[llvm-branch-commits] [clang] [llvm] [Offload][Lang] Add blocking to LaunchKernel and Memcpy (PR #218049)

[Kevin Sala Penades via llvm-branch-commits]

================
@@ -45,8 +46,21 @@ ol_result_t __llvmLaunchKernelImpl(const char *KernelID, dim3 GridDim,
   LaunchSizeArgs.GroupSize.z = BlockDim.z;
   LaunchSizeArgs.DynSharedMemory = DynamicSharedMem;
 
-  ol_queue_handle_t Queue = Stream ? reinterpret_cast<StreamTy *>(Stream)->Queue
-                                   : ThreadState.getDefaultQueue();
+  StreamTy *LaunchStream = Stream ? reinterpret_cast<StreamTy *>(Stream)
+                                  : ThreadState.getDefaultStream();
+  if (!LaunchStream || !State.isStreamRegistered(LaunchStream) ||
----------------
kevinsala wrote:

Please remove this check for now. We can re-enable it as a debugging check in separate PR.

```suggestion
 !State.isStreamRegistered(LaunchStream)
```

https://github.com/llvm/llvm-project/pull/218049