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.

Sign up for GitHub

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

Jump to bottom

Add PTX wrapping functions for TMA features #379

Merged

miscco merged 7 commits into NVIDIA:main from ahendriksen:add-cp-async-ptx-wrappers

Sep 13, 2023

Contributor

ahendriksen commented Aug 28, 2023 •

edited

Loading

Description

closes #359

Add PTX wrapping functions for TMA features.

Checklist (still TODO)

New or existing tests cover these changes.
The documentation is up to date with these changes. (There is a working draft of this functionality for the CUDA programming guide)

miscco reviewed

View reviewed changes

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

libcudacxx/include/cuda/std/detail/libcxx/include/__config Outdated Show resolved Hide resolved

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch from 368ceda to 9c1af7a Compare

August 28, 2023 15:31

jrhemstad reviewed

View reviewed changes

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

jrhemstad reviewed

View reviewed changes

libcudacxx/include/cuda/std/detail/libcxx/include/version Show resolved Hide resolved

jrhemstad reviewed

View reviewed changes

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

jrhemstad reviewed

View reviewed changes

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

jrhemstad reviewed

View reviewed changes

libcudacxx/include/cuda/std/detail/libcxx/include/__cuda/barrier.h Outdated Show resolved Hide resolved

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch 3 times, most recently from d5a6e67 to 81fa05f Compare

August 31, 2023 19:52

ahendriksen marked this pull request as ready for review

September 1, 2023 12:41

ahendriksen requested review from a team as code owners

September 1, 2023 12:41

ahendriksen requested review from alliepiper and miscco and removed request for a team

September 1, 2023 12:41

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch from 2543821 to 8ba4aa5 Compare

September 4, 2023 11:28

miscco reviewed

View reviewed changes

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk.pass.cpp Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_1d.pass.cpp Outdated Show resolved Hide resolved

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch 3 times, most recently from e2f7979 to 70018f6 Compare

September 5, 2023 15:08

Contributor Author

ahendriksen commented Sep 5, 2023

I have:

removed superfluous empty lines
replaced __trap by assert in the tests
replaced bar.arrive_tx by cuda::device::arrive_tx
made it possible to skip kernel execution in the NVRTC tests 189bffe.
Made the experimental feature flag available on host compilation trajectory as well, following this discussion.

In terms of documentation, we can either rely on the CUDA programming guide, or perhaps add a section "Experimental API" in addition to the existing "Standard API" and "Extended API" sections.

This PR is now completely ready for (final?) review.

jrhemstad approved these changes

View reviewed changes

Collaborator

jrhemstad left a comment

Merge after #358

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch 2 times, most recently from b1df77a to a00e849 Compare

September 12, 2023 09:55

ahendriksen mentioned this pull request

[FEA]: Allow tests to skip kernel launch when run with NVRTC #433

Closed

1 task

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch from a910014 to 1341f23 Compare

September 12, 2023 12:11

ahendriksen mentioned this pull request

Add NVRTC_SKIP_KERNEL_RUN tag to compile, but skip running NVRTC test #434

Merged

2 tasks

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch 2 times, most recently from 98a7864 to 7028458 Compare

September 12, 2023 13:48

ahendriksen added 4 commits

September 12, 2023 16:01


          Add feature flag

75d9816


          Add cuda::device::experimental namespace

a9da695


          Add PTX wrapping functions

9607f3d


          Add tests

fe67868

ahendriksen force-pushed the add-cp-async-ptx-wrappers branch from 7028458 to fe67868 Compare

September 12, 2023 14:03

miscco reviewed

View reviewed changes

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_feature_test.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_2d.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_generic.h Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_generic.h Outdated Show resolved Hide resolved

ahendriksen and others added 3 commits

September 12, 2023 20:27


          Apply suggestions from code review

c65a523

Co-authored-by: Michael Schellenberger Costa <[email protected]>


          Extract bulk tensor host test code into header

13503da


          Clean up CUTensorMap creation

bc31fd1

ahendriksen commented

View reviewed changes

Contributor Author

ahendriksen left a comment

Thanks for the review @miscco. I have implemented your feedback.

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_2d.pass.cpp Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_generic.h Outdated Show resolved Hide resolved

libcudacxx/.upstream-tests/test/cuda/barrier/cp_async_bulk_tensor_generic.h Outdated Show resolved Hide resolved

miscco approved these changes

View reviewed changes

miscco merged commit abb7cb7 into NVIDIA:main

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet