* N (code in SoC repo) Long Term Prediction
* Y intensity stereo
* Y channel coupling
- * N frequency domain prediction
+ * Y frequency domain prediction
* Y Perceptual Noise Substitution
* Y Mid/Side stereo
* N Scalable Inverse AAC Quantization
#include "avcodec.h"
+#include "internal.h"
#include "bitstream.h"
#include "dsputil.h"
#include "lpc.h"
int extension_flag, ret;
if(get_bits1(gb)) { // frameLengthFlag
- av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
+ ff_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
return -1;
}
skip_bits_long(&gb, i);
switch (ac->m4ac.object_type) {
+ case AOT_AAC_MAIN:
case AOT_AAC_LC:
if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
return -1;
return previous_val * 1664525 + 1013904223;
}
+static void reset_predict_state(PredictorState * ps) {
+ ps->r0 = 0.0f;
+ ps->r1 = 0.0f;
+ ps->cor0 = 0.0f;
+ ps->cor1 = 0.0f;
+ ps->var0 = 1.0f;
+ ps->var1 = 1.0f;
+}
+
+static void reset_all_predictors(PredictorState * ps) {
+ int i;
+ for (i = 0; i < MAX_PREDICTORS; i++)
+ reset_predict_state(&ps[i]);
+}
+
+static void reset_predictor_group(PredictorState * ps, int group_num) {
+ int i;
+ for (i = group_num-1; i < MAX_PREDICTORS; i+=30)
+ reset_predict_state(&ps[i]);
+}
+
static av_cold int aac_decode_init(AVCodecContext * avccontext) {
AACContext * ac = avccontext->priv_data;
int i;
}
#ifndef CONFIG_HARDCODED_TABLES
- for (i = 0; i < 316; i++)
+ for (i = 0; i < 428; i++)
ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
#endif /* CONFIG_HARDCODED_TABLES */
- INIT_VLC_STATIC(&vlc_scalefactors, 7, sizeof(ff_aac_scalefactor_code)/sizeof(ff_aac_scalefactor_code[0]),
+ INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
352);
skip_bits_long(gb, 8 * count);
}
+static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) {
+ int sfb;
+ if (get_bits1(gb)) {
+ ics->predictor_reset_group = get_bits(gb, 5);
+ if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
+ av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
+ return -1;
+ }
+ }
+ for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
+ ics->prediction_used[sfb] = get_bits1(gb);
+ }
+ return 0;
+}
+
/**
* Decode Individual Channel Stream info; reference: table 4.6.
*
ics->swb_offset = swb_offset_128[ac->m4ac.sampling_index];
ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
ics->tns_max_bands = tns_max_bands_128[ac->m4ac.sampling_index];
+ ics->predictor_present = 0;
} else {
ics->max_sfb = get_bits(gb, 6);
ics->num_windows = 1;
ics->swb_offset = swb_offset_1024[ac->m4ac.sampling_index];
ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
ics->tns_max_bands = tns_max_bands_1024[ac->m4ac.sampling_index];
- if (get_bits1(gb)) {
- av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
- memset(ics, 0, sizeof(IndividualChannelStream));
- return -1;
+ ics->predictor_present = get_bits1(gb);
+ ics->predictor_reset_group = 0;
+ if (ics->predictor_present) {
+ if (ac->m4ac.object_type == AOT_AAC_MAIN) {
+ if (decode_prediction(ac, ics, gb)) {
+ memset(ics, 0, sizeof(IndividualChannelStream));
+ return -1;
+ }
+ } else if (ac->m4ac.object_type == AOT_AAC_LC) {
+ av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
+ memset(ics, 0, sizeof(IndividualChannelStream));
+ return -1;
+ } else {
+ ff_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
+ memset(ics, 0, sizeof(IndividualChannelStream));
+ return -1;
+ }
}
}
"%s (%d) out of range.\n", sf_str[1], offset[1]);
return -1;
}
- sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset];
+ sf[idx] = -ff_aac_pow2sf_tab[ offset[1] + sf_offset + 100];
}
}else {
for(; i < run_end; i++, idx++) {
const int c = 1024/ics->num_windows;
const uint16_t * offsets = ics->swb_offset;
