falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
83122 Commits
32 Branches
283MB
Tree: a6191d098a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a6191d098a'
${ noResults }
FFmpeg/libavcodec/dcaenc.c

998 lines
31KB
Raw Blame History 

  1. /*

  2.  * DCA encoder

  3.  * Copyright (C) 2008-2012 Alexander E. Patrakov

  4.  *               2010 Benjamin Larsson

  5.  *               2011 Xiang Wang

  6.  *

  7.  * This file is part of FFmpeg.

  8.  *

  9.  * FFmpeg is free software; you can redistribute it and/or

  10.  * modify it under the terms of the GNU Lesser General Public

  11.  * License as published by the Free Software Foundation; either

  12.  * version 2.1 of the License, or (at your option) any later version.

  13.  *

  14.  * FFmpeg is distributed in the hope that it will be useful,

  15.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  16.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  17.  * Lesser General Public License for more details.

  18.  *

  19.  * You should have received a copy of the GNU Lesser General Public

  20.  * License along with FFmpeg; if not, write to the Free Software

  21.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  22.  */

  23. 

  24. #include "libavutil/avassert.h"

  25. #include "libavutil/channel_layout.h"

  26. #include "libavutil/common.h"

  27. #include "libavutil/ffmath.h"

  28. #include "avcodec.h"

  29. #include "dca.h"

  30. #include "dcadata.h"

  31. #include "dcaenc.h"

  32. #include "internal.h"

  33. #include "mathops.h"

  34. #include "put_bits.h"

  35. 

  36. #define MAX_CHANNELS 6

  37. #define DCA_MAX_FRAME_SIZE 16384

  38. #define DCA_HEADER_SIZE 13

  39. #define DCA_LFE_SAMPLES 8

  40. 

  41. #define DCAENC_SUBBANDS 32

  42. #define SUBFRAMES 1

  43. #define SUBSUBFRAMES 2

  44. #define SUBBAND_SAMPLES (SUBFRAMES * SUBSUBFRAMES * 8)

  45. #define AUBANDS 25

  46. 

  47. typedef struct DCAEncContext {

  48.     PutBitContext pb;

  49.     int frame_size;

  50.     int frame_bits;

  51.     int fullband_channels;

  52.     int channels;

  53.     int lfe_channel;

  54.     int samplerate_index;

  55.     int bitrate_index;

  56.     int channel_config;

  57.     const int32_t *band_interpolation;

  58.     const int32_t *band_spectrum;

  59.     int lfe_scale_factor;

  60.     softfloat lfe_quant;

  61.     int32_t lfe_peak_cb;

  62.     const int8_t *channel_order_tab;  ///< channel reordering table, lfe and non lfe

  63. 

  64.     int32_t history[MAX_CHANNELS][512]; /* This is a circular buffer */

  65.     int32_t subband[MAX_CHANNELS][DCAENC_SUBBANDS][SUBBAND_SAMPLES];

  66.     int32_t quantized[MAX_CHANNELS][DCAENC_SUBBANDS][SUBBAND_SAMPLES];

  67.     int32_t peak_cb[MAX_CHANNELS][DCAENC_SUBBANDS];

  68.     int32_t downsampled_lfe[DCA_LFE_SAMPLES];

  69.     int32_t masking_curve_cb[SUBSUBFRAMES][256];

  70.     int abits[MAX_CHANNELS][DCAENC_SUBBANDS];

  71.     int scale_factor[MAX_CHANNELS][DCAENC_SUBBANDS];

  72.     softfloat quant[MAX_CHANNELS][DCAENC_SUBBANDS];

  73.     int32_t eff_masking_curve_cb[256];

  74.     int32_t band_masking_cb[32];

  75.     int32_t worst_quantization_noise;

  76.     int32_t worst_noise_ever;

  77.     int consumed_bits;

  78. } DCAEncContext;

  79. 

  80. static int32_t cos_table[2048];

  81. static int32_t band_interpolation[2][512];

  82. static int32_t band_spectrum[2][8];

  83. static int32_t auf[9][AUBANDS][256];

  84. static int32_t cb_to_add[256];

  85. static int32_t cb_to_level[2048];

  86. static int32_t lfe_fir_64i[512];

  87. 

  88. /* Transfer function of outer and middle ear, Hz -> dB */

  89. static double hom(double f)

  90. {

  91.     double f1 = f / 1000;

  92. 

  93.     return -3.64 * pow(f1, -0.8)

  94.            + 6.8 * exp(-0.6 * (f1 - 3.4) * (f1 - 3.4))

  95.            - 6.0 * exp(-0.15 * (f1 - 8.7) * (f1 - 8.7))

  96.            - 0.0006 * (f1 * f1) * (f1 * f1);

  97. }

  98. 

  99. static double gammafilter(int i, double f)

  100. {

  101.     double h = (f - fc[i]) / erb[i];

  102. 

  103.     h = 1 + h * h;

  104.     h = 1 / (h * h);

  105.     return 20 * log10(h);

  106. }

  107. 

  108. static int encode_init(AVCodecContext *avctx)

  109. {

  110.     DCAEncContext *c = avctx->priv_data;

  111.     uint64_t layout = avctx->channel_layout;

  112.     int i, min_frame_bits;

  113. 

  114.     c->fullband_channels = c->channels = avctx->channels;

  115.     c->lfe_channel = (avctx->channels == 3 || avctx->channels == 6);

  116.     c->band_interpolation = band_interpolation[1];

  117.     c->band_spectrum = band_spectrum[1];

  118.     c->worst_quantization_noise = -2047;

  119.     c->worst_noise_ever = -2047;

  120. 

