ATRAC+ decoder
[ffmpeg.git] / libavcodec / atrac3plusdec.c
1 /*
2  * ATRAC3+ compatible decoder
3  *
4  * Copyright (c) 2010-2013 Maxim Poliakovski
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22
23 /**
24  * @file
25  * Sony ATRAC3+ compatible decoder.
26  *
27  * Container formats used to store its data:
28  * RIFF WAV (.at3) and Sony OpenMG (.oma, .aa3).
29  *
30  * Technical description of this codec can be found here:
31  * http://wiki.multimedia.cx/index.php?title=ATRAC3plus
32  *
33  * Kudos to Benjamin Larsson and Michael Karcher
34  * for their precious technical help!
35  */
36
37 #include <stdint.h>
38 #include <string.h>
39
40 #include "libavutil/channel_layout.h"
41 #include "libavutil/float_dsp.h"
42 #include "avcodec.h"
43 #include "get_bits.h"
44 #include "internal.h"
45 #include "atrac.h"
46 #include "atrac3plus.h"
47
48 typedef struct ATRAC3PContext {
49     GetBitContext gb;
50     AVFloatDSPContext fdsp;
51
52     DECLARE_ALIGNED(32, float, samples)[2][ATRAC3P_FRAME_SAMPLES];  ///< quantized MDCT spectrum
53     DECLARE_ALIGNED(32, float, mdct_buf)[2][ATRAC3P_FRAME_SAMPLES]; ///< output of the IMDCT
54     DECLARE_ALIGNED(32, float, time_buf)[2][ATRAC3P_FRAME_SAMPLES]; ///< output of the gain compensation
55     DECLARE_ALIGNED(32, float, outp_buf)[2][ATRAC3P_FRAME_SAMPLES];
56
57     AtracGCContext gainc_ctx;   ///< gain compensation context
58     FFTContext mdct_ctx;
59     FFTContext ipqf_dct_ctx;    ///< IDCT context used by IPQF
60
61     Atrac3pChanUnitCtx *ch_units;   ///< global channel units
62
63     int num_channel_blocks;     ///< number of channel blocks
64     uint8_t channel_blocks[5];  ///< channel configuration descriptor
65     uint64_t my_channel_layout; ///< current channel layout
66 } ATRAC3PContext;
67
68 static av_cold int atrac3p_decode_close(AVCodecContext *avctx)
69 {
70     av_free(((ATRAC3PContext *)(avctx->priv_data))->ch_units);
71
72     return 0;
73 }
74
75 static av_cold int set_channel_params(ATRAC3PContext *ctx,
76                                       AVCodecContext *avctx)
77 {
78     memset(ctx->channel_blocks, 0, sizeof(ctx->channel_blocks));
79
80     switch (avctx->channel_layout) {
81     case AV_CH_FRONT_LEFT:
82     case AV_CH_LAYOUT_MONO:
83         ctx->num_channel_blocks = 1;
84         ctx->channel_blocks[0]  = CH_UNIT_MONO;
85         break;
86     case AV_CH_LAYOUT_STEREO:
87         ctx->num_channel_blocks = 1;
88         ctx->channel_blocks[0]  = CH_UNIT_STEREO;
89         break;
90     case AV_CH_LAYOUT_SURROUND:
91         ctx->num_channel_blocks = 2;
92         ctx->channel_blocks[0]  = CH_UNIT_STEREO;
93         ctx->channel_blocks[1]  = CH_UNIT_MONO;
94         break;
95     case AV_CH_LAYOUT_4POINT0:
96         ctx->num_channel_blocks = 3;
97         ctx->channel_blocks[0]  = CH_UNIT_STEREO;
98         ctx->channel_blocks[1]  = CH_UNIT_MONO;
99         ctx->channel_blocks[2]  = CH_UNIT_MONO;
100         break;
101     case AV_CH_LAYOUT_5POINT1_BACK:
102         ctx->num_channel_blocks = 4;
103         ctx->channel_blocks[0]  = CH_UNIT_STEREO;
104         ctx->channel_blocks[1]  = CH_UNIT_MONO;
105         ctx->channel_blocks[2]  = CH_UNIT_STEREO;
106         ctx->channel_blocks[3]  = CH_UNIT_MONO;
107         break;
108     case AV_CH_LAYOUT_6POINT1_BACK:
109         ctx->num_channel_blocks = 5;
110         ctx->channel_blocks[0]  = CH_UNIT_STEREO;
111         ctx->channel_blocks[1]  = CH_UNIT_MONO;
112         ctx->channel_blocks[2]  = CH_UNIT_STEREO;
113         ctx->channel_blocks[3]  = CH_UNIT_MONO;
114         ctx->channel_blocks[4]  = CH_UNIT_MONO;
115         break;
116     case AV_CH_LAYOUT_7POINT1:
117         ctx->num_channel_blocks = 5;
118         ctx->channel_blocks[0]  = CH_UNIT_STEREO;
119         ctx->channel_blocks[1]  = CH_UNIT_MONO;
120         ctx->channel_blocks[2]  = CH_UNIT_STEREO;
121         ctx->channel_blocks[3]  = CH_UNIT_STEREO;
