[GitHub] [tvm] jcf94 commented on a change in pull request #7147: [CUDA][PASS]Legalize tensorcore

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

jcf94 commented on a change in pull request #7147:
URL: https://github.com/apache/tvm/pull/7147#discussion_r547889252



##########
File path: python/tvm/topi/cuda/conv2d_alter_op.py
##########
@@ -345,4 +347,49 @@ def _conv2d_legalize(attrs, inputs, arg_types):
             else:
                 out = relay.nn.conv2d(data, kernel, **new_attrs)
             return out
+    elif data_dtype in ['float16', 'float32']:

Review comment:
       #7146 added data type transform, so it's fine that both float16 and float32 can all be processed to support tensorcore.
   But I'll still suggest @Meteorix to consider add some options in the whole flow that enables to close the tensor core feature manually, or add more strictly conditions here. In case if someone just don't want to use tensorcore or some shape/datatype cause special bugs.




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