ssahasra wrote: > FWIW, SWDEV-509877 requests for builtins/intrinsics for 128-bit atomic loads > and stores for global / flat / ds memory spaces. They require a memory > ordering for each of those builtins, but they would be okay with `relaxed` -- > which should match behavior as implemented in this PR (though, please correct > me if wrong here).
This PR implements "non-atomic accesses with explicit cache control". The proposed official name for them is "store-available" and "load-visible" operations. They do have a scope argument, which can make them look like atomics, but there is no promise of atomicity on these intrinsics. https://github.com/llvm/llvm-project/pull/172090 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
