yading@10: /*
yading@10:  * Copyright (c) 2001-2003 The ffmpeg Project
yading@10:  *
yading@10:  * first version by Francois Revol (revol@free.fr)
yading@10:  * fringe ADPCM codecs (e.g., DK3, DK4, Westwood)
yading@10:  *   by Mike Melanson (melanson@pcisys.net)
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "avcodec.h"
yading@10: #include "put_bits.h"
yading@10: #include "bytestream.h"
yading@10: #include "adpcm.h"
yading@10: #include "adpcm_data.h"
yading@10: #include "internal.h"
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * ADPCM encoders
yading@10:  * See ADPCM decoder reference documents for codec information.
yading@10:  */
yading@10: 
yading@10: typedef struct TrellisPath {
yading@10:     int nibble;
yading@10:     int prev;
yading@10: } TrellisPath;
yading@10: 
yading@10: typedef struct TrellisNode {
yading@10:     uint32_t ssd;
yading@10:     int path;
yading@10:     int sample1;
yading@10:     int sample2;
yading@10:     int step;
yading@10: } TrellisNode;
yading@10: 
yading@10: typedef struct ADPCMEncodeContext {
yading@10:     ADPCMChannelStatus status[6];
yading@10:     TrellisPath *paths;
yading@10:     TrellisNode *node_buf;
yading@10:     TrellisNode **nodep_buf;
yading@10:     uint8_t *trellis_hash;
yading@10: } ADPCMEncodeContext;
yading@10: 
yading@10: #define FREEZE_INTERVAL 128
yading@10: 
yading@10: static av_cold int adpcm_encode_close(AVCodecContext *avctx);
yading@10: 
yading@10: static av_cold int adpcm_encode_init(AVCodecContext *avctx)
yading@10: {
yading@10:     ADPCMEncodeContext *s = avctx->priv_data;
yading@10:     uint8_t *extradata;
yading@10:     int i;
yading@10:     int ret = AVERROR(ENOMEM);
yading@10: 
yading@10:     if (avctx->channels > 2) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "only stereo or mono is supported\n");
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10: 
yading@10:     if (avctx->trellis && (unsigned)avctx->trellis > 16U) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "invalid trellis size\n");
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10: 
yading@10:     if (avctx->trellis) {
yading@10:         int frontier  = 1 << avctx->trellis;
yading@10:         int max_paths =  frontier * FREEZE_INTERVAL;
yading@10:         FF_ALLOC_OR_GOTO(avctx, s->paths,
yading@10:                          max_paths * sizeof(*s->paths), error);
yading@10:         FF_ALLOC_OR_GOTO(avctx, s->node_buf,
yading@10:                          2 * frontier * sizeof(*s->node_buf),  error);
yading@10:         FF_ALLOC_OR_GOTO(avctx, s->nodep_buf,
yading@10:                          2 * frontier * sizeof(*s->nodep_buf), error);
yading@10:         FF_ALLOC_OR_GOTO(avctx, s->trellis_hash,
yading@10:                          65536 * sizeof(*s->trellis_hash), error);
yading@10:     }
yading@10: 
yading@10:     avctx->bits_per_coded_sample = av_get_bits_per_sample(avctx->codec->id);
yading@10: 
yading@10:     switch (avctx->codec->id) {
yading@10:     case AV_CODEC_ID_ADPCM_IMA_WAV:
yading@10:         /* each 16 bits sample gives one nibble
yading@10:            and we have 4 bytes per channel overhead */
yading@10:         avctx->frame_size = (BLKSIZE - 4 * avctx->channels) * 8 /
yading@10:                             (4 * avctx->channels) + 1;
yading@10:         /* seems frame_size isn't taken into account...
