xref: /third_party/ffmpeg/libavcodec/dcadec.c (revision cabdff1a)
1/*
2 * Copyright (C) 2016 foo86
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#include "libavutil/opt.h"
22#include "libavutil/channel_layout.h"
23#include "libavutil/thread.h"
24
25#include "codec_internal.h"
26#include "dcadec.h"
27#include "dcahuff.h"
28#include "dca_syncwords.h"
29#include "profiles.h"
30
31#define MIN_PACKET_SIZE     16
32#define MAX_PACKET_SIZE     0x104000
33
34int ff_dca_set_channel_layout(AVCodecContext *avctx, int *ch_remap, int dca_mask)
35{
36    static const uint8_t dca2wav_norm[28] = {
37         2,  0, 1, 9, 10,  3,  8,  4,  5,  9, 10, 6, 7, 12,
38        13, 14, 3, 6,  7, 11, 12, 14, 16, 15, 17, 8, 4,  5,
39    };
40
41    static const uint8_t dca2wav_wide[28] = {
42         2,  0, 1, 4,  5,  3,  8,  4,  5,  9, 10, 6, 7, 12,
43        13, 14, 3, 9, 10, 11, 12, 14, 16, 15, 17, 8, 4,  5,
44    };
45
46    DCAContext *s = avctx->priv_data;
47
48    int dca_ch, wav_ch, nchannels = 0;
49
50    av_channel_layout_uninit(&avctx->ch_layout);
51    if (s->output_channel_order == CHANNEL_ORDER_CODED) {
52        for (dca_ch = 0; dca_ch < DCA_SPEAKER_COUNT; dca_ch++)
53            if (dca_mask & (1U << dca_ch))
54                ch_remap[nchannels++] = dca_ch;
55        avctx->ch_layout.order       = AV_CHANNEL_ORDER_UNSPEC;
56        avctx->ch_layout.nb_channels = nchannels;
57    } else {
58        int wav_mask = 0;
59        int wav_map[18];
60        const uint8_t *dca2wav;
61        if (dca_mask == DCA_SPEAKER_LAYOUT_7POINT0_WIDE ||
62            dca_mask == DCA_SPEAKER_LAYOUT_7POINT1_WIDE)
63            dca2wav = dca2wav_wide;
64        else
65            dca2wav = dca2wav_norm;
66        for (dca_ch = 0; dca_ch < 28; dca_ch++) {
67            if (dca_mask & (1 << dca_ch)) {
68                wav_ch = dca2wav[dca_ch];
69                if (!(wav_mask & (1 << wav_ch))) {
70                    wav_map[wav_ch] = dca_ch;
71                    wav_mask |= 1 << wav_ch;
72                }
73            }
74        }
75        for (wav_ch = 0; wav_ch < 18; wav_ch++)
76            if (wav_mask & (1 << wav_ch))
77                ch_remap[nchannels++] = wav_map[wav_ch];
78
79        av_channel_layout_from_mask(&avctx->ch_layout, wav_mask);
80    }
81
82    return nchannels;
83}
84
85void ff_dca_downmix_to_stereo_fixed(DCADSPContext *dcadsp, int32_t **samples,
86                                    int *coeff_l, int nsamples, int ch_mask)
87{
88    int pos, spkr, max_spkr = av_log2(ch_mask);
89    int *coeff_r = coeff_l + av_popcount(ch_mask);
90
91    av_assert0(DCA_HAS_STEREO(ch_mask));
92
93    // Scale left and right channels
94    pos = (ch_mask & DCA_SPEAKER_MASK_C);
95    dcadsp->dmix_scale(samples[DCA_SPEAKER_L], coeff_l[pos    ], nsamples);
96    dcadsp->dmix_scale(samples[DCA_SPEAKER_R], coeff_r[pos + 1], nsamples);
97
98    // Downmix remaining channels
99    for (spkr = 0; spkr <= max_spkr; spkr++) {
100        if (!(ch_mask & (1U << spkr)))
101            continue;
102
103        if (*coeff_l && spkr != DCA_SPEAKER_L)
104            dcadsp->dmix_add(samples[DCA_SPEAKER_L], samples[spkr],
105                             *coeff_l, nsamples);
106
107        if (*coeff_r && spkr != DCA_SPEAKER_R)
108            dcadsp->dmix_add(samples[DCA_SPEAKER_R], samples[spkr],
109                             *coeff_r, nsamples);
110
111        coeff_l++;
112        coeff_r++;
113    }
114}
115
116void ff_dca_downmix_to_stereo_float(AVFloatDSPContext *fdsp, float **samples,
117                                    int *coeff_l, int nsamples, int ch_mask)
118{
119    int pos, spkr, max_spkr = av_log2(ch_mask);
120    int *coeff_r = coeff_l + av_popcount(ch_mask);
121    const float scale = 1.0f / (1 << 15);
122
123    av_assert0(DCA_HAS_STEREO(ch_mask));
124
125    // Scale left and right channels
126    pos = (ch_mask & DCA_SPEAKER_MASK_C);
127    fdsp->vector_fmul_scalar(samples[DCA_SPEAKER_L], samples[DCA_SPEAKER_L],
