00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022 #include "avcodec.h"
00023 #include "wma.h"
00024
00025 #undef NDEBUG
00026 #include <assert.h>
00027
00028
00029 static int encode_init(AVCodecContext * avctx){
00030 WMACodecContext *s = avctx->priv_data;
00031 int i, flags1, flags2;
00032 uint8_t *extradata;
00033
00034 s->avctx = avctx;
00035
00036 if(avctx->channels > MAX_CHANNELS)
00037 return -1;
00038
00039 if(avctx->bit_rate < 24*1000)
00040 return -1;
00041
00042
00043 flags1 = 0;
00044 flags2 = 1;
00045 if (avctx->codec->id == CODEC_ID_WMAV1) {
00046 extradata= av_malloc(4);
00047 avctx->extradata_size= 4;
00048 AV_WL16(extradata, flags1);
00049 AV_WL16(extradata+2, flags2);
00050 } else if (avctx->codec->id == CODEC_ID_WMAV2) {
00051 extradata= av_mallocz(10);
00052 avctx->extradata_size= 10;
00053 AV_WL32(extradata, flags1);
00054 AV_WL16(extradata+4, flags2);
00055 }else
00056 assert(0);
00057 avctx->extradata= extradata;
00058 s->use_exp_vlc = flags2 & 0x0001;
00059 s->use_bit_reservoir = flags2 & 0x0002;
00060 s->use_variable_block_len = flags2 & 0x0004;
00061
00062 ff_wma_init(avctx, flags2);
00063
00064
00065 for(i = 0; i < s->nb_block_sizes; i++)
00066 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0);
00067
00068 avctx->block_align=
00069 s->block_align= avctx->bit_rate*(int64_t)s->frame_len / (avctx->sample_rate*8);
00070
00071 avctx->frame_size= s->frame_len;
00072
00073 return 0;
00074 }
00075
00076
00077 static void apply_window_and_mdct(AVCodecContext * avctx, signed short * audio, int len) {
00078 WMACodecContext *s = avctx->priv_data;
00079 int window_index= s->frame_len_bits - s->block_len_bits;
00080 int i, j, channel;
00081 const float * win = s->windows[window_index];
00082 int window_len = 1 << s->block_len_bits;
00083 float n = window_len/2;
00084
00085 for (channel = 0; channel < avctx->channels; channel++) {
00086 memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len);
00087 j = channel;
00088 for (i = 0; i < len; i++, j += avctx->channels){
00089 s->output[i+window_len] = audio[j] / n * win[window_len - i - 1];
00090 s->frame_out[channel][i] = audio[j] / n * win[i];
00091 }
00092 ff_mdct_calc(&s->mdct_ctx[window_index], s->coefs[channel], s->output);
00093 }
00094 }
00095
00096
00097 static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
00098 int n;
00099 const uint16_t *ptr;
00100 float v, *q, max_scale, *q_end;
00101
00102 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
00103 q = s->exponents[ch];
00104 q_end = q + s->block_len;
00105 max_scale = 0;
00106 while (q < q_end) {
00107
00108 v = pow(10, *exp_param++ * (1.0 / 16.0));
00109 max_scale= FFMAX(max_scale, v);
00110 n = *ptr++;
00111 do {
00112 *q++ = v;
00113 } while (--n);
00114 }
00115 s->max_exponent[ch] = max_scale;
00116 }
00117
00118 static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
00119 int last_exp;
00120 const uint16_t *ptr;
00121 float *q, *q_end;
00122
00123 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
00124 q = s->exponents[ch];
00125 q_end = q + s->block_len;
00126 if (s->version == 1) {
00127 last_exp= *exp_param++;
00128 assert(last_exp-10 >= 0 && last_exp-10 < 32);
00129 put_bits(&s->pb, 5, last_exp - 10);
00130 q+= *ptr++;
00131 }else
00132 last_exp = 36;
00133 while (q < q_end) {
00134 int exp = *exp_param++;
00135 int code = exp - last_exp + 60;
00136 assert(code >= 0 && code < 120);
00137 put_bits(&s->pb, ff_wma_scale_huffbits[code], ff_wma_scale_huffcodes[code]);
