Add MVS 2022 Support

[AnderBiguri]

related #347

[tsadakane]

Hi,

CUDA 11.6 released. It supports VS2022.

• https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
  • The cudaDeviceSynchronize() function used for on-device fork/join parallelism is deprecated...
  • CUDA 11.6 officially supports the latest VS2022 as host compiler.

I tried...

As for python, the package was build successfully and example.py seemed to work.

As for Matlab, >> mex -setup -v could not find VS2022; 2021b does not support VS2022 yet.

• https://jp.mathworks.com/support/requirements/supported-compilers.html

FYI

[AnderBiguri]

Fantastik @tsadakane . This would need indeed a change in MATLAB. Even if mex seems not to support it, a correct xml file may make it work.

[tsadakane]

Modifying the XML file to specify 2022 as the c++ compiler used by nvcc was straight forward. ([15.0, 16.0) => [16.0, 17.0) ).

But, there is another issue. It seems that mex searches for cl (vc compiler) if the file list only includes .cpp files. If vs2019 is installed too, no problem. but if only vs2022 was installed, it fails. Adding a dummy .cu file solves this problem:

        mex -largeArrayDims ./Utilities/IO/VarianCBCT/mexReadXim.cpp ../Common/CUDA/gpuUtils.cu -outdir ./Mex_files/win64

Of course, gpuUtils.cu is unnecessary for mexReadXim.cpp. Ugly solution.

What do you think?

[AnderBiguri]

I'd say go for it. Its ugly, but so is the way nvcc and mex search for compilers, so I'd say if it works, its fine!

[tsadakane]

@AnderBiguri , I think you can close this issue.

Before being closed, I report that I could run d03_generateData.m successfully under the following environment:

• Windows 10 64bit
• VS2022
• CUDA 11.6
• GeForce GTX 1060 6GB
• MATLAB R2022a

[AnderBiguri]

Fantastic! Good to know!