Statistics
| Branch: | Revision:

ffmpeg / libavcodec / wmaenc.c @ 26f548bb

History | View | Annotate | Download (12.5 KB)

1
/*
2
 * WMA compatible encoder
3
 * Copyright (c) 2007 Michael Niedermayer
4
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * Libav is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with Libav; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
#include "avcodec.h"
23
#include "wma.h"
24

    
25
#undef NDEBUG
26
#include <assert.h>
27

    
28

    
29
static int encode_init(AVCodecContext * avctx){
30
    WMACodecContext *s = avctx->priv_data;
31
    int i, flags1, flags2;
32
    uint8_t *extradata;
33

    
34
    s->avctx = avctx;
35

    
36
    if(avctx->channels > MAX_CHANNELS)
37
        return -1;
38

    
39
    if(avctx->bit_rate < 24*1000)
40
        return -1;
41

    
42
    /* extract flag infos */
43
    flags1 = 0;
44
    flags2 = 1;
45
    if (avctx->codec->id == CODEC_ID_WMAV1) {
46
        extradata= av_malloc(4);
47
        avctx->extradata_size= 4;
48
        AV_WL16(extradata, flags1);
49
        AV_WL16(extradata+2, flags2);
50
    } else if (avctx->codec->id == CODEC_ID_WMAV2) {
51
        extradata= av_mallocz(10);
52
        avctx->extradata_size= 10;
53
        AV_WL32(extradata, flags1);
54
        AV_WL16(extradata+4, flags2);
55
    }else
56
        assert(0);
57
    avctx->extradata= extradata;
58
    s->use_exp_vlc = flags2 & 0x0001;
59
    s->use_bit_reservoir = flags2 & 0x0002;
60
    s->use_variable_block_len = flags2 & 0x0004;
61

    
62
    ff_wma_init(avctx, flags2);
63

    
64
    /* init MDCT */
65
    for(i = 0; i < s->nb_block_sizes; i++)
66
        ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0);
67

    
68
    avctx->block_align=
69
    s->block_align= avctx->bit_rate*(int64_t)s->frame_len / (avctx->sample_rate*8);
70
//av_log(NULL, AV_LOG_ERROR, "%d %d %d %d\n", s->block_align, avctx->bit_rate, s->frame_len, avctx->sample_rate);
71
    avctx->frame_size= s->frame_len;
72

    
73
    return 0;
74
}
75

    
76

    
77
static void apply_window_and_mdct(AVCodecContext * avctx, const signed short * audio, int len) {
78
    WMACodecContext *s = avctx->priv_data;
79
    int window_index= s->frame_len_bits - s->block_len_bits;
80
    FFTContext *mdct = &s->mdct_ctx[window_index];
81
    int i, j, channel;
82
    const float * win = s->windows[window_index];
83
    int window_len = 1 << s->block_len_bits;
84
    float n = window_len/2;
85

    
86
    for (channel = 0; channel < avctx->channels; channel++) {
87
        memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len);
88
        j = channel;
89
        for (i = 0; i < len; i++, j += avctx->channels){
90
            s->output[i+window_len]  = audio[j] / n * win[window_len - i - 1];
91
            s->frame_out[channel][i] = audio[j] / n * win[i];
92
        }
93
        mdct->mdct_calc(mdct, s->coefs[channel], s->output);
94
    }
95
}
96

    
97
//FIXME use for decoding too
98
static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
99
    int n;
100
    const uint16_t *ptr;
101
    float v, *q, max_scale, *q_end;
102

    
103
    ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
104
    q = s->exponents[ch];
105
    q_end = q + s->block_len;
106
    max_scale = 0;
107
    while (q < q_end) {
108
        /* XXX: use a table */
109
        v = pow(10, *exp_param++ * (1.0 / 16.0));
110
        max_scale= FFMAX(max_scale, v);
111
        n = *ptr++;
112
        do {
113
            *q++ = v;
114
        } while (--n);
115
    }
116
    s->max_exponent[ch] = max_scale;
117
}
118

    
119
static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
120
    int last_exp;
121
    const uint16_t *ptr;
122
    float *q, *q_end;
123

    
124
    ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
125
    q = s->exponents[ch];
126
    q_end = q + s->block_len;
127
    if (s->version == 1) {
128
        last_exp= *exp_param++;
129
        assert(last_exp-10 >= 0 && last_exp-10 < 32);
130
        put_bits(&s->pb, 5, last_exp - 10);
131
        q+= *ptr++;
132
    }else
133
        last_exp = 36;
134
    while (q < q_end) {
135
        int exp = *exp_param++;
136
        int code = exp - last_exp + 60;
137
        assert(code >= 0 && code < 120);
138
        put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]);
139
        /* XXX: use a table */
140
        q+= *ptr++;
141
        last_exp= exp;
142
    }
143
}
144

