2 * Direct Stream Transfer (DST) decoder
3 * Copyright (c) 2014 Peter Ross <pross@xvid.org>
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
24 * Direct Stream Transfer (DST) decoder
25 * ISO/IEC 14496-3 Part 3 Subpart 10: Technical description of lossless coding of oversampled audio
28 #include "libavutil/avassert.h"
29 #include "libavutil/intreadwrite.h"
37 #define DST_MAX_CHANNELS 6
38 #define DST_MAX_ELEMENTS (2 * DST_MAX_CHANNELS)
40 #define DSD_FS44(sample_rate) (sample_rate * 8LL / 44100)
42 #define DST_SAMPLES_PER_FRAME(sample_rate) (588 * DSD_FS44(sample_rate))
44 static const int8_t fsets_code_pred_coeff[3][3] = {
50 static const int8_t probs_code_pred_coeff[3][3] = {
56 typedef struct ArithCoder {
62 typedef struct Table {
63 unsigned int elements;
64 unsigned int length[DST_MAX_ELEMENTS];
65 int coeff[DST_MAX_ELEMENTS][128];
68 typedef struct DSTContext {
74 DECLARE_ALIGNED(16, uint8_t, status)[DST_MAX_CHANNELS][16];
75 DECLARE_ALIGNED(16, int16_t, filter)[DST_MAX_ELEMENTS][16][256];
76 DSDContext dsdctx[DST_MAX_CHANNELS];
79 static av_cold int decode_init(AVCodecContext *avctx)
81 DSTContext *s = avctx->priv_data;
84 if (avctx->channels > DST_MAX_CHANNELS) {
85 avpriv_request_sample(avctx, "Channel count %d", avctx->channels);
86 return AVERROR_PATCHWELCOME;
89 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
91 for (i = 0; i < avctx->channels; i++)
92 memset(s->dsdctx[i].buf, 0x69, sizeof(s->dsdctx[i].buf));
99 static int read_map(GetBitContext *gb, Table *t, unsigned int map[DST_MAX_CHANNELS], int channels)
104 if (!get_bits1(gb)) {
105 for (ch = 1; ch < channels; ch++) {
106 int bits = av_log2(t->elements) + 1;
107 map[ch] = get_bits(gb, bits);
108 if (map[ch] == t->elements) {
110 if (t->elements >= DST_MAX_ELEMENTS)
111 return AVERROR_INVALIDDATA;
112 } else if (map[ch] > t->elements) {
113 return AVERROR_INVALIDDATA;
117 memset(map, 0, sizeof(*map) * DST_MAX_CHANNELS);
122 static av_always_inline int get_sr_golomb_dst(GetBitContext *gb, unsigned int k)
124 int v = get_ur_golomb(gb, k, get_bits_left(gb), 0);
125 if (v && get_bits1(gb))
130 static void read_uncoded_coeff(GetBitContext *gb, int *dst, unsigned int elements,
131 int coeff_bits, int is_signed, int offset)
135 for (i = 0; i < elements; i++) {
136 dst[i] = (is_signed ? get_sbits(gb, coeff_bits) : get_bits(gb, coeff_bits)) + offset;
140 static int read_table(GetBitContext *gb, Table *t, const int8_t code_pred_coeff[3][3],
141 int length_bits, int coeff_bits, int is_signed, int offset)
143 unsigned int i, j, k;
144 for (i = 0; i < t->elements; i++) {
145 t->length[i] = get_bits(gb, length_bits) + 1;
146 if (!get_bits1(gb)) {
147 read_uncoded_coeff(gb, t->coeff[i], t->length[i], coeff_bits, is_signed, offset);
149 int method = get_bits(gb, 2), lsb_size;
151 return AVERROR_INVALIDDATA;
153 read_uncoded_coeff(gb, t->coeff[i], method + 1, coeff_bits, is_signed, offset);
155 lsb_size = get_bits(gb, 3);
156 for (j = method + 1; j < t->length[i]; j++) {
158 for (k = 0; k < method + 1; k++)
159 x += code_pred_coeff[method][k] * t->coeff[i][j - k - 1];
160 c = get_sr_golomb_dst(gb, lsb_size);
166 if (c < offset || c >= offset + (1<<coeff_bits))
167 return AVERROR_INVALIDDATA;
176 static void ac_init(ArithCoder *ac, GetBitContext *gb)
179 ac->c = get_bits(gb, 12);
183 static av_always_inline void ac_get(ArithCoder *ac, GetBitContext *gb, int p, int *e)
185 unsigned int k = (ac->a >> 8) | ((ac->a >> 7) & 1);
186 unsigned int q = k * p;
187 unsigned int a_q = ac->a - q;
198 int n = 11 - av_log2(ac->a);
200 if (get_bits_left(gb) < n)
202 ac->c = (ac->c << n) | get_bits(gb, n);
206 static uint8_t prob_dst_x_bit(int c)
208 return (ff_reverse[c & 127] >> 1) + 1;
211 static void build_filter(int16_t table[DST_MAX_ELEMENTS][16][256], const Table *fsets)
215 for (i = 0; i < fsets->elements; i++) {
216 int length = fsets->length[i];
218 for (j = 0; j < 16; j++) {
219 int total = av_clip(length - j * 8, 0, 8);
221 for (k = 0; k < 256; k++) {
224 for (l = 0; l < total; l++)