float *coef_base = coef;
+ static const float sign_lookup[] = { 1.0f, -1.0f };
for (g = 0; g < ics->num_windows; g++)
memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb]));
memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float));
}
}else if (cur_band_type == NOISE_BT) {
- const float scale = sf[idx] / ((offsets[i+1] - offsets[i]) * PNS_MEAN_ENERGY);
for (group = 0; group < ics->group_len[g]; group++) {
+ float scale;
+ float band_energy = 0;
for (k = offsets[i]; k < offsets[i+1]; k++) {
ac->random_state = lcg_random(ac->random_state);
- coef[group*128+k] = ac->random_state * scale;
+ coef[group*128+k] = ac->random_state;
+ band_energy += coef[group*128+k]*coef[group*128+k];
+ }
+ scale = sf[idx] / sqrtf(band_energy);
+ for (k = offsets[i]; k < offsets[i+1]; k++) {
+ coef[group*128+k] *= scale;
}
}
}else if (cur_band_type != INTENSITY_BT2 && cur_band_type != INTENSITY_BT) {
}
vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
if (is_cb_unsigned) {
- for (j = 0; j < dim; j++)
- if (vq_ptr[j])
- coef[coef_tmp_idx + j] = 1 - 2*(int)get_bits1(gb);
+ if (vq_ptr[0]) coef[coef_tmp_idx ] = sign_lookup[get_bits1(gb)];
+ if (vq_ptr[1]) coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
+ if (dim == 4) {
+ if (vq_ptr[2]) coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
+ if (vq_ptr[3]) coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
+ }
}else {
- for (j = 0; j < dim; j++)
- coef[coef_tmp_idx + j] = 1.0f;
+ coef[coef_tmp_idx ] = 1.0f;
+ coef[coef_tmp_idx + 1] = 1.0f;
+ if (dim == 4) {
+ coef[coef_tmp_idx + 2] = 1.0f;
+ coef[coef_tmp_idx + 3] = 1.0f;
+ }
}
if (cur_band_type == ESC_BT) {
for (j = 0; j < 2; j++) {
return -1;
}
n = (1<<n) + get_bits(gb, n);
- coef[coef_tmp_idx + j] *= cbrtf(fabsf(n)) * n;
+ coef[coef_tmp_idx + j] *= cbrtf(n) * n;
}else
coef[coef_tmp_idx + j] *= vq_ptr[j];
}
}else
- for (j = 0; j < dim; j++)
- coef[coef_tmp_idx + j] *= vq_ptr[j];
- for (j = 0; j < dim; j++)
- coef[coef_tmp_idx + j] *= sf[idx];
+ {
+ coef[coef_tmp_idx ] *= vq_ptr[0];
+ coef[coef_tmp_idx + 1] *= vq_ptr[1];
+ if (dim == 4) {
+ coef[coef_tmp_idx + 2] *= vq_ptr[2];
+ coef[coef_tmp_idx + 3] *= vq_ptr[3];
+ }
+ }
+ coef[coef_tmp_idx ] *= sf[idx];
+ coef[coef_tmp_idx + 1] *= sf[idx];
+ if (dim == 4) {
+ coef[coef_tmp_idx + 2] *= sf[idx];
+ coef[coef_tmp_idx + 3] *= sf[idx];
+ }
}
}
}
return 0;
}
+static av_always_inline float flt16_round(float pf) {
+ int exp;
+ pf = frexpf(pf, &exp);
+ pf = ldexpf(roundf(ldexpf(pf, 8)), exp-8);
+ return pf;
+}
+
+static av_always_inline float flt16_even(float pf) {
+ int exp;
+ pf = frexpf(pf, &exp);
+ pf = ldexpf(rintf(ldexpf(pf, 8)), exp-8);
+ return pf;
+}
+
+static av_always_inline float flt16_trunc(float pf) {
+ int exp;
+ pf = frexpf(pf, &exp);
+ pf = ldexpf(truncf(ldexpf(pf, 8)), exp-8);
+ return pf;
+}
+
+static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) {
+ const float a = 0.953125; // 61.0/64
+ const float alpha = 0.90625; // 29.0/32
+ float e0, e1;
+ float pv;
+ float k1, k2;
+
+ k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
+ k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
+
+ pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
+ if (output_enable)
+ *coef += pv * ac->sf_scale;
+
+ e0 = *coef / ac->sf_scale;
+ e1 = e0 - k1 * ps->r0;
+
+ ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
+ ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
+ ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
+ ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
+
+ ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
+ ps->r0 = flt16_trunc(a * e0);
+}
+
+/**
+ * Apply AAC-Main style frequency domain prediction.
