[LV] Use vscale for tuning to improve branch weight estimates #144733
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
In addBranchWeightToMiddleTerminator we attempt to add branch weights to the middle block terminator. We pessimistically assume vscale=1, whereas we can improve the estimate by using the value of vscale used for tuning.
|
@llvm/pr-subscribers-vectorizers Author: David Sherwood (david-arm) ChangesIn addBranchWeightToMiddleTerminator we attempt to add branch weights to the middle block terminator. We pessimistically assume vscale=1, whereas we can improve the estimate by using the value of vscale used for tuning. Full diff: https://github.com/llvm/llvm-project/pull/144733.diff 4 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index f1470fd1f7314..3445cfa355e8a 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7330,9 +7330,11 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
VPlanTransforms::runPass(VPlanTransforms::unrollByUF, BestVPlan, BestUF,
OrigLoop->getHeader()->getContext());
VPlanTransforms::runPass(VPlanTransforms::materializeBroadcasts, BestVPlan);
- if (hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator()))
+ if (hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator())) {
+ std::optional<unsigned> VScale = CM.getVScaleForTuning();
VPlanTransforms::runPass(VPlanTransforms::addBranchWeightToMiddleTerminator,
- BestVPlan, BestVF);
+ BestVPlan, BestVF, VScale);
+ }
VPlanTransforms::optimizeForVFAndUF(BestVPlan, BestVF, BestUF, PSE);
VPlanTransforms::simplifyRecipes(BestVPlan, *Legal->getWidestInductionType());
VPlanTransforms::narrowInterleaveGroups(
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 05a0e15f9a199..f474ee94ec5e5 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -3213,8 +3213,8 @@ void VPlanTransforms::narrowInterleaveGroups(VPlan &Plan, ElementCount VF,
/// Add branch weight metadata, if the \p Plan's middle block is terminated by a
/// BranchOnCond recipe.
-void VPlanTransforms::addBranchWeightToMiddleTerminator(VPlan &Plan,
- ElementCount VF) {
+void VPlanTransforms::addBranchWeightToMiddleTerminator(
+ VPlan &Plan, ElementCount VF, std::optional<unsigned> VScale) {
VPBasicBlock *MiddleVPBB = Plan.getMiddleBlock();
auto *MiddleTerm =
dyn_cast_or_null<VPInstruction>(MiddleVPBB->getTerminator());
@@ -3226,6 +3226,8 @@ void VPlanTransforms::addBranchWeightToMiddleTerminator(VPlan &Plan,
"must have a BranchOnCond");
// Assume that `TripCount % VectorStep ` is equally distributed.
unsigned VectorStep = Plan.getUF() * VF.getKnownMinValue();
+ if (VF.isScalable() && VScale.has_value())
+ VectorStep *= *VScale;
assert(VectorStep > 0 && "trip count should not be zero");
MDBuilder MDB(Plan.getScalarHeader()->getIRBasicBlock()->getContext());
MDNode *BranchWeights =
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 5a03bdb7c6882..32c4c48275121 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -237,7 +237,8 @@ struct VPlanTransforms {
/// Add branch weight metadata, if the \p Plan's middle block is terminated by
/// a BranchOnCond recipe.
- static void addBranchWeightToMiddleTerminator(VPlan &Plan, ElementCount VF);
+ static void addBranchWeightToMiddleTerminator(VPlan &Plan, ElementCount VF,
+ std::optional<unsigned> VScale);
};
} // namespace llvm
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll b/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll
index 9435c544fc812..1f619898ea788 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll
@@ -92,7 +92,7 @@ for.cond.cleanup: ; preds = %for.body
; CHECK-V1-IC1: [[LOOP1]] = distinct !{[[LOOP1]], [[META2:![0-9]+]], [[META3:![0-9]+]]}
; CHECK-V1-IC1: [[META2]] = !{!"llvm.loop.isvectorized", i32 1}
; CHECK-V1-IC1: [[META3]] = !{!"llvm.loop.unroll.runtime.disable"}
-; CHECK-V1-IC1: [[PROF4]] = !{!"branch_weights", i32 1, i32 3}
+; CHECK-V1-IC1: [[PROF4]] = !{!"branch_weights", i32 1, i32 7}
; CHECK-V1-IC1: [[PROF5]] = !{!"branch_weights", i32 0, i32 0}
; CHECK-V1-IC1: [[LOOP6]] = distinct !{[[LOOP6]], [[META3]], [[META2]]}
;.
