Cross-compile `linux_aarch64` & `linux_ppc64le` #58

conda-forge-admin · 2023-12-19T03:12:26Z

Enables cross-compilation for linux_aarch64 & linux_ppc64le builds (as opposed to emulation, which was used before). This significantly improves the build times of those platforms and should simplify maintenance here.

Was able to do a bit of cleanup in the process thanks to other fixes that have been made in the interim.

Did need to add one workaround for Windows. Though hopefully this can be dropped in the 12.2 upgrade with an upstream fix ( conda-forge/cuda-nvcc-impl-feedstock#4 )

Fixes #34

conda-forge-webservices · 2023-12-19T03:12:36Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

…nda-forge-pinning 2023.12.18.20.43.37

jakirkham · 2023-12-19T03:15:29Z

@conda-forge-admin , please re-render

…nda-forge-pinning 2023.12.18.20.43.37

Borrows the pinning of `cuda-version` from `host`.

Place Python 3.12 `skip` by itself. It is something we plan to drop once we are able. Also place CUDA `skip` by itself. This is something we need to hold onto long term. Drop macOS `skip` as it already falls out of the CUDA skip. So is redundant.

Instead of having the CUDA version defined to build against in various places, consolidate down to just the `version` used by `cuda-python` at the beginning of the recipe. The propagate this value into how the `skip`s are configured to select the relevant CUDA version. Also leverage this in how `cuda-version` is constrained in the `requirements` during the build phase. This should simplify the upgrade process of `cuda-python` for new versions (including major CUDA version bumps).

jakirkham · 2023-12-19T03:35:05Z

@conda-forge-admin , please re-render

github-actions · 2023-12-19T03:37:00Z

Hi! This is the friendly automated conda-forge-webservice.

I tried to rerender for you, but it looks like there was nothing to do.

This message was generated by GitHub actions workflow run https://github.com/conda-forge/cuda-python-feedstock/actions/runs/7257034521.

jakirkham · 2023-12-19T04:29:19Z

Seeing errors like this on CI:

  Error compiling Cython file:
  ------------------------------------------------------------
  ...
  cdef cudaError_t _cudaProfilerStart() nogil except ?cudaErrorCallRequiresNewerDriver:
      cdef cudaError_t err
      err = m_global.lazyInitContextState()
      if err != cudaSuccess:
          return err
      err = <cudaError_t>ccuda._cuProfilerStart()
           ^
  ------------------------------------------------------------

  cuda/_lib/ccudart/ccudart.pyx:3625:10: Casting temporary Python object to non-numeric non-Python type

Think these are masking the real issue, which is that cuda-python is having trouble finding cuda-profiler-api. This comes back to the fact that we encountered some issues with splayed layouts in the cuda-python recipe before ( #33 (comment) ) ( NVIDIA/cuda-python#46 )

We worked around this issue previously by placing some dependencies (like cuda-profiler-api) in requirements/build. However those packages will be installed to match the build_platform (not the target_platform), which means their targets contents will be in a directory determined by build_platform (and not target_platform where we search for headers and libraries). Hence this is causing headers for dependencies (like cuda-profiler-api) not to be found, which leads to opaque errors like those above (as Cython assumes these are Python objects as it doesn't see the C APIs as available)

Previously we pointed `CUDA_HOME` to `BUILD_PREFIX`. Though ideally we would point to `PREFIX` as that is where headers and libraries for the intended `target_platform` would be. So make this change. This may require more changes given `cuda-python` does not support splayed layouts.

