Test for faad_imdct issue

gcc-test1.c

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

typedef unsigned short uint16_t;
float constdata[1024];

inline static void ComplexMult(float *y1, float *y2, float x1, float x2, float c1, float c2)
{
    *y1 = (x1 * c1) + (x2 * c2);
    *y2 = (x2 * c1) - (x1 * c2);
}

// incorrect behavior on -O3
// (implicit cast of N2 to int causes vectorization to produce wrong results)
void __attribute__((noinline)) imdct(unsigned int N, float *xin, float *xout)
{
    float Z1[1024];

    uint16_t N2 = N >> 1;
    uint16_t N4 = N >> 2;

    for (uint16_t k = 0; k < N4; k++)
        ComplexMult(&Z1[k * 2 + 1], &Z1[k * 2 + 0], xin[k * 2], xin[N2 - 1 - k * 2], constdata[k * 2 + 0], constdata[k * 2 + 1]);

    for (uint16_t k = 0; k < N4; k++)
        ComplexMult(&xout[k * 2], &xout[k * 2 + 1], Z1[k * 2 + 1], Z1[k * 2 + 0], constdata[k * 2 + 0], constdata[k * 2 + 1]);
}

// correct behavior on -O3
void __attribute__((noinline)) imdct2(unsigned int N, float *xin, float *xout)
{
    float Z1[1024];

    uint16_t N2 = N >> 1;
    uint16_t N4 = N >> 2;

    for (uint16_t k = 0; k < N4; k++)
        ComplexMult(&Z1[k * 2 + 1], &Z1[k * 2 + 0], xin[k * 2], xin[(unsigned int)N2 - 1 - k * 2], constdata[k * 2 + 0], constdata[k * 2 + 1]);

    for (uint16_t k = 0; k < N4; k++)
        ComplexMult(&xout[k * 2], &xout[k * 2 + 1], Z1[k * 2 + 1], Z1[k * 2 + 0], constdata[k * 2 + 0], constdata[k * 2 + 1]);
}

// test bed to compare results of the two functions
int main(int argc, char *argv[])
{
    float xin[1024], xout[1024], xout2[1024];

    for (unsigned int i = 0; i < 1024; i++)
    {
        constdata[i] = 1;
        xin[i] = 1;
    }
    memset(xout, 0, sizeof(xout));
    imdct(2048, xin, xout);

    for (unsigned int i = 0; i < 1024; i++)
    {
        constdata[i] = 1;
        xin[i] = 1;
    }
    memset(xout2, 0, sizeof(xout2));
    imdct2(2048, xin, xout2);

    if (memcmp(xout, xout2, sizeof(xout)) == 0)
    {
        printf("pass\n");
    }
    else
    {
        printf("fail\n");

        for (unsigned int i = 0; i < 16; i++)
            printf("%f = %f\n", xout[i], xout2[i]);
    }
    return 0;
}