yading@10: /*
yading@10:  * AC-3 encoder float/fixed template
yading@10:  * Copyright (c) 2000 Fabrice Bellard
yading@10:  * Copyright (c) 2006-2011 Justin Ruggles <justin.ruggles@gmail.com>
yading@10:  * Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * AC-3 encoder float/fixed template
yading@10:  */
yading@10: 
yading@10: #include <stdint.h>
yading@10: 
yading@10: #include "libavutil/internal.h"
yading@10: 
yading@10: /* prototypes for static functions in ac3enc_fixed.c and ac3enc_float.c */
yading@10: 
yading@10: static void scale_coefficients(AC3EncodeContext *s);
yading@10: 
yading@10: static void apply_window(void *dsp, SampleType *output,
yading@10:                          const SampleType *input, const SampleType *window,
yading@10:                          unsigned int len);
yading@10: 
yading@10: static int normalize_samples(AC3EncodeContext *s);
yading@10: 
yading@10: static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
yading@10: 
yading@10: static CoefType calc_cpl_coord(CoefSumType energy_ch, CoefSumType energy_cpl);
yading@10: 
yading@10: static void sum_square_butterfly(AC3EncodeContext *s, CoefSumType sum[4],
yading@10:                                  const CoefType *coef0, const CoefType *coef1,
yading@10:                                  int len);
yading@10: 
yading@10: int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
yading@10: {
yading@10:     int ch;
yading@10: 
yading@10:     FF_ALLOC_OR_GOTO(s->avctx, s->windowed_samples, AC3_WINDOW_SIZE *
yading@10:                      sizeof(*s->windowed_samples), alloc_fail);
yading@10:     FF_ALLOC_OR_GOTO(s->avctx, s->planar_samples, s->channels * sizeof(*s->planar_samples),
yading@10:                      alloc_fail);
yading@10:     for (ch = 0; ch < s->channels; ch++) {
yading@10:         FF_ALLOCZ_OR_GOTO(s->avctx, s->planar_samples[ch],
yading@10:                           (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples),
yading@10:                           alloc_fail);
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: alloc_fail:
yading@10:     return AVERROR(ENOMEM);
yading@10: }
yading@10: 
yading@10: 
yading@10: /*
yading@10:  * Copy input samples.
yading@10:  * Channels are reordered from FFmpeg's default order to AC-3 order.
yading@10:  */
yading@10: static void copy_input_samples(AC3EncodeContext *s, SampleType **samples)
yading@10: {
yading@10:     int ch;
yading@10: 
yading@10:     /* copy and remap input samples */
yading@10:     for (ch = 0; ch < s->channels; ch++) {
yading@10:         /* copy last 256 samples of previous frame to the start of the current frame */
yading@10:         memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
yading@10:                AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
yading@10: 
yading@10:         /* copy new samples for current frame */
yading@10:         memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE],
yading@10:                samples[s->channel_map[ch]],
yading@10:                AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0]));
yading@10:     }
yading@10: }
yading@10: 
yading@10: 
yading@10: /*
yading@10:  * Apply the MDCT to input samples to generate frequency coefficients.
yading@10:  * This applies the KBD window and normalizes the input to reduce precision
yading@10:  * loss due to fixed-point calculations.
yading@10:  */
yading@10: static void apply_mdct(AC3EncodeContext *s)
yading@10: {
yading@10:     int blk, ch;
yading@10: 
yading@10:     for (ch = 0; ch < s->channels; ch++) {
yading@10:         for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:             AC3Block *block = &s->blocks[blk];
yading@10:             const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
yading@10: 
yading@10: #if CONFIG_AC3ENC_FLOAT
yading@10:             apply_window(&s->fdsp, s->windowed_samples, input_samples,
yading@10:                          s->mdct_window, AC3_WINDOW_SIZE);
yading@10: #else
yading@10:             apply_window(&s->dsp, s->windowed_samples, input_samples,
yading@10:                          s->mdct_window, AC3_WINDOW_SIZE);
yading@10: #endif
yading@10: 
yading@10:             if (s->fixed_point)
yading@10:                 block->coeff_shift[ch+1] = normalize_samples(s);
yading@10: 
yading@10:             s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1],
yading@10:                                s->windowed_samples);
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: 
yading@10: /*
yading@10:  * Calculate coupling channel and coupling coordinates.
yading@10:  */
yading@10: static void apply_channel_coupling(AC3EncodeContext *s)
yading@10: {
yading@10:     LOCAL_ALIGNED_16(CoefType, cpl_coords,      [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
yading@10: #if CONFIG_AC3ENC_FLOAT
yading@10:     LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
yading@10: #else
yading@10:     int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
yading@10: #endif
yading@10:     int av_uninit(blk), ch, bnd, i, j;
yading@10:     CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
yading@10:     int cpl_start, num_cpl_coefs;
yading@10: 
yading@10:     memset(cpl_coords,       0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
yading@10: #if CONFIG_AC3ENC_FLOAT
yading@10:     memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
yading@10: #endif
yading@10: 
yading@10:     /* align start to 16-byte boundary. align length to multiple of 32.
yading@10:         note: coupling start bin % 4 will always be 1 */
yading@10:     cpl_start     = s->start_freq[CPL_CH] - 1;
yading@10:     num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
yading@10:     cpl_start     = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
yading@10: 
yading@10:     /* calculate coupling channel from fbw channels */
yading@10:     for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:         AC3Block *block = &s->blocks[blk];
yading@10:         CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
yading@10:         if (!block->cpl_in_use)
yading@10:             continue;
yading@10:         memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
yading@10:         for (ch = 1; ch <= s->fbw_channels; ch++) {
yading@10:             CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
yading@10:             if (!block->channel_in_cpl[ch])
yading@10:                 continue;
yading@10:             for (i = 0; i < num_cpl_coefs; i++)
yading@10:                 cpl_coef[i] += ch_coef[i];
yading@10:         }
yading@10: 
yading@10:         /* coefficients must be clipped in order to be encoded */
yading@10:         clip_coefficients(&s->dsp, cpl_coef, num_cpl_coefs);
yading@10:     }
yading@10: 
yading@10:     /* calculate energy in each band in coupling channel and each fbw channel */
yading@10:     /* TODO: possibly use SIMD to speed up energy calculation */
yading@10:     bnd = 0;
yading@10:     i = s->start_freq[CPL_CH];
yading@10:     while (i < s->cpl_end_freq) {
yading@10:         int band_size = s->cpl_band_sizes[bnd];
yading@10:         for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
yading@10:             for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:                 AC3Block *block = &s->blocks[blk];
yading@10:                 if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
yading@10:                     continue;
yading@10:                 for (j = 0; j < band_size; j++) {
yading@10:                     CoefType v = block->mdct_coef[ch][i+j];
yading@10:                     MAC_COEF(energy[blk][ch][bnd], v, v);
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         i += band_size;
yading@10:         bnd++;
yading@10:     }
yading@10: 
yading@10:     /* calculate coupling coordinates for all blocks for all channels */
yading@10:     for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:         AC3Block *block  = &s->blocks[blk];
yading@10:         if (!block->cpl_in_use)
yading@10:             continue;
yading@10:         for (ch = 1; ch <= s->fbw_channels; ch++) {
yading@10:             if (!block->channel_in_cpl[ch])
yading@10:                 continue;
yading@10:             for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
yading@10:                 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
yading@10:                                                           energy[blk][CPL_CH][bnd]);
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     /* determine which blocks to send new coupling coordinates for */
yading@10:     for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:         AC3Block *block  = &s->blocks[blk];
yading@10:         AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
yading@10: 
yading@10:         memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
yading@10: 
yading@10:         if (block->cpl_in_use) {
yading@10:             /* send new coordinates if this is the first block, if previous
yading@10:              * block did not use coupling but this block does, the channels
yading@10:              * using coupling has changed from the previous block, or the
yading@10:              * coordinate difference from the last block for any channel is
yading@10:              * greater than a threshold value. */
yading@10:             if (blk == 0 || !block0->cpl_in_use) {
yading@10:                 for (ch = 1; ch <= s->fbw_channels; ch++)
yading@10:                     block->new_cpl_coords[ch] = 1;
yading@10:             } else {
yading@10:                 for (ch = 1; ch <= s->fbw_channels; ch++) {
yading@10:                     if (!block->channel_in_cpl[ch])
yading@10:                         continue;
