diff options
| author | Florian Hahn <flo@fhahn.com> | 2025-07-18 22:58:19 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-07-18 21:58:19 +0100 |
| commit | 004c67ea257039e4e98abc26dd4ac6e8f3d7a171 (patch) | |
| tree | 24d7ae02719807bb7d7aa3b7a6ffeb57c75c0913 /llvm/lib/Transforms/Utils/LoopUtils.cpp | |
| parent | 695660cdfd1ca65cd6e02e6950d10c990dfa0036 (diff) | |
| download | llvm-004c67ea257039e4e98abc26dd4ac6e8f3d7a171.zip llvm-004c67ea257039e4e98abc26dd4ac6e8f3d7a171.tar.gz llvm-004c67ea257039e4e98abc26dd4ac6e8f3d7a171.tar.bz2 | |
[LV] Vectorize maxnum/minnum w/o fast-math flags. (#148239)
Update LV to vectorize maxnum/minnum reductions without fast-math flags,
by adding an extra check in the loop if any inputs to maxnum/minnum are
NaN, due to maxnum/minnum behavior w.r.t to signaling NaNs. Signed-zeros
are already handled consistently by maxnum/minnum.
If any input is NaN,
*exit the vector loop,
*compute the reduction result up to the vector iteration that contained
NaN inputs and
* resume in the scalar loop
New recurrence kinds are added for reductions using maxnum/minnum
without fast-math flags.
PR: https://github.com/llvm/llvm-project/pull/148239
Diffstat (limited to 'llvm/lib/Transforms/Utils/LoopUtils.cpp')
| -rw-r--r-- | llvm/lib/Transforms/Utils/LoopUtils.cpp | 10 |
1 files changed, 8 insertions, 2 deletions
diff --git a/llvm/lib/Transforms/Utils/LoopUtils.cpp b/llvm/lib/Transforms/Utils/LoopUtils.cpp index 200d1fb..e7623aa 100644 --- a/llvm/lib/Transforms/Utils/LoopUtils.cpp +++ b/llvm/lib/Transforms/Utils/LoopUtils.cpp @@ -938,8 +938,10 @@ constexpr Intrinsic::ID llvm::getReductionIntrinsicID(RecurKind RK) { case RecurKind::UMin: return Intrinsic::vector_reduce_umin; case RecurKind::FMax: + case RecurKind::FMaxNum: return Intrinsic::vector_reduce_fmax; case RecurKind::FMin: + case RecurKind::FMinNum: return Intrinsic::vector_reduce_fmin; case RecurKind::FMaximum: return Intrinsic::vector_reduce_fmaximum; @@ -1037,8 +1039,10 @@ Intrinsic::ID llvm::getMinMaxReductionIntrinsicOp(RecurKind RK) { case RecurKind::SMax: return Intrinsic::smax; case RecurKind::FMin: + case RecurKind::FMinNum: return Intrinsic::minnum; case RecurKind::FMax: + case RecurKind::FMaxNum: return Intrinsic::maxnum; case RecurKind::FMinimum: return Intrinsic::minimum; @@ -1096,9 +1100,9 @@ Value *llvm::createMinMaxOp(IRBuilderBase &Builder, RecurKind RK, Value *Left, Value *Right) { Type *Ty = Left->getType(); if (Ty->isIntOrIntVectorTy() || - (RK == RecurKind::FMinimum || RK == RecurKind::FMaximum || + (RK == RecurKind::FMinNum || RK == RecurKind::FMaxNum || + RK == RecurKind::FMinimum || RK == RecurKind::FMaximum || RK == RecurKind::FMinimumNum || RK == RecurKind::FMaximumNum)) { - // TODO: Add float minnum/maxnum support when FMF nnan is set. Intrinsic::ID Id = getMinMaxReductionIntrinsicOp(RK); return Builder.CreateIntrinsic(Ty, Id, {Left, Right}, nullptr, "rdx.minmax"); @@ -1308,6 +1312,8 @@ Value *llvm::createSimpleReduction(IRBuilderBase &Builder, Value *Src, case RecurKind::UMin: case RecurKind::FMax: case RecurKind::FMin: + case RecurKind::FMinNum: + case RecurKind::FMaxNum: case RecurKind::FMinimum: case RecurKind::FMaximum: case RecurKind::FMinimumNum: |