  121.     if (!layout) {

  122.         av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "

  123.                                       "encoder will guess the layout, but it "

  124.                                       "might be incorrect.\n");

  125.         layout = av_get_default_channel_layout(avctx->channels);

  126.     }

  127.     switch (layout) {

  128.     case AV_CH_LAYOUT_MONO:         c->channel_config = 0; break;

  129.     case AV_CH_LAYOUT_STEREO:       c->channel_config = 2; break;

  130.     case AV_CH_LAYOUT_2_2:          c->channel_config = 8; break;

  131.     case AV_CH_LAYOUT_5POINT0:      c->channel_config = 9; break;

  132.     case AV_CH_LAYOUT_5POINT1:      c->channel_config = 9; break;

  133.     default:

  134.         av_log(avctx, AV_LOG_ERROR, "Unsupported channel layout!\n");

  135.         return AVERROR_PATCHWELCOME;

  136.     }

  137. 

  138.     if (c->lfe_channel) {

  139.         c->fullband_channels--;

  140.         c->channel_order_tab = channel_reorder_lfe[c->channel_config];

  141.     } else {

  142.         c->channel_order_tab = channel_reorder_nolfe[c->channel_config];

  143.     }

  144. 

  145.     for (i = 0; i < 9; i++) {

  146.         if (sample_rates[i] == avctx->sample_rate)

  147.             break;

  148.     }

  149.     if (i == 9)

  150.         return AVERROR(EINVAL);

  151.     c->samplerate_index = i;

  152. 

  153.     if (avctx->bit_rate < 32000 || avctx->bit_rate > 3840000) {

  154.         av_log(avctx, AV_LOG_ERROR, "Bit rate %"PRId64" not supported.", (int64_t)avctx->bit_rate);

  155.         return AVERROR(EINVAL);

  156.     }

  157.     for (i = 0; ff_dca_bit_rates[i] < avctx->bit_rate; i++)

  158.         ;

  159.     c->bitrate_index = i;

  160.     c->frame_bits = FFALIGN((avctx->bit_rate * 512 + avctx->sample_rate - 1) / avctx->sample_rate, 32);

  161.     min_frame_bits = 132 + (493 + 28 * 32) * c->fullband_channels + c->lfe_channel * 72;

  162.     if (c->frame_bits < min_frame_bits || c->frame_bits > (DCA_MAX_FRAME_SIZE << 3))

  163.         return AVERROR(EINVAL);

  164. 

  165.     c->frame_size = (c->frame_bits + 7) / 8;

  166. 

  167.     avctx->frame_size = 32 * SUBBAND_SAMPLES;

  168. 

  169.     if (!cos_table[0]) {

  170.         int j, k;

  171. 

  172.         cos_table[0] = 0x7fffffff;

  173.         cos_table[512] = 0;

  174.         cos_table[1024] = -cos_table[0];

  175.         for (i = 1; i < 512; i++) {

  176.             cos_table[i]   = (int32_t)(0x7fffffff * cos(M_PI * i / 1024));

  177.             cos_table[1024-i] = -cos_table[i];

  178.             cos_table[1024+i] = -cos_table[i];

  179.             cos_table[2048-i] = cos_table[i];

  180.         }

  181.         for (i = 0; i < 2048; i++) {

  182.             cb_to_level[i] = (int32_t)(0x7fffffff * ff_exp10(-0.005 * i));

  183.         }

  184. 

  185.         for (k = 0; k < 32; k++) {

  186.             for (j = 0; j < 8; j++) {

  187.                 lfe_fir_64i[64 * j + k] = (int32_t)(0xffffff800000ULL * ff_dca_lfe_fir_64[8 * k + j]);

  188.                 lfe_fir_64i[64 * (7-j) + (63 - k)] = (int32_t)(0xffffff800000ULL * ff_dca_lfe_fir_64[8 * k + j]);

  189.             }

  190.         }

  191. 

  192.         for (i = 0; i < 512; i++) {

  193.             band_interpolation[0][i] = (int32_t)(0x1000000000ULL * ff_dca_fir_32bands_perfect[i]);

  194.             band_interpolation[1][i] = (int32_t)(0x1000000000ULL * ff_dca_fir_32bands_nonperfect[i]);

  195.         }

  196. 

  197.         for (i = 0; i < 9; i++) {

  198.             for (j = 0; j < AUBANDS; j++) {

  199.                 for (k = 0; k < 256; k++) {

  200.                     double freq = sample_rates[i] * (k + 0.5) / 512;

  201. 

  202.                     auf[i][j][k] = (int32_t)(10 * (hom(freq) + gammafilter(j, freq)));

  203.                 }

  204.             }

  205.         }

  206. 

  207.         for (i = 0; i < 256; i++) {

  208.             double add = 1 + ff_exp10(-0.01 * i);

  209.             cb_to_add[i] = (int32_t)(100 * log10(add));

  210.         }

  211.         for (j = 0; j < 8; j++) {

  212.             double accum = 0;

  213.             for (i = 0; i < 512; i++) {

  214.                 double reconst = ff_dca_fir_32bands_perfect[i] * ((i & 64) ? (-1) : 1);

  215.                 accum += reconst * cos(2 * M_PI * (i + 0.5 - 256) * (j + 0.5) / 512);

  216.             }

  217.             band_spectrum[0][j] = (int32_t)(200 * log10(accum));

  218.         }

  219.         for (j = 0; j < 8; j++) {

  220.             double accum = 0;

  221.             for (i = 0; i < 512; i++) {

  222.                 double reconst = ff_dca_fir_32bands_nonperfect[i] * ((i & 64) ? (-1) : 1);

  223.                 accum += reconst * cos(2 * M_PI * (i + 0.5 - 256) * (j + 0.5) / 512);

  224.             }

  225.             band_spectrum[1][j] = (int32_t)(200 * log10(accum));

  226.         }

  227.     }

  228.     return 0;

  229. }

  230. 

