[llvm] [IR] Memory locations for floating-point operations (PR #226826)

via llvm-commits llvm-commits at lists.llvm.org
Thu Oct 1 04:08:18 PDT 2026


================
@@ -285,7 +285,7 @@ static void addMemoryAttrs(const SCCNodeSet &SCCNodes, AARGetterT &&AARGetter,
     ME |= FnME;
     RecursiveArgME |= FnRecursiveArgME;
     // Reached bottom of the lattice, we will not be able to improve the result.
-    if (ME == MemoryEffects::unknown())
+    if (ME == MemoryEffects::unknown_mem())
----------------
CarolineConcatto wrote:

For all  the Transform changes if you keep this checking MemoryEffects::unknown() and the memory set(ME = MemoryEffects(ME.getModRef());) you still set all locations, what would be a problem? 
 I dont understand what is the problem of setting all bits for now and checking all bits for the memory location.

https://github.com/llvm/llvm-project/pull/226826


More information about the llvm-commits mailing list