Skip to content
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

CI: 11/08/24 upstream sync #135

Merged
merged 26 commits into from
Nov 12, 2024
Merged

CI: 11/08/24 upstream sync #135

merged 26 commits into from
Nov 12, 2024

Conversation

github-actions[bot]
Copy link

@github-actions github-actions bot commented Nov 8, 2024

Daily sync with upstream

andportnoy and others added 26 commits October 31, 2024 12:33
…nchmarking-scripts

PiperOrigin-RevId: 694095765
Previously we didn't really fully discharge squeezing the indexed
dims before applying other GMEM transforms, leading to potential
failures because they were not anticipating the increased rank.

PiperOrigin-RevId: 694098739
So far all of our layouts have been tailored to a limited set of use
cases we've tried so far, but they're still not general enough to
handle all of the register layouts needed for WGMMA or mixed precision
matmuls (incl. intermediate steps during conversions). Instead of adding
more special cases, I decided to adopt XLA tiled layouts and they do seem
to work quite well!

This change only lays the groundwork for the new layout system. Future
changes will build upon them to add new features and eventually replace
`WGMMA_LAYOUT` altogether.

PiperOrigin-RevId: 694105514
Remove unnecessary skip statements. Also added tests for bf16 types.

PiperOrigin-RevId: 694130207
PiperOrigin-RevId: 694223298
PiperOrigin-RevId: 694236671
This test only takes around 30s on most hardware platforms, it does not need 10 shards.

PiperOrigin-RevId: 694243316
PiperOrigin-RevId: 694259486
…ce_put works correctly in presence of copy/donate options that user specified.

This change only supports pinned_host -> pinned_host copies on the same device. HBM -> HBM copies don't work yet and donation also doesn't work in PJRT.

This CL also sets up the plumbing from JAX to PJRT so that in the future support for missing features can be added easily.

Fixes jax-ml#24521

PiperOrigin-RevId: 694274616
… GPU dialect.

Also add a lowering rule for `mosaic_gpu.initialize_barrier`.

PiperOrigin-RevId: 694276698
The compilation cache has a dependency on etils.epath if the
cache is not on a local filesystem.

PiperOrigin-RevId: 694311585
@charleshofer charleshofer self-requested a review November 12, 2024 17:15
@charleshofer charleshofer merged commit 0b970b8 into rocm-main Nov 12, 2024
7 checks passed
@charleshofer charleshofer deleted the ci-upstream-sync-13_1 branch November 12, 2024 17:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.