1 files changed, 35 insertions, 8 deletions

diff --git a/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 055fbb00871f..601d2454c1e1 100644
--- a/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/contrib/llvm-project/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -7379,6 +7379,8 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
             continue;
           if (Idx >= static_cast<int>(CommonVF))
             Idx = E1Mask[Idx - CommonVF] + VF;
+          else
+            Idx = E1Mask[Idx];
         }
         CommonVF = VF;
       }
@@ -12986,8 +12988,8 @@ void BoUpSLP::scheduleBlock(BlockScheduling *BS) {
     for (ScheduleData *BundleMember = Picked; BundleMember;
          BundleMember = BundleMember->NextInBundle) {
       Instruction *PickedInst = BundleMember->Inst;
-      if (PickedInst->getNextNode() != LastScheduledInst)
-        PickedInst->moveBefore(LastScheduledInst);
+      if (PickedInst->getNextNonDebugInstruction() != LastScheduledInst)
+        PickedInst->moveAfter(LastScheduledInst->getPrevNode());
       LastScheduledInst = PickedInst;
     }
 
@@ -13181,7 +13183,7 @@ void BoUpSLP::computeMinimumValueSizes() {
   // We only attempt to truncate integer expressions.
   auto &TreeRoot = VectorizableTree[0]->Scalars;
   auto *TreeRootIT = dyn_cast<IntegerType>(TreeRoot[0]->getType());
-  if (!TreeRootIT)
+  if (!TreeRootIT || VectorizableTree.front()->State == TreeEntry::NeedToGather)
     return;
 
   // Ensure the roots of the vectorizable tree don't form a cycle.
@@ -14792,8 +14794,17 @@ public:
             LocalExternallyUsedValues[RdxVal];
         // Update LocalExternallyUsedValues for the scalar, replaced by
         // extractelement instructions.
+        DenseMap<Value *, Value *> ReplacementToExternal;
+        for (const std::pair<Value *, Value *> &Pair : ReplacedExternals)
+          ReplacementToExternal.try_emplace(Pair.second, Pair.first);
         for (const std::pair<Value *, Value *> &Pair : ReplacedExternals) {
-          auto *It = ExternallyUsedValues.find(Pair.first);
+          Value *Ext = Pair.first;
+          auto RIt = ReplacementToExternal.find(Ext);
+          while (RIt != ReplacementToExternal.end()) {
+            Ext = RIt->second;
+            RIt = ReplacementToExternal.find(Ext);
+          }
+          auto *It = ExternallyUsedValues.find(Ext);
           if (It == ExternallyUsedValues.end())
             continue;
           LocalExternallyUsedValues[Pair.second].append(It->second);
@@ -15214,6 +15225,19 @@ private:
     assert(IsSupportedHorRdxIdentityOp &&
            "The optimization of matched scalar identity horizontal reductions "
            "must be supported.");
+    auto *VTy = cast<FixedVectorType>(VectorizedValue->getType());
+    if (VTy->getElementType() != VL.front()->getType()) {
+      VectorizedValue = Builder.CreateIntCast(
+          VectorizedValue,
+          FixedVectorType::get(VL.front()->getType(), VTy->getNumElements()),
+          any_of(VL, [&](Value *R) {
+            KnownBits Known = computeKnownBits(
+                R, cast<Instruction>(ReductionOps.front().front())
+                       ->getModule()
+                       ->getDataLayout());
+            return !Known.isNonNegative();
+          }));
+    }
     switch (RdxKind) {
     case RecurKind::Add: {
       // root = mul prev_root, <1, 1, n, 1>
@@ -16217,10 +16241,13 @@ bool SLPVectorizerPass::vectorizeGEPIndices(BasicBlock *BB, BoUpSLP &R) {
       SetVector<Value *> Candidates(GEPList.begin(), GEPList.end());
 
       // Some of the candidates may have already been vectorized after we
-      // initially collected them. If so, they are marked as deleted, so remove
-      // them from the set of candidates.
-      Candidates.remove_if(
-          [&R](Value *I) { return R.isDeleted(cast<Instruction>(I)); });
+      // initially collected them or their index is optimized to constant value.
+      // If so, they are marked as deleted, so remove them from the set of
+      // candidates.
+      Candidates.remove_if([&R](Value *I) {
+        return R.isDeleted(cast<Instruction>(I)) ||
+               isa<Constant>(cast<GetElementPtrInst>(I)->idx_begin()->get());
+      });
 
       // Remove from the set of candidates all pairs of getelementptrs with
       // constant differences. Such getelementptrs are likely not good