diff options
author | Alex Converse <alex.converse@gmail.com> | 2010-06-19 14:14:51 +0000 |
---|---|---|
committer | Alex Converse <alex.converse@gmail.com> | 2010-06-19 14:14:51 +0000 |
commit | a20639017bfca0490bb1799575714f22bf470b4f (patch) | |
tree | 7ece4b88c997a02342f75f91d1b928dd67257681 /libavcodec/ps.c | |
parent | 4d49a5a785220c3430739de8b1e340a4ea3f1864 (diff) | |
download | ffmpeg-a20639017bfca0490bb1799575714f22bf470b4f.tar.gz |
Add HE-AAC v2 support to the AAC decoder.
Originally committed as revision 23647 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/ps.c')
-rw-r--r-- | libavcodec/ps.c | 1124 |
1 files changed, 1124 insertions, 0 deletions
diff --git a/libavcodec/ps.c b/libavcodec/ps.c new file mode 100644 index 0000000000..1f2922639d --- /dev/null +++ b/libavcodec/ps.c @@ -0,0 +1,1124 @@ +/* + * MPEG-4 Parametric Stereo decoding functions + * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> + * + * 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 <stdint.h> +#include "libavutil/mathematics.h" +#include "avcodec.h" +#include "get_bits.h" +#include "ps.h" +#include "ps_tablegen.h" +#include "psdata.c" + +#define PS_BASELINE 0 + +#define numQMFSlots 32 //numTimeSlots * RATE + +static const int8_t num_env_tab[2][4] = { + { 0, 1, 2, 4, }, + { 1, 2, 3, 4, }, +}; + +static const int8_t nr_iidicc_par_tab[] = { + 10, 20, 34, 10, 20, 34, +}; + +static const int8_t nr_iidopd_par_tab[] = { + 5, 11, 17, 5, 11, 17, +}; + +enum { + huff_iid_df1, + huff_iid_dt1, + huff_iid_df0, + huff_iid_dt0, + huff_icc_df, + huff_icc_dt, + huff_ipd_df, + huff_ipd_dt, + huff_opd_df, + huff_opd_dt, +}; + +static const int huff_iid[] = { + huff_iid_df0, + huff_iid_df1, + huff_iid_dt0, + huff_iid_dt1, +}; + +static VLC vlc_ps[10]; + +/** + * Read Inter-channel Intensity Difference parameters from the bitstream. + * + * @param avctx contains the current codec context + * @param gb pointer to the input bitstream + * @param ps pointer to the Parametric Stereo context + * @param e envelope to decode + * @param dt 1: time delta-coded, 0: frequency delta-coded + */ +static int iid_data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, int e, int dt) +{ + int b; + int table_idx = huff_iid[2*dt+ps->iid_quant]; + VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; + if (dt) { + int e_prev = e ? e - 1 : ps->num_env_old - 1; + e_prev = FFMAX(e_prev, 0); + for (b = 0; b < ps->nr_iid_par; b++) { + ps->iid_par[e][b] = ps->iid_par[e_prev][b] + + get_vlc2(gb, vlc_table, 9, 3) - + huff_offset[table_idx]; + if (FFABS(ps->iid_par[e][b]) > 7 + 8 * ps->iid_quant) + goto err; + } + } else { + int prev = 0; + for (b = 0; b < ps->nr_iid_par; b++) { + prev += get_vlc2(gb, vlc_table, 9, 3) - + huff_offset[table_idx]; + ps->iid_par[e][b] = prev; + if (FFABS(ps->iid_par[e][b]) > 7 + 8 * ps->iid_quant) + goto err; + } + } + return 0; +err: + av_log(avctx, AV_LOG_ERROR, "illegal iid\n"); + return -1; +} + +/** + * Read Inter-Channel Coherence parameters from the bitstream. + * + * @param avctx contains the current codec context + * @param gb pointer to the input bitstream + * @param ps pointer to the Parametric Stereo context + * @param e envelope to decode + * @param dt 1: time delta-coded, 0: frequency delta-coded + */ +static int icc_data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, int e, int dt) +{ + int b; + int table_idx = dt ? huff_icc_dt : huff_icc_df; + VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; + if (dt) { + int e_prev = e ? e - 1 : ps->num_env_old - 1; + e_prev = FFMAX(e_prev, 0); + for (b = 0; b < ps->nr_icc_par; b++) { + ps->icc_par[e][b] = ps->icc_par[e_prev][b] + get_vlc2(gb, vlc_table, 9, 3) - huff_offset[table_idx]; + if (ps->icc_par[e][b] > 7U) + goto err; + } + } else { + int prev = 0; + for (b = 0; b < ps->nr_icc_par; b++) { + prev += get_vlc2(gb, vlc_table, 9, 3) - huff_offset[table_idx]; + ps->icc_par[e][b] = prev; + if (ps->icc_par[e][b] > 7U) + goto err; + } + } + return 0; +err: + av_log(avctx, AV_LOG_ERROR, "illegal icc\n"); + return -1; +} + +/** + * Read Inter-channel Phase Difference parameters from the bitstream. + * + * @param gb pointer to the input bitstream + * @param ps pointer to the Parametric Stereo context + * @param e envelope to decode + * @param dt 1: time delta-coded, 0: frequency