Caffe源码学习（1）：Blob

笔者在caffe上做各种实验有一段时间了，但一直都只是在修改配置文件或者存在某些新的idea却难以实现的地步，很多时候实现一些idea需要深入到底层去修改或者添加一下新的layer等，这样也就要求对caffe的底层源码有一个较为深层的理解，这个系列的博客将记录和分享在学习caffe源码中的体会和过程，作为一个EE转CS的编程菜鸟，如有错误，希望读者指正。
caffe大致上可以分为Blob, Layer, Net, Solver这四个大的模块。

Solver: An interface for classes that perform     optimization on NetsNet: Connects Layers together into a directed acyclic graph (DAG) specified by a NetParameterLayer: An interface for the units of computation which can be composed into a NetBlob: A wrapper around SyncedMemory holders serving as the basic computational unit through which Layers, Nets, and Solvers interact

Blob作为caffe数据流通的一个基本数据结构，层与层之间的数据是通过blob来实现的，具有cpu和gpu之间同步的能力，可以看作一个四维数组(num, channels, heights, width)。主要代码位于/home/xxx/caffe/include/blob.hpp。

*1. 主要变量*

shared_ptr<SyncedMemory> data_;shared_ptr<SyncedMemory> diff_;shared_ptr<SyncedMemory> shape_data_;vector<int> shape_;int count_;int capacity_;

Blob作为一个基本的数据结构，内部变量并不复杂，首先是data指针，采用的是boost库的一个智能指针shared_ptr。这一部分主要用来申请内存存储data用于正向传播。diff_用来存储误差，shape_data 和 shape_都是用来存储Blob的形状，只是一个是老版本，一个是新版本，count表示Blob中的元素个数，也就是 num * channel * height * width，capacity表示当前元素的个数。
caffe基于blob进行存储和交换数据，为了便于优化，Blob提供了统一的内存接口存储某种类型的数据，可以根据cpu到gpu的同步需要，屏蔽cpu/gpu在计算上的开销。

*2. 主要函数*

template <typename Dtype>class Blob { public:  Blob()       : data_(), diff_(), count_(0), capacity_(0) {}  /// @brief Deprecated; use <code>Blob(const vector<int>& shape)</code>.  explicit Blob(const int num, const int channels, const int height,      const int width);  explicit Blob(const vector<int>& shape);  /// @brief Deprecated; use <code>Reshape(const vector<int>& shape)</code>.  void Reshape(const int num, const int channels, const int height,      const int width);

Blob作为一个基础的数据结构，其中构造函数开辟了一个内存空间来存储数据，Reshape()函数在layer中的reshape和forward中来adjust dimension，同时会改变Blob的大小，此时内存将会被重新分配。
blob中的index是可以从负坐标开始读的，这一点类似于python，对于blob中的四个基本变量，num channel height width，可以直接通过shape(0), shape(1), shape(2), shape(3)。

计算offset

inline int offset(const int n, const int c = 0, const int h = 0, const int w = 0)inline int offset(const vector<int>& indices)

offset计算的方式也支持两种方式，直接指定n,c,h,w或者放到一个vector中进行计算，偏差是根据对应的n,c,h,w，返回的offset是

((n * channels() + c) * height() + h) * width() + w

从一个blob中copy数据 ，通过开关控制是否copy_diff,如果是False则copy data。reshape控制是否需要reshape。

inline Dtype data_at(const int n, const int c, const int h, const int w)inline Dtype diff_at(const int n, const int c, const int h, const int w)inline Dtype data_at(const vector<int>& index)inline Dtype diff_at(const vector<int>& index)inline const shared_ptr<SyncedMemory>& data()inline const shared_ptr<SyncedMemory>& diff()

这一部分函数主要通过给定的位置访问数据，根据位置计算与数据起始的偏差offset，在通过cpu_data*指针获得地址。

const Dtype* cpu_data() const;void set_cpu_data(Dtype* data);const int* gpu_shape() const;const Dtype* gpu_data() const;const Dtype* cpu_diff() const;const Dtype* gpu_diff() const;Dtype* mutable_cpu_data();Dtype* mutable_gpu_data();Dtype* mutable_cpu_diff();Dtype* mutable_gpu_diff();

data就是存储前向传递的信息的数据，diff指的是神经网络在反向传播时候的梯度。

void FromProto(const BlobProto& proto, bool reshape = true);void ToProto(BlobProto* proto, bool write_diff = false) const;

