yangjunpro commented on a change in pull request #5284: [Topi] Tensorcore
support for Conv3D
URL: https://github.com/apache/incubator-tvm/pull/5284#discussion_r407191106
##########
File path: python/tvm/relay/op/nn/_nn.py
##########
@@ -183,6 +193,42 @@ def alter_op_layout_conv3d(attrs, inputs, tinfos,
out_type):
"""Alternate the layout of conv3d"""
return topi.nn.conv3d_alter_layout(attrs, inputs, tinfos, out_type)
[email protected]_convert_op_layout("nn.conv3d")
+def convert_conv3d(attrs, inputs, tinfos, desired_layout):
+ """Convert Layout pass registration for conv3d op.
+
+ Parameters
+ ----------
+ attrs : tvm.ir.Attrs
+ Attributes of current convolution
+ inputs : list of tvm.relay.Expr
+ The args of the Relay expr to be legalized
+ tinfos : list of types
+ List of input and output types
+ desired_layout : str
+ The desired layout
+
+ Returns
+ -------
+ result : tvm.relay.Expr
+ The transformed expr
+ """
+ # pylint: disable=import-outside-toplevel
+ from tvm import relay
+ data, weight = inputs
+ new_attrs = dict(attrs)
+ if desired_layout == 'NCDHW':
+ new_attrs['data_layout'] = desired_layout
Review comment:
Better to move the common attribute assignment out of the if/elif branches.
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
