本征矩阵的就地元素类型转换

Question

I want to cast an integer matrix to a floating point matrix, such that:

1. data is not copied.
2. no new memory is allocated
3. the new floating point view of the data is mutable.

newest try:

#include "Eigen/Eigen"
#include <iostream>
int main(){
    using namespace Eigen;
    MatrixXi x(3,3);
    x.fill(0);
    double* ptr_x = (double*)(x.data());
    Map<MatrixXd> y(ptr_x, x.rows(), x.cols());
    x(0,0) = 100;
    y = x.cast<double>();
    y(1,1) = 0.5f;
    std::cout << y << "\n";
}

Run time error:

a.out: malloc.c:2405: 

    sysmalloc: 
        Assertion `(old_top == initial_top (av) && old_size == 0) 
        || ((unsigned long) (old_size) >= MINSIZE 
        && prev_inuse (old_top) 
        && ((unsigned long) old_end & (pagesize - 1)) == 0)' 
    failed.

Aborted (core dumped)

The following does not compile:

#include "Eigen/Eigen"
#include <stdint.h>

#include <iostream>
int main(){
    using namespace Eigen;
    MatrixXi x(3,3);
    x.fill(0);
    float* ptr_x = (float*)(x.data());

    Map<MatrixXd> y(ptr_x, x.rows(), x.cols());
    x(0,0) = 100;
    y(1,1) = 0.5f;
    y = x.cast<float>();
    std::cout << y << "\n";
}

I guess the CastXpr<NewType>::Type might work ( documentation ). But I cant figure out how to use that.

The CastXpr seems to be an unaryExpr : How do I in-place modify each element of a 1D Array?

Answer 1

First of all, you cannot cast it into a double (without allocating more space) matrix since a double is 8 bytes in memory while an int is 4 bytes.

I think you can simply cast a float pointer to the original Matrix. Following code worked for me.

#include "Eigen/Eigen"
#include <iostream>
int main()
{
    using namespace Eigen;
    MatrixXi x(3, 3);
    x.fill(2);
    MatrixXf* float_ptr = (MatrixXf*) &x;
    float_ptr->operator()(2,2) = 42.1f;

    std::cout << "float cast(2,2): " << 
        float_ptr->operator()(2, 2) << std::endl;
    std::cout << "float cast(1,1): " <<
        float_ptr->operator()(1, 1) << std::endl;
    std::cout << "x(1,1): " << x(1, 1) << std::endl;
}

output:

float cast(2,2): 42.1
float cast(1,1): 2.8026e-45 (which is ~2)
x(1,1): 2
Press any key to continue . . .

So... As long as you use this pointer, the allocated object will act as a floating point matrix but please keep that in mind that you cannot use 'x' as if it's a floating point matrix since any function call using 'x' will cause the allocated memory to be interpreted as an integer matrix

For example: since we have changed the original (2,2) from int to a float, if you try to retrieve it with 'x' you will see something like this.

    .
    .
    .
    std::cout << "float cast(2,2): " << 
        float_ptr->operator()(2, 2) << std::endl;
    std::cout << "x(2,2): " << x(2, 2) << std::endl;

the output:

float cast(2,2): 42.1
x(2,2): 1109943910

Answer 2

After fixing the types, the following finally works.

Program:

#include "Eigen/Eigen"
#include <iostream>
int main(){
    using namespace Eigen;
    MatrixXi x(3,3);
    x.fill(0);
    float* ptr_x = (float*)(x.data());

    Map<MatrixXf> y(ptr_x, x.rows(), x.cols());

    x << 1,2,3,4,5,6,7,8,9;
    y = x.cast<float>();
    y /= 2;
    std::cout << y << "\n";
}

Result:

0.5   1 1.5
  2 2.5   3
3.5   4 4.5