cuda-samples
cuda-samples copied to clipboard
Update Makefiles for HPC SDK
When using CUDA from the HPC SDK (NVHPC), compiling the examples using make doesn't only require setting CUDA_PATH
(e.g. /opt/nvidia/hpc_sdk/Linux_x86_64/22.3/cuda
), but also (at least for samples in 4_CUDA_Libraries
) to set the library path (nvcc doesn't know it?) by e.g. adding
-L/opt/nvidia/hpc_sdk/Linux_x86_64/22.3/math_libs/11.6/lib64
to LDFLAGS
. It would be helpful for beginners if the Makefiles would as a first step detect from the CUDA_PATH
that one is using the HPC SDK and modify LDFLAGS
on its own. As a second step one might want to at least try out a few default CUDA_PATH
s instead of specifying just one (this causes problems also with users of the normal CUDA SDK, see e.g. #127). Even better would be code to generally construct the CUDA_PATH
as long as e.g. nvcc
is on the PATH
.
If one already reworks these Makefiles, one might want to outsource all the repeated elements of the Makefiles into the Common
folder (DRY principle).