这两个函数主要是将数据序列化，存储到BlobProto，这里说到Proto是谷歌的一个数据序列化的存储格式，可以实现语言、平台无关、可扩展的序列化结构数据格式。Caffe里面数据的存储都采用这一结构.
总结一下Blob.hpp的一些函数：
Reshape()可以改变一个Blob的大小；
ReshapeLike()为data和diff重新分配了一块空间，大小和另一个Blob一样。
Num_axes()返回的是Blob的大小
Count()计算得到 num * channels * height * width
Offset() 可得到输入Blob数据的(n,c,h,w)的偏移量位置
CopyFrom()从source拷贝数据，copy_diff来作为标志区分拷贝的是data还是diff。
FromProto()从proto读数据进俩，其实就是反序列化
ToProto()把blob数据保存到proto中
ShareData() / ShareDiff() 从other的blob复制data和diff的值

#ifndef CAFFE_BLOB_HPP_#define CAFFE_BLOB_HPP_#include <algorithm>#include <string>#include <vector>#include "caffe/common.hpp"  //单例化caffe类，并且封装了boost和cuda随机数生成的函数，提供了统一接口#include "caffe/proto/caffe.pb.h"#include "caffe/syncedmem.hpp"/*主要是分配内存和释放内存。class yncedMemory定义了内存分配管理和CPU与GPU之间同步的函数。 Blob会使用SyncedMem自动决定什么时候去copy data以提高运行效率，通常情况是仅当gnu或cpu修改后有copy操作。 */const int kMaxBlobAxes = 32; //在头文件中为它添加 extern 声明,以使其能被多个文件共享namespace caffe {/** * @brief A wrapper around SyncedMemory holders serving as the basic *        computational unit through which Layer%s, Net%s, and Solver%s *        interact. * * TODO(dox): more thorough description. */template <typename Dtype>   //模板类，虚拟类型Dtypeclass Blob { public:  Blob() //构造函数：初始化列表 {空函数体}       : data_(), diff_(), count_(0), capacity_(0) {}  //当构造函数被声明 explicit 时,编译器将不使用它作为转换操作符。  /// @brief Deprecated; use <code>Blob(const vector<int>& shape)</code>.  explicit Blob(const int num, const int channels, const int height,      const int width);   //可以通过设置数据维度（N,C,H,W）初始化  //const 传递过来的参数在函数内不可以改变(无意义，因为本身就是形参)  //const引用参数在函数内为常量不可变  explicit Blob(const vector<int>& shape); //也可以通过传入vector<int>直接传入维数  /// @brief Deprecated; use <code>Reshape(const vector<int>& shape)</code>.  void Reshape(const int num, const int channels, const int height,      const int width);  /**   * @brief Change the dimensions of the blob, allocating new memory if   *        necessary.   *   * This function can be called both to create an initial allocation   * of memory, and to adjust the dimensions of a top blob during Layer::Reshape   * or Layer::Forward. When changing the size of blob, memory will only be   * reallocated if sufficient memory does not already exist, and excess memory   * will never be freed.   *   * Note that reshaping an input blob and immediately calling Net::Backward is   *   * an error; either Net::Forward or Net::Reshape need to be called to   * propagate the new input shape to higher layers.   */  void Reshape(const vector<int>& shape);  void Reshape(const BlobShape& shape);  void ReshapeLike(const Blob& other);  //内联函数 通过内联函数，编译器不需要跳转到内存其他地址去执行函数调用，也不需要保留函数调用时的现场数据。  // const 成员函数，任何不会修改数据成员的函数都应该声明为const 类型。  // 输出blob的形状  inline string shape_string() const { //    ostringstream stream;    for (int i = 0; i < shape_.size(); ++i) {      stream << shape_[i] << " ";    }    stream << "(" << count_ << ")";    return stream.str();  }  inline const vector<int>& shape() const { return shape_; }  /**   * @brief Returns the dimension of the index-th axis (or the negative index-th   *        axis from the end, if index is negative).   *   * @param index the axis index, which may be negative as it will be   *        "canonicalized" using CanonicalAxisIndex.   *        Dies on out of range index.   */  inline int shape(int index) const {  //根据索引返回维数，对于维数(N,C,H,W),shape(0)返回N,shape(-1)返回W。    return shape_[CanonicalAxisIndex(index)];  }  inline int num_axes() const { return shape_.size(); }    //返回Blob维度数，对于维数(N,C,H,W)，返回4  inline int count() const { return count_; }               //返回Blob维度数，对于维数(N,C,H,W)，返回N×C×H×W  /**   * @brief Compute the volume of a slice; i.e., the product of dimensions   *        among a range of axes.   *   * @param start_axis The first axis to include in the slice.   *   * @param end_axis The first axis to exclude from the slice.   */  //对于维数(N,C,H,W)，count(0, 3)返回N×C×H  inline int count(int start_axis, int end_axis) const {    CHECK_LE(start_axis, end_axis);    CHECK_GE(start_axis, 0);    CHECK_GE(end_axis, 0);    CHECK_LE(start_axis, num_axes());    CHECK_LE(end_axis, num_axes());    int count = 1;    for (int i = start_axis; i < end_axis; ++i) {      count *= shape(i);    }    return count;  }  /**   * @brief Compute the volume of a slice spanning from a particular first   *        axis to the final axis.   *   * @param start_axis The first axis to include in the slice.   */  //对于维数(N,C,H,W)，count(1)返回C×H×W  inline int count(int start_axis) const {    return count(start_axis, num_axes());  }  /**   * @brief Returns the 'canonical' version of a (usually) user-specified axis,   *        allowing for negative indexing (e.g., -1 for the last axis).   *   * @param axis_index the axis index.   *        If 0 <= index < num_axes(), return index.   *        If -num_axes <= index <= -1, return (num_axes() - (-index)),   *        e.g., the last axis index (num_axes() - 1) if index == -1,   *        the second to last if index == -2, etc.   *        Dies on out of range index.   */  inline int CanonicalAxisIndex(int axis_index) const {    CHECK_GE(axis_index, -num_axes())        << "axis " << axis_index << " out of range for " << num_axes()        << "-D Blob with shape " << shape_string();    CHECK_LT(axis_index, num_axes())        << "axis " << axis_index << " out of range for " << num_axes()        << "-D Blob with shape " << shape_string();    if (axis_index < 0) {      return axis_index + num_axes();    }    return axis_index;  }  /// @brief Deprecated legacy shape accessor num: use shape(0) instead.  inline int num() const { return LegacyShape(0); }  /// @brief Deprecated legacy shape accessor channels: use shape(1) instead.  inline int channels() const { return LegacyShape(1); }  /// @brief Deprecated legacy shape accessor height: use shape(2) instead.  inline int height() const { return LegacyShape(2); }  /// @brief Deprecated legacy shape accessor width: use shape(3) instead.  inline int width() const { return LegacyShape(3); }  inline int LegacyShape(int index) const {    CHECK_LE(num_axes(), 4)        << "Cannot use legacy accessors on Blobs with > 4 axes.";    CHECK_LT(index, 4);    CHECK_GE(index, -4);    if (index >= num_axes() || index < -num_axes()) {      // Axis is out of range, but still in [0, 3] (or [-4, -1] for reverse      // indexing) -- this special case simulates the one-padding used to fill      // extraneous axes of legacy blobs.      return 1;    }    return shape(index);  }  inline int offset(const int n, const int c = 0, const int h = 0,      const int w = 0) const {    //计算物理偏移量，(n,c,h,w)的偏移量为((n∗C+c)∗H+h)∗W+w    CHECK_GE(n, 0);    CHECK_LE(n, num());    CHECK_GE(channels(), 0);    CHECK_LE(c, channels());    CHECK_GE(height(), 0);    CHECK_LE(h, height());    CHECK_GE(width(), 0);    CHECK_LE(w, width());    return ((n * channels() + c) * height() + h) * width() + w;  }  inline int offset(const vector<int>& indices) const {    CHECK_LE(indices.size(), num_axes());    int offset = 0;    for (int i = 0; i < num_axes(); ++i) {      offset *= shape(i);      if (indices.size() > i) {        CHECK_GE(indices[i], 0);        CHECK_LT(indices[i], shape(i));        offset += indices[i];      }    }    return offset;  }  /**   * @brief Copy from a source Blob.   *   * @param source the Blob to copy from   * @param copy_diff: if false, copy the data; if true, copy the diff   * @param reshape: if false, require this Blob to be pre-shaped to the shape   *        of other (and die otherwise); if true, Reshape this Blob to other's   *        shape if necessary   */  void CopyFrom(const Blob<Dtype>& source, bool copy_diff = false,   //从source拷贝数据， copy_diff来作为标志区分是拷贝data还是diff。      