CODEWITHSUNDEEP
csegmentation-faultfftw
FacundoGFlores
FacundoGFlores

Reputation: 8118

Using fftw3 inside a function produce segmentation fault

Testing the following code I get the right answers:

#include <fftw3.h>
void dump_vector(float *a, int size)
{
    int i;
    for (i = 0; i < size; i++) {
        printf("%f\t", a[i]);
    }
    printf("\n");
}

int main()
{
    float a[] = {1, 2, 3, 4};
    dump_vector(a, 4);
    fftw_plan plan = fftw_plan_r2r_1d(2 * 2, a, a, FFTW_REDFT10, FFTW_ESTIMATE);
    fftw_execute(plan);
    dump_vector(a, 4);
}

Result:

./main
1.000000    2.000000    3.000000    4.000000    
3.992188    4.501953    -334878449985782808576.000000   3.886292

But I want make a naive wrap for dct so:

#include <fftw3.h>
void dump_vector(float *a, int size)
{
    int i;
    for (i = 0; i < size; i++) {
        printf("%f\t", a[i]);
    }
    printf("\n");
}

void dct(float * array, int xsize, int ysize)
{
    fftw_plan plan = fftw_plan_r2r_1d(xsize * ysize, array, array, FFTW_REDFT10, FFTW_ESTIMATE);
    fftw_execute(plan);
}


int main()
{
    float a[] = {1, 2, 3, 4};
    dump_vector(a, 4);
    dct(a, 2, 2);
    dump_vector(a, 4);
}

Result:

./main
1.000000    2.000000    3.000000    4.000000    
3.992188    4.501953    -334878449985782808576.000000   3.886292    
[1]    10880 segmentation fault  ./main

Is there something that I am missing?

I need to do this because then I want to generate a .so file for using it in another app.

Upvotes: 1

Views: 527

Answers (1)

francis
francis

Reputation: 9817

The problem is much simplier than expected: fftw_plan_r2r_1d perform DCTs/DSTs on arrays of double.

As is, the array float a[] is too small: 4 float are stored in 4x4=16 bytes, while fftw_plan_r2r_1d expects 4 double (32 bytes). Hence, fftw_plan_r2r_1d will attempt to read pass the end of the actual array. This triggers undefined behaviors such as segmentation fault or wrong results.

Two solutions:

  • Change float a[] for double a[] and dump_vector(float *a,... for dump_vector(double *a,...
  • Use fftwf_plan_r2r_1d() to build a fftwf_plan for arrays of float. See http://www.fftw.org/doc/Precision.html : the library libfftw3f.a must be linked by adding -lfftw3f to the compiler command.

Upvotes: 1

Related Questions