Libav 0.7.1
libavcodec/alacenc.c
Go to the documentation of this file.
00001 
00022 #include "avcodec.h"
00023 #include "put_bits.h"
00024 #include "dsputil.h"
00025 #include "lpc.h"
00026 #include "mathops.h"
00027 
00028 #define DEFAULT_FRAME_SIZE        4096
00029 #define DEFAULT_SAMPLE_SIZE       16
00030 #define MAX_CHANNELS              8
00031 #define ALAC_EXTRADATA_SIZE       36
00032 #define ALAC_FRAME_HEADER_SIZE    55
00033 #define ALAC_FRAME_FOOTER_SIZE    3
00034 
00035 #define ALAC_ESCAPE_CODE          0x1FF
00036 #define ALAC_MAX_LPC_ORDER        30
00037 #define DEFAULT_MAX_PRED_ORDER    6
00038 #define DEFAULT_MIN_PRED_ORDER    4
00039 #define ALAC_MAX_LPC_PRECISION    9
00040 #define ALAC_MAX_LPC_SHIFT        9
00041 
00042 #define ALAC_CHMODE_LEFT_RIGHT    0
00043 #define ALAC_CHMODE_LEFT_SIDE     1
00044 #define ALAC_CHMODE_RIGHT_SIDE    2
00045 #define ALAC_CHMODE_MID_SIDE      3
00046 
00047 typedef struct RiceContext {
00048     int history_mult;
00049     int initial_history;
00050     int k_modifier;
00051     int rice_modifier;
00052 } RiceContext;
00053 
00054 typedef struct AlacLPCContext {
00055     int lpc_order;
00056     int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
00057     int lpc_quant;
00058 } AlacLPCContext;
00059 
00060 typedef struct AlacEncodeContext {
00061     int compression_level;
00062     int min_prediction_order;
00063     int max_prediction_order;
00064     int max_coded_frame_size;
00065     int write_sample_size;
00066     int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
00067     int32_t predictor_buf[DEFAULT_FRAME_SIZE];
00068     int interlacing_shift;
00069     int interlacing_leftweight;
00070     PutBitContext pbctx;
00071     RiceContext rc;
00072     AlacLPCContext lpc[MAX_CHANNELS];
00073     LPCContext lpc_ctx;
00074     AVCodecContext *avctx;
00075 } AlacEncodeContext;
00076 
00077 
00078 static void init_sample_buffers(AlacEncodeContext *s, const int16_t *input_samples)
00079 {
00080     int ch, i;
00081 
00082     for(ch=0;ch<s->avctx->channels;ch++) {
00083         const int16_t *sptr = input_samples + ch;
00084         for(i=0;i<s->avctx->frame_size;i++) {
00085             s->sample_buf[ch][i] = *sptr;
00086             sptr += s->avctx->channels;
00087         }
00088     }
00089 }
00090 
00091 static void encode_scalar(AlacEncodeContext *s, int x, int k, int write_sample_size)
00092 {
00093     int divisor, q, r;
00094 
00095     k = FFMIN(k, s->rc.k_modifier);
00096     divisor = (1<<k) - 1;
00097     q = x / divisor;
00098     r = x % divisor;
00099 
00100     if(q > 8) {
00101         // write escape code and sample value directly
00102         put_bits(&s->pbctx, 9, ALAC_ESCAPE_CODE);
00103         put_bits(&s->pbctx, write_sample_size, x);
00104     } else {
00105         if(q)
00106             put_bits(&s->pbctx, q, (1<<q) - 1);
00107         put_bits(&s->pbctx, 1, 0);
00108 
00109         if(k != 1) {
00110             if(r > 0)
00111                 put_bits(&s->pbctx, k, r+1);
00112             else
00113                 put_bits(&s->pbctx, k-1, 0);
00114         }
00115     }
00116 }
00117 
00118 static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
00119 {
00120     put_bits(&s->pbctx, 3,  s->avctx->channels-1);          // No. of channels -1
00121     put_bits(&s->pbctx, 16, 0);                             // Seems to be zero
00122     put_bits(&s->pbctx, 1,  1);                             // Sample count is in the header
00123     put_bits(&s->pbctx, 2,  0);                             // FIXME: Wasted bytes field
00124     put_bits(&s->pbctx, 1,  is_verbatim);                   // Audio block is verbatim
00125     put_bits32(&s->pbctx, s->avctx->frame_size);            // No. of samples in the frame
00126 }
00127 
00128 static void calc_predictor_params(AlacEncodeContext *s, int ch)
00129 {
00130     int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
00131     int shift[MAX_LPC_ORDER];
00132     int opt_order;
00133 
00134     if (s->compression_level == 1) {
00135         s->lpc[ch].lpc_order = 6;
00136         s->lpc[ch].lpc_quant = 6;
00137         s->lpc[ch].lpc_coeff[0] =  160;
00138         s->lpc[ch].lpc_coeff[1] = -190;
00139         s->lpc[ch].lpc_coeff[2] =  170;
00140         s->lpc[ch].lpc_coeff[3] = -130;
00141         s->lpc[ch].lpc_coeff[4] =   80;
00142         s->lpc[ch].lpc_coeff[5] =  -25;
00143     } else {
00144         opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, s->sample_buf[ch],
00145                                       s->avctx->frame_size,
00146                                       s->min_prediction_order,
