Carla/source/modules/zita-resampler/vresampler.cc

// ----------------------------------------------------------------------------
//
//  Copyright (C) 2006-2023 Fons Adriaensen <fons@linuxaudio.org>
//
//  This program is free software; you can redistribute it and/or modify
//  it under the terms of the GNU General Public License as published by
//  the Free Software Foundation; either version 3 of the License, or
//  (at your option) any later version.
//
//  This program is distributed in the hope that it will be useful,
//  but WITHOUT ANY WARRANTY; without even the implied warranty of
//  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
//  GNU General Public License for more details.
//
//  You should have received a copy of the GNU General Public License
//  along with this program.  If not, see <http://www.gnu.org/licenses/>.
//
// ----------------------------------------------------------------------------


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

#include "vresampler.h"

#undef ENABLE_VEC4
#ifndef CARLA_OS_WIN
# if defined(__SSE2_MATH__)
#  define ENABLE_VEC4
#  include <xmmintrin.h>
# elif defined(__ARM_NEON) || defined(__ARM_NEON__)
#  define ENABLE_VEC4
#  include <arm_neon.h>
# endif
#endif


VResampler::VResampler (void) noexcept :
    _table (0),
    _nchan (0),
    _buff  (0),
    _c1 (0),
    _c2 (0)
{
    reset ();
}


VResampler::~VResampler (void)
{
    clear ();
}


bool VResampler::setup (double       ratio,
                        unsigned int nchan,
                        unsigned int hlen)
{
    return setup (ratio, nchan, hlen, 1.0 - 2.6 / hlen);
}


bool VResampler::setup (double       ratio,
                        unsigned int nchan,
                        unsigned int hlen,
                        double       frel)
{
    unsigned int       hl, mi, n;
    double             dp;
    Resampler_table    *T = 0;

    if (!nchan || (hlen < 8) || (hlen > 96) || (64 * ratio < 1) || (ratio > 256))
    {
        clear ();
        return false;
    }

    dp = NPHASE / ratio;
    hl = hlen;
    mi = 32;
    if (ratio < 1.0)
    {
        frel *= ratio;
        hl = (unsigned int)(ceil (hl / ratio));
        mi = (unsigned int)(ceil (mi / ratio));
    }
#ifdef ENABLE_VEC4
    hl = (hl + 3) & ~3;
#endif
    T = Resampler_table::create (frel, hl, NPHASE);
    clear ();
    if (T)
    {
        _table = T;
        n = nchan * (2 * hl + mi);
#ifdef ENABLE_VEC4
        posix_memalign ((void **)(&_buff), 16, n * sizeof (float));
        posix_memalign ((void **)(&_c1), 16, hl * sizeof (float));
        posix_memalign ((void **)(&_c2), 16, hl * sizeof (float));
#else
        _buff  = new float [n];
        _c1 = new float [hl];
        _c2 = new float [hl];
#endif
        _nchan = nchan;
        _ratio = ratio;
        _inmax = mi;
        _pstep = dp;
        _qstep = dp;
        _wstep = 1;
        return reset ();
    }
    else return false;
}


void VResampler::clear (void)
{
    Resampler_table::destroy (_table);
#ifdef ENABLE_VEC4
    free (_buff);
    free (_c1);
    free (_c2);
#else
    delete[] _buff;
    delete[] _c1;
    delete[] _c2;
#endif
    _buff = 0;
    _c1 = 0;
    _c2 = 0;
    _table = 0;
    _nchan = 0;
    _inmax = 0;
    _pstep = 0;
    _qstep = 0;
    _wstep = 1;
    reset ();
}


void VResampler::set_phase (double p)
{
    if (!_table) return;
    _phase = (p - floor (p)) * _table->_np;
}


void VResampler::set_rrfilt (double t)
{
    if (!_table) return;
    _wstep =  (t < 1) ? 1 : 1 - exp (-1 / t);
}


void VResampler::set_rratio (double r)
{
    if (!_table) return;
    if (r > 16.0) r = 16.0;
    if (r < 0.95) r = 0.95;
    _qstep = _table->_np / (_ratio * r);
}


double VResampler::inpdist (void) const noexcept
{
    if (!_table) return 0;
    return (int)(_table->_hl + 1 - _nread) - _phase / _table->_np;
}


int VResampler::inpsize (void) const noexcept
{
    if (!_table) return 0;
    return 2 * _table->_hl;
}


bool VResampler::reset (void) noexcept
{
    if (!_table) return false;

    inp_count = 0;
    out_count = 0;
    inp_data = 0;
    out_data = 0;
    _index = 0;
    _nread = 0;
    _nzero = 0;
    _phase = 0;
    if (_table)
    {
        _nread = 2 * _table->_hl;
        return true;
    }
    return false;
}


bool VResampler::process (void)
{
    int            nr, np, hl, nz, di, i, n;
    unsigned int   in, j;
    double         ph, dp, dd;
    float          a, b, *p1, *p2, *q1, *q2;

    if (!_table) return false;

    hl = _table->_hl;
    np = _table->_np;
    in = _index;
    nr = _nread;
    nz = _nzero;
    ph = _phase;
    dp = _pstep;

    p1 = _buff + in;
    p2 = p1 + 2 * hl - nr;
    di = 2 * hl + _inmax;

    while (out_count)
    {
        while (nr && inp_count)
        {
            if (inp_data)
            {
                for (j = 0; j < _nchan; j++) p2 [j * di] = inp_data [j];
                inp_data += _nchan;
                nz = 0;
            }
            else
            {
                for (j = 0; j < _nchan; j++) p2 [j * di] = 0;
                if (nz < 2 * hl) nz++;
            }
            p2++;
            nr--;
            inp_count--;
        }
        if (nr) break;

        if (out_data)
        {
            if (nz < 2 * hl)
            {
                n = (unsigned int) ph;
                b = (float)(ph - n);
                a = 1.0f - b;
                q1 = _table->_ctab + hl * n;
                q2 = _table->_ctab + hl * (np - n);

