VPlan/PatternMatch: introduce m_c_Mul (NFC)

[artagnon]

Introduce a commutative version of m_Mul, and simplify a usage based on it.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Ramkumar Ramachandra (artagnon)

Changes

Introduce a commutative version of m_Mul, and simplify a usage based on it.

---

Full diff: https://github.com/llvm/llvm-project/pull/93950.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h (+12-7)
• (modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+1-2)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
index 0587468807435..499aea4c6b75a 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
@@ -268,10 +268,16 @@ m_Binary(const Op0_t &Op0, const Op1_t &Op1) {
   return AllBinaryRecipe_match<Op0_t, Op1_t, Opcode, Commutative>(Op0, Op1);
 }
 
-template <typename Op0_t, typename Op1_t>
-inline AllBinaryRecipe_match<Op0_t, Op1_t, Instruction::Mul>
+template <typename Op0_t, typename Op1_t, bool Commutative = false>
+inline AllBinaryRecipe_match<Op0_t, Op1_t, Instruction::Mul, Commutative>
 m_Mul(const Op0_t &Op0, const Op1_t &Op1) {
-  return m_Binary<Instruction::Mul, Op0_t, Op1_t>(Op0, Op1);
+  return m_Binary<Instruction::Mul, Op0_t, Op1_t, Commutative>(Op0, Op1);
+}
+
+template <typename Op0_t, typename Op1_t, bool Commutative = true>
+inline AllBinaryRecipe_match<Op0_t, Op1_t, Instruction::Mul, Commutative>
+m_c_Mul(const Op0_t &Op0, const Op1_t &Op1) {
+  return m_Binary<Instruction::Mul, Op0_t, Op1_t, Commutative>(Op0, Op1);
 }
 
 /// Match a binary OR operation. Note that while conceptually the operands can
@@ -284,11 +290,10 @@ m_BinaryOr(const Op0_t &Op0, const Op1_t &Op1) {
   return m_Binary<Instruction::Or, Op0_t, Op1_t, Commutative>(Op0, Op1);
 }
 
-template <typename Op0_t, typename Op1_t>
-inline AllBinaryRecipe_match<Op0_t, Op1_t, Instruction::Or,
-                             /*Commutative*/ true>
+template <typename Op0_t, typename Op1_t, bool Commutative = true>
+inline AllBinaryRecipe_match<Op0_t, Op1_t, Instruction::Or, Commutative>
 m_c_BinaryOr(const Op0_t &Op0, const Op1_t &Op1) {
-  return m_BinaryOr<Op0_t, Op1_t, /*Commutative*/ true>(Op0, Op1);
+  return m_BinaryOr<Op0_t, Op1_t, Commutative>(Op0, Op1);
 }
 
 template <typename Op0_t, typename Op1_t>
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 422579ea8b84f..c23187065101c 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -948,8 +948,7 @@ static void simplifyRecipe(VPRecipeBase &R, VPTypeAnalysis &TypeInfo) {
     return;
   }
 
-  if (match(&R, m_CombineOr(m_Mul(m_VPValue(A), m_SpecificInt(1)),
-                            m_Mul(m_SpecificInt(1), m_VPValue(A)))))
+  if (match(&R, m_c_Mul(m_VPValue(A), m_SpecificInt(1))))
     return R.getVPSingleValue()->replaceAllUsesWith(A);
 }
 

[artagnon]

Addressed review.

[fhahn]

LGTM, thanks!

[ayalz]

Mentioned in #92539 (comment)

Simplifies existing functionality, worth an NFC suffix in the title.

[artagnon]

This codepath is quite well-tested:

  LLVM :: Transforms/LoopVectorize/ARM/mve-gather-scatter-tailpred.ll
  LLVM :: Transforms/LoopVectorize/RISCV/strided-accesses.ll
  LLVM :: Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
  LLVM :: Transforms/LoopVectorize/runtime-checks-hoist.ll
  LLVM :: Transforms/LoopVectorize/select-reduction.ll
  LLVM :: Transforms/LoopVectorize/version-stride-with-integer-casts.ll