A little confusion of CUDA macros

[QingleiCao]

There are DPLASMA_GPU_WITH_CUDA, DPLASMA_HAVE_CUDA, PARSEC_HAVE_CUDA, PARSEC_GPU_WITH_CUDA, and PARSEC_HAVE_DEV_CUDA_SUPPORT. What's the purpose of each?

[bosilca]

Easy ;) Aside from the obvious, aka. parsec one are related to parsec while the DPLASMA one to DPLASMA, these allows to have a parsec with GPU but a DPLASMA without. *_GPU_WITH_CUDA is a build option while *_HAVE_CUDA is the code option superseded lately by *_HAVE_DEV_CUDA_SUPPORT. They all exists in the headers files, but I think the best to use in code development is _HAVE_DEV_XYZ_SUPPORT.

Moreover, these are also available for other accelerators, such as HIP and LEVEL_ZERO