[llvm] c5f40bf - [InstCombine] Fold `X!=Y ? ctz(X^Y, true) : BW -> ctz(X^Y, false)` (#128483)

[via llvm-commits]

Author: Yingwei Zheng
Date: 2025-02-24T17:35:46+08:00
New Revision: c5f40bf024ee2d62478c8036fb174d75ecabe51f

URL: https://github.com/llvm/llvm-project/commit/c5f40bf024ee2d62478c8036fb174d75ecabe51f
DIFF: https://github.com/llvm/llvm-project/commit/c5f40bf024ee2d62478c8036fb174d75ecabe51f.diff

LOG: [InstCombine] Fold `X!=Y ? ctz(X^Y, true) : BW -> ctz(X^Y, false)` (#128483)

Proof: https://alive2.llvm.org/ce/z/mzL6W2
Closes https://github.com/llvm/llvm-project/issues/128441.

Added: 
    

Modified: 
    llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
    llvm/test/Transforms/InstCombine/select-cmp-cttz-ctlz.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index e621a0b7fe596..dca969e160bb7 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1227,8 +1227,10 @@ static Value *foldSelectCttzCtlz(ICmpInst *ICI, Value *TrueVal, Value *FalseVal,
 
   // (X == 0) ? BitWidth : ctz(X)
   // (X == -1) ? BitWidth : ctz(~X)
+  // (X == Y) ? BitWidth : ctz(X ^ Y)
   if ((X != CmpLHS || !match(CmpRHS, m_Zero())) &&
-      (!match(X, m_Not(m_Specific(CmpLHS))) || !match(CmpRHS, m_AllOnes())))
+      (!match(X, m_Not(m_Specific(CmpLHS))) || !match(CmpRHS, m_AllOnes())) &&
+      !match(X, m_c_Xor(m_Specific(CmpLHS), m_Specific(CmpRHS))))
     return nullptr;
 
   IntrinsicInst *II = cast<IntrinsicInst>(Count);

diff  --git a/llvm/test/Transforms/InstCombine/select-cmp-cttz-ctlz.ll b/llvm/test/Transforms/InstCombine/select-cmp-cttz-ctlz.ll
index 2cb70e85f435f..52a32e19f57ef 100644
--- a/llvm/test/Transforms/InstCombine/select-cmp-cttz-ctlz.ll
+++ b/llvm/test/Transforms/InstCombine/select-cmp-cttz-ctlz.ll
@@ -657,6 +657,67 @@ define i16 @test_multiuse_trunc_undef(i64 %x, ptr %p) {
   ret i16 %cond
 }
 
+define i64 @test_pr128441(i64 %x, i64 %y) {
+; CHECK-LABEL: @test_pr128441(
+; CHECK-NEXT:    [[XOR:%.*]] = xor i64 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT:    [[SEL:%.*]] = call range(i64 0, 65) i64 @llvm.cttz.i64(i64 [[XOR]], i1 false)
+; CHECK-NEXT:    ret i64 [[SEL]]
+;
+  %iszero = icmp ne i64 %x, %y
+  %xor = xor i64 %x, %y
+  %cttz = call i64 @llvm.cttz.i64(i64 %xor, i1 true)
+  %sel = select i1 %iszero, i64 %cttz, i64 64
+  ret i64 %sel
+}
+
+define i64 @test_pr128441_commuted1(i64 %x, i64 %y) {
+; CHECK-LABEL: @test_pr128441_commuted1(
+; CHECK-NEXT:    [[XOR:%.*]] = xor i64 [[Y:%.*]], [[X:%.*]]
+; CHECK-NEXT:    [[SEL:%.*]] = call range(i64 0, 65) i64 @llvm.cttz.i64(i64 [[XOR]], i1 false)
+; CHECK-NEXT:    ret i64 [[SEL]]
+;
+  %iszero = icmp ne i64 %x, %y
+  %xor = xor i64 %y, %x
+  %cttz = call i64 @llvm.cttz.i64(i64 %xor, i1 true)
+  %sel = select i1 %iszero, i64 %cttz, i64 64
+  ret i64 %sel
+}
+
+define i64 @test_pr128441_commuted2(i64 %x, i64 %y) {
+; CHECK-LABEL: @test_pr128441_commuted2(
+; CHECK-NEXT:    [[XOR:%.*]] = xor i64 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT:    [[SEL:%.*]] = call range(i64 0, 65) i64 @llvm.cttz.i64(i64 [[XOR]], i1 false)
+; CHECK-NEXT:    ret i64 [[SEL]]
+;
+  %iszero = icmp eq i64 %x, %y
+  %xor = xor i64 %x, %y
+  %cttz = call i64 @llvm.cttz.i64(i64 %xor, i1 true)
+  %sel = select i1 %iszero, i64 64, i64 %cttz
+  ret i64 %sel
+}
+
+define i64 @test_pr128441_commuted3_negative(i64 %x, i64 %y) {
+; CHECK-LABEL: @test_pr128441_commuted3_negative(
+; CHECK-NEXT:    ret i64 64
+;
+  %iszero = icmp eq i64 %x, %y
+  %xor = xor i64 %y, %x
+  %cttz = call i64 @llvm.cttz.i64(i64 %xor, i1 true)
+  %sel = select i1 %iszero, i64 %cttz, i64 64
+  ret i64 %sel
+}
+
+define i64 @test_pr128441_commuted4_negative(i64 %x, i64 %y) {
+; CHECK-LABEL: @test_pr128441_commuted4_negative(
+; CHECK-NEXT:    ret i64 64
+;
+  %iszero = icmp ne i64 %x, %y
+  %xor = xor i64 %x, %y
+  %cttz = call i64 @llvm.cttz.i64(i64 %xor, i1 true)
+  %sel = select i1 %iszero, i64 64, i64 %cttz
+  ret i64 %sel
+}
+
 declare i16 @llvm.ctlz.i16(i16, i1)
 declare i32 @llvm.ctlz.i32(i32, i1)
 declare i64 @llvm.ctlz.i64(i64, i1)