nikic wrote: > > As far as I understand it, the hint here is specifically a cache hint. > > That would be true for now, but in the future this intrinsic will be used for > adding other types of hints to atomic store in the future. For example in > [ARM-software/acle#430](https://github.com/ARM-software/acle/pull/430) we > will extend the allowed hints to support Contention management hints. So > therefore I think using !mem.cache_hint wouldn't be possible here. Though, if > we generalized this mechanism to support any kind of hint we could use it > here.
cc @YonahGoldberg but I think this still fits within the spirit of `!mem.cache_hint`. Though I'll say that the description of these hints is really not great: > Ensures that the memory write effect of the next instruction occurs before > any other effects from other threads. That doesn't sound like a hint, that sounds like a change to the memory model! Elsewhere I found this description: > That there is a performance benefit if a particular memory access is > sequenced before memory accesses from other threads of execution in the > coherence order to the same location. Whew, that makes more sense... https://github.com/llvm/llvm-project/pull/198316 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
