为什么 Eigen 矩阵到 C 数组的转换会为前两个索引提供垃圾值?

Why is Eigen matrix to C array conversion giving garbage values for the first two indices?

我有一个要转换为 C 数组的特征矩阵。我可以用下面的例子重现这个问题。

#include <iostream>
#include <Eigen/Core>

int *test()
{
    Eigen::MatrixXi arr = Eigen::MatrixXi::Ones(6,1);
    // just to check
    arr(4)=3;
    arr(5)=19;
    return arr.data();
}

int main()
{
    int *c_arr;
    c_arr = test();

    for (int i=0; i<6;++i)
    {
        std::cout << c_arr[i] << std::endl;
    }

    return 0;
}

输出:

0
0
1
1
3
19

现在,如果我从 test 函数中打印转换后的 C 数组值,则这些值是正确的。但是,如果我打印 main 中的值(如上所示),前两个索引总是垃圾。所以我想知道函数调用中发生了什么?我用不同的特征矩阵(类型、大小)尝试过这个,我得到了相同的结果。

首先我会说我不是 100% 熟悉 Eigen 库(只是出于好奇下载它来查看它)并且它的文档有点缺乏但是你的问题是一个基本的 C 问题可以通过几种方式补救。

首先,我们将首先解释您的代码中发生了什么以提供垃圾值:

int *test()
{
    /* create an auto scoped variable on the stack;
       this variable is only "visible" to this function
       and any references to it or it's underlying data
       outside the scope of this function will result
       in "undefined behaviour" */
    Eigen::MatrixXi arr = Eigen::MatrixXi::Ones(6,1);
    arr(4)=3;
    arr(5)=19;
    /* arr.data() is defined as returning a pointer to the scalar underlying type (or
    a C-style array in other words). Regardless of the type being returned, it is pointer based
    and you are returning a pointer to a location in memory, not the actual data being held in
    the memory. */
    return arr.data();
} /* the variable arr is destroyed here since we left function scope and the return value (the pointer location)
is put in the return register and "program flow" is returned back to the main function where the pointer being
returned now points to "invalid" memory */

int main()
{
    int *c_arr; // create a pointer type that can reference int types
    c_arr = test(); // point it to the result of the test function (see notes above)
    /* c_arr now points to a memory location returned from test, but since the
    arr variable no longer exists here, when you go through and print the values pointed
    to at those memory locations you will get what is at those locations and could be "anything"
    except a valid reference to the original arr variable and it's underlying data. */

    for (int i=0; i<6;++i)
    {
        std::cout << c_arr[i] << std::endl;
    }

    return 0;
}

这就是原因,至于如何解决它,有几种方法可以解决您的问题;一种是将 return 数组作为变量传递给 test 函数(例如 void test(int*& val)),然后您可以选择为测试函数中的变量分配新内存,或者假设用户已经这样做了,并且还必须假设用户会自己清理并调用 delete[](不只是 delete,因为您正在对数据数组进行操作)。

但这有很多注意事项,需要知道要分配多少 space 并确保在完成后释放。我不确定为什么你特别需要一个 C 风格的数组,但由于你使用的是 C++,如果你使用一些可用的 STL 和容器函数来帮助你,可能会更谨慎,例如:

#include <iostream>
#include <vector>
#include <Eigen/Core>

std::vector<int> test()
{
    Eigen::MatrixXi arr = Eigen::MatrixXi::Ones(6,1);
    arr(4)=3;
    arr(5)=19;
    // we need the size so we know how big of a container to allocate
    std::size_t sz = arr.innerSize() * arr.outerSize();
    std::vector<int> ret(sz);
    // get a temporary C array pointer so we can reference the data
    int* tmp = arr.data();
    // copy from tmp[0] to tmp[sz] and insert the data into the first element of ret
    std::copy(tmp, tmp+sz, ret.begin());
    // return the (copied) data
    return ret;
}

int main()
{
    std::vector<int> c_arr = test();
    // c_arr now points to valid data it holds and can be iterated on
    for (std::size_t i = 0; i < c_arr.size(); ++i) {
        std::cout << c_arr[i] << std::endl;
    }
    // if you need a C-style array from here, you can easily copy the data
    // from the vector to your C-array
    return 0;
}

我研究了使用 class 的 cast() 函数,但无法完全弄清楚语法,以使其比仅以上述方式复制它更不痛苦,因为它看起来像你' d 必须将 cast 函数调用到不同的 Eigen 类型,然后从那里再次转换,但知道有一个 cast 函数和其他方法来获取 [=] 的基础数据20=] classes 如果你需要访问它。

希望能帮到你。