00147                                       s->max_prediction_order,
00148                                       ALAC_MAX_LPC_PRECISION, coefs, shift,
00149                                       FF_LPC_TYPE_LEVINSON, 0,
00150                                       ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
00151 
00152         s->lpc[ch].lpc_order = opt_order;
00153         s->lpc[ch].lpc_quant = shift[opt_order-1];
00154         memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
00155     }
00156 }
00157 
00158 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
00159 {
00160     int i, best;
00161     int32_t lt, rt;
00162     uint64_t sum[4];
00163     uint64_t score[4];
00164 
00165     /* calculate sum of 2nd order residual for each channel */
00166     sum[0] = sum[1] = sum[2] = sum[3] = 0;
00167     for(i=2; i<n; i++) {
00168         lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
00169         rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
00170         sum[2] += FFABS((lt + rt) >> 1);
00171         sum[3] += FFABS(lt - rt);
00172         sum[0] += FFABS(lt);
00173         sum[1] += FFABS(rt);
00174     }
00175 
00176     /* calculate score for each mode */
00177     score[0] = sum[0] + sum[1];
00178     score[1] = sum[0] + sum[3];
00179     score[2] = sum[1] + sum[3];
00180     score[3] = sum[2] + sum[3];
00181 
00182     /* return mode with lowest score */
00183     best = 0;
00184     for(i=1; i<4; i++) {
00185         if(score[i] < score[best]) {
00186             best = i;
00187         }
00188     }
00189     return best;
00190 }
00191 
00192 static void alac_stereo_decorrelation(AlacEncodeContext *s)
00193 {
00194     int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
00195     int i, mode, n = s->avctx->frame_size;
00196     int32_t tmp;
00197 
00198     mode = estimate_stereo_mode(left, right, n);
00199 
00200     switch(mode)
00201     {
00202         case ALAC_CHMODE_LEFT_RIGHT:
00203             s->interlacing_leftweight = 0;
00204             s->interlacing_shift = 0;
00205             break;
00206 
00207         case ALAC_CHMODE_LEFT_SIDE:
00208             for(i=0; i<n; i++) {
00209                 right[i] = left[i] - right[i];
00210             }
00211             s->interlacing_leftweight = 1;
00212             s->interlacing_shift = 0;
00213             break;
00214 
00215         case ALAC_CHMODE_RIGHT_SIDE:
00216             for(i=0; i<n; i++) {
00217                 tmp = right[i];
00218                 right[i] = left[i] - right[i];
00219                 left[i] = tmp + (right[i] >> 31);
00220             }
00221             s->interlacing_leftweight = 1;
00222             s->interlacing_shift = 31;
00223             break;
00224 
00225         default:
00226             for(i=0; i<n; i++) {
00227                 tmp = left[i];
00228                 left[i] = (tmp + right[i]) >> 1;
00229                 right[i] = tmp - right[i];
00230             }
00231             s->interlacing_leftweight = 1;
00232             s->interlacing_shift = 1;
00233             break;
00234     }
00235 }
00236 
00237 static void alac_linear_predictor(AlacEncodeContext *s, int ch)
00238 {
00239     int i;
00240     AlacLPCContext lpc = s->lpc[ch];
00241 
00242     if(lpc.lpc_order == 31) {
00243         s->predictor_buf[0] = s->sample_buf[ch][0];
00244 
00245         for(i=1; i<s->avctx->frame_size; i++)
00246             s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1];
00247 
00248         return;
00249     }
00250 
00251     // generalised linear predictor
00252 
00253     if(lpc.lpc_order > 0) {
00254         int32_t *samples  = s->sample_buf[ch];
00255         int32_t *residual = s->predictor_buf;
00256 
00257         // generate warm-up samples
00258         residual[0] = samples[0];
00259         for(i=1;i<=lpc.lpc_order;i++)
00260             residual[i] = samples[i] - samples[i-1];
00261 
00262         // perform lpc on remaining samples
00263         for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
00264             int sum = 1 << (lpc.lpc_quant - 1), res_val, j;
00265 
00266             for (j = 0; j < lpc.lpc_order; j++) {
00267                 sum += (samples[lpc.lpc_order-j] - samples[0]) *
00268                         lpc.lpc_coeff[j];
00269             }
00270 
00271             sum >>= lpc.lpc_quant;
00272             sum += samples[0];
00273             residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum,
00274                                       s->write_sample_size);
00275             res_val = residual[i];
00276 
00277             if(res_val) {
00278                 int index = lpc.lpc_order - 1;
