aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/aacenc_pred.c
blob: c0e5e6e3b67327c101c3c483d95e605a329d92e2 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
/*
 * AAC encoder main-type prediction
 * Copyright (C) 2015 Rostislav Pehlivanov
 *
 * 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
 */

/**
 * @file
 * AAC encoder Intensity Stereo
 * @author Rostislav Pehlivanov ( atomnuker gmail com )
 */

#include "aactab.h"
#include "aacenc_pred.h"
#include "aacenc_utils.h"
#include "aacenc_is.h"            /* <- Needed for common window distortions */
#include "aacenc_quantization.h"

#define RESTORE_PRED(sce, sfb) \
        if (sce->ics.prediction_used[sfb]) {\
            sce->ics.prediction_used[sfb] = 0;\
            sce->band_type[sfb] = sce->band_alt[sfb];\
        }

static inline float flt16_round(float pf)
{
    union av_intfloat32 tmp;
    tmp.f = pf;
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
    return tmp.f;
}

static inline float flt16_even(float pf)
{
    union av_intfloat32 tmp;
    tmp.f = pf;
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
    return tmp.f;
}

static inline float flt16_trunc(float pf)
{
    union av_intfloat32 pun;
    pun.f = pf;
    pun.i &= 0xFFFF0000U;
    return pun.f;
}

static inline void predict(PredictorState *ps, float *coef, float *rcoef, int set)
{
    float k2;
    const float a     = 0.953125; // 61.0 / 64
    const float alpha = 0.90625;  // 29.0 / 32
    const float   k1 = ps->k1;
    const float   r0 = ps->r0,     r1 = ps->r1;
    const float cor0 = ps->cor0, cor1 = ps->cor1;
    const float var0 = ps->var0, var1 = ps->var1;
    const float e0 = *coef - ps->x_est;
    const float e1 = e0 - k1 * r0;

    if (set)
        *coef = e0;

    ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
    ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
    ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
    ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
    ps->r1   = flt16_trunc(a * (r0 - k1 * e0));
    ps->r0   = flt16_trunc(a * e0);

    /* Prediction for next frame */
    ps->k1   = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
    k2       = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
    *rcoef   = ps->x_est = flt16_round(ps->k1*ps->r0 + k2*ps->r1);
}

static inline void reset_predict_state(PredictorState *ps)
{
    ps->r0    = 0.0f;
    ps->r1    = 0.0f;
    ps->k1    = 0.0f;
    ps->cor0  = 0.0f;
    ps->cor1  = 0.0f;
    ps->var0  = 1.0f;
    ps->var1  = 1.0f;
    ps->x_est = 0.0f;
}

static inline void reset_all_predictors(PredictorState *ps)
{
    int i;
    for (i = 0; i < MAX_PREDICTORS; i++)
        reset_predict_state(&ps[i]);
}

static inline void reset_predictor_group(SingleChannelElement *sce, int group_num)
{
    int i;
    PredictorState *ps = sce->predictor_state;
    for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
        reset_predict_state(&ps[i]);
}

void ff_aac_apply_main_pred(AACEncContext *s, SingleChannelElement *sce)
{
    int sfb, k;
    const int pmax = FFMIN(sce->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]);

    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
        for (sfb = 0; sfb < pmax; sfb++) {
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
                predict(&sce->predictor_state[k], &sce->coeffs[k], &sce->prcoeffs[k],
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
            }
        }
        if (sce->ics.predictor_reset_group) {
            reset_predictor_group(sce, sce->ics.predictor_reset_group);
        }
    } else {
        reset_all_predictors(sce->predictor_state);
    }
}