This reverts commit 5ffc9e5. It looks like `LIBRARY_PREFIX` is not defined when `script_env`s are set. Given this, revert to `PREFIX`, which is defined when setting `script_env`s.

jakirkham

Looks like a lot of the previous issues on Linux that were seen here got cleaned up. Am guessing this came out of changes to how headers and libraries are searched for ( conda-forge/cuda-nvcc-feedstock#16 )

Windows ran into issues finding crt headers though. It doesn't have the same setup as Linux though. So we lack some of the same flexibility in setting flags. This should go away once crt headers can be added cleanly as a dependency in CUDA 12.2. For now added a simple workaround for Windows that with more details below

jakirkham · 2023-12-19T05:54:02Z

recipe/meta.yaml

  script:
+    - set "INCLUDE=%INCLUDE%;%BUILD_PREFIX%\Library\include"  # [win64]


On Windows, the builds ran into an issue finding the crt headers on CI:

building 'cuda._lib.ccudart.ccudart' extension creating build\temp.win-amd64-cpython-311\Release\cuda\_lib\ccudart "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.29.30133\bin\HostX64\x64\cl.exe" /c /nologo /O2 /W3 /GL /DNDEBUG /MD -I.\cuda -ID:\bld\cuda-python_1702961331834\_h_env -ID:\bld\cuda-python_1702961331834\_h_env\Library/include -ID:\bld\cuda-python_1702961331834\_h_env\include -ID:\bld\cuda-python_1702961331834\_h_env\Include "-IC:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.29.30133\include" "-IC:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.29.30133\ATLMFC\include" "-IC:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\VS\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\um" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\shared" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\winrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\cppwinrt" "-IC:\Program Files (x86)\Windows Kits\NETFXSDK\4.8\include\um" -ID:\bld\cuda-python_1702961331834\_h_env\Library\include -ID:\bld\cuda-python_1702961331834\_h_env\Library\include -ID:\bld\cuda-python_1702961331834\_h_env\Library\include\targets\x64 /d1trimfile:D:\bld\cuda-python_1702961331834\work /EHsc /Tpcuda/_lib/ccudart\ccudart.cpp /Fobuild\temp.win-amd64-cpython-311\Release\cuda/_lib/ccudart\ccudart.obj ccudart.cpp D:\bld\cuda-python_1702961331834\_h_env\Library/include\vector_types.h(65): fatal error C1083: Cannot open include file: 'crt/host_defines.h': No such file or directory error: command 'C:\\Program Files\\Microsoft Visual Studio\\2022\\Enterprise\\VC\\Tools\\MSVC\\14.29.30133\\bin\\HostX64\\x64\\cl.exe' failed with exit code 2 error: subprocess-exited-with-error

The crt headers are part of the compiler currently. This will be broken out as their own package in CUDA 12.2 ( conda-forge/cuda-nvcc-impl-feedstock#4 )

For now this adds the NVCC includes to the end of INCLUDE so they are searched as well (after exhausting all other search paths)

This fixes the Windows build issue with crt headers

Tracking this in issue: #74

Attempting to address this with PR: #75

jakirkham · 2023-12-19T06:34:39Z

@conda-forge-admin , please re-render

github-actions · 2023-12-19T06:36:36Z

Hi! This is the friendly automated conda-forge-webservice.

I tried to rerender for you, but it looks like there was nothing to do.

This message was generated by GitHub actions workflow run https://github.com/conda-forge/cuda-python-feedstock/actions/runs/7258263787.

jakirkham · 2023-12-19T23:18:01Z

Thanks all! 🙏

Going to go ahead and merge this so we can use cross-compilation for the upgrades. Hope that is ok 🙂

dummy commit for rerendering

809e2e0

conda-forge-admin requested review from jakirkham, leofang, m3vaz and mmccarty as code owners December 19, 2023 03:12

conda-forge-webservices bot mentioned this pull request Dec 19, 2023

Use cross-compilation instead #34

Closed

MNT: Re-rendered with conda-build 3.27.0, conda-smithy 3.30.1, and co…

e05bfd1

…nda-forge-pinning 2023.12.18.20.43.37

jakirkham marked this pull request as draft December 19, 2023 03:14

Switch to cross-compiling for arch builds

2e633e3

MNT: Re-rendered with conda-build 3.27.0, conda-smithy 3.30.1, and co…

3b43512

…nda-forge-pinning 2023.12.18.20.43.37

This comment was marked as off-topic.

