[llvm] [VPlan] Introduce vputils::getSingleScalarClone (PR #161667)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-vectorizers

Author: Ramkumar Ramachandra (artagnon)

<details>
<summary>Changes</summary>



---
Full diff: https://github.com/llvm/llvm-project/pull/161667.diff


3 Files Affected:

- (modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+6-22) 
- (modified) llvm/lib/Transforms/Vectorize/VPlanUtils.cpp (+19) 
- (modified) llvm/lib/Transforms/Vectorize/VPlanUtils.h (+4) 


``````````diff
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index acdb37996a443..2b3f4af8fa3ad 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -189,19 +189,7 @@ static bool sinkScalarOperands(VPlan &Plan) {
     if (NeedsDuplicating) {
       if (ScalarVFOnly)
         continue;
-      VPSingleDefRecipe *Clone;
-      if (auto *SinkCandidateRepR =
-              dyn_cast<VPReplicateRecipe>(SinkCandidate)) {
-        // TODO: Handle converting to uniform recipes as separate transform,
-        // then cloning should be sufficient here.
-        Instruction *I = SinkCandidate->getUnderlyingInstr();
-        Clone = new VPReplicateRecipe(I, SinkCandidate->operands(), true,
-                                      nullptr /*Mask*/, *SinkCandidateRepR);
-        // TODO: add ".cloned" suffix to name of Clone's VPValue.
-      } else {
-        Clone = SinkCandidate->clone();
-      }
-
+      VPSingleDefRecipe *Clone = vputils::getSingleScalarClone(SinkCandidate);
       Clone->insertBefore(SinkCandidate);
       SinkCandidate->replaceUsesWithIf(Clone, [SinkTo](VPUser &U, unsigned) {
         return cast<VPRecipeBase>(&U)->getParent() != SinkTo;
@@ -666,8 +654,7 @@ static void legalizeAndOptimizeInductions(VPlan &Plan) {
       if (!vputils::isSingleScalar(Def) && !vputils::onlyFirstLaneUsed(Def))
         continue;
 
-      auto *Clone = new VPReplicateRecipe(Def->getUnderlyingInstr(),
-                                          Def->operands(), /*IsUniform*/ true);
+      VPSingleDefRecipe *Clone = vputils::getSingleScalarClone(Def);
       Clone->insertAfter(Def);
       Def->replaceAllUsesWith(Clone);
     }
@@ -1309,15 +1296,14 @@ static void narrowToSingleScalarRecipes(VPlan &Plan) {
       auto *RepOrWidenR = cast<VPSingleDefRecipe>(&R);
       if (RepR && isa<StoreInst>(RepR->getUnderlyingInstr()) &&
           vputils::isSingleScalar(RepR->getOperand(1))) {
-        auto *Clone = new VPReplicateRecipe(
-            RepOrWidenR->getUnderlyingInstr(), RepOrWidenR->operands(),
-            true /*IsSingleScalar*/, nullptr /*Mask*/, *RepR /*Metadata*/);
+        auto *Clone =
+            cast<VPReplicateRecipe>(vputils::getSingleScalarClone(RepOrWidenR));
         Clone->insertBefore(RepOrWidenR);
         auto *Ext = new VPInstruction(VPInstruction::ExtractLastElement,
                                       {Clone->getOperand(0)});
         Ext->insertBefore(Clone);
         Clone->setOperand(0, Ext);
-        RepR->eraseFromParent();
+        RepOrWidenR->eraseFromParent();
         continue;
       }
 
@@ -1332,9 +1318,7 @@ static void narrowToSingleScalarRecipes(VPlan &Plan) {
           }))
         continue;
 
-      auto *Clone = new VPReplicateRecipe(RepOrWidenR->getUnderlyingInstr(),
-                                          RepOrWidenR->operands(),
-                                          true /*IsSingleScalar*/);
+      VPSingleDefRecipe *Clone = vputils::getSingleScalarClone(RepOrWidenR);
       Clone->insertBefore(RepOrWidenR);
       RepOrWidenR->replaceAllUsesWith(Clone);
     }
diff --git a/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp b/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
index 059993043dcda..8a59bfc7abfe0 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
@@ -250,3 +250,22 @@ vputils::getRecipesForUncountableExit(VPlan &Plan,
 
   return UncountableCondition;
 }
+
+VPSingleDefRecipe *vputils::getSingleScalarClone(VPSingleDefRecipe *R) {
+  return TypeSwitch<VPSingleDefRecipe *, VPSingleDefRecipe *>(R)
+      .Case<VPInstruction, VPWidenRecipe, VPWidenCastRecipe,
+            VPWidenSelectRecipe, VPWidenCallRecipe, VPReplicateRecipe>(
+          [](auto *I) {
+            return new VPReplicateRecipe(I->getUnderlyingInstr(), I->operands(),
+                                         /*IsSingleScalar*/ true,
+                                         /*Mask*/ nullptr,
+                                         /*Metadata*/ *I);
+          })
+      .Case<VPWidenGEPRecipe>([](auto *I) {
+        // WidenGEP does not have metadata.
+        return new VPReplicateRecipe(I->getUnderlyingInstr(), I->operands(),
+                                     /*IsSingleScalar*/ true, /*Mask*/ nullptr);
+      })
+      .Case<VPScalarIVStepsRecipe>([](auto *I) { return I->clone(); })
+      .Default([](auto *I) { return nullptr; });
+}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanUtils.h b/llvm/lib/Transforms/Vectorize/VPlanUtils.h
index 0222b0aa81063..4979ade0c2764 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanUtils.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanUtils.h
@@ -116,6 +116,10 @@ std::optional<VPValue *>
 getRecipesForUncountableExit(VPlan &Plan,
                              SmallVectorImpl<VPRecipeBase *> &Recipes,
                              SmallVectorImpl<VPRecipeBase *> &GEPs);
+
+/// Returns a single-scalar version of \p R if possible, creating a fresh
+/// single-scalar VPReplicateRecipe or just cloning the recipe.
+VPSingleDefRecipe *getSingleScalarClone(VPSingleDefRecipe *R);
 } // namespace vputils
 
 //===----------------------------------------------------------------------===//

``````````

</details>


https://github.com/llvm/llvm-project/pull/161667