  231. static inline int32_t cos_t(int x)

  232. {

  233.     return cos_table[x & 2047];

  234. }

  235. 

  236. static inline int32_t sin_t(int x)

  237. {

  238.     return cos_t(x - 512);

  239. }

  240. 

  241. static inline int32_t half32(int32_t a)

  242. {

  243.     return (a + 1) >> 1;

  244. }

  245. 

  246. static inline int32_t mul32(int32_t a, int32_t b)

  247. {

  248.     int64_t r = (int64_t)a * b + 0x80000000ULL;

  249.     return r >> 32;

  250. }

  251. 

  252. static void subband_transform(DCAEncContext *c, const int32_t *input)

  253. {

  254.     int ch, subs, i, k, j;

  255. 

  256.     for (ch = 0; ch < c->fullband_channels; ch++) {

  257.         /* History is copied because it is also needed for PSY */

  258.         int32_t hist[512];

  259.         int hist_start = 0;

  260.         const int chi = c->channel_order_tab[ch];

  261. 

  262.         memcpy(hist, &c->history[ch][0], 512 * sizeof(int32_t));

  263. 

  264.         for (subs = 0; subs < SUBBAND_SAMPLES; subs++) {

  265.             int32_t accum[64];

  266.             int32_t resp;

  267.             int band;

  268. 

  269.             /* Calculate the convolutions at once */

  270.             memset(accum, 0, 64 * sizeof(int32_t));

  271. 

  272.             for (k = 0, i = hist_start, j = 0;

  273.                     i < 512; k = (k + 1) & 63, i++, j++)

  274.                 accum[k] += mul32(hist[i], c->band_interpolation[j]);

  275.             for (i = 0; i < hist_start; k = (k + 1) & 63, i++, j++)

  276.                 accum[k] += mul32(hist[i], c->band_interpolation[j]);

  277. 

  278.             for (k = 16; k < 32; k++)

  279.                 accum[k] = accum[k] - accum[31 - k];

  280.             for (k = 32; k < 48; k++)

  281.                 accum[k] = accum[k] + accum[95 - k];

  282. 

  283.             for (band = 0; band < 32; band++) {

  284.                 resp = 0;

  285.                 for (i = 16; i < 48; i++) {

  286.                     int s = (2 * band + 1) * (2 * (i + 16) + 1);

  287.                     resp += mul32(accum[i], cos_t(s << 3)) >> 3;

  288.                 }

  289. 

  290.                 c->subband[ch][band][subs] = ((band + 1) & 2) ? -resp : resp;

  291.             }

  292. 

  293.             /* Copy in 32 new samples from input */

  294.             for (i = 0; i < 32; i++)

  295.                 hist[i + hist_start] = input[(subs * 32 + i) * c->channels + chi];

  296. 

  297.             hist_start = (hist_start + 32) & 511;

  298.         }

  299.     }

  300. }

  301. 

  302. static void lfe_downsample(DCAEncContext *c, const int32_t *input)

  303. {

  304.     /* FIXME: make 128x LFE downsampling possible */

  305.     const int lfech = lfe_index[c->channel_config];

  306.     int i, j, lfes;

  307.     int32_t hist[512];

  308.     int32_t accum;

  309.     int hist_start = 0;

  310. 

  311.     memcpy(hist, &c->history[c->channels - 1][0], 512 * sizeof(int32_t));

  312. 

  313.     for (lfes = 0; lfes < DCA_LFE_SAMPLES; lfes++) {

  314.         /* Calculate the convolution */

  315.         accum = 0;

  316. 

  317.         for (i = hist_start, j = 0; i < 512; i++, j++)

  318.             accum += mul32(hist[i], lfe_fir_64i[j]);

  319.         for (i = 0; i < hist_start; i++, j++)

  320.             accum += mul32(hist[i], lfe_fir_64i[j]);

  321. 

  322.         c->downsampled_lfe[lfes] = accum;

  323. 

  324.         /* Copy in 64 new samples from input */

  325.         for (i = 0; i < 64; i++)

  326.             hist[i + hist_start] = input[(lfes * 64 + i) * c->channels + lfech];

  327. 

  328.         hist_start = (hist_start + 64) & 511;

  329.     }

  330. }

  331. 

  332. typedef struct {

  333.     int32_t re;

  334.     int32_t im;

  335. } cplx32;

  336. 

  337. static void fft(const int32_t in[2 * 256], cplx32 out[256])

  338. {

  339.     cplx32 buf[256], rin[256], rout[256];

  340.     int i, j, k, l;

  341. 

  342.     /* do two transforms in parallel */

  343.     for (i = 0; i < 256; i++) {

  344.         /* Apply the Hann window */

  345.         rin[i].re = mul32(in[2 * i], 0x3fffffff - (cos_t(8 * i + 2) >> 1));

  346.         rin[i].im = mul32(in[2 * i + 1], 0x3fffffff - (cos_t(8 * i + 6) >> 1));

  347.     }

  348.     /* pre-rotation */

  349.     for (i = 0; i < 256; i++) {

  350.         buf[i].re = mul32(cos_t(4 * i + 2), rin[i].re)

  351.                   - mul32(sin_t(4 * i + 2), rin[i].im);

  352.         buf[i].im = mul32(cos_t(4 * i + 2), rin[i].im)

  353.                   + mul32(sin_t(4 * i + 2), rin[i].re);

  354.     }

  355. 

