InstSimplify: support floating-point equivalences #115152
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
llvm:analysis
|
@llvm/pr-subscribers-llvm-transforms Author: Ramkumar Ramachandra (artagnon) ChangesSince cd16b07 (IR: introduce CmpInst::isEquivalence), there is now an isEquivalence routine in CmpInst that we can use to determine equivalence in simplifySelectWithICmpEq. Implement this, extending the code from integer-equalities to integer and floating-point equivalences. Full diff: https://github.com/llvm/llvm-project/pull/115152.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index 2cb2612bf611e3..198707c5667c8c 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -4617,10 +4617,10 @@ static Value *simplifySelectWithFakeICmpEq(Value *CmpLHS, Value *CmpRHS,
/// Try to simplify a select instruction when its condition operand is an
/// integer equality comparison.
-static Value *simplifySelectWithICmpEq(Value *CmpLHS, Value *CmpRHS,
- Value *TrueVal, Value *FalseVal,
- const SimplifyQuery &Q,
- unsigned MaxRecurse) {
+static Value *simplifySelectWithEquivalence(Value *CmpLHS, Value *CmpRHS,
+ Value *TrueVal, Value *FalseVal,
+ const SimplifyQuery &Q,
+ unsigned MaxRecurse) {
if (simplifyWithOpReplaced(FalseVal, CmpLHS, CmpRHS, Q.getWithoutUndef(),
/* AllowRefinement */ false,
/* DropFlags */ nullptr, MaxRecurse) == TrueVal)
@@ -4635,23 +4635,21 @@ static Value *simplifySelectWithICmpEq(Value *CmpLHS, Value *CmpRHS,
/// Try to simplify a select instruction when its condition operand is an
/// integer comparison.
-static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
- Value *FalseVal,
- const SimplifyQuery &Q,
- unsigned MaxRecurse) {
+static Value *simplifySelectWithCmpCond(Value *CondVal, Value *TrueVal,
+ Value *FalseVal, const SimplifyQuery &Q,
+ unsigned MaxRecurse) {
ICmpInst::Predicate Pred;
Value *CmpLHS, *CmpRHS;
- if (!match(CondVal, m_ICmp(Pred, m_Value(CmpLHS), m_Value(CmpRHS))))
+ if (!match(CondVal, m_Cmp(Pred, m_Value(CmpLHS), m_Value(CmpRHS))))
return nullptr;
+ auto *CI = cast<CmpInst>(CondVal);
if (Value *V = simplifyCmpSelOfMaxMin(CmpLHS, CmpRHS, Pred, TrueVal, FalseVal))
return V;
- // Canonicalize ne to eq predicate.
- if (Pred == ICmpInst::ICMP_NE) {
- Pred = ICmpInst::ICMP_EQ;
+ // Canonicalize the equivalence, of which equality is a subset.
+ if (CI->isEquivalence(/*Invert=*/true))
std::swap(TrueVal, FalseVal);
- }
// Check for integer min/max with a limit constant:
// X > MIN_INT ? X : MIN_INT --> X
@@ -4659,9 +4657,7 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
if (TrueVal->getType()->isIntOrIntVectorTy()) {
Value *X, *Y;
SelectPatternFlavor SPF =
- matchDecomposedSelectPattern(cast<ICmpInst>(CondVal), TrueVal, FalseVal,
- X, Y)
- .Flavor;
+ matchDecomposedSelectPattern(CI, TrueVal, FalseVal, X, Y).Flavor;
if (SelectPatternResult::isMinOrMax(SPF) && Pred == getMinMaxPred(SPF)) {
APInt LimitC = getMinMaxLimit(getInverseMinMaxFlavor(SPF),
X->getType()->getScalarSizeInBits());
@@ -4670,7 +4666,7 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
}
}
- if (Pred == ICmpInst::ICMP_EQ && match(CmpRHS, m_Zero())) {
+ if (CI->isEquality() && match(CmpRHS, m_Zero())) {
Value *X;
const APInt *Y;
if (match(CmpLHS, m_And(m_Value(X), m_APInt(Y))))
@@ -4698,7 +4694,7 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
// (ShAmt == 0) ? X : fshl(X, X, ShAmt) --> fshl(X, X, ShAmt)
// (ShAmt == 0) ? X : fshr(X, X, ShAmt) --> fshr(X, X, ShAmt)
if (match(FalseVal, isRotate) && TrueVal == X && CmpLHS == ShAmt &&
- Pred == ICmpInst::ICMP_EQ)
+ CI->isEquality())
return FalseVal;
// X == 0 ? abs(X) : -abs(X) --> -abs(X)
@@ -4720,12 +4716,12 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
// If we have a scalar equality comparison, then we know the value in one of
// the arms of the select. See if substituting this value into the arm and
// simplifying the result yields the same value as the other arm.
