使用TensorFlow编译C ++代码而不使用Bazel

时间:2017-07-26 20:20:52

标签: c++ tensorflow compilation

我希望将TensorFlow及其ANN功能用于我的计算机制目的,其中大多数代码都是用c ++编写的。是否可以在不使用Bazel的同时使用TensorFlow .h文件进行c ++编译?如果是这样的话,我真的很感激一个例子(到目前为止还没有找到任何在线版本)。 感谢

编辑:我做了,但我无法关注。让我举一个例子,也许我们可以从那里开始。我使用的是ubuntu 16.10,gcc(Ubuntu 6.2.0-5ubuntu12)6.2.0 20161005和Python 2.7.12+。我从源代码安装了bazel,并且还克隆了TF存储库(〜/ Desktop / tensorflow)。从(https://www.tensorflow.org/api_guides/cc/guide)稍微修改一下的例子,我在example.cc:

#include "tensorflow/cc/client/client_session.h"
#include "tensorflow/cc/ops/standard_ops.h"
#include "tensorflow/core/framework/tensor.h"
#include <iostream>

int main() {
  using namespace tensorflow;
  using namespace tensorflow::ops;
  using namespace std;
  Scope root = Scope::NewRootScope();
  // Matrix A = [3 2; -1 0]
  auto A = Const(root, { {3.f, 2.f}, {-1.f, 0.f}});
  // Vector b = [3 5]
  auto b = Const(root, { {3.f, 5.f}});
  // v = Ab^T
  auto v = MatMul(root.WithOpName("v"), A, b, MatMul::TransposeB(true));
  std::vector<Tensor> outputs;
  ClientSession session(root);
  // Run and fetch v
  TF_CHECK_OK(session.Run({v}, &outputs));
  // Expect outputs[0] == [19; -3]
  LOG(INFO) << outputs[0].matrix<float>();
  return 0;
  cout<<"compiled correctly!"<<endl;
}

它位于〜/ Desktop / tensorflow / tensorflow / cc / example中。我的BUILD文件 - 也在〜/ Desktop / tensorflow / tensorflow / cc / example中 - 读取:

cc_binary(
    name = "example",
    srcs = ["example.cc"],
    deps = [
        "//tensorflow/cc:cc_ops",
        "//tensorflow/cc:client_session",
        "//tensorflow/core:tensorflow",
    ],
)

我尝试使用:

从〜/ Desktop / tensorflow编译
bazel build tensorflow/cc/example/...

这就是我得到的:

INFO: Found 1 target...
Target //tensorflow/cc/example:example up-to-date:
  bazel-bin/tensorflow/cc/example/example
INFO: Elapsed time: 0.381s, Critical Path: 0.00s

然后当我去〜/ Desktop / tensorflow / bazel-bin / tensorflow / cc / example并运行:

./example 

我明白了:

2017-07-27 09:58:39.906578: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-07-27 09:58:39.906628: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-07-27 09:58:39.906636: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2017-07-27 09:58:39.906641: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2017-07-27 09:58:39.906646: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
2017-07-27 09:58:39.907751: I tensorflow/cc/example/example.cc:22] 19
-3

任何帮助都会非常感激,因为我试图用手包围这个。谢谢你的耐心等待。

3 个答案:

答案 0 :(得分:1)

在不使用Bazel等任何构建工具的情况下构建基于tensorflow框架的c ++代码的步骤。

  1. 从以下链接克隆/下载github的张量流。

    由于pip install tensorflow只会为python安装tensorflow。

    https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/makefile

  2. 要运行your_code.cc(您要构建的c ++文件,请对c和c ++使用.cc.cpp由于某种原因不起作用)

    将.cc文件和.pb文件放在tensorflow / tensorflow / tools / benchmark /

    在vim编辑器中编辑Makefile中的以下行 (tensorflow / tensorflow /了contrib /生成文件/生成文件)

    BENCHMARK_NAME := $(BINDIR)benchmark
    

    而不是基准,请提供可执行文件的名称

    BENCHMARK_SRCS := \
    tensorflow/core/util/reporter.cc \
    tensorflow/tools/benchmark/benchmark_model.cc \
    tensorflow/tools/benchmark/benchmark_model_main.cc
    