yading@10:            have to buffer the samples :-( */
yading@10:         avctx->block_align = BLKSIZE;
yading@10:         avctx->bits_per_coded_sample = 4;
yading@10:         break;
yading@10:     case AV_CODEC_ID_ADPCM_IMA_QT:
yading@10:         avctx->frame_size  = 64;
yading@10:         avctx->block_align = 34 * avctx->channels;
yading@10:         break;
yading@10:     case AV_CODEC_ID_ADPCM_MS:
yading@10:         /* each 16 bits sample gives one nibble
yading@10:            and we have 7 bytes per channel overhead */
yading@10:         avctx->frame_size = (BLKSIZE - 7 * avctx->channels) * 2 / avctx->channels + 2;
yading@10:         avctx->bits_per_coded_sample = 4;
yading@10:         avctx->block_align    = BLKSIZE;
yading@10:         if (!(avctx->extradata = av_malloc(32 + FF_INPUT_BUFFER_PADDING_SIZE)))
yading@10:             goto error;
yading@10:         avctx->extradata_size = 32;
yading@10:         extradata = avctx->extradata;
yading@10:         bytestream_put_le16(&extradata, avctx->frame_size);
yading@10:         bytestream_put_le16(&extradata, 7); /* wNumCoef */
yading@10:         for (i = 0; i < 7; i++) {
yading@10:             bytestream_put_le16(&extradata, ff_adpcm_AdaptCoeff1[i] * 4);
yading@10:             bytestream_put_le16(&extradata, ff_adpcm_AdaptCoeff2[i] * 4);
yading@10:         }
yading@10:         break;
yading@10:     case AV_CODEC_ID_ADPCM_YAMAHA:
yading@10:         avctx->frame_size  = BLKSIZE * 2 / avctx->channels;
yading@10:         avctx->block_align = BLKSIZE;
yading@10:         break;
yading@10:     case AV_CODEC_ID_ADPCM_SWF:
yading@10:         if (avctx->sample_rate != 11025 &&
yading@10:             avctx->sample_rate != 22050 &&
yading@10:             avctx->sample_rate != 44100) {
yading@10:             av_log(avctx, AV_LOG_ERROR, "Sample rate must be 11025, "
yading@10:                    "22050 or 44100\n");
yading@10:             ret = AVERROR(EINVAL);
yading@10:             goto error;
yading@10:         }
yading@10:         avctx->frame_size = 512 * (avctx->sample_rate / 11025);
yading@10:         break;
yading@10:     default:
yading@10:         ret = AVERROR(EINVAL);
yading@10:         goto error;
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: error:
yading@10:     adpcm_encode_close(avctx);
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: static av_cold int adpcm_encode_close(AVCodecContext *avctx)
yading@10: {
yading@10:     ADPCMEncodeContext *s = avctx->priv_data;
yading@10:     av_freep(&s->paths);
yading@10:     av_freep(&s->node_buf);
yading@10:     av_freep(&s->nodep_buf);
yading@10:     av_freep(&s->trellis_hash);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: 
yading@10: static inline uint8_t adpcm_ima_compress_sample(ADPCMChannelStatus *c,
yading@10:                                                 int16_t sample)
yading@10: {
yading@10:     int delta  = sample - c->prev_sample;
yading@10:     int nibble = FFMIN(7, abs(delta) * 4 /
yading@10:                        ff_adpcm_step_table[c->step_index]) + (delta < 0) * 8;
yading@10:     c->prev_sample += ((ff_adpcm_step_table[c->step_index] *
yading@10:                         ff_adpcm_yamaha_difflookup[nibble]) / 8);
yading@10:     c->prev_sample = av_clip_int16(c->prev_sample);
yading@10:     c->step_index  = av_clip(c->step_index + ff_adpcm_index_table[nibble], 0, 88);
yading@10:     return nibble;
yading@10: }
yading@10: 
yading@10: static inline uint8_t adpcm_ima_qt_compress_sample(ADPCMChannelStatus *c,
yading@10:                                                    int16_t sample)
yading@10: {
yading@10:     int delta  = sample - c->prev_sample;
yading@10:     int diff, step = ff_adpcm_step_table[c->step_index];
yading@10:     int nibble = 8*(delta < 0);
yading@10: 
yading@10:     delta= abs(delta);
yading@10:     diff = delta + (step >> 3);
yading@10: 
yading@10:     if (delta >= step) {
yading@10:         nibble |= 4;
yading@10:         delta  -= step;
yading@10:     }
yading@10:     step >>= 1;
yading@10:     if (delta >= step) {
yading@10:         nibble |= 2;
yading@10:         delta  -= step;
yading@10:     }
yading@10:     step >>= 1;
yading@10:     if (delta >= step) {
yading@10:         nibble |= 1;
yading@10:         delta  -= step;
yading@10:     }
yading@10:     diff -= delta;
yading@10: 
yading@10:     if (nibble & 8)