yading@10:                     if (!block0->channel_in_cpl[ch]) {
yading@10:                         block->new_cpl_coords[ch] = 1;
yading@10:                     } else {
yading@10:                         CoefSumType coord_diff = 0;
yading@10:                         for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
yading@10:                             coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
yading@10:                                                 cpl_coords[blk  ][ch][bnd]);
yading@10:                         }
yading@10:                         coord_diff /= s->num_cpl_bands;
yading@10:                         if (coord_diff > NEW_CPL_COORD_THRESHOLD)
yading@10:                             block->new_cpl_coords[ch] = 1;
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     /* calculate final coupling coordinates, taking into account reusing of
yading@10:        coordinates in successive blocks */
yading@10:     for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
yading@10:         blk = 0;
yading@10:         while (blk < s->num_blocks) {
yading@10:             int av_uninit(blk1);
yading@10:             AC3Block *block  = &s->blocks[blk];
yading@10: 
yading@10:             if (!block->cpl_in_use) {
yading@10:                 blk++;
yading@10:                 continue;
yading@10:             }
yading@10: 
yading@10:             for (ch = 1; ch <= s->fbw_channels; ch++) {
yading@10:                 CoefSumType energy_ch, energy_cpl;
yading@10:                 if (!block->channel_in_cpl[ch])
yading@10:                     continue;
yading@10:                 energy_cpl = energy[blk][CPL_CH][bnd];
yading@10:                 energy_ch = energy[blk][ch][bnd];
yading@10:                 blk1 = blk+1;
yading@10:                 while (!s->blocks[blk1].new_cpl_coords[ch] && blk1 < s->num_blocks) {
yading@10:                     if (s->blocks[blk1].cpl_in_use) {
yading@10:                         energy_cpl += energy[blk1][CPL_CH][bnd];
yading@10:                         energy_ch += energy[blk1][ch][bnd];
yading@10:                     }
yading@10:                     blk1++;
yading@10:                 }
yading@10:                 cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
yading@10:             }
yading@10:             blk = blk1;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     /* calculate exponents/mantissas for coupling coordinates */
yading@10:     for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:         AC3Block *block = &s->blocks[blk];
yading@10:         if (!block->cpl_in_use)
yading@10:             continue;
yading@10: 
yading@10: #if CONFIG_AC3ENC_FLOAT
yading@10:         s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
yading@10:                                    cpl_coords[blk][1],
yading@10:                                    s->fbw_channels * 16);
yading@10: #endif
yading@10:         s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
yading@10:                                     fixed_cpl_coords[blk][1],
yading@10:                                     s->fbw_channels * 16);
yading@10: 
yading@10:         for (ch = 1; ch <= s->fbw_channels; ch++) {
yading@10:             int bnd, min_exp, max_exp, master_exp;
yading@10: 
yading@10:             if (!block->new_cpl_coords[ch])
yading@10:                 continue;
yading@10: 
yading@10:             /* determine master exponent */
yading@10:             min_exp = max_exp = block->cpl_coord_exp[ch][0];
yading@10:             for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
yading@10:                 int exp = block->cpl_coord_exp[ch][bnd];
yading@10:                 min_exp = FFMIN(exp, min_exp);
yading@10:                 max_exp = FFMAX(exp, max_exp);
yading@10:             }
yading@10:             master_exp = ((max_exp - 15) + 2) / 3;
yading@10:             master_exp = FFMAX(master_exp, 0);
yading@10:             while (min_exp < master_exp * 3)
yading@10:                 master_exp--;
yading@10:             for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
yading@10:                 block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
yading@10:                                                         master_exp * 3, 0, 15);
yading@10:             }
yading@10:             block->cpl_master_exp[ch] = master_exp;
yading@10: 
yading@10:             /* quantize mantissas */
yading@10:             for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
yading@10:                 int cpl_exp  = block->cpl_coord_exp[ch][bnd];
yading@10:                 int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
yading@10:                 if (cpl_exp == 15)
yading@10:                     cpl_mant >>= 1;
yading@10:                 else
yading@10:                     cpl_mant -= 16;
yading@10: 
yading@10:                 block->cpl_coord_mant[ch][bnd] = cpl_mant;
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     if (CONFIG_EAC3_ENCODER && s->eac3)
yading@10:         ff_eac3_set_cpl_states(s);
yading@10: }
yading@10: 
yading@10: 
yading@10: /*
yading@10:  * Determine rematrixing flags for each block and band.
