KellenSunderland commented on a change in pull request #13398: Use dynamic omp
schedule for sparse dot with large matrix
URL: https://github.com/apache/incubator-mxnet/pull/13398#discussion_r236126238
##########
File path: src/operator/tensor/dot-inl.h
##########
@@ -791,17 +791,31 @@ inline void DotCsrDnsDnsImpl(const OpContext& ctx,
s, num_threads, data_out.dptr<DType>());
}
num_threads = mxnet_op::get_num_threads<cpu>(data_out.shape_[0]);
+ bool dynamic = false;
+ if (data_out.shape_[0] > 1024 * 10) {
Review comment:
Nit: could make "1024 * 10" a const as well and give a meaningful name like
min_parallel_work_threshold.
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
