C ++中的整数的三维数组

C ++中的整数的三维数组

Three dimensional arrays of integers in C++

我想找出在C ++中使用指针算术/动态内存分配,或者使用STL技术(例如向量)来实现整数的三维数组的安全方法。

本质上,我希望整数数组的尺寸看起来像:

1
[ x ][ y ][ z ]

x和y的范围是20-6000
z是已知的,等于4。


看一下Boost多维数组库。这是一个示例(改编自Boost文档):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
#include"boost/multi_array.hpp"

int main() {
  // Create a 3D array that is 20 x 30 x 4
  int x = 20;
  int y = 30;
  int z = 4;

  typedef boost::multi_array<int, 3> array_type;
  typedef array_type::index index;
  array_type my_array(boost::extents[x][y][z]);

  // Assign values to the elements
  int values = 0;
  for (index i = 0; i != x; ++i) {
    for (index j = 0; j != y; ++j) {
      for (index k = 0; k != z; ++k) {
        my_array[i][j][k] = values++;
      }
    }
  }
}

以下是在每个阵列的一块内存中使用C或C ++创建3D阵列的一种直接方法。无需使用BOOST(即使它很好),也无需在具有多个间接访问的行之间分配分配(这很糟糕,因为在访问数据时通常会带来很大的性能损失,并且会碎片化内存)。

唯一要了解的是,没有多维数组,只有数组(数组)的数组。最里面的索引是内存中最远的。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
#include <stdio.h>
#include <stdlib.h>

int main(){

    {
        // C Style Static 3D Arrays
        int a[10][20][30];
        a[9][19][29] = 10;
        printf("a[9][19][29]=%d\
"
, a[9][19][29]);
    }

    {
        // C Style dynamic 3D Arrays
        int (*a)[20][30];
        a = (int (*)[20][30])malloc(10*20*30*sizeof(int));
        a[9][19][29] = 10;
        printf("a[9][19][29]=%d\
"
, a[9][19][29]);
        free(a);
    }

    {
        // C++ Style dynamic 3D Arrays
        int (*a)[20][30];
        a = new int[10][20][30];
        a[9][19][29] = 10;
        printf("a[9][19][29]=%d\
"
, a[9][19][29]);
        delete [] a;
    }

}

对于您的实际问题,由于可能存在两个未知维度,因此我的建议存在一个问题,因为它仅允许一个未知维度。有几种方法可以解决这个问题。

好消息是,现在使用变量可与C一起使用,称为变量长度数组。您在这里查看详细信息。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
    int x = 100;
    int y = 200;
    int z = 30;

    {
        // C Style Static 3D Arrays
        int a[x][y][z];
        a[99][199][29] = 10;
        printf("a[99][199][29]=%d\
"
, a[99][199][29]);
    }

    {
        // C Style dynamic 3D Arrays
        int (*a)[y][z];
        a = (int (*)[y][z])malloc(x*y*z*sizeof(int));
        a[99][199][29] = 10;
        printf("a[99][199][29]=%d\
"
, a[99][199][29]);
        free(a);
    }

如果使用C ++,最简单的方法可能是使用运算符重载来坚持数组语法:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
    {
        class ThreeDArray {
            class InnerTwoDArray {
                int * data;
                size_t y;
                size_t z;
                public:
                InnerTwoDArray(int * data, size_t y, size_t z)
                    : data(data), y(y), z(z) {}

                public:
                int * operator [](size_t y){ return data + y*z; }
            };

            int * data;
            size_t x;
            size_t y;
            size_t z;
            public:
            ThreeDArray(size_t x, size_t y, size_t z) : x(x), y(y), z(z) {
                data = (int*)malloc(x*y*z*sizeof data);
            }

            ~ThreeDArray(){ free(data); }

            InnerTwoDArray operator [](size_t x){
                return InnerTwoDArray(data + x*y*z, y, z);
            }
        };

        ThreeDArray a(x, y, z);
        a[99][199][29] = 10;
        printf("a[99][199][29]=%d\
"
, a[99][199][29]);
    }

上面的代码在访问InnerTwoDArray时有一些间接开销(但是好的编译器可能会对其进行优化),但是仅使用一个内存块来分配在堆上的数组。通常这是最有效的选择。

显然,即使上面的代码仍然简单明了,STL或BOOST也能很好地完成,因此不需要重新发明轮子。我仍然相信知道它可以轻松完成很有趣。


每对方括号都是解引用操作(应用于指针时)。例如,以下几对代码行是等效的:

