I’m trying to run a scientific application that uses GPU accelerated compression library Bitcomp. It runs without problem on x86 systems, but when I try to run it on an ARM-base system with NVIDIA A100 GPUs, the executions fail due to segmentation. After running it with GDB, I discovered that the segmentation comes from the Bitcomp library. The application comes with the libbitcomp.a library for many versions of CUDA for x86, but only 10.0 for aarch64. I search for the libbitcomp.a library for more recent CUDA versions, but I could not find anything. I also searched for the library’s source in the hope that I would be able to compile it, but without success.
Does anyone know where to find the Bitcomp library for newer CUDA versions on aarch64 systems?
Weird because on NVIDIA’s nvCOMP website, I can download nvCOMP for Linux SBSA systems. However, when I try to build it, it asks me to provide the location of Bitcomp CMake configuration files. The application I’m trying to run on the system came with a CUDA 10.0 version of bitcomp.a library together with the CMake config files. I used this file to try to build nvCOMP to see if it would generate a newer bitcomp.a, but without success.