122         ctx->channel_blocks[4]  = CH_UNIT_MONO;
123         break;
124     default:
125         av_log(avctx, AV_LOG_ERROR,
126                "Unsupported channel layout: %"PRIx64"!\n", avctx->channel_layout);
127         return AVERROR_INVALIDDATA;
128     }
129
130     return 0;
131 }
132
133 static av_cold int atrac3p_decode_init(AVCodecContext *avctx)
134 {
135     ATRAC3PContext *ctx = avctx->priv_data;
136     int i, ch, ret;
137
138     ff_atrac3p_init_vlcs();
139
140     avpriv_float_dsp_init(&ctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
141
142     /* initialize IPQF */
143     ff_mdct_init(&ctx->ipqf_dct_ctx, 5, 1, 32.0 / 32768.0);
144
145     ff_atrac3p_init_imdct(avctx, &ctx->mdct_ctx);
146
147     ff_atrac_init_gain_compensation(&ctx->gainc_ctx, 6, 2);
148
149     ff_atrac3p_init_wave_synth();
150
151     if ((ret = set_channel_params(ctx, avctx)) < 0)
152         return ret;
153
154     ctx->my_channel_layout = avctx->channel_layout;
155
156     ctx->ch_units = av_mallocz(sizeof(*ctx->ch_units) *
157                                ctx->num_channel_blocks);
158     if (!ctx->ch_units) {
159         atrac3p_decode_close(avctx);
160         return AVERROR(ENOMEM);
161     }
162
163     for (i = 0; i < ctx->num_channel_blocks; i++) {
164         for (ch = 0; ch < 2; ch++) {
165             ctx->ch_units[i].channels[ch].ch_num          = ch;
166             ctx->ch_units[i].channels[ch].wnd_shape       = &ctx->ch_units[i].channels[ch].wnd_shape_hist[0][0];
167             ctx->ch_units[i].channels[ch].wnd_shape_prev  = &ctx->ch_units[i].channels[ch].wnd_shape_hist[1][0];
168             ctx->ch_units[i].channels[ch].gain_data       = &ctx->ch_units[i].channels[ch].gain_data_hist[0][0];
169             ctx->ch_units[i].channels[ch].gain_data_prev  = &ctx->ch_units[i].channels[ch].gain_data_hist[1][0];
170             ctx->ch_units[i].channels[ch].tones_info      = &ctx->ch_units[i].channels[ch].tones_info_hist[0][0];
171             ctx->ch_units[i].channels[ch].tones_info_prev = &ctx->ch_units[i].channels[ch].tones_info_hist[1][0];
172         }
173
174         ctx->ch_units[i].waves_info      = &ctx->ch_units[i].wave_synth_hist[0];
175         ctx->ch_units[i].waves_info_prev = &ctx->ch_units[i].wave_synth_hist[1];
176     }
177
178     avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
179
180     return 0;
181 }
182
183 static void decode_residual_spectrum(Atrac3pChanUnitCtx *ctx,
184                                      float out[2][ATRAC3P_FRAME_SAMPLES],
185                                      int num_channels,
186                                      AVCodecContext *avctx)
187 {
188     int i, sb, ch, qu, nspeclines, RNG_index;
189     float *dst, q;
190     int16_t *src;
191     /* calculate RNG table index for each subband */
192     int sb_RNG_index[ATRAC3P_SUBBANDS] = { 0 };
193
194     if (ctx->mute_flag) {
195         for (ch = 0; ch < num_channels; ch++)
196             memset(out[ch], 0, ATRAC3P_FRAME_SAMPLES * sizeof(*out[ch]));
197         return;
198     }
199
200     for (qu = 0, RNG_index = 0; qu < ctx->used_quant_units; qu++)
201         RNG_index += ctx->channels[0].qu_sf_idx[qu] +
202                      ctx->channels[1].qu_sf_idx[qu];
203
204     for (sb = 0; sb < ctx->num_coded_subbands; sb++, RNG_index += 128)
205         sb_RNG_index[sb] = RNG_index & 0x3FC;
206
207     /* inverse quant and power compensation */
208     for (ch = 0; ch < num_channels; ch++) {
209         /* clear channel's residual spectrum */
210         memset(out[ch], 0, ATRAC3P_FRAME_SAMPLES * sizeof(*out[ch]));
211
212         for (qu = 0; qu < ctx->used_quant_units; qu++) {
213             src        = &ctx->channels[ch].spectrum[ff_atrac3p_qu_to_spec_pos[qu]];
214             dst        = &out[ch][ff_atrac3p_qu_to_spec_pos[qu]];
215             nspeclines = ff_atrac3p_qu_to_spec_pos[qu + 1] -
216                          ff_atrac3p_qu_to_spec_pos[qu];