|
|
@llvm/pr-subscribers-llvm-transforms Author: David Sherwood (david-arm) ChangesIn addBranchWeightToMiddleTerminator we attempt to add branch weights to the middle block terminator. We pessimistically assume vscale=1, whereas we can improve the estimate by using the value of vscale used for tuning. Full diff: https://github.com/llvm/llvm-project/pull/144733.diff 4 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index f1470fd1f7314..3445cfa355e8a 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7330,9 +7330,11 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
VPlanTransforms::runPass(VPlanTransforms::unrollByUF, BestVPlan, BestUF,
OrigLoop->getHeader()->getContext());
VPlanTransforms::runPass(VPlanTransforms::materializeBroadcasts, BestVPlan);
- if (hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator()))
+ if (hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator())) {
+ std::optional<unsigned> VScale = CM.getVScaleForTuning();
VPlanTransforms::runPass(VPlanTransforms::addBranchWeightToMiddleTerminator,
- BestVPlan, BestVF);
+ BestVPlan, BestVF, VScale);
+ }
VPlanTransforms::optimizeForVFAndUF(BestVPlan, BestVF, BestUF, PSE);
VPlanTransforms::simplifyRecipes(BestVPlan, *Legal->getWidestInductionType());
VPlanTransforms::narrowInterleaveGroups(
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 05a0e15f9a199..f474ee94ec5e5 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -3213,8 +3213,8 @@ void VPlanTransforms::narrowInterleaveGroups(VPlan &Plan, ElementCount VF,
/// Add branch weight metadata, if the \p Plan's middle block is terminated by a
/// BranchOnCond recipe.
-void VPlanTransforms::addBranchWeightToMiddleTerminator(VPlan &Plan,
- ElementCount VF) {
+void VPlanTransforms::addBranchWeightToMiddleTerminator(
+ VPlan &Plan, ElementCount VF, std::optional<unsigned> VScale) {
VPBasicBlock *MiddleVPBB = Plan.getMiddleBlock();
auto *MiddleTerm =
dyn_cast_or_null<VPInstruction>(MiddleVPBB->getTerminator());
@@ -3226,6 +3226,8 @@ void VPlanTransforms::addBranchWeightToMiddleTerminator(VPlan &Plan,
"must have a BranchOnCond");
// Assume that `TripCount % VectorStep ` is equally distributed.
unsigned VectorStep = Plan.getUF() * VF.getKnownMinValue();
+ if (VF.isScalable() && VScale.has_value())
+ VectorStep *= *VScale;
assert(VectorStep > 0 && "trip count should not be zero");
MDBuilder MDB(Plan.getScalarHeader()->getIRBasicBlock()->getContext());
MDNode *BranchWeights =
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 5a03bdb7c6882..32c4c48275121 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -237,7 +237,8 @@ struct VPlanTransforms {
/// Add branch weight metadata, if the \p Plan's middle block is terminated by
/// a BranchOnCond recipe.
- static void addBranchWeightToMiddleTerminator(VPlan &Plan, ElementCount VF);
+ static void addBranchWeightToMiddleTerminator(VPlan &Plan, ElementCount VF,
+ std::optional<unsigned> VScale);
};
} // namespace llvm
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll b/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll
index 9435c544fc812..1f619898ea788 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/check-prof-info.ll
@@ -92,7 +92,7 @@ for.cond.cleanup: ; preds = %for.body
; CHECK-V1-IC1: [[LOOP1]] = distinct !{[[LOOP1]], [[META2:![0-9]+]], [[META3:![0-9]+]]}
; CHECK-V1-IC1: [[META2]] = !{!"llvm.loop.isvectorized", i32 1}
; CHECK-V1-IC1: [[META3]] = !{!"llvm.loop.unroll.runtime.disable"}
-; CHECK-V1-IC1: [[PROF4]] = !{!"branch_weights", i32 1, i32 3}
+; CHECK-V1-IC1: [[PROF4]] = !{!"branch_weights", i32 1, i32 7}
; CHECK-V1-IC1: [[PROF5]] = !{!"branch_weights", i32 0, i32 0}
; CHECK-V1-IC1: [[LOOP6]] = distinct !{[[LOOP6]], [[META3]], [[META2]]}
;.
|
david-arm
commented
Jun 18, 2025
| @@ -7330,9 +7330,11 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan( | |||
| VPlanTransforms::runPass(VPlanTransforms::unrollByUF, BestVPlan, BestUF, | |||
| OrigLoop->getHeader()->getContext()); | |||
| VPlanTransforms::runPass(VPlanTransforms::materializeBroadcasts, BestVPlan); | |||
| if (hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator())) | |||
| if (hasBranchWeightMD(*OrigLoop->getLoopLatch()->getTerminator())) { | |||
| std::optional<unsigned> VScale = CM.getVScaleForTuning(); | |||
There was a problem hiding this comment.
NOTE: I'm forced to create a temporary variable here because runPass takes values by reference.
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.
In addBranchWeightToMiddleTerminator we attempt to add branch weights to the middle block terminator. We pessimistically assume vscale=1, whereas we can improve the estimate by using the value of vscale used for tuning.