diff --git a/pycuda/gpuarray.py b/pycuda/gpuarray.py
index 97630e3bdeb572e2d881deeea840d1ff1bce3cc0..a1a3f3f34848f9e1ca7eb7b9da2b31098cb0cb65 100644
--- a/pycuda/gpuarray.py
+++ b/pycuda/gpuarray.py
@@ -627,6 +627,10 @@ class GPUArray:
 
     def fill(self, value, stream=None):
         """fills the array with the specified value"""
+        if not self.flags.forc:
+            raise RuntimeError(
+                "only contiguous arrays may be used as arguments to this operation")
+
         func = elementwise.get_fill_kernel(self.dtype)
         func.prepared_async_call(
             self._grid, self._block, stream, value, self.gpudata, self.mem_size
@@ -650,8 +654,7 @@ class GPUArray:
     ):
         if not self.flags.forc:
             raise RuntimeError(
-                "only contiguous arrays may " "be used as arguments to this operation"
-            )
+                "only contiguous arrays may be used as arguments to this operation")
 
         if self.dtype == np.float64 and allow_double_hack:
             if channels != 1:
diff --git a/test/test_gpuarray.py b/test/test_gpuarray.py
index fb6a20fcffc6943bf814d8a2fb0d421ba17cb8f9..fc7b67366c509089fd3a26b4e3c183fde1ca9269 100644
--- a/test/test_gpuarray.py
+++ b/test/test_gpuarray.py
@@ -1192,6 +1192,9 @@ class TestGPUArray:
 
                 contig = arr.flags.c_contiguous or arr.flags.f_contiguous
 
+                if not contig:
+                    continue
+
                 # Output matches order of input.
                 # Non-contiguous becomes C-contiguous
                 new_z = func(arr, order="A")