delta-coded + */ +static void ipd_data(GetBitContext *gb, PSContext *ps, int e, int dt) +{ + int b; + int table_idx = dt ? huff_ipd_dt : huff_ipd_df; + VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; + if (dt) { + int e_prev = e ? e - 1 : ps->num_env_old - 1; + e_prev = FFMAX(e_prev, 0); + for (b = 0; b < ps->nr_ipdopd_par; b++) { + ps->ipd_par[e][b] = (ps->ipd_par[e_prev][b] + get_vlc2(gb, vlc_table, 9, 1)) & 0x07; + } + } else { + int prev = 0; + for (b = 0; b < ps->nr_ipdopd_par; b++) { + prev += get_vlc2(gb, vlc_table, 9, 3); + prev &= 0x07; + ps->ipd_par[e][b] = prev; + } + } +} + +/** + * Read Overall Phase Difference parameters from the bitstream. + * + * @param gb pointer to the input bitstream + * @param ps pointer to the Parametric Stereo context + * @param e envelope to decode + * @param dt 1: time delta-coded, 0: frequency delta-coded + */ +static void opd_data(GetBitContext *gb, PSContext *ps, int e, int dt) +{ + int b; + int table_idx = dt ? huff_opd_dt : huff_opd_df; + VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; + if (dt) { + int e_prev = e ? e - 1 : ps->num_env_old - 1; + e_prev = FFMAX(e_prev, 0); + for (b = 0; b < ps->nr_ipdopd_par; b++) { + ps->opd_par[e][b] = (ps->opd_par[e_prev][b] + get_vlc2(gb, vlc_table, 9, 1)) & 0x07; + } + } else { + int prev = 0; + for (b = 0; b < ps->nr_ipdopd_par; b++) { + prev += get_vlc2(gb, vlc_table, 9, 3); + prev &= 0x07; + ps->opd_par[e][b] = prev; + } + } +} + +static int ps_extension(GetBitContext *gb, PSContext *ps, int ps_extension_id) +{ + int e; + int count = get_bits_count(gb); + + if (ps_extension_id) + return 0; + + ps->enable_ipdopd = get_bits1(gb); + if (ps->enable_ipdopd) { + for (e = 0; e < ps->num_env; e++) { + int dt = get_bits1(gb); + ipd_data(gb, ps, e, dt); + dt = get_bits1(gb); + opd_data(gb, ps, e, dt); + } + } + skip_bits1(gb); //reserved_ps + return get_bits_count(gb) - count; +} + +static void ipdopd_reset(int8_t *opd_hist, int8_t *ipd_hist) +{ + int i; + for (i = 0; i < PS_MAX_NR_IPDOPD; i++) { + opd_hist[i] = 0; + ipd_hist[i] = 0; + } +} + +int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb_host, PSContext *ps, int bits_left) +{ + int e; + int bit_count_start = get_bits_count(gb_host); + int header; + int bits_consumed; + GetBitContext gbc = *gb_host, *gb = &gbc; + + header = get_bits1(gb); + if (header) { //enable_ps_header + ps->enable_iid = get_bits1(gb); + if (ps->enable_iid) { + ps->iid_mode = get_bits(gb, 3); + if (ps->iid_mode > 5) { + av_log(avctx, AV_LOG_ERROR, "iid_mode %d is reserved.\n", + ps->iid_mode); + goto err; + } + ps->nr_iid_par = nr_iidicc_par_tab[ps->iid_mode]; + ps->iid_quant = ps->iid_mode > 2; + ps->nr_ipdopd_par = nr_iidopd_par_tab[ps->iid_mode]; + } + ps->enable_icc = get_bits1(gb); + if (ps->enable_icc) { + ps->icc_mode = get_bits(gb, 3); + if (ps->icc_mode > 5) { + av_log(avctx, AV_LOG_ERROR, "icc_mode %d is reserved.\n", + ps->icc_mode); + goto err; + } + ps->nr_icc_par = nr_iidicc_par_tab[ps->icc_mode]; + } + ps->enable_ext = get_bits1(gb); + } + + ps->frame_class = get_bits1(gb); + ps->num_env_old = ps->num_env; + ps->num_env = num_env_tab[ps->frame_class][get_bits(gb, 2)]; + + ps->border_position[0] = -1; + if (ps->frame_class) { + for (e = 1; e <= ps->num_env; e++) + ps->border_position[e] = get_bits(gb, 5); + } else + for (e = 1; e <= ps->num_env; e++) + ps->border_position[e] = e * numQMFSlots / ps->num_env - 1; + + if (ps->enable_iid) { + for (e = 0; e < ps->num_env; e++) { + int dt = get_bits1(gb); + if (iid_data(avctx, gb, ps, e, dt)) + goto err; + } + } else + memset(ps->iid_par, 0, sizeof(ps->iid_par)); + + if (ps->enable_icc) + for (e = 0; e < ps->num_env; e++) { + int dt = get_bits1(gb); + if (icc_data(avctx, gb, ps, e, dt)) + goto err; + } + else + memset(ps->icc_par, 0, sizeof(ps->icc_par)); + + if (ps->enable_ext) { + int cnt = get_bits(gb, 4); + if (cnt == 15) { + cnt += get_bits(gb, 8); + } + cnt *= 8; + while (cnt > 7) { + int ps_extension_id = get_bits(gb, 2); + cnt -= 2 + ps_extension(gb, ps, ps_extension_id); + } + if (cnt < 0) { + av_log(avctx, AV_LOG_ERROR, "ps extension overflow %d", cnt); + goto err; + } + skip_bits(gb, cnt); + } + + ps->enable_ipdopd &= !PS_BASELINE; + + //Fix up envelopes + if (!ps->num_env || ps->border_position[ps->num_env] < numQMFSlots - 