- if (Pred == ICmpInst::ICMP_EQ) {
- if (Value *V = simplifySelectWithICmpEq(CmpLHS, CmpRHS, TrueVal, FalseVal,
- Q, MaxRecurse))
+ if (CI->isEquivalence() || CI->isEquivalence(/*Invert=*/true)) {
+ if (Value *V = simplifySelectWithEquivalence(CmpLHS, CmpRHS, TrueVal,
+ FalseVal, Q, MaxRecurse))
return V;
- if (Value *V = simplifySelectWithICmpEq(CmpRHS, CmpLHS, TrueVal, FalseVal,
- Q, MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(CmpRHS, CmpLHS, TrueVal,
+ FalseVal, Q, MaxRecurse))
return V;
Value *X;
@@ -4734,11 +4730,11 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
if (match(CmpLHS, m_Or(m_Value(X), m_Value(Y))) &&
match(CmpRHS, m_Zero())) {
// (X | Y) == 0 implies X == 0 and Y == 0.
- if (Value *V = simplifySelectWithICmpEq(X, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(X, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
- if (Value *V = simplifySelectWithICmpEq(Y, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(Y, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
}
@@ -4746,11 +4742,11 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
if (match(CmpLHS, m_And(m_Value(X), m_Value(Y))) &&
match(CmpRHS, m_AllOnes())) {
// (X & Y) == -1 implies X == -1 and Y == -1.
- if (Value *V = simplifySelectWithICmpEq(X, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(X, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
- if (Value *V = simplifySelectWithICmpEq(Y, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(Y, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
}
}
@@ -4952,7 +4948,7 @@ static Value *simplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal,
}
if (Value *V =
- simplifySelectWithICmpCond(Cond, TrueVal, FalseVal, Q, MaxRecurse))
+ simplifySelectWithCmpCond(Cond, TrueVal, FalseVal, Q, MaxRecurse))
return V;
if (Value *V = simplifySelectWithFCmp(Cond, TrueVal, FalseVal, Q))
diff --git a/llvm/test/Transforms/InstSimplify/select-equivalence-fp.ll b/llvm/test/Transforms/InstSimplify/select-equivalence-fp.ll
new file mode 100644
index 00000000000000..a59139246b00a6
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/select-equivalence-fp.ll
@@ -0,0 +1,156 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+
+define float @select_fcmp_fsub_oeq(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_oeq(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 0.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_oeq_zero(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_oeq_zero(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp oeq float [[Y:%.*]], 0.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float [[FADD]], float 2.000000e+00
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp oeq float %y, 0.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 2.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_ueq(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_ueq(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp ueq float [[Y:%.*]], 2.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float [[FADD]], float 0.000000e+00
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp ueq float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 0.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_ueq_nnan(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_ueq_nnan(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp nnan ueq float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 0.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_une(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_une(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp une float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 0., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_une_zero(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_une_zero(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp une float [[Y:%.*]], 0.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float 2.000000e+00, float [[FADD]]
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp une float %y, 0.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 2., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_one(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_one(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp one float [[Y:%.*]], 2.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float 0.000000e+00, float [[FADD]]
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp one float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 0., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_one_nnan(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_one_nnan(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp nnan one float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 0., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fadd(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fadd(
+; CHECK-NEXT: ret float 4.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 2.
+ %fadd = fadd float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 4.