225 v += (((k >> l) & 1) * 2 - 1) * fsets->coeff[i][j * 8 + l];
232 static int decode_frame(AVCodecContext *avctx, void *data,
233 int *got_frame_ptr, AVPacket *avpkt)
235 unsigned samples_per_frame = DST_SAMPLES_PER_FRAME(avctx->sample_rate);
236 unsigned map_ch_to_felem[DST_MAX_CHANNELS];
237 unsigned map_ch_to_pelem[DST_MAX_CHANNELS];
238 unsigned i, ch, same_map, dst_x_bit;
239 unsigned half_prob[DST_MAX_CHANNELS];
240 const int channels = avctx->channels;
241 DSTContext *s = avctx->priv_data;
242 GetBitContext *gb = &s->gb;
243 ArithCoder *ac = &s->ac;
244 AVFrame *frame = data;
249 if (avpkt->size <= 1)
250 return AVERROR_INVALIDDATA;
252 frame->nb_samples = samples_per_frame / 8;
253 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
255 dsd = frame->data[0];
256 pcm = (float *)frame->data[0];
258 if ((ret = init_get_bits8(gb, avpkt->data, avpkt->size)) < 0)
261 if (!get_bits1(gb)) {
264 return AVERROR_INVALIDDATA;
265 memcpy(frame->data[0], avpkt->data + 1, FFMIN(avpkt->size - 1, frame->nb_samples * avctx->channels));
269 /* Segmentation (10.4, 10.5, 10.6) */
271 if (!get_bits1(gb)) {
272 avpriv_request_sample(avctx, "Not Same Segmentation");
273 return AVERROR_PATCHWELCOME;
276 if (!get_bits1(gb)) {
277 avpriv_request_sample(avctx, "Not Same Segmentation For All Channels");
278 return AVERROR_PATCHWELCOME;
281 if (!get_bits1(gb)) {
282 avpriv_request_sample(avctx, "Not End Of Channel Segmentation");
283 return AVERROR_PATCHWELCOME;
286 /* Mapping (10.7, 10.8, 10.9) */
288 same_map = get_bits1(gb);
290 if ((ret = read_map(gb, &s->fsets, map_ch_to_felem, avctx->channels)) < 0)
294 s->probs.elements = s->fsets.elements;
295 memcpy(map_ch_to_pelem, map_ch_to_felem, sizeof(map_ch_to_felem));
297 avpriv_request_sample(avctx, "Not Same Mapping");
298 if ((ret = read_map(gb, &s->probs, map_ch_to_pelem, avctx->channels)) < 0)
302 /* Half Probability (10.10) */
304 for (ch = 0; ch < avctx->channels; ch++)
305 half_prob[ch] = get_bits1(gb);
307 /* Filter Coef Sets (10.12) */
309 ret = read_table(gb, &s->fsets, fsets_code_pred_coeff, 7, 9, 1, 0);
313 /* Probability Tables (10.13) */
315 ret = read_table(gb, &s->probs, probs_code_pred_coeff, 6, 7, 0, 1);
319 /* Arithmetic Coded Data (10.11) */
322 return AVERROR_INVALIDDATA;
325 build_filter(s->filter, &s->fsets);
327 memset(s->status, 0xAA, sizeof(s->status));
328 memset(dsd, 0, frame->nb_samples * 4 * avctx->channels);
330 ac_get(ac, gb, prob_dst_x_bit(s->fsets.coeff[0][0]), &dst_x_bit);
332 for (i = 0; i < samples_per_frame; i++) {
333 for (ch = 0; ch < channels; ch++) {
334 const unsigned felem = map_ch_to_felem[ch];
335 int16_t (*filter)[256] = s->filter[felem];
336 uint8_t *status = s->status[ch];
337 int prob, residual, v;
339 #define F(x) filter[(x)][status[(x)]]
340 const int16_t predict = F( 0) + F( 1) + F( 2) + F( 3) +
341 F( 4) + F( 5) + F( 6) + F( 7) +
342 F( 8) + F( 9) + F(10) + F(11) +
343 F(12) + F(13) + F(14) + F(15);
346 if (!half_prob[ch] || i >= s->fsets.length[felem]) {
347 unsigned pelem = map_ch_to_pelem[ch];
348 unsigned index = FFABS(predict) >> 3;
349 prob = s->probs.coeff[pelem][FFMIN(index, s->probs.length[pelem] - 1)];
354 if (ac->overread > 16)
355 return AVERROR_INVALIDDATA;
357 ac_get(ac, gb, prob, &residual);
358 v = ((predict >> 15) ^ residual) & 1;
359 dsd[((i >> 3) * channels + ch) << 2] |= v << (7 - (i & 0x7 ));
361 AV_WL64A(status + 8, (AV_RL64A(status + 8) << 1) | ((AV_RL64A(status) >> 63) & 1));
362 AV_WL64A(status, (AV_RL64A(status) << 1) | v);
367 for (i = 0; i < avctx->channels; i++) {
368 ff_dsd2pcm_translate(&s->dsdctx[i], frame->nb_samples, 0,
369 frame->data[0] + i * 4,
370 avctx->channels * 4, pcm + i, avctx->channels);
378 AVCodec ff_dst_decoder = {
380 .long_name = NULL_IF_CONFIG_SMALL("DST (Digital Stream Transfer)"),
381 .type = AVMEDIA_TYPE_AUDIO,
382 .id = AV_CODEC_ID_DST,
383 .priv_data_size = sizeof(DSTContext),
385 .decode = decode_frame,
386 .capabilities = AV_CODEC_CAP_DR1,
387 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLT,
388 AV_SAMPLE_FMT_NONE },