1
2
x = myArray[4];
x = *(myArray+4);

1
2
x = myArray[2][7];
x = *((*(myArray+2))+7);

要使用建议的语法,您只需对第一个取消引用返回的值进行取消引用。

1
2
3
4
5
6
7
8
9
int*** myArray = (some allocation method, keep reading);
//
// All in one line:
int   value = myArray[x][y][z];
//
// Separated to multiple steps:
int** deref1 = myArray[x];
int*  deref2 = deref1[y];
int   value = deref2[z];

要分配此数组,您只需要认识到实际上没有三维整数数组即可。您有一个整数数组的数组。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
// Start by allocating an array for array of arrays
int*** myArray = new int**[X_MAXIMUM];

// Allocate an array for each element of the first array
for(int x = 0; x < X_MAXIMUM; ++x)
{
    myArray[x] = new int*[Y_MAXIMUM];

    // Allocate an array of integers for each element of this array
    for(int y = 0; y < Y_MAXIMUM; ++y)
    {
        myArray[x][y] = new int[Z_MAXIMUM];

        // Specify an initial value (if desired)
        for(int z = 0; z < Z_MAXIMUM; ++z)
        {
            myArray[x][y][z] = -1;
        }
    }
}

取消分配此数组与分配它类似的过程:

1
2
3
4
5
6
7
8
9
10
11
for(int x = 0; x < X_MAXIMUM; ++x)
{
    for(int y = 0; y < Y_MAXIMUM; ++y)
    {
        delete[] myArray[x][y];
    }

    delete[] myArray[x];
}

delete[] myArray;

与向量:

1
std::vector< std::vector< std::vector< int > > > array3d;

如果已经添加了元素,则可以使用array3d [x] [y] [z]访问每个元素。 (例如,通过push_back)


与使用new / delete相比,使用STL管理内存有很多优点。如何表示数据的选择取决于您计划如何使用它。一个建议是隐藏执行决策并为一维STL向量提供三维获取/设置方法的类。

如果您确实认为需要创建自定义3d矢量类型,请先研究Boost。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
// a class that does something in 3 dimensions

class MySimpleClass
{
public:

  MySimpleClass(const size_t inWidth, const size_t inHeight, const size_t inDepth) :
   mWidth(inWidth), mHeight(inHeight), mDepth(inDepth)
   {
       mArray.resize(mWidth * mHeight * mDepth);
   }


  // inline for speed
  int Get(const size_t inX, const size_t inY, const size_t inZ) {
     return mArray[(inZ * mWidth * mHeight) + (mY * mWidth) + mX];
  }

  void Set(const size_t inX, const size_t inY, const size_t inZ, const int inVal) {
     return mArray[(inZ * mWidth * mHeight) + (mY * mWidth) + mX];
  }

  // doing something uniform with the data is easier if it's not a vector of vectors
  void DoSomething()
  {
     std::transform(mArray.begin(), mArray.end(), mArray.begin(), MyUnaryFunc);
  }

private:

  // dimensions of data
  size_t mWidth;
  size_t mHeight;
  size_t mDepth;

  // data buffer
  std::vector< int > mArray;
};

应该注意的是,出于所有意图和目的,您只处理2D数组,因为已知第三(也是最低有效)维度。

如果您事先不知道数组每个维度中将有多少个条目,那么使用STL或Boost是非常好的方法,因为它们将为您提供动态内存分配,如果您的数据集是保持基本上是静态的,或者大多数情况下仅接收新条目而不删除很多条目。

但是,如果您事先了解数据集的知识,例如总共将存储大约多少个项目,或者如果要稀疏地填充数组,则最好使用某种哈希/存储桶函数,并使用XYZ索引作为键。在这种情况下,假设每个维度的条目数不超过8192(13位),则可以使用40位(5字节)的密钥来获得。或者,假设始终有4 x Z个条目,则只需使用26位XY键即可。这是速度,内存使用和动态分配之间更有效的折衷之一。


Pieter的建议当然很好,但是您要记住的一件事是,在构建大型阵列的情况下,它可能会非常缓慢。每次向量容量更改时,都必须复制所有数据(向量的n个向量)。


推荐阅读