_t SrcSize, OrderedNode *Res
SetNZ_ZeroCV(SrcSize, Res);
}

void OpDispatchBuilder::CalculateFlags_ShiftLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res, Src1, Src2](){
const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
SetNZ_ZeroCV(SrcSize, Res);

// Extract the last bit shifted in to CF
auto Size = _Constant(SrcSize * 8);
auto ShiftAmt = SrcSize >= 4 ? _Neg(OpSize, Src2) : _Sub(OpSize, Size, Src2);
auto LastBit = _Lshr(OpSize, Src1, ShiftAmt);
SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit, 0, true);

CalculatePF(Res);

// AF
// Undefined
_InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);

// In the case of left shift. OF is only set from the result of <Top Source Bit> XOR <Top Result Bit>
// When Shift > 1 then OF is undefined
auto OFXor = _Xor(OpSize, Src1, Res);
SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(OFXor, SrcSize * 8 - 1, true);
});
}

void OpDispatchBuilder::CalculateFlags_ShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res, Src1, Src2](){
const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
SetNZ_ZeroCV(SrcSize, Res);

// Extract the last bit shifted in to CF
auto ShiftAmt = _Sub(OpSize::i64Bit, Src2, _Constant(1));
const auto CFSize = IR::SizeToOpSize(std::max<uint8_t>(4u, SrcSize));
auto LastBit = _Lshr(CFSize, Src1, ShiftAmt);
SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit, 0, true);

CalculatePF(Res);

// AF
// Undefined
_InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);

// Only defined when Shift is 1 else undefined
// OF flag is set if a sign change occurred
auto val = _Xor(OpSize, Src1, Res);
SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(val, SrcSize * 8 - 1, true);
});
}

void OpDispatchBuilder::CalculateFlags_SignShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res, Src1, Src2](){
// SF/ZF/OF
SetNZ_ZeroCV(SrcSize, Res);

// Extract the last bit shifted in to CF
const auto CFSize = IR::SizeToOpSize(std::max<uint32_t>(4u, GetOpSize(Src1)));
auto ShiftAmt = _Sub(OpSize::i64Bit, Src2, _Constant(1));
auto LastBit = _Lshr(CFSize, Src1, ShiftAmt);
SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit, 0, true);

CalculatePF(Res);

// AF
// Undefined
_InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
});
}

void OpDispatchBuilder::CalculateFlags_ShiftLeftImmediate(uint8_t SrcSize, OrderedNode *UnmaskedRes, OrderedNode *Src1, uint64_t Shift) {
// No flags changed if shift is zero
if (Shift == 0) return;