128                             coeff_l[pos    ] * scale, nsamples);
129    fdsp->vector_fmul_scalar(samples[DCA_SPEAKER_R], samples[DCA_SPEAKER_R],
130                             coeff_r[pos + 1] * scale, nsamples);
131
132    // Downmix remaining channels
133    for (spkr = 0; spkr <= max_spkr; spkr++) {
134        if (!(ch_mask & (1U << spkr)))
135            continue;
136
137        if (*coeff_l && spkr != DCA_SPEAKER_L)
138            fdsp->vector_fmac_scalar(samples[DCA_SPEAKER_L], samples[spkr],
139                                     *coeff_l * scale, nsamples);
140
141        if (*coeff_r && spkr != DCA_SPEAKER_R)
142            fdsp->vector_fmac_scalar(samples[DCA_SPEAKER_R], samples[spkr],
143                                     *coeff_r * scale, nsamples);
144
145        coeff_l++;
146        coeff_r++;
147    }
148}
149
150static int dcadec_decode_frame(AVCodecContext *avctx, AVFrame *frame,
151                               int *got_frame_ptr, AVPacket *avpkt)
152{
153    DCAContext *s = avctx->priv_data;
154    const uint8_t *input = avpkt->data;
155    int input_size = avpkt->size;
156    int i, ret, prev_packet = s->packet;
157    uint32_t mrk;
158
159    if (input_size < MIN_PACKET_SIZE || input_size > MAX_PACKET_SIZE) {
160        av_log(avctx, AV_LOG_ERROR, "Invalid packet size\n");
161        return AVERROR_INVALIDDATA;
162    }
163
164    // Convert input to BE format
165    mrk = AV_RB32(input);
166    if (mrk != DCA_SYNCWORD_CORE_BE && mrk != DCA_SYNCWORD_SUBSTREAM) {
167        av_fast_padded_malloc(&s->buffer, &s->buffer_size, input_size);
168        if (!s->buffer)
169            return AVERROR(ENOMEM);
170
171        for (i = 0, ret = AVERROR_INVALIDDATA; i < input_size - MIN_PACKET_SIZE + 1 && ret < 0; i++)
172            ret = avpriv_dca_convert_bitstream(input + i, input_size - i, s->buffer, s->buffer_size);
173
174        if (ret < 0) {
175            av_log(avctx, AV_LOG_ERROR, "Not a valid DCA frame\n");
176            return ret;
177        }
178
179        input      = s->buffer;
180        input_size = ret;
181    }
182
183    s->packet = 0;
184
185    // Parse backward compatible core sub-stream
186    if (AV_RB32(input) == DCA_SYNCWORD_CORE_BE) {
187        int frame_size;
188
189        if ((ret = ff_dca_core_parse(&s->core, input, input_size)) < 0)
190            return ret;
191
192        s->packet |= DCA_PACKET_CORE;
193
194        // EXXS data must be aligned on 4-byte boundary
195        frame_size = FFALIGN(s->core.frame_size, 4);
196        if (input_size - 4 > frame_size) {
197            input      += frame_size;
198            input_size -= frame_size;
199        }
200    }
201
202    if (!s->core_only) {
203        DCAExssAsset *asset = NULL;
204
205        // Parse extension sub-stream (EXSS)
206        if (AV_RB32(input) == DCA_SYNCWORD_SUBSTREAM) {
207            if ((ret = ff_dca_exss_parse(&s->exss, input, input_size)) < 0) {
208                if (avctx->err_recognition & AV_EF_EXPLODE)
209                    return ret;
210            } else {
211                s->packet |= DCA_PACKET_EXSS;
212                asset = &s->exss.assets[0];
213            }
214        }
215
216        // Parse XLL component in EXSS
217        if (asset && (asset->extension_mask & DCA_EXSS_XLL)) {
218            if ((ret = ff_dca_xll_parse(&s->xll, input, asset)) < 0) {
219                // Conceal XLL synchronization error
220                if (ret == AVERROR(EAGAIN)
221                    && (prev_packet & DCA_PACKET_XLL)
222                    && (s->packet & DCA_PACKET_CORE))
223                    s->packet |= DCA_PACKET_XLL | DCA_PACKET_RECOVERY;
224                else if (ret == AVERROR(ENOMEM) || (avctx->err_recognition & AV_EF_EXPLODE))
225                    return ret;
226            } else {
227                s->packet |= DCA_PACKET_XLL;
228            }
229        }
230
