Update to 1.22.2 and CUDA 12.9 -- use cirun for compilation #149

conda-forge-admin · 2025-05-21T09:03:32Z

Note that only builds triggered by maintainers of the feedstock (and core)
who have accepted the terms of service and privacy policy will run
on Github actions via Cirun.

Maintainers have accepted the terms of service and privacy policy
at https://github.com/Quansight/open-gpu-server

Closes #150
Closes #151
Closes #152
Closes #158

…penstack-cpu-xlarge using Cirun

…nda-forge-pinning 2025.05.20.06.20.57

conda-forge-admin · 2025-05-21T09:05:02Z

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/meta.yaml) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe/meta.yaml:

ℹ️ The recipe is not parsable by parser conda-souschef (grayskull). This parser is not currently used by conda-forge, but may be in the future. We are collecting information to see which recipes are compatible with grayskull.
ℹ️ The recipe is not parsable by parser conda-recipe-manager. The recipe can only be automatically migrated to the new v1 format if it is parseable by conda-recipe-manager.

_{This message was generated by GitHub Actions workflow run https://github.com/conda-forge/conda-forge-webservices/actions/runs/19783971603. Examine the logs at this URL for more detail.}

cbourjau · 2025-07-09T08:54:37Z

I'm a bit lost on how to move this forward or which documentation to follow. Do all maintainers have to accept the ToS? Is there an obvious next step to take here, @hmaarrfk?

cbourjau · 2025-08-04T09:21:50Z

The 1.22.0 builds fail due to a hash-issue with some of the external dependencies pulled in by cmake. The issue is reportedly fixed in 1.22.1. I changed this PR to target that version and rerendered.

cbourjau · 2025-08-13T22:24:53Z

The Cirun fails with "No runners configuration found after labels filtering" (see here). @h-vetinari stated here that all maintainers need to agree to the ToS by going through the motions described here. This is not the case, yet. May this be the issue? The following list tracks which maintainers have agreed:

It would be great if those missing could create the appropriate PRs so that we can finally get this over the finish line ❤️ .

h-vetinari

@h-vetinari stated here that all maintainers need to agree to the ToS by going through the motions described here.

Not all listed maintainers need to accept, but anyone who wants to trigger CI. Things ran fine here already, but you've misconfigured things, and jobs with unknown labels fail.

This must be fixed, because as it stands, you're wasting an enormous amount of resources by doing each job three times

recipe/conda_build_config.yaml

conda-forge.yml

CUDA 12.8 added support for architectures `sm_100`, `sm_101` and `sm_120`, while CUDA 12.9 further added `sm_103` and `sm_121`. To build for these, maintainers will need to modify their existing list of specified architectures (e.g. `CMAKE_CUDA_ARCHITECTURES`, `TORCH_CUDA_ARCH_LIST`, etc.) for their package. A good balance between broad support and storage footprint (resp. compilation time) is to add `sm_100` and `sm_120`. Since CUDA 12.8, the conda-forge nvcc package now sets `CUDAARCHS` and `TORCH_CUDA_ARCH_LIST` in its activation script to a string containing all of the supported real architectures plus the virtual architecture of the latest. Recipes for packages who use these variables to control their build but do not want to build for all supported architectures will need to override these variables in their build script. ref: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#new-features

…5.08.31.03.07.54 Other tools: - conda-build 25.7.0 - rattler-build 0.46.0 - rattler-build-conda-compat 1.4.5

recipe/conda_build_config.yaml

cbourjau · 2025-09-26T12:03:49Z

@h-vetinari, do you see a way to unblock this PR? I'm a bit lost here...

leofang · 2025-11-04T06:08:24Z

It seems we are cross-compiling here? Any chance the wrong libcudadevrt is fed to the linker?

h-vetinari · 2025-11-04T06:25:48Z

It seems we are cross-compiling here? Any chance the wrong libcudadevrt is fed to the linker?

Hey Leo, thanks for the quick response! We're cross-compiling aarch+CUDA in many places (including on this feedstock with CUDA 12.6), so I'd be surprised if a more or less standard setup (that we've been using for years) would suddenly stop working. We'd be seeing this in a lot more places.

cbourjau · 2025-11-10T15:30:40Z

At least the xlarge runners are enough to get through the build without OOM'ing.

