add more test for conv2d

tests/test_reduction_ops.py

@@ -575,6 +575,12 @@ def test_accuracy_conv1d(shape, kernel, stride, padding, dtype):
     ((32, 16, 8, 8), (32, 4, 4, 4), 4),
     ((18, 16, 4, 4), (16, 8, 2, 2), 2),
     ((9, 16, 4, 4), (128, 8, 2, 2), 2),
+    ((32, 8, 8, 8), (32, 8, 3, 3), 1),
+    ((18, 16, 5, 5), (16, 16, 3, 3), 1),
+    ((9, 16, 7, 7), (128, 4, 3, 3), 4),
+    ((32, 16, 9, 9), (32, 4, 5, 5), 4),
+    ((18, 16, 11, 11), (16, 8, 3, 3), 2),
+    ((9, 16, 6, 6), (128, 8, 3, 3), 2),
 ]
 
 
@@ -623,6 +629,12 @@ def test_accuracy_conv2d(shape, kernel, stride, padding, groups, dtype):
     ((32, 16, 8, 8), (32, 1, 4, 4), (4, 4)),
     ((18, 8, 4, 4), (16, 1, 2, 2), (2, 2)),
     ((9, 4, 4, 4), (128, 1, 2, 2), (2, 2)),
+    ((32, 4, 8, 8), (32, 1, 3, 3), (3, 3)),
+    ((18, 16, 13, 13), (16, 1, 5, 5), (5, 5)),
+    ((9, 32, 8, 8), (128, 1, 3, 3), (3, 3)),
+    ((32, 16, 9, 9), (32, 1, 5, 5), (5, 5)),
+    ((18, 8, 7, 7), (16, 1, 3, 3), (3, 3)),
+    ((9, 4, 6, 6), (128, 1, 3, 3), (3, 3)),
 ]
 
 
@@ -635,13 +647,11 @@ def test_accuracy_depthwise2d(
     shape_input, shape_weight, kernel, stride, padding, dtype
 ):
     inp = torch.randn(shape_input, dtype=dtype, device="cuda", requires_grad=True)
-    ref_inp = to_reference(inp, True)
     torch.backends.cudnn.allow_tf32 = False
     weight = torch.randn(shape_weight, dtype=dtype, device="cuda")
-    ref_weight = to_reference(weight, True)
     ref_out = torch._C._nn._conv_depthwise2d(
-        ref_inp,
-        ref_weight,
+        inp,
+        weight,
         kernel,
         bias=None,
         stride=stride,