FFmpeg/libavcodec/ffwavesynth.c

/*
 * Wavesynth pseudo-codec
 * Copyright (c) 2011 Nicolas George
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/intreadwrite.h"
#include "libavutil/log.h"
#include "avcodec.h"
#include "internal.h"


#define SIN_BITS 14
#define WS_MAX_CHANNELS 32
#define INF_TS 0x7FFFFFFFFFFFFFFF

#define PINK_UNIT 128

/*
   Format of the extradata and packets

   THIS INFORMATION IS NOT PART OF THE PUBLIC API OR ABI.
   IT CAN CHANGE WITHOUT NOTIFICATION.

   All numbers are in little endian.

   The codec extradata define a set of intervals with uniform content.
   Overlapping intervals are added together.

   extradata:
       uint32      number of intervals
       ...         intervals

   interval:
       int64       start timestamp; time_base must be 1/sample_rate;
                   start timestamps must be in ascending order
       int64       end timestamp
       uint32      type
       uint32      channels mask
       ...         additional information, depends on type

   sine interval (type fourcc "SINE"):
       int32       start frequency, in 1/(1<<16) Hz
       int32       end frequency
       int32       start amplitude, 1<<16 is the full amplitude
       int32       end amplitude
       uint32      start phase, 0 is sin(0), 0x20000000 is sin(pi/2), etc.;
                   n | (1<<31) means to match the phase of previous channel #n

   pink noise interval (type fourcc "NOIS"):
       int32       start amplitude
       int32       end amplitude

   The input packets encode the time and duration of the requested segment.

   packet:
       int64       start timestamp
       int32       duration

*/

enum ws_interval_type {
    WS_SINE  = MKTAG('S','I','N','E'),
    WS_NOISE = MKTAG('N','O','I','S'),
};

struct ws_interval {
    int64_t ts_start, ts_end;
    uint64_t phi0, dphi0, ddphi;
    uint64_t amp0, damp;
    uint64_t phi, dphi, amp;
    uint32_t channels;
    enum ws_interval_type type;
    int next;
};

struct wavesynth_context {
    int64_t cur_ts;
    int64_t next_ts;
    int32_t *sin;
    struct ws_interval *inter;
    uint32_t dither_state;
    uint32_t pink_state;
    int32_t pink_pool[PINK_UNIT];
    unsigned pink_need, pink_pos;
    int nb_inter;
    int cur_inter;
    int next_inter;
};

#define LCG_A 1284865837
#define LCG_C 4150755663
#define LCG_AI 849225893 /* A*AI = 1 [mod 1<<32] */

static uint32_t lcg_next(uint32_t *s)
{
    *s = *s * LCG_A + LCG_C;
    return *s;
}

static void lcg_seek(uint32_t *s, uint32_t dt)
{
    uint32_t a, c, t = *s;

    a = LCG_A;
    c = LCG_C;
    while (dt) {
        if (dt & 1)
            t = a * t + c;
        c *= a + 1; /* coefficients for a double step */
        a *= a;
        dt >>= 1;
    }
    *s = t;
}

/* Emulate pink noise by summing white noise at the sampling frequency,
 * white noise at half the sampling frequency (each value taken twice),
 * etc., with a total of 8 octaves.
 * This is known as the Voss-McCartney algorithm. */

static void pink_fill(struct wavesynth_context *ws)
{
    int32_t vt[7] = { 0 }, v = 0;
    int i, j;

    ws->pink_pos = 0;
    if (!ws->pink_need)
        return;
    for (i = 0; i < PINK_UNIT; i++) {
        for (j = 0; j < 7; j++) {
            if ((i >> j) & 1)
                break;
            v -= vt[j];
            vt[j] = (int32_t)lcg_next(&ws->pink_state) >> 3;
            v += vt[j];
        }
        ws->pink_pool[i] = v + ((int32_t)lcg_next(&ws->pink_state) >> 3);
    }
    lcg_next(&ws->pink_state); /* so we use exactly 256 steps */
}