00138
00139 q+= *ptr++;
00140 last_exp= exp;
00141 }
00142 }
00143
00144 static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
00145 int v, bsize, ch, coef_nb_bits, parse_exponents;
00146 float mdct_norm;
00147 int nb_coefs[MAX_CHANNELS];
00148 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};
00149
00150
00151 if (s->use_variable_block_len) {
00152 assert(0);
00153 }else{
00154
00155 s->next_block_len_bits = s->frame_len_bits;
00156 s->prev_block_len_bits = s->frame_len_bits;
00157 s->block_len_bits = s->frame_len_bits;
00158 }
00159
00160 s->block_len = 1 << s->block_len_bits;
00161
00162 bsize = s->frame_len_bits - s->block_len_bits;
00163
00164
00165 v = s->coefs_end[bsize] - s->coefs_start;
00166 for(ch = 0; ch < s->nb_channels; ch++)
00167 nb_coefs[ch] = v;
00168 {
00169 int n4 = s->block_len / 2;
00170 mdct_norm = 1.0 / (float)n4;
00171 if (s->version == 1) {
00172 mdct_norm *= sqrt(n4);
00173 }
00174 }
00175
00176 if (s->nb_channels == 2) {
00177 put_bits(&s->pb, 1, s->ms_stereo= 1);
00178 }
00179
00180 for(ch = 0; ch < s->nb_channels; ch++) {
00181 s->channel_coded[ch] = 1;
00182 if (s->channel_coded[ch]) {
00183 init_exp(s, ch, fixed_exp);
00184 }
00185 }
00186
00187 for(ch = 0; ch < s->nb_channels; ch++) {
00188 if (s->channel_coded[ch]) {
00189 int16_t *coefs1;
00190 float *coefs, *exponents, mult;
00191 int i, n;
00192
00193 coefs1 = s->coefs1[ch];
00194 exponents = s->exponents[ch];
00195 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
00196 mult *= mdct_norm;
00197 coefs = src_coefs[ch];
00198 if (s->use_noise_coding && 0) {
00199 assert(0);
00200 } else {
00201 coefs += s->coefs_start;
00202 n = nb_coefs[ch];
00203 for(i = 0;i < n; i++){
00204 double t= *coefs++ / (exponents[i] * mult);
00205 if(t<-32768 || t>32767)
00206 return -1;
00207
00208 coefs1[i] = lrint(t);
00209 }
00210 }
00211 }
00212 }
00213
00214 v = 0;
00215 for(ch = 0; ch < s->nb_channels; ch++) {
00216 int a = s->channel_coded[ch];
00217 put_bits(&s->pb, 1, a);
00218 v |= a;
00219 }
00220
00221 if (!v)
00222 return 1;
00223
00224 for(v= total_gain-1; v>=127; v-= 127)
00225 put_bits(&s->pb, 7, 127);
00226 put_bits(&s->pb, 7, v);
00227
00228 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
00229
00230 if (s->use_noise_coding) {
00231 for(ch = 0; ch < s->nb_channels; ch++) {
00232 if (s->channel_coded[ch]) {
00233 int i, n;
00234 n = s->exponent_high_sizes[bsize];
00235 for(i=0;i<n;i++) {
00236 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
00237 if (0)
00238 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
00239 }
00240 }
00241 }
00242 }
00243
00244 parse_exponents = 1;
00245 if (s->block_len_bits != s->frame_len_bits) {
00246 put_bits(&s->pb, 1, parse_exponents);
00247 }
00248
00249 if (parse_exponents) {
00250 for(ch = 0; ch < s->nb_channels; ch++) {
00251 if (s->channel_coded[ch]) {
00252 if (s->use_exp_vlc) {
00253 encode_exp_vlc(s, ch, fixed_exp);
00254 } else {
00255 assert(0);
00256
00257 }
00258 }
00259 }
00260 } else {
00261 assert(0);
00262 }
00263
00264 for(ch = 0; ch < s->nb_channels; ch++) {
00265 if (s->channel_coded[ch]) {
00266 int run, tindex;
00267 int16_t *ptr, *eptr;
00268 tindex = (ch == 1 && s->ms_stereo);
00269 ptr = &s->coefs1[ch][0];
00270 eptr = ptr + nb_coefs[ch];
00271
00272 run=0;
00273 for(;ptr < eptr; ptr++){
00274 if(*ptr){
00275 int level= *ptr;
00276 int abs_level= FFABS(level);
00277 int code= 0;
00278 if(abs_level <= s->coef_vlcs[tindex]->max_level){
00279 if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