yading@10:         c->prev_sample -= diff;
yading@10:     else
yading@10:         c->prev_sample += diff;
yading@10: 
yading@10:     c->prev_sample = av_clip_int16(c->prev_sample);
yading@10:     c->step_index  = av_clip(c->step_index + ff_adpcm_index_table[nibble], 0, 88);
yading@10: 
yading@10:     return nibble;
yading@10: }
yading@10: 
yading@10: static inline uint8_t adpcm_ms_compress_sample(ADPCMChannelStatus *c,
yading@10:                                                int16_t sample)
yading@10: {
yading@10:     int predictor, nibble, bias;
yading@10: 
yading@10:     predictor = (((c->sample1) * (c->coeff1)) +
yading@10:                 (( c->sample2) * (c->coeff2))) / 64;
yading@10: 
yading@10:     nibble = sample - predictor;
yading@10:     if (nibble >= 0)
yading@10:         bias =  c->idelta / 2;
yading@10:     else
yading@10:         bias = -c->idelta / 2;
yading@10: 
yading@10:     nibble = (nibble + bias) / c->idelta;
yading@10:     nibble = av_clip(nibble, -8, 7) & 0x0F;
yading@10: 
yading@10:     predictor += ((nibble & 0x08) ? (nibble - 0x10) : nibble) * c->idelta;
yading@10: 
yading@10:     c->sample2 = c->sample1;
yading@10:     c->sample1 = av_clip_int16(predictor);
yading@10: 
yading@10:     c->idelta = (ff_adpcm_AdaptationTable[nibble] * c->idelta) >> 8;
yading@10:     if (c->idelta < 16)
yading@10:         c->idelta = 16;
yading@10: 
yading@10:     return nibble;
yading@10: }
yading@10: 
yading@10: static inline uint8_t adpcm_yamaha_compress_sample(ADPCMChannelStatus *c,
yading@10:                                                    int16_t sample)
yading@10: {
yading@10:     int nibble, delta;
yading@10: 
yading@10:     if (!c->step) {
yading@10:         c->predictor = 0;
yading@10:         c->step      = 127;
yading@10:     }
yading@10: 
yading@10:     delta = sample - c->predictor;
yading@10: 
yading@10:     nibble = FFMIN(7, abs(delta) * 4 / c->step) + (delta < 0) * 8;
yading@10: 
yading@10:     c->predictor += ((c->step * ff_adpcm_yamaha_difflookup[nibble]) / 8);
yading@10:     c->predictor = av_clip_int16(c->predictor);
yading@10:     c->step = (c->step * ff_adpcm_yamaha_indexscale[nibble]) >> 8;
yading@10:     c->step = av_clip(c->step, 127, 24567);
yading@10: 
yading@10:     return nibble;
yading@10: }
yading@10: 
yading@10: static void adpcm_compress_trellis(AVCodecContext *avctx,
yading@10:                                    const int16_t *samples, uint8_t *dst,
yading@10:                                    ADPCMChannelStatus *c, int n, int stride)
yading@10: {
yading@10:     //FIXME 6% faster if frontier is a compile-time constant
yading@10:     ADPCMEncodeContext *s = avctx->priv_data;
yading@10:     const int frontier = 1 << avctx->trellis;
yading@10:     const int version  = avctx->codec->id;
yading@10:     TrellisPath *paths       = s->paths, *p;
yading@10:     TrellisNode *node_buf    = s->node_buf;
yading@10:     TrellisNode **nodep_buf  = s->nodep_buf;
yading@10:     TrellisNode **nodes      = nodep_buf; // nodes[] is always sorted by .ssd
yading@10:     TrellisNode **nodes_next = nodep_buf + frontier;
yading@10:     int pathn = 0, froze = -1, i, j, k, generation = 0;
yading@10:     uint8_t *hash = s->trellis_hash;
yading@10:     memset(hash, 0xff, 65536 * sizeof(*hash));
yading@10: 
yading@10:     memset(nodep_buf, 0, 2 * frontier * sizeof(*nodep_buf));
yading@10:     nodes[0]          = node_buf + frontier;
yading@10:     nodes[0]->ssd     = 0;
yading@10:     nodes[0]->path    = 0;
yading@10:     nodes[0]->step    = c->step_index;
yading@10:     nodes[0]->sample1 = c->sample1;
yading@10:     nodes[0]->sample2 = c->sample2;
yading@10:     if (version == AV_CODEC_ID_ADPCM_IMA_WAV ||
yading@10:         version == AV_CODEC_ID_ADPCM_IMA_QT  ||
yading@10:         version == AV_CODEC_ID_ADPCM_SWF)
yading@10:         nodes[0]->sample1 = c->prev_sample;
yading@10:     if (version == AV_CODEC_ID_ADPCM_MS)
yading@10:         nodes[0]->step = c->idelta;
yading@10:     if (version == AV_CODEC_ID_ADPCM_YAMAHA) {
yading@10:         if (c->step == 0) {
yading@10:             nodes[0]->step    = 127;
yading@10:             nodes[0]->sample1 = 0;
yading@10:         } else {
yading@10:             nodes[0]->step    = c->step;
yading@10:             nodes[0]->sample1 = c->predictor;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     for (i = 0; i < n; i++) {
