[GitHub] [incubator-mxnet] Neutron3529 commented on a change in pull request #20188: [BUGFIX]try avoid the error in operator/tensor/amp_cast.h

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

Neutron3529 commented on a change in pull request #20188:
URL: https://github.com/apache/incubator-mxnet/pull/20188#discussion_r616061336



##########
File path: 3rdparty/mshadow/mshadow/base.h
##########
@@ -1364,7 +1364,8 @@ struct minimum {
   default:                                          \
     LOG(FATAL) << "Unknown type enum " << type;     \
   }
-
+// amp_cast.h is using this MSHADOW_TYPE_SWITCH_WITH_BOOL in order to
+// avoid 'Unsupport enum type 12' error.

Review comment:
       Actually I put this comment as a TODO...
   It is more harder than I thought to make bf16 work.
   I had tried my best, but I have no idea about convolution ops.
   This is a very very dirty fix and should be modified by another PR.
   
   this PR only enable bf16 on dense layers
   There are lots of works todo in the future.
   
   Is it OK to mark a TODO on this comment?
   Or simply delete this comment?




-- 
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