+ */
+static void apply_prediction(AACContext * ac, SingleChannelElement * sce) {
+ int sfb, k;
+
+ if (!sce->ics.predictor_initialized) {
+ reset_all_predictors(sce->ics.predictor_state);
+ sce->ics.predictor_initialized = 1;
+ }
+
+ if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
+ for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
+ for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
+ predict(ac, &sce->ics.predictor_state[k], &sce->coeffs[k],
+ sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
+ }
+ }
+ if (sce->ics.predictor_reset_group)
+ reset_predictor_group(sce->ics.predictor_state, sce->ics.predictor_reset_group);
+ } else
+ reset_all_predictors(sce->ics.predictor_state);
+}
+
/**
* Decode an individual_channel_stream payload; reference: table 4.44.
*
if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
return -1;
if (get_bits1(gb)) {
- av_log_missing_feature(ac->avccontext, "SSR", 1);
+ ff_log_missing_feature(ac->avccontext, "SSR", 1);
return -1;
}
}
if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
return -1;
+
+ if(ac->m4ac.object_type == AOT_AAC_MAIN)
+ apply_prediction(ac, sce);
+
return 0;
}
*/
static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) {
int num_gain = 0;
- int c, g, sfb, ret, idx = 0;
+ int c, g, sfb, ret;
int sign;
float scale;
SingleChannelElement * sce = &che->ch[0];
if (coup->ch_select[c] == 3)
num_gain++;
} else
- coup->ch_select[c] = 1;
+ coup->ch_select[c] = 2;
}
coup->coupling_point += get_bits1(gb);
return ret;
for (c = 0; c < num_gain; c++) {
+ int idx = 0;
int cge = 1;
int gain = 0;
float gain_cache = 1.;
if (c) {
cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
- gain_cache = pow(scale, gain);
+ gain_cache = pow(scale, -gain);
}
for (g = 0; g < sce->ics.num_window_groups; g++) {
for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
if (t) {
int s = 1;
+ t = gain += t;
if (sign) {
s -= 2 * (t & 0x1);
t >>= 1;
}
- gain += t;
- gain_cache = pow(scale, gain) * s;
+ gain_cache = pow(scale, -t) * s;
}
}
coup->gain[c][idx] = gain_cache;
*/
static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
// TODO : sbr_extension implementation
- av_log_missing_feature(ac->avccontext, "SBR", 0);
+ ff_log_missing_feature(ac->avccontext, "SBR", 0);
skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
return cnt;
}
const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
float * buf = ac->buf_mdct;
- DECLARE_ALIGNED(16, float, temp[128]);
+ float * temp = ac->temp;
int i;
// imdct
*
* @param index index into coupling gain array
*/
-static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index) {
- IndividualChannelStream * ics = &cc->ch[0].ics;
+static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
+ IndividualChannelStream * ics = &cce->ch[0].ics;
const uint16_t * offsets = ics->swb_offset;
- float * dest = sce->coeffs;
- const float * src = cc->ch[0].coeffs;
+ float * dest = target->coeffs;
+ const float * src = cce->ch[0].coeffs;
int g, i, group, k, idx = 0;
if(ac->m4ac.object_type == AOT_AAC_LTP) {
av_log(ac->avccontext, AV_LOG_ERROR,
}
for (g = 0; g < ics->num_window_groups; g++) {
for (i = 0; i < ics->max_sfb; i++, idx++) {
- if (cc->ch[0].band_type[idx] != ZERO_BT) {
+ if (cce->ch[0].band_type[idx] != ZERO_BT) {
for (group = 0; group < ics->group_len[g]; group++) {
for (k = offsets[i]; k < offsets[i+1]; k++) {
// XXX dsputil-ize
