10157
|
1 /*
|
|
2 * Atrac 1 compatible decoder
|
|
3 * Copyright (c) 2009 Maxim Poliakovski
|
|
4 * Copyright (c) 2009 Benjamin Larsson
|
|
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 libavcodec/atrac1.c
|
|
25 * Atrac 1 compatible decoder.
|
|
26 * This decoder handles raw ATRAC1 data.
|
|
27 */
|
|
28
|
|
29 /* Many thanks to Tim Craig for all the help! */
|
|
30
|
|
31 #include <math.h>
|
|
32 #include <stddef.h>
|
|
33 #include <stdio.h>
|
|
34
|
|
35 #include "avcodec.h"
|
|
36 #include "get_bits.h"
|
|
37 #include "dsputil.h"
|
|
38
|
|
39 #include "atrac.h"
|
|
40 #include "atrac1data.h"
|
|
41
|
|
42 #define AT1_MAX_BFU 52 ///< max number of block floating units in a sound unit
|
|
43 #define AT1_SU_SIZE 212 ///< number of bytes in a sound unit
|
|
44 #define AT1_SU_SAMPLES 512 ///< number of samples in a sound unit
|
|
45 #define AT1_FRAME_SIZE AT1_SU_SIZE * 2
|
|
46 #define AT1_SU_MAX_BITS AT1_SU_SIZE * 8
|
|
47 #define AT1_MAX_CHANNELS 2
|
|
48
|
|
49 #define AT1_QMF_BANDS 3
|
|
50 #define IDX_LOW_BAND 0
|
|
51 #define IDX_MID_BAND 1
|
|
52 #define IDX_HIGH_BAND 2
|
|
53
|
|
54 /**
|
|
55 * Sound unit struct, one unit is used per channel
|
|
56 */
|
|
57 typedef struct {
|
|
58 int log2_block_count[AT1_QMF_BANDS]; ///< log2 number of blocks in a band
|
|
59 int num_bfus; ///< number of Block Floating Units
|
|
60 int idwls[AT1_MAX_BFU]; ///< the word length indexes for each BFU
|
|
61 int idsfs[AT1_MAX_BFU]; ///< the scalefactor indexes for each BFU
|
|
62 float* spectrum[2];
|
10197
|
63 DECLARE_ALIGNED_16(float, spec1[AT1_SU_SAMPLES]); ///< mdct buffer
|
|
64 DECLARE_ALIGNED_16(float, spec2[AT1_SU_SAMPLES]); ///< mdct buffer
|
|
65 DECLARE_ALIGNED_16(float, fst_qmf_delay[46]); ///< delay line for the 1st stacked QMF filter
|
|
66 DECLARE_ALIGNED_16(float, snd_qmf_delay[46]); ///< delay line for the 2nd stacked QMF filter
|
|
67 DECLARE_ALIGNED_16(float, last_qmf_delay[256+23]); ///< delay line for the last stacked QMF filter
|
10157
|
68 } AT1SUCtx;
|
|
69
|
|
70 /**
|
|
71 * The atrac1 context, holds all needed parameters for decoding
|
|
72 */
|
|
73 typedef struct {
|
|
74 AT1SUCtx SUs[AT1_MAX_CHANNELS]; ///< channel sound unit
|
10197
|
75 DECLARE_ALIGNED_16(float, spec[AT1_SU_SAMPLES]); ///< the mdct spectrum buffer
|
10185
|
76
|
10197
|
77 DECLARE_ALIGNED_16(float, low[256]);
|
|
78 DECLARE_ALIGNED_16(float, mid[256]);
|
|
79 DECLARE_ALIGNED_16(float, high[512]);
|
10157
|
80 float* bands[3];
|
10197
|
81 DECLARE_ALIGNED_16(float, out_samples[AT1_MAX_CHANNELS][AT1_SU_SAMPLES]);
|
10157
|
82 MDCTContext mdct_ctx[3];
|
|
83 int channels;
|
|
84 DSPContext dsp;
|
|
85 } AT1Ctx;
|
|
86
|
10185
|
87 DECLARE_ALIGNED_16(static float, short_window[32]);
|
10157
|
88
|
|
89 /** size of the transform in samples in the long mode for each QMF band */
|
|
90 static const uint16_t samples_per_band[3] = {128, 128, 256};
|
|
91 static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
|
|
92
|
|
93
|
10170
|
94 static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
|
|
95 int rev_spec)
|
10157
|
96 {
|
|
97 MDCTContext* mdct_context;
|
|
98 int transf_size = 1 << nbits;
|
|
99
|
10197
|
100 mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];
|
10157
|
101
|
|
102 if (rev_spec) {
|
|
103 int i;
|
10197
|
104 for (i = 0; i < transf_size / 2; i++)
|
10170
|
105 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
|
10157
|
106 }
|
10170
|
107 ff_imdct_half(mdct_context, out, spec);
|
10157
|
108 }
|
|
109
|
|
110
|
|
