Add cmake files in onnxruntime-cpp

[diegoferigo-rai]

Checklist

• Used a personal fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

[conda-forge-admin]

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/22118185634. Examine the logs at this URL for more detail.}

[traversaro]

@diegoferigo-rai if you are still iterating on this and you are not running the build locally, can you temporary skip most of the builds leaving just a combination? Only macOS builds are running now, but macOS runners are a bit scarce, so it would be ideal to run the full macOS jobs matrix only if necessary.

[diegoferigo-rai]

can you temporary skip most of the builds leaving just a combination?

I wasn't sure how to configure the workflow to skip the non-Linux jobs temporarily. Did you mean I should modify the CI configuration file directly, or is there another method?

In any case, I've finished iterating on the PR and the changes should be good to go.

[traversaro]

Ok! The next step is to trigger the build on the cirun server, but I think it is a bit congested at the moment due to PyTorch 2.10 + CUDA 13.0 + Python 3.14, I would wait the go from @h-vetinari for doing that.

[h-vetinari]

Indeed the server is pretty congested right now -- no resources available until pytorch and tensorflow finish.

Actually, we also need to backport the CMake fix for pytorch to v2.9 (and there's other stuff piling up), so unless this PR addresses a super-urgent issue (haven't looked closely), I'd ask to wait for 2-3 days or so. I'll try to start CI here if there's some capacity available in the meantime (given that the builds here don't take very long).

[diegoferigo-rai]

Thank you all for the support. This isn't urgent, so it can wait a couple more days. I'm also looking forward to having CUDA 13 support here!

[traversaro]

Just to clarify in case anyone reads here, the fallback until CMake config files are actually installed is to use Findonnxruntime.cmake file, like https://github.com/gbionics/bipedal-locomotion-framework/blob/592d1cdfe07e40963a2c501423bbd1ccfa899e61/cmake/Findonnxruntime.cmake#L10 .

[diegoferigo-rai]

@h-vetinari any update on this PR? Mind triggering CI again?

[diegoferigo-rai]

Thanks @h-vetinari, hopefully fixed in 6c897e4. I missed a step on windows.

[diegoferigo-rai]

It seems that there are problems downloading this zip file on windows. The URL looks valid though.

[1/9] Performing download step (download, verify and extract) for 'dlpack-populate'
-- Downloading...
   dst='C:/bld/onnxruntime_1770368734649/work/build-ci/Release/_deps/dlpack-subbuild/dlpack-populate-prefix/src/5c210da409e7f1e51ddf445134a4376fdbd70d7d.zip'
   timeout='none'
   inactivity timeout='none'
-- Using src='https://github.com/dmlc/dlpack/archive/5c210da409e7f1e51ddf445134a4376fdbd70d7d.zip'
CMake Error at dlpack-subbuild/dlpack-populate-prefix/src/dlpack-populate-stamp/download-dlpack-populate.cmake:163 (message):
  Each download failed!

    error: downloading 'https://github.com/dmlc/dlpack/archive/5c210da409e7f1e51ddf445134a4376fdbd70d7d.zip' failed
          status_code: 22
          status_string: "HTTP response code said error"
          log:
          --- LOG BEGIN ---
          Host github.com:443 was resolved.

  IPv6: (none)

  IPv4: 20.26.156.215

    Trying 20.26.156.215:443...

  schannel: disabled automatic use of client certificate

  ALPN: curl offers http/1.1

  schannel: added 136 certificate(s) from CA file
  'C:\bld\onnxruntime_1770368734649\_h_env\Library\ssl\cacert.pem'

  schannel: connection hostname (github.com) validated against certificate
  name (github.com)

  ALPN: server accepted http/1.1

  Established connection to github.com (20.26.156.215 port 443) from 10.0.0.4
  port 49208

  using HTTP/1.x

  GET /dmlc/dlpack/archive/5c210da409e7f1e51ddf445134a4376fdbd70d7d.zip
  HTTP/1.1

  Host: github.com

  User-Agent: curl/8.18.0

  Accept: */*

  

