[InstCombine] Add check for flag propagation in foldSelectIntoOp
#173735
Conversation
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-llvm-transforms Author: Yunbo Ni (cardigan1008) ChangesFixes #162003 (comment). The current flag propagation assumes that if a select has both For example, assume we have: The current transform would turn This patch add check for operators before flag propagation to avoid Alive2: https://alive2.llvm.org/ce/z/o0MJmS Full diff: https://github.com/llvm/llvm-project/pull/173735.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index f52bac5e600cb..67d1845832725 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -555,8 +555,15 @@ Instruction *InstCombinerImpl::foldSelectIntoOp(SelectInst &SI, Value *TrueVal,
// Examples: -inf + +inf = NaN, -inf - -inf = NaN, 0 * inf = NaN
// Specifically, if the original select has both ninf and nnan, we can
// safely propagate the flag.
+ // Note: This property holds for fadd, fsub, and fmul, but does not
+ // hold for fdiv (e.g. A / Inf == 0.0).
+ bool CanInferFiniteOperandsFromResult =
+ TVI->getOpcode() == Instruction::FAdd ||
+ TVI->getOpcode() == Instruction::FSub ||
+ TVI->getOpcode() == Instruction::FMul;
NewSelFMF.setNoInfs(TVI->hasNoInfs() ||
- (NewSelFMF.noInfs() && NewSelFMF.noNaNs()));
+ (CanInferFiniteOperandsFromResult &&
+ NewSelFMF.noInfs() && NewSelFMF.noNaNs()));
cast<Instruction>(NewSel)->setFastMathFlags(NewSelFMF);
}
NewSel->takeName(TVI);
diff --git a/llvm/test/Transforms/InstCombine/select-binop-foldable-floating-point.ll b/llvm/test/Transforms/InstCombine/select-binop-foldable-floating-point.ll
index c14dd469f1a6e..83fa28a406f75 100644
--- a/llvm/test/Transforms/InstCombine/select-binop-foldable-floating-point.ll
+++ b/llvm/test/Transforms/InstCombine/select-binop-foldable-floating-point.ll
@@ -409,3 +409,14 @@ define float @select_nnan_fdiv_invalid(i1 %cond, float %A, float %B) {
%D = select nnan i1 %cond, float %C, float %A
ret float %D
}
+
+define float @select_fpclass_fdiv_nnan_ninf(i1 %cond, float %A, float %B) {
+; CHECK-LABEL: @select_fpclass_fdiv_nnan_ninf(
+; CHECK-NEXT: [[C:%.*]] = select nnan i1 [[COND:%.*]], float [[B:%.*]], float 1.000000e+00
+; CHECK-NEXT: [[D:%.*]] = fdiv float [[A:%.*]], [[C]]
+; CHECK-NEXT: ret float [[D]]
+;
+ %C = fdiv float %A, %B
+ %D = select ninf nnan i1 %cond, float %C, float %A
+ ret float %D
+}
|
Fixes #162003 (comment).
The current flag propagation assumes that if a select has both
ninfandnnan, then the operands of the folded operation must be finite. While this assumption holds forfadd,fsub, andfmul, it does not hold forfdiv.For example, assume we have:
The current transform would turn
fdiv A, B; select ninf nnan cond, A/B, A;intoA / (select ninf nnan cond, B, 1.0). Ifcondis true, the inner select returnsB = +Inf, and due to the propagatedninf, this becomes poison.This patch add check for operators before flag propagation to avoid
fdivcases.Alive2: https://alive2.llvm.org/ce/z/o0MJmS