Building with a recent version of CUDA (12.8.1, 12.9.0, etc.) results in a large number of warnings like
nvcc warning : Support for offline compilation for architectures prior to '<compute/sm/lto>_75' will be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
Older architectures are still very much supported and largely in use (e.g. NVIDIA V100 is sm_70), so dropping support for them does not make much sense.
Is there a way to pass -Wno-deprecated-gpu-targets to nvcc ?
Should it be passed by default ?
Building with a recent version of CUDA (12.8.1, 12.9.0, etc.) results in a large number of warnings like
Older architectures are still very much supported and largely in use (e.g. NVIDIA V100 is
sm_70), so dropping support for them does not make much sense.Is there a way to pass
-Wno-deprecated-gpu-targetstonvcc?Should it be passed by default ?