[ https://issues.apache.org/jira/browse/MXNET-97?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Ni Hui updated MXNET-97: ------------------------ Labels: performance (was: ) > implement DepthwiseConv2dBackwardFilterKernel from tensorflow codebase > ---------------------------------------------------------------------- > > Key: MXNET-97 > URL: https://issues.apache.org/jira/browse/MXNET-97 > Project: Apache MXNet > Issue Type: Improvement > Components: MXNet Engine > Reporter: Ni Hui > Priority: Major > Labels: performance > > The current mxnet implementation calls __syncthreads() function too much, > which is extemely slow. > The new code comes from tensorflow, but the variable names are adjusted for > consistency. > My model uses depthwise conv heavily, and now its training time per iteration > is over 5x faster on single P40 gpu. ( old 92s vs new 18s ) -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org