/* If inc = 0 you can check if this returns 0 to see if you can reset freely */
static inline int update_counters(IndividualChannelStream *ics, int inc)
{
    int i;
    for (i = 1; i < 31; i++) {
        ics->predictor_reset_count[i] += inc;
        if (ics->predictor_reset_count[i] > PRED_RESET_FRAME_MIN)
            return i; /* Reset this immediately */
    }
    return 0;
}

void ff_aac_adjust_common_prediction(AACEncContext *s, ChannelElement *cpe)
{
    int start, w, w2, g, i, count = 0;
    SingleChannelElement *sce0 = &cpe->ch[0];
    SingleChannelElement *sce1 = &cpe->ch[1];
    const int pmax0 = FFMIN(sce0->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]);
    const int pmax1 = FFMIN(sce1->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]);
    const int pmax  = FFMIN(pmax0, pmax1);

    if (!cpe->common_window ||
        sce0->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE ||
        sce1->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE)
        return;

    for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
        start = 0;
        for (g = 0; g < sce0->ics.num_swb; g++) {
            int sfb = w*16+g;
            int sum = sce0->ics.prediction_used[sfb] + sce1->ics.prediction_used[sfb];
            float ener0 = 0.0f, ener1 = 0.0f, ener01 = 0.0f;
            struct AACISError ph_err1, ph_err2, *erf;
            if (sfb < PRED_SFB_START || sfb > pmax || sum != 2) {
                RESTORE_PRED(sce0, sfb);
                RESTORE_PRED(sce1, sfb);
                start += sce0->ics.swb_sizes[g];
                continue;
            }
            for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
                for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
                    float coef0 = sce0->pcoeffs[start+(w+w2)*128+i];
                    float coef1 = sce1->pcoeffs[start+(w+w2)*128+i];
                    ener0  += coef0*coef0;
                    ener1  += coef1*coef1;
                    ener01 += (coef0 + coef1)*(coef0 + coef1);
                }
            }
            ph_err1 = ff_aac_is_encoding_err(s, cpe, start, w, g,
                                             ener0, ener1, ener01, 1, -1);
            ph_err2 = ff_aac_is_encoding_err(s, cpe, start, w, g,
                                             ener0, ener1, ener01, 1, +1);
            erf = ph_err1.error < ph_err2.error ? &ph_err1 : &ph_err2;
            if (erf->pass) {
                sce0->ics.prediction_used[sfb] = 1;
                sce1->ics.prediction_used[sfb] = 1;
                count++;
            } else {
                RESTORE_PRED(sce0, sfb);
                RESTORE_PRED(sce1, sfb);
            }
            start += sce0->ics.swb_sizes[g];
        }
    }

    sce1->ics.predictor_present = sce0->ics.predictor_present = !!count;
}

static void update_pred_resets(SingleChannelElement *sce)
{
    int i, max_group_id_c, max_frame = 0;
    float avg_frame = 0.0f;
    IndividualChannelStream *ics = &sce->ics;

    /* Update the counters and immediately update any frame behind schedule */
    if ((ics->predictor_reset_group = update_counters(&sce->ics, 1)))
        return;

    for (i = 1; i < 31; i++) {
        /* Count-based */
        if (ics->predictor_reset_count[i] > max_frame) {
            max_group_id_c = i;
            max_frame = ics->predictor_reset_count[i];
        }
        avg_frame = (ics->predictor_reset_count[i] + avg_frame)/2;
    }

    if (max_frame > PRED_RESET_MIN) {
        ics->predictor_reset_group = max_group_id_c;
    } else {
        ics->predictor_reset_group = 0;
    }
}

void ff_aac_search_for_pred(AACEncContext *s, SingleChannelElement *sce)
{
    int sfb, i, count = 0, cost_coeffs = 0, cost_pred = 0;
    const int pmax = FFMIN(sce->ics.max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]);
    float *O34  = &s->scoefs[128*0], *P34 = &s->scoefs[128*1];
    float *SENT = &s->scoefs[128*2], *S34 = &s->scoefs[128*3];
    float *QERR = &s->scoefs[128*4];

    if (sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
        sce->ics.predictor_present = 0;
        return;
    }

    if (!sce->ics.predictor_initialized) {
        reset_all_predictors(sce->predictor_state);
        sce->ics.predictor_initialized = 1;
        memcpy(sce->prcoeffs, sce->coeffs, 1024*sizeof(float));
        for (i = 1; i < 31; i++)
            sce->ics.predictor_reset_count[i] = i;
    }

    update_pred_resets(sce);
    memcpy(sce->band_alt, sce->band_type, sizeof(sce->band_type));

    for (sfb = PRED_SFB_START; sfb < pmax; sfb++) {
        int cost1, cost2, cb_p;
        float dist1, dist2, dist_spec_err = 0.0f;
        const int cb_n = sce->band_type[sfb];
        const int start_coef = sce->ics.swb_offset[sfb];
        const int num_coeffs = sce->ics.swb_offset[sfb + 1] - start_coef;
        const FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[sfb];

        if (start_coef + num_coeffs > MAX_PREDICTORS ||
            (s->cur_channel && sce->band_type[sfb] >= INTENSITY_BT2) ||
            sce->band_type[sfb] == NOISE_BT)
            continue;