1) { + //Create a fake envelope + int source = ps->num_env ? ps->num_env - 1 : ps->num_env_old - 1; + if (source >= 0 && source != ps->num_env) { + if (ps->enable_iid && ps->num_env_old > 1) { + memcpy(ps->iid_par+ps->num_env, ps->iid_par+source, sizeof(ps->iid_par[0])); + } + if (ps->enable_icc && ps->num_env_old > 1) { + memcpy(ps->icc_par+ps->num_env, ps->icc_par+source, sizeof(ps->icc_par[0])); + } + if (ps->enable_ipdopd && ps->num_env_old > 1) { + memcpy(ps->ipd_par+ps->num_env, ps->ipd_par+source, sizeof(ps->ipd_par[0])); + memcpy(ps->opd_par+ps->num_env, ps->opd_par+source, sizeof(ps->opd_par[0])); + } + } + ps->num_env++; + ps->border_position[ps->num_env] = numQMFSlots - 1; + } + + + ps->is34bands_old = ps->is34bands; + if (!PS_BASELINE && (ps->enable_iid || ps->enable_icc)) + ps->is34bands = (ps->enable_iid && ps->nr_iid_par == 34) || + (ps->enable_icc && ps->nr_icc_par == 34); + + //Baseline + if (!ps->enable_ipdopd) { + memset(ps->ipd_par, 0, sizeof(ps->ipd_par)); + memset(ps->opd_par, 0, sizeof(ps->opd_par)); + } + + if (header) + ps->start = 1; + + bits_consumed = get_bits_count(gb) - bit_count_start; + if (bits_consumed <= bits_left) { + skip_bits_long(gb_host, bits_consumed); + return bits_consumed; + } + av_log(avctx, AV_LOG_ERROR, "Expected to read %d PS bits actually read %d.\n", bits_left, bits_consumed); +err: + ps->start = 0; + skip_bits_long(gb_host, bits_left); + return bits_left; +} + +/** Split one subband into 2 subsubbands with a symmetric real filter. + * The filter must have its non-center even coefficients equal to zero. */ +static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[7], int len, int reverse) +{ + int i, j; + for (i = 0; i < len; i++) { + float re_in = filter[6] * in[6+i][0]; //real inphase + float re_op = 0.0f; //real out of phase + float im_in = filter[6] * in[6+i][1]; //imag inphase + float im_op = 0.0f; //imag out of phase + for (j = 0; j < 6; j += 2) { + re_op += filter[j+1] * (in[i+j+1][0] + in[12-j-1+i][0]); + im_op += filter[j+1] * (in[i+j+1][1] + in[12-j-1+i][1]); + } + out[ reverse][i][0] = re_in + re_op; + out[ reverse][i][1] = im_in + im_op; + out[!reverse][i][0] = re_in - re_op; + out[!reverse][i][1] = im_in - im_op; + } +} + +/** Split one subband into 6 subsubbands with a complex filter */ +static void hybrid6_cx(float (*in)[2], float (*out)[32][2], const float (*filter)[7][2], int len) +{ + int i, j, ssb; + int N = 8; + float temp[8][2]; + + for (i = 0; i < len; i++) { + for (ssb = 0; ssb < N; ssb++) { + float sum_re = filter[ssb][6][0] * in[i+6][0], sum_im = filter[ssb][6][0] * in[i+6][1]; + for (j = 0; j < 6; j++) { + float in0_re = in[i+j][0]; + float in0_im = in[i+j][1]; + float in1_re = in[i+12-j][0]; + float in1_im = in[i+12-j][1]; + sum_re += filter[ssb][j][0] * (in0_re + in1_re) - filter[ssb][j][1] * (in0_im - in1_im); + sum_im += filter[ssb][j][0] * (in0_im + in1_im) + filter[ssb][j][1] * (in0_re - in1_re); + } + temp[ssb][0] = sum_re; + temp[ssb][1] = sum_im; + } + out[0][i][0] = temp[6][0]; + out[0][i][1] = temp[6][1]; + out[1][i][0] = temp[7][0]; + out[1][i][1] = temp[7][1]; + out[2][i][0] = temp[0][0]; + out[2][i][1] = temp[0][1]; + out[3][i][0] = temp[1][0]; + out[3][i][1] = temp[1][1]; + out[4][i][0] = temp[2][0] + temp[5][0]; + out[4][i][1] = temp[2][1] + temp[5][1]; + out[5][i][0] = temp[3][0] + temp[4][0]; + out[5][i][1] = temp[3][1] + temp[4][1]; + } +} + +static void hybrid4_8_12_cx(float (*in)[2], float (*out)[32][2], const float (*filter)[7][2], int N, int len) +{ + int i, j, ssb; + + for (i = 0; i < len; i++) { + for (ssb = 0; ssb < N; ssb++) { + float sum_re = filter[ssb][6][0] * in[i+6][0], sum_im = filter[ssb][6][0] * in[i+6][1]; + for (j = 0; j < 6; j++) { + float in0_re = in[i+j][0]; + float in0_im = in[i+j][1]; + float in1_re = in[i+12-j][0]; + float in1_im = in[i+12-j][1]; + sum_re += filter[ssb][j][0] * (in0_re + in1_re) - filter[ssb][j][1] * (in0_im - in1_im); + sum_im += filter[ssb][j][0] * (in0_im + in1_im) + filter[ssb][j][1] * (in0_re - in1_re); + } + out[ssb][i][0] = sum_re; + out[ssb][i][1] = sum_im; + } + } +} + +static void hybrid_analysis(float out[91][32][2], float in[5][44][2], float L[2][38][64], int is34, int len) +{ + int i, j; + for (i = 0; i < 5; i++) { + for (j = 0; j < 