yading@10:         TrellisNode *t = node_buf + frontier*(i&1);
yading@10:         TrellisNode **u;
yading@10:         int sample   = samples[i * stride];
yading@10:         int heap_pos = 0;
yading@10:         memset(nodes_next, 0, frontier * sizeof(TrellisNode*));
yading@10:         for (j = 0; j < frontier && nodes[j]; j++) {
yading@10:             // higher j have higher ssd already, so they're likely
yading@10:             // to yield a suboptimal next sample too
yading@10:             const int range = (j < frontier / 2) ? 1 : 0;
yading@10:             const int step  = nodes[j]->step;
yading@10:             int nidx;
yading@10:             if (version == AV_CODEC_ID_ADPCM_MS) {
yading@10:                 const int predictor = ((nodes[j]->sample1 * c->coeff1) +
yading@10:                                        (nodes[j]->sample2 * c->coeff2)) / 64;
yading@10:                 const int div  = (sample - predictor) / step;
yading@10:                 const int nmin = av_clip(div-range, -8, 6);
yading@10:                 const int nmax = av_clip(div+range, -7, 7);
yading@10:                 for (nidx = nmin; nidx <= nmax; nidx++) {
yading@10:                     const int nibble = nidx & 0xf;
yading@10:                     int dec_sample   = predictor + nidx * step;
yading@10: #define STORE_NODE(NAME, STEP_INDEX)\
yading@10:                     int d;\
yading@10:                     uint32_t ssd;\
yading@10:                     int pos;\
yading@10:                     TrellisNode *u;\
yading@10:                     uint8_t *h;\
yading@10:                     dec_sample = av_clip_int16(dec_sample);\
yading@10:                     d = sample - dec_sample;\
yading@10:                     ssd = nodes[j]->ssd + d*d;\
yading@10:                     /* Check for wraparound, skip such samples completely. \
yading@10:                      * Note, changing ssd to a 64 bit variable would be \
yading@10:                      * simpler, avoiding this check, but it's slower on \
yading@10:                      * x86 32 bit at the moment. */\
yading@10:                     if (ssd < nodes[j]->ssd)\
yading@10:                         goto next_##NAME;\
yading@10:                     /* Collapse any two states with the same previous sample value. \
yading@10:                      * One could also distinguish states by step and by 2nd to last
yading@10:                      * sample, but the effects of that are negligible.
yading@10:                      * Since nodes in the previous generation are iterated
yading@10:                      * through a heap, they're roughly ordered from better to
yading@10:                      * worse, but not strictly ordered. Therefore, an earlier
yading@10:                      * node with the same sample value is better in most cases
yading@10:                      * (and thus the current is skipped), but not strictly
yading@10:                      * in all cases. Only skipping samples where ssd >=
yading@10:                      * ssd of the earlier node with the same sample gives
yading@10:                      * slightly worse quality, though, for some reason. */ \
yading@10:                     h = &hash[(uint16_t) dec_sample];\
yading@10:                     if (*h == generation)\
yading@10:                         goto next_##NAME;\
yading@10:                     if (heap_pos < frontier) {\
yading@10:                         pos = heap_pos++;\
yading@10:                     } else {\
yading@10:                         /* Try to replace one of the leaf nodes with the new \
yading@10:                          * one, but try a different slot each time. */\
yading@10:                         pos = (frontier >> 1) +\
yading@10:                               (heap_pos & ((frontier >> 1) - 1));\
yading@10:                         if (ssd > nodes_next[pos]->ssd)\
yading@10:                             goto next_##NAME;\
yading@10:                         heap_pos++;\
yading@10:                     }\
yading@10:                     *h = generation;\
yading@10:                     u  = nodes_next[pos];\
yading@10:                     if (!u) {\
yading@10:                         av_assert1(pathn < FREEZE_INTERVAL << avctx->trellis);\
yading@10:                         u = t++;\