/**
 * @return  (1<<64) * a / b, without overflow, if a < b
 */
static uint64_t frac64(uint64_t a, uint64_t b)
{
    uint64_t r = 0;
    int i;

    if (b < (uint64_t)1 << 32) { /* b small, use two 32-bits steps */
        a <<= 32;
        return ((a / b) << 32) | ((a % b) << 32) / b;
    }
    if (b < (uint64_t)1 << 48) { /* b medium, use four 16-bits steps */
        for (i = 0; i < 4; i++) {
            a <<= 16;
            r = (r << 16) | (a / b);
            a %= b;
        }
        return r;
    }
    for (i = 63; i >= 0; i--) {
        if (a >= (uint64_t)1 << 63 || a << 1 >= b) {
            r |= (uint64_t)1 << i;
            a = (a << 1) - b;
        } else {
            a <<= 1;
        }
    }
    return r;
}

static uint64_t phi_at(struct ws_interval *in, int64_t ts)
{
    uint64_t dt = ts - (uint64_t)in->ts_start;
    uint64_t dt2 = dt & 1 ? /* dt * (dt - 1) / 2 without overflow */
                   dt * ((dt - 1) >> 1) : (dt >> 1) * (dt - 1);
    return in->phi0 + dt * in->dphi0 + dt2 * in->ddphi;
}

static void wavesynth_seek(struct wavesynth_context *ws, int64_t ts)
{
    int *last, i;
    struct ws_interval *in;

    last = &ws->cur_inter;
    for (i = 0; i < ws->nb_inter; i++) {
        in = &ws->inter[i];
        if (ts < in->ts_start)
            break;
        if (ts >= in->ts_end)
            continue;
        *last = i;
        last = &in->next;
        in->phi  = phi_at(in, ts);
        in->dphi = in->dphi0 + (ts - in->ts_start) * in->ddphi;
        in->amp  = in->amp0  + (ts - in->ts_start) * in->damp;
    }
    ws->next_inter = i;
    ws->next_ts = i < ws->nb_inter ? ws->inter[i].ts_start : INF_TS;
    *last = -1;
    lcg_seek(&ws->dither_state, (uint32_t)ts - (uint32_t)ws->cur_ts);
    if (ws->pink_need) {
        uint64_t pink_ts_cur  = (ws->cur_ts + (uint64_t)PINK_UNIT - 1) & ~(PINK_UNIT - 1);
        uint64_t pink_ts_next = ts & ~(PINK_UNIT - 1);
        int pos = ts & (PINK_UNIT - 1);
        lcg_seek(&ws->pink_state, (uint32_t)(pink_ts_next - pink_ts_cur) * 2);
        if (pos) {
            pink_fill(ws);
            ws->pink_pos = pos;
        } else {
            ws->pink_pos = PINK_UNIT;
        }
    }
    ws->cur_ts = ts;
}

static int wavesynth_parse_extradata(AVCodecContext *avc)
{
    struct wavesynth_context *ws = avc->priv_data;
    struct ws_interval *in;
    uint8_t *edata, *edata_end;
    int32_t f1, f2, a1, a2;
    uint32_t phi;
    int64_t dphi1, dphi2, dt, cur_ts = -0x8000000000000000;
    int i;

