The CUDA version on Windows

[ancientflow]

I attempted to compile the CUDA-accelerated WarpX on the Windows system, using VS2022 as the IDE and CUDA version 12.4. However, numerous errors occurred during the compilation process. Among the 7 projects I selected, amrex, ablastr, buildInfoapp, openpmd, and zero_check were all compiled successfully, while lib and app failed. I would like to know if there is a solution to this issue, or if compiling GPU-accelerated WarpX on Windows is not feasible (I have already compiled the openmp version of WarpX on the same platform). Thanks.

[ax3l]

Hi @ancientflow!

We have no CI test running to install WarpX with CUDA on native Windows, but we have one in AMReX, which is 95% identical to WarpX install logic and uses MSVC from VS:
https://github.com/AMReX-Codes/amrex/blob/25.08/.github/workflows/windows.yml#L149-L194
Does that approach work for you?

Alternatively, there is a good documentation from Nvidia how to compile and run CUDA programs on Windows WSL:
https://docs.nvidia.com/cuda/wsl-user-guide/index.html
Have you tried that yet?

Our development team uses Linux and macOS and we do deploy to Linux/macOS/Windows, e.g., via conda-forge. But we do not add a Windows+CUDA build yet, due to low demand. But if you are interested in this, we could:

• find out what is needed to be done
• maybe add a CI runner
• maybe update our conda builds to add a CUDA variant.

[ancientflow]

Thank you for your reply. According to the link about AMReX mentioned above, I successfully compiled the CUDA version of AMReX using msvc. However, when compiling the lib project under the solution WarpX, I still encountered errors. I attempted to solve these problems, but failed. I noticed that the compiler reported all the errors coming from the file AMReX_MLCellLinOp.H, yet amrex was successfully compiled. However, since the development was carried out on Linux, I will try to compile it on that system.