yading@10:                         nodes_next[pos] = u;\
yading@10:                         u->path = pathn++;\
yading@10:                     }\
yading@10:                     u->ssd  = ssd;\
yading@10:                     u->step = STEP_INDEX;\
yading@10:                     u->sample2 = nodes[j]->sample1;\
yading@10:                     u->sample1 = dec_sample;\
yading@10:                     paths[u->path].nibble = nibble;\
yading@10:                     paths[u->path].prev   = nodes[j]->path;\
yading@10:                     /* Sift the newly inserted node up in the heap to \
yading@10:                      * restore the heap property. */\
yading@10:                     while (pos > 0) {\
yading@10:                         int parent = (pos - 1) >> 1;\
yading@10:                         if (nodes_next[parent]->ssd <= ssd)\
yading@10:                             break;\
yading@10:                         FFSWAP(TrellisNode*, nodes_next[parent], nodes_next[pos]);\
yading@10:                         pos = parent;\
yading@10:                     }\
yading@10:                     next_##NAME:;
yading@10:                     STORE_NODE(ms, FFMAX(16,
yading@10:                                (ff_adpcm_AdaptationTable[nibble] * step) >> 8));
yading@10:                 }
yading@10:             } else if (version == AV_CODEC_ID_ADPCM_IMA_WAV ||
yading@10:                        version == AV_CODEC_ID_ADPCM_IMA_QT  ||
yading@10:                        version == AV_CODEC_ID_ADPCM_SWF) {
yading@10: #define LOOP_NODES(NAME, STEP_TABLE, STEP_INDEX)\
yading@10:                 const int predictor = nodes[j]->sample1;\
yading@10:                 const int div = (sample - predictor) * 4 / STEP_TABLE;\
yading@10:                 int nmin = av_clip(div - range, -7, 6);\
yading@10:                 int nmax = av_clip(div + range, -6, 7);\
yading@10:                 if (nmin <= 0)\
yading@10:                     nmin--; /* distinguish -0 from +0 */\
yading@10:                 if (nmax < 0)\
yading@10:                     nmax--;\
yading@10:                 for (nidx = nmin; nidx <= nmax; nidx++) {\
yading@10:                     const int nibble = nidx < 0 ? 7 - nidx : nidx;\
yading@10:                     int dec_sample = predictor +\
yading@10:                                     (STEP_TABLE *\
yading@10:                                      ff_adpcm_yamaha_difflookup[nibble]) / 8;\
yading@10:                     STORE_NODE(NAME, STEP_INDEX);\
yading@10:                 }
yading@10:                 LOOP_NODES(ima, ff_adpcm_step_table[step],
yading@10:                            av_clip(step + ff_adpcm_index_table[nibble], 0, 88));
yading@10:             } else { //AV_CODEC_ID_ADPCM_YAMAHA
yading@10:                 LOOP_NODES(yamaha, step,
yading@10:                            av_clip((step * ff_adpcm_yamaha_indexscale[nibble]) >> 8,
yading@10:                                    127, 24567));
yading@10: #undef LOOP_NODES
yading@10: #undef STORE_NODE
yading@10:             }
yading@10:         }
yading@10: 
yading@10:         u = nodes;
yading@10:         nodes = nodes_next;
yading@10:         nodes_next = u;
yading@10: 
yading@10:         generation++;
yading@10:         if (generation == 255) {
yading@10:             memset(hash, 0xff, 65536 * sizeof(*hash));
yading@10:             generation = 0;
yading@10:         }
yading@10: 
yading@10:         // prevent overflow
yading@10:         if (nodes[0]->ssd > (1 << 28)) {
yading@10:             for (j = 1; j < frontier && nodes[j]; j++)
yading@10:                 nodes[j]->ssd -= nodes[0]->ssd;
yading@10:             nodes[0]->ssd = 0;
yading@10:         }
yading@10: 
yading@10:         // merge old paths to save memory
yading@10:         if (i == froze + FREEZE_INTERVAL) {
yading@10:             p = &paths[nodes[0]->path];
yading@10:             for (k = i; k > froze; k--) {
yading@10:                 dst[k] = p->nibble;
yading@10:                 p = &paths[p->prev];
yading@10:             }
yading@10:             froze = i;
yading@10:             pathn = 0;
yading@10:             // other nodes might use paths that don't coincide with the frozen one.
yading@10:             // checking which nodes do so is too slow, so just kill them all.
yading@10:             // this also slightly improves quality, but I don't know why.
