Arch Linux User Repository

I will look into it, thanks

Can you perhaps consider adding a hook file, so that whenever cuda and cudnn are updated, lc0 gets rebuilt too? See https://wiki.archlinux.org/title/pacman#Hooks

The reason I'm asking is that cuda runtime doesn't like it when the version used to build lc0 is different than the current version, so now I need to rebuild lc0 manually after updating cuda every so often. Ideally this process should be automated IMO.

It seems your cuda/cudnn packages are outdated. Run pls pacman -Syu in order to update your system and then install lc0 itself.

Build fails with

The Meson build system
Version: 0.51.0
Source dir: /home/eric/.cache/pacaur/lc0-cudnn-git/src/lc0-cudnn
Build dir: /home/eric/.cache/pacaur/lc0-cudnn-git/src/lc0-cudnn/build/release
Build type: native build
DEPRECATION: Option uses prefix "build_", which is reserved for Meson. This will become an error in the future.
Project name: lc0
Project version: undefined
C++ compiler for the build machine: c++ (gcc 9.1.0 "c++ (GCC) 9.1.0")
C++ compiler for the host machine: c++ (gcc 9.1.0 "c++ (GCC) 9.1.0")
Build machine cpu family: x86_64
Build machine cpu: x86_64
Library libprotobuf found: YES
Program protoc found: YES (/usr/bin/protoc)
Library pthread found: YES
Library dl found: YES
Library libtensorflow_cc found: YES
Found pkg-config: /usr/bin/pkg-config (1.6.1)
Found CMake: /usr/bin/cmake (3.14.5)
Run-time dependency accelerate found: NO (tried pkgconfig and cmake)
Library mkl_rt found: YES
Library openblas.dll found: NO
Library openblas found: YES
Library OpenCL found: YES
Run-time dependency opencl found: NO (tried pkgconfig and cmake)
Library cublas found: YES
Library cudnn found: YES
Library cudart found: YES
Program nvcc found: YES (/opt/cuda/bin/nvcc)
Dependency zlib found: YES (cached)
Dependency gtest found: YES (cached)
Build targets in project: 8
Found ninja-1.9.0 at /usr/bin/ninja
[3/26] Generating 'lc0@exe/common_kernels.o'.
FAILED: lc0@exe/common_kernels.o 
/opt/cuda/bin/nvcc -c ../../src/neural/cuda/common_kernels.cu -o lc0@exe/common_kernels.o -I /home/eric/.cache/pacaur/lc0-cudnn-git/src/lc0-cudnn/src --std=c++14 -Xcompiler -fPIC -I /opt/cuda/include/ -I /usr/local/cuda/include/
/usr/include/c++/9.1.0/bits/stl_function.h(437): error: identifier "__builtin_is_constant_evaluated" is undefined

1 error detected in the compilation of "/tmp/tmpxft_00006df4_00000000-6_common_kernels.cpp1.ii".
[4/26] Generating 'lc0_lib@sha/common_kernels.o'.
FAILED: lc0_lib@sha/common_kernels.o 
/opt/cuda/bin/nvcc -c ../../src/neural/cuda/common_kernels.cu -o lc0_lib@sha/common_kernels.o -I /home/eric/.cache/pacaur/lc0-cudnn-git/src/lc0-cudnn/src --std=c++14 -Xcompiler -fPIC -I /opt/cuda/include/ -I /usr/local/cuda/include/
/usr/include/c++/9.1.0/bits/stl_function.h(437): error: identifier "__builtin_is_constant_evaluated" is undefined

1 error detected in the compilation of "/tmp/tmpxft_00006dfd_00000000-6_common_kernels.cpp1.ii".
[5/26] Generating 'lc0@exe/fp16_kernels.o'.
FAILED: lc0@exe/fp16_kernels.o 
/opt/cuda/bin/nvcc -arch=compute_70 -code=sm_70 -c ../../src/neural/cuda/fp16_kernels.cu -o lc0@exe/fp16_kernels.o -I /home/eric/.cache/pacaur/lc0-cudnn-git/src/lc0-cudnn/src --std=c++14 -Xcompiler -fPIC -I /opt/cuda/include/ -I /usr/local/cuda/include/
/usr/include/c++/9.1.0/bits/stl_function.h(437): error: identifier "__builtin_is_constant_evaluated" is undefined

1 error detected in the compilation of "/tmp/tmpxft_00006df5_00000000-6_fp16_kernels.cpp1.ii".
[6/26] Generating 'lc0_lib@sha/fp16_kernels.o'.
FAILED: lc0_lib@sha/fp16_kernels.o 
/opt/cuda/bin/nvcc -arch=compute_70 -code=sm_70 -c ../../src/neural/cuda/fp16_kernels.cu -o lc0_lib@sha/fp16_kernels.o -I /home/eric/.cache/pacaur/lc0-cudnn-git/src/lc0-cudnn/src --std=c++14 -Xcompiler -fPIC -I /opt/cuda/include/ -I /usr/local/cuda/include/
/usr/include/c++/9.1.0/bits/stl_function.h(437): error: identifier "__builtin_is_constant_evaluated" is undefined

1 error detected in the compilation of "/tmp/tmpxft_00006dfe_00000000-6_fp16_kernels.cpp1.ii".
[17/26] Compiling C++ object 'lc0@exe/src_mcts_search.cc.o'.
ninja: build stopped: subcommand failed.
==> ERROR: A failure occurred in build().
    Aborting...
:: failed to build lc0-cudnn-git package(s)

I suspect devs changed the way how protobuf lib is packaged (beforehand it compiled statically). I didn't catch it, because I have this lib installed as dependency of some other soft. Will add it as dependency immediately.

Output of ldd $(which lc0)

$ ldd $(which lc0)
    linux-vdso.so.1 (0x00007fffee7ee000)
    libprotobuf-lite.so => not found
    libprotobuf.so => not found
    libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f0e329ef000)
    libcublas.so.10.0 => /opt/cuda/lib64/libcublas.so.10.0 (0x00007f0e2e459000)
    libcudnn.so.7 => /usr/lib/libcudnn.so.7 (0x00007f0e19632000)
    libcudart.so.10.0 => /opt/cuda/lib64/libcudart.so.10.0 (0x00007f0e193b8000)
    libz.so.1 => /usr/lib/libz.so.1 (0x00007f0e1919f000)
    libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x00007f0e19010000)
    libm.so.6 => /usr/lib/libm.so.6 (0x00007f0e18e8b000)
    libgcc_s.so.1 => /usr/lib/libgcc_s.so.1 (0x00007f0e18e71000)
    libc.so.6 => /usr/lib/libc.so.6 (0x00007f0e18cad000)
    /lib64/ld-linux-x86-64.so.2 => /usr/lib64/ld-linux-x86-64.so.2 (0x00007f0e32c40000)
    librt.so.1 => /usr/lib/librt.so.1 (0x00007f0e18ca3000)
    libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f0e18c9c000)

Output of pacman -Qs is empty.

somehow lc0 got linked against libprotobuf-lite.so which, apparently, is nonexistent in your system can you post your output of ldd $(which lc0) and pacman -Qs protobuf?

I am getting this error:

lc0: error while loading shared libraries: libprotobuf-lite.so: cannot open shared object file: No such file or directory

I don't know where to start trying to fix this.

Thank you! Your fix works for me.