217
218             if (ctx->channels[ch].qu_wordlen[qu] > 0) {
219                 q = ff_atrac3p_sf_tab[ctx->channels[ch].qu_sf_idx[qu]] *
220                     ff_atrac3p_mant_tab[ctx->channels[ch].qu_wordlen[qu]];
221                 for (i = 0; i < nspeclines; i++)
222                     dst[i] = src[i] * q;
223             }
224         }
225
226         for (sb = 0; sb < ctx->num_coded_subbands; sb++)
227             ff_atrac3p_power_compensation(ctx, ch, &out[ch][0],
228                                           sb_RNG_index[sb], sb);
229     }
230
231     if (ctx->unit_type == CH_UNIT_STEREO) {
232         for (sb = 0; sb < ctx->num_coded_subbands; sb++) {
233             if (ctx->swap_channels[sb]) {
234                 for (i = 0; i < ATRAC3P_SUBBAND_SAMPLES; i++)
235                     FFSWAP(float, out[0][sb * ATRAC3P_SUBBAND_SAMPLES + i],
236                                   out[1][sb * ATRAC3P_SUBBAND_SAMPLES + i]);
237             }
238
239             /* flip coefficients' sign if requested */
240             if (ctx->negate_coeffs[sb])
241                 for (i = 0; i < ATRAC3P_SUBBAND_SAMPLES; i++)
242                     out[1][sb * ATRAC3P_SUBBAND_SAMPLES + i] = -(out[1][sb * ATRAC3P_SUBBAND_SAMPLES + i]);
243         }
244     }
245 }
246
247 static void reconstruct_frame(ATRAC3PContext *ctx, Atrac3pChanUnitCtx *ch_unit,
248                               int num_channels, AVCodecContext *avctx)
249 {
250     int ch, sb;
251
252     for (ch = 0; ch < num_channels; ch++) {
253         for (sb = 0; sb < ch_unit->num_subbands; sb++) {
254             /* inverse transform and windowing */
255             ff_atrac3p_imdct(&ctx->fdsp, &ctx->mdct_ctx,
256                              &ctx->samples[ch][sb * ATRAC3P_SUBBAND_SAMPLES],
257                              &ctx->mdct_buf[ch][sb * ATRAC3P_SUBBAND_SAMPLES],
258                              (ch_unit->channels[ch].wnd_shape_prev[sb] << 1) +
259                              ch_unit->channels[ch].wnd_shape[sb], sb);
260
261             /* gain compensation and overlapping */
262             ff_atrac_gain_compensation(&ctx->gainc_ctx,
263                                        &ctx->mdct_buf[ch][sb * ATRAC3P_SUBBAND_SAMPLES],
264                                        &ch_unit->prev_buf[ch][sb * ATRAC3P_SUBBAND_SAMPLES],
265                                        &ch_unit->channels[ch].gain_data_prev[sb],
266                                        &ch_unit->channels[ch].gain_data[sb],
267                                        ATRAC3P_SUBBAND_SAMPLES,
268                                        &ctx->time_buf[ch][sb * ATRAC3P_SUBBAND_SAMPLES]);
269         }
270
271         /* zero unused subbands in both output and overlapping buffers */
272         memset(&ch_unit->prev_buf[ch][ch_unit->num_subbands * ATRAC3P_SUBBAND_SAMPLES],
273                0,
274                (ATRAC3P_SUBBANDS - ch_unit->num_subbands) *
275                ATRAC3P_SUBBAND_SAMPLES *
276                sizeof(ch_unit->prev_buf[ch][ch_unit->num_subbands * ATRAC3P_SUBBAND_SAMPLES]));
277         memset(&ctx->time_buf[ch][ch_unit->num_subbands * ATRAC3P_SUBBAND_SAMPLES],
278                0,
279                (ATRAC3P_SUBBANDS - ch_unit->num_subbands) *
280                ATRAC3P_SUBBAND_SAMPLES *
281                sizeof(ctx->time_buf[ch][ch_unit->num_subbands * ATRAC3P_SUBBAND_SAMPLES]));
282
283         /* resynthesize and add tonal signal */
284         if (ch_unit->waves_info->tones_present ||
285             ch_unit->waves_info_prev->tones_present) {
286             for (sb = 0; sb < ch_unit->num_subbands; sb++)
287                 if (ch_unit->channels[ch].tones_info[sb].num_wavs ||
288                     ch_unit->channels[ch].tones_info_prev[sb].num_wavs) {
289                     ff_atrac3p_generate_tones(ch_unit, &ctx->fdsp, ch, sb,
