2 * WMA compatible encoder
3 * Copyright (c) 2007 Michael Niedermayer
5 * This file is part of FFmpeg.
7 * FFmpeg 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.
12 * FFmpeg 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 #include "libavutil/avassert.h"
28 static int encode_init(AVCodecContext * avctx){
29 WMACodecContext *s = avctx->priv_data;
30 int i, flags1, flags2, block_align;
35 if(avctx->channels > MAX_CHANNELS) {
36 av_log(avctx, AV_LOG_ERROR, "too many channels: got %i, need %i or fewer\n",
37 avctx->channels, MAX_CHANNELS);
38 return AVERROR(EINVAL);
41 if (avctx->sample_rate > 48000) {
42 av_log(avctx, AV_LOG_ERROR, "sample rate is too high: %d > 48kHz\n",
44 return AVERROR(EINVAL);
47 if(avctx->bit_rate < 24*1000) {
48 av_log(avctx, AV_LOG_ERROR, "bitrate too low: got %i, need 24000 or higher\n",
50 return AVERROR(EINVAL);
53 /* extract flag infos */
56 if (avctx->codec->id == AV_CODEC_ID_WMAV1) {
57 extradata= av_malloc(4);
58 avctx->extradata_size= 4;
59 AV_WL16(extradata, flags1);
60 AV_WL16(extradata+2, flags2);
61 } else if (avctx->codec->id == AV_CODEC_ID_WMAV2) {
62 extradata= av_mallocz(10);
63 avctx->extradata_size= 10;
64 AV_WL32(extradata, flags1);
65 AV_WL16(extradata+4, flags2);
68 avctx->extradata= extradata;
69 s->use_exp_vlc = flags2 & 0x0001;
70 s->use_bit_reservoir = flags2 & 0x0002;
71 s->use_variable_block_len = flags2 & 0x0004;
72 if (avctx->channels == 2)
75 ff_wma_init(avctx, flags2);
78 for(i = 0; i < s->nb_block_sizes; i++)
79 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0);
81 block_align = avctx->bit_rate * (int64_t)s->frame_len /
82 (avctx->sample_rate * 8);
83 block_align = FFMIN(block_align, MAX_CODED_SUPERFRAME_SIZE);
84 avctx->block_align = block_align;
86 avctx->frame_size = avctx->delay = s->frame_len;
92 static void apply_window_and_mdct(AVCodecContext * avctx, const AVFrame *frame)
94 WMACodecContext *s = avctx->priv_data;
95 float **audio = (float **)frame->extended_data;
96 int len = frame->nb_samples;
97 int window_index= s->frame_len_bits - s->block_len_bits;
98 FFTContext *mdct = &s->mdct_ctx[window_index];
100 const float * win = s->windows[window_index];
101 int window_len = 1 << s->block_len_bits;
102 float n = 2.0 * 32768.0 / window_len;
104 for (ch = 0; ch < avctx->channels; ch++) {
105 memcpy(s->output, s->frame_out[ch], window_len * sizeof(*s->output));
106 s->fdsp.vector_fmul_scalar(s->frame_out[ch], audio[ch], n, len);
107 s->fdsp.vector_fmul_reverse(&s->output[window_len], s->frame_out[ch], win, len);
108 s->fdsp.vector_fmul(s->frame_out[ch], s->frame_out[ch], win, len);
109 mdct->mdct_calc(mdct, s->coefs[ch], s->output);
113 //FIXME use for decoding too
114 static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
117 float v, *q, max_scale, *q_end;
119 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
120 q = s->exponents[ch];
121 q_end = q + s->block_len;
124 /* XXX: use a table */
125 v = pow(10, *exp_param++ * (1.0 / 16.0));
126 max_scale= FFMAX(max_scale, v);
132 s->max_exponent[ch] = max_scale;
135 static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
140 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
141 q = s->exponents[ch];
142 q_end = q + s->block_len;
143 if (s->version == 1) {
144 last_exp= *exp_param++;
145 av_assert0(last_exp-10 >= 0 && last_exp-10 < 32);
146 put_bits(&s->pb, 5, last_exp - 10);
151 int exp = *exp_param++;
152 int code = exp - last_exp + 60;
153 av_assert1(code >= 0 && code < 120);
154 put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]);
155 /* XXX: use a table */
161 static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
162 int v, bsize, ch, coef_nb_bits, parse_exponents;
164 int nb_coefs[MAX_CHANNELS];
165 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};
167 //FIXME remove duplication relative to decoder
168 if (s->use_variable_block_len) {
169 av_assert0(0); //FIXME not implemented
171 /* fixed block len */
172 s->next_block_len_bits = s->frame_len_bits;
173 s->prev_block_len_bits = s->frame_len_bits;
174 s->block_len_bits = s->frame_len_bits;
177 s->block_len = 1 << s->block_len_bits;
178 // assert((s->block_pos + s->block_len) <= s->frame_len);
179 bsize = s->frame_len_bits - s->block_len_bits;
182 v = s->coefs_end[bsize] - s->coefs_start;
183 for (ch = 0; ch < s->avctx->channels; ch++)
186 int n4 = s->block_len / 2;
187 mdct_norm = 1.0 / (float)n4;
188 if (s->version == 1) {
189 mdct_norm *= sqrt(n4);
193 if (s->avctx->channels == 2) {
194 put_bits(&s->pb, 1, !!s->ms_stereo);
197 for (ch = 0; ch < s->avctx->channels; ch++) {
