t-vi commented on issue #5686: URL: https://github.com/apache/incubator-tvm/issues/5686#issuecomment-638753031
This sounds similar to me to the symptoms discussed in the recent posts in #5600. So quite likely, making the cuda softmax schedule specific to cuda would fix this. Of course, implementing shuffle reductions where they exist is the alternative... I'm not entirely sure how to improve the error message (because the root cause is that warp shuffle reductions are not supported on the target). ---------------------------------------------------------------- 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]