yading@10:             memset(nodes + 1, 0, (frontier - 1) * sizeof(TrellisNode*));
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     p = &paths[nodes[0]->path];
yading@10:     for (i = n - 1; i > froze; i--) {
yading@10:         dst[i] = p->nibble;
yading@10:         p = &paths[p->prev];
yading@10:     }
yading@10: 
yading@10:     c->predictor  = nodes[0]->sample1;
yading@10:     c->sample1    = nodes[0]->sample1;
yading@10:     c->sample2    = nodes[0]->sample2;
yading@10:     c->step_index = nodes[0]->step;
yading@10:     c->step       = nodes[0]->step;
yading@10:     c->idelta     = nodes[0]->step;
yading@10: }
yading@10: 
yading@10: static int adpcm_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
yading@10:                               const AVFrame *frame, int *got_packet_ptr)
yading@10: {
yading@10:     int n, i, ch, st, pkt_size, ret;
yading@10:     const int16_t *samples;
yading@10:     int16_t **samples_p;
yading@10:     uint8_t *dst;
yading@10:     ADPCMEncodeContext *c = avctx->priv_data;
yading@10:     uint8_t *buf;
yading@10: 
yading@10:     samples = (const int16_t *)frame->data[0];
yading@10:     samples_p = (int16_t **)frame->extended_data;
yading@10:     st = avctx->channels == 2;
yading@10: 
yading@10:     if (avctx->codec_id == AV_CODEC_ID_ADPCM_SWF)
yading@10:         pkt_size = (2 + avctx->channels * (22 + 4 * (frame->nb_samples - 1)) + 7) / 8;
yading@10:     else
yading@10:         pkt_size = avctx->block_align;
yading@10:     if ((ret = ff_alloc_packet2(avctx, avpkt, pkt_size)) < 0)
yading@10:         return ret;
yading@10:     dst = avpkt->data;
yading@10: 
yading@10:     switch(avctx->codec->id) {
yading@10:     case AV_CODEC_ID_ADPCM_IMA_WAV:
yading@10:     {
yading@10:         int blocks, j;
yading@10: 
yading@10:         blocks = (frame->nb_samples - 1) / 8;
yading@10: 
yading@10:         for (ch = 0; ch < avctx->channels; ch++) {
yading@10:             ADPCMChannelStatus *status = &c->status[ch];
yading@10:             status->prev_sample = samples_p[ch][0];
yading@10:             /* status->step_index = 0;
yading@10:                XXX: not sure how to init the state machine */
yading@10:             bytestream_put_le16(&dst, status->prev_sample);
yading@10:             *dst++ = status->step_index;
yading@10:             *dst++ = 0; /* unknown */
yading@10:         }
yading@10: 
yading@10:         /* stereo: 4 bytes (8 samples) for left, 4 bytes for right */
yading@10:         if (avctx->trellis > 0) {
yading@10:             FF_ALLOC_OR_GOTO(avctx, buf, avctx->channels * blocks * 8, error);
yading@10:             for (ch = 0; ch < avctx->channels; ch++) {
yading@10:                 adpcm_compress_trellis(avctx, &samples_p[ch][1],
yading@10:                                        buf + ch * blocks * 8, &c->status[ch],
yading@10:                                        blocks * 8, 1);
yading@10:             }
yading@10:             for (i = 0; i < blocks; i++) {
yading@10:                 for (ch = 0; ch < avctx->channels; ch++) {
yading@10:                     uint8_t *buf1 = buf + ch * blocks * 8 + i * 8;
yading@10:                     for (j = 0; j < 8; j += 2)
yading@10:                         *dst++ = buf1[j] | (buf1[j + 1] << 4);
yading@10:                 }
yading@10:             }
yading@10:             av_free(buf);
yading@10:         } else {
yading@10:             for (i = 0; i < blocks; i++) {
yading@10:                 for (ch = 0; ch < avctx->channels; ch++) {
yading@10:                     ADPCMChannelStatus *status = &c->status[ch];
yading@10:                     const int16_t *smp = &samples_p[ch][1 + i * 8];
yading@10:                     for (j = 0; j < 8; j += 2) {
yading@10:                         uint8_t v = adpcm_ima_compress_sample(status, smp[j    ]);
yading@10:                         v        |= adpcm_ima_compress_sample(status, smp[j + 1]) << 4;
yading@10:                         *dst++ = v;
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         break;
yading@10:     }
yading@10:     case AV_CODEC_ID_ADPCM_IMA_QT:
yading@10:     {
yading@10:         PutBitContext pb;
yading@10:         init_put_bits(&pb, dst, pkt_size * 8);
yading@10: 
yading@10:         for (ch = 0; ch < avctx->channels; ch++) {
yading@10:             ADPCMChannelStatus *status = &c->status[ch];
yading@10:             put_bits(&pb, 9, (status->prev_sample & 0xFFFF) >> 7);
yading@10:             put_bits(&pb, 7,  status->step_index);
yading@10:             if (avctx->trellis > 0) {
yading@10:                 uint8_t buf[64];
