https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71660
--- Comment #11 from Peter Cordes <peter at cordes dot ca> --- (In reply to Thiago Macieira from comment #10) > Actually, PR 65146 points out that the problem is not efficiency but > correctness. An under-aligned type could cross a cacheline boundary and thus > fail to be atomic in the first place. As I pointed out there, you technically could solve the correctness problem by checking alignment and falling back to locking for objects where a plain 8B load or 8B store wouldn't be atomic. That's what I meant by "efficiently lock-free". And we're talking about *huge* inefficiencies here compared to always being able to inline and SSE load/store. That would let you keep struct layouts the same, but it would still be an ABI change, since everything has to agree about which objects are lock-free and which aren't. Now that I think about it, all of my suggested fixes on PR 65146 are effectively ABI changes. > Those structures were disasters waiting to happen. Yes, exactly. Basically any existing binaries compiled with a gcc that allows under-aligned atomic objects are unsafe, so keeping compatibility with them is not important.