00280 code= run + s->int_table[tindex][abs_level-1];
00281 }
00282
00283 assert(code < s->coef_vlcs[tindex]->n);
00284 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
00285
00286 if(code == 0){
00287 if(1<<coef_nb_bits <= abs_level)
00288 return -1;
00289
00290
00291
00292 if(abs_level == 0x71B && (s->avctx->flags & CODEC_FLAG_BITEXACT)) abs_level=0x71A;
00293
00294 put_bits(&s->pb, coef_nb_bits, abs_level);
00295 put_bits(&s->pb, s->frame_len_bits, run);
00296 }
00297 put_bits(&s->pb, 1, level < 0);
00298 run=0;
00299 }else{
00300 run++;
00301 }
00302 }
00303 if(run)
00304 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
00305 }
00306 if (s->version == 1 && s->nb_channels >= 2) {
00307 align_put_bits(&s->pb);
00308 }
00309 }
00310 return 0;
00311 }
00312
00313 static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
00314 init_put_bits(&s->pb, buf, buf_size);
00315
00316 if (s->use_bit_reservoir) {
00317 assert(0);
00318 }else{
00319 if(encode_block(s, src_coefs, total_gain) < 0)
00320 return INT_MAX;
00321 }
00322
00323 align_put_bits(&s->pb);
00324
00325 return put_bits_count(&s->pb)/8 - s->block_align;
00326 }
00327
00328 static int encode_superframe(AVCodecContext *avctx,
00329 unsigned char *buf, int buf_size, void *data){
00330 WMACodecContext *s = avctx->priv_data;
00331 short *samples = data;
00332 int i, total_gain;
00333
00334 s->block_len_bits= s->frame_len_bits;
00335 s->block_len = 1 << s->block_len_bits;
00336
00337 apply_window_and_mdct(avctx, samples, avctx->frame_size);
00338
00339 if (s->ms_stereo) {
00340 float a, b;
00341 int i;
00342
00343 for(i = 0; i < s->block_len; i++) {
00344 a = s->coefs[0][i]*0.5;
00345 b = s->coefs[1][i]*0.5;
00346 s->coefs[0][i] = a + b;
00347 s->coefs[1][i] = a - b;
00348 }
00349 }
00350
00351 #if 1
00352 total_gain= 128;
00353 for(i=64; i; i>>=1){
00354 int error= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
00355 if(error<0)
00356 total_gain-= i;
00357 }
00358 #else
00359 total_gain= 90;
00360 best= encode_frame(s, s->coefs, buf, buf_size, total_gain);
00361 for(i=32; i; i>>=1){
00362 int scoreL= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
00363 int scoreR= encode_frame(s, s->coefs, buf, buf_size, total_gain+i);
00364 av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain);
00365 if(scoreL < FFMIN(best, scoreR)){
00366 best = scoreL;
00367 total_gain -= i;
00368 }else if(scoreR < best){
00369 best = scoreR;
00370 total_gain += i;
00371 }
00372 }
00373 #endif
00374
00375 encode_frame(s, s->coefs, buf, buf_size, total_gain);
00376 assert((put_bits_count(&s->pb) & 7) == 0);
00377 i= s->block_align - (put_bits_count(&s->pb)+7)/8;
00378 assert(i>=0);
00379 while(i--)
00380 put_bits(&s->pb, 8, 'N');
00381
00382 flush_put_bits(&s->pb);
00383 return pbBufPtr(&s->pb) - s->pb.buf;
00384 }
00385
00386 AVCodec wmav1_encoder =
00387 {
00388 "wmav1",
00389 CODEC_TYPE_AUDIO,
00390 CODEC_ID_WMAV1,
00391 sizeof(WMACodecContext),
00392 encode_init,
00393 encode_superframe,
00394 ff_wma_end,
00395 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
00396 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
00397 };
00398
00399 AVCodec wmav2_encoder =
00400 {
00401 "wmav2",
00402 CODEC_TYPE_AUDIO,
00403 CODEC_ID_WMAV2,
00404 sizeof(WMACodecContext),
00405 encode_init,
00406 encode_superframe,
00407 ff_wma_end,
00408 .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
00409 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
00410 };