It seems like we might still be OOM'ing, does it not: https://github.com/conda-forge/onnxruntime-feedstock/actions/runs/19094250500/job/54550686105?pr=149#step:3:3757

h-vetinari · 2025-11-10T15:37:21Z

Indeed. But d14dff9 was only experimental to see if it would change anything about

nvlink fatal   : elfLink linker library load error (target: sm_50)

(it didn't...) and the commit before that ran fine on linux-64. We can however try 2xlarge, just send a PR for using cirun-openstack-cpu-2xlarge to https://github.com/conda-forge/admin-requests using the template

…5.11.14.11.30.28 Other tools: - conda-build 25.9.0 - rattler-build 0.49.0 - rattler-build-conda-compat 1.4.9

…5.11.15.23.15.53 Other tools: - conda-build 25.9.0 - rattler-build 0.49.0 - rattler-build-conda-compat 1.4.9

cbourjau · 2025-11-27T10:09:42Z

Should we consider skipping cuda builds on linux-aarch64 for now?

h-vetinari · 2025-11-27T21:13:41Z

Should we consider skipping cuda builds on linux-aarch64 for now?

Fine by me. I don't know what's wrong with

nvlink fatal   : elfLink linker library load error (target: sm_80)

I've tried dropping the older arches, but the same error remains even for stuff that's definitely still supported, and I don't know how to debug this

hmaarrfk · 2025-11-27T21:27:42Z

after you drop things, ping @dslarm, he has been quite interested in maintain arm.

Perhaps he can help troubleshoot things.

…5.11.29.08.28.23

cbourjau · 2025-12-01T09:13:37Z

I skipped the aarch64 cuda builds and the cuda+novec builds. On the other hand, I re-added the Windows-novec builds. I think this is finally ready to go 😍!

cbourjau

I'll merge by the end of today if there are no objections. Thanks to everybody involved already!

hmaarrfk · 2025-12-08T14:01:52Z

i updated the title a bit.

cbourjau · 2025-12-08T20:48:23Z

@dslarm I'm afraid we had to remove CUDA builds from linux-aarch64 to get this PR over the finish line. Are you interested in taking a look at this at some point?

automatic conda-forge administrator and others added 2 commits May 21, 2025 09:00

Enable cirun-openstack-cpu-medium, cirun-openstack-cpu-large, cirun-o…

b9ba48f

…penstack-cpu-xlarge using Cirun

MNT: Re-rendered with conda-build 25.4.2, conda-smithy 3.48.1, and co…

860120d

…nda-forge-pinning 2025.05.20.06.20.57

conda-forge-admin requested review from cbourjau, hmaarrfk, janjagusch, jtilly, traversaro and xhochy as code owners May 21, 2025 09:03

Add win_64

b3ff776

cbourjau mentioned this pull request Jul 9, 2025

Add cbourjau Quansight/open-gpu-server#58

Merged

2 tasks

hmaarrfk closed this Jul 9, 2025

hmaarrfk reopened this Jul 9, 2025

Target 1.22.1 and rerender

724363c

h-vetinari requested changes Aug 13, 2025

View reviewed changes

recipe/conda_build_config.yaml Outdated Show resolved Hide resolved

recipe/conda_build_config.yaml Outdated Show resolved Hide resolved

conda-forge.yml Outdated Show resolved Hide resolved

remove obsolete compiler pins

4a58215

h-vetinari changed the title ~~Update feedstock to use cirun-openstack-cpu-medium, cirun-openstack-cpu-large, cirun-openstack-cpu-xlarge with Cirun~~ Update feedstock to use cirun Aug 31, 2025

h-vetinari and others added 6 commits August 31, 2025 15:55

fix github_actions_labels

6f6c652

collect patches in folder

3852d46

delete obsolete skips

ca4dd4e

MNT: Re-rendered with conda-smithy 3.52.0 and conda-forge-pinning 202…

dce3af0

…5.08.31.03.07.54 Other tools: - conda-build 25.7.0 - rattler-build 0.46.0 - rattler-build-conda-compat 1.4.5

