cmake --install . does not work

[vicentebolea]

The generate Makefile/ninja.rules does not have any install rule. While I understand that cuSZ comes with the build.py script, this makes things complicated when trying to use SZ from other CMake projects.

[jtian0]

Hi @vicentebolea,

Thank you for pointing out, and I will improve: I left out the installation.

The intention of build.py is to avoid lengthy commands when invoking cmake or extra configuration in CMakeLists.txt regarding CUDA arch. But I agree that the script should not be a trouble when interoperating with cmake.

[vicentebolea]

I could also possibly help cmakefying this

On Fri, Feb 4, 2022, 6:28 PM Nate @.***> wrote:

Hi @vicentebolea https://github.com/vicentebolea,

Thank you for pointing out, and I will improve: I left out the installation.

The intention of build.py is to avoid lengthy commands when invoking cmake or extra configuration in CMakeLists.txt regarding CUDA arch. But I agree that the script should not be a trouble when interoperating with cmake.

— Reply to this email directly, view it on GitHub https://github.com/szcompressor/cuSZ/issues/65#issuecomment-1030427622, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAHFOFRMRVZBHGFIWGZ5BGLUZROJRANCNFSM5NHMS6FA . You are receiving this because you were mentioned.Message ID: @.***>

[bozhang-hpc]

Hi,

I'm also experiencing this cmake install issue. By the way, is it possible to build shared libraries? Some CRAY supercomputers doesn't have good support to static linkage.

Bests,

Bo

[jtian0]

Hello @Zhang690683220,

We have a new branch where shared libraries are enabled. The branch sees building process problems because of the dependency modules (such as NVIDIA::CUB). Can you tell which CUDA version are you using? I might come up with a quick fix (by dropping compatibility).

CC: @dingwentao

[bozhang-hpc]

Hi,

I'm using cuda/11.5 and maybe upgrade to 11.7 later. If you have the access to perlmutter, can you do a test on it? Because the math library like cublas and other stuffs are strangely managed on that machine.

Bo

[jtian0]

Hi @Zhang690683220, we just released a new version, v0.3.1, corresponding to https://github.com/szcompressor/cuSZ/commit/4f9c9f162d5b34e9a5dbdd87d6cfcae7b8f6037e.

It is dynamically linked by default https://github.com/szcompressor/cuSZ/blob/4f9c9f162d5b34e9a5dbdd87d6cfcae7b8f6037e/CMakeLists.txt#L6 We tested on Perlmutter with the following modules loaded

module load cmake/3.22.0
module load gcc/10.3.0
module load cudatoolkit/11.5

And it works in our initial test.

Please refer to the document to build from the source code. If you have any questions, please tell us. Thank you for trying out our software.

[jtian0]

Hi @vicentebolea, I am sorry for the late reply. We spent months resolving most compiling issues. We currently have full support for cmake compiling, and the previous build.py is retired. Everything about building from the source code is documented here.

Thank you for reaching out to help! We appreciate it.

[vicentebolea]

@jtian0 great! Thanks for help, this was a long waited feature.