- dest[group*128+k] += cc->coup.gain[index][idx] * src[group*128+k];
+ dest[group*128+k] += cce->coup.gain[index][idx] * src[group*128+k];
}
}
}
*
* @param index index into coupling gain array
*/
-static void apply_independent_coupling(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index) {
+static void apply_independent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
int i;
for (i = 0; i < 1024; i++)
- sce->ret[i] += cc->coup.gain[index][0] * (cc->ch[0].ret[i] - ac->add_bias);
+ target->ret[i] += cce->coup.gain[index][0] * (cce->ch[0].ret[i] - ac->add_bias);
}
/**
* @param apply_coupling_method pointer to (in)dependent coupling function
*/
static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
- void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index))
+ enum RawDataBlockType type, int elem_id, enum CouplingPoint coupling_point,
+ void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index))
{
- int c;
- int index = 0;
- ChannelCoupling * coup = &cc->coup;
- for (c = 0; c <= coup->num_coupled; c++) {
- if (ac->che[coup->type[c]][coup->id_select[c]]) {
- if (coup->ch_select[c] != 2) {
- apply_coupling_method(ac, &ac->che[coup->type[c]][coup->id_select[c]]->ch[0], cc, index);
- if (coup->ch_select[c] != 0)
- index++;
+ int i, c;
+
+ for (i = 0; i < MAX_ELEM_ID; i++) {
+ ChannelElement *cce = ac->che[TYPE_CCE][i];
+ int index = 0;
+
+ if (cce && cce->coup.coupling_point == coupling_point) {
+ ChannelCoupling * coup = &cce->coup;
+
+ for (c = 0; c <= coup->num_coupled; c++) {
+ if (coup->type[c] == type && coup->id_select[c] == elem_id) {
+ if (coup->ch_select[c] != 1) {
+ apply_coupling_method(ac, &cc->ch[0], cce, index);
+ if (coup->ch_select[c] != 0)
+ index++;
+ }
+ if (coup->ch_select[c] != 2)
+ apply_coupling_method(ac, &cc->ch[1], cce, index++);
+ } else
+ index += 1 + (coup->ch_select[c] == 3);
}
- if (coup->ch_select[c] != 1)
- apply_coupling_method(ac, &ac->che[coup->type[c]][coup->id_select[c]]->ch[1], cc, index++);
- } else {
- av_log(ac->avccontext, AV_LOG_ERROR,
- "coupling target %sE[%d] not available\n",
- coup->type[c] == TYPE_CPE ? "CP" : "SC", coup->id_select[c]);
- break;
}
}
}
*/
static void spectral_to_sample(AACContext * ac) {
int i, type;
- for (i = 0; i < MAX_ELEM_ID; i++) {
- for(type = 0; type < 4; type++) {
+ for(type = 3; type >= 0; type--) {
+ for (i = 0; i < MAX_ELEM_ID; i++) {
ChannelElement *che = ac->che[type][i];
if(che) {
- if(che->coup.coupling_point == BEFORE_TNS)
- apply_channel_coupling(ac, che, apply_dependent_coupling);
+ if(type <= TYPE_CPE)
+ apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
if(che->ch[0].tns.present)
apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
if(che->ch[1].tns.present)
apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
- if(che->coup.coupling_point == BETWEEN_TNS_AND_IMDCT)
- apply_channel_coupling(ac, che, apply_dependent_coupling);
- imdct_and_windowing(ac, &che->ch[0]);
+ if(type <= TYPE_CPE)
+ apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
+ if(type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
+ imdct_and_windowing(ac, &che->ch[0]);
if(type == TYPE_CPE)
imdct_and_windowing(ac, &che->ch[1]);
- if(che->coup.coupling_point == AFTER_IMDCT)
- apply_channel_coupling(ac, che, apply_independent_coupling);
+ if(type <= TYPE_CCE)
+ apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
}
}
}