!fixup dont transform reductions with intermediate stores.

Author: fhahn

llvm/include/llvm/Analysis/TargetTransformInfo.h

@@ -638,6 +638,8 @@ class TargetTransformInfo {
     /// Fall back to the generic logic to determine whether multi-exit unrolling
     /// is profitable if set to false.
     bool RuntimeUnrollMultiExit;
+
+    DenseMap<PHINode *, RecurrenceDescriptor> ParallelizeReductions;
   };
 
   /// Get target-customized preferences for the generic loop unrolling

llvm/lib/Transforms/Utils/LoopUnroll.cpp

@@ -1210,12 +1210,12 @@ MDNode *llvm::GetUnrollMetadata(MDNode *LoopID, StringRef Name) {
 std::optional<RecurrenceDescriptor>
 llvm::canParallelizeReductionWhenUnrolling(PHINode &Phi, Loop *L,
                                            ScalarEvolution *SE) {
-  RecurrenceDescriptor RedDes;
-  if (!RecurrenceDescriptor::isReductionPHI(&Phi, L, RedDes,
+  RecurrenceDescriptor RdxDesc;
+  if (!RecurrenceDescriptor::isReductionPHI(&Phi, L, RdxDesc,
                                             /*DemandedBits=*/nullptr,
                                             /*AC=*/nullptr, /*DT=*/nullptr, SE))
     return std::nullopt;
-  RecurKind RK = RedDes.getRecurrenceKind();
+  RecurKind RK = RdxDesc.getRecurrenceKind();
   // Skip unsupported reductions.
   // TODO: Handle additional reductions, including FP and min-max
   // reductions.
@@ -1225,6 +1225,9 @@ llvm::canParallelizeReductionWhenUnrolling(PHINode &Phi, Loop *L,
       RecurrenceDescriptor::isMinMaxRecurrenceKind(RK))
     return std::nullopt;
 
+  if (RdxDesc.IntermediateStore)
+    return std::nullopt;
+
   // Don't unroll reductions with constant ops; those can be folded to a
   // single induction update.
   if (any_of(cast<Instruction>(Phi.getIncomingValueForBlock(L->getLoopLatch()))
@@ -1239,5 +1242,5 @@ llvm::canParallelizeReductionWhenUnrolling(PHINode &Phi, Loop *L,
           &Phi))
     return std::nullopt;
 
-  return RedDes;
+  return RdxDesc;
 }