00279                 int neg = (res_val < 0);
00280 
00281                 while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) {
00282                     int val = samples[0] - samples[lpc.lpc_order - index];
00283                     int sign = (val ? FFSIGN(val) : 0);
00284 
00285                     if(neg)
00286                         sign*=-1;
00287 
00288                     lpc.lpc_coeff[index] -= sign;
00289                     val *= sign;
00290                     res_val -= ((val >> lpc.lpc_quant) *
00291                             (lpc.lpc_order - index));
00292                     index--;
00293                 }
00294             }
00295             samples++;
00296         }
00297     }
00298 }
00299 
00300 static void alac_entropy_coder(AlacEncodeContext *s)
00301 {
00302     unsigned int history = s->rc.initial_history;
00303     int sign_modifier = 0, i, k;
00304     int32_t *samples = s->predictor_buf;
00305 
00306     for(i=0;i < s->avctx->frame_size;) {
00307         int x;
00308 
00309         k = av_log2((history >> 9) + 3);
00310 
00311         x = -2*(*samples)-1;
00312         x ^= (x>>31);
00313 
00314         samples++;
00315         i++;
00316 
00317         encode_scalar(s, x - sign_modifier, k, s->write_sample_size);
00318 
00319         history += x * s->rc.history_mult
00320                    - ((history * s->rc.history_mult) >> 9);
00321 
00322         sign_modifier = 0;
00323         if(x > 0xFFFF)
00324             history = 0xFFFF;
00325 
00326         if((history < 128) && (i < s->avctx->frame_size)) {
00327             unsigned int block_size = 0;
00328 
00329             k = 7 - av_log2(history) + ((history + 16) >> 6);
00330 
00331             while((*samples == 0) && (i < s->avctx->frame_size)) {
00332                 samples++;
00333                 i++;
00334                 block_size++;
00335             }
00336             encode_scalar(s, block_size, k, 16);
00337 
00338             sign_modifier = (block_size <= 0xFFFF);
00339 
00340             history = 0;
00341         }
00342 
00343     }
00344 }
00345 
00346 static void write_compressed_frame(AlacEncodeContext *s)
00347 {
00348     int i, j;
00349 
00350     if(s->avctx->channels == 2)
00351         alac_stereo_decorrelation(s);
00352     put_bits(&s->pbctx, 8, s->interlacing_shift);
00353     put_bits(&s->pbctx, 8, s->interlacing_leftweight);
00354 
00355     for(i=0;i<s->avctx->channels;i++) {
00356 
00357         calc_predictor_params(s, i);
00358 
00359         put_bits(&s->pbctx, 4, 0);  // prediction type : currently only type 0 has been RE'd
00360         put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant);
00361 
00362         put_bits(&s->pbctx, 3, s->rc.rice_modifier);
00363         put_bits(&s->pbctx, 5, s->lpc[i].lpc_order);
00364         // predictor coeff. table
00365         for(j=0;j<s->lpc[i].lpc_order;j++) {
00366             put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]);
00367         }
00368     }
00369 
00370     // apply lpc and entropy coding to audio samples
00371 
00372     for(i=0;i<s->avctx->channels;i++) {
00373         alac_linear_predictor(s, i);
00374         alac_entropy_coder(s);
00375     }
00376 }
00377 
00378 static av_cold int alac_encode_init(AVCodecContext *avctx)
00379 {
00380     AlacEncodeContext *s    = avctx->priv_data;
00381     int ret;
00382     uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1);
00383 
00384     avctx->frame_size      = DEFAULT_FRAME_SIZE;
00385     avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE;
00386 
00387     if(avctx->sample_fmt != AV_SAMPLE_FMT_S16) {
00388         av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
00389         return -1;
00390     }
00391 
00392     // Set default compression level
00393     if(avctx->compression_level == FF_COMPRESSION_DEFAULT)
00394         s->compression_level = 2;
00395     else
00396         s->compression_level = av_clip(avctx->compression_level, 0, 2);
00397 
00398     // Initialize default Rice parameters
00399     s->rc.history_mult    = 40;
00400     s->rc.initial_history = 10;
00401     s->rc.k_modifier      = 14;
00402     s->rc.rice_modifier   = 4;
00403 
00404     s->max_coded_frame_size = 8 + (avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample>>3);
00405 
00406     s->write_sample_size  = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes
00407 
00408     AV_WB32(alac_extradata,    ALAC_EXTRADATA_SIZE);
00409     AV_WB32(alac_extradata+4,  MKBETAG('a','l','a','c'));
00410     AV_WB32(alac_extradata+12, avctx->frame_size);
00411     AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample);
