[libc-commits] [libc] [libc] Add memory fence utility to the GPU utilities (PR #91756)

[Matt Arsenault via libc-commits]

================
@@ -140,6 +140,12 @@ LIBC_INLINE uint32_t get_lane_size() {
   __builtin_amdgcn_fence(__ATOMIC_ACQUIRE, "workgroup");
 }
 
+/// Waits for all pending memory operations to complete in program order.
+[[clang::convergent]] LIBC_INLINE void memory_fence() {
+  __builtin_amdgcn_s_waitcnt(0);
+  __builtin_amdgcn_fence(__ATOMIC_ACQ_REL, "");
----------------
arsenm wrote:

Probably just shouldn't have any fence utilities that don't have an explicit ordering and scope (or at least name what they are in the function name) 

https://github.com/llvm/llvm-project/pull/91756