  356.     for (j = 256, l = 1; j != 1; j >>= 1, l <<= 1) {

  357.         for (k = 0; k < 256; k += j) {

  358.             for (i = k; i < k + j / 2; i++) {

  359.                 cplx32 sum, diff;

  360.                 int t = 8 * l * i;

  361. 

  362.                 sum.re = buf[i].re + buf[i + j / 2].re;

  363.                 sum.im = buf[i].im + buf[i + j / 2].im;

  364. 

  365.                 diff.re = buf[i].re - buf[i + j / 2].re;

  366.                 diff.im = buf[i].im - buf[i + j / 2].im;

  367. 

  368.                 buf[i].re = half32(sum.re);

  369.                 buf[i].im = half32(sum.im);

  370. 

  371.                 buf[i + j / 2].re = mul32(diff.re, cos_t(t))

  372.                                   - mul32(diff.im, sin_t(t));

  373.                 buf[i + j / 2].im = mul32(diff.im, cos_t(t))

  374.                                   + mul32(diff.re, sin_t(t));

  375.             }

  376.         }

  377.     }

  378.     /* post-rotation */

  379.     for (i = 0; i < 256; i++) {

  380.         int b = ff_reverse[i];

  381.         rout[i].re = mul32(buf[b].re, cos_t(4 * i))

  382.                    - mul32(buf[b].im, sin_t(4 * i));

  383.         rout[i].im = mul32(buf[b].im, cos_t(4 * i))

  384.                    + mul32(buf[b].re, sin_t(4 * i));

  385.     }

  386.     for (i = 0; i < 256; i++) {

  387.         /* separate the results of the two transforms */

  388.         cplx32 o1, o2;

  389. 

  390.         o1.re =  rout[i].re - rout[255 - i].re;

  391.         o1.im =  rout[i].im + rout[255 - i].im;

  392. 

  393.         o2.re =  rout[i].im - rout[255 - i].im;

  394.         o2.im = -rout[i].re - rout[255 - i].re;

  395. 

  396.         /* combine them into one long transform */

  397.         out[i].re = mul32( o1.re + o2.re, cos_t(2 * i + 1))

  398.                   + mul32( o1.im - o2.im, sin_t(2 * i + 1));

  399.         out[i].im = mul32( o1.im + o2.im, cos_t(2 * i + 1))

  400.                   + mul32(-o1.re + o2.re, sin_t(2 * i + 1));

  401.     }

  402. }

  403. 

  404. static int32_t get_cb(int32_t in)

  405. {

  406.     int i, res;

  407. 

  408.     res = 0;

  409.     if (in < 0)

  410.         in = -in;

  411.     for (i = 1024; i > 0; i >>= 1) {

  412.         if (cb_to_level[i + res] >= in)

  413.             res += i;

  414.     }

  415.     return -res;

  416. }

  417. 

  418. static int32_t add_cb(int32_t a, int32_t b)

  419. {

  420.     if (a < b)

  421.         FFSWAP(int32_t, a, b);

  422. 

  423.     if (a - b >= 256)

  424.         return a;

  425.     return a + cb_to_add[a - b];

  426. }

  427. 

  428. static void adjust_jnd(int samplerate_index,

  429.                        const int32_t in[512], int32_t out_cb[256])

  430. {

  431.     int32_t power[256];

  432.     cplx32 out[256];

  433.     int32_t out_cb_unnorm[256];

  434.     int32_t denom;

  435.     const int32_t ca_cb = -1114;

  436.     const int32_t cs_cb = 928;

  437.     int i, j;

  438. 

  439.     fft(in, out);

  440. 

  441.     for (j = 0; j < 256; j++) {

  442.         power[j] = add_cb(get_cb(out[j].re), get_cb(out[j].im));

  443.         out_cb_unnorm[j] = -2047; /* and can only grow */

  444.     }

  445. 

  446.     for (i = 0; i < AUBANDS; i++) {

  447.         denom = ca_cb; /* and can only grow */

  448.         for (j = 0; j < 256; j++)

  449.             denom = add_cb(denom, power[j] + auf[samplerate_index][i][j]);

  450.         for (j = 0; j < 256; j++)

  451.             out_cb_unnorm[j] = add_cb(out_cb_unnorm[j],

  452.                     -denom + auf[samplerate_index][i][j]);

  453.     }

  454. 

  455.     for (j = 0; j < 256; j++)

  456.         out_cb[j] = add_cb(out_cb[j], -out_cb_unnorm[j] - ca_cb - cs_cb);

  457. }

  458. 

  459. typedef void (*walk_band_t)(DCAEncContext *c, int band1, int band2, int f,

  460.                             int32_t spectrum1, int32_t spectrum2, int channel,

  461.                             int32_t * arg);

  462. 

  463. static void walk_band_low(DCAEncContext *c, int band, int channel,

  464.                           walk_band_t walk, int32_t *arg)

  465. {

  466.     int f;

  467. 

  468.     if (band == 0) {

  469.         for (f = 0; f < 4; f++)

  470.             walk(c, 0, 0, f, 0, -2047, channel, arg);

  471.     } else {

  472.         for (f = 0; f < 8; f++)

  473.             walk(c, band, band - 1, 8 * band - 4 + f,

  474.                     c->band_spectrum[7 - f], c->band_spectrum[f], channel, arg);

  475.     }

  476. }

  477. 

  478. static void walk_band_high(DCAEncContext *c, int band, int channel,

  479.                            walk_band_t walk, int32_t *arg)

  480. {

  481.     int f;

  482. 

