masahi commented on PR #14817: URL: https://github.com/apache/tvm/pull/14817#issuecomment-1565288152
I've updated my vk 4K matmul test https://github.com/masahi/tensorir-experiment/blob/master/vk_cooperative_matrix_nv/test_4k.py to use this extension support. It gets ~90 TFLOPs on RTX4080. The cool part is that by changing the target from vk to cuda the exact same schedule / script works with the same performance. https://github.com/masahi/tensorir-experiment/blob/master/vk_cooperative_matrix_nv/test_4k.py#L171-L172 -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
