Baunsgaard opened a new pull request #1078: URL: https://github.com/apache/systemds/pull/1078
This PR is sort of an question. Is it okay for our MKL to assume that the output matrix is fully dense with all values !=0? If so then we have an potential improved performance. I have observed up to 30% in some cases (most likely due to inter run variance) but around 10% on average, and it is because the native call to matrix multiplication has to transfer the matrix out and and then into java again and then afterwards it loops through all values to count 0's forcing another iteration through the matrix once more. ---------------------------------------------------------------- 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: us...@infra.apache.org
