Build llama runner fail due to pthread

[CHNtentes]

Hi, When I follow the insrtuction here to build llama runner on pc, it throws an error:

(executorch) v2x@v2x-OMEN-Desktop:/hdd_2/ltg/executorch$ cmake --build cmake-out/examples/models/llama2 -j8 --config Release                              [  8%] Building CXX object custom_ops/CMakeFiles/custom_ops.dir/hdd_2/ltg/executorch/extension/parallel/thread_parallel.cpp.o
[ 16%] Building CXX object runner/CMakeFiles/llama_runner.dir/__/tokenizer/bpe_tokenizer.cpp.o
[ 25%] Building CXX object custom_ops/CMakeFiles/custom_ops.dir/op_sdpa.cpp.o
[ 41%] Building CXX object runner/CMakeFiles/llama_runner.dir/runner.cpp.o
[ 41%] Building CXX object runner/CMakeFiles/llama_runner.dir/__/sampler/sampler.cpp.o
[ 58%] Building CXX object runner/CMakeFiles/llama_runner.dir/hdd_2/ltg/executorch/kernels/optimized/blas/CPUBlas.cpp.o
[ 58%] Building CXX object runner/CMakeFiles/llama_runner.dir/hdd_2/ltg/executorch/extension/evalue_util/print_evalue.cpp.o
/hdd_2/ltg/executorch/examples/models/llama2/custom_ops/op_sdpa.cpp: In lambda function:
/hdd_2/ltg/executorch/examples/models/llama2/custom_ops/op_sdpa.cpp:420:17: note: the ABI for passing parameters with 32-byte alignment has changed in GCC 4.6
  420 |                 [scaling_factor](Vec x, Vec y) {
      |                 ^
[ 66%] Linking CXX shared library libllama_runner.so
[ 66%] Built target llama_runner
[ 75%] Linking CXX static library libcustom_ops.a
[ 75%] Built target custom_ops
[ 83%] Building CXX object CMakeFiles/llama_main.dir/main.cpp.o
[ 91%] Building CXX object CMakeFiles/llama_main.dir/hdd_2/ltg/executorch/backends/xnnpack/threadpool/cpuinfo_utils.cpp.o
[100%] Linking CXX executable llama_main
/usr/bin/ld: CMakeFiles/llama_main.dir/hdd_2/ltg/executorch/backends/xnnpack/threadpool/cpuinfo_utils.cpp.o: in function `torch::executorch::cpuinfo::get_num_performant_cores()':
cpuinfo_utils.cpp:(.text+0x909): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libxnnpack_backend.a(threadpool.cpp.o): in function `torch::executorch::threadpool::get_threadpool()':
threadpool.cpp:(.text._ZN5torch10executorch10threadpool14get_threadpoolEv+0x98): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libpthreadpool.a(pthreads.c.o): in function `pthreadpool_create':
pthreads.c:(.text+0x32d): undefined reference to `pthread_create'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libpthreadpool.a(pthreads.c.o): in function `pthreadpool_destroy':
pthreads.c:(.text+0x67e): undefined reference to `pthread_join'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libcpuinfo.a(init.c.o): in function `cpuinfo_initialize':
init.c:(.text+0x17): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libXNNPACK.a(init.c.o): in function `xnn_initialize':
init.c:(.text+0x61): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libXNNPACK.a(hardware-config.c.o): in function `xnn_init_hardware_config':
hardware-config.c:(.text+0xf6): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libXNNPACK.a(argmaxpool-config.c.o): in function `xnn_init_f32_argmaxpool_config':
argmaxpool-config.c:(.text+0x87): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libXNNPACK.a(avgpool-config.c.o): in function `xnn_init_f16_avgpool_config':
avgpool-config.c:(.text+0xe0): undefined reference to `pthread_once'
/usr/bin/ld: /hdd_2/ltg/executorch/cmake-out/lib/libXNNPACK.a(avgpool-config.c.o):avgpool-config.c:(.text+0x111): more undefined references to `pthread_once' follow
collect2: error: ld returned 1 exit status
make[2]: *** [CMakeFiles/llama_main.dir/build.make:133: llama_main] Error 1
make[1]: *** [CMakeFiles/Makefile2:119: CMakeFiles/llama_main.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

I ask ChatGPT and it tells me to add 'pthread' to target_link_libraries in examples/models/llama2/CMakeLists.txt, and it solves the issue effectively. I just wonder if this issue is only happening on my pc?