bool reshape = false);  inline Dtype data_at(const int n, const int c, const int h,      const int w) const {    return cpu_data()[offset(n, c, h, w)];  }  inline Dtype diff_at(const int n, const int c, const int h,      const int w) const {    return cpu_diff()[offset(n, c, h, w)];  }  inline Dtype data_at(const vector<int>& index) const {    return cpu_data()[offset(index)];  }  inline Dtype diff_at(const vector<int>& index) const {    return cpu_diff()[offset(index)];  }  inline const shared_ptr<SyncedMemory>& data() const {    CHECK(data_);    return data_;  }  inline const shared_ptr<SyncedMemory>& diff() const {    CHECK(diff_);    return diff_;  }\  /*    // 假定数据在 CPU 上进行初始化，我们有一个 blob    const Dtype* foo;    Dtype* bar;    foo = blob.gpu_data(); // 数据从 CPU 复制到 GPU    foo = blob.cpu_data(); // 没有数据复制，两者都有最新的内容    bar = blob.mutable_gpu_data(); // 没有数据复制    // ... 一些操作 ...    bar = blob.mutable_gpu_data(); // 仍在 GPU，没有数据复制    foo = blob.cpu_data(); // 由于 GPU 修改了数值，数据从 GPU 复制到 CPU    foo = blob.gpu_data(); //没有数据复制，两者都有最新的内容    bar = blob.mutable_cpu_data(); // 依旧没有数据复制    bar = blob.mutable_gpu_data(); //数据从 CPU 复制到 GPU    bar = blob.mutable_cpu_data(); //数据从 GPU 复制到 CPU   */  const Dtype* cpu_data() const;  //数据访问，const方式只读，不允许改写数据  void set_cpu_data(Dtype* data);  const int* gpu_shape() const;  const Dtype* gpu_data() const;  const Dtype* cpu_diff() const;  const Dtype* gpu_diff() const;  Dtype* mutable_cpu_data();      //mutable方式可改写数据（对diff_的访问也是类似的）  Dtype* mutable_gpu_data();  Dtype* mutable_cpu_diff();  Dtype* mutable_gpu_diff();  void Update();  void FromProto(const BlobProto& proto, bool reshape = true); //从proto读数据进来，其实就是反序列化  void ToProto(BlobProto* proto, bool write_diff = false) const; //blob数据保存到proto中  /// @brief Compute the sum of absolute values (L1 norm) of the data.  Dtype asum_data() const;  /// @brief Compute the sum of absolute values (L1 norm) of the diff.  Dtype asum_diff() const;  /// @brief Compute the sum of squares (L2 norm squared) of the data.  Dtype sumsq_data() const;  /// @brief Compute the sum of squares (L2 norm squared) of the diff.  Dtype sumsq_diff() const;  /// @brief Scale the blob data by a constant factor.  void scale_data(Dtype scale_factor);  /// @brief Scale the blob diff by a constant factor.  void scale_diff(Dtype scale_factor);  /**   * @brief Set the data_ shared_ptr to point to the SyncedMemory holding the   *        data_ of Blob other -- useful in Layer%s which simply perform a copy   *        in their Forward pass.   *   * This deallocates the SyncedMemory holding this Blob's data_, as   * shared_ptr calls its destructor when reset with the "=" operator.   */  void ShareData(const Blob& other); //Blob& other 赋值给data_  /**   * @brief Set the diff_ shared_ptr to point to the SyncedMemory holding the   *        diff_ of Blob other -- useful in Layer%s which simply perform a copy   *        in their Forward pass.   *   * This deallocates the SyncedMemory holding this Blob's diff_, as   * shared_ptr calls its destructor when reset with the "=" operator.   */  void ShareDiff(const Blob& other); //Blob& other 赋值给diff_  bool ShapeEquals(const BlobProto& other); protected:  shared_ptr<SyncedMemory> data_; //存储前向传递数据  shared_ptr<SyncedMemory> diff_; //存储反向传递梯度  shared_ptr<SyncedMemory> shape_data_;  vector<int> shape_;  //参数维度  int count_; //Blob存储的元素个数（shape_所有元素乘积）  int capacity_;//当前Blob的元素个数（控制动态分配）  DISABLE_COPY_AND_ASSIGN(Blob);};  // class Blob}  // namespace caffe#endif  // CAFFE_BLOB_HPP_