290                                               &ctx->time_buf[ch][sb * 128]);
291                 }
292         }
293
294         /* subband synthesis and acoustic signal output */
295         ff_atrac3p_ipqf(&ctx->ipqf_dct_ctx, &ch_unit->ipqf_ctx[ch],
296                         &ctx->time_buf[ch][0], &ctx->outp_buf[ch][0]);
297     }
298
299     /* swap window shape and gain control buffers. */
300     for (ch = 0; ch < num_channels; ch++) {
301         FFSWAP(uint8_t *, ch_unit->channels[ch].wnd_shape,
302                ch_unit->channels[ch].wnd_shape_prev);
303         FFSWAP(AtracGainInfo *, ch_unit->channels[ch].gain_data,
304                ch_unit->channels[ch].gain_data_prev);
305         FFSWAP(Atrac3pWavesData *, ch_unit->channels[ch].tones_info,
306                ch_unit->channels[ch].tones_info_prev);
307     }
308
309     FFSWAP(Atrac3pWaveSynthParams *, ch_unit->waves_info, ch_unit->waves_info_prev);
310 }
311
312 static int atrac3p_decode_frame(AVCodecContext *avctx, void *data,
313                                 int *got_frame_ptr, AVPacket *avpkt)
314 {
315     ATRAC3PContext *ctx = avctx->priv_data;
316     AVFrame *frame      = data;
317     int i, ret, ch_unit_id, ch_block = 0, out_ch_index = 0, channels_to_process;
318     float **samples_p = (float **)frame->extended_data;
319
320     frame->nb_samples = ATRAC3P_FRAME_SAMPLES;
321     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
322         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
323         return ret;
324     }
325
326     if ((ret = init_get_bits8(&ctx->gb, avpkt->data, avpkt->size)) < 0)
327         return ret;
328
329     if (get_bits1(&ctx->gb)) {
330         av_log(avctx, AV_LOG_ERROR, "Invalid start bit!\n");
331         return AVERROR_INVALIDDATA;
332     }
333
334     while (get_bits_left(&ctx->gb) >= 2 &&
335            (ch_unit_id = get_bits(&ctx->gb, 2)) != CH_UNIT_TERMINATOR) {
336         if (ch_unit_id == CH_UNIT_EXTENSION) {
337             avpriv_report_missing_feature(avctx, "Channel unit extension");
338             return AVERROR_PATCHWELCOME;
339         }
340         if (ch_block >= ctx->num_channel_blocks ||
341             ctx->channel_blocks[ch_block] != ch_unit_id) {
342             av_log(avctx, AV_LOG_ERROR,
343                    "Frame data doesn't match channel configuration!\n");
344             return AVERROR_INVALIDDATA;
345         }
346
347         ctx->ch_units[ch_block].unit_type = ch_unit_id;
348         channels_to_process               = ch_unit_id + 1;
349
350         if ((ret = ff_atrac3p_decode_channel_unit(&ctx->gb,
351                                                   &ctx->ch_units[ch_block],
352                                                   channels_to_process,
353                                                   avctx)) < 0)
354             return ret;
355
356         decode_residual_spectrum(&ctx->ch_units[ch_block], ctx->samples,
357                                  channels_to_process, avctx);
358         reconstruct_frame(ctx, &ctx->ch_units[ch_block],
359                           channels_to_process, avctx);
360
361         for (i = 0; i < channels_to_process; i++)
362             memcpy(samples_p[out_ch_index + i], ctx->outp_buf[i],
363                    ATRAC3P_FRAME_SAMPLES * sizeof(**samples_p));
364
365         ch_block++;
366         out_ch_index += channels_to_process;
367     }
368
369     *got_frame_ptr = 1;
370
371     return avctx->block_align;
372 }
373
374 AVCodec ff_atrac3p_decoder = {
375     .name           = "atrac3plus",
376     .long_name      = NULL_IF_CONFIG_SMALL("ATRAC3+ (Adaptive TRansform Acoustic Coding 3+)"),
377     .type           = AVMEDIA_TYPE_AUDIO,
378     .id             = AV_CODEC_ID_ATRAC3P,
379     .priv_data_size = sizeof(ATRAC3PContext),
380     .init           = atrac3p_decode_init,
381     .close          = atrac3p_decode_close,
382     .decode         = atrac3p_decode_frame,
383 };