mboehm7 commented on pull request #1174: URL: https://github.com/apache/systemds/pull/1174#issuecomment-770297542
LGTM - thanks for the extension @sebwrede, this is great. I think it's the right step to first call the existing conv2d (for which internal Java/C++/GPU operators with dedicated implementations for combinations of dense/sparse inputs exist). If we measure substantial overhead of conv1d via conv2d then we could go ahead and add the different kernels. Btw, I see that the conv2d / pooling layers might be confusing as _builtin refers to wrappers around natively supported operators, while in scripts/builtin we have dml-bodied functions. Perhaps we should remove the script-based conv2d and pooling layers altogether. ---------------------------------------------------------------- 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: [email protected]
