[AMDGPU] Fix missing IsExact flag when expanding vector binary operator

[shiltian]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Shilei Tian (shiltian)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/86712.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp (+3)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
index bddf3d958a1ae6..b8ac81f1316721 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
@@ -1594,6 +1594,9 @@ bool AMDGPUCodeGenPrepareImpl::visitBinaryOperator(BinaryOperator &I) {
           }
         }
 
+        if (auto *NewEltI = dyn_cast<Instruction>(NewElt))
+          NewEltI->setIsExact(I.isExact());
+
         NewDiv = Builder.CreateInsertElement(NewDiv, NewElt, N);
       }
     } else {

[shiltian]

I'll update the test case in #83155 once that patch is landed to test this patch.

[arsenm]

Missing test

[arsenm]

I'll update the test case in #83155 once that patch is landed to test this patch.

The code is already tested, just add a case with exact

[shiltian]

I'll update the test case in #83155 once that patch is landed to test this patch.

The code is already tested, just add a case with exact

Done.

[jayfoad]

LGTM