cccl missing in matx_cpm_cache

[CrreeL]

Hi! I'm a newb to the MatX community. I'm working on a simulation software project with MatX. I need to work in an offline server cluster, so I followed the instructions in "MatX in Offline Environments" section in MatX 0.9.4 documentation.
My cmake options in internet-enabled environment were "cmake -DMATX_NVTX_FLAGS=ON -DMATX_BUILD_32_BIT=ON -DMATX_EN_FILEIO=ON -DMATX_EN_CUDA_LINEINFO=ON -DMATX_EN_CUTENSOR=ON ..".
I have cuda toolkit 13.0 and cccl 3.0.0 in my docker container, while MatX still thied to fetch cccl 3.2.0.
However, when I transfered MatX Source code and CPM cache to offline system and performed cmake, I got stuck at
-- Finding CCCL...
-- CPM: adding package CCCL@3.2.0
It really bothered me, untill I found that cccl 3.2.0 source code was missing in matx_cpm_cache/cccl.
Finally I solved the problem by copying build/_deps/cccl-src to matx_cpm_cache/cccl and renamed it with 26015dd9d94d91cabbd4dc612f7f2484e2dc5556.
I wonder if I ignored some details or it was a bug. If it was a bug, could you please have it fixed, so that MatX would be more friendly to offline users!

[cliffburdick]

@CrreeL have you considered installing MatX directly into the container as part of your Dockerfile? If you do a make install after make it should put CCCL into the proper location.

[CrreeL]

So glad to hear from you!
Yes, exactly, I didn't take the "MatX Installed to the System" approach, instead, I followed the instructions in "MatX as a Subdirectory" section in README.md. I prefer "MatX as a Subdirectory", to keep my docker image simple, and I think "MatX as a Subdirectory" will be easier for me to follow new versions of MatX in the future.

[CrreeL]

BTW, my project was based on armadillo (a matlab-like C++ library for linear algebra on CPU) and I dopted cuDSS as sparse linear solver. I want to trnasfer the whole project to GPU for speed up, and was desperate for a "GPU-version armadillo", untill I hear about MatX. Appreciate it a lot that your team developed MatX to fill the niche!

[cliffburdick]

That's great! I think you probably noticed we have cuDSS support now using the solve function. If there's any functionality from armadillo we're missing please let us know.