Cuda build feature issue

[cpacia]

Either lurk-rs or one of its dependencies appear to be checking to see if you have the nvidia-cuda-toolkit installed locally and linking to it regardless of whether or not you use the cuda build feature.

This would possibly explain why I would notice the GPU firing even though I wasn't building with the CUDA feature flag.

I only noticed this because I'm linking to to Rust from Go via CGO and you need to set the CGO directives based on whether the rust binary is linking to cuda or not.... which is dependent on what you have installed on your system and not what build features you use. The workaround I'm using is a Go build tag where, if you have the nvidia-cuda-toolkit installed you use the build tag, if you don't you omit it.

I don't know if this is a bug but I wasn't really expecting that behavior.

[winston-h-zhang]

This is a known issue: lurk-lang/arecibo#259

I'm actually pushing fixes as I type this 😅

[winston-h-zhang]

@cpacia It should be fixed now, please cargo update -p grumpkin-msm

[cpacia]

Thanks @winston-h-zhang I'll check it out

[samuelburnham]

Closing as fixed by lurk-lab/grumpkin-msm#9 and lurk-lang/arecibo#279, and tested locally in lurk-rs.