*correctly* fix the default cuda context init in pycuda accuracy tests...

Author: vincefn

pyvkfft/accuracy.py

@@ -89,7 +89,7 @@ def init_ctx(backend, gpu_name=None, opencl_platform=None, verbose=False):
             else:
                 raise RuntimeError("Selected backend is pycuda, but no device found")
         gpu_ctx_dic["pycuda"] = (d, d.retain_primary_context())
-        gpu_ctx_dic["pycuda"].push()
+        gpu_ctx_dic["pycuda"][1].push()
         if verbose:
             print("Selected device for pycuda: %s" % d.name())
     elif backend == "pyopencl":