-
Notifications
You must be signed in to change notification settings - Fork 62
Do not generate 2D block loads with sizes > the block size #4092
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
Comments
Here's a specific instance where we seem to be generating tiles that are too big:
The block size here is A matrix: These are 32x32 loads - but the block size outer dim is 16. If we dump the DPAS layout using
then we see that the rows 16-32 are never used
So, we should not be creating a 32x32 load. |
Currently we attempt to generate the largest 2D block load given hardware constraints. If the block size is smaller than the constrained load size, we will load more data than is required in each 2D block load invocation. This seems wasteful and makes it difficult to generate the Load -> DPAS mapping using layouts.
The text was updated successfully, but these errors were encountered: