This patch is predecessor of D70223. Since the patch is to optimize vector load-update-store pattern, some of the tests might be affected. The two test cases modified in this patch is to check PPC's swap removal, not related to stores. So it's better to add volatile flag to them, for eliminating unintended test break.
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
My concern is store and store volatile are so different in semantics that it might break original test intention.
Comment Actions
Agree with Kai. You need some evidence to show that, this won't break the test point.
Comment Actions
The two tests are testing functionality of swap removal which runs on machine functions. volatile won't affect generated MIR here.
Comment Actions
I agree with @lkail and @steven.zhang , I don't think we should update these testcases just to avoid updating them in another patch.
I believe keeping them as it is will catch more regression than adding volatile to them .