111 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
|
|
112 {
|
10197
|
113 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
|
|
114 unsigned int start_pos, ref_pos = 0 pos = 0;
|
10157
|
115
|
10197
|
116 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
|
10157
|
117 band_samples = samples_per_band[band_num];
|
|
118 log2_block_count = su->log2_block_count[band_num];
|
|
119
|
|
120 /* number of mdct blocks in the current QMF band: 1 - for long mode */
|
|
121 /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
|
|
122 num_blocks = 1 << log2_block_count;
|
|
123
|
|
124 /* mdct block size in samples: 128 (long mode, low & mid bands), */
|
|
125 /* 256 (long mode, high band) and 32 (short mode, all bands) */
|
|
126 block_size = band_samples >> log2_block_count;
|
|
127
|
|
128 /* calc transform size in bits according to the block_size_mode */
|
|
129 nbits = mdct_long_nbits[band_num] - log2_block_count;
|
|
130
|
10197
|
131 if (nbits != 5 && nbits != 7 && nbits != 8)
|
10157
|
132 return -1;
|
|
133
|
|
134 if (num_blocks == 1) {
|
10189
|
135 /* long blocks */
|
10157
|
136 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos], nbits, band_num);
|
|
137 pos += block_size; // move to the next mdct block in the spectrum
|
10185
|
138
|
|
139 /* overlap and window long blocks */
|
10197
|
140 q->dsp.vector_fmul_window(q->bands[band_num], &su->spectrum[1][ref_pos + band_samples - 16],
|
|
141 &su->spectrum[0][ref_pos], short_window, 0, 16);
|
|
142 memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float));
|
10157
|
143 } else {
|
10189
|
144 /* short blocks */
|
10185
|
145 float *prev_buf;
|
10189
|
146 start_pos = 0;
|
10197
|
147 prev_buf = &su->spectrum[1][ref_pos + band_samples - 16];
|
|
148 for (; num_blocks != 0; num_blocks--) {
|
|
149 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos + start_pos], 5, band_num);
|
10157
|
150
|
|
151 /* overlap and window between short blocks */
|
10189
|
152 q->dsp.vector_fmul_window(&q->bands[band_num][start_pos], prev_buf,
|
10197
|
153 &su->spectrum[0][ref_pos + start_pos], short_window, 0, 16);
|
10185
|
154
|
10197
|
155 prev_buf = &su->spectrum[0][ref_pos+start_pos + 16];
|
10157
|
156 start_pos += 32; // use hardcoded block_size
|
|
157 pos += 32;
|
|
158 }
|
|
159 }
|
|
160 ref_pos += band_samples;
|
|
161 }
|
|
162
|
|
163 /* Swap buffers so the mdct overlap works */
|
|
164 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
|
|
165
|
|
166 return 0;
|
|
167 }
|
|
168
|
10170
|
169 /**
|
|
170 * Parse the block size mode byte
|
|
171 */
|
10157
|
172
|
10170
|
173 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
|
10157
|
174 {
|
|
175 int log2_block_count_tmp, i;
|
|
176
|
10197
|
177 for (i = 0; i < 2; i++) {
|
10157
|
178 /* low and mid band */
|
|
179 log2_block_count_tmp = get_bits(gb, 2);
|
|
180 if (log2_block_count_tmp & 1)
|
|
181 return -1;
|
10170
|
182 log2_block_cnt[i] = 2 - log2_block_count_tmp;
|
10157
|
183 }
|
|
184
|
|
185 /* high band */
|
|
186 log2_block_count_tmp = get_bits(gb, 2);
|
|
187 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
|
|
188 return -1;
|
10170
|
189 log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
|
10157
|
190
|
|
191 skip_bits(gb, 2);
|
|
192 return 0;
|
|
193 }
|
|
194
|
|
195
|
10170
|
196 static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
|
|
197 float spec[AT1_SU_SAMPLES])
|
10157
|
198 {
|
|
199 int bits_used, band_num, bfu_num, i;
|
|
200
|
|