38; j++) { + in[i][j+6][0] = L[0][j][i]; + in[i][j+6][1] = L[1][j][i]; + } + } + if(is34) { + hybrid4_8_12_cx(in[0], out, f34_0_12, 12, len); + hybrid4_8_12_cx(in[1], out+12, f34_1_8, 8, len); + hybrid4_8_12_cx(in[2], out+20, f34_2_4, 4, len); + hybrid4_8_12_cx(in[3], out+24, f34_2_4, 4, len); + hybrid4_8_12_cx(in[4], out+28, f34_2_4, 4, len); + for (i = 0; i < 59; i++) { + for (j = 0; j < len; j++) { + out[i+32][j][0] = L[0][j][i+5]; + out[i+32][j][1] = L[1][j][i+5]; + } + } + } else { + hybrid6_cx(in[0], out, f20_0_8, len); + hybrid2_re(in[1], out+6, g1_Q2, len, 1); + hybrid2_re(in[2], out+8, g1_Q2, len, 0); + for (i = 0; i < 61; i++) { + for (j = 0; j < len; j++) { + out[i+10][j][0] = L[0][j][i+3]; + out[i+10][j][1] = L[1][j][i+3]; + } + } + } + //update in_buf + for (i = 0; i < 5; i++) { + memcpy(in[i], in[i]+32, 6 * sizeof(in[i][0])); + } +} + +static void hybrid_synthesis(float out[2][38][64], float in[91][32][2], int is34, int len) +{ + int i, n; + if(is34) { + for (n = 0; n < len; n++) { + memset(out[0][n], 0, 5*sizeof(out[0][n][0])); + memset(out[1][n], 0, 5*sizeof(out[1][n][0])); + for(i = 0; i < 12; i++) { + out[0][n][0] += in[ i][n][0]; + out[1][n][0] += in[ i][n][1]; + } + for(i = 0; i < 8; i++) { + out[0][n][1] += in[12+i][n][0]; + out[1][n][1] += in[12+i][n][1]; + } + for(i = 0; i < 4; i++) { + out[0][n][2] += in[20+i][n][0]; + out[1][n][2] += in[20+i][n][1]; + out[0][n][3] += in[24+i][n][0]; + out[1][n][3] += in[24+i][n][1]; + out[0][n][4] += in[28+i][n][0]; + out[1][n][4] += in[28+i][n][1]; + } + } + for (i = 0; i < 59; i++) { + for (n = 0; n < len; n++) { + out[0][n][i+5] = in[i+32][n][0]; + out[1][n][i+5] = in[i+32][n][1]; + } + } + } else { + for (n = 0; n < len; n++) { + out[0][n][0] = in[0][n][0] + in[1][n][0] + in[2][n][0] + + in[3][n][0] + in[4][n][0] + in[5][n][0]; + out[1][n][0] = in[0][n][1] + in[1][n][1] + in[2][n][1] + + in[3][n][1] + in[4][n][1] + in[5][n][1]; + out[0][n][1] = in[6][n][0] + in[7][n][0]; + out[1][n][1] = in[6][n][1] + in[7][n][1]; + out[0][n][2] = in[8][n][0] + in[9][n][0]; + out[1][n][2] = in[8][n][1] + in[9][n][1]; + } + for (i = 0; i < 61; i++) { + for (n = 0; n < len; n++) { + out[0][n][i+3] = in[i+10][n][0]; + out[1][n][i+3] = in[i+10][n][1]; + } + } + } +} + +/// All-pass filter decay slope +#define DECAY_SLOPE 0.05f +/// Number of frequency bands that can be addressed by the parameter index, b(k) +static const int NR_PAR_BANDS[] = { 20, 34 }; +/// Number of frequency bands that can be addressed by the sub subband index, k +static const int NR_BANDS[] = { 71, 91 }; +/// Start frequency band for the all-pass filter decay slope +static const int DECAY_CUTOFF[] = { 10, 32 }; +/// Number of all-pass filer bands +static const int NR_ALLPASS_BANDS[] = { 30, 50 }; +/// First stereo band using the short one sample delay +static const int SHORT_DELAY_BAND[] = { 42, 62 }; + +/** Table 8.46 */ +static void map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full) +{ + int b; + if (full) + b = 9; + else { + b = 4; + par_mapped[10] = 0; + } + for (; b >= 0; b--) { + par_mapped[2*b+1] = par_mapped[2*b] = par[b]; + } +} + +static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full) +{ + par_mapped[ 0] = (2*par[ 0] + par[ 1]) / 3; + par_mapped[ 1] = ( par[ 1] + 2*par[ 2]) / 3; + par_mapped[ 2] = (2*par[ 3] + par[ 4]) / 3; + par_mapped[ 3] = ( par[ 4] + 2*par[ 5]) / 3; + par_mapped[ 4] = ( par[ 6] + par[ 7]) / 2; + par_mapped[ 5] = ( par[ 8] + par[ 9]) / 2; + par_mapped[ 6] = par[10]; + par_mapped[ 7] = par[11]; + par_mapped[ 8] = ( par[12] + par[13]) / 2; + par_mapped[ 9] = ( par[14] + par[15]) / 2; + par_mapped[10] = par[16]; + if (full) { + par_mapped[11] = par[17]; + par_mapped[12] = par[18]; + par_mapped[13] = par[19]; + par_mapped[14] = ( par[20] + par[21]) / 2; + par_mapped[15] = ( par[22] + par[23]) / 2; + par_mapped[16] = ( par[24] + par[25]) / 2; + par_mapped[17] = ( par[26] + par[27]) / 2; + par_mapped[18] = ( par[28] + par[29] + par[30] + par[31]) / 4; + par_mapped[19] = ( par[32] + par[33]) / 2; + } +} + +static void map_val_34_to_20(float par[PS_MAX_NR_IIDICC]) +{ + par[ 0] = (2*par[ 0] + par[ 1]) * 0.33333333f; + par[ 1] = ( par[ 1] + 2*par[ 2]) * 0.33333333f; + par[ 2] = (2*par[ 3] + par[ 4]) * 0.33333333f; + par[ 3] = ( par[ 4] + 2*par[ 5]) * 0.33333333f; + par[ 4] = ( par[ 6] + par[ 7]) * 0.5f; + par[ 5] = ( par[ 