cusignal icon indicating copy to clipboard operation
cusignal copied to clipboard

[FEA] Add aarch64 build support into CI/CD for cuSignal

Open awthomp opened this issue 5 years ago • 8 comments

cuSignal 0.13 currently has bare-bones instructions for getting cuSignal built and running using Anaconda on a Jetson Embedded GPU. Since the main cuSignal developers are primarily working on PCIe versions of GPUs, some merged PRs may break aarch64 builds; this recently came up with a Numba versioning issue. Since cuSignal has a significant number of users on a Jetson Nano, TX2, or Xavier, we should simplify the SW delivery to these users and ensure a positive experience.

awthomp avatar Apr 13 '20 12:04 awthomp

The core cuSignal dependencies on the Jetson GPUs are:

  • scipy
  • numpy
  • numba>=0.49
  • cupy >= 7.2

Scipy, Numba, and Numpy all have anaconda images built for aarch64 and are available via conda-forge or numba channels. CuPy, however, is seemingly blocked in their aarch64 build.

See: https://github.com/cupy/cupy/issues/3196 https://github.com/AnacondaRecipes/cudatoolkit-feedstock/issues/3

The current work-around for source installs is to install pip into the conda environment and then run pip install cupy>=7.2.

awthomp avatar May 07 '20 12:05 awthomp

This issue has been marked stale due to no recent activity in the past 30d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be marked rotten if there is no activity in the next 60d.

github-actions[bot] avatar Feb 16 '21 20:02 github-actions[bot]

This issue has been marked rotten due to no recent activity in the past 90d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

github-actions[bot] avatar Feb 16 '21 20:02 github-actions[bot]

Marked rotten? Huh?

I have this working with conda-forge in a Conda environment on my Jetsons in https://github.com/edgyR/edgyR-containers.git FWIW ... if there's a CI/CD provider I can use for free I'm all ears. :-)

znmeb avatar Feb 16 '21 23:02 znmeb

@znmeb -- yeah, there was a recent CI change that made all older issues without active conversation/PRs 'rotten'. I want to keep this one open. To be honest, I've been working more on the datacenter rather than edge side and will restart the conversation on where we are with Jetson CI/CD with RAPIDS.

awthomp avatar Feb 17 '21 00:02 awthomp

TBH my bottleneck can be fixed with pure arm64 Ubuntu - my builder images are what takes the time and those don't touch the GPU. But the only thing I've been able to find to get those done is paid time on AWS Gravitons, and I'm suspecting one of the builds needs more than 4 GB of RAM.

znmeb avatar Feb 17 '21 01:02 znmeb

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

github-actions[bot] avatar Mar 19 '21 01:03 github-actions[bot]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

github-actions[bot] avatar Nov 23 '21 20:11 github-actions[bot]