[GitHub] [tvm] Wheest commented on a change in pull request #7142: Asymmetric padding and dilation in conv2d workload

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

Wheest commented on a change in pull request #7142:
URL: https://github.com/apache/tvm/pull/7142#discussion_r549681173



##########
File path: python/tvm/topi/nn/depthwise_conv2d.py
##########
@@ -38,20 +38,26 @@
         "out_filter",
         "hkernel",
         "wkernel",
-        "hpad",
-        "wpad",
+        "padt",
+        "padl",
+        "padb",
+        "padr",
+        "hdilation",
+        "wdilation",
         "hstride",
         "wstride",
     ],
 )
 
 
-def _get_workload(data, kernel, stride, padding, out_dtype):
+def _get_workload(data, kernel, stride, padding, dilation, out_dtype):
     """ Get the workload structure. """
     _, in_channel, height, width = [x.value for x in data.shape]
     channel, channel_multiplier, kh, kw = [x.value for x in kernel.shape]
     out_channel = channel * channel_multiplier
-    HPAD, WPAD, _, _ = get_pad_tuple(padding, kernel)
+    pt, pl, pb, pr = get_pad_tuple(padding, kernel)
+    hdilation, wdilation = dilation if isinstance(dilation, (tuple, list)) else (dilation, dilation)

Review comment:
       Should this also apply to their names in the Workload tuples?  Or just the variables in the `_get_workload()` functions?
   
   I named them `hdilation`, and `wdilation` in the Workload to follow the `"hstride"`, `"wstride",` style.




----------------------------------------------------------------
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