diff --git a/paddle/fluid/operators/math/depthwise_conv.cu b/paddle/fluid/operators/math/depthwise_conv.cu
index 240cec14dc2662d154a0ef476e197ab446674a54..a372f6fa718e5db6b8de5d77391e0171d82f18dd 100644
--- a/paddle/fluid/operators/math/depthwise_conv.cu
+++ b/paddle/fluid/operators/math/depthwise_conv.cu
@@ -487,8 +487,12 @@ class DepthwiseConvFunctor<platform::CUDADeviceContext, T,
     check_case(1, 2, 3);
     check_case(1, 2, 5);
     check_case(1, 2, -1);
-    check_case(0, 0, 3);
-    check_case(0, 0, 5);
+    check_case(2, 1, 3);
+    check_case(2, 1, 5);
+    check_case(2, 1, -1);
+    check_case(2, 2, 3);
+    check_case(2, 2, 5);
+    check_case(2, 2, -1);
     check_case(0, 0, -1);
 // NOTE(liangdun): 0,0 for other case
 // add other case if needed, e.g. check_case(2^n,1)
diff --git a/python/paddle/fluid/tests/unittests/test_conv2d_op.py b/python/paddle/fluid/tests/unittests/test_conv2d_op.py
index dfc98c2f471f5f3fe9da8e015686cc40c0186845..725953b67df3c48f79e292fef9a465d1781fc692 100644
--- a/python/paddle/fluid/tests/unittests/test_conv2d_op.py
+++ b/python/paddle/fluid/tests/unittests/test_conv2d_op.py
@@ -208,6 +208,48 @@ class TestWith1x1(TestConv2dOp):
         self.groups = 3
 
 
+class TestWithDepthWise3x3(TestConv2dOp):
+    def init_test_case(self):
+        self.pad = [1, 1]
+        self.stride = [1, 1]
+        self.input_size = [3, 4, 10, 10]  # NCHW
+        assert np.mod(self.input_size[1], self.groups) == 0
+        f_c = self.input_size[1] // self.groups
+        self.filter_size = [8, f_c, 3, 3]
+
+    def init_dilation(self):
+        self.dilations = [2, 2]
+
+    def init_group(self):
+        self.groups = 4
+
+
+class TestWithDepthWise5x5(TestConv2dOp):
+    def init_test_case(self):
+        self.pad = [0, 0]
+        self.stride = [1, 1]
+        self.input_size = [2, 4, 10, 10]  # NCHW
+        assert np.mod(self.input_size[1], self.groups) == 0
+        f_c = self.input_size[1] // self.groups
+        self.filter_size = [8, f_c, 5, 5]
+
+    def init_group(self):
+        self.groups = 4
+
+
+class TestWithDepthWise7x7(TestConv2dOp):
+    def init_test_case(self):
+        self.pad = [1, 1]
+        self.stride = [2, 2]
+        self.input_size = [2, 8, 10, 10]  # NCHW
+        assert np.mod(self.input_size[1], self.groups) == 0
+        f_c = self.input_size[1] // self.groups
+        self.filter_size = [16, f_c, 7, 7]
+
+    def init_group(self):
+        self.groups = 8
+
+
 class TestWithDilation(TestConv2dOp):
     def init_test_case(self):
         self.pad = [0, 0]