231        // Parse LBR component in EXSS
232        if (asset && (asset->extension_mask & DCA_EXSS_LBR)) {
233            if ((ret = ff_dca_lbr_parse(&s->lbr, input, asset)) < 0) {
234                if (ret == AVERROR(ENOMEM) || (avctx->err_recognition & AV_EF_EXPLODE))
235                    return ret;
236            } else {
237                s->packet |= DCA_PACKET_LBR;
238            }
239        }
240
241        // Parse core extensions in EXSS or backward compatible core sub-stream
242        if ((s->packet & DCA_PACKET_CORE)
243            && (ret = ff_dca_core_parse_exss(&s->core, input, asset)) < 0)
244            return ret;
245    }
246
247    // Filter the frame
248    if (s->packet & DCA_PACKET_LBR) {
249        if ((ret = ff_dca_lbr_filter_frame(&s->lbr, frame)) < 0)
250            return ret;
251    } else if (s->packet & DCA_PACKET_XLL) {
252        if (s->packet & DCA_PACKET_CORE) {
253            int x96_synth = -1;
254
255            // Enable X96 synthesis if needed
256            if (s->xll.chset[0].freq == 96000 && s->core.sample_rate == 48000)
257                x96_synth = 1;
258
259            if ((ret = ff_dca_core_filter_fixed(&s->core, x96_synth)) < 0)
260                return ret;
261
262            // Force lossy downmixed output on the first core frame filtered.
263            // This prevents audible clicks when seeking and is consistent with
264            // what reference decoder does when there are multiple channel sets.
265            if (!(prev_packet & DCA_PACKET_RESIDUAL) && s->xll.nreschsets > 0
266                && s->xll.nchsets > 1) {
267                av_log(avctx, AV_LOG_VERBOSE, "Forcing XLL recovery mode\n");
268                s->packet |= DCA_PACKET_RECOVERY;
269            }
270
271            // Set 'residual ok' flag for the next frame
272            s->packet |= DCA_PACKET_RESIDUAL;
273        }
274
275        if ((ret = ff_dca_xll_filter_frame(&s->xll, frame)) < 0) {
276            // Fall back to core unless hard error
277            if (!(s->packet & DCA_PACKET_CORE))
278                return ret;
279            if (ret != AVERROR_INVALIDDATA || (avctx->err_recognition & AV_EF_EXPLODE))
280                return ret;
281            if ((ret = ff_dca_core_filter_frame(&s->core, frame)) < 0)
282                return ret;
283        }
284    } else if (s->packet & DCA_PACKET_CORE) {
285        if ((ret = ff_dca_core_filter_frame(&s->core, frame)) < 0)
286            return ret;
287        if (s->core.filter_mode & DCA_FILTER_MODE_FIXED)
288            s->packet |= DCA_PACKET_RESIDUAL;
289    } else {
290        av_log(avctx, AV_LOG_ERROR, "No valid DCA sub-stream found\n");
291        if (s->core_only)
292            av_log(avctx, AV_LOG_WARNING, "Consider disabling 'core_only' option\n");
293        return AVERROR_INVALIDDATA;
294    }
295
296    *got_frame_ptr = 1;
297
298    return avpkt->size;
299}
300
301static av_cold void dcadec_flush(AVCodecContext *avctx)
302{
303    DCAContext *s = avctx->priv_data;
304
305    ff_dca_core_flush(&s->core);
306    ff_dca_xll_flush(&s->xll);
307    ff_dca_lbr_flush(&s->lbr);
308
309    s->packet &= DCA_PACKET_MASK;
310}
311
312static av_cold int dcadec_close(AVCodecContext *avctx)
313{
314    DCAContext *s = avctx->priv_data;
315
316    ff_dca_core_close(&s->core);
317    ff_dca_xll_close(&s->xll);
318    ff_dca_lbr_close(&s->lbr);
319
320    av_freep(&s->buffer);
321    s->buffer_size = 0;
322
323    return 0;
324}
325
326static av_cold void dcadec_init_static(void)
327{
328    ff_dca_lbr_init_tables();
329    ff_dca_init_vlcs();
330}
331
332static av_cold int dcadec_init(AVCodecContext *avctx)
333{
334    static AVOnce init_static_once = AV_ONCE_INIT;
335    DCAContext *s = avctx->priv_data;
336
337    s->avctx = avctx;
338    s->core.avctx = avctx;
339    s->exss.avctx = avctx;
340    s->xll.avctx = avctx;
341    s->lbr.avctx = avctx;
342
343    if (ff_dca_core_init(&s->core) < 0)