  483.     if (band == 31) {

  484.         for (f = 0; f < 4; f++)

  485.             walk(c, 31, 31, 256 - 4 + f, 0, -2047, channel, arg);

  486.     } else {

  487.         for (f = 0; f < 8; f++)

  488.             walk(c, band, band + 1, 8 * band + 4 + f,

  489.                     c->band_spectrum[f], c->band_spectrum[7 - f], channel, arg);

  490.     }

  491. }

  492. 

  493. static void update_band_masking(DCAEncContext *c, int band1, int band2,

  494.                                 int f, int32_t spectrum1, int32_t spectrum2,

  495.                                 int channel, int32_t * arg)

  496. {

  497.     int32_t value = c->eff_masking_curve_cb[f] - spectrum1;

  498. 

  499.     if (value < c->band_masking_cb[band1])

  500.         c->band_masking_cb[band1] = value;

  501. }

  502. 

  503. static void calc_masking(DCAEncContext *c, const int32_t *input)

  504. {

  505.     int i, k, band, ch, ssf;

  506.     int32_t data[512];

  507. 

  508.     for (i = 0; i < 256; i++)

  509.         for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)

  510.             c->masking_curve_cb[ssf][i] = -2047;

  511. 

  512.     for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)

  513.         for (ch = 0; ch < c->fullband_channels; ch++) {

  514.             const int chi = c->channel_order_tab[ch];

  515. 

  516.             for (i = 0, k = 128 + 256 * ssf; k < 512; i++, k++)

  517.                 data[i] = c->history[ch][k];

  518.             for (k -= 512; i < 512; i++, k++)

  519.                 data[i] = input[k * c->channels + chi];

  520.             adjust_jnd(c->samplerate_index, data, c->masking_curve_cb[ssf]);

  521.         }

  522.     for (i = 0; i < 256; i++) {

  523.         int32_t m = 2048;

  524. 

  525.         for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)

  526.             if (c->masking_curve_cb[ssf][i] < m)

  527.                 m = c->masking_curve_cb[ssf][i];

  528.         c->eff_masking_curve_cb[i] = m;

  529.     }

  530. 

  531.     for (band = 0; band < 32; band++) {

  532.         c->band_masking_cb[band] = 2048;

  533.         walk_band_low(c, band, 0, update_band_masking, NULL);

  534.         walk_band_high(c, band, 0, update_band_masking, NULL);

  535.     }

  536. }

  537. 

  538. static void find_peaks(DCAEncContext *c)

  539. {

  540.     int band, ch;

  541. 

  542.     for (ch = 0; ch < c->fullband_channels; ch++)

  543.         for (band = 0; band < 32; band++) {

  544.             int sample;

  545.             int32_t m = 0;

  546. 

  547.             for (sample = 0; sample < SUBBAND_SAMPLES; sample++) {

  548.                 int32_t s = abs(c->subband[ch][band][sample]);

  549.                 if (m < s)

  550.                     m = s;

  551.             }

  552.             c->peak_cb[ch][band] = get_cb(m);

  553.         }

  554. 

  555.     if (c->lfe_channel) {

  556.         int sample;

  557.         int32_t m = 0;

  558. 

  559.         for (sample = 0; sample < DCA_LFE_SAMPLES; sample++)

  560.             if (m < abs(c->downsampled_lfe[sample]))

  561.                 m = abs(c->downsampled_lfe[sample]);

  562.         c->lfe_peak_cb = get_cb(m);

  563.     }

  564. }

  565. 

  566. static const int snr_fudge = 128;

  567. #define USED_1ABITS 1

  568. #define USED_NABITS 2

  569. #define USED_26ABITS 4

  570. 

  571. static int init_quantization_noise(DCAEncContext *c, int noise)

  572. {

  573.     int ch, band, ret = 0;

  574. 

  575.     c->consumed_bits = 132 + 493 * c->fullband_channels;

  576.     if (c->lfe_channel)

  577.         c->consumed_bits += 72;

  578. 

  579.     /* attempt to guess the bit distribution based on the prevoius frame */

  580.     for (ch = 0; ch < c->fullband_channels; ch++) {

  581.         for (band = 0; band < 32; band++) {

  582.             int snr_cb = c->peak_cb[ch][band] - c->band_masking_cb[band] - noise;

  583. 

  584.             if (snr_cb >= 1312) {

  585.                 c->abits[ch][band] = 26;

  586.                 ret |= USED_26ABITS;

  587.             } else if (snr_cb >= 222) {

  588.                 c->abits[ch][band] = 8 + mul32(snr_cb - 222, 69000000);

  589.                 ret |= USED_NABITS;

  590.             } else if (snr_cb >= 0) {

  591.                 c->abits[ch][band] = 2 + mul32(snr_cb, 106000000);

  592.                 ret |= USED_NABITS;

  593.             } else {

  594.                 c->abits[ch][band] = 1;

  595.                 ret |= USED_1ABITS;

  596.             }

  597.         }

  598.     }

  599. 

  600.     for (ch = 0; ch < c->fullband_channels; ch++)

  601.         for (band = 0; band < 32; band++) {

  602.             c->consumed_bits += bit_consumption[c->abits[ch][band]];

  603.         }

  604. 

  605.     return ret;

  606. }

  607. 

  608. static void assign_bits(DCAEncContext *c)

  609. {

  610.     /* Find the bounds where the binary search should work */

  611.     int low, high, down;

  612.     int used_abits = 0;

  613. 

