ARM Neon optimisation does not change performance

[nicastel]

In https://github.com/tomojitakasu/PocketSDR/blob/master/src/sdr_func.c the dot_IQ_code function (called by corr_std) contains some NEON/AVX2 optimised code but it does not seems to do anything are the non optimised code is executed anyway at the end of the function. I would have expected a #else

The performance metics obtained via test/sdr_func_test.py are also unaffected by the activation or not of the neon optimisation.

// inner product of IQ data and code -------------------------------------------
static void dot_IQ_code(const sdr_cpx16_t *IQ, const sdr_cpx16_t *code, int N,
    float s, sdr_cpx_t *c)
{
    int i = 0;
    (*c)[0] = (*c)[1] = 0.0f;
#if defined(AVX2)
 ...
#elif defined(NEON)
...
#endif
    for ( ; i < N; i++) {
        (*c)[0] += IQ[i].I * code[i].I;
        (*c)[1] += IQ[i].Q * code[i].Q;
    }
    (*c)[0] *= s * SDR_CSCALE;
    (*c)[1] *= s * SDR_CSCALE;

[tomojitakasu]

Use test/utest/sdr_func_c_test to measure the performance. With my Raspi 4 (4GB, Raspi OS 64bit, Debian version 12.5), its test_05 reports:
with -DNEON

        mix_carr corr_std corr_fft
 12000    0.0251   0.0475   0.7805
 16000    0.0329   0.0626   0.8443
 24000    0.0497   0.0934   1.5235
 32000    0.0664   0.1240   2.2704
 32768    0.0681   0.1272   2.5413
 48000    0.1001   0.1858   4.2515
 65536    0.1359   0.2527   6.2994
 96000    0.1990   0.3818  12.1114

without -DNEON

        mix_carr corr_std corr_fft
 12000    0.0247   0.0935   0.7519
 16000    0.0329   0.1233   0.8333
 24000    0.0492   0.1845   1.4773
 32000    0.0655   0.2447   2.2628
 32768    0.0673   0.2506   2.1643
 48000    0.0989   0.3669   4.2197
 65536    0.1343   0.4999   6.5492
 96000    0.1979   0.7353  12.0928