NVIDIA
[QST] Python Conv2d example?
Hello, I would like to implement a custom Pytorch kernel using the CUTLASS 2D convolution. I saw that you released Python scripts in release 2.9 to launch a GEMM kernel from Python. Is there a similar script to launch a 2D convolution fprop kernel? Same for wgrad and dgrad? I am currently looking at how to transpose what you did for GEMM but I was wondering whether you had already done the work. That would be very helpful. Thanks a lot!
Hello! Thank you for expressing interest.
There is not a similar script for convolution, but there could be. Nearly all of the details are implemented already or follow the existing pattern. I expect it would be low effort.
Could I work with you to add fprop initially? I can devote some time this week and submit a pull request to you for evaluation.
Once that works and you have had the opportunity to try it out, perhaps you would be willing to add additional steps for to complete the convolution gradient passes? :)
Hi Andrew, Yes that would be great, thank you! Sure no problem, I could try to complete the gradient versions once fprop is done, this way I'll see if I understood the design or not.
This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}
This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.