        /* Normal coefficients */
        abs_pow34_v(O34, &sce->coeffs[start_coef], num_coeffs);
        dist1 = quantize_and_encode_band_cost(s, NULL, &sce->coeffs[start_coef], NULL,
                                              O34, num_coeffs, sce->sf_idx[sfb],
                                              cb_n, s->lambda / band->threshold, INFINITY, &cost1, 0);
        cost_coeffs += cost1;

        /* Encoded coefficients - needed for #bits, band type and quant. error */
        for (i = 0; i < num_coeffs; i++)
            SENT[i] = sce->coeffs[start_coef + i] - sce->prcoeffs[start_coef + i];
        abs_pow34_v(S34, SENT, num_coeffs);
        if (cb_n < RESERVED_BT)
            cb_p = find_min_book(find_max_val(1, num_coeffs, S34), sce->sf_idx[sfb]);
        else
            cb_p = cb_n;
        quantize_and_encode_band_cost(s, NULL, SENT, QERR, S34, num_coeffs,
                                      sce->sf_idx[sfb], cb_p, s->lambda / band->threshold, INFINITY,
                                      &cost2, 0);

        /* Reconstructed coefficients - needed for distortion measurements */
        for (i = 0; i < num_coeffs; i++)
            sce->prcoeffs[start_coef + i] += QERR[i] != 0.0f ? (sce->prcoeffs[start_coef + i] - QERR[i]) : 0.0f;
        abs_pow34_v(P34, &sce->prcoeffs[start_coef], num_coeffs);
        if (cb_n < RESERVED_BT)
            cb_p = find_min_book(find_max_val(1, num_coeffs, P34), sce->sf_idx[sfb]);
        else
            cb_p = cb_n;
        dist2 = quantize_and_encode_band_cost(s, NULL, &sce->prcoeffs[start_coef], NULL,
                                              P34, num_coeffs, sce->sf_idx[sfb],
                                              cb_p, s->lambda / band->threshold, INFINITY, NULL, 0);
        for (i = 0; i < num_coeffs; i++)
            dist_spec_err += (O34[i] - P34[i])*(O34[i] - P34[i]);
        dist_spec_err *= s->lambda / band->threshold;
        dist2 += dist_spec_err;

        if (dist2 <= dist1 && cb_p <= cb_n) {
            cost_pred += cost2;
            sce->ics.prediction_used[sfb] = 1;
            sce->band_alt[sfb]  = cb_n;
            sce->band_type[sfb] = cb_p;
            count++;
        } else {
            cost_pred += cost1;
            sce->band_alt[sfb] = cb_p;
        }
    }

    if (count && cost_coeffs < cost_pred) {
        count = 0;
        for (sfb = PRED_SFB_START; sfb < pmax; sfb++)
            RESTORE_PRED(sce, sfb);
        memset(&sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
    }

    sce->ics.predictor_present = !!count;
}

/**
 * Encoder predictors data.
 */
void ff_aac_encode_main_pred(AACEncContext *s, SingleChannelElement *sce)
{
    int sfb;
    IndividualChannelStream *ics = &sce->ics;
    const int pmax = FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[s->samplerate_index]);

    if (!ics->predictor_present)
        return;

    put_bits(&s->pb, 1, !!ics->predictor_reset_group);
    if (ics->predictor_reset_group)
        put_bits(&s->pb, 5, ics->predictor_reset_group);
    for (sfb = 0; sfb < pmax; sfb++)
        put_bits(&s->pb, 1, ics->prediction_used[sfb]);
}