V6.2.0

[KrisThielemans]

No description provided.

[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.

[KrisThielemans]

• Some Linux cuda jobs with python 3.12 fail in initial Python import test with

  ImportError: libcudart.so.12: cannot open shared object file: No such file or directory
  

  Checking a few log files, it looks like these jobs installed libparallelproj-cpu, as opposed to *cuda. This leads to 2 questions:
  • Why do some cuda jobs not select the cuda version of libparallelproj. @gschramm any idea?
  • What is the STIR recipe missing to get libcudart if we wouldn't import libparallelproj? the run section does have

     - __cuda  # [cuda_compiler_version != "None"]
    

• Linux cuda jobs with python 3.11 or older (and hence install a root_base version that makes STIR switch to C++ 20) fail in CMake with

  CMake Error in src/recon_buildblock/CMakeLists.txt:
  Target "recon_buildblock" requires the language dialect "CUDA20" .  But the
  current compiler "NVIDIA" does not support this, or CMake does not know the
  flags to enable it.
  

  which is the same error as in https://stackoverflow.com/a/75038504 where the answer says you need CMake 3.25.2 for C++ 20 support and CUDA, but the build is already using CMake 3.30.2.
• Linux cuda jobs that do not have the above problems fail with

   Running OSMAPOSL CUDA/OSMAPOSL_test_sim_Parallelproj_CUDARDP.par 
   ./run_test_simulate_and_recon.sh: line 118:  4484 Segmentation fault 
  

  This is because there's no CUDA run-time on the azure infrastructure. We should therefore test for nvidia-smi here as well.
• osx jobs fail with

  ld: library not found for -latomic
  

  these use clang. This will be due to the work-around I introduced in STIR for CLang OpenMP linking errors, see https://gitlab.kitware.com/cmake/cmake/-/issues/26037
• Windows CUDA jobs fail with

  import stir
  File "D:\bld\stir_1724185476851\_test_env\lib\site-packages\stir.py", line 12, in <module>
    from _stir import *
  ImportError: DLL load failed while importing _stir: The specified module could not be found.
  

  which is likely the same problem as libcudart.so not found. The DLL checks above report

  WARNING (stir,Library/bin/extract_segments.exe): $RPATH/cudart64_12.dll not found in packages, sysroot(s) nor the missing_dso_whitelist.
  .. is this binary repackaging?
  

  However, I see this problem in at least one job that does install libparallel-proj-cuda, e.g. here

[gschramm]

Hi Kris,

1. No idea why the builds with cuda comp v12 want to install the cpu version of libarallelproj :/
2. Regarding libcudart, the recipe for libparallelproj is below. There are extra cuda-version requirements for host and run, but I don't remember why we added those (with the help of the conda-forge people). Maybe those will solve the problems.

    requirements:
      build:
        - {{ stdlib("c") }}
        - {{ compiler('c') }}
        - {{ compiler('cxx') }}  # [cuda_compiler_version != "None"]
        - {{ compiler('cuda') }}  # [cuda_compiler_version != "None"]
        - cmake >=3.23
        - ninja
        - python
      host:
        - libgomp      # [linux]
        - llvm-openmp  # [osx]
        - cuda-version {{ cuda_compiler_version }}  # [(cuda_compiler_version or "None") != "None"]
      run:
        - __cuda  # [cuda_compiler_version != "None"]
        - cuda-version {{ cuda_compiler_version }}  # [(cuda_compiler_version or "None") != "None"]

[KrisThielemans]

There are extra cuda-version requirements for host and run,

I did check that cuda-version is installed, so I think that won't help. but doing this check more carefully, I see in this job

• cuda_compiler_version: '12.0'
• install of 12.6
• install of 12.0
• 12.0 and cuda-cudart-static in the build section
• 12.6 (and no cuda libraries) in the host section
• cuda-version >=12.0,<13 and __cuda in the run section
• Building success but the test run installs cuda-version 12.6 but no cuda libraries

There's differences therefore between host and build in both version and installed libraries. As everything builds fine, I suspect that this is ok (and maybe we don't need cuda at all on the host). My feeling is that the problem is really just the last bullet. I suppose I could explicitly add cuda-cudart in the run section.... (I have no idea what __cuda adds).

