-
Notifications
You must be signed in to change notification settings - Fork 195
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
[FEA] Padding support for mdspan
and mdarray
.
#497
Comments
This issue has been labeled |
I'll start working on this. |
I added #663 as draft for discussion. For now I have chosen not to implement a full custom layout as this would require changes to the |
This issue has been labeled |
I have added a new PR with a full static approach #725. |
This issue has been labeled |
Please see also mdspan PR kokkos/mdspan#180 and discussion on PR #725; thanks! |
To clarify my above comment: |
mdspan PR kokkos/mdspan#180 ( |
This issue has been labeled |
This is a different approach / followup PR of #663 for issue #497. I implemented a `layout_padded_general` within raft to statically enforce padding on mdpsan accesses. * The layout has template parameters for `ValueType`, `StorageOrder `(default `row_major_t`), and `ByteAlignment `(default 128) * in order to *not* require changes upstream I skipped `submdspan `functionality right now. I have a branch on a mdspan fork where I tested this though (https://github.com/mfoerste4/mdspan/tree/layout_padded). Authors: - Malte Förster (https://github.com/mfoerste4) Approvers: - Artem M. Chirkin (https://github.com/achirkin) - Corey J. Nolet (https://github.com/cjnolet) URL: #725
The integration of
mdspan
introduced in #437 allocates the memory based on the functionrequired_span_size
from layouts. We can customize the layout classes to change the required size and pad the last dimension of the contiguous array. This way we can use optimizations like vector load with the new mdspan.The text was updated successfully, but these errors were encountered: