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
29 static int encode_init(AVCodecContext * avctx){
30 WMACodecContext *s = avctx->priv_data;
31 int i, flags1, flags2;
36 if(avctx->channels > MAX_CHANNELS) {
37 av_log(avctx, AV_LOG_ERROR, "too many channels: got %i, need %i or fewer",
38 avctx->channels, MAX_CHANNELS);
39 return AVERROR(EINVAL);
42 if(avctx->bit_rate < 24*1000) {
43 av_log(avctx, AV_LOG_ERROR, "bitrate too low: got %i, need 24000 or higher\n",
45 return AVERROR(EINVAL);
48 /* extract flag infos */
51 if (avctx->codec->id == CODEC_ID_WMAV1) {
52 extradata= av_malloc(4);
53 avctx->extradata_size= 4;
54 AV_WL16(extradata, flags1);
55 AV_WL16(extradata+2, flags2);
56 } else if (avctx->codec->id == CODEC_ID_WMAV2) {
57 extradata= av_mallocz(10);
58 avctx->extradata_size= 10;
59 AV_WL32(extradata, flags1);
60 AV_WL16(extradata+4, flags2);
63 avctx->extradata= extradata;
64 s->use_exp_vlc = flags2 & 0x0001;
65 s->use_bit_reservoir = flags2 & 0x0002;
66 s->use_variable_block_len = flags2 & 0x0004;
68 ff_wma_init(avctx, flags2);
71 for(i = 0; i < s->nb_block_sizes; i++)
72 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0);
75 s->block_align= avctx->bit_rate*(int64_t)s->frame_len / (avctx->sample_rate*8);
76 //av_log(NULL, AV_LOG_ERROR, "%d %d %d %d\n", s->block_align, avctx->bit_rate, s->frame_len, avctx->sample_rate);
77 avctx->frame_size= s->frame_len;
83 static void apply_window_and_mdct(AVCodecContext * avctx, const signed short * audio, int len) {
84 WMACodecContext *s = avctx->priv_data;
85 int window_index= s->frame_len_bits - s->block_len_bits;
86 FFTContext *mdct = &s->mdct_ctx[window_index];
88 const float * win = s->windows[window_index];
89 int window_len = 1 << s->block_len_bits;
90 float n = window_len/2;
92 for (channel = 0; channel < avctx->channels; channel++) {
93 memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len);
95 for (i = 0; i < len; i++, j += avctx->channels){
96 s->output[i+window_len] = audio[j] / n * win[window_len - i - 1];
97 s->frame_out[channel][i] = audio[j] / n * win[i];
99 mdct->mdct_calc(mdct, s->coefs[channel], s->output);
103 //FIXME use for decoding too
104 static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
107 float v, *q, max_scale, *q_end;
109 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
110 q = s->exponents[ch];
111 q_end = q + s->block_len;
114 /* XXX: use a table */
115 v = pow(10, *exp_param++ * (1.0 / 16.0));
116 max_scale= FFMAX(max_scale, v);
122 s->max_exponent[ch] = max_scale;
125 static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
130 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
131 q = s->exponents[ch];
132 q_end = q + s->block_len;
133 if (s->version == 1) {
134 last_exp= *exp_param++;
135 assert(last_exp-10 >= 0 && last_exp-10 < 32);
136 put_bits(&s->pb, 5, last_exp - 10);
141 int exp = *exp_param++;
142 int code = exp - last_exp + 60;
143 assert(code >= 0 && code < 120);
144 put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]);
145 /* XXX: use a table */
151 static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
152 int v, bsize, ch, coef_nb_bits, parse_exponents;
154 int nb_coefs[MAX_CHANNELS];
155 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};
157 //FIXME remove duplication relative to decoder
158 if (s->use_variable_block_len) {
159 assert(0); //FIXME not implemented
161 /* fixed block len */
162 s->next_block_len_bits = s->frame_len_bits;
163 s->prev_block_len_bits = s->frame_len_bits;
164 s->block_len_bits = s->frame_len_bits;
167 s->block_len = 1 << s->block_len_bits;
168 // assert((s->block_pos + s->block_len) <= s->frame_len);
169 bsize = s->frame_len_bits - s->block_len_bits;
172 v = s->coefs_end[bsize] - s->coefs_start;
173 for(ch = 0; ch < s->nb_channels; ch++)
176 int n4 = s->block_len / 2;
177 mdct_norm = 1.0 / (float)n4;
178 if (s->version == 1) {
179 mdct_norm *= sqrt(n4);
183 if (s->nb_channels == 2) {
184 put_bits(&s->pb, 1, s->ms_stereo= 1);
187 for(ch = 0; ch < s->nb_channels; ch++) {
188 s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always
189 if (s->channel_coded[ch]) {
190 init_exp(s, ch, fixed_exp);
194 for(ch = 0; ch < s->nb_channels; ch++) {
195 if (s->channel_coded[ch]) {
