Rebuild for windows_cuda

[regro-cf-autotick-bot]

This PR has been triggered in an effort to update windows_cuda.

Notes and instructions for merging this PR:

1. Please merge the PR only after the tests have passed.
2. Feel free to push to the bot's branch to update this PR if needed.

Please note that if you close this PR we presume that the feedstock has been rebuilt, so if you are going to perform the rebuild yourself don't close this PR until the your rebuild has been merged.

This package has the following downstream children:

And potentially more.

If this PR was opened in error or needs to be updated please add the bot-rerun label to this PR. The bot will close this PR and schedule another one. If you do not have permissions to add this label, you can use the phrase @conda-forge-admin, please rerun bot in a PR comment to have the conda-forge-admin add it for you.

_{This PR was created by the regro-cf-autotick-bot.
The regro-cf-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. If you would like a local version of this bot, you might consider using rever. Rever is a tool for automating software releases and forms the backbone of the bot's conda-forge PRing capability. Rever is both conda (conda install -c conda-forge rever) and pip (pip install re-ver) installable.
Finally, feel free to drop us a line if there are any issues!
This PR was generated by https://github.com/regro/autotick-bot/actions/runs/379456304, please use this URL for debugging}

[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.

[h-vetinari]

Requires windows builds to make any sense, therefore blocked on #17.

[h-vetinari]

@jaimergp I saw you added the windows packaging for nvcc: conda-forge/nvcc-feedstock#49

Not sure if I need to change something else, but it's not working out of the box here... Even though I see

"Exporting and adding $CUDA_PATH ('C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0') to $PATH"

in the conda-forge build setup, this variable is then apparently empty by the time we get around to activate.bat:

(D:\bld\faiss-split_1607560645796\_h_env) (base) D:\bld\faiss-split_1607560645796\work>call "C:\Miniconda\Scripts\..\condabin\conda.bat" activate --stack "D:\bld\faiss-split_1607560645796\_build_env" 

(base) D:\bld\faiss-split_1607560645796\work>if defined CUDA_HOME (set "CUDA_HOME_CONDA_NVCC_BACKUP=" ) 

(base) D:\bld\faiss-split_1607560645796\work>if defined CUDA_PATH (
set "CUDA_PATH_CONDA_NVCC_BACKUP="  
 if not defined CUDA_HOME (set "CUDA_HOME=" ) 
) 

(base) D:\bld\faiss-split_1607560645796\work>if defined CFLAGS (set "CFLAGS_CONDA_NVCC_BACKUP=" ) 

(base) D:\bld\faiss-split_1607560645796\work>if defined CPPFLAGS (set "CPPFLAGS_CONDA_NVCC_BACKUP=" ) 

(base) D:\bld\faiss-split_1607560645796\work>if defined CXXFLAGS (set "CXXFLAGS_CONDA_NVCC_BACKUP=" ) 

(base) D:\bld\faiss-split_1607560645796\work>if not defined CUDA_PATH (
for /F "usebackq tokens=*" %a in (`where nvcc.exe`) do set "CUDA_NVCC_EXECUTABLE=%a"   || goto :error  
 if "" == "" (
echo "Cannot determine CUDA_PATH: nvcc.exe not in PATH"  
 exit /b 1 
)  else (for /F "usebackq tokens=*" %a in (`python -c "from pathlib import Path; print(Path('').parents[1])"`) do set "CUDA_PATH=%a"   || goto :error ) 
) 
INFO: Could not find files for the given pattern(s).
"Cannot determine CUDA_PATH: nvcc.exe not in PATH"

[jaimergp]

@h-vetinari Yes, I saw this happen in other recipe too. Tracking here: conda-forge/nvcc-feedstock#53. Please add a comment there pointing to this PR so I have more data points. I should have some free time next weeks and will look into this.

I have tried a different variable export mechanism in this PR conda-forge/conda-forge-ci-setup-feedstock#131, using the activation scripts of conda-forge-ci-setup, but it didn't make a difference for prismatic.

I am not sure why this doesn't work here either, but a thing to notice is that both this repo and prismatic are multi-output recipes. It does work in single output recipes like OpenMM, though 🤷 I wonder if this is an issue at the conda build level, where some env variables do not survive long enough to reach here.

The workaround for now is, unfortunately, setting %CUDA_PATH% manually to:

set "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%cuda_compiler_version%"

[h-vetinari]

The workaround for now is, unfortunately, setting %CUDA_PATH% manually to:

set "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%cuda_compiler_version%"

@jaimergp Thanks for the info & workaround, but I have no idea how to provide it. The builds-scripts are too late (at least for the nvcc activation script), and setting it in azure-pipelines-win.yml didn't work either...

[jaimergp]

Since nvcc activation is failing in this recipe, we might need to bypass it for now.

What's the manual way of telling CMake where to find CUDA in this project? I normally need to set up CUDA_TOOLKIT_ROOT_DIR (with forward slashes): https://github.com/conda-forge/openmm-feedstock/blob/master/recipe/bld.bat#L10

You'll also need these lines: https://github.com/conda-forge/nvcc-feedstock/blob/master/recipe/windows/activate.bat#L59-L61

and make nvcc use -ccbin: https://github.com/conda-forge/nvcc-feedstock/blob/master/recipe/windows/nvcc_windows.bat

These are a lot of "workarounds" needed so you can also choose to wait a bit until I debug the multi-output problem with the activation script (hopefully before the year ends!).