[mergennachin]

Yeah, I think it happens on PC.

We added that as part of "Common Issues and Mitigations" section in https://github.com/pytorch/executorch/blob/main/examples/models/llama2/README.md#common-issues-and-mitigations

[mergennachin]

Are there any further actionable things to do with this github issue?

[CHNtentes]

Thanks for your reply. I'll close this issue.

[CHNtentes]

Are there any further actionable things to do with this github issue?

@mergennachin I built llama runner on pc successfully, but when I build it for android, it again gives me another similar error:

(executorch) v2x@v2x-OMEN-Desktop:/hdd_2/ltg/executorch$ cmake --build cmake-out-android/examples/models/llama2 -j16 --config Release                     [ 16%] Building CXX object runner/CMakeFiles/llama_runner.dir/runner.cpp.o
[ 16%] Building CXX object runner/CMakeFiles/llama_runner.dir/__/tokenizer/bpe_tokenizer.cpp.o
[ 25%] Building CXX object runner/CMakeFiles/llama_runner.dir/hdd_2/ltg/executorch/extension/evalue_util/print_evalue.cpp.o
[ 33%] Building CXX object runner/CMakeFiles/llama_runner.dir/hdd_2/ltg/executorch/kernels/optimized/blas/CPUBlas.cpp.o
[ 41%] Building CXX object runner/CMakeFiles/llama_runner.dir/__/sampler/sampler.cpp.o
[ 58%] Building CXX object custom_ops/CMakeFiles/custom_ops.dir/hdd_2/ltg/executorch/extension/parallel/thread_parallel.cpp.o
[ 58%] Building CXX object custom_ops/CMakeFiles/custom_ops.dir/op_sdpa.cpp.o
[ 66%] Linking CXX static library libllama_runner.a
[ 66%] Built target llama_runner
[ 75%] Linking CXX static library libcustom_ops.a
[ 75%] Built target custom_ops
[ 83%] Building CXX object CMakeFiles/llama_main.dir/main.cpp.o
[ 91%] Building CXX object CMakeFiles/llama_main.dir/hdd_2/ltg/executorch/backends/xnnpack/threadpool/cpuinfo_utils.cpp.o
[100%] Linking CXX executable llama_main
ld.lld: error: unable to find library -lpthread
clang++: error: linker command failed with exit code 1 (use -v to see invocation)
make[2]: *** [CMakeFiles/llama_main.dir/build.make:133: llama_main] Error 1
make[1]: *** [CMakeFiles/Makefile2:119: CMakeFiles/llama_main.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

I already add pthread to target link libraries in examples/models/llama2/CMakeLists.txt, and I don't know what to do now...

[mcr229]

@kirklandsign

[CHNtentes]

I actually found a temporary solution, just remove "-lpthread" from cmake-out-android/examples/models/llama2/CMakeFiles/llama_main.dir/link.txt and the build succeeded. It seems that the project does not require -lpthread after all, not sure why it was added though.

[kirklandsign]

Hi @CHNtentes for this part, do you mean for Android or PC?

I actually found a temporary solution, just remove "-lpthread" from cmake-out-android/examples/models/llama2/CMakeFiles/llama_main.dir/link.txt and the build succeeded. It seems that the project does not require -lpthread after all, not sure why it was added though.

[CHNtentes]

Hi @CHNtentes for this part, do you mean for Android or PC?

I actually found a temporary solution, just remove "-lpthread" from cmake-out-android/examples/models/llama2/CMakeFiles/llama_main.dir/link.txt and the build succeeded. It seems that the project does not require -lpthread after all, not sure why it was added though.

It is for Android. You can find it in filename.

[kirklandsign]

It is for Android. You can find it in filename.

OK for android I didn't use -lpthread and it worked for me. Android has its own libc bionic so the behavior might be different.

[kirklandsign]

All resolved?