197 float *coefs, *exponents, mult;
200 coefs1 = s->coefs1[ch];
201 exponents = s->exponents[ch];
202 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
204 coefs = src_coefs[ch];
205 if (s->use_noise_coding && 0) {
206 assert(0); //FIXME not implemented
208 coefs += s->coefs_start;
210 for(i = 0;i < n; i++){
211 double t= *coefs++ / (exponents[i] * mult);
212 if(t<-32768 || t>32767)
215 coefs1[i] = lrint(t);
222 for(ch = 0; ch < s->nb_channels; ch++) {
223 int a = s->channel_coded[ch];
224 put_bits(&s->pb, 1, a);
231 for(v= total_gain-1; v>=127; v-= 127)
232 put_bits(&s->pb, 7, 127);
233 put_bits(&s->pb, 7, v);
235 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
237 if (s->use_noise_coding) {
238 for(ch = 0; ch < s->nb_channels; ch++) {
239 if (s->channel_coded[ch]) {
241 n = s->exponent_high_sizes[bsize];
243 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
245 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
252 if (s->block_len_bits != s->frame_len_bits) {
253 put_bits(&s->pb, 1, parse_exponents);
256 if (parse_exponents) {
257 for(ch = 0; ch < s->nb_channels; ch++) {
258 if (s->channel_coded[ch]) {
259 if (s->use_exp_vlc) {
260 encode_exp_vlc(s, ch, fixed_exp);
262 assert(0); //FIXME not implemented
263 // encode_exp_lsp(s, ch);
268 assert(0); //FIXME not implemented
271 for(ch = 0; ch < s->nb_channels; ch++) {
272 if (s->channel_coded[ch]) {
275 tindex = (ch == 1 && s->ms_stereo);
276 ptr = &s->coefs1[ch][0];
277 eptr = ptr + nb_coefs[ch];
280 for(;ptr < eptr; ptr++){
283 int abs_level= FFABS(level);
285 if(abs_level <= s->coef_vlcs[tindex]->max_level){
286 if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
287 code= run + s->int_table[tindex][abs_level-1];
290 assert(code < s->coef_vlcs[tindex]->n);
291 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
294 if(1<<coef_nb_bits <= abs_level)
298 //Workaround minor rounding differences for the regression tests, FIXME we should find and replace the problematic float by fixpoint for reg tests
299 if(abs_level == 0x71B && (s->avctx->flags & CODEC_FLAG_BITEXACT)) abs_level=0x71A;
301 put_bits(&s->pb, coef_nb_bits, abs_level);
302 put_bits(&s->pb, s->frame_len_bits, run);
304 put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere
311 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
313 if (s->version == 1 && s->nb_channels >= 2) {
314 align_put_bits(&s->pb);
320 static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
321 init_put_bits(&s->pb, buf, buf_size);
323 if (s->use_bit_reservoir) {
324 assert(0);//FIXME not implemented
326 if(encode_block(s, src_coefs, total_gain) < 0)
330 align_put_bits(&s->pb);
332 return put_bits_count(&s->pb)/8 - s->block_align;
335 static int encode_superframe(AVCodecContext *avctx,
336 unsigned char *buf, int buf_size, void *data){
337 WMACodecContext *s = avctx->priv_data;
338 const short *samples = data;
341 s->block_len_bits= s->frame_len_bits; //required by non variable block len
342 s->block_len = 1 << s->block_len_bits;
344 apply_window_and_mdct(avctx, samples, avctx->frame_size);
350 for(i = 0; i < s->block_len; i++) {
351 a = s->coefs[0][i]*0.5;
352 b = s->coefs[1][i]*0.5;
353 s->coefs[0][i] = a + b;
354 s->coefs[1][i] = a - b;
361 int error= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
367 best= encode_frame(s, s->coefs, buf, buf_size, total_gain);
369 int scoreL= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
370 int scoreR= encode_frame(s, s->coefs, buf, buf_size, total_gain+i);
371 av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain);
372 if(scoreL < FFMIN(best, scoreR)){
375 }else if(scoreR < best){
382 encode_frame(s, s->coefs, buf, buf_size, total_gain);
383 assert((put_bits_count(&s->pb) & 7) == 0);
384 i= s->block_align - (put_bits_count(&s->pb)+7)/8;
387 put_bits(&s->pb, 8, 'N');
389 flush_put_bits(&s->pb);
390 return put_bits_ptr(&s->pb) - s->pb.buf;
393 AVCodec ff_wmav1_encoder =
398 sizeof(WMACodecContext),
402 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
403 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
406 AVCodec ff_wmav2_encoder =
411 sizeof(WMACodecContext),
415 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
416 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),