+ ret float %sel
+}
+
+define <2 x float> @select_fcmp_fadd_vec(<2 x float> %x, <2 x float> %y) {
+; CHECK-LABEL: @select_fcmp_fadd_vec(
+; CHECK-NEXT: ret <2 x float> <float 4.000000e+00, float 4.000000e+00>
+;
+ %fcmp = fcmp oeq <2 x float> %y, <float 2., float 2.>
+ %fadd = fadd <2 x float> %y, <float 2., float 2.>
+ %sel = select <2 x i1> %fcmp, <2 x float> %fadd, <2 x float> <float 4., float 4.>
+ ret <2 x float> %sel
+}
+
+
+define float @select_fcmp_fdiv(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fdiv(
+; CHECK-NEXT: ret float 1.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 2.
+ %fdiv = fdiv float %y, 2.
+ %sel = select i1 %fcmp, float %fdiv, float 1.
+ ret float %sel
+}
+
+define float @select_fcmp_frem(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_frem(
+; CHECK-NEXT: ret float 1.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 3.
+ %frem = frem float %y, 2.
+ %sel = select i1 %fcmp, float %frem, float 1.
+ ret float %sel
+}
+
+define <2 x float> @select_fcmp_insertelement(<2 x float> %x, <2 x float> %y) {
+; CHECK-LABEL: @select_fcmp_insertelement(
+; CHECK-NEXT: ret <2 x float> <float 4.000000e+00, float 2.000000e+00>
+;
+ %fcmp = fcmp oeq <2 x float> %y, <float 2., float 2.>
+ %insert = insertelement <2 x float> %y, float 4., i64 0
+ %sel = select <2 x i1> %fcmp, <2 x float> %insert, <2 x float> <float 4., float 2.>
+ ret <2 x float> %sel
+}
+
+define <4 x float> @select_fcmp_shufflevector_select(<4 x float> %x, <4 x float> %y) {
+; CHECK-LABEL: @select_fcmp_shufflevector_select(
+; CHECK-NEXT: ret <4 x float> <float poison, float 2.000000e+00, float poison, float 2.000000e+00>
+;
+ %fcmp = fcmp oeq <4 x float> %y, <float 2., float 2., float 2., float 2.>
+ %shuffle = shufflevector <4 x float> %y, <4 x float> poison, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+ %sel = select <4 x i1> %fcmp, <4 x float> %shuffle, <4 x float> <float poison, float 2., float poison, float 2.>
+ ret <4 x float> %sel
+}
+
|
|
@llvm/pr-subscribers-llvm-analysis Author: Ramkumar Ramachandra (artagnon) ChangesSince cd16b07 (IR: introduce CmpInst::isEquivalence), there is now an isEquivalence routine in CmpInst that we can use to determine equivalence in simplifySelectWithICmpEq. Implement this, extending the code from integer-equalities to integer and floating-point equivalences. Full diff: https://github.com/llvm/llvm-project/pull/115152.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index 2cb2612bf611e3..198707c5667c8c 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -4617,10 +4617,10 @@ static Value *simplifySelectWithFakeICmpEq(Value *CmpLHS, Value *CmpRHS,
/// Try to simplify a select instruction when its condition operand is an
/// integer equality comparison.
-static Value *simplifySelectWithICmpEq(Value *CmpLHS, Value *CmpRHS,
- Value *TrueVal, Value *FalseVal,
- const SimplifyQuery &Q,
- unsigned MaxRecurse) {
+static Value *simplifySelectWithEquivalence(Value *CmpLHS, Value *CmpRHS,
+ Value *TrueVal, Value *FalseVal,
+ const SimplifyQuery &Q,
+ unsigned MaxRecurse) {
if (simplifyWithOpReplaced(FalseVal, CmpLHS, CmpRHS, Q.getWithoutUndef(),
/* AllowRefinement */ false,
/* DropFlags */ nullptr, MaxRecurse) == TrueVal)
@@ -4635,23 +4635,21 @@ static Value *simplifySelectWithICmpEq(Value *CmpLHS, Value *CmpRHS,
/// Try to simplify a select instruction when its condition operand is an
/// integer comparison.
-static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
- Value *FalseVal,
- const SimplifyQuery &Q,
- unsigned MaxRecurse) {
+static Value *simplifySelectWithCmpCond(Value *CondVal, Value *TrueVal,
+ Value *FalseVal, const SimplifyQuery &Q,
+ unsigned MaxRecurse) {
ICmpInst::Predicate Pred;
Value *CmpLHS, *CmpRHS;
- if (!match(CondVal, m_ICmp(Pred, m_Value(CmpLHS), m_Value(CmpRHS))))
+ if (!match(CondVal, m_Cmp(Pred, m_Value(CmpLHS), m_Value(CmpRHS))))
return nullptr;
+ auto *CI = cast<CmpInst>(CondVal);
if (Value *V = simplifyCmpSelOfMaxMin(CmpLHS, CmpRHS, Pred, TrueVal, FalseVal))
return V;
- // Canonicalize ne to eq predicate.