DEBUG: only python 3.12

df40b0e

h-vetinari force-pushed the cirun-1747818023 branch from 9ff191d to df40b0e Compare August 31, 2025 04:58

h-vetinari previously requested changes Aug 31, 2025

View reviewed changes

recipe/conda_build_config.yaml Show resolved Hide resolved

traversaro mentioned this pull request Sep 3, 2025

Add @traversaro to the list of allowed users Quansight/open-gpu-server#67

Merged

2 tasks

don't rely on CMAKE_CUDA_ARCHITECTURES=all-major

d14dff9

cbourjau mentioned this pull request Nov 10, 2025

Increase runner size for onnxruntime conda-forge/admin-requests#1746

Merged

h-vetinari added 3 commits November 15, 2025 00:13

delete cudnn migrator

ef933d5

increase size of linux runner again

ef9f731

MNT: Re-rendered with conda-smithy 3.53.0 and conda-forge-pinning 202…

5c36e75

…5.11.14.11.30.28 Other tools: - conda-build 25.9.0 - rattler-build 0.49.0 - rattler-build-conda-compat 1.4.9

This was referenced Nov 14, 2025

Smithy does not delete closed cudnn910 migration conda-forge/conda-smithy#2424

Closed

allow cpu-4xlarge for onnxruntime-feedstock conda-forge/.cirun#127

Merged

h-vetinari added 6 commits November 16, 2025 10:21

4xlarge

a4d6344

MNT: Re-rendered with conda-smithy 3.53.1 and conda-forge-pinning 202…

1e49732

…5.11.15.23.15.53 Other tools: - conda-build 25.9.0 - rattler-build 0.49.0 - rattler-build-conda-compat 1.4.9

drop sm_50 on aarch

da32832

increase timeout

9575e4a

thin out CUDA arches due to LTO hanging; remove sm_60 on aarch

1c53bec

revert to all-major on linux-64; drop sm_70 on linux-aarch64

8430fb6

cbourjau added 2 commits November 29, 2025 12:32

MNT: Re-rendered with conda-smithy 3.53.3 and conda-forge-pinning 202…

13101bd

…5.11.29.08.28.23

Revert changes to recipe/build.sh

2d0e9bc

cbourjau approved these changes Dec 8, 2025

View reviewed changes

hmaarrfk changed the title ~~Update feedstock to use cirun~~ Update to 1.22.2 and CUDA 12.9 -- use cirun for compilation Dec 8, 2025

cbourjau merged commit 4cee211 into conda-forge:main Dec 8, 2025
98 of 100 checks passed

This was referenced Dec 9, 2025

Try external runner for Windows + Cuda #139

Closed

Try Megabuilds #134

Closed

Uh oh!

Update to 1.22.2 and CUDA 12.9 -- use cirun for compilation #149

Update to 1.22.2 and CUDA 12.9 -- use cirun for compilation #149

Uh oh!

Conversation

conda-forge-admin commented May 21, 2025 • edited by h-vetinari Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

conda-forge-admin commented May 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cbourjau commented Jul 9, 2025

Uh oh!

cbourjau commented Aug 4, 2025

Uh oh!

cbourjau commented Aug 13, 2025

Uh oh!

h-vetinari left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cbourjau commented Sep 26, 2025

Uh oh!

leofang commented Nov 4, 2025

Uh oh!

h-vetinari commented Nov 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cbourjau commented Nov 10, 2025

Uh oh!

h-vetinari commented Nov 10, 2025

Uh oh!

cbourjau commented Nov 27, 2025

Uh oh!

h-vetinari commented Nov 27, 2025

Uh oh!

hmaarrfk commented Nov 27, 2025

Uh oh!

cbourjau commented Dec 1, 2025

Uh oh!

cbourjau left a comment

Choose a reason for hiding this comment

Uh oh!

hmaarrfk commented Dec 8, 2025

Uh oh!

Uh oh!

cbourjau commented Dec 8, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants

conda-forge-admin commented May 21, 2025 •

edited by h-vetinari

Loading

conda-forge-admin commented May 21, 2025 •

edited

Loading

h-vetinari commented Nov 4, 2025 •

edited

Loading