[InstCombine] Fix FMF propagation in copysign Mag, (copysign ?, X) -> copysign Mag, X

[dtcxzyw]

Closes #121432.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Yingwei Zheng (dtcxzyw)

Changes

Closes #121432.

---

Full diff: https://github.com/llvm/llvm-project/pull/121574.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp (+3-1)
• (modified) llvm/test/Transforms/InstCombine/copysign.ll (+11-1)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index fd38738e3be80b..6624a40a781d9f 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -2644,8 +2644,10 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
     // Propagate sign argument through nested calls:
     // copysign Mag, (copysign ?, X) --> copysign Mag, X
     Value *X;
-    if (match(Sign, m_Intrinsic<Intrinsic::copysign>(m_Value(), m_Value(X))))
+    if (match(Sign, m_Intrinsic<Intrinsic::copysign>(m_Value(), m_Value(X)))) {
+      II->andIRFlags(Sign);
       return replaceOperand(*II, 1, X);
+    }
 
     // Clear sign-bit of constant magnitude:
     // copysign -MagC, X --> copysign MagC, X
diff --git a/llvm/test/Transforms/InstCombine/copysign.ll b/llvm/test/Transforms/InstCombine/copysign.ll
index abc707acf0cd3f..2c92557534708c 100644
--- a/llvm/test/Transforms/InstCombine/copysign.ll
+++ b/llvm/test/Transforms/InstCombine/copysign.ll
@@ -82,7 +82,7 @@ define float @not_known_positive_sign_arg(float %x, float %y) {
 
 define float @copysign_sign_arg(float %x, float %y, float %z) {
 ; CHECK-LABEL: @copysign_sign_arg(
-; CHECK-NEXT:    [[R:%.*]] = call ninf float @llvm.copysign.f32(float [[X:%.*]], float [[Z:%.*]])
+; CHECK-NEXT:    [[R:%.*]] = call float @llvm.copysign.f32(float [[X:%.*]], float [[Z:%.*]])
 ; CHECK-NEXT:    ret float [[R]]
 ;
   %s = call reassoc float @llvm.copysign.f32(float %y, float %z)
@@ -90,6 +90,16 @@ define float @copysign_sign_arg(float %x, float %y, float %z) {
   ret float %r
 }
 
+define float @copysign_sign_arg_nnan(float %x, float %y, float %z) {
+; CHECK-LABEL: @copysign_sign_arg_nnan(
+; CHECK-NEXT:    [[R:%.*]] = call nnan float @llvm.copysign.f32(float [[X:%.*]], float [[Z:%.*]])
+; CHECK-NEXT:    ret float [[R]]
+;
+  %s = call nnan float @llvm.copysign.f32(float %y, float %z)
+  %r = call nnan float @llvm.copysign.f32(float %x, float %s)
+  ret float %r
+}
+
 define float @fneg_mag(float %x, float %y) {
 ; CHECK-LABEL: @fneg_mag(
 ; CHECK-NEXT:    [[R:%.*]] = call float @llvm.copysign.f32(float [[X:%.*]], float [[Y:%.*]])

[nikic]

Can we create a new call here instead of modifying the operand? Otherwise there would still be a potential problem with nofpclass on the argument.

[arsenm]

We really ought to add FMF arguments to all these CreateFunctions, the guard + set is so clumsy

[dtcxzyw]

I am working on a prototype.

[nikic]

LGTM