  614.     init_quantization_noise(c, c->worst_quantization_noise);

  615.     low = high = c->worst_quantization_noise;

  616.     if (c->consumed_bits > c->frame_bits) {

  617.         while (c->consumed_bits > c->frame_bits) {

  618.             av_assert0(used_abits != USED_1ABITS);

  619.             low = high;

  620.             high += snr_fudge;

  621.             used_abits = init_quantization_noise(c, high);

  622.         }

  623.     } else {

  624.         while (c->consumed_bits <= c->frame_bits) {

  625.             high = low;

  626.             if (used_abits == USED_26ABITS)

  627.                 goto out; /* The requested bitrate is too high, pad with zeros */

  628.             low -= snr_fudge;

  629.             used_abits = init_quantization_noise(c, low);

  630.         }

  631.     }

  632. 

  633.     /* Now do a binary search between low and high to see what fits */

  634.     for (down = snr_fudge >> 1; down; down >>= 1) {

  635.         init_quantization_noise(c, high - down);

  636.         if (c->consumed_bits <= c->frame_bits)

  637.             high -= down;

  638.     }

  639.     init_quantization_noise(c, high);

  640. out:

  641.     c->worst_quantization_noise = high;

  642.     if (high > c->worst_noise_ever)

  643.         c->worst_noise_ever = high;

  644. }

  645. 

  646. static void shift_history(DCAEncContext *c, const int32_t *input)

  647. {

  648.     int k, ch;

  649. 

  650.     for (k = 0; k < 512; k++)

  651.         for (ch = 0; ch < c->channels; ch++) {

  652.             const int chi = c->channel_order_tab[ch];

  653. 

  654.             c->history[ch][k] = input[k * c->channels + chi];

  655.         }

  656. }

  657. 

  658. static int32_t quantize_value(int32_t value, softfloat quant)

  659. {

  660.     int32_t offset = 1 << (quant.e - 1);

  661. 

  662.     value = mul32(value, quant.m) + offset;

  663.     value = value >> quant.e;

  664.     return value;

  665. }

  666. 

  667. static int calc_one_scale(int32_t peak_cb, int abits, softfloat *quant)

  668. {

  669.     int32_t peak;

  670.     int our_nscale, try_remove;

  671.     softfloat our_quant;

  672. 

  673.     av_assert0(peak_cb <= 0);

  674.     av_assert0(peak_cb >= -2047);

  675. 

  676.     our_nscale = 127;

  677.     peak = cb_to_level[-peak_cb];

  678. 

  679.     for (try_remove = 64; try_remove > 0; try_remove >>= 1) {

  680.         if (scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e <= 17)

  681.             continue;

  682.         our_quant.m = mul32(scalefactor_inv[our_nscale - try_remove].m, stepsize_inv[abits].m);

  683.         our_quant.e = scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e - 17;

  684.         if ((ff_dca_quant_levels[abits] - 1) / 2 < quantize_value(peak, our_quant))

  685.             continue;

  686.         our_nscale -= try_remove;

  687.     }

  688. 

  689.     if (our_nscale >= 125)

  690.         our_nscale = 124;

  691. 

  692.     quant->m = mul32(scalefactor_inv[our_nscale].m, stepsize_inv[abits].m);

  693.     quant->e = scalefactor_inv[our_nscale].e + stepsize_inv[abits].e - 17;

  694.     av_assert0((ff_dca_quant_levels[abits] - 1) / 2 >= quantize_value(peak, *quant));

  695. 

  696.     return our_nscale;

  697. }

  698. 

  699. static void calc_scales(DCAEncContext *c)

  700. {

  701.     int band, ch;

  702. 

  703.     for (ch = 0; ch < c->fullband_channels; ch++)

  704.         for (band = 0; band < 32; band++)

  705.             c->scale_factor[ch][band] = calc_one_scale(c->peak_cb[ch][band],

  706.                                                        c->abits[ch][band],

  707.                                                        &c->quant[ch][band]);

  708. 

  709.     if (c->lfe_channel)

  710.         c->lfe_scale_factor = calc_one_scale(c->lfe_peak_cb, 11, &c->lfe_quant);

  711. }

  712. 

  713. static void quantize_all(DCAEncContext *c)

  714. {

  715.     int sample, band, ch;

  716. 

  717.     for (ch = 0; ch < c->fullband_channels; ch++)

  718.         for (band = 0; band < 32; band++)

  719.             for (sample = 0; sample < SUBBAND_SAMPLES; sample++)

  720.                 c->quantized[ch][band][sample] = quantize_value(c->subband[ch][band][sample], c->quant[ch][band]);

  721. }

  722. 

  723. static void put_frame_header(DCAEncContext *c)

  724. {

  725.     /* SYNC */

  726.     put_bits(&c->pb, 16, 0x7ffe);

  727.     put_bits(&c->pb, 16, 0x8001);

  728. 

  729.     /* Frame type: normal */

  730.     put_bits(&c->pb, 1, 1);

  731. 

  732.     /* Deficit sample count: none */

  733.     put_bits(&c->pb, 5, 31);

  734. 

  735.     /* CRC is not present */

  736.     put_bits(&c->pb, 1, 0);

  737. 

  738.     /* Number of PCM sample blocks */

  739.     put_bits(&c->pb, 7, SUBBAND_SAMPLES - 1);

  740. 

  741.     /* Primary frame byte size */

  742.     put_bits(&c->pb, 14, c->frame_size - 1);

  743. 

  744.     /* Audio channel arrangement */

  745.     put_bits(&c->pb, 6, c->channel_config);

  746. 

  747.     /* Core audio sampling frequency */

  748.     put_bits(&c->pb, 4, bitstream_sfreq[c->samplerate_index]);

  749. 

  750.     /* Transmission bit rate */

  751.     put_bits(&c->pb, 5, c->bitrate_index);

  752. 

