Cuda CI windows

[Jimver]

I looked at the CI and it seems you use a bit outdated version of getting cuda in your cmakelists.txt

I now use find_package(CUDAToolkit) which was introduced in CMake 3.17, so I updated those to 3.17 as well, not sure if that impacts other stuff though. But the CI seems to run fine on my fork.

I also changed floor to floorf and ceil to ceilf, see: https://devtalk.blender.org/t/cuda-compile-error-windows-10/17886/2

[Haydnspass]

Hey @Jimver
really awesome that you are actually contributing! Thanks already for that.
Unfortunately, although the Job indicates green, CMAke fails in your job on windows with a broken CUDA compiler which is quite odd:

-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.2/bin/nvcc.exe - broken
-- Configuring incomplete, errors occurred!
See also "C:/Miniconda/envs/test/conda-bld/spline_1616761487310/work/python/build/temp.win-amd64-3.9/Release/CMakeFiles/CMakeOutput.log".
See also "C:/Miniconda/envs/test/conda-bld/spline_1616761487310/work/python/build/temp.win-amd64-3.9/Release/CMakeFiles/CMakeError.log".
Traceback (most recent call last):

[Haydnspass]

@Jimver It's quite strange, isn't it? I wonder if I made something weird in the CMakeLists or conda build that leads to this?
Any new insights?

[Jimver]

Yeah I debugged it for a bit and managed to get the CMakeError.log as an artifact. See: https://github.com/TuragaLab/SplinePSF/suites/2359120187/artifacts/49944202

First part of the error log file:

Looking for a CUDA compiler failed with the following output:
-- The CUDA compiler identification is unknown
-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.2/bin/nvcc.exe
-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.2/bin/nvcc.exe - broken
CMake Error at C:/Miniconda/envs/test/conda-bld/spline_1616863076905/_build_env/Library/share/cmake-3.17/Modules/CMakeTestCUDACompiler.cmake:46 (message):
  The CUDA compiler

    "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.2/bin/nvcc.exe"

  is not able to compile a simple test program.

  It fails with the following output:

    Change Dir: C:/Miniconda/envs/test/conda-bld/spline_1616863076905/work/python/build/temp.win-amd64-3.7/Release/CMakeFiles/CheckCUDA/CMakeFiles/CMakeTmp
    
    Run Build Command(s):C:/Miniconda/envs/test/conda-bld/spline_1616863076905/_build_env/Library/bin/ninja.exe cmTC_d1238 && [1/2] Building CUDA object CMakeFiles\cmTC_d1238.dir\main.cu.obj
    FAILED: CMakeFiles/cmTC_d1238.dir/main.cu.obj 
    cmd.exe /C "C:\PROGRA~1\NVIDIA~2\CUDA\v11.2\bin\nvcc.exe     -x cu -c main.cu -o CMakeFiles\cmTC_d1238.dir\main.cu.obj && C:\PROGRA~1\NVIDIA~2\CUDA\v11.2\bin\nvcc.exe     -x cu -M main.cu -MT CMakeFiles\cmTC_d1238.dir\main.cu.obj -o CMakeFiles\cmTC_d1238.dir\main.cu.obj.d"
    nvcc fatal   : Could not set up the environment for Microsoft Visual Studio using 'C:/Program Files (x86)/Microsoft Visual Studio/2019/Enterprise/VC/Tools/MSVC/14.28.29910/bin/HostX64/x64/../../../../../../../VC/Auxiliary/Build/vcvars64.bat'
    ninja: build stopped: subcommand failed.

So it seems to fail when calling vcvars64.bat which I belive is used to setup the dev environment on Windows.

Afterwards I set VSCMD_DEBUG to 3 which makes the vcvars64.bat print more information, but now the CMakeError.log is hundreds of lines so I haven't looked through it all yet. If you want to take a look it's in the artifact I linked above.

[Jimver]

Ok after a bit of digging I found to following at line 2386-2396 in CMakeError.log:

...
C:\Miniconda\envs\test\conda-bld\spline_1616863076905\work\python\build\temp.win-amd64-3.7\Release\CMakeFiles\3.17.0\CompilerIdCUDA>call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\vsdevcmd\ext\roslyn.bat" 

C:\Miniconda\envs\test\conda-bld\spline_1616863076905\work\python\build\temp.win-amd64-3.7\Release\CMakeFiles\3.17.0\CompilerIdCUDA>set __VSCMD_script_err_count=0 

C:\Miniconda\envs\test\conda-bld\spline_1616863076905\work\python\build\temp.win-amd64-3.7\Release\CMakeFiles\3.17.0\CompilerIdCUDA>if "" NEQ "" goto :test 

C:\Miniconda\envs\test\conda-bld\spline_1616863076905\work\python\build\temp.win-amd64-3.7\Release\CMakeFiles\3.17.0\CompilerIdCUDA>if "" NEQ "" goto :clean_env 
The input line is too long.
The syntax of the command is incorrect.
C:\Miniconda\envs\test\conda-bld\spline_1616863076905\work\python\build\temp.win-amd64-3.7\Release\CMakeFiles\3.17.0\CompilerIdCUDA>set "PATH=C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\MSBuild\Current\bin\Roslyn;at"#$ C:/Program Files (x86)/Microsoft Visual Studio/2019/Enterprise/VC/Tools/MSVC/14.28.29910/bin/HostX64/x64/../../../../../../../VC/Auxiliary/Build/vcvars64.bat
nvcc fatal   : Could not set up the environment for Microsoft Visual Studio using 'C:/Program Files (x86)/Microsoft Visual Studio/2019/Enterprise/VC/Tools/MSVC/14.28.29910/bin/HostX64/x64/../../../../../../../VC/Auxiliary/Build/vcvars64.bat'
...

So it seems to call roslyn.bat. I found it on my local machine at C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\Tools\vsdevcmd\ext\roslyn.bat and the first lines are:

set __VSCMD_script_err_count=0
if "%VSCMD_TEST%" NEQ "" goto :test
if "%VSCMD_ARG_CLEAN_ENV%" NEQ "" goto :clean_env

@REM ------------------------------------------------------------------------
:start

set "PATH=%VSINSTALLDIR%MSBuild\Current\bin\Roslyn;%PATH%"

goto :end
...

So my suspicion is that it fails at the following line:

set "PATH=%VSINSTALLDIR%MSBuild\Current\bin\Roslyn;%PATH%"

As in the error log it says this:

The input line is too long.
The syntax of the command is incorrect.
C:\Miniconda\envs\test\conda-bld\spline_1616863076905\work\python\build\temp.win-amd64-3.7\Release\CMakeFiles\3.17.0\CompilerIdCUDA>set "PATH=C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\MSBuild\Current\bin\Roslyn;at"#$ C:/Program Files (x86)/Microsoft Visual Studio/2019/Enterprise/VC/Tools/MSVC/14.28.29910/bin/HostX64/x64/../../../../../../../VC/Auxiliary/Build/vcvars64.bat

So maybe the PATH is becoming too large? Not sure if that is tunable.

[Haydnspass]

A note for myself: https://stackoverflow.com/questions/43184251/cmake-command-line-too-long-windows