mxnet-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From GitBox <...@apache.org>
Subject [GitHub] mseeger commented on issue #8502: Fix of log10_grad, log2_grad
Date Thu, 01 Jan 1970 00:00:00 GMT
mseeger commented on issue #8502: Fix of log10_grad, log2_grad
URL: https://github.com/apache/incubator-mxnet/pull/8502#issuecomment-341065583
 
 
   @piiswrong The change you are asking for (good catch).
   
   BTW: One open point here is whether all DType's are supported. I think that mshadow::half::half2_t
is not, because conversion to/from float is missing.
   I am not much interested in half-precision arithmetic, so would not want to own doing this,
but if somebody works on CUDA half2 support and needs help with mshadow_op.h, please loop
me in.
   
   I note that CUDA has own implementations for a subset of math.h functions for its half
types. But I also have the feeling that this is never the bottleneck for DNNs, and so casting
to/from float for them seems OK to me.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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


With regards,
Apache Git Services

Mime
View raw message