[PATCH] D149966: [SLP]Include cost of the reshuffling for same nodes with resizing.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev updated this revision to Diff 538619.
ABataev added a comment.

Address comment


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149966/new/

https://reviews.llvm.org/D149966

Files:
  llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
  llvm/test/Transforms/SLPVectorizer/X86/crash_clear_undefs.ll


Index: llvm/test/Transforms/SLPVectorizer/X86/crash_clear_undefs.ll
===================================================================
--- llvm/test/Transforms/SLPVectorizer/X86/crash_clear_undefs.ll
+++ llvm/test/Transforms/SLPVectorizer/X86/crash_clear_undefs.ll
@@ -9,7 +9,7 @@
 ; YAML-NEXT:  Function:        foo
 ; YAML-NEXT:  Args:
 ; YAML-NEXT:    - String:          'SLP vectorized with cost '
-; YAML-NEXT:    - Cost:            '-6'
+; YAML-NEXT:    - Cost:            '-5'
 ; YAML-NEXT:    - String:          ' and with tree size '
 ; YAML-NEXT:    - TreeSize:        '10'
 ; YAML-NEXT:  ...
Index: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -7001,6 +7001,17 @@
         const TreeEntry *E2 = P2.get<const TreeEntry *>();
         if (E->Scalars.size() == E2->Scalars.size())
           CommonVF = VF = E->Scalars.size();
+      } else {
+        // P2 is empty, check that we have same node + reshuffle (if any).
+        if (E->Scalars.size() == Mask.size() && VF != Mask.size()) {
+          VF = E->Scalars.size();
+          SmallVector<int> CommonMask(Mask.begin(), Mask.end());
+          ::addMask(CommonMask, E->ReuseShuffleIndices);
+          V1 = Constant::getNullValue(
+              FixedVectorType::get(E->Scalars.front()->getType(), VF));
+          return BaseShuffleAnalysis::createShuffle<InstructionCost>(
+              V1, nullptr, CommonMask, Builder);
+        }
       }
       V1 = Constant::getNullValue(
           FixedVectorType::get(E->Scalars.front()->getType(), VF));
@@ -7281,7 +7292,17 @@
             dbgs()
             << "SLP: perfect diamond match for gather bundle that starts with "
             << *VL.front() << ".\n");
-        return 0;
+        // Restore the mask for previous partially matched values.
+        for (auto [I, V] : enumerate(E->Scalars)) {
+          if (isa<PoisonValue>(V)) {
+            Mask[I] = PoisonMaskElem;
+            continue;
+          }
+          if (Mask[I] == PoisonMaskElem)
+            Mask[I] = Entries.front()->findLaneForValue(V);
+        }
+        Estimator.add(Entries.front(), Mask);
+        return Estimator.finalize(E->ReuseShuffleIndices);
       }
       if (!Resized) {
         unsigned VF1 = Entries.front()->getVectorFactor();


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D149966.538619.patch
Type: text/x-patch
Size: 2423 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230710/a9bb51a3/attachment.bin>