[llvm] [LoopFlatten] Invalidate SCEV on unflattening bail (PR #211819)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 24 08:03:20 PDT 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: Arda Serdar Pektezol (pektezol)
<details>
<summary>Changes</summary>
LoopFlatten misses invalidating SCEV on an early !CanFlatten bail.
Fixes #<!-- -->203176
Co-authored-by: Steve Gustaman <stevegustaman@<!-- -->kaist.ac.kr>
---
Full diff: https://github.com/llvm/llvm-project/pull/211819.diff
2 Files Affected:
- (modified) llvm/lib/Transforms/Scalar/LoopFlatten.cpp (+4-1)
- (added) llvm/test/Transforms/LoopFlatten/invalidate-scev-pr203176.ll (+118)
``````````diff
diff --git a/llvm/lib/Transforms/Scalar/LoopFlatten.cpp b/llvm/lib/Transforms/Scalar/LoopFlatten.cpp
index e48c47f1b4b89..da2e81aab3cbd 100644
--- a/llvm/lib/Transforms/Scalar/LoopFlatten.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopFlatten.cpp
@@ -929,8 +929,11 @@ static bool FlattenLoopPair(FlattenInfo &FI, DominatorTree *DT, LoopInfo *LI,
// 'RepeatedInstructionThreshold' is set to only 2, which can probably be
// relaxed. Because this is making a code change (the IV widening, but not
// the flattening), we return true here.
- if (FI.Widened && !CanFlatten)
+ if (FI.Widened && !CanFlatten) {
+ SE->forgetLoop(FI.OuterLoop);
+ SE->forgetBlockAndLoopDispositions();
return true;
+ }
// If we have widened and can perform the transformation, do that here.
if (CanFlatten)
diff --git a/llvm/test/Transforms/LoopFlatten/invalidate-scev-pr203176.ll b/llvm/test/Transforms/LoopFlatten/invalidate-scev-pr203176.ll
new file mode 100644
index 0000000000000..e76736fe96194
--- /dev/null
+++ b/llvm/test/Transforms/LoopFlatten/invalidate-scev-pr203176.ll
@@ -0,0 +1,118 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -S -passes='loop(loop-flatten),loop-vectorize' | FileCheck %s
+
+define i64 @f() {
+; CHECK-LABEL: define i64 @f() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i16> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[VEC_PHI]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536
+; CHECK-NEXT: br i1 [[TMP0]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[TMP1:%.*]] = call i16 @llvm.vector.reduce.or.v2i16(<2 x i16> [[VEC_PHI]])
+; CHECK-NEXT: br label %[[FOR_COND_CLEANUP3_I:.*]]
+; CHECK: [[FOR_COND_CLEANUP3_I]]:
+; CHECK-NEXT: [[SEED_1_I:%.*]] = zext i16 [[TMP1]] to i64
+; CHECK-NEXT: [[FLOW_MIX_I:%.*]] = xor i64 [[SEED_1_I]], 1
+; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
+; CHECK: [[LOOP_HEADER]]:
+; CHECK-NEXT: [[I:%.*]] = phi i64 [ 0, %[[FOR_COND_CLEANUP3_I]] ], [ [[I_NEXT:%.*]], %[[TEST1_EXIT:.*]] ]
+; CHECK-NEXT: [[TRIP_COUNT:%.*]] = add i64 [[FLOW_MIX_I]], 1
+; CHECK-NEXT: [[LOOP_COND:%.*]] = icmp ult i64 [[I]], [[TRIP_COUNT]]
+; CHECK-NEXT: br i1 [[LOOP_COND]], label %[[LOOP_BODY:.*]], label %[[LOOP_EXIT:.*]]
+; CHECK: [[LOOP_BODY]]:
+; CHECK-NEXT: [[LOOPINJECT_1:%.*]] = trunc i64 [[I]] to i32
+; CHECK-NEXT: br i1 false, label %[[FOR_INNER_PREHEADER_I_PREHEADER:.*]], label %[[TEST1_EXIT]]
+; CHECK: [[FOR_INNER_PREHEADER_I_PREHEADER]]:
+; CHECK-NEXT: [[FLATTEN_TRIPCOUNT:%.*]] = mul i32 [[LOOPINJECT_1]], 0
+; CHECK-NEXT: br label %[[FOR_INNER_PREHEADER_I:.*]]
+; CHECK: [[FOR_INNER_PREHEADER_I]]:
+; CHECK-NEXT: [[I_I:%.*]] = phi i32 [ [[INC2_I:%.*]], %[[FOR_OUTER_I:.*]] ], [ 0, %[[FOR_INNER_PREHEADER_I_PREHEADER]] ]