344        return AVERROR(ENOMEM);
345
346    if (ff_dca_lbr_init(&s->lbr) < 0)
347        return AVERROR(ENOMEM);
348
349    ff_dcadsp_init(&s->dcadsp);
350    s->core.dcadsp = &s->dcadsp;
351    s->xll.dcadsp = &s->dcadsp;
352    s->lbr.dcadsp = &s->dcadsp;
353
354    s->crctab = av_crc_get_table(AV_CRC_16_CCITT);
355
356#if FF_API_OLD_CHANNEL_LAYOUT
357FF_DISABLE_DEPRECATION_WARNINGS
358    if (avctx->request_channel_layout & AV_CH_LAYOUT_NATIVE)
359        s->output_channel_order = CHANNEL_ORDER_CODED;
360
361    if (avctx->request_channel_layout & ~AV_CH_LAYOUT_NATIVE) {
362        av_channel_layout_uninit(&s->downmix_layout);
363        av_channel_layout_from_mask(&s->downmix_layout, avctx->request_channel_layout & ~AV_CH_LAYOUT_NATIVE);
364    }
365FF_ENABLE_DEPRECATION_WARNINGS
366#endif
367
368    if (s->downmix_layout.nb_channels) {
369        if (!av_channel_layout_compare(&s->downmix_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_STEREO) ||
370            !av_channel_layout_compare(&s->downmix_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_STEREO_DOWNMIX))
371            s->request_channel_layout = DCA_SPEAKER_LAYOUT_STEREO;
372        else if (!av_channel_layout_compare(&s->downmix_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_5POINT0))
373            s->request_channel_layout = DCA_SPEAKER_LAYOUT_5POINT0;
374        else if (!av_channel_layout_compare(&s->downmix_layout, &(AVChannelLayout)AV_CHANNEL_LAYOUT_5POINT1))
375            s->request_channel_layout = DCA_SPEAKER_LAYOUT_5POINT1;
376        else
377            av_log(avctx, AV_LOG_WARNING, "Invalid downmix layout\n");
378    }
379
380    ff_thread_once(&init_static_once, dcadec_init_static);
381
382    return 0;
383}
384
385#define OFFSET(x) offsetof(DCAContext, x)
386#define PARAM AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM
387
388static const AVOption dcadec_options[] = {
389    { "core_only", "Decode core only without extensions", OFFSET(core_only), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, PARAM },
390
391    { "channel_order", "Order in which the channels are to be exported",
392        OFFSET(output_channel_order), AV_OPT_TYPE_INT,
393        { .i64 = CHANNEL_ORDER_DEFAULT }, 0, 1, PARAM, "channel_order" },
394      { "default", "normal libavcodec channel order", 0, AV_OPT_TYPE_CONST,
395        { .i64 = CHANNEL_ORDER_DEFAULT }, .flags = PARAM, "channel_order" },
396      { "coded",    "order in which the channels are coded in the bitstream",
397        0, AV_OPT_TYPE_CONST, { .i64 = CHANNEL_ORDER_CODED }, .flags = PARAM, "channel_order" },
398
399    { "downmix", "Request a specific channel layout from the decoder", OFFSET(downmix_layout),
400        AV_OPT_TYPE_CHLAYOUT, {.str = NULL}, .flags = PARAM },
401
402    { NULL }
403};
404
405static const AVClass dcadec_class = {
406    .class_name = "DCA decoder",
407    .item_name  = av_default_item_name,
408    .option     = dcadec_options,
409    .version    = LIBAVUTIL_VERSION_INT,
410    .category   = AV_CLASS_CATEGORY_DECODER,
411};
412
413const FFCodec ff_dca_decoder = {
414    .p.name         = "dca",
415    .p.long_name    = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
416    .p.type         = AVMEDIA_TYPE_AUDIO,
417    .p.id           = AV_CODEC_ID_DTS,
418    .priv_data_size = sizeof(DCAContext),
419    .init           = dcadec_init,
420    FF_CODEC_DECODE_CB(dcadec_decode_frame),
421    .close          = dcadec_close,
422    .flush          = dcadec_flush,
423    .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
424    .p.sample_fmts  = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S32P,
425                                                      AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE },
426    .p.priv_class   = &dcadec_class,
427    .p.profiles     = NULL_IF_CONFIG_SMALL(ff_dca_profiles),
428    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
429};
430