    
145
static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
146
    int v, bsize, ch, coef_nb_bits, parse_exponents;
147
    float mdct_norm;
148
    int nb_coefs[MAX_CHANNELS];
149
    static const int fixed_exp[25]={20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20};
150

    
151
    //FIXME remove duplication relative to decoder
152
    if (s->use_variable_block_len) {
153
        assert(0); //FIXME not implemented
154
    }else{
155
        /* fixed block len */
156
        s->next_block_len_bits = s->frame_len_bits;
157
        s->prev_block_len_bits = s->frame_len_bits;
158
        s->block_len_bits = s->frame_len_bits;
159
    }
160

    
161
    s->block_len = 1 << s->block_len_bits;
162
//     assert((s->block_pos + s->block_len) <= s->frame_len);
163
    bsize = s->frame_len_bits - s->block_len_bits;
164

    
165
    //FIXME factor
166
    v = s->coefs_end[bsize] - s->coefs_start;
167
    for(ch = 0; ch < s->nb_channels; ch++)
168
        nb_coefs[ch] = v;
169
    {
170
        int n4 = s->block_len / 2;
171
        mdct_norm = 1.0 / (float)n4;
172
        if (s->version == 1) {
173
            mdct_norm *= sqrt(n4);
174
        }
175
    }
176

    
177
    if (s->nb_channels == 2) {
178
        put_bits(&s->pb, 1, s->ms_stereo= 1);
179
    }
180

    
181
    for(ch = 0; ch < s->nb_channels; ch++) {
182
        s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always
183
        if (s->channel_coded[ch]) {
184
            init_exp(s, ch, fixed_exp);
185
        }
186
    }
187

    
188
    for(ch = 0; ch < s->nb_channels; ch++) {
189
        if (s->channel_coded[ch]) {
190
            WMACoef *coefs1;
191
            float *coefs, *exponents, mult;
192
            int i, n;
193

    
194
            coefs1 = s->coefs1[ch];
195
            exponents = s->exponents[ch];
196
            mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
197
            mult *= mdct_norm;
198
            coefs = src_coefs[ch];
199
            if (s->use_noise_coding && 0) {
200
                assert(0); //FIXME not implemented
201
            } else {
202
                coefs += s->coefs_start;
203
                n = nb_coefs[ch];
204
                for(i = 0;i < n; i++){
205
                    double t= *coefs++ / (exponents[i] * mult);
206
                    if(t<-32768 || t>32767)
207
                        return -1;
208

    
209
                    coefs1[i] = lrint(t);
210
                }
211
            }
212
        }
213
    }
214

    
215
    v = 0;
216
    for(ch = 0; ch < s->nb_channels; ch++) {
217
        int a = s->channel_coded[ch];
218
        put_bits(&s->pb, 1, a);
219
        v |= a;
220
    }
221

    
222
    if (!v)
223
        return 1;
224

    
225
    for(v= total_gain-1; v>=127; v-= 127)
226
        put_bits(&s->pb, 7, 127);
227
    put_bits(&s->pb, 7, v);
228

    
229
    coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
230

    
231
    if (s->use_noise_coding) {
232
        for(ch = 0; ch < s->nb_channels; ch++) {
233
            if (s->channel_coded[ch]) {
234
                int i, n;
235
                n = s->exponent_high_sizes[bsize];
236
                for(i=0;i<n;i++) {
237
                    put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
238
                    if (0)
239
                        nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
240
                }
241
            }
242
        }
243
    }
244

    
245
    parse_exponents = 1;
246
    if (s->block_len_bits != s->frame_len_bits) {
247
        put_bits(&s->pb, 1, parse_exponents);
248
    }
249

    
250
    if (parse_exponents) {
251
        for(ch = 0; ch < s->nb_channels; ch++) {
252
            if (s->channel_coded[ch]) {
253
                if (s->use_exp_vlc) {
254
                    encode_exp_vlc(s, ch, fixed_exp);
255
                } else {
256
                    assert(0); //FIXME not implemented
257
//                    encode_exp_lsp(s, ch);
258
                }
259
            }
260
        }
261
    } else {
262
        assert(0); //FIXME not implemented
263
    }
264

    
265
    for(ch = 0; ch < s->nb_channels; ch++) {
266
        if (s->channel_coded[ch]) {
267
            int run, tindex;
268
            WMACoef *ptr, *eptr;
269
            tindex = (ch == 1 && s->ms_stereo);
270
            ptr = &s->coefs1[ch][0];
271
            eptr = ptr + nb_coefs[ch];
272