#if defined(__SSE2_MATH__) && !defined(CARLA_OS_WIN)
                __m128 C1, C2, Q1, Q2, S;
                C1 = _mm_load1_ps (&a);
                C2 = _mm_load1_ps (&b);
                for (i = 0; i < hl; i += 4)
                {
                    Q1 = _mm_load_ps (q1 + i);
                    Q2 = _mm_load_ps (q1 + i + hl);
                    S = _mm_add_ps (_mm_mul_ps (Q1, C1), _mm_mul_ps (Q2, C2));
                    _mm_store_ps (_c1 + i, S);
                    Q1 = _mm_load_ps (q2 + i);
                    Q2 = _mm_load_ps (q2 + i - hl);
                    S = _mm_add_ps (_mm_mul_ps (Q1, C1), _mm_mul_ps (Q2, C2));
                    _mm_store_ps (_c2 + i, S);
                }
                for (j = 0; j < _nchan; j++)
                {
                    q1 = p1 + j * di;
                    q2 = p2 + j * di;
                    S = _mm_setzero_ps ();
                    for (i = 0; i < hl; i += 4)
                    {
                        C1 = _mm_load_ps (_c1 + i);
                        Q1 = _mm_loadu_ps (q1);
                        q2 -= 4;
                        S = _mm_add_ps (S, _mm_mul_ps (C1, Q1));
                        C2 = _mm_loadr_ps (_c2 + i);
                        Q2 = _mm_loadu_ps (q2);
                        q1 += 4;
                        S = _mm_add_ps (S, _mm_mul_ps (C2, Q2));
                    }
                    *out_data++ = S [0] + S [1] + S [2] + S [3];
                }

#elif (defined(__ARM_NEON) || defined(__ARM_NEON__)) && !defined(CARLA_OS_WIN)
                // ARM64 version by Nicolas Belin <nbelin@baylibre.com>
                float32x4_t *C1 = (float32x4_t *)_c1;
                float32x4_t *C2 = (float32x4_t *)_c2;
                float32x4_t S, T;
                for (i = 0; i < (hl>>2); i++)
                {
                    T = vmulq_n_f32 (vld1q_f32 (q1 + hl), b);
                    C1 [i] = vmlaq_n_f32 (T, vld1q_f32 (q1), a);
                    T = vmulq_n_f32 (vld1q_f32 (q2 - hl), b);
                    C2 [i] = vmlaq_n_f32 (T, vld1q_f32 (q2), a);
                    q2 += 4;
                    q1 += 4;
                }
                for (j = 0; j < _nchan; j++)
                {
                    q1 = p1 + j * di;
                    q2 = p2 + j * di - 4;
                    T = vrev64q_f32 (vld1q_f32 (q2));
                    S = vmulq_f32 (vextq_f32 (T, T, 2), C2 [0]);
                    S = vmlaq_f32 (S, vld1q_f32 (q1), C1 [0]);
                    for (i = 1; i < (hl>>2); i++)
                    {
                        q2 -= 4;
                        q1 += 4;
                        T = vrev64q_f32 (vld1q_f32 (q2));
                        S = vmlaq_f32 (S, vextq_f32 (T, T, 2), C2 [i]);
                        S = vmlaq_f32 (S, vld1q_f32 (q1), C1 [i]);
                    }
                    *out_data++ = vaddvq_f32 (S);
                }

#else
                float s;
                for (i = 0; i < hl; i++)
                {
                    _c1 [i] = a * q1 [i] + b * q1 [i + hl];
                    _c2 [i] = a * q2 [i] + b * q2 [i - hl];
                }
                for (j = 0; j < _nchan; j++)
                {
                    q1 = p1 + j * di;
                    q2 = p2 + j * di;
                    s = 1e-30f;
                    for (i = 0; i < hl; i++)
                    {
                        q2--;
                        s += *q1 * _c1 [i] + *q2 * _c2 [i];
                        q1++;
                    }
                    *out_data++ = s - 1e-30f;
                }
#endif
            }
            else
            {
                for (j = 0; j < _nchan; j++) *out_data++ = 0;
            }
        }
        out_count--;

        dd =  _qstep - dp;
        if (fabs (dd) < 1e-20) dp = _qstep;
        else dp += _wstep * dd;
        ph += dp;
        if (ph >= np)
        {
            nr = (unsigned int) floor (ph / np);
            ph -= nr * np;;
            in += nr;
            p1 += nr;

            if (in >= _inmax)
            {
                n = 2 * hl - nr;
                p2 = _buff;
                for (j = 0; j < _nchan; j++)
                {
                    memmove (p2 + j * di, p1 + j * di, n * sizeof (float));
                }
                in = 0;
                p1 = _buff;
                p2 = p1 + n;
            }
        }
    }

    _index = in;
    _nread = nr;
    _phase = ph;
    _pstep = dp;
    _nzero = nz;

    return true;
}