00412     AV_WB8 (alac_extradata+21, avctx->channels);
00413     AV_WB32(alac_extradata+24, s->max_coded_frame_size);
00414     AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_coded_sample); // average bitrate
00415     AV_WB32(alac_extradata+32, avctx->sample_rate);
00416 
00417     // Set relevant extradata fields
00418     if(s->compression_level > 0) {
00419         AV_WB8(alac_extradata+18, s->rc.history_mult);
00420         AV_WB8(alac_extradata+19, s->rc.initial_history);
00421         AV_WB8(alac_extradata+20, s->rc.k_modifier);
00422     }
00423 
00424     s->min_prediction_order = DEFAULT_MIN_PRED_ORDER;
00425     if(avctx->min_prediction_order >= 0) {
00426         if(avctx->min_prediction_order < MIN_LPC_ORDER ||
00427            avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) {
00428             av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
00429                 return -1;
00430         }
00431 
00432         s->min_prediction_order = avctx->min_prediction_order;
00433     }
00434 
00435     s->max_prediction_order = DEFAULT_MAX_PRED_ORDER;
00436     if(avctx->max_prediction_order >= 0) {
00437         if(avctx->max_prediction_order < MIN_LPC_ORDER ||
00438            avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) {
00439             av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
00440                 return -1;
00441         }
00442 
00443         s->max_prediction_order = avctx->max_prediction_order;
00444     }
00445 
00446     if(s->max_prediction_order < s->min_prediction_order) {
00447         av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
00448                s->min_prediction_order, s->max_prediction_order);
00449         return -1;
00450     }
00451 
00452     avctx->extradata = alac_extradata;
00453     avctx->extradata_size = ALAC_EXTRADATA_SIZE;
00454 
00455     avctx->coded_frame = avcodec_alloc_frame();
00456     avctx->coded_frame->key_frame = 1;
00457 
00458     s->avctx = avctx;
00459     ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size, s->max_prediction_order,
00460                       FF_LPC_TYPE_LEVINSON);
00461 
00462     return ret;
00463 }
00464 
00465 static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
00466                              int buf_size, void *data)
00467 {
00468     AlacEncodeContext *s = avctx->priv_data;
00469     PutBitContext *pb = &s->pbctx;
00470     int i, out_bytes, verbatim_flag = 0;
00471 
00472     if(avctx->frame_size > DEFAULT_FRAME_SIZE) {
00473         av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n");
00474         return -1;
00475     }
00476 
00477     if(buf_size < 2*s->max_coded_frame_size) {
00478         av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n");
00479         return -1;
00480     }
00481 
00482 verbatim:
00483     init_put_bits(pb, frame, buf_size);
00484 
00485     if((s->compression_level == 0) || verbatim_flag) {
00486         // Verbatim mode
00487         const int16_t *samples = data;
00488         write_frame_header(s, 1);
00489         for(i=0; i<avctx->frame_size*avctx->channels; i++) {
00490             put_sbits(pb, 16, *samples++);
00491         }
00492     } else {
00493         init_sample_buffers(s, data);
00494         write_frame_header(s, 0);
00495         write_compressed_frame(s);
00496     }
00497 
00498     put_bits(pb, 3, 7);
00499     flush_put_bits(pb);
00500     out_bytes = put_bits_count(pb) >> 3;
00501 
00502     if(out_bytes > s->max_coded_frame_size) {
00503         /* frame too large. use verbatim mode */
00504         if(verbatim_flag || (s->compression_level == 0)) {
00505             /* still too large. must be an error. */
00506             av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
00507             return -1;
00508         }
00509         verbatim_flag = 1;
00510         goto verbatim;
00511     }
00512 
00513     return out_bytes;
00514 }
00515 
00516 static av_cold int alac_encode_close(AVCodecContext *avctx)
00517 {
00518     AlacEncodeContext *s = avctx->priv_data;
00519     ff_lpc_end(&s->lpc_ctx);
00520     av_freep(&avctx->extradata);
00521     avctx->extradata_size = 0;
00522     av_freep(&avctx->coded_frame);
00523     return 0;
00524 }
00525 
00526 AVCodec ff_alac_encoder = {
00527     "alac",
00528     AVMEDIA_TYPE_AUDIO,
00529     CODEC_ID_ALAC,
00530     sizeof(AlacEncodeContext),
00531     alac_encode_init,
00532     alac_encode_frame,
00533     alac_encode_close,
00534     .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
00535     .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE},
00536     .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
00537 };