[ConstantFolding][SVE] Do not fold fcmp of denormal without known mode. #150614
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
This is a follow on to llvm#115407 that introduced code which bypasses the splat handling for scalable vectors. To maintain existing tests I have moved the early return until after the splat handling so all vector types are treated equally.
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-llvm-analysis Author: Paul Walker (paulwalker-arm) ChangesThis is a follow on to Full diff: https://github.com/llvm/llvm-project/pull/150614.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index ec78386def24f..acccd0988693e 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1374,7 +1374,7 @@ Constant *llvm::FlushFPConstant(Constant *Operand, const Instruction *Inst,
if (ConstantFP *CFP = dyn_cast<ConstantFP>(Operand))
return flushDenormalConstantFP(CFP, Inst, IsOutput);
- if (isa<ConstantAggregateZero, UndefValue, ConstantExpr>(Operand))
+ if (isa<ConstantAggregateZero, UndefValue>(Operand))
return Operand;
Type *Ty = Operand->getType();
@@ -1390,6 +1390,9 @@ Constant *llvm::FlushFPConstant(Constant *Operand, const Instruction *Inst,
Ty = VecTy->getElementType();
}
+ if (isa<ConstantExpr>(Operand))
+ return Operand;
+
if (const auto *CV = dyn_cast<ConstantVector>(Operand)) {
SmallVector<Constant *, 16> NewElts;
for (unsigned i = 0, e = CV->getNumOperands(); i != e; ++i) {
diff --git a/llvm/test/Transforms/SCCP/no-fold-fcmp-dynamic-denormal-mode-issue114947.ll b/llvm/test/Transforms/SCCP/no-fold-fcmp-dynamic-denormal-mode-issue114947.ll
index 1eb5417d5ffdf..525bf31824f9a 100644
--- a/llvm/test/Transforms/SCCP/no-fold-fcmp-dynamic-denormal-mode-issue114947.ll
+++ b/llvm/test/Transforms/SCCP/no-fold-fcmp-dynamic-denormal-mode-issue114947.ll
@@ -109,7 +109,8 @@ define <vscale x 2 x i1> @fold_fcmp_nondenormal_double_ieee_dynamic_scalable_vec
define <vscale x 2 x i1> @no_fold_fcmp_denormal_double_ieee_dynamic_scalaable_vector_splat() #0 {
; CHECK-LABEL: define <vscale x 2 x i1> @no_fold_fcmp_denormal_double_ieee_dynamic_scalaable_vector_splat(
; CHECK-SAME: ) #[[ATTR0]] {
-; CHECK-NEXT: ret <vscale x 2 x i1> splat (i1 true)
+; CHECK-NEXT: [[CMP:%.*]] = fcmp une <vscale x 2 x double> splat (double 0x8000000000000), zeroinitializer
+; CHECK-NEXT: ret <vscale x 2 x i1> [[CMP]]
;
%cmp = fcmp une <vscale x 2 x double> splat (double 0x8000000000000), zeroinitializer
ret <vscale x 2 x i1> %cmp
|
arsenm
approved these changes
Jul 25, 2025
| @@ -1390,6 +1390,9 @@ Constant *llvm::FlushFPConstant(Constant *Operand, const Instruction *Inst, | |||
| Ty = VecTy->getElementType(); | |||
| } | |||
|
|
|||
| if (isa<ConstantExpr>(Operand)) | |||
| return Operand; | |||
There was a problem hiding this comment.
Is there some reason we don't just return nullptr here?
There was a problem hiding this comment.
I suspect that means we should handle the special case of icmp ueq %x, %x before we flush, not that we should pretend that flushing succeeded. But I guess we can leave it for now.
dtcxzyw
reviewed
Jul 26, 2025
llvm/test/Transforms/SCCP/no-fold-fcmp-dynamic-denormal-mode-issue114947.ll
Outdated
Show resolved
Hide resolved
|
Typo in title |
7c7f44c to
e6341b6
Compare
efriedma-quic
approved these changes
Jul 28, 2025
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.
This is a follow on to
#115407 that introduced code which bypasses the splat handling for scalable vectors. To maintain existing tests I have moved the early return until after the splat handling so all vector types are treated equally.