    
273
            run=0;
274
            for(;ptr < eptr; ptr++){
275
                if(*ptr){
276
                    int level= *ptr;
277
                    int abs_level= FFABS(level);
278
                    int code= 0;
279
                    if(abs_level <= s->coef_vlcs[tindex]->max_level){
280
                        if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
281
                            code= run + s->int_table[tindex][abs_level-1];
282
                    }
283

    
284
                    assert(code < s->coef_vlcs[tindex]->n);
285
                    put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
286

    
287
                    if(code == 0){
288
                        if(1<<coef_nb_bits <= abs_level)
289
                            return -1;
290

    
291

    
292
                        //Workaround minor rounding differences for the regression tests, FIXME we should find and replace the problematic float by fixpoint for reg tests
293
                        if(abs_level == 0x71B && (s->avctx->flags & CODEC_FLAG_BITEXACT)) abs_level=0x71A;
294

    
295
                        put_bits(&s->pb, coef_nb_bits, abs_level);
296
                        put_bits(&s->pb, s->frame_len_bits, run);
297
                    }
298
                    put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere
299
                    run=0;
300
                }else{
301
                    run++;
302
                }
303
            }
304
            if(run)
305
                put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
306
        }
307
        if (s->version == 1 && s->nb_channels >= 2) {
308
            align_put_bits(&s->pb);
309
        }
310
    }
311
    return 0;
312
}
313

    
314
static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
315
    init_put_bits(&s->pb, buf, buf_size);
316

    
317
    if (s->use_bit_reservoir) {
318
        assert(0);//FIXME not implemented
319
    }else{
320
        if(encode_block(s, src_coefs, total_gain) < 0)
321
            return INT_MAX;
322
    }
323

    
324
    align_put_bits(&s->pb);
325

    
326
    return put_bits_count(&s->pb)/8 - s->block_align;
327
}
328

    
329
static int encode_superframe(AVCodecContext *avctx,
330
                            unsigned char *buf, int buf_size, void *data){
331
    WMACodecContext *s = avctx->priv_data;
332
    const short *samples = data;
333
    int i, total_gain;
334

    
335
    s->block_len_bits= s->frame_len_bits; //required by non variable block len
336
    s->block_len = 1 << s->block_len_bits;
337

    
338
    apply_window_and_mdct(avctx, samples, avctx->frame_size);
339

    
340
    if (s->ms_stereo) {
341
        float a, b;
342
        int i;
343

    
344
        for(i = 0; i < s->block_len; i++) {
345
            a = s->coefs[0][i]*0.5;
346
            b = s->coefs[1][i]*0.5;
347
            s->coefs[0][i] = a + b;
348
            s->coefs[1][i] = a - b;
349
        }
350
    }
351

    
352
#if 1
353
    total_gain= 128;
354
    for(i=64; i; i>>=1){
355
        int error= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
356
        if(error<0)
357
            total_gain-= i;
358
    }
359
#else
360
    total_gain= 90;
361
    best= encode_frame(s, s->coefs, buf, buf_size, total_gain);
362
    for(i=32; i; i>>=1){
363
        int scoreL= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
364
        int scoreR= encode_frame(s, s->coefs, buf, buf_size, total_gain+i);
365
        av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain);
366
        if(scoreL < FFMIN(best, scoreR)){
367
            best = scoreL;
368
            total_gain -= i;
369
        }else if(scoreR < best){
370
            best = scoreR;
371
            total_gain += i;
372
        }
373
    }
374
#endif
375

    
376
    encode_frame(s, s->coefs, buf, buf_size, total_gain);
377
    assert((put_bits_count(&s->pb) & 7) == 0);
378
    i= s->block_align - (put_bits_count(&s->pb)+7)/8;
379
    assert(i>=0);
380
    while(i--)
381
        put_bits(&s->pb, 8, 'N');
382

    
383
    flush_put_bits(&s->pb);
384
    return put_bits_ptr(&s->pb) - s->pb.buf;
385
}
386

    
387
AVCodec ff_wmav1_encoder =
388
{
389
    "wmav1",
390
    AVMEDIA_TYPE_AUDIO,
391
    CODEC_ID_WMAV1,
392
    sizeof(WMACodecContext),
393
    encode_init,
394
    encode_superframe,
395
    ff_wma_end,
396
    .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
397
    .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
398
};
399

    
400
AVCodec ff_wmav2_encoder =
401
{
402
    "wmav2",
403
    AVMEDIA_TYPE_AUDIO,
404
    CODEC_ID_WMAV2,
405
    sizeof(WMACodecContext),
406
    encode_init,
407
    encode_superframe,
408
    ff_wma_end,
409
    .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
410
    .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
411
};