[llvm] 54c6e1c - [SLP] Move a non-power-of-two bailout down slightly
Philip Reames via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 11 14:33:53 PDT 2024
Author: Philip Reames
Date: 2024-09-11T14:33:45-07:00
New Revision: 54c6e1c3f51758469cc06cbcc2ad28af210fc004
URL: https://github.com/llvm/llvm-project/commit/54c6e1c3f51758469cc06cbcc2ad28af210fc004
DIFF: https://github.com/llvm/llvm-project/commit/54c6e1c3f51758469cc06cbcc2ad28af210fc004.diff
LOG: [SLP] Move a non-power-of-two bailout down slightly
The first part of CheckForShuffledLoads isn't doing any subvector
analysis, so it's perfectly safe for arbitrary VL.
Added:
Modified:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index c89a50fc7bd429..00d9f2909d71e2 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -4819,12 +4819,6 @@ BoUpSLP::LoadsState BoUpSLP::canVectorizeLoads(
// representation is better than just gather.
auto CheckForShuffledLoads = [&, &TTI = *TTI](Align CommonAlignment,
bool ProfitableGatherPointers) {
- // FIXME: The following code has not been updated for non-power-of-2
- // vectors. The splitting logic here does not cover the original
- // vector if the vector factor is not a power of two. FIXME
- if (!has_single_bit(VL.size()))
- return false;
-
// Compare masked gather cost and loads + insert subvector costs.
TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
auto [ScalarGEPCost, VectorGEPCost] =
@@ -4874,6 +4868,13 @@ BoUpSLP::LoadsState BoUpSLP::canVectorizeLoads(
constexpr unsigned ListLimit = 4;
if (!TryRecursiveCheck || VL.size() < ListLimit)
return MaskedGatherCost - GatherCost >= -SLPCostThreshold;
+
+ // FIXME: The following code has not been updated for non-power-of-2
+ // vectors. The splitting logic here does not cover the original
+ // vector if the vector factor is not a power of two. FIXME
+ if (!has_single_bit(VL.size()))
+ return false;
+
unsigned Sz = DL->getTypeSizeInBits(ScalarTy);
unsigned MinVF = getMinVF(2 * Sz);
DemandedElts.clearAllBits();
More information about the llvm-commits
mailing list