yading@10:                 adpcm_compress_trellis(avctx, &samples_p[ch][1], buf, status,
yading@10:                                        64, 1);
yading@10:                 for (i = 0; i < 64; i++)
yading@10:                     put_bits(&pb, 4, buf[i ^ 1]);
yading@10:             } else {
yading@10:                 for (i = 0; i < 64; i += 2) {
yading@10:                     int t1, t2;
yading@10:                     t1 = adpcm_ima_qt_compress_sample(status, samples_p[ch][i    ]);
yading@10:                     t2 = adpcm_ima_qt_compress_sample(status, samples_p[ch][i + 1]);
yading@10:                     put_bits(&pb, 4, t2);
yading@10:                     put_bits(&pb, 4, t1);
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10: 
yading@10:         flush_put_bits(&pb);
yading@10:         break;
yading@10:     }
yading@10:     case AV_CODEC_ID_ADPCM_SWF:
yading@10:     {
yading@10:         PutBitContext pb;
yading@10:         init_put_bits(&pb, dst, pkt_size * 8);
yading@10: 
yading@10:         n = frame->nb_samples - 1;
yading@10: 
yading@10:         // store AdpcmCodeSize
yading@10:         put_bits(&pb, 2, 2);    // set 4-bit flash adpcm format
yading@10: 
yading@10:         // init the encoder state
yading@10:         for (i = 0; i < avctx->channels; i++) {
yading@10:             // clip step so it fits 6 bits
yading@10:             c->status[i].step_index = av_clip(c->status[i].step_index, 0, 63);
yading@10:             put_sbits(&pb, 16, samples[i]);
yading@10:             put_bits(&pb, 6, c->status[i].step_index);
yading@10:             c->status[i].prev_sample = samples[i];
yading@10:         }
yading@10: 
yading@10:         if (avctx->trellis > 0) {
yading@10:             FF_ALLOC_OR_GOTO(avctx, buf, 2 * n, error);
yading@10:             adpcm_compress_trellis(avctx, samples + avctx->channels, buf,
yading@10:                                    &c->status[0], n, avctx->channels);
yading@10:             if (avctx->channels == 2)
yading@10:                 adpcm_compress_trellis(avctx, samples + avctx->channels + 1,
yading@10:                                        buf + n, &c->status[1], n,
yading@10:                                        avctx->channels);
yading@10:             for (i = 0; i < n; i++) {
yading@10:                 put_bits(&pb, 4, buf[i]);
yading@10:                 if (avctx->channels == 2)
yading@10:                     put_bits(&pb, 4, buf[n + i]);
yading@10:             }
yading@10:             av_free(buf);
yading@10:         } else {
yading@10:             for (i = 1; i < frame->nb_samples; i++) {
yading@10:                 put_bits(&pb, 4, adpcm_ima_compress_sample(&c->status[0],
yading@10:                          samples[avctx->channels * i]));
yading@10:                 if (avctx->channels == 2)
yading@10:                     put_bits(&pb, 4, adpcm_ima_compress_sample(&c->status[1],
yading@10:                              samples[2 * i + 1]));
yading@10:             }
yading@10:         }
yading@10:         flush_put_bits(&pb);
yading@10:         break;
yading@10:     }
yading@10:     case AV_CODEC_ID_ADPCM_MS:
yading@10:         for (i = 0; i < avctx->channels; i++) {
yading@10:             int predictor = 0;
yading@10:             *dst++ = predictor;
yading@10:             c->status[i].coeff1 = ff_adpcm_AdaptCoeff1[predictor];
yading@10:             c->status[i].coeff2 = ff_adpcm_AdaptCoeff2[predictor];
yading@10:         }
yading@10:         for (i = 0; i < avctx->channels; i++) {
yading@10:             if (c->status[i].idelta < 16)
yading@10:                 c->status[i].idelta = 16;
yading@10:             bytestream_put_le16(&dst, c->status[i].idelta);
yading@10:         }
yading@10:         for (i = 0; i < avctx->channels; i++)
yading@10:             c->status[i].sample2= *samples++;
yading@10:         for (i = 0; i < avctx->channels; i++) {
yading@10:             c->status[i].sample1 = *samples++;
yading@10:             bytestream_put_le16(&dst, c->status[i].sample1);
yading@10:         }
yading@10:         for (i = 0; i < avctx->channels; i++)
yading@10:             bytestream_put_le16(&dst, c->status[i].sample2);
yading@10: 
yading@10:         if (avctx->trellis > 0) {
yading@10:             n = avctx->block_align - 7 * avctx->channels;
yading@10:             FF_ALLOC_OR_GOTO(avctx, buf, 2 * n, error);
yading@10:             if (avctx->channels == 1) {
yading@10:                 adpcm_compress_trellis(avctx, samples, buf, &c->status[0], n,
yading@10:                                        avctx->channels);
