jwfromm commented on a change in pull request #7147:
URL: https://github.com/apache/tvm/pull/7147#discussion_r547655745
##########
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:
Why are we checking for 'float32'? My understanding is that tensorcores
only work on smaller datatypes. You should consider looking at some of the
support TVM has for running int8 and int4 workloads on tensorcores as well.
----------------------------------------------------------------
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]
