Non-argument streams are now correctly passed in the CuPy and NVCUDA backends

Author: fjwillemsen

kernel_tuner/backends/cupy.py

@@ -198,6 +198,8 @@ def run_kernel(self, func, gpu_args, threads, grid, stream=None):
             of the grid
         :type grid: tuple(int, int)
         """
+        if stream is None:
+            stream = self.stream
         func(grid, threads, gpu_args, stream=stream, shared_mem=self.smem_size)
 
     def memset(self, allocation, value, size):

kernel_tuner/backends/nvcuda.py

@@ -278,6 +278,8 @@ def run_kernel(self, func, gpu_args, threads, grid, stream=None):
             of the grid
         :type grid: tuple(int, int)
         """
+        if stream is None:
+            stream = self.stream
         arg_types = list()
         for arg in gpu_args:
             if isinstance(arg, cuda.CUdeviceptr):