    if (avc->extradata_size < 4)
        return AVERROR(EINVAL);
    edata = avc->extradata;
    edata_end = edata + avc->extradata_size;
    ws->nb_inter = AV_RL32(edata);
    edata += 4;
    if (ws->nb_inter < 0 || (edata_end - edata) / 24 < ws->nb_inter)
        return AVERROR(EINVAL);
    ws->inter = av_calloc(ws->nb_inter, sizeof(*ws->inter));
    if (!ws->inter)
        return AVERROR(ENOMEM);
    for (i = 0; i < ws->nb_inter; i++) {
        in = &ws->inter[i];
        if (edata_end - edata < 24)
            return AVERROR(EINVAL);
        in->ts_start = AV_RL64(edata +  0);
        in->ts_end   = AV_RL64(edata +  8);
        in->type     = AV_RL32(edata + 16);
        in->channels = AV_RL32(edata + 20);
        edata += 24;
        if (in->ts_start < cur_ts ||
            in->ts_end <= in->ts_start ||
            (uint64_t)in->ts_end - in->ts_start > INT64_MAX
        )
            return AVERROR(EINVAL);
        cur_ts = in->ts_start;
        dt = in->ts_end - in->ts_start;
        switch (in->type) {
            case WS_SINE:
                if (edata_end - edata < 20 || avc->sample_rate <= 0)
                    return AVERROR(EINVAL);
                f1  = AV_RL32(edata +  0);
                f2  = AV_RL32(edata +  4);
                a1  = AV_RL32(edata +  8);
                a2  = AV_RL32(edata + 12);
                phi = AV_RL32(edata + 16);
                edata += 20;
                dphi1 = frac64(f1, (int64_t)avc->sample_rate << 16);
                dphi2 = frac64(f2, (int64_t)avc->sample_rate << 16);
                in->dphi0 = dphi1;
                in->ddphi = (int64_t)(dphi2 - (uint64_t)dphi1) / dt;
                if (phi & 0x80000000) {
                    phi &= ~0x80000000;
                    if (phi >= i)
                        return AVERROR(EINVAL);
                    in->phi0 = phi_at(&ws->inter[phi], in->ts_start);
                } else {
                    in->phi0 = (uint64_t)phi << 33;
                }
                break;
            case WS_NOISE:
                if (edata_end - edata < 8)
                    return AVERROR(EINVAL);
                a1  = AV_RL32(edata +  0);
                a2  = AV_RL32(edata +  4);
                edata += 8;
                break;
            default:
                return AVERROR(EINVAL);
        }
        in->amp0 = (uint64_t)a1 << 32;
        in->damp = (int64_t)(((uint64_t)a2 << 32) - ((uint64_t)a1 << 32)) / dt;
    }
    if (edata != edata_end)
        return AVERROR(EINVAL);
    return 0;
}

static av_cold int wavesynth_init(AVCodecContext *avc)
{
    struct wavesynth_context *ws = avc->priv_data;
    int i, r;

    if (avc->channels > WS_MAX_CHANNELS) {
        av_log(avc, AV_LOG_ERROR,
               "This implementation is limited to %d channels.\n",
               WS_MAX_CHANNELS);
        return AVERROR(EINVAL);
    }
    r = wavesynth_parse_extradata(avc);
    if (r < 0) {
        av_log(avc, AV_LOG_ERROR, "Invalid intervals definitions.\n");
        goto fail;
    }
    ws->sin = av_malloc(sizeof(*ws->sin) << SIN_BITS);
    if (!ws->sin) {
        r = AVERROR(ENOMEM);
        goto fail;
    }
    for (i = 0; i < 1 << SIN_BITS; i++)
        ws->sin[i] = floor(32767 * sin(2 * M_PI * i / (1 << SIN_BITS)));
    ws->dither_state = MKTAG('D','I','T','H');
    for (i = 0; i < ws->nb_inter; i++)
        ws->pink_need += ws->inter[i].type == WS_NOISE;
    ws->pink_state = MKTAG('P','I','N','K');
    ws->pink_pos = PINK_UNIT;
    wavesynth_seek(ws, 0);
    avc->sample_fmt = AV_SAMPLE_FMT_S16;
    return 0;

fail:
    av_freep(&ws->inter);
    av_freep(&ws->sin);
    return r;
}

static void wavesynth_synth_sample(struct wavesynth_context *ws, int64_t ts,
                                   int32_t *channels)
{
    int32_t amp, *cv;
    unsigned val;
    struct ws_interval *in;
    int i, *last, pink;
    uint32_t c, all_ch = 0;