- if (Pred == ICmpInst::ICMP_NE) {
- Pred = ICmpInst::ICMP_EQ;
+ // Canonicalize the equivalence, of which equality is a subset.
+ if (CI->isEquivalence(/*Invert=*/true))
std::swap(TrueVal, FalseVal);
- }
// Check for integer min/max with a limit constant:
// X > MIN_INT ? X : MIN_INT --> X
@@ -4659,9 +4657,7 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
if (TrueVal->getType()->isIntOrIntVectorTy()) {
Value *X, *Y;
SelectPatternFlavor SPF =
- matchDecomposedSelectPattern(cast<ICmpInst>(CondVal), TrueVal, FalseVal,
- X, Y)
- .Flavor;
+ matchDecomposedSelectPattern(CI, TrueVal, FalseVal, X, Y).Flavor;
if (SelectPatternResult::isMinOrMax(SPF) && Pred == getMinMaxPred(SPF)) {
APInt LimitC = getMinMaxLimit(getInverseMinMaxFlavor(SPF),
X->getType()->getScalarSizeInBits());
@@ -4670,7 +4666,7 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
}
}
- if (Pred == ICmpInst::ICMP_EQ && match(CmpRHS, m_Zero())) {
+ if (CI->isEquality() && match(CmpRHS, m_Zero())) {
Value *X;
const APInt *Y;
if (match(CmpLHS, m_And(m_Value(X), m_APInt(Y))))
@@ -4698,7 +4694,7 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
// (ShAmt == 0) ? X : fshl(X, X, ShAmt) --> fshl(X, X, ShAmt)
// (ShAmt == 0) ? X : fshr(X, X, ShAmt) --> fshr(X, X, ShAmt)
if (match(FalseVal, isRotate) && TrueVal == X && CmpLHS == ShAmt &&
- Pred == ICmpInst::ICMP_EQ)
+ CI->isEquality())
return FalseVal;
// X == 0 ? abs(X) : -abs(X) --> -abs(X)
@@ -4720,12 +4716,12 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
// If we have a scalar equality comparison, then we know the value in one of
// the arms of the select. See if substituting this value into the arm and
// simplifying the result yields the same value as the other arm.
- if (Pred == ICmpInst::ICMP_EQ) {
- if (Value *V = simplifySelectWithICmpEq(CmpLHS, CmpRHS, TrueVal, FalseVal,
- Q, MaxRecurse))
+ if (CI->isEquivalence() || CI->isEquivalence(/*Invert=*/true)) {
+ if (Value *V = simplifySelectWithEquivalence(CmpLHS, CmpRHS, TrueVal,
+ FalseVal, Q, MaxRecurse))
return V;
- if (Value *V = simplifySelectWithICmpEq(CmpRHS, CmpLHS, TrueVal, FalseVal,
- Q, MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(CmpRHS, CmpLHS, TrueVal,
+ FalseVal, Q, MaxRecurse))
return V;
Value *X;
@@ -4734,11 +4730,11 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
if (match(CmpLHS, m_Or(m_Value(X), m_Value(Y))) &&
match(CmpRHS, m_Zero())) {
// (X | Y) == 0 implies X == 0 and Y == 0.
- if (Value *V = simplifySelectWithICmpEq(X, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(X, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
- if (Value *V = simplifySelectWithICmpEq(Y, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(Y, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
}
@@ -4746,11 +4742,11 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
if (match(CmpLHS, m_And(m_Value(X), m_Value(Y))) &&
match(CmpRHS, m_AllOnes())) {
// (X & Y) == -1 implies X == -1 and Y == -1.
