[InstCombine] Remove OneUse clause from ((A+B)+B) → A+2*B

[rj-jesus]

This patch removes the m_OneUse check that was being done for the following patterns:
(A + RHS) + RHS --> A + (RHS << 1)
LHS + (A + LHS) --> A + (LHS << 1)

This may increase the instruction count if (A + RHS) or (A + LHS) are used more than once but will usually help with LICM later on if A is an induction variable.

This is exemplified here: https://godbolt.org/z/q94vbf5e9 - the first instruction of the second iteration ([1,0]) can start much earlier on with the patch, allowing for significantly higher IPC.

I've benchmarked this with RAJAPerf on a Grace system and see an overall performance uplift of 2% across the whole benchmark which comes mainly from ~45% uplifts in three kernels. Worst I saw was a 1.2% slowdown on a single kernel which I look into later.

Please let me know if I should move this elsewhere!

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Ricardo Jesus (rj-jesus)

Changes

This patch removes the m_OneUse check that was being done for the following patterns:
(A + RHS) + RHS --> A + (RHS << 1)
LHS + (A + LHS) --> A + (LHS << 1)

This may increase the instruction count if (A + RHS) or (A + LHS) are used more than once but will usually help with LICM later on if A is an induction variable.

This is exemplified here: https://godbolt.org/z/q94vbf5e9 - the first instruction of the second iteration ([1,0]) can start much earlier on with the patch, allowing for significantly higher IPC.

I've benchmarked this with RAJAPerf on a Grace system and see an overall performance uplift of 2% across the whole benchmark which comes mainly from ~45% uplifts in three kernels. Worst I saw was a 1.2% slowdown on a single kernel which I look into later.

Please let me know if I should move this elsewhere!

---

Full diff: https://github.com/llvm/llvm-project/pull/80705.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp (+5-2)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 8a00b75a1f7404..984f7a619d4ab8 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -1521,11 +1521,14 @@ Instruction *InstCombinerImpl::visitAdd(BinaryOperator &I) {
     return BinaryOperator::CreateSub(A, B);
 
   // (A + RHS) + RHS --> A + (RHS << 1)
-  if (match(LHS, m_OneUse(m_c_Add(m_Value(A), m_Specific(RHS)))))
+  // This may increase instruction count if (A + RHS) is used more than once
+  // but helps make (RHS << 1) suitable for potential LICM.
+  if (match(LHS, m_c_Add(m_Value(A), m_Specific(RHS))))
     return BinaryOperator::CreateAdd(A, Builder.CreateShl(RHS, 1, "reass.add"));
 
   // LHS + (A + LHS) --> A + (LHS << 1)
-  if (match(RHS, m_OneUse(m_c_Add(m_Value(A), m_Specific(LHS)))))
+  // Same as above.
+  if (match(RHS, m_c_Add(m_Value(A), m_Specific(LHS))))
     return BinaryOperator::CreateAdd(A, Builder.CreateShl(LHS, 1, "reass.add"));
 
   {