[llvm] 4959025 - [SLP]Fix non-determinism in reused elements analysis

[Alexey Bataev via llvm-commits]

Author: Alexey Bataev
Date: 2025-03-06T10:12:49-08:00
New Revision: 4959025bbc9e64cc58a1dfa0d9fe1c7162b3ade3

URL: https://github.com/llvm/llvm-project/commit/4959025bbc9e64cc58a1dfa0d9fe1c7162b3ade3
DIFF: https://github.com/llvm/llvm-project/commit/4959025bbc9e64cc58a1dfa0d9fe1c7162b3ade3.diff

LOG: [SLP]Fix non-determinism in reused elements analysis

Need to use consistent storages for unique elements, when going to
iterate over them to avoid non-determinism in reused elements analysis.

Fixes #130082

Added: 
    llvm/test/Transforms/SLPVectorizer/X86/buildvectors-with-same-parents.ll

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 3bb8e38905aae..4decf5bec9514 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -3801,7 +3801,7 @@ class BoUpSLP {
   SetVector<const TreeEntry *> PostponedGathers;
 
   using ValueToGatherNodesMap =
-      DenseMap<Value *, SmallPtrSet<const TreeEntry *, 4>>;
+      DenseMap<Value *, SmallSetVector<const TreeEntry *, 4>>;
   ValueToGatherNodesMap ValueToGatherNodes;
 
   /// A list of the load entries (node indices), which can be vectorized using
@@ -8328,8 +8328,9 @@ class PHIHandler {
       }
       return;
     }
-    SmallDenseMap<BasicBlock *, SmallVector<unsigned>, 4> Blocks;
-    for (unsigned I : seq<unsigned>(0, Main->getNumIncomingValues())) {
+    SmallMapVector<BasicBlock *, SmallVector<unsigned>, 4>
+        Blocks;
+    for (unsigned I : seq<unsigned>(Main->getNumIncomingValues())) {
       BasicBlock *InBB = Main->getIncomingBlock(I);
       if (!DT.isReachableFromEntry(InBB)) {
         Operands[I].assign(Phis.size(), PoisonValue::get(Main->getType()));
@@ -8344,7 +8345,7 @@ class PHIHandler {
         continue;
       }
       auto *P = cast<PHINode>(V);
-      for (unsigned I : seq<unsigned>(0, P->getNumIncomingValues())) {
+      for (unsigned I : seq<unsigned>(P->getNumIncomingValues())) {
         BasicBlock *InBB = P->getIncomingBlock(I);
         if (InBB == Main->getIncomingBlock(I)) {
           if (isa_and_nonnull<PoisonValue>(Operands[I][Idx]))
@@ -8352,17 +8353,18 @@ class PHIHandler {
           Operands[I][Idx] = P->getIncomingValue(I);
           continue;
         }
-        auto It = Blocks.find(InBB);
+        auto *It = Blocks.find(InBB);
         if (It == Blocks.end())
           continue;
         Operands[It->second.front()][Idx] = P->getIncomingValue(I);
       }
     }
     for (const auto &P : Blocks) {
-      if (P.getSecond().size() <= 1)
+      ArrayRef<unsigned> IncomingValues = P.second;
+      if (IncomingValues.size() <= 1)
         continue;
-      unsigned BasicI = P.getSecond().front();
-      for (unsigned I : ArrayRef(P.getSecond()).drop_front()) {
+      unsigned BasicI = IncomingValues.front();
+      for (unsigned I : IncomingValues.drop_front()) {
         assert(all_of(enumerate(Operands[I]),
                       [&](const auto &Data) {
                         return !Data.value() ||

diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/buildvectors-with-same-parents.ll b/llvm/test/Transforms/SLPVectorizer/X86/buildvectors-with-same-parents.ll
new file mode 100644
index 0000000000000..daf9501f1b4ea
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/buildvectors-with-same-parents.ll
@@ -0,0 +1,43 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu -slp-threshold=-99999 < %s | FileCheck %s
+
+define i32 @test(i32 %0) {
+; CHECK-LABEL: define i32 @test(
+; CHECK-SAME: i32 [[TMP0:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> <i32 0, i32 poison>, i32 [[TMP0]], i32 1
+; CHECK-NEXT:    switch i32 0, label %[[BCI_130:.*]] [
+; CHECK-NEXT:      i32 -4, label %[[BCI_96:.*]]
+; CHECK-NEXT:      i32 3, label %[[BCI_130]]
+; CHECK-NEXT:      i32 1, label %[[BCI_130]]
+; CHECK-NEXT:      i32 9, label %[[BCI_108:.*]]
+; CHECK-NEXT:      i32 0, label %[[BCI_130]]
+; CHECK-NEXT:    ]
+; CHECK:       [[BCI_130]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = phi <2 x i32> [ zeroinitializer, %[[BCI_108]] ], [ [[TMP1]], %[[BCI_96]] ], [ [[TMP1]], %[[ENTRY]] ], [ [[TMP1]], %[[ENTRY]] ], [ [[TMP1]], %[[ENTRY]] ], [ [[TMP1]], %[[ENTRY]] ]
+; CHECK-NEXT:    ret i32 0
+; CHECK:       [[BCI_108]]:
+; CHECK-NEXT:    br label %[[BCI_130]]
+; CHECK:       [[BCI_96]]:
+; CHECK-NEXT:    br label %[[BCI_130]]
+;
+entry:
+  switch i32 0, label %bci_130 [
+  i32 -4, label %bci_96
+  i32 3, label %bci_130
+  i32 1, label %bci_130
+  i32 9, label %bci_108
+  i32 0, label %bci_130
+  ]
+
+bci_130:
+  %1 = phi i32 [ 0, %bci_108 ], [ %0, %bci_96 ], [ %0, %entry ], [ %0, %entry ], [ %0, %entry ], [ %0, %entry ]
+  %local_2_10 = phi i32 [ 0, %bci_108 ], [ 0, %bci_96 ], [ 0, %entry ], [ 0, %entry ], [ 0, %entry ], [ 0, %entry ]
+  ret i32 0
+
+bci_108:
+  br label %bci_130
+
+bci_96:
+  br label %bci_130
+}