[PATCH] D148806: [SLP]Fix the cost for the extractelements, used in several nodes.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev updated this revision to Diff 515461.
ABataev added a comment.

Rebase


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D148806/new/

https://reviews.llvm.org/D148806

Files:
  llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
  llvm/test/Transforms/SLPVectorizer/X86/reused-extractelements.ll


Index: llvm/test/Transforms/SLPVectorizer/X86/reused-extractelements.ll
===================================================================
--- llvm/test/Transforms/SLPVectorizer/X86/reused-extractelements.ll
+++ llvm/test/Transforms/SLPVectorizer/X86/reused-extractelements.ll
@@ -8,7 +8,7 @@
 ; YAML-NEXT: Function:        g
 ; YAML-NEXT: Args:
 ; YAML-NEXT:   - String:          'SLP vectorized with cost '
-; YAML-NEXT:   - Cost:            '-2'
+; YAML-NEXT:   - Cost:            '-1'
 ; YAML-NEXT:   - String:          ' and with tree size '
 ; YAML-NEXT:   - TreeSize:        '4'
 
Index: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -2433,7 +2433,8 @@
 
   /// \returns the cost of the vectorizable entry.
   InstructionCost getEntryCost(const TreeEntry *E,
-                               ArrayRef<Value *> VectorizedVals);
+                               ArrayRef<Value *> VectorizedVals,
+                               SmallPtrSetImpl<Value *> &CheckedExtracts);
 
   /// This is the recursive part of buildTree.
   void buildTree_rec(ArrayRef<Value *> Roots, unsigned Depth,
@@ -6793,6 +6794,7 @@
   InstructionCost Cost = 0;
   ArrayRef<Value *> VectorizedVals;
   BoUpSLP &R;
+  SmallPtrSetImpl<Value *> &CheckedExtracts;
   constexpr static TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
 
   InstructionCost getBuildVectorCost(ArrayRef<Value *> VL, Value *Root) {
@@ -6923,8 +6925,10 @@
 
 public:
   ShuffleCostEstimator(TargetTransformInfo &TTI,
-                       ArrayRef<Value *> VectorizedVals, BoUpSLP &R)
-      : TTI(TTI), VectorizedVals(VectorizedVals), R(R) {}
+                       ArrayRef<Value *> VectorizedVals, BoUpSLP &R,
+                       SmallPtrSetImpl<Value *> &CheckedExtracts)
+      : TTI(TTI), VectorizedVals(VectorizedVals), R(R),
+        CheckedExtracts(CheckedExtracts) {}
   Value *adjustExtracts(const TreeEntry *E, ArrayRef<int> Mask) {
     if (Mask.empty())
       return nullptr;
@@ -6936,7 +6940,6 @@
     if (VecNumParts == VecTy->getNumElements())
       return nullptr;
     DenseMap<Value *, int> ExtractVectorsTys;
-    SmallPtrSet<Value *, 4> CheckedExtracts;
     for (auto [I, V] : enumerate(VL)) {
       // Ignore non-extractelement scalars.
       if (isa<UndefValue>(V) || (!Mask.empty() && Mask[I] == UndefMaskElem))
@@ -7061,8 +7064,9 @@
   }
 };
 
-InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
-                                      ArrayRef<Value *> VectorizedVals) {
+InstructionCost
+BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
+                      SmallPtrSetImpl<Value *> &CheckedExtracts) {
   ArrayRef<Value *> VL = E->Scalars;
 
   Type *ScalarTy = VL[0]->getType();
@@ -7089,7 +7093,8 @@
       return 0;
     if (isa<InsertElementInst>(VL[0]))
       return InstructionCost::getInvalid();
-    ShuffleCostEstimator Estimator(*TTI, VectorizedVals, *this);
+    ShuffleCostEstimator Estimator(*TTI, VectorizedVals, *this,
+                                   CheckedExtracts);
     unsigned VF = E->getVectorFactor();
     SmallVector<int> ReuseShuffleIndicies(E->ReuseShuffleIndices.begin(),
                                           E->ReuseShuffleIndices.end());
@@ -8223,6 +8228,7 @@
 
   unsigned BundleWidth = VectorizableTree[0]->Scalars.size();
 
+  SmallPtrSet<Value *, 4> CheckedExtracts;
   for (unsigned I = 0, E = VectorizableTree.size(); I < E; ++I) {
     TreeEntry &TE = *VectorizableTree[I];
     if (TE.State == TreeEntry::NeedToGather) {
@@ -8238,7 +8244,7 @@
       }
     }
 
-    InstructionCost C = getEntryCost(&TE, VectorizedVals);
+    InstructionCost C = getEntryCost(&TE, VectorizedVals, CheckedExtracts);
     Cost += C;
     LLVM_DEBUG(dbgs() << "SLP: Adding cost " << C
                       << " for bundle that starts with " << *TE.Scalars[0]


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D148806.515461.patch
Type: text/x-patch
Size: 4002 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230420/97f21e22/attachment.bin>