[clang] [llvm] [mlir] [OpenMP] Migrate GPU Reductions CodeGen from Clang to OMPIRBuilder (PR #80343)

[Johannes Doerfert via cfe-commits]

================
@@ -2953,7 +2953,7 @@ int bar(int n){
 // CHECK3-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0:%.*]], ptr [[TMP4]], i32 [[TMP5]]
 // CHECK3-NEXT:    [[C:%.*]] = getelementptr inbounds [[STRUCT__GLOBALIZED_LOCALS_TY_0]], ptr [[TMP8]], i32 0, i32 0
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i8, ptr [[TMP7]], align 1
-// CHECK3-NEXT:    store i8 [[TMP9]], ptr [[C]], align 4
----------------
jdoerfert wrote:

Similar to above, this affects only accesses to a static global. Even if the new alignment is lower than optimal, we should be able to reconstruct it from the allocation. This is fine, IMHO.

https://github.com/llvm/llvm-project/pull/80343