diff --git a/topi/python/topi/cuda/conv2d_winograd.py b/topi/python/topi/cuda/conv2d_winograd.py
index 50c485290bbf..eb004d79fd9b 100644
--- a/topi/python/topi/cuda/conv2d_winograd.py
+++ b/topi/python/topi/cuda/conv2d_winograd.py
@@ -17,6 +17,7 @@
 # pylint: disable=invalid-name,unused-variable,unused-argument
 """Winograd template for cuda backend"""
 
+import logging
 import tvm
 from tvm import autotvm
 
@@ -27,6 +28,8 @@
 from ..nn.winograd_util import winograd_transform_matrices
 
 
+logger = logging.getLogger('conv2d_winograd')
+
 def _infer_tile_size(data, kernel):
     N, CI, H, W = get_const_tuple(data.shape)
 
@@ -42,26 +45,25 @@ def winograd_cuda(cfg, data, kernel, strides, padding, dilation, layout, out_dty
 
     N, CI, H, W = get_const_tuple(data.shape)
 
+    if isinstance(dilation, int):
+        dilation_h = dilation_w = dilation
+    else:
+        dilation_h, dilation_w = dilation
+    HSTR, WSTR = (strides, strides) if isinstance(strides, int) else strides
+
     if not pre_computed: # kernel tensor is raw tensor, do strict check
-        if isinstance(dilation, int):
-            dilation_h = dilation_w = dilation
-        else:
-            dilation_h, dilation_w = dilation
         if dilation_h != 1 or dilation_w != 1:
-            kernel = dilate(kernel, (1, 1, dilation_h, dilation_w))
-
+            kernel = dilation(kernel, (1, 1, dilation_h, dilation_w))
         CO, CI, KH, KW = get_const_tuple(kernel.shape)
-        HPAD, WPAD, _, _ = nn.get_pad_tuple(padding, kernel)
-        HSTR, WSTR = (strides, strides) if isinstance(strides, int) else strides
         assert HSTR == 1 and WSTR == 1 and KH == KW
-    else:                   # kernel tensor is pre-transfomred. this op is created by
-                            # alter op layout, do not check
+    else:
+        # kernel tensor is pre-transfomred. this op is created by alter op layout.
         # dilation is not supported
-        HSTR = WSTR = 1
-        HPAD = WPAD = 1
-        KH = KW = 3
         _, _, CI, CO = get_const_tuple(kernel.shape)
+        KH = KW = 3
+        assert HSTR == 1 and WSTR == 1 and dilation_h == 1 and dilation_w == 1
 
+    HPAD, WPAD, _, _ = nn.get_pad_tuple(padding, kernel)
     data_pad = nn.pad(data, (0, 0, HPAD, WPAD), (0, 0, HPAD, WPAD), name="data_pad")
 
     r = KW
@@ -384,7 +386,7 @@ def _alter_conv2d_layout(attrs, inputs, tinfos, F):
             return F.nn.conv2d(*copy_inputs, **new_attrs)
 
         if attrs.get_int_tuple("dilation") != (1, 1):
-            warnings.warn("Does not support weight pre-transform for dilated convolution.")
+            logger.warning("Does not support weight pre-transform for dilated convolution.")
             return None
 
         # pre-compute weight transformation in winograd
diff --git a/topi/tests/python/common.py b/topi/tests/python/common.py
index 0bacd61129b7..4e0a45be0a22 100644
--- a/topi/tests/python/common.py
+++ b/topi/tests/python/common.py
@@ -40,4 +40,5 @@ def _query_inside(self, target, workload):
         cfg = FallbackConfigEntity()
         cfg.template_key = 'int8'
         self.memory[key] = cfg
+        cfg.is_fallback = False
         return cfg
diff --git a/topi/tests/python/test_topi_conv2d_winograd.py b/topi/tests/python/test_topi_conv2d_winograd.py
index 908317ebf55e..5974dad20f88 100644
--- a/topi/tests/python/test_topi_conv2d_winograd.py
+++ b/topi/tests/python/test_topi_conv2d_winograd.py
@@ -99,6 +99,7 @@ def _query_inside(self, target, workload):
         cfg = FallbackConfigEntity()
         cfg.template_key = 'winograd'
         self.memory[key] = cfg
+        cfg.is_fallback = False
         return cfg