Enable PowerPC

[jaimergp]

Checklist

• Used a 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-linter]

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.

[jaimergp]

@conda-forge-admin, please rerender

[simonbray]

Error: /bin/csh not found on your system! Install csh or tcsh and rerun configure.
I fixed this with the patch file included in the OSX PR.

[jaimergp]

I managed to reproduce the NetCDF error locally. Attaching log here for reference.

netcdf_config.log

[jaimergp]

Further local attempts have been frustrated by a problem during NetCDF compilation :/

config.log

[jaimergp]

The netcdf-fortran folder contained a separate log (attached). All the relevant errors point to the same problem:

configure:5326: /home/conda/build_artifacts/ambertools_1576779864067/_build_env/bin/powerpc64le-conda_cos7-linux-gnu-gfortran -c -fPIC -fopenmp -mcpu=power8 -mtune=power8 -mpower8-fusion -mpower8-vector -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O3 -pipe -isystem /home/conda/build_artifacts/ambertools_1576779864067/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_place/include -fdebug-prefix-map=/home/conda/build_artifacts/ambertools_1576779864067/work=/usr/local/src/conda/ambertools-19.10 -fdebug-prefix-map=/home/conda/build_artifacts/ambertools_1576779864067/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_place=/usr/local/src/conda-prefix -O3 -mtune=native -msse -mfpmath=sse -x f95 conftest.f90 >&5
f951: Error: unrecognized command line option '-msse'
f951: Error: unrecognized command line option '-mfpmath=sse'
configure:5326: $? = 1
configure: failed program was:
|       program main
|
|       end
configure:5335: result: unknown
configure:5338: error: Fortran could not compile .f90 files

It tries to use SSE in PowerPC, which is not available. It also overrides some flags already defined by the conda compilers (like mtune).

[simonbray]

Maybe you could avoid compiling netcdf by including it as a dependency and using the --with-netcdf flag in the configure script?

[jaimergp]

No ppc64le available for libnetcdf :( I'll have to submit a PR for that one too. Yay!

[jaimergp]

Oh wait, defaults has it. Let me try! Thanks!

[jaimergp]

Updates on my progress:

• I had to patch AmberTools/src/configure2 to pass ppc64le manually to fftw-3.3 configuration. Some sse flag is still around and I couldn't find it, so for now, -nosse is being used - it shouldn't have an effect on ppc64le anyway
• boost fails to build, resulting in two (unimportant?) programs not getting compiled

Ideally, we should patch AmberTools to use the external boost and fftw libraries (and maybe others?), but I am not familiar with their source and I don't even know where to begin.

[jaimergp]

@conda-forge-admin, please restart ci

[simonbray]

@conda-forge-admin, please restart ci

[jaimergp]

I think that Travis error is unrelated to us. It might have to do with the domain? (.org vs .com)?

[conda-forge-linter]

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

I tried to re-render for you but ran into some issues, please ping conda-forge/core for further assistance. You can also try re-rendering locally.

[simonbray]

Try merging this: jaimergp#1

[jaimergp]

Thanks!

[jaimergp]

The MacOS error seems to be linked to ld (heh, see what I did there?), as commented in conda-forge/superlu_dist-feedstock#4 (comment) (see error). I'll follow their suggestion and pin ld in MacOS for now.

[jaimergp]

I think some configuration must be being ignored by the CMake system. I managed to get my OSX VM running again so I'll debug locally now. Thanks for the patience!

[simonbray]

The OSX stuff can potentially be done in another PR if you prefer? You already did a lot more here than just PowerPC builds.

[jaimergp]

I'd be fine with that. Shall we merge into a different branch that uploads to a different label for now, or just straight to master?

[simonbray]

I think straight to master is fine. What alternative label would you propose, if we did it that way?

[jaimergp]

testing_ambertools, probably. We used testing in openmm, but turns out other people had the same idea (ie, matplotlib), so if you use -c conda-forge/label/testing you will pull beta matplotlib packages too (accidentally, I presume). I wouldn't like to release a broken packages (imports and -h messages work ok, but we have no clue about the tests).

[jaimergp]

We can also wait and build locally for Linux to test the package, with no branches or labels involved. This might be even easier.

[simonbray]

So far I just did a few tests locally. But I can see the advantage of testing a package that was built using the conda-forge infrastructure too. Either option is fine with me.

[jaimergp]

Did you use the Python script build-locally.py in Linux 64? That should be an exact reproduction of the CI environments thanks to the Docker images, so if you tested that, we should be good. If you used a different approach, maybe we should double check with the py script just in case.

[simonbray]

Oh, to be clear, I was referring to the original recipe I wrote. I did not test this PR (yet - I can help a bit if you like).

[jaimergp]

In the meantime I keep testing OSX. I managed to reproduce the same error locally, and found other issue where they have the same (unsolved?) problem: AnacondaRecipes/aggregate#112

I tried setting the sysroot manually, but it has no effect. We need cc1 (available through gcc, but this is not compiled for osx)... might try with the old version?

[jaimergp]

I ran some tests locally for Linux 64 and seems to be ok. They are not exhaustive, but main components (cpptraj, antechamber, tleap) look ok! I say we can merge to master after disabling OSX here. I'll re-render in the next commit.

[simonbray]

Thanks @jaimergp!

[jaimergp]

You are welcome!