Use rdma-core package (instead of CDT) & add to linux_aarch64

[jakirkham]

Part of issue: conda-forge/cuda-feedstock#28

Previously we used the CDT for rdma-core. However it would be better to use the package. Especially so as this CDT is not provided on AlmaLinux 8. This makes that change. It is also aligned with how we use rdma-core elsewhere.

Additionally it looks like rdma-core would be needed on linux_aarch64, but we didn't have it there before. So this makes sure to add it.

---

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-webservices]

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.

[jakirkham]

Not seeing Azure here. Going to try restarting

@conda-forge-admin , please restart CI

[jakirkham]

Looks like that fixed it

[jakirkham]

Am seeing the following on CI

sed: can't read /home/conda/feedstock_root/build_artifacts/libcufile_1710268372716/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_p/lib/pkgconfig/cufile*.pc: No such file or directory

Based on looking locally, this appears to be happening because ${PREFIX}/lib/pkgconfig already exists. So the line below moves pkg-config into ${PREFIX}/lib/pkgconfig, which gives us ${PREFIX}/lib/pkgconfig/pkg-config. Where instead we would want all the .pc files in ${PREFIX}/lib/pkgconfig

libcufile-feedstock/recipe/build.sh

Line 10 in 7f8e1ec

[[ -d pkg-config ]] && mv pkg-config ${PREFIX}/lib/pkgconfig

IOW this looks like a bug in build.sh

[jakirkham]

@conda-forge-admin , please re-render

[github-actions]

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

I tried to rerender for you, but it looks like I wasn't able to push to the drop_cdt branch of jakirkham-feedstocks/libcufile-feedstock. Did you check the "Allow edits from maintainers" box?

NOTE: Our webservices cannot push to PRs from organization accounts or PRs from forks made from organization forks because of GitHub permissions. Please fork the feedstock directly from conda-forge into your personal GitHub account.

This message was generated by GitHub actions workflow run https://github.com/conda-forge/libcufile-feedstock/actions/runs/8626284824.

[jakirkham]

@conda-forge-admin , please re-render

[jakirkham]

@conda-forge-admin , please lint

[conda-forge-webservices]

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.

[jakirkham]

@conda-forge-admin , please re-render

[github-actions]

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/libcufile-feedstock/actions/runs/8626433350.

[adibbley]

I'm not sure the rdma-core package is providing what we needed from the cdt package originally.

One example from the 12.4.1 build log I see:

INFO (libcufile,targets/x86_64-linux/lib/libcufile_rdma.so.1.9.1): Needed DSO x86_64-conda-linux-gnu/sysroot/usr/lib64/libmlx5.so.1 found in CDT/compiler package conda-forge/noarch::libibverbs-cos7-x86_64==22.4=h9b0a68f_1105

And here I see:

WARNING (libcufile,targets/x86_64-linux/lib/libcufile_rdma.so.1.9.1): $RPATH/libmlx5.so.1 not found in packages, sysroot(s) nor the missing_dso_whitelist.

[jakirkham]

@conda-forge-admin , please re-render

[github-actions]

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

I tried to rerender for you, but it looks like I wasn't able to push to the drop_cdt branch of jakirkham-feedstocks/libcufile-feedstock. Did you check the "Allow edits from maintainers" box?

NOTE: Our webservices cannot push to PRs from organization accounts or PRs from forks made from organization forks because of GitHub permissions. Please fork the feedstock directly from conda-forge into your personal GitHub account.

This message was generated by GitHub actions workflow run https://github.com/conda-forge/libcufile-feedstock/actions/runs/9654546805.

[leofang]

It seems the rerendering failed 😅

[conda-forge-webservices]

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

I was trying to look for recipes to lint for you, but it appears we have a merge conflict.
Please try to merge or rebase with the base branch to resolve this conflict.

Please ping the 'conda-forge/core' team (using the @ notation in a comment) if you believe this is a bug.

[conda-forge-webservices]

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.