[h-vetinari]

These are a lot of "workarounds" needed so you can also choose to wait a bit until I debug the multi-output problem with the activation script (hopefully before the year ends!).

Thanks for the quick reply! I don't mind adding work-arounds (if we're talking about a handful of lines) and removing them once fixed. I actually just realized that I had not successfully committed the changes in azure-pipelines-win.yml (due to doing git add . in /recipe 🤦‍♂️), but anyway, that would have not survived a re-render, so I prefer your approach.

I'm hoping that cmake is smart enough to add the -ccbin, at least it is on linux.

[jaimergp]

I have included a fallback mechanism that will try to look for CUDA in the default location, in conda-forge/nvcc-feedstock#52

In the meantime, let's try adding these lines first thing in your bld.bat:

set "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%cuda_compiler_version%"
set "ORIG_PKG_VERSION=%PKG_VERSION%"
set "PKG_VERSION=%cuda_compiler_version%"
call %BUILD_PREFIX%\etc\conda\activate.d\nvcc_activate.bat
set "PKG_VERSION=%ORIG_PKG_VERSION%"
set "ORIG_PKG_VERSION="

Maybe it's PREFIX instead of BUILD_PREFIX but you'll see :)

[jaimergp]

A new nvcc wrapper was released so you might be lucky now. Otherwise... well, I'll try to help you next week by debugging on my Windows machine. Sorry for all the trouble!

[h-vetinari]

A new nvcc wrapper was released so you might be lucky now. Otherwise... well, I'll try to help you next week by debugging on my Windows machine. Sorry for all the trouble!

To the contrary, thank you for your help on this! :)

[h-vetinari]

@jaimergp
Sounds like we made a step forward, but that some quotes are missing around the include-path, and it gets split by spaces into things the compiler doesn't know anything about:

NVIDIA
      cl /c /IC:\Program /Zi /W1 /WX- /diagnostics:classic /Od /Ob0 /D WIN32 /D _WINDOWS /D "CMAKE_INTDIR=\"Debug\"" /D _MBCS /Gm- /EHsc /RTC1 /MDd /GS /fp:precise /Qspectre /Zc:wchar_t /Zc:forScope /Zc:inline /GR /Fo"cmTC_b6410.dir\Debug\\" /Fd"cmTC_b6410.dir\Debug\vc141.pdb" /Gd /TP /errorReport:queue  Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\include "D:\bld\faiss-split_1607680204242\work\_build\CMakeFiles\CMakeTmp\testCXXCompiler.cxx"
    c1xx : fatal error C1083: Cannot open source file: 'Files\NVIDIA': No such file or directory [D:\bld\faiss-split_1607680204242\work\_build\CMakeFiles\CMakeTmp\cmTC_b6410.vcxproj]
      GPU
      
    c1xx : fatal error C1083: Cannot open source file: 'GPU': No such file or directory [D:\bld\faiss-split_1607680204242\work\_build\CMakeFiles\CMakeTmp\cmTC_b6410.vcxproj]
      Computing
    c1xx : fatal error C1083: Cannot open source file: 'Computing': No such file or directory [D:\bld\faiss-split_1607680204242\work\_build\CMakeFiles\CMakeTmp\cmTC_b6410.vcxproj]
      include
    c1xx : fatal error C1083: Cannot open source file: 'Toolkit\CUDA\v10.0\include': No such file or directory [D:\bld\faiss-split_1607680204242\work\_build\CMakeFiles\CMakeTmp\cmTC_b6410.vcxproj]
      testCXXCompiler.cxx

[jaimergp]

Maybe we need some kind of quotes here, but this should have been caught by the tests 🤔

I'll need to open a new PR for this. If you want you can try to define CFLAGS and friends manually, using the example output in cuda_compiler_versionNone, and see where that gets you.

[h-vetinari]

Seems I started this too closely on the heels of conda-forge/nvcc-feedstock#57 being merged - the builds here still picked up the old build nvcc (9 instead of 10) - let's try again...

[h-vetinari]

It appears that this traces back to cudaGetDeviceCount from cuda_profiler_api.h, which seems to segfault on windows (but not linux) when there are no actual GPUs. @kkraus14, would you be able to comment?

[jaimergp]

Take into account that the CUDA installation on Windows lacks the drivers, since the Nvidia installer refuses to install them due to the missing GPU.

[jaimergp]

Check how we handle this at OpenMM. We can create packages with the full test suite (which does not run on the CI) and then run them locally.

[h-vetinari]

@jaimergp
This just built successfully for the first time. 🥳

Thanks so much for your help debugging this 🙏 🙃

[github-actions]

Hi! This is the friendly conda-forge automerge bot!

I considered the following status checks when analyzing this PR:

• linter: passed
• azure: passed

Thus the PR was passing and merged! Have a great day!

[jaimergp]

Wow, this makes me so glad! Congratulations, this was quite the journey! 🎉

[h-vetinari]

Thanks for the kind words!

Journey isn't over yet, unfortunately - ran the gpu-specific test suite locally and it has a bunch of segfaults 😒
Opened a PR to mark them broken, though I really should have tested the builds from artefacts of the PR directly... 🤦‍♂️

[jaimergp]

Soon...