  753.     /* Embedded down mix: disabled */

  754.     put_bits(&c->pb, 1, 0);

  755. 

  756.     /* Embedded dynamic range flag: not present */

  757.     put_bits(&c->pb, 1, 0);

  758. 

  759.     /* Embedded time stamp flag: not present */

  760.     put_bits(&c->pb, 1, 0);

  761. 

  762.     /* Auxiliary data flag: not present */

  763.     put_bits(&c->pb, 1, 0);

  764. 

  765.     /* HDCD source: no */

  766.     put_bits(&c->pb, 1, 0);

  767. 

  768.     /* Extension audio ID: N/A */

  769.     put_bits(&c->pb, 3, 0);

  770. 

  771.     /* Extended audio data: not present */

  772.     put_bits(&c->pb, 1, 0);

  773. 

  774.     /* Audio sync word insertion flag: after each sub-frame */

  775.     put_bits(&c->pb, 1, 0);

  776. 

  777.     /* Low frequency effects flag: not present or 64x subsampling */

  778.     put_bits(&c->pb, 2, c->lfe_channel ? 2 : 0);

  779. 

  780.     /* Predictor history switch flag: on */

  781.     put_bits(&c->pb, 1, 1);

  782. 

  783.     /* No CRC */

  784.     /* Multirate interpolator switch: non-perfect reconstruction */

  785.     put_bits(&c->pb, 1, 0);

  786. 

  787.     /* Encoder software revision: 7 */

  788.     put_bits(&c->pb, 4, 7);

  789. 

  790.     /* Copy history: 0 */

  791.     put_bits(&c->pb, 2, 0);

  792. 

  793.     /* Source PCM resolution: 16 bits, not DTS ES */

  794.     put_bits(&c->pb, 3, 0);

  795. 

  796.     /* Front sum/difference coding: no */

  797.     put_bits(&c->pb, 1, 0);

  798. 

  799.     /* Surrounds sum/difference coding: no */

  800.     put_bits(&c->pb, 1, 0);

  801. 

  802.     /* Dialog normalization: 0 dB */

  803.     put_bits(&c->pb, 4, 0);

  804. }

  805. 

  806. static void put_primary_audio_header(DCAEncContext *c)

  807. {

  808.     static const int bitlen[11] = { 0, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3 };

  809.     static const int thr[11]    = { 0, 1, 3, 3, 3, 3, 7, 7, 7, 7, 7 };

  810. 

  811.     int ch, i;

  812.     /* Number of subframes */

  813.     put_bits(&c->pb, 4, SUBFRAMES - 1);

  814. 

  815.     /* Number of primary audio channels */

  816.     put_bits(&c->pb, 3, c->fullband_channels - 1);

  817. 

  818.     /* Subband activity count */

  819.     for (ch = 0; ch < c->fullband_channels; ch++)

  820.         put_bits(&c->pb, 5, DCAENC_SUBBANDS - 2);

  821. 

  822.     /* High frequency VQ start subband */

  823.     for (ch = 0; ch < c->fullband_channels; ch++)

  824.         put_bits(&c->pb, 5, DCAENC_SUBBANDS - 1);

  825. 

  826.     /* Joint intensity coding index: 0, 0 */

  827.     for (ch = 0; ch < c->fullband_channels; ch++)

  828.         put_bits(&c->pb, 3, 0);

  829. 

  830.     /* Transient mode codebook: A4, A4 (arbitrary) */

  831.     for (ch = 0; ch < c->fullband_channels; ch++)

  832.         put_bits(&c->pb, 2, 0);

  833. 

  834.     /* Scale factor code book: 7 bit linear, 7-bit sqrt table (for each channel) */

  835.     for (ch = 0; ch < c->fullband_channels; ch++)

  836.         put_bits(&c->pb, 3, 6);

  837. 

  838.     /* Bit allocation quantizer select: linear 5-bit */

  839.     for (ch = 0; ch < c->fullband_channels; ch++)

  840.         put_bits(&c->pb, 3, 6);

  841. 

  842.     /* Quantization index codebook select: dummy data

  843.        to avoid transmission of scale factor adjustment */

  844.     for (i = 1; i < 11; i++)

  845.         for (ch = 0; ch < c->fullband_channels; ch++)

  846.             put_bits(&c->pb, bitlen[i], thr[i]);

  847. 

  848.     /* Scale factor adjustment index: not transmitted */

  849.     /* Audio header CRC check word: not transmitted */

  850. }

  851. 

  852. static void put_subframe_samples(DCAEncContext *c, int ss, int band, int ch)

  853. {

  854.     if (c->abits[ch][band] <= 7) {

  855.         int sum, i, j;

  856.         for (i = 0; i < 8; i += 4) {

  857.             sum = 0;

  858.             for (j = 3; j >= 0; j--) {

  859.                 sum *= ff_dca_quant_levels[c->abits[ch][band]];

  860.                 sum += c->quantized[ch][band][ss * 8 + i + j];

  861.                 sum += (ff_dca_quant_levels[c->abits[ch][band]] - 1) / 2;

  862.             }

  863.             put_bits(&c->pb, bit_consumption[c->abits[ch][band]] / 4, sum);

  864.         }

  865.     } else {

  866.         int i;

  867.         for (i = 0; i < 8; i++) {

  868.             int bits = bit_consumption[c->abits[ch][band]] / 16;

  869.             put_sbits(&c->pb, bits, c->quantized[ch][band][ss * 8 + i]);

  870.         }

  871.     }

  872. }

  873. 

  874. static void put_subframe(DCAEncContext *c, int subframe)

  875. {

  876.     int i, band, ss, ch;

  877. 

