Hello friends,
I decided to return back to oneAPI and play a bit with it. Unfortunately, I’m facing compatibility issues. The 2023.1.0 plug-in does not work due to libstdc++
version.
[opc@sws lib]$ ldd libpi_cuda.so
./libpi_cuda.so: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by ./libpi_cuda.so)
linux-vdso.so.1 (0x00007fff159ea000)
libcuda.so.1 => /usr/lib/x86_64-linux-gnu/libcuda.so.1 (0x00007f276ff16000)
libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f276fcfd000)
libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007f276f968000)
libc.so.6 => /lib64/libc.so.6 (0x00007f276f5a3000)
/lib64/ld-linux-x86-64.so.2 (0x00007f2771bf7000)
libm.so.6 => /lib64/libm.so.6 (0x00007f276f221000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007f276f01d000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f276edfd000)
librt.so.1 => /lib64/librt.so.1 (0x00007f276ebf5000)
As you can see, my RHEL8 based system uses older version of libsted++
.
My question is: What Linux distributions are supported by oneAPI for NVIDIA GPUs? It is obviously not the same as the oneAPI toolchain itself, which is quite unfortunate. I’ve installed Intel’s compiler from theirs official YUM repository and it works. Also note that the previous version 2023.0.0 worked.
I’ve also tried to look for the source code of the plug-in to re-compile it myself, but I’ve failed to find it.
There is NVIDIA GPUs support in open source version of oneAPI toolchain (Intel’s LLVM git repo) which works as expected, but I don’t know what is the difference between the downloadable binary and the source code in Intel’s repo.
Do you have any hints on how to use oneAPI for NVIDIA GPUs on RHEL8-based distros?
Thanks in advance.