    删除这些文件并在此处添加源代码(.cc)。

  3. 下一步在Makefile中注释这一行:

    all: $(LIB_PATH) $(BENCHMARK_NAME)
    

    并添加以下行:

    all: $(BENCHMARK_NAME) 
    
  4. 接下来运行make文件,转到根目录(tesnroflow /)并输入以下命令。

    make -f tensorflow/contrib/makefile/Makefile
    
  5. 如果您遇到任何错误,只需输入以下命令,

    export HOST_NSYNC_LIB=`/home/nivedita_s/Downloads/tensorflow-master/tensorflow/contrib/makefile/compile_nsync.sh`
    export TARGET_NSYNC_LIB="$HOST_NSYNC_LIB"
    
  6. 如果您已正确执行了所有过程,那么将在下面提到的此文件夹中创建可执行文件,

    tensorflow/tensorflow/contrib/makefile/gen/bin/
    
  7. 将以您为可执行文件指定的名称创建输出。 如果./executable_file给出了任何库问题,请按照进一步的步骤(a和b)进行操作。

    • vim ~/.profile在最后添加此行,

      export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/path_to_the_tensorflow/tensorflow/contrib/makefile/downloads/protobuf/src/.libs
      
    • 运行此命令

      source ~/.profile
      
  8. 现在您应该能够在该终端中运行您的程序。

答案 1 :(得分:0)

最简单的方法是将所有必要的文件复制到本地项目中:

cd /git/github.com/tensorflow
git clone git@github.com:tensorflow/tensorflow.git
# build TensorFlow once (tensorflow:libtensorflow_cc.so, tensorflow:libtensorflow.so)
cd project
mkdir tensorflow
cp /git/github.com/tensorflow/tensorflow/bazel-tensorflow/tensorflow/cc tensorflow/cc
cp /git/github.com/tensorflow/tensorflow/bazel-genfiles/tensorflow/cc tensorflow

有了这些文件,一个非常基本的CMakeLists.txt可以完成这项任务:

cmake_minimum_required( VERSION 2.8 )

set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED ON)

set(TensorFlow_ABI 0)
set(TensorFlow_INCLUDE_DIRS "${HOME}/.local/lib/python2.7/site-packages/tensorflow/include")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_GLIBCXX_USE_CXX11_ABI=${TensorFlow_ABI}")
set(CMAKE_EXE_LINKER_FLAGS  "${CMAKE_EXE_LINKER_FLAGS} -D_GLIBCXX_USE_CXX11_ABI=${TensorFlow_ABI}" )


project( TFMyProject )

include_directories(SYSTEM "${TensorFlow_INCLUDE_DIRS}/external/nsync/public")
include_directories(SYSTEM ${TensorFlow_INCLUDE_DIRS})
include_directories(SYSTEM ".")

add_executable (example example.cc)
TARGET_LINK_LIBRARIES(example -Wl,--allow-multiple-definition -Wl,--whole-archive "/git/github.com/tensorflow/bazel-bin/tensorflow/libtensorflow_cc.so" -Wl,--no-whole-archive)
TARGET_LINK_LIBRARIES(example -Wl,--allow-multiple-definition -Wl,--whole-archive "/git/github.com/tensorflow/bazel-bin/tensorflow/libtensorflow_framework.so" -Wl,--no-whole-archive)

甚至可以使用a python script设置这些设置。

答案 2 :(得分:0)

如果您不想过多地修改环境,则有两种可能。一种是使用Floop的tensorflow_cc项目编译C ++ API,然后将其安装在系统上。另一种可能性是我的打包项目tensorflow_cpp_packaging,为Tensorflow的C和C ++ API提供Debian软件包。这两个项目都使用CMake(而不是Bazel)来支持源文件的C ++编译。有时,我会编译最新的Tensorflow版本,它们在release部分中适用于64位CPU和Raspberry Pi。

从Tensorflow的角度来看,tensorflow_cc的优势是可以构建GPU支持,而我的项目只能使用CPU进行推理。

如果您选择使用任何一个项目,我仍然可以推荐my short tutorial我写过的有关如何在Python中冻结Tensorflow模型并使用C或C ++ API加载并用于推理的内容。