198 s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always
199 if (s->channel_coded[ch]) {
200 init_exp(s, ch, fixed_exp);
204 for (ch = 0; ch < s->avctx->channels; ch++) {
205 if (s->channel_coded[ch]) {
207 float *coefs, *exponents, mult;
210 coefs1 = s->coefs1[ch];
211 exponents = s->exponents[ch];
212 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
214 coefs = src_coefs[ch];
215 if (s->use_noise_coding && 0) {
216 av_assert0(0); //FIXME not implemented
218 coefs += s->coefs_start;
220 for(i = 0;i < n; i++){
221 double t= *coefs++ / (exponents[i] * mult);
222 if(t<-32768 || t>32767)
225 coefs1[i] = lrint(t);
232 for (ch = 0; ch < s->avctx->channels; ch++) {
233 int a = s->channel_coded[ch];
234 put_bits(&s->pb, 1, a);
241 for(v= total_gain-1; v>=127; v-= 127)
242 put_bits(&s->pb, 7, 127);
243 put_bits(&s->pb, 7, v);
245 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
247 if (s->use_noise_coding) {
248 for (ch = 0; ch < s->avctx->channels; ch++) {
249 if (s->channel_coded[ch]) {
251 n = s->exponent_high_sizes[bsize];
253 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
255 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
262 if (s->block_len_bits != s->frame_len_bits) {
263 put_bits(&s->pb, 1, parse_exponents);
266 if (parse_exponents) {
267 for (ch = 0; ch < s->avctx->channels; ch++) {
268 if (s->channel_coded[ch]) {
269 if (s->use_exp_vlc) {
270 encode_exp_vlc(s, ch, fixed_exp);
272 av_assert0(0); //FIXME not implemented
273 // encode_exp_lsp(s, ch);
278 av_assert0(0); //FIXME not implemented
281 for (ch = 0; ch < s->avctx->channels; ch++) {
282 if (s->channel_coded[ch]) {
285 tindex = (ch == 1 && s->ms_stereo);
286 ptr = &s->coefs1[ch][0];
287 eptr = ptr + nb_coefs[ch];
290 for(;ptr < eptr; ptr++){
293 int abs_level= FFABS(level);
295 if(abs_level <= s->coef_vlcs[tindex]->max_level){
296 if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
297 code= run + s->int_table[tindex][abs_level-1];
300 av_assert2(code < s->coef_vlcs[tindex]->n);
301 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
304 if(1<<coef_nb_bits <= abs_level)
307 put_bits(&s->pb, coef_nb_bits, abs_level);
308 put_bits(&s->pb, s->frame_len_bits, run);
310 put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere
317 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
319 if (s->version == 1 && s->avctx->channels >= 2) {
320 avpriv_align_put_bits(&s->pb);
326 static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
327 init_put_bits(&s->pb, buf, buf_size);
329 if (s->use_bit_reservoir) {
330 av_assert0(0);//FIXME not implemented
332 if(encode_block(s, src_coefs, total_gain) < 0)
336 avpriv_align_put_bits(&s->pb);
338 return put_bits_count(&s->pb) / 8 - s->avctx->block_align;
341 static int encode_superframe(AVCodecContext *avctx, AVPacket *avpkt,
342 const AVFrame *frame, int *got_packet_ptr)
344 WMACodecContext *s = avctx->priv_data;
345 int i, total_gain, ret, error;
347 s->block_len_bits= s->frame_len_bits; //required by non variable block len
348 s->block_len = 1 << s->block_len_bits;
350 apply_window_and_mdct(avctx, frame);
356 for(i = 0; i < s->block_len; i++) {
357 a = s->coefs[0][i]*0.5;
358 b = s->coefs[1][i]*0.5;
359 s->coefs[0][i] = a + b;
360 s->coefs[1][i] = a - b;
364 if ((ret = ff_alloc_packet2(avctx, avpkt, 2 * MAX_CODED_SUPERFRAME_SIZE)) < 0)
369 error = encode_frame(s, s->coefs, avpkt->data, avpkt->size,
375 while(total_gain <= 128 && error > 0)
376 error = encode_frame(s, s->coefs, avpkt->data, avpkt->size, total_gain++);
377 av_assert0((put_bits_count(&s->pb) & 7) == 0);
378 i= avctx->block_align - (put_bits_count(&s->pb)+7)/8;
381 put_bits(&s->pb, 8, 'N');
383 flush_put_bits(&s->pb);
384 av_assert0(put_bits_ptr(&s->pb) - s->pb.buf == avctx->block_align);
386 if (frame->pts != AV_NOPTS_VALUE)
387 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay);
389 avpkt->size = avctx->block_align;
394 #if CONFIG_WMAV1_ENCODER
395 AVCodec ff_wmav1_encoder = {
397 .type = AVMEDIA_TYPE_AUDIO,
398 .id = AV_CODEC_ID_WMAV1,
399 .priv_data_size = sizeof(WMACodecContext),
401 .encode2 = encode_superframe,
403 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
404 AV_SAMPLE_FMT_NONE },
405 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
408 #if CONFIG_WMAV2_ENCODER
409 AVCodec ff_wmav2_encoder = {
411 .type = AVMEDIA_TYPE_AUDIO,
412 .id = AV_CODEC_ID_WMAV2,
413 .priv_data_size = sizeof(WMACodecContext),
415 .encode2 = encode_superframe,
417 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
418 AV_SAMPLE_FMT_NONE },
419 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),