Avoid building cuda files every build

[JAicewizard]

Although I am not really sure how to fix this, it would be nice if the CUDA files could not be regenerated on every build.

Currently, the cuda files are regenerated every build, and thus also recompiled every build. This compilation can take quite a while, especially in CI.

It would be nice if there were a way to avoid regenerating this every time, so that:
1: cmake is able to understand it doesn't need to recompile these files
2: These files can be cached in CI (and then the benefit of 1 would also work in CI).

[mnorris11]

@tarang-jain @cjnolet is this an nvidia-specific option in CMake we should be enabling?

[tarang-jain]

There are CMake flags for enabling / disabling both -- vanilla GPU and cuVS

-DFAISS_ENABLE_GPU=OFF
-DFAISS_ENABLE_CUVS=OFF

[JAicewizard]

This disables building for GPU, which is not what I want. I do need to build for the gpu, just preferably without regenerating sourcefiles

[tarang-jain]

Oh I see what you are saying. Not sure if this will help, but I was using ccache a long time ago. Perhaps these flags help:

-DCMAKE_CUDA_COMPILER_LAUNCHER=ccache \
-DCMAKE_C_COMPILER_LAUNCHER=ccache \
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache \

If this does not work, I'd have to defer to the original authors / maintainers of this repo.

[JAicewizard]

One option I just thought of would be to check the (hash of) the contents of the file, before writing? If the contents is the same, keep the file as-is to prevent recompiling. Does this sound like a reasonable fix?