anko-intel commented on a change in pull request #20753:
URL: https://github.com/apache/incubator-mxnet/pull/20753#discussion_r826931929
##########
File path: src/operator/subgraph/dnnl/dnnl_subgraph_property.cc
##########
@@ -60,6 +61,9 @@ MXNET_REGISTER_SUBGRAPH_PROPERTY(ONEDNN_QUANTIZE,
SgDNNLPostQuantizeAlignScalePr
MXNET_REGISTER_SUBGRAPH_PROPERTY(ONEDNN_QUANTIZE, SgDNNLFCSumFuseProperty)
.set_attr("quantize", true);
+MXNET_REGISTER_SUBGRAPH_BACKEND(ONEDNN_AMP).set_attr("context",
Context::CPU());
Review comment:
What about mixing quantization with AMP? should it also work for
ONEDNN_QUANTIZE?
If we put it to ONEDNN what should be an order of passes? Could AMP work on
already fused operators ?
How we can run only AMP pass if we wish?
So rather it seems to be a not good idea.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
