ibsidorenko opened a new pull request, #13854: URL: https://github.com/apache/tvm/pull/13854
This commit adds new Relay operation `qnn.dense_pack` that supports different weights layout (`nn.dense` and `qnn.dense` do not support this attribute). This new operation is full analog of `nn.contrib_dense_pack` operation but in QNN space. With this PR, current QNN Dense can achieve ~10x performance gain on Hexagon target without QNN canonicalization (through the use of `vrmpy` intrinsic). Also, this PR includes slight performance improvement for `qnn.mul` (without QNN canonicalization). -- 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]
