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

[Sophia Herrmann via cfe-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) ||
+      LaunchStream->Device != Device)
+    return &InvalidConfigurationError;
+
+  if (LaunchStream->Kind == QueueKind::LegacyDefault) {
+    ol_result_t Result = waitOnBlockingStreams();
+    if (Result != OL_SUCCESS)
+      return Result;
+  } else if (LaunchStream->Kind == QueueKind::ExplicitBlocking) {
+    ol_result_t Result = waitOnLegacyDefaultStream(LaunchStream, Device);
----------------
jellytabby wrote:

Done

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