[gschramm]

https://conda-forge.org/docs/maintainer/knowledge_base/#cuda-builds

Conda exposes the maximum CUDA version supported by the installed Nvidia drivers through a virtual package named __cuda. By default, conda will install the highest version available for the packages involved. To override this behaviour, you can define a CONDA_OVERRIDE_CUDA environment variable. More details in the Conda docs.

[KrisThielemans]

@gchramm, why do you have

      ignore_run_exports:
        - cudatoolkit  # [(cuda_compiler_version or "None").startswith("11")]

No ignore for 12?

[KrisThielemans]

More info on conda-forge/conda-forge.github.io#1963

[KrisThielemans]

and https://github.com/conda-forge/cuda-feedstock/blob/main/recipe/doc/recipe_guide.md. Seems that the cuda-version might be a good idea after all. There's far too much there for me to spend time on though. Sigh.

[gschramm]

Indeed challenging to keep up to date. That's why I usually ask for help on the conda-forge gitter (which usually works very well)

[KrisThielemans]

@conda-forge/core I hope you can help. The main change here is that STIR now has CUDA code itself, so we need to take that into account. Summarising above problems related to CUDA:

• For CUDA12 jobs, libcudart.so.12 (or DLL) is missing for the test, so the "import stir` test fails. Do we need an explicit dependency?
• The hope was that if CUDA is detected, libparallelproj would choose its CUDA version as well, but that doesn't seem to be guaranteed.

Obviously, your help with the CMake issue would also be appreciated, but I'm not expecting any.

[hmaarrfk]

See https://github.com/conda-forge/pytorch-cpu-feedstock/blob/main/recipe/meta.yaml for an example of how to build cuda recipes.

[KrisThielemans]

• Linux CUDA 12 jobs now work (older CUDA versoins fail due to unrelated CMake C++-20 problem).
• Windows CUDA 12 fail with cuda-driver-dev does not exist. Older CUDA version still fail with DLL not found. That's not a surprise as the previous fix was only for CUDA 12, but I have no idea what to do for CUDA 11. Anyone?
• OSX jobs still fail due to STIR CMake clang problem mentioned above

[hmaarrfk]

but I have no idea what to do for CUDA 11. Anyone?

its probably fine to drop in 2024 at the feedstock level. don't let it hold you back.

If somebody wants it back they can make a PR and bring back support.

[hmaarrfk]

I think this can be used as an example for windows maybe conda-forge/pytorch-cpu-feedstock#231

[conda-forge-webservices]

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found some lint.

Here's what I've got...

For recipe/meta.yaml:

• Selectors are suggested to take a <two spaces>#<one space>[<expression>] form. See lines [26, 28]

[KrisThielemans]

@conda-forge-admin please rerender

[conda-forge-webservices]

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found some lint.

Here's what I've got...

For recipe/meta.yaml:

• Selectors are suggested to take a <two spaces>#<one space>[<expression>] form. See lines [26, 28]
• The feedstock has no .ci_support files and thus will not build any packages.

[conda-forge-webservices]

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found some lint.

Here's what I've got...

For recipe/meta.yaml:

• The feedstock has no .ci_support files and thus will not build any packages.

[KrisThielemans]

@conda-forge-admin please rerender

[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/stir-feedstock/actions/runs/10522325040.

[KrisThielemans]

@conda-forge-admin please rerender

[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/stir-feedstock/actions/runs/10522647113.

[KrisThielemans]

Seems that I know skip everything and can't get rid of it. @casperdcl any ideas?

[KrisThielemans]

@conda-forge-admin, please lint

[conda-forge-webservices]

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found some lint.

Here's what I've got...

For recipe/meta.yaml:

• The feedstock has no .ci_support files and thus will not build any packages.

[gschramm]

@KrisThielemans : When I look at the "Files Changed" tab of this PR, it seems that indeed all the files .ci_support files have been deleted (not sure if that was done by any of the bots)

[KrisThielemans]

yup, the rerender. I'm a bit confused by the sequence here. I had some white-space issues in the recipe etc, but I suppose I just have the wrong skip clause (trying to do only cuda 12 on Windows, and skipping osx for now)

[h-vetinari]

@conda-forge-admin, please rerender

[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.

[KrisThielemans]

@conda-forge-admin, please rerender

[KrisThielemans]

Thanks @h-vetinari for fixing my (stupid) mistake with the "bare" skip and the suggestion. As always, great support from the core!

Hopefully we're there now.

[KrisThielemans]

@conda-forge-admin, please rerender