-
Notifications
You must be signed in to change notification settings - Fork 3.7k
[microNPU] Cascader performance model bugfixes #10510
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
3a54cf7 to
8745836
Compare
ekalda
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks @jacobbohlin, one comment, otherwise looks good! :)
| min(output_shape[2] * output_shape[4], self._max_block_shape.depth), | ||
| min(output_shape[3], self._max_block_shape.width), | ||
| _round_up(min(output_shape[1], max_height), self._micro_block.height), | ||
| min(output_shape[2] * output_shape[4], max_width), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Probably shouldn't take min with max_width
* Fixed incorrect num_blocks calculations for both BufferModes. * Fixed similar issues with Read/Write byte calculations. * Fixed an issue where the 'partkernel' flag was not propagated to the performance estimation code. * Fixed single buffering check incorrectly used output shape and block rather than the input shape and block. * Fixed block config not aligned to micro block for Elementwise. Change-Id: Ide6b231bc1a17c65bed20129d2179a215ada14b2
Changed incorrect usage of 'max_width' to 'max_depth'.
3a54cf7 to
07e7603
Compare
ekalda
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! :)
manupak
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM!
|
Thanks @jacobbohlin @ekalda ! |
This PR fixes some bugs related to the microNPU performance model in the cascader.
This PR builds on and includes the changes in #10508 and will remain as a draft until that is merged.