[GitHub] [incubator-tvm] apivovarov commented on a change in pull request #4472: Workaround to make conv2d_transpose compilation for CUDA work

[GitBox <gi...@apache.org>]

apivovarov commented on a change in pull request #4472: Workaround to make conv2d_transpose compilation for CUDA work
URL: https://github.com/apache/incubator-tvm/pull/4472#discussion_r354992237
 
 

 ##########
 File path: topi/python/topi/cuda/conv2d_transpose_nchw.py
 ##########
 @@ -186,7 +186,9 @@ def _callback(op):
 
             if cfg.is_fallback:
                 N, F, Y, X = get_const_tuple(conv.shape)
-                _fallback_schedule(N, F, Y, X)
+                # Workaround to make CUDA compilation work. Issue #4470
 
 Review comment:
   I checked more kernel and strides combinations and found that the error happens when kernel is equal to strides, e.g.
   ```
   # kernel and strides when compilation for CUDA fails
   2x2 and (2,2)
   3x3 and (3,3)
   4x4 and (4,4)
   5x5 and (5,5)
   2x3 and (2,3)
   3x2 and (3,2)
   ```

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services