  Request completely sent off

  schannel: remote party requests renegotiation

  schannel: renegotiating SSL/TLS connection

  schannel: reusing certificate store from cache

  schannel: connection hostname (github.com) validated against certificate
  name (github.com)

  schannel: SSL/TLS connection renegotiated

  schannel: remote party requests renegotiation

  schannel: renegotiating SSL/TLS connection

  schannel: reusing certificate store from cache

  schannel: connection hostname (github.com) validated against certificate
  name (github.com)

  schannel: SSL/TLS connection renegotiated

  HTTP/1.1 503 Service Unavailable

  Content-Length: 190

  Content-Type: text/plain

  Date: Fri, 06 Feb 2026 09:12:03 GMT

  Server: github.com

  X-Frame-Options: DENY

  Strict-Transport-Security: max-age=31536000; includeSubDomains; preload

  X-GitHub-Request-Id: 0408:66BA2:45ED1D:56259A:6985B064

  The requested URL returned error: 503

  

  closing connection #0



          --- LOG END ---

[diegoferigo]

Apologies for the ping, @h-vetinari. I'm really hoping to have the exported targets in v1.24 (PR #169). After checking the logs, it seems that the download failure occurred during the initial bat script configuration, before the lines I added were executed. Could you please trigger a re-run? Thanks!

[h-vetinari]

I restarted one of the failed jobs now. :)

[diegoferigo-rai]

Now the build fails later on Windows, when it runs cmake --install:

(%BUILD_PREFIX%) %SRC_DIR%>cmake --install build-ci --config Release --prefix "install-ci" 
CMake Error: Not a file: C:/bld/onnxruntime_1770676394245/work/build-ci/cmake_install.cmake
CMake Error: Error processing file: C:/bld/onnxruntime_1770676394245/work/build-ci/cmake_install.cmake

I do not have a Windows machine, and manually iterating on CI here is quite painful. If no one is available to take a look at the Windows side, I will update this PR to target only linux and macos for now. Does that sound reasonable to you @h-vetinari? Windows support could be implemented in another PR.

[h-vetinari]

manually iterating on CI here is quite painful. If no one is available to take a look at the Windows side, I will update this PR to target only linux and macos for now. Does that sound reasonable to you @h-vetinari?

No, that does not sound acceptable. I'll switch this back to azure for you so you can debug on your merry way, but we don't simply drop a platform because we can't get something to work (especially because this looks like a pretty trivial problem about the naming of folders and/or targets). I agree that debugging in CI is not fun, but that's the way it is, unless you want to organise yourself a suitable machine or VM.

[diegoferigo-rai]

Thanks for switching the CI, that will make iterating much easier!

I'm happy to look into the Windows side, though my initial suggestion was based on the idea that having a feature available for a subset of platforms is often better for the community than having none at all. I worry that requiring contributors to set up complex local environments (like VMs) can create a high barrier to entry and discourage future contributions, which I'm sure is something we'd all like to avoid.

That said, I appreciate your support, and I'm more than happy to finalize it for all platforms now that the overhead is lower for everyone and I can iterate my way on CI.

[diegoferigo-rai]

CI is now passing also on windows! @h-vetinari let me know if there is anything else I can help with.

[diegoferigo]

Friendly ping @h-vetinari. This is ready to go, perhaps @cbourjau or another maintainer could help with the final steps to get this merged?

[diegoferigo]

@conda-forge-admin, please rerender

[h-vetinari]

It looks like #169 is further along. I suggest this PR be folded into some of the follow-up changes (that is, unless #169 gets modified again, then it could be folded into that PR).

[cbourjau]

Closed via #169

[diegoferigo-rai]

Thanks @cbourjau for your contribution in #169! I'm waiting for the CUDA versions of the linux-aarch64 packages to land on the conda-forge channel before testing them on an AGX Thor. If you don't hear from me soon, that means everything worked perfectly :)

[cbourjau]

I'm waiting for the CUDA versions of the linux-aarch64 packages to land on the conda-forge channel before testing them on an AGX Thor. If you don't hear from me soon, that means everything worked perfectly :)

I'm afraid that will be a longer wait 😬 ! We currently don't build CUDA versions on linux-aarch64:

onnxruntime-feedstock/recipe/meta.yaml

Line 40 in 3c0038c

skip: true # [(cuda_compiler_version != "None") and aarch64]

[diegoferigo-rai]

Ah, got it, thanks for the info! I missed that skip, which explains why there are no packages. What's the limitation at the moment?

[cbourjau]

We ran into build errors in the recent past, but I'm afraid I forgot to retry it on the latest release. Would you like to give it a spin? If it works, we can include it in 1.24.3, which seems to be imminent upstream.

[diegoferigo-rai]

Worth checking #172 🤞