Skip to content

[InstCombine] Fold gep of exact unsigned division #82334

New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Merged
merged 1 commit into from
Feb 20, 2024
Merged

Conversation

nikic
Copy link
Contributor

@nikic nikic commented Feb 20, 2024

Extend the transform added in #76458 to also handle unsigned division. X exact/ Y * Y == X holds independently of whether the division is signed or unsigned.

Proofs: https://alive2.llvm.org/ce/z/wFd5Ec

Extend the transform added in llvm#76458
to also handle unsigned division. X exact/ Y * Y == X holds
independently of whether the division is signed or unsigned.

Proofs: https://alive2.llvm.org/ce/z/wFd5Ec
@llvmbot
Copy link
Member

llvmbot commented Feb 20, 2024

@llvm/pr-subscribers-llvm-transforms

Author: Nikita Popov (nikic)

Changes

Extend the transform added in #76458 to also handle unsigned division. X exact/ Y * Y == X holds independently of whether the division is signed or unsigned.

Proofs: https://alive2.llvm.org/ce/z/wFd5Ec


Full diff: https://github.com/llvm/llvm-project/pull/82334.diff

2 Files Affected:

  • (modified) llvm/lib/Transforms/InstCombine/InstructionCombining.cpp (+3-3)
  • (modified) llvm/test/Transforms/InstCombine/getelementptr.ll (+2-4)
diff --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
index 7450f39c1e7641..4dbb79f7bef887 100644
--- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -2615,10 +2615,10 @@ Instruction *InstCombinerImpl::visitGetElementPtrInst(GetElementPtrInst &GEP) {
         Value *V;
         if ((has_single_bit(TyAllocSize) &&
              match(GEP.getOperand(1),
-                   m_Exact(m_AShr(m_Value(V),
-                                  m_SpecificInt(countr_zero(TyAllocSize)))))) ||
+                   m_Exact(m_Shr(m_Value(V),
+                                 m_SpecificInt(countr_zero(TyAllocSize)))))) ||
             match(GEP.getOperand(1),
-                  m_Exact(m_SDiv(m_Value(V), m_SpecificInt(TyAllocSize))))) {
+                  m_Exact(m_IDiv(m_Value(V), m_SpecificInt(TyAllocSize))))) {
           GetElementPtrInst *NewGEP = GetElementPtrInst::Create(
               Builder.getInt8Ty(), GEP.getPointerOperand(), V);
           NewGEP->setIsInBounds(GEP.isInBounds());
diff --git a/llvm/test/Transforms/InstCombine/getelementptr.ll b/llvm/test/Transforms/InstCombine/getelementptr.ll
index a9c619f3fddd05..04b0c196ab5108 100644
--- a/llvm/test/Transforms/InstCombine/getelementptr.ll
+++ b/llvm/test/Transforms/InstCombine/getelementptr.ll
@@ -1477,8 +1477,7 @@ define ptr @gep_sdiv(ptr %p, i64 %off) {
 
 define ptr @gep_udiv(ptr %p, i64 %off) {
 ; CHECK-LABEL: @gep_udiv(
-; CHECK-NEXT:    [[INDEX:%.*]] = udiv exact i64 [[OFF:%.*]], 7
-; CHECK-NEXT:    [[PTR:%.*]] = getelementptr [[STRUCT_C:%.*]], ptr [[P:%.*]], i64 [[INDEX]]
+; CHECK-NEXT:    [[PTR:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 [[OFF:%.*]]
 ; CHECK-NEXT:    ret ptr [[PTR]]
 ;
   %index = udiv exact i64 %off, 7
@@ -1518,8 +1517,7 @@ define ptr @gep_ashr(ptr %p, i64 %off) {
 
 define ptr @gep_lshr(ptr %p, i64 %off) {
 ; CHECK-LABEL: @gep_lshr(
-; CHECK-NEXT:    [[INDEX:%.*]] = lshr exact i64 [[OFF:%.*]], 2
-; CHECK-NEXT:    [[PTR:%.*]] = getelementptr i32, ptr [[P:%.*]], i64 [[INDEX]]
+; CHECK-NEXT:    [[PTR:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 [[OFF:%.*]]
 ; CHECK-NEXT:    ret ptr [[PTR]]
 ;
   %index = lshr exact i64 %off, 2

dtcxzyw added a commit to dtcxzyw/llvm-opt-benchmark that referenced this pull request Feb 20, 2024
dtcxzyw added a commit to dtcxzyw/llvm-opt-benchmark that referenced this pull request Feb 20, 2024
Copy link
Member

@dtcxzyw dtcxzyw left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Thanks!

@nikic nikic merged commit 26d4afc into llvm:main Feb 20, 2024
@nikic nikic deleted the gep-div-exact branch February 20, 2024 11:48
@nikic nikic added this to the LLVM 18.X Release milestone Feb 20, 2024
@nikic
Copy link
Contributor Author

nikic commented Feb 20, 2024

/cherry-pick ec2c770 26d4afc

llvmbot pushed a commit to llvmbot/llvm-project that referenced this pull request Feb 20, 2024
Extend the transform added in
llvm#76458 to also handle unsigned
division. X exact/ Y * Y == X holds independently of whether the
division is signed or unsigned.

Proofs: https://alive2.llvm.org/ce/z/wFd5Ec
(cherry picked from commit 26d4afc)
@llvmbot
Copy link
Member

llvmbot commented Feb 20, 2024

/pull-request #82347

# for free to join this conversation on GitHub. Already have an account? # to comment
Projects
Development

Successfully merging this pull request may close these issues.

3 participants