[OpenMP] Remove hack around missing atomic load (llvm#122781)

jhuber6 · web-flow · commit 1c00d0d7768f · 2025-01-16T15:17:15.000-06:00
Summary:
We used to do a fetch add of zero to approximate a load. This is because
the NVPTX backend didn't handle this properly. It's not an issue anymore
so simply use the proper atomic builtin.
diff --git a/offload/DeviceRTL/include/Synchronization.h b/offload/DeviceRTL/include/Synchronization.h
@@ -66,7 +66,7 @@ V add(Ty *Address, V Val, atomic::OrderingTy Ordering) {
 
 template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>
 V load(Ty *Address, atomic::OrderingTy Ordering) {
-  return add(Address, Ty(0), Ordering);
+  return __scoped_atomic_load_n(Address, Ordering, __MEMORY_SCOPE_DEVICE);
 }
 
 template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>

Original file line number	Diff line number	Diff line change
`@@ -66,7 +66,7 @@ V add(Ty *Address, V Val, atomic::OrderingTy Ordering) {`
`66`	`66`
`67`	`67`	`template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>`
`68`	`68`	`V load(Ty *Address, atomic::OrderingTy Ordering) {`
`69`		`- return add(Address, Ty(0), Ordering);`
	`69`	`+ return __scoped_atomic_load_n(Address, Ordering, __MEMORY_SCOPE_DEVICE);`
`70`	`70`	`}`
`71`	`71`
`72`	`72`	`template <typename Ty, typename V = utils::remove_addrspace_t<Ty>>`