8] + par[ 9]) * 0.5f; + par[ 6] = par[10]; + par[ 7] = par[11]; + par[ 8] = ( par[12] + par[13]) * 0.5f; + par[ 9] = ( par[14] + par[15]) * 0.5f; + par[10] = par[16]; + par[11] = par[17]; + par[12] = par[18]; + par[13] = par[19]; + par[14] = ( par[20] + par[21]) * 0.5f; + par[15] = ( par[22] + par[23]) * 0.5f; + par[16] = ( par[24] + par[25]) * 0.5f; + par[17] = ( par[26] + par[27]) * 0.5f; + par[18] = ( par[28] + par[29] + par[30] + par[31]) * 0.25f; + par[19] = ( par[32] + par[33]) * 0.5f; +} + +static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full) +{ + if (full) { + par_mapped[33] = par[9]; + par_mapped[32] = par[9]; + par_mapped[31] = par[9]; + par_mapped[30] = par[9]; + par_mapped[29] = par[9]; + par_mapped[28] = par[9]; + par_mapped[27] = par[8]; + par_mapped[26] = par[8]; + par_mapped[25] = par[8]; + par_mapped[24] = par[8]; + par_mapped[23] = par[7]; + par_mapped[22] = par[7]; + par_mapped[21] = par[7]; + par_mapped[20] = par[7]; + par_mapped[19] = par[6]; + par_mapped[18] = par[6]; + par_mapped[17] = par[5]; + par_mapped[16] = par[5]; + } else { + par_mapped[16] = 0; + } + par_mapped[15] = par[4]; + par_mapped[14] = par[4]; + par_mapped[13] = par[4]; + par_mapped[12] = par[4]; + par_mapped[11] = par[3]; + par_mapped[10] = par[3]; + par_mapped[ 9] = par[2]; + par_mapped[ 8] = par[2]; + par_mapped[ 7] = par[2]; + par_mapped[ 6] = par[2]; + par_mapped[ 5] = par[1]; + par_mapped[ 4] = par[1]; + par_mapped[ 3] = par[1]; + par_mapped[ 2] = par[0]; + par_mapped[ 1] = par[0]; + par_mapped[ 0] = par[0]; +} + +static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full) +{ + if (full) { + par_mapped[33] = par[19]; + par_mapped[32] = par[19]; + par_mapped[31] = par[18]; + par_mapped[30] = par[18]; + par_mapped[29] = par[18]; + par_mapped[28] = par[18]; + par_mapped[27] = par[17]; + par_mapped[26] = par[17]; + par_mapped[25] = par[16]; + par_mapped[24] = par[16]; + par_mapped[23] = par[15]; + par_mapped[22] = par[15]; + par_mapped[21] = par[14]; + par_mapped[20] = par[14]; + par_mapped[19] = par[13]; + par_mapped[18] = par[12]; + par_mapped[17] = par[11]; + } + par_mapped[16] = par[10]; + par_mapped[15] = par[ 9]; + par_mapped[14] = par[ 9]; + par_mapped[13] = par[ 8]; + par_mapped[12] = par[ 8]; + par_mapped[11] = par[ 7]; + par_mapped[10] = par[ 6]; + par_mapped[ 9] = par[ 5]; + par_mapped[ 8] = par[ 5]; + par_mapped[ 7] = par[ 4]; + par_mapped[ 6] = par[ 4]; + par_mapped[ 5] = par[ 3]; + par_mapped[ 4] = (par[ 2] + par[ 3]) / 2; + par_mapped[ 3] = par[ 2]; + par_mapped[ 2] = par[ 1]; + par_mapped[ 1] = (par[ 0] + par[ 1]) / 2; + par_mapped[ 0] = par[ 0]; +} + +static void map_val_20_to_34(float par[PS_MAX_NR_IIDICC]) +{ + par[33] = par[19]; + par[32] = par[19]; + par[31] = par[18]; + par[30] = par[18]; + par[29] = par[18]; + par[28] = par[18]; + par[27] = par[17]; + par[26] = par[17]; + par[25] = par[16]; + par[24] = par[16]; + par[23] = par[15]; + par[22] = par[15]; + par[21] = par[14]; + par[20] = par[14]; + par[19] = par[13]; + par[18] = par[12]; + par[17] = par[11]; + par[16] = par[10]; + par[15] = par[ 9]; + par[14] = par[ 9]; + par[13] = par[ 8]; + par[12] = par[ 8]; + par[11] = par[ 7]; + par[10] = par[ 6]; + par[ 9] = par[ 5]; + par[ 8] = par[ 5]; + par[ 7] = par[ 4]; + par[ 6] = par[ 4]; + par[ 5] = par[ 3]; + par[ 4] = (par[ 2] + par[ 3]) * 0.5f; + par[ 3] = par[ 2]; + par[ 2] = par[ 1]; + par[ 1] = (par[ 0] + par[ 1]) * 0.5f; + par[ 0] = par[ 0]; +} + +static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[32][2], int is34) +{ + float power[34][PS_QMF_TIME_SLOTS] = {{0}}; + float transient_gain[34][PS_QMF_TIME_SLOTS]; + float *peak_decay_nrg = ps->peak_decay_nrg; + float *power_smooth = ps->power_smooth; + float *peak_decay_diff_smooth = ps->peak_decay_diff_smooth; + float (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay; + float (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay; + const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20; + const float peak_decay_factor = 0.76592833836465f; + const float transient_impact = 1.5f; + const float a_smooth = 0.25f; //< Smoothing coefficient + int i, k, m, n; + int n0 = 0, nL = 32; + static const int link_delay[] = { 3, 4, 5 }; + static const float a[] = { 0.65143905753106f, + 0.56471812200776f, + 0.48954165955695f }; + + if (is34 != ps->is34bands_old) { + memset(ps->peak_decay_nrg, 0, sizeof(ps->peak_decay_nrg)); + memset(ps->power_smooth, 0, sizeof(ps->power_smooth)); + memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth)); + memset(ps->delay, 0, sizeof(ps->delay)); + memset(ps->ap_delay, 0, sizeof(ps->ap_delay)); + } + + for (n = n0; n < nL; n++) { + for (k = 0; k < NR_BANDS[is34]; k++) { + int i = k_to_i[k]; + power[i][n] += s[k][n][0] * s[k][n][0] + s[k][n][1] * s[k][n][1]; + } + } + + //Transient detection + for (i = 0; i < NR_PAR_BANDS[is34]; i++) { + for (n = n0; n < nL; n++) { + float decayed_peak = peak_decay_factor * peak_decay_nrg[i]; + float denom; + peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]); + power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]); + peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]); + denom = transient_impact * peak_decay_diff_smooth[i]; + transient_gain[i][n] = (denom > power_smooth[i]) ? + power_smooth[i] / denom : 1.0f; + } + } + + //Decorrelation and transient reduction + // PS_AP_LINKS - 1 + // ----- + // | | Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k] + //H[k][z] = z^-2 * phi_fract[k] * | | ---------------------------------------------------------------- + // | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m] + // m = 0 + //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z] + for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) { + int b = k_to_i[k]; + float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]); + float ag[PS_AP_LINKS]; + g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f); + memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0])); + memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0])); + for (m = 0; m < PS_AP_LINKS; m++) { + memcpy(ap_delay[k][m], ap_delay[k][m]+numQMFSlots, 5*sizeof(ap_delay[k][m][0])); + ag[m] = a[m] * g_decay_slope; + } + for (n = n0; n < nL; n++) { + float in_re = delay[k][n+PS_MAX_DELAY-2][0] * phi_fract[is34][k][0] - + delay[k][n+PS_MAX_DELAY-2][1] * phi_fract[is34][k][1]; + float in_im = delay[k][n+PS_MAX_DELAY-2][0] * phi_fract[is34][k][1] + + delay[k][n+PS_MAX_DELAY-2][1] * phi_fract[is34][k][0]; + for (m = 0; m < PS_AP_LINKS; m++) { + float a_re = ag[m] * in_re; + float a_im = ag[m] * in_im; + float link_delay_re = ap_delay[k][m][n+5-link_delay[m]][0]; + float link_delay_im = ap_delay[k][m][n+5-link_delay[m]][1]; + float fractional_delay_re = Q_fract_allpass[is34][k][m][0]; + float fractional_delay_im = Q_fract_allpass[is34][k][m][1]; + ap_delay[k][m][n+5][0] = in_re; + ap_delay[k][m][n+5][1] = in_im; + in_re = link_delay_re * fractional_delay_re - link_delay_im * fractional_delay_im - a_re; + in_im = link_delay_re * fractional_delay_im + link_delay_im * fractional_delay_re - a_im; + ap_delay[k][m][n+5][0] += ag[m] * in_re; + ap_delay[k][m][n+5][1] += ag[m] * in_im; + } + out[k][n][0] = transient_gain[b][n] * in_re; + out[k][n][1] = transient_gain[b][n] * in_im; + } + } + for (; k < SHORT_DELAY_BAND[is34]; k++) { + memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0])); + memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0])); + for (n = n0; n < nL; n++) { + //H = delay 14 + out[k][n][0] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-14][0]; + out[k][n][1] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-14][1]; + } + } + for (; k < NR_BANDS[is34]; k++) { + memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0])); + memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0])); + for (n = n0; n < nL; n++) { + //H = delay 1 + out[k][n][0] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-1][0]; + out[k][n][1] = transient_gain[k_to_i[k]][n] * delay[k][n+PS_MAX_DELAY-1][1]; + } + } +} + +static void remap34(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC], + int8_t (*par)[PS_MAX_NR_IIDICC], + int num_par, int num_env, int full) +{ + int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped; + int e; + if (num_par == 20 || num_par == 11) { + for (e = 0; e < num_env; e++) { + map_idx_20_to_34(par_mapped[e], par[e], full); + } + } else if (num_par == 10 || num_par == 5) { + for (e = 0; e < num_env; e++) { + map_idx_10_to_34(par_mapped[e], par[e], full); + } + } else { + *p_par_mapped = par; + } +} + +static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC], + int8_t (*par)[PS_MAX_NR_IIDICC], + int num_par, int num_env, int full) +{ + int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped; + int e; + if (num_par == 34 || num_par == 17) { + for (e = 0; e < num_env; e++) { + map_idx_34_to_20(par_mapped[e], par[e], full); + } + } else if (num_par == 10 || num_par == 5) { + for (e = 0; e < num_env; e++) { + map_idx_10_to_20(par_mapped[e], par[e], full); + } + } else { + *p_par_mapped = par; + } +} + +static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2], int is34) +{ + int e, b, k, n; + + float (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11; + float (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12; + float (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21; + float (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22; + int8_t *opd_hist = ps->opd_hist; + int8_t *ipd_hist = ps->ipd_hist; + int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC]; + int8_t icc_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC]; + int8_t ipd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC]; + int8_t opd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC]; + int8_t (*iid_mapped)[PS_MAX_NR_IIDICC] = iid_mapped_buf; + int8_t (*icc_mapped)[PS_MAX_NR_IIDICC] = icc_mapped_buf; + int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf; + int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf; + const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20; + const float (*H_LUT)[8][4] = (PS_BASELINE || ps->icc_mode < 3) ? HA : HB; + + //Remapping + for (b = 0; b < PS_MAX_NR_IIDICC; b++) { + H11[0][0][b] = H11[0][ps->num_env_old][b]; + H12[0][0][b] = H12[0][ps->num_env_old][b]; + H21[0][0][b] = H21[0][ps->num_env_old][b]; + H22[0][0][b] = H22[0][ps->num_env_old][b]; + H11[1][0][b] = H11[1][ps->num_env_old][b]; + H12[1][0][b] = H12[1][ps->num_env_old][b]; + H21[1][0][b] = H21[1][ps->num_env_old][b]; + H22[1][0][b] = H22[1][ps->num_env_old][b]; + } + if (is34) { + remap34(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1); + remap34(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1); + if (ps->enable_ipdopd) { + remap34(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0); + remap34(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0); + } + if (!ps->is34bands_old) { + map_val_20_to_34(H11[0][0]); + map_val_20_to_34(H11[1][0]); + map_val_20_to_34(H12[0][0]); + map_val_20_to_34(H12[1][0]); + map_val_20_to_34(H21[0][0]); + map_val_20_to_34(H21[1][0]); + map_val_20_to_34(H22[0][0]); + map_val_20_to_34(H22[1][0]); + ipdopd_reset(ipd_hist, opd_hist); + } + } else { + remap20(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1); + remap20(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1); + if (ps->enable_ipdopd) { + remap20(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0); + remap20(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0); + } + if (ps->is34bands_old) { + map_val_34_to_20(H11[0][0]); + map_val_34_to_20(H11[1][0]); + map_val_34_to_20(H12[0][0]); + map_val_34_to_20(H12[1][0]); + map_val_34_to_20(H21[0][0]); + map_val_34_to_20(H21[1][0]); + map_val_34_to_20(H22[0][0]); + map_val_34_to_20(H22[1][0]); + ipdopd_reset(ipd_hist, opd_hist); + } + } + + //Mixing + for (e = 0; e < ps->num_env; e++) { + for (b = 0; b < NR_PAR_BANDS[is34]; b++) { + float h11, h12, h21, h22; + h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][0]; + h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][1]; + h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][2]; + h22 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][3]; + if (!PS_BASELINE && ps->enable_ipdopd && b < ps->nr_ipdopd_par) { + //The spec say says to only run this smoother when enable_ipdopd + //is set but the reference decoder appears to run it constantly + float h11i, h12i, h21i, h22i; + float ipd_adj_re, ipd_adj_im; + int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b]; + int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b]; + float opd_re = pd_re_smooth[opd_idx]; + float opd_im = pd_im_smooth[opd_idx]; + float ipd_re = pd_re_smooth[ipd_idx]; + float ipd_im = pd_im_smooth[ipd_idx]; + opd_hist[b] = opd_idx & 0x3F; + ipd_hist[b] = ipd_idx & 0x3F; + + ipd_adj_re = opd_re*ipd_re + opd_im*ipd_im; + ipd_adj_im = opd_im*ipd_re - opd_re*ipd_im; + h11i = h11 * opd_im; + h11 = h11 * opd_re; + h12i = h12 * ipd_adj_im; + h12 = h12 * ipd_adj_re; + h21i = h21 * opd_im; + h21 = h21 * opd_re; + h22i = h22 * ipd_adj_im; + h22 = h22 * ipd_adj_re; + H11[1][e+1][b] = h11i; + H12[1][e+1][b] = h12i; + H21[1][e+1][b] = h21i; + H22[1][e+1][b] = h22i; + } + H11[0][e+1][b] = h11; + H12[0][e+1][b] = h12; + H21[0][e+1][b] = h21; + H22[0][e+1][b] = h22; + } + for (k = 0; k < NR_BANDS[is34]; k++) { + float h11r, h12r, h21r, h22r; + float h11i, h12i, h21i, h22i; + float h11r_step, h12r_step, h21r_step, h22r_step; + float h11i_step, h12i_step, h21i_step, h22i_step; + int start = ps->border_position[e]; + int stop = ps->border_position[e+1]; + float width = 1.f / (stop - start); + b = k_to_i[k]; + h11r = H11[0][e][b]; + h12r = H12[0][e][b]; + h21r = H21[0][e][b]; + h22r = H22[0][e][b]; + if (!PS_BASELINE && ps->enable_ipdopd) { + //Is this necessary? ps_04_new seems unchanged + if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) { + h11i = -H11[1][e][b]; + h12i = -H12[1][e][b]; + h21i = -H21[1][e][b]; + h22i = -H22[1][e][b]; + } else { + h11i = H11[1][e][b]; + h12i = H12[1][e][b]; + h21i = H21[1][e][b]; + h22i = H22[1][e][b]; + } + } + //Interpolation + h11r_step = (H11[0][e+1][b] - h11r) * width; + h12r_step = (H12[0][e+1][b] - h12r) * width; + h21r_step = (H21[0][e+1][b] - h21r) * width; + h22r_step = (H22[0][e+1][b] - h22r) * width; + if (!PS_BASELINE && ps->enable_ipdopd) { + h11i_step = (H11[1][e+1][b] - h11i) * width; + h12i_step = (H12[1][e+1][b] - h12i) * width; + h21i_step = (H21[1][e+1][b] - h21i) * width; + h22i_step = (H22[1][e+1][b] - h22i) * width; + } + for (n = start + 1; n <= stop; n++) { + //l is s, r is d + float l_re = l[k][n][0]; + float l_im = l[k][n][1]; + float r_re = r[k][n][0]; + float r_im = r[k][n][1]; + h11r += h11r_step; + h12r += h12r_step; + h21r += h21r_step; + h22r += h22r_step; + if (!PS_BASELINE && ps->enable_ipdopd) { + h11i += h11i_step; + h12i += h12i_step; + h21i += h21i_step; + h22i += h22i_step; + + l[k][n][0] = h11r*l_re + h21r*r_re - h11i*l_im - h21i*r_im; + l[k][n][1] = h11r*l_im + h21r*r_im + h11i*l_re + h21i*r_re; + r[k][n][0] = h12r*l_re + h22r*r_re - h12i*l_im - h22i*r_im; + r[k][n][1] = h12r*l_im + h22r*r_im + h12i*l_re + h22i*r_re; + } else { + l[k][n][0] = h11r*l_re + h21r*r_re; + l[k][n][1] = h11r*l_im + h21r*r_im; + r[k][n][0] = h12r*l_re + h22r*r_re; + r[k][n][1] = h12r*l_im + h22r*r_im; + } + } + } + } +} + +int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float R[2][38][64], int top) +{ + float Lbuf[91][32][2]; + float Rbuf[91][32][2]; + const int len = 32; + int is34 = ps->is34bands; + + top += NR_BANDS[is34] - 64; + memset(ps->delay+top, 0, (NR_BANDS[is34] - top)*sizeof(ps->delay[0])); + if (top < NR_ALLPASS_BANDS[is34]) + memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0])); + + hybrid_analysis(Lbuf, ps->in_buf, L, is34, len); + decorrelation(ps, Rbuf, Lbuf, is34); + stereo_processing(ps, Lbuf, Rbuf, is34); + hybrid_synthesis(L, Lbuf, is34, len); + hybrid_synthesis(R, Rbuf, is34, len); + + return 0; +} + +#define PS_INIT_VLC_STATIC(num, size) \ + INIT_VLC_STATIC(&vlc_ps[num], 9, ps_tmp[num].table_size / ps_tmp[num].elem_size, \ + ps_tmp[num].ps_bits, 1, 1, \ + ps_tmp[num].ps_codes, ps_tmp[num].elem_size, ps_tmp[num].elem_size, \ + size); + +#define PS_VLC_ROW(name) \ + { name ## _codes, name ## _bits, sizeof(name ## _codes), sizeof(name ## _codes[0]) } + +av_cold void ff_ps_init(void) { + // Syntax initialization + static const struct { + const void *ps_codes, *ps_bits; + const unsigned int table_size, elem_size; + } ps_tmp[] = { + PS_VLC_ROW(huff_iid_df1), + PS_VLC_ROW(huff_iid_dt1), + PS_VLC_ROW(huff_iid_df0), + PS_VLC_ROW(huff_iid_dt0), + PS_VLC_ROW(huff_icc_df), + PS_VLC_ROW(huff_icc_dt), + PS_VLC_ROW(huff_ipd_df), + PS_VLC_ROW(huff_ipd_dt), + PS_VLC_ROW(huff_opd_df), + PS_VLC_ROW(huff_opd_dt), + }; + + PS_INIT_VLC_STATIC(0, 1544); + PS_INIT_VLC_STATIC(1, 832); + PS_INIT_VLC_STATIC(2, 1024); + PS_INIT_VLC_STATIC(3, 1036); + PS_INIT_VLC_STATIC(4, 544); + PS_INIT_VLC_STATIC(5, 544); + PS_INIT_VLC_STATIC(6, 512); + PS_INIT_VLC_STATIC(7, 512); + PS_INIT_VLC_STATIC(8, 512); + PS_INIT_VLC_STATIC(9, 512); + + ps_tableinit(); +} + +av_cold void ff_ps_ctx_init(PSContext *ps) +{ + ipdopd_reset(ps->ipd_hist, ps->opd_hist); +} |