- if (Value *V = simplifySelectWithICmpEq(X, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(X, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
- if (Value *V = simplifySelectWithICmpEq(Y, CmpRHS, TrueVal, FalseVal, Q,
- MaxRecurse))
+ if (Value *V = simplifySelectWithEquivalence(Y, CmpRHS, TrueVal, FalseVal,
+ Q, MaxRecurse))
return V;
}
}
@@ -4952,7 +4948,7 @@ static Value *simplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal,
}
if (Value *V =
- simplifySelectWithICmpCond(Cond, TrueVal, FalseVal, Q, MaxRecurse))
+ simplifySelectWithCmpCond(Cond, TrueVal, FalseVal, Q, MaxRecurse))
return V;
if (Value *V = simplifySelectWithFCmp(Cond, TrueVal, FalseVal, Q))
diff --git a/llvm/test/Transforms/InstSimplify/select-equivalence-fp.ll b/llvm/test/Transforms/InstSimplify/select-equivalence-fp.ll
new file mode 100644
index 00000000000000..a59139246b00a6
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/select-equivalence-fp.ll
@@ -0,0 +1,156 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+
+define float @select_fcmp_fsub_oeq(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_oeq(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 0.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_oeq_zero(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_oeq_zero(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp oeq float [[Y:%.*]], 0.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float [[FADD]], float 2.000000e+00
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp oeq float %y, 0.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 2.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_ueq(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_ueq(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp ueq float [[Y:%.*]], 2.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float [[FADD]], float 0.000000e+00
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp ueq float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 0.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_ueq_nnan(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_ueq_nnan(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp nnan ueq float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 0.
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_une(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_une(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp une float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 0., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_une_zero(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_une_zero(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp une float [[Y:%.*]], 0.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float 2.000000e+00, float [[FADD]]
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp une float %y, 0.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 2., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_one(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_one(
+; CHECK-NEXT: [[FCMP:%.*]] = fcmp one float [[Y:%.*]], 2.000000e+00
+; CHECK-NEXT: [[FADD:%.*]] = fsub float [[Y]], 2.000000e+00
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[FCMP]], float 0.000000e+00, float [[FADD]]
+; CHECK-NEXT: ret float [[SEL]]
+;
+ %fcmp = fcmp one float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 0., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fsub_one_nnan(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fsub_one_nnan(
+; CHECK-NEXT: ret float 0.000000e+00
+;
+ %fcmp = fcmp nnan one float %y, 2.
+ %fadd = fsub float %y, 2.
+ %sel = select i1 %fcmp, float 0., float %fadd
+ ret float %sel
+}
+
+define float @select_fcmp_fadd(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fadd(
+; CHECK-NEXT: ret float 4.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 2.
+ %fadd = fadd float %y, 2.
+ %sel = select i1 %fcmp, float %fadd, float 4.
+ ret float %sel
+}
+
+define <2 x float> @select_fcmp_fadd_vec(<2 x float> %x, <2 x float> %y) {
+; CHECK-LABEL: @select_fcmp_fadd_vec(
+; CHECK-NEXT: ret <2 x float> <float 4.000000e+00, float 4.000000e+00>
+;
+ %fcmp = fcmp oeq <2 x float> %y, <float 2., float 2.>
+ %fadd = fadd <2 x float> %y, <float 2., float 2.>
+ %sel = select <2 x i1> %fcmp, <2 x float> %fadd, <2 x float> <float 4., float 4.>
+ ret <2 x float> %sel
+}
+
+
+define float @select_fcmp_fdiv(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_fdiv(
+; CHECK-NEXT: ret float 1.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 2.
+ %fdiv = fdiv float %y, 2.
+ %sel = select i1 %fcmp, float %fdiv, float 1.
+ ret float %sel
+}
+
+define float @select_fcmp_frem(float %x, float %y) {
+; CHECK-LABEL: @select_fcmp_frem(
+; CHECK-NEXT: ret float 1.000000e+00
+;
+ %fcmp = fcmp oeq float %y, 3.
+ %frem = frem float %y, 2.
+ %sel = select i1 %fcmp, float %frem, float 1.