yading@10:                 for (i = 0; i < n; i += 2)
yading@10:                     *dst++ = (buf[i] << 4) | buf[i + 1];
yading@10:             } else {
yading@10:                 adpcm_compress_trellis(avctx, samples,     buf,
yading@10:                                        &c->status[0], n, avctx->channels);
yading@10:                 adpcm_compress_trellis(avctx, samples + 1, buf + n,
yading@10:                                        &c->status[1], n, avctx->channels);
yading@10:                 for (i = 0; i < n; i++)
yading@10:                     *dst++ = (buf[i] << 4) | buf[n + i];
yading@10:             }
yading@10:             av_free(buf);
yading@10:         } else {
yading@10:             for (i = 7 * avctx->channels; i < avctx->block_align; i++) {
yading@10:                 int nibble;
yading@10:                 nibble  = adpcm_ms_compress_sample(&c->status[ 0], *samples++) << 4;
yading@10:                 nibble |= adpcm_ms_compress_sample(&c->status[st], *samples++);
yading@10:                 *dst++  = nibble;
yading@10:             }
yading@10:         }
yading@10:         break;
yading@10:     case AV_CODEC_ID_ADPCM_YAMAHA:
yading@10:         n = frame->nb_samples / 2;
yading@10:         if (avctx->trellis > 0) {
yading@10:             FF_ALLOC_OR_GOTO(avctx, buf, 2 * n * 2, error);
yading@10:             n *= 2;
yading@10:             if (avctx->channels == 1) {
yading@10:                 adpcm_compress_trellis(avctx, samples, buf, &c->status[0], n,
yading@10:                                        avctx->channels);
yading@10:                 for (i = 0; i < n; i += 2)
yading@10:                     *dst++ = buf[i] | (buf[i + 1] << 4);
yading@10:             } else {
yading@10:                 adpcm_compress_trellis(avctx, samples,     buf,
yading@10:                                        &c->status[0], n, avctx->channels);
yading@10:                 adpcm_compress_trellis(avctx, samples + 1, buf + n,
yading@10:                                        &c->status[1], n, avctx->channels);
yading@10:                 for (i = 0; i < n; i++)
yading@10:                     *dst++ = buf[i] | (buf[n + i] << 4);
yading@10:             }
yading@10:             av_free(buf);
yading@10:         } else
yading@10:             for (n *= avctx->channels; n > 0; n--) {
yading@10:                 int nibble;
yading@10:                 nibble  = adpcm_yamaha_compress_sample(&c->status[ 0], *samples++);
yading@10:                 nibble |= adpcm_yamaha_compress_sample(&c->status[st], *samples++) << 4;
yading@10:                 *dst++  = nibble;
yading@10:             }
yading@10:         break;
yading@10:     default:
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10: 
yading@10:     avpkt->size = pkt_size;
yading@10:     *got_packet_ptr = 1;
yading@10:     return 0;
yading@10: error:
yading@10:     return AVERROR(ENOMEM);
yading@10: }
yading@10: 
yading@10: static const enum AVSampleFormat sample_fmts[] = {
yading@10:     AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
yading@10: };
yading@10: 
yading@10: static const enum AVSampleFormat sample_fmts_p[] = {
yading@10:     AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_NONE
yading@10: };
yading@10: 
yading@10: #define ADPCM_ENCODER(id_, name_, sample_fmts_, long_name_) \
yading@10: AVCodec ff_ ## name_ ## _encoder = {                        \
yading@10:     .name           = #name_,                               \
yading@10:     .type           = AVMEDIA_TYPE_AUDIO,                   \
yading@10:     .id             = id_,                                  \
yading@10:     .priv_data_size = sizeof(ADPCMEncodeContext),           \
yading@10:     .init           = adpcm_encode_init,                    \
yading@10:     .encode2        = adpcm_encode_frame,                   \
yading@10:     .close          = adpcm_encode_close,                   \
yading@10:     .sample_fmts    = sample_fmts_,                         \
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL(long_name_),     \
yading@10: }
yading@10: 
yading@10: ADPCM_ENCODER(AV_CODEC_ID_ADPCM_IMA_QT,  adpcm_ima_qt,  sample_fmts_p, "ADPCM IMA QuickTime");
yading@10: ADPCM_ENCODER(AV_CODEC_ID_ADPCM_IMA_WAV, adpcm_ima_wav, sample_fmts_p, "ADPCM IMA WAV");
yading@10: ADPCM_ENCODER(AV_CODEC_ID_ADPCM_MS,      adpcm_ms,      sample_fmts,   "ADPCM Microsoft");
yading@10: ADPCM_ENCODER(AV_CODEC_ID_ADPCM_SWF,     adpcm_swf,     sample_fmts,   "ADPCM Shockwave Flash");
yading@10: ADPCM_ENCODER(AV_CODEC_ID_ADPCM_YAMAHA,  adpcm_yamaha,  sample_fmts,   "ADPCM Yamaha");