yading@10:  */
yading@10: static void compute_rematrixing_strategy(AC3EncodeContext *s)
yading@10: {
yading@10:     int nb_coefs;
yading@10:     int blk, bnd;
yading@10:     AC3Block *block, *block0 = NULL;
yading@10: 
yading@10:     if (s->channel_mode != AC3_CHMODE_STEREO)
yading@10:         return;
yading@10: 
yading@10:     for (blk = 0; blk < s->num_blocks; blk++) {
yading@10:         block = &s->blocks[blk];
yading@10:         block->new_rematrixing_strategy = !blk;
yading@10: 
yading@10:         block->num_rematrixing_bands = 4;
yading@10:         if (block->cpl_in_use) {
yading@10:             block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
yading@10:             block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
yading@10:             if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
yading@10:                 block->new_rematrixing_strategy = 1;
yading@10:         }
yading@10:         nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
yading@10: 
yading@10:         if (!s->rematrixing_enabled) {
yading@10:             block0 = block;
yading@10:             continue;
yading@10:         }
yading@10: 
yading@10:         for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
yading@10:             /* calculate calculate sum of squared coeffs for one band in one block */
yading@10:             int start = ff_ac3_rematrix_band_tab[bnd];
yading@10:             int end   = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
yading@10:             CoefSumType sum[4];
yading@10:             sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
yading@10:                                  block->mdct_coef[2] + start, end - start);
yading@10: 
yading@10:             /* compare sums to determine if rematrixing will be used for this band */
yading@10:             if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
yading@10:                 block->rematrixing_flags[bnd] = 1;
yading@10:             else
yading@10:                 block->rematrixing_flags[bnd] = 0;
yading@10: 
yading@10:             /* determine if new rematrixing flags will be sent */
yading@10:             if (blk &&
yading@10:                 block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
yading@10:                 block->new_rematrixing_strategy = 1;
yading@10:             }
yading@10:         }
yading@10:         block0 = block;
yading@10:     }
yading@10: }
yading@10: 
yading@10: 
yading@10: int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
yading@10:                            const AVFrame *frame, int *got_packet_ptr)
yading@10: {
yading@10:     AC3EncodeContext *s = avctx->priv_data;
yading@10:     int ret;
yading@10: 
yading@10:     if (s->options.allow_per_frame_metadata) {
yading@10:         ret = ff_ac3_validate_metadata(s);
yading@10:         if (ret)
yading@10:             return ret;
yading@10:     }
yading@10: 
yading@10:     if (s->bit_alloc.sr_code == 1 || s->eac3)
yading@10:         ff_ac3_adjust_frame_size(s);
yading@10: 
yading@10:     copy_input_samples(s, (SampleType **)frame->extended_data);
yading@10: 
yading@10:     apply_mdct(s);
yading@10: 
yading@10:     if (s->fixed_point)
yading@10:         scale_coefficients(s);
yading@10: 
yading@10:     clip_coefficients(&s->dsp, s->blocks[0].mdct_coef[1],
yading@10:                       AC3_MAX_COEFS * s->num_blocks * s->channels);
yading@10: 
yading@10:     s->cpl_on = s->cpl_enabled;
yading@10:     ff_ac3_compute_coupling_strategy(s);
yading@10: 
yading@10:     if (s->cpl_on)
yading@10:         apply_channel_coupling(s);
yading@10: 
yading@10:     compute_rematrixing_strategy(s);
yading@10: 
yading@10:     if (!s->fixed_point)
yading@10:         scale_coefficients(s);
yading@10: 
yading@10:     ff_ac3_apply_rematrixing(s);
yading@10: 
yading@10:     ff_ac3_process_exponents(s);
yading@10: 
yading@10:     ret = ff_ac3_compute_bit_allocation(s);
yading@10:     if (ret) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n");
yading@10:         return ret;
yading@10:     }
yading@10: 
yading@10:     ff_ac3_group_exponents(s);
yading@10: 
yading@10:     ff_ac3_quantize_mantissas(s);
yading@10: 
yading@10:     if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size)) < 0)
yading@10:         return ret;
yading@10:     ff_ac3_output_frame(s, avpkt->data);
yading@10: 
yading@10:     if (frame->pts != AV_NOPTS_VALUE)
yading@10:         avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay);
yading@10: 
yading@10:     *got_packet_ptr = 1;
yading@10:     return 0;
yading@10: }