  878.     /* Subsubframes count */

  879.     put_bits(&c->pb, 2, SUBSUBFRAMES -1);

  880. 

  881.     /* Partial subsubframe sample count: dummy */

  882.     put_bits(&c->pb, 3, 0);

  883. 

  884.     /* Prediction mode: no ADPCM, in each channel and subband */

  885.     for (ch = 0; ch < c->fullband_channels; ch++)

  886.         for (band = 0; band < DCAENC_SUBBANDS; band++)

  887.             put_bits(&c->pb, 1, 0);

  888. 

  889.     /* Prediction VQ address: not transmitted */

  890.     /* Bit allocation index */

  891.     for (ch = 0; ch < c->fullband_channels; ch++)

  892.         for (band = 0; band < DCAENC_SUBBANDS; band++)

  893.             put_bits(&c->pb, 5, c->abits[ch][band]);

  894. 

  895.     if (SUBSUBFRAMES > 1) {

  896.         /* Transition mode: none for each channel and subband */

  897.         for (ch = 0; ch < c->fullband_channels; ch++)

  898.             for (band = 0; band < DCAENC_SUBBANDS; band++)

  899.                 put_bits(&c->pb, 1, 0); /* codebook A4 */

  900.     }

  901. 

  902.     /* Scale factors */

  903.     for (ch = 0; ch < c->fullband_channels; ch++)

  904.         for (band = 0; band < DCAENC_SUBBANDS; band++)

  905.             put_bits(&c->pb, 7, c->scale_factor[ch][band]);

  906. 

  907.     /* Joint subband scale factor codebook select: not transmitted */

  908.     /* Scale factors for joint subband coding: not transmitted */

  909.     /* Stereo down-mix coefficients: not transmitted */

  910.     /* Dynamic range coefficient: not transmitted */

  911.     /* Stde information CRC check word: not transmitted */

  912.     /* VQ encoded high frequency subbands: not transmitted */

  913. 

  914.     /* LFE data: 8 samples and scalefactor */

  915.     if (c->lfe_channel) {

  916.         for (i = 0; i < DCA_LFE_SAMPLES; i++)

  917.             put_bits(&c->pb, 8, quantize_value(c->downsampled_lfe[i], c->lfe_quant) & 0xff);

  918.         put_bits(&c->pb, 8, c->lfe_scale_factor);

  919.     }

  920. 

  921.     /* Audio data (subsubframes) */

  922.     for (ss = 0; ss < SUBSUBFRAMES ; ss++)

  923.         for (ch = 0; ch < c->fullband_channels; ch++)

  924.             for (band = 0; band < DCAENC_SUBBANDS; band++)

  925.                     put_subframe_samples(c, ss, band, ch);

  926. 

  927.     /* DSYNC */

  928.     put_bits(&c->pb, 16, 0xffff);

  929. }

  930. 

  931. static int encode_frame(AVCodecContext *avctx, AVPacket *avpkt,

  932.                         const AVFrame *frame, int *got_packet_ptr)

  933. {

  934.     DCAEncContext *c = avctx->priv_data;

  935.     const int32_t *samples;

  936.     int ret, i;

  937. 

  938.     if ((ret = ff_alloc_packet2(avctx, avpkt, c->frame_size, 0)) < 0)

  939.         return ret;

  940. 

  941.     samples = (const int32_t *)frame->data[0];

  942. 

  943.     subband_transform(c, samples);

  944.     if (c->lfe_channel)

  945.         lfe_downsample(c, samples);

  946. 

  947.     calc_masking(c, samples);

  948.     find_peaks(c);

  949.     assign_bits(c);

  950.     calc_scales(c);

  951.     quantize_all(c);

  952.     shift_history(c, samples);

  953. 

  954.     init_put_bits(&c->pb, avpkt->data, avpkt->size);

  955.     put_frame_header(c);

  956.     put_primary_audio_header(c);

  957.     for (i = 0; i < SUBFRAMES; i++)

  958.         put_subframe(c, i);

  959. 

  960. 

  961.     for (i = put_bits_count(&c->pb); i < 8*c->frame_size; i++)

  962.         put_bits(&c->pb, 1, 0);

  963. 

  964.     flush_put_bits(&c->pb);

  965. 

  966.     avpkt->pts      = frame->pts;

  967.     avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);

  968.     avpkt->size     = put_bits_count(&c->pb) >> 3;

  969.     *got_packet_ptr = 1;

  970.     return 0;

  971. }

  972. 

  973. static const AVCodecDefault defaults[] = {

  974.     { "b",          "1411200" },

  975.     { NULL },

  976. };

  977. 

  978. AVCodec ff_dca_encoder = {

  979.     .name                  = "dca",

  980.     .long_name             = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),

  981.     .type                  = AVMEDIA_TYPE_AUDIO,

  982.     .id                    = AV_CODEC_ID_DTS,

  983.     .priv_data_size        = sizeof(DCAEncContext),

  984.     .init                  = encode_init,

  985.     .encode2               = encode_frame,

  986.     .capabilities          = AV_CODEC_CAP_EXPERIMENTAL,

  987.     .sample_fmts           = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32,

  988.                                                             AV_SAMPLE_FMT_NONE },

  989.     .supported_samplerates = sample_rates,

  990.     .channel_layouts       = (const uint64_t[]) { AV_CH_LAYOUT_MONO,

  991.                                                   AV_CH_LAYOUT_STEREO,

  992.                                                   AV_CH_LAYOUT_2_2,

  993.                                                   AV_CH_LAYOUT_5POINT0,

  994.                                                   AV_CH_LAYOUT_5POINT1,

  995.                                                   0 },

  996.     .defaults              = defaults,

  997. };