+; CHECK-NEXT: [[FLATTEN_ARRAYIDX_I2:%.*]] = getelementptr i32, ptr null, i32 [[I_I]]
+; CHECK-NEXT: br label %[[FOR_INNER_I:.*]]
+; CHECK: [[FOR_INNER_I]]:
+; CHECK-NEXT: [[J_I:%.*]] = phi i32 [ 0, %[[FOR_INNER_PREHEADER_I]] ]
+; CHECK-NEXT: [[MUL_I1:%.*]] = mul i32 [[I_I]], [[LOOPINJECT_1]]
+; CHECK-NEXT: [[GEP_I:%.*]] = getelementptr i32, ptr null, i32 [[MUL_I1]]
+; CHECK-NEXT: [[ARRAYIDX_I2:%.*]] = getelementptr i32, ptr [[GEP_I]], i32 [[J_I]]
+; CHECK-NEXT: store i32 0, ptr [[GEP_I]], align 4
+; CHECK-NEXT: [[INC1_I:%.*]] = add i32 [[J_I]], 1
+; CHECK-NEXT: [[CMP2_I:%.*]] = icmp ult i32 [[INC1_I]], [[LOOPINJECT_1]]
+; CHECK-NEXT: br label %[[FOR_OUTER_I]]
+; CHECK: [[FOR_OUTER_I]]:
+; CHECK-NEXT: [[INC2_I]] = add i32 [[I_I]], 1
+; CHECK-NEXT: [[CMP1_I:%.*]] = icmp ult i32 [[INC2_I]], [[FLATTEN_TRIPCOUNT]]
+; CHECK-NEXT: br i1 [[CMP1_I]], label %[[FOR_INNER_PREHEADER_I]], label %[[TEST1_EXIT_LOOPEXIT:.*]]
+; CHECK: [[TEST1_EXIT_LOOPEXIT]]:
+; CHECK-NEXT: br label %[[TEST1_EXIT]]
+; CHECK: [[TEST1_EXIT]]:
+; CHECK-NEXT: [[I_NEXT]] = add i64 [[I]], 1
+; CHECK-NEXT: br label %[[LOOP_HEADER]]
+; CHECK: [[LOOP_EXIT]]:
+; CHECK-NEXT: ret i64 0
+;
+entry:
+ br label %for.body4.i
+
+for.cond.cleanup3.i: ; preds = %for.body4.i
+ %seed.1.i = zext i16 %add5.i to i64
+ %flow.mix.i = xor i64 %seed.1.i, 1
+ br label %loop.header
+
+for.body4.i: ; preds = %for.body4.i, %entry
+ %j.011.i = phi i16 [ 0, %entry ], [ %inc.i, %for.body4.i ]
+ %sum.110.i = phi i16 [ 0, %entry ], [ %add5.i, %for.body4.i ]
+ %add5.i = or i16 0, %sum.110.i
+ %inc.i = add i16 %j.011.i, 1
+ %exitcond.not.i = icmp eq i16 %inc.i, 0
+ br i1 %exitcond.not.i, label %for.cond.cleanup3.i, label %for.body4.i
+
+loop.header: ; preds = %test1.exit, %for.cond.cleanup3.i
+ %i = phi i64 [ 0, %for.cond.cleanup3.i ], [ %i.next, %test1.exit ]
+ %trip.count = add i64 %flow.mix.i, 1
+ %loop.cond = icmp ult i64 %i, %trip.count
+ br i1 %loop.cond, label %loop.body, label %loop.exit
+
+loop.body: ; preds = %loop.header
+ %loopinject.1 = trunc i64 %i to i32
+ br i1 false, label %for.inner.preheader.i, label %test1.exit
+
+for.inner.preheader.i: ; preds = %for.outer.i, %loop.body
+ %i.i = phi i32 [ %inc2.i, %for.outer.i ], [ 0, %loop.body ]
+ br label %for.inner.i
+
+for.inner.i: ; preds = %for.inner.i, %for.inner.preheader.i
+ %j.i = phi i32 [ 0, %for.inner.preheader.i ], [ %inc1.i, %for.inner.i ]
+ %mul.i1 = mul i32 %i.i, %loopinject.1
+ %gep.i = getelementptr i32, ptr null, i32 %mul.i1
+ %arrayidx.i2 = getelementptr i32, ptr %gep.i, i32 %j.i
+ store i32 0, ptr %gep.i, align 4
+ %inc1.i = add i32 %j.i, 1
+ %cmp2.i = icmp ult i32 %inc1.i, %loopinject.1
+ br i1 %cmp2.i, label %for.inner.i, label %for.outer.i
+
+for.outer.i: ; preds = %for.inner.i
+ %inc2.i = add i32 %i.i, 1
+ %cmp1.i = icmp ult i32 %inc2.i, 0
+ br i1 %cmp1.i, label %for.inner.preheader.i, label %test1.exit
+
+test1.exit: ; preds = %for.outer.i, %loop.body
+ %i.next = add i64 %i, 1
+ br label %loop.header
+
+loop.exit: ; preds = %loop.header
+ ret i64 0
+
+; uselistorder directives
+ uselistorder i16 %add5.i, { 1, 0 }
+}
+;.
+; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+;.
``````````
</details>
https://github.com/llvm/llvm-project/pull/211819
More information about the llvm-commits
mailing list