Sign in to view

jakirkham changed the title ~~MNT: rerender~~ Cross-compile linux_aarch64 & linux_ppc64le Dec 19, 2023

jakirkham added 3 commits December 18, 2023 19:31

Add cuda-version to build too

7b459c8

Borrows the pinning of `cuda-version` from `host`.

Break up skips

b209cd7

Place Python 3.12 `skip` by itself. It is something we plan to drop once we are able. Also place CUDA `skip` by itself. This is something we need to hold onto long term. Drop macOS `skip` as it already falls out of the CUDA skip. So is redundant.

Add Python cross-compilation dependencies

6698452

jakirkham added 4 commits December 18, 2023 20:40

Set CUDA_HOME to LIBRARY_PREFIX on Windows

5ffc9e5

Check BUILD_PREFIX headers last on Windows

9226503

Revert "Set CUDA_HOME to LIBRARY_PREFIX on Windows"

d5bd2bd

This reverts commit 5ffc9e5. It looks like `LIBRARY_PREFIX` is not defined when `script_env`s are set. Given this, revert to `PREFIX`, which is defined when setting `script_env`s.

jakirkham reviewed Dec 19, 2023

View reviewed changes

jakirkham marked this pull request as ready for review December 19, 2023 06:31

Bump build/number to 4

bd30a28

jakirkham mentioned this pull request Dec 19, 2023

cuda-python v12.1.0 #39

Merged

3 tasks

bdice approved these changes Dec 19, 2023

View reviewed changes

leofang approved these changes Dec 19, 2023

View reviewed changes

jakirkham merged commit 16af942 into conda-forge:main Dec 19, 2023
18 checks passed

jakirkham mentioned this pull request Mar 7, 2024

Drop environment variable workarounds in build #74

Closed

leofang mentioned this pull request Jun 25, 2024

Splayed layout support NVIDIA/cuda-python#46

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Cross-compile `linux_aarch64` & `linux_ppc64le` #58

Cross-compile `linux_aarch64` & `linux_ppc64le` #58

conda-forge-admin commented Dec 19, 2023 •

edited by jakirkham

Loading

conda-forge-webservices bot commented Dec 19, 2023

jakirkham commented Dec 19, 2023

This comment was marked as off-topic.

jakirkham commented Dec 19, 2023

github-actions bot commented Dec 19, 2023

jakirkham commented Dec 19, 2023

jakirkham left a comment

jakirkham Dec 19, 2023

jakirkham Mar 7, 2024

jakirkham Mar 7, 2024

jakirkham commented Dec 19, 2023

github-actions bot commented Dec 19, 2023

jakirkham commented Dec 19, 2023

		script:
		- set "INCLUDE=%INCLUDE%;%BUILD_PREFIX%\Library\include" # [win64]

Cross-compile linux_aarch64 & linux_ppc64le #58

Cross-compile linux_aarch64 & linux_ppc64le #58

Conversation

conda-forge-admin commented Dec 19, 2023 • edited by jakirkham Loading

conda-forge-webservices bot commented Dec 19, 2023

jakirkham commented Dec 19, 2023

This comment was marked as off-topic.

jakirkham commented Dec 19, 2023

github-actions bot commented Dec 19, 2023

jakirkham commented Dec 19, 2023

jakirkham left a comment

Choose a reason for hiding this comment

jakirkham Dec 19, 2023

Choose a reason for hiding this comment

jakirkham Mar 7, 2024

Choose a reason for hiding this comment

jakirkham Mar 7, 2024

Choose a reason for hiding this comment

jakirkham commented Dec 19, 2023

github-actions bot commented Dec 19, 2023

jakirkham commented Dec 19, 2023

Cross-compile `linux_aarch64` & `linux_ppc64le` #58

Cross-compile `linux_aarch64` & `linux_ppc64le` #58

conda-forge-admin commented Dec 19, 2023 •

edited by jakirkham

Loading