[jakirkham]

Not sure why it wouldn't find the libraries, they are all there

rdma-core-52.0-hcccb83c_0/lib
├── libefa.so -> libefa.so.1.3.52.0
├── libefa.so.1 -> libefa.so.1.3.52.0
├── libefa.so.1.3.52.0
├── libhns.so -> libhns.so.1.0.52.0
├── libhns.so.1 -> libhns.so.1.0.52.0
├── libhns.so.1.0.52.0
├── libibmad.so -> libibmad.so.5.3.52.0
├── libibmad.so.5 -> libibmad.so.5.3.52.0
├── libibmad.so.5.3.52.0
├── libibnetdisc.so -> libibnetdisc.so.5.1.52.0
├── libibnetdisc.so.5 -> libibnetdisc.so.5.1.52.0
├── libibnetdisc.so.5.1.52.0
├── libibumad.so -> libibumad.so.3.2.52.0
├── libibumad.so.3 -> libibumad.so.3.2.52.0
├── libibumad.so.3.2.52.0
├── libibverbs.so -> libibverbs.so.1.14.52.0
├── libibverbs.so.1 -> libibverbs.so.1.14.52.0
├── libibverbs.so.1.14.52.0
├── libmana.so -> libmana.so.1.0.52.0
├── libmana.so.1 -> libmana.so.1.0.52.0
├── libmana.so.1.0.52.0
├── libmlx4.so -> libmlx4.so.1.0.52.0
├── libmlx4.so.1 -> libmlx4.so.1.0.52.0
├── libmlx4.so.1.0.52.0
├── libmlx5.so -> libmlx5.so.1.24.52.0
├── libmlx5.so.1 -> libmlx5.so.1.24.52.0
├── libmlx5.so.1.24.52.0
├── librdmacm.so -> librdmacm.so.1.3.52.0
├── librdmacm.so.1 -> librdmacm.so.1.3.52.0
└── librdmacm.so.1.3.52.0

[leofang]

I seem to recall we encountered a similar situation elsewhere (but can't find the example now), where a CTK component in $PREFIX/targets/<arch>/lib depends on non-CTK packages in $PREFIX/lib. Maybe we need to patch rpath manually for the cuFILE libraries?

[conda-forge-webservices]

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.

[jakirkham]

Looks like we need libnuma based on this CI warning:

WARNING (gds-tools,gds/tools/gdsio): $RPATH/libnuma.so.1 not found in packages, sysroot(s) nor the missing_dso_whitelist.

However that is already here

libcufile-feedstock/recipe/meta.yaml

Line 161 in 63ee783

- libnuma # [linux]

Or an RPATH fix?

[jakirkham]

Yeah it looks like there is something up with the rpath for gds-tools based on this line from CI:

WARNING (gds-tools,gds/tools/gdsio): $RPATH/libcufile.so.0 not found in packages, sysroot(s) nor the missing_dso_whitelist.

As there is already libcufile present in the dependencies

libcufile-feedstock/recipe/meta.yaml

Line 164 in 6b64cd2

- {{ pin_subpackage("libcufile", max_pin="x") }}

[jakirkham]

Yep, this is related to issue: conda-forge/cuda-feedstock#10

[jakirkham]

@conda-forge-admin , please restart CI

[billysuh7]

@conda-forge-admin, please restart ci

[billysuh7]

Sorry I forgot @h-vetinari to approve this. LGTM.

[jakirkham]

Thanks Billy! 🙏

Were you meaning PR: conda-forge/cuda-nvcc-impl-feedstock#25 ?

[billysuh7]

Thanks Billy! 🙏

Were you meaning PR: conda-forge/cuda-nvcc-impl-feedstock#25 ?

That is all set also, I approved it and put more comment. Thanks.

[jakirkham]

Thanks Billy! 🙏

[jakirkham]

The conclusion Billy and I reached after looking into the warnings above more closely is there is likely a remaining RPATH issue in this package, which we will need to look into. This will be follow up work

xref: conda-forge/cuda-feedstock#10 (comment)