[clang] [llvm] Remove __arm_atomic_store_with_stshh from llvm (PR #192419)

David Green via cfe-commits cfe-commits at lists.llvm.org
Thu Apr 16 05:57:32 PDT 2026


davemgreen wrote:

> Yeah that's exactly the bug from what I understand. Of course it will be good to fix, as this effects other builtins like __builtin_arm_stlex, but currently this is not priority for us. We are also redisigning the ACLE for this builtin and we plan to implement it as an atomic store with metadata, so most of it will need to be redone anyway. We are removing it for now to reduce the amount of wrong code being generated.

Do you have another example of it going wrong? The one you pointed to looks like it would perform the same optimization/miscompile for a normal call, which could presumably contain atomics too if the reasoning is correct.  https://godbolt.org/z/MddbeWKTe

> I think this builtin would be in LLVM 22.1.2 onwards. I don't know how the policy works for minor releases regarding this ?

Do you have a link to the patch that went into the branch? I didn't see it, but maybe it had a different name. https://github.com/llvm/llvm-project/pull/181386 was the one on trunk I believe, which was after the branch.

https://github.com/llvm/llvm-project/pull/192419


More information about the cfe-commits mailing list