[llvm] [flang][cuda] Use the provided stream in kernel launch (PR #135267)

[Zhen Wang via llvm-commits]

================
@@ -74,9 +75,9 @@ void RTDEF(CUFLaunchKernel)(const void *kernel, intptr_t gridX, intptr_t gridY,
     Fortran::runtime::Terminator terminator{__FILE__, __LINE__};
     terminator.Crash("Too many invalid grid dimensions");
   }
-  cudaStream_t cuStream = 0; // TODO stream managment
-  CUDA_REPORT_IF_ERROR(
-      cudaLaunchKernel(kernel, gridDim, blockDim, params, smem, cuStream));
+  cudaStream_t deafultStream = 0;
----------------
wangzpgi wrote:

same here

https://github.com/llvm/llvm-project/pull/135267