[llvm] [ConstantFolding] Fix nvvm_round folding on PPC (PR #149837)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-analysis

Author: Lewis Crawford (LewisCrawford)

<details>
<summary>Changes</summary>

Fix a failing test for constant-folding the nvvm_round intrinsic. The original implementation added in #<!-- -->141233 used a native libm call to the "round" function, but on PPC this produces +0.0 if the input is -0.0, which caused a test failure.

This patch updates it to use APFloat functions instead of native libm calls to ensure cross-platform consistency.

---
Full diff: https://github.com/llvm/llvm-project/pull/149837.diff


1 Files Affected:

- (modified) llvm/lib/Analysis/ConstantFolding.cpp (+9-5) 


``````````diff
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index f5a88b6e0368e..eb7369fcc7513 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -2677,11 +2677,15 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
 
       case Intrinsic::nvvm_round_ftz_f:
       case Intrinsic::nvvm_round_f:
-      case Intrinsic::nvvm_round_d:
-        return ConstantFoldFP(
-            round, APF, Ty,
-            nvvm::GetNVVMDenromMode(
-                nvvm::UnaryMathIntrinsicShouldFTZ(IntrinsicID)));
+      case Intrinsic::nvvm_round_d: {
+        // Use APFloat implementation instead of native libm call, as some
+        // implementations (e.g. on PPC) do not preserve the sign of negative 0.
+        APFloat Res = nvvm::UnaryMathIntrinsicShouldFTZ(IntrinsicID)
+                          ? FTZPreserveSign(APF)
+                          : APF;
+        Res.roundToIntegral(APFloat::rmNearestTiesToAway);
+        return ConstantFP::get(Ty->getContext(), U);
+      }
 
       case Intrinsic::nvvm_saturate_ftz_f:
       case Intrinsic::nvvm_saturate_d:

``````````

</details>


https://github.com/llvm/llvm-project/pull/149837