+ ret float %sel
+}
+
+define <2 x float> @select_fcmp_insertelement(<2 x float> %x, <2 x float> %y) {
+; CHECK-LABEL: @select_fcmp_insertelement(
+; CHECK-NEXT: ret <2 x float> <float 4.000000e+00, float 2.000000e+00>
+;
+ %fcmp = fcmp oeq <2 x float> %y, <float 2., float 2.>
+ %insert = insertelement <2 x float> %y, float 4., i64 0
+ %sel = select <2 x i1> %fcmp, <2 x float> %insert, <2 x float> <float 4., float 2.>
+ ret <2 x float> %sel
+}
+
+define <4 x float> @select_fcmp_shufflevector_select(<4 x float> %x, <4 x float> %y) {
+; CHECK-LABEL: @select_fcmp_shufflevector_select(
+; CHECK-NEXT: ret <4 x float> <float poison, float 2.000000e+00, float poison, float 2.000000e+00>
+;
+ %fcmp = fcmp oeq <4 x float> %y, <float 2., float 2., float 2., float 2.>
+ %shuffle = shufflevector <4 x float> %y, <4 x float> poison, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+ %sel = select <4 x i1> %fcmp, <4 x float> %shuffle, <4 x float> <float poison, float 2., float poison, float 2.>
+ ret <4 x float> %sel
+}
+
|
nikic
reviewed
Nov 6, 2024
arsenm
reviewed
Nov 7, 2024
nikic
requested changes
Nov 7, 2024
There was a problem hiding this comment.
I think this is missing tests for the non-refining code path. Try something along the lines of folding (x == 1) ? y : (x * y) to x * y.
And related to that, I think we need to be extra careful about derefinement here. For integers, the only derefinement we really have to worry about is related to undef and poison, which the existing code already does. But for FP we also have to worry about non-determinism relating to libcalls and NaN behavior.
For example, if we have something like x == C : sin(C) : sin(x), then we should generally not fold that to sin(x), because we cannot guarantee that sin(x) at runtime folds to the same sin(C) that the host libm produced.
We should be able to avoid this by passing the non-determinism flag to the constant folding API in simplifyWithOpReplaced, for the non-refining case.
In prepraration to extend select folding to include floating-points using CmpInst::isEquivalence, cover it with tests first.
Since cd16b07 (IR: introduce CmpInst::isEquivalence), there is now an isEquivalence routine in CmpInst that we can use to determine equivalence in simplifySelectWithICmpEq. Implement this, extending the code from integer-equalities to integer and floating-point equivalences.
Thanks for catching this subtle point. I do think |
nikic
reviewed
Nov 11, 2024
nikic
reviewed
Nov 11, 2024
nikic
reviewed
Nov 11, 2024
|
Alive reports a verification failure for one of the tests: https://alive2.llvm.org/ce/z/eHm-rS ----------------------------------------
define half @src(half %x, half %y) {
#0:
%fcmp = fcmp oeq half %x, 0x3c00
%fmul = fmul half %y, %x
%sel = select i1 %fcmp, half %y, half %fmul
ret half %sel
}
=>
define half @tgt(half %x, half %y) {
#0:
%fmul = fmul half %y, %x
ret half %fmul
}
Transformation doesn't verify!
ERROR: Value mismatch
Example:
half %x = #x3c00 (1)
half %y = #xfc80 (SNaN)
Source:
i1 %fcmp = #x1 (1)
half %fmul = #x7e00 (QNaN)
half %sel = #xfc80 (SNaN)
Target:
half %fmul = #x7c80 (SNaN)
Source value: #xfc80 (SNaN)
Target value: #x7c80 (SNaN)The problem is that the fmul may return a different NaN value than just passing the value through. cc @nunoplopes |
|
Yep, the optimization needs the 'nnan' flag. @jcranmer-intel to confirm. |
|
I was wondering why this gets folded at all, given that this is the AllowRefinement=false case. I think the reason is this code: llvm-project/llvm/lib/Analysis/InstructionSimplify.cpp Lines 4347 to 4352 in 76f2589 We assume that folding |
|
I've put up #120098 to fix this. |
nikic
added a commit
that referenced
this pull request
Dec 17, 2024
If x is NaN, then fmul (x, 1) may produce a different NaN value. Our float semantics explicitly permit folding fmul (x, 1) to x, but we can't do this when we're replacing a select input, as selects are supposed to preserve the exact bitwise value. Fixes #115152 (comment).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Since cd16b07 (IR: introduce CmpInst::isEquivalence), there is now an isEquivalence routine in CmpInst that we can use to determine equivalence in simplifySelectWithICmpEq. Implement this, extending the code from integer-equalities to integer and floating-point equivalences.