HTK代码阅读之Vector/Matrix 内存管理

 

Vector/Matrix 内存管理

/* ------------- Vector/Matrix Memory Management -------------- */

/*
Vector的实际数据都是存储在都是从1到n的下标，下标0记录此vector有多少个元素。需要多申请一个元素的字节大小。
Matrix本质上是一块连续内存（和本人原来分配矩阵内存方式不太一样）。首先申请所有需要的内存大小size.然后从0到第R个元素可以认为是一个vector，第0个元素记录该矩阵一共有几行，第1到R个元素用于保存第1行到第R行的起始地址,每行都有C（列数）+1个元素。所以一个Matrix总共需要的内存size=(R+1)*sizeof(Ptr)+R* vectorElemSize(C)
Shared vectors and matrices have an extra 2 * sizeof(Ptr) bytes 
  prepended to hold a usage count and a hook.
*/

Matrix内存结构示意

TriMatrix内存结构示意

sdf

所需内存R*sizeof(float)+C*sizeof(float)+sizeof(float)*C(C+1)/2

跟代码表述不同，含义相同。请参考代码阅读理解。

 

 /*Vectors are pointers to arrays of float (ie float*); matrices are pointers

  to an array of vectors (ie float**).  All indexing is v[1..n] and

  m[1..r][1..c].  The actual size of each vector is stored in v[0].  For

  matrices, the row lengths (number of columns) are stored in every

  row since they are genuine vectors, the number of rows is stored in m[0].

  Triangular matrices are the same except that the rows increase in length

  with the first row being a vector of one element.

  Short and IntVecs are arranged in a similar way to Vectors.

 

  Shared vectors and matrices have an extra 2 * sizeof(Ptr) bytes

  prepended to hold a usage count and a hook.

*/

/*下面类型函数：申请size个元素的vector总共需要多大空间*/

/* EXPORT->vectorElemSize: size of vectors for creating heaps */

size_t ShortVecElemSize(int size) { return (size+1)*sizeof(short); }

size_t IntVecElemSize(int size) { return (size+1)*sizeof(int); }

size_t VectorElemSize(int size) { return (size+1)*sizeof(float); }

size_t DVectorElemSize(int size){ return (size+1)*sizeof(double);}

size_t SVectorElemSize(int size){ return (size+1)*sizeof(float)+2*sizeof(Ptr); }*/

 

后面的不一一列举了