    i = ws->cur_inter;
    last = &ws->cur_inter;
    if (ws->pink_pos == PINK_UNIT)
        pink_fill(ws);
    pink = ws->pink_pool[ws->pink_pos++] >> 16;
    while (i >= 0) {
        in = &ws->inter[i];
        i = in->next;
        if (ts >= in->ts_end) {
            *last = i;
            continue;
        }
        last = &in->next;
        amp = in->amp >> 32;
        in->amp  += in->damp;
        switch (in->type) {
            case WS_SINE:
                val = amp * (unsigned)ws->sin[in->phi >> (64 - SIN_BITS)];
                in->phi  += in->dphi;
                in->dphi += in->ddphi;
                break;
            case WS_NOISE:
                val = amp * (unsigned)pink;
                break;
            default:
                val = 0;
        }
        all_ch |= in->channels;
        for (c = in->channels, cv = channels; c; c >>= 1, cv++)
            if (c & 1)
                *cv += (unsigned)val;
    }
    val = (int32_t)lcg_next(&ws->dither_state) >> 16;
    for (c = all_ch, cv = channels; c; c >>= 1, cv++)
        if (c & 1)
            *cv += val;
}

static void wavesynth_enter_intervals(struct wavesynth_context *ws, int64_t ts)
{
    int *last, i;
    struct ws_interval *in;

    last = &ws->cur_inter;
    for (i = ws->cur_inter; i >= 0; i = ws->inter[i].next)
        last = &ws->inter[i].next;
    for (i = ws->next_inter; i < ws->nb_inter; i++) {
        in = &ws->inter[i];
        if (ts < in->ts_start)
            break;
        if (ts >= in->ts_end)
            continue;
        *last = i;
        last = &in->next;
        in->phi = in->phi0;
        in->dphi = in->dphi0;
        in->amp = in->amp0;
    }
    ws->next_inter = i;
    ws->next_ts = i < ws->nb_inter ? ws->inter[i].ts_start : INF_TS;
    *last = -1;
}

static int wavesynth_decode(AVCodecContext *avc, void *rframe, int *rgot_frame,
                            AVPacket *packet)
{
    struct wavesynth_context *ws = avc->priv_data;
    AVFrame *frame = rframe;
    int64_t ts;
    int duration;
    int s, c, r;
    int16_t *pcm;
    int32_t channels[WS_MAX_CHANNELS];

    *rgot_frame = 0;
    if (packet->size != 12)
        return AVERROR_INVALIDDATA;
    ts = AV_RL64(packet->data);
    if (ts != ws->cur_ts)
        wavesynth_seek(ws, ts);
    duration = AV_RL32(packet->data + 8);
    if (duration <= 0)
        return AVERROR(EINVAL);
    frame->nb_samples = duration;
    r = ff_get_buffer(avc, frame, 0);
    if (r < 0)
        return r;
    pcm = (int16_t *)frame->data[0];
    for (s = 0; s < duration; s++, ts+=(uint64_t)1) {
        memset(channels, 0, avc->channels * sizeof(*channels));
        if (ts >= ws->next_ts)
            wavesynth_enter_intervals(ws, ts);
        wavesynth_synth_sample(ws, ts, channels);
        for (c = 0; c < avc->channels; c++)
            *(pcm++) = channels[c] >> 16;
    }
    ws->cur_ts += (uint64_t)duration;
    *rgot_frame = 1;
    return packet->size;
}

static av_cold int wavesynth_close(AVCodecContext *avc)
{
    struct wavesynth_context *ws = avc->priv_data;

    av_freep(&ws->sin);
    av_freep(&ws->inter);
    return 0;
}

AVCodec ff_ffwavesynth_decoder = {
    .name           = "wavesynth",
    .long_name      = NULL_IF_CONFIG_SMALL("Wave synthesis pseudo-codec"),
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = AV_CODEC_ID_FFWAVESYNTH,
    .priv_data_size = sizeof(struct wavesynth_context),
    .init           = wavesynth_init,
    .close          = wavesynth_close,
    .decode         = wavesynth_decode,
    .capabilities   = AV_CODEC_CAP_DR1,
};