CentOS环境下配置使用GPU加速的dlib库

1. 下载安装dlib

配置CUDA加速环境：

    git clone https://github.com/davisking/dlib.git
    cd dlib
    mkdir build
    cd build
    cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1
    cmake --build .

过程中出现错误：

(dlib-gpu) [xl@GPU1 build]$ cmake --build .
[  0%] Building NVCC (Device) object dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cuda_dlib.cu.o
/home/xl/dlib-gpu/dlib/cuda/cuda_dlib.cu(1762): error: calling a constexpr __host__ function("log1p") from a __device__ function("cuda_log1pexp") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

1 error detected in the compilation of "/tmp/tmpxft_00001c85_00000000-6_cuda_dlib.cpp1.ii".
CMake Error at dlib_generated_cuda_dlib.cu.o.Release.cmake:267 (message):
  Error generating file
  /home/xl/dlib-gpu/build/dlib/CMakeFiles/dlib.dir/cuda/./dlib_generated_cuda_dlib.cu.o


gmake[2]: *** [dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cuda_dlib.cu.o] Error 1
gmake[1]: *** [dlib/CMakeFiles/dlib.dir/all] Error 2
gmake: *** [all] Error 2
(dlib-gpu) [xl@GPU1 build]$ cmake --build . --expt-relaxed-constexpr
Unknown argument --expt-relaxed-constexpr
Usage: cmake --build <dir> [options] [-- [native-options]]
Options:
  <dir>          = Project binary directory to be built.
  --target <tgt> = Build <tgt> instead of default targets.
                   May only be specified once.
  --config <cfg> = For multi-configuration tools, choose <cfg>.
  --clean-first  = Build target 'clean' first, then build.
                   (To clean only, use --target 'clean'.)
  --use-stderr   = Ignored.  Behavior is default in CMake >= 3.0.
  --             = Pass remaining options to the native tool.
(dlib-gpu) [xl@GPU1 build]$

之后按照提示，找到出错的文件：

定位到出现错的代码行：cuda_dlib.cu(1762)，即cuda_dlib.cu中的第1762行
1762行的内容如下：

这是一行关于涉及数学计算的代码，查了一下std::exp(x)和std::log1p(x)两个函数的用法：

std::exp(x) ：https://zh.cppreference.com/w/cpp/numeric/math/exp

std::log1p(x)：https://en.cppreference.com/w/cpp/numeric/math/log1p

也不知为什么将std::log1p()函数替换为std::log1pf()函数后，问题解决了

可能是因为：




重新编译：

(dlib-gpu) [xl@GPU1 build]$ cmake --build .
[  0%] Building NVCC (Device) object dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cuda_dlib.cu.o
Scanning dependencies of target dlib
[  0%] Building CXX object dlib/CMakeFiles/dlib.dir/base64/base64_kernel_1.cpp.o
[  0%] Building CXX object dlib/CMakeFiles/dlib.dir/bigint/bigint_kernel_1.cpp.o
[  0%] Building CXX object dlib/CMakeFiles/dlib.dir/bigint/bigint_kernel_2.cpp.o
[  0%] Building CX