[llvm] [LV] Vectorize predicated loop-invariant loads via masked load + broadcast (PR #200361)

[via llvm-commits]

github-actions[bot] wrote:

<!--LLVM CODE FORMAT COMMENT: {clang-format}-->


:warning: C/C++ code formatter, clang-format found issues in your code. :warning:

<details>
<summary>
You can test this locally with the following command:
</summary>

``````````bash
git-clang-format --diff origin/main HEAD --extensions h,cpp -- llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.cpp llvm/lib/Transforms/Vectorize/LoopVectorize.cpp llvm/lib/Transforms/Vectorize/VPlan.h llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp llvm/lib/Transforms/Vectorize/VPlanTransforms.h --diff_from_common_commit
``````````

:warning:
The reproduction instructions above might return results for more than one PR
in a stack if you are using a stacked PR workflow. You can limit the results by
changing `origin/main` to the base branch/commit you want to compare against.
:warning:

</details>

<details>
<summary>
View the diff from clang-format here.
</summary>

``````````diff
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 9c4fe7b2d..0ffe933bc 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -3815,9 +3815,9 @@ public:
   InstructionCost computeCost(ElementCount VF, VPCostContext &Ctx) const;
 
   /// Return the cost of a masked widened memory operation with opcode \p Opcode
-  /// (Load or Store), vectorized data type \p DataTy, alignment \p Alignment and
-  /// address space \p AS. This is a static helper so callers can price a masked
-  /// memory access without first constructing a recipe.
+  /// (Load or Store), vectorized data type \p DataTy, alignment \p Alignment
+  /// and address space \p AS. This is a static helper so callers can price a
+  /// masked memory access without first constructing a recipe.
   static InstructionCost computeMaskedCost(unsigned Opcode, Type *DataTy,
                                            Align Alignment, unsigned AS,
                                            VPCostContext &Ctx);
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 991dc331a..582240667 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -1331,7 +1331,7 @@ InstructionCost VPInstruction::computeExtractElementCost(VectorType *VecTy,
 
 InstructionCost
 VPInstruction::computeShuffleCost(TargetTransformInfo::ShuffleKind Kind,
-                                 VectorType *VecTy, VPCostContext &Ctx) {
+                                  VectorType *VecTy, VPCostContext &Ctx) {
   return Ctx.TTI.getShuffleCost(Kind, VecTy, VecTy, /*Mask=*/{}, Ctx.CostKind,
                                 /*Index=*/0);
 }
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 86811e3e7..cd55c06f3 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -3935,8 +3935,8 @@ void VPlanTransforms::widenPredicatedInvariantLoads(VPlan &Plan,
         Builder.createNaryOp(VPInstruction::Broadcast, {FalseI1}, DL);
     VPValue *Lane0Idx =
         Plan.getOrAddLiveIn(ConstantInt::get(Type::getInt32Ty(Ctx), 0));
-    VPValue *VMask = Builder.createNaryOp(
-        Instruction::InsertElement, {ZeroVMask, AnyActive, Lane0Idx}, DL);
+    VPValue *VMask = Builder.createNaryOp(Instruction::InsertElement,
+                                          {ZeroVMask, AnyActive, Lane0Idx}, DL);
 
     // Consecutive=true selects the masked-load (not masked-gather) lowering in
     // VPWidenLoadRecipe::execute. The access is stride-0, not stride-1; lane 0
@@ -3947,10 +3947,11 @@ void VPlanTransforms::widenPredicatedInvariantLoads(VPlan &Plan,
 
     // Broadcast lane 0 of the masked load across all lanes, reusing the
     // existing extractelement + broadcast recipes.
-    VPValue *Lane0 = Builder.createNaryOp(
-        Instruction::ExtractElement, {MaskedLoad->getVPSingleValue(), Lane0Idx},
-        DL);
-    VPValue *Bcast = Builder.createNaryOp(VPInstruction::Broadcast, {Lane0}, DL);
+    VPValue *Lane0 =
+        Builder.createNaryOp(Instruction::ExtractElement,
+                             {MaskedLoad->getVPSingleValue(), Lane0Idx}, DL);
+    VPValue *Bcast =
+        Builder.createNaryOp(VPInstruction::Broadcast, {Lane0}, DL);
 
     RepR->replaceAllUsesWith(Bcast);
     RepR->eraseFromParent();

``````````

</details>


https://github.com/llvm/llvm-project/pull/200361