201 /* parse the info byte (2nd byte) telling how much BFUs were coded */
|
|
202 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
|
|
203
|
|
204 /* calc number of consumed bits:
|
|
205 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
|
|
206 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
|
|
207 bits_used = su->num_bfus * 10 + 32 +
|
|
208 bfu_amount_tab2[get_bits(gb, 2)] +
|
|
209 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
|
|
210
|
|
211 /* get word length index (idwl) for each BFU */
|
10197
|
212 for (i = 0; i < su->num_bfus; i++)
|
10157
|
213 su->idwls[i] = get_bits(gb, 4);
|
|
214
|
|
215 /* get scalefactor index (idsf) for each BFU */
|
10197
|
216 for (i = 0; i < su->num_bfus; i++)
|
10157
|
217 su->idsfs[i] = get_bits(gb, 6);
|
|
218
|
|
219 /* zero idwl/idsf for empty BFUs */
|
|
220 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
|
|
221 su->idwls[i] = su->idsfs[i] = 0;
|
|
222
|
|
223 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
|
10197
|
224 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
|
|
225 for (bfu_num = bfu_bands_t[band_num]; bfu_num < bfu_bands_t[band_num+1]; bfu_num++) {
|
10157
|
226 int pos;
|
|
227
|
|
228 int num_specs = specs_per_bfu[bfu_num];
|
|
229 int word_len = !!su->idwls[bfu_num] + su->idwls[bfu_num];
|
|
230 float scale_factor = sf_table[su->idsfs[bfu_num]];
|
|
231 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
|
|
232
|
|
233 /* check for bitstream overflow */
|
|
234 if (bits_used > AT1_SU_MAX_BITS)
|
|
235 return -1;
|
|
236
|
|
237 /* get the position of the 1st spec according to the block size mode */
|
|
238 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
|
|
239
|
|
240 if (word_len) {
|
10170
|
241 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
|
10157
|
242
|
10197
|
243 for (i = 0; i < num_specs; i++) {
|
10157
|
244 /* read in a quantized spec and convert it to
|
|
245 * signed int and then inverse quantization
|
|
246 */
|
|
247 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
|
|
248 }
|
|
249 } else { /* word_len = 0 -> empty BFU, zero all specs in the emty BFU */
|
10197
|
250 memset(&spec[pos], 0, num_specs * sizeof(float));
|
10157
|
251 }
|
|
252 }
|
|
253 }
|
|
254
|
|
255 return 0;
|
|
256 }
|
|
257
|
|
258
|
|
259 void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
|
|
260 {
|
10197
|
261 float temp[256];
|
|
262 float iqmf_temp[512 + 46];
|
10157
|
263
|
|
264 /* combine low and middle bands */
|
|
265 atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
|
|
266
|
|
267 /* delay the signal of the high band by 23 samples */
|
10197
|
268 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 23);
|
|
269 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float) * 256);
|
10157
|
270
|
|
271 /* combine (low + middle) and high bands */
|
|
272 atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
|
|
273 }
|
|
274
|
|
275
|
10170
|
276 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
|
|
277 int *data_size, AVPacket *avpkt)
|
10157
|
278 {
|
|
279 const uint8_t *buf = avpkt->data;
|
10170
|
280 int buf_size = avpkt->size;
|
|
281 AT1Ctx *q = avctx->priv_data;
|
10157
|
282 int ch, ret, i;
|
|
283 GetBitContext gb;
|
|
284 float* samples = data;
|
|
285
|
|
286
|
|
287 if (buf_size < 212 * q->channels) {
|
|
288 av_log(q,AV_LOG_ERROR,"Not enought data to decode!\n");
|
|
289 return -1;
|
|
290 }
|
|
291
|
10197
|
292 for (ch = 0; ch < q->channels; ch++) {
|
10157
|
293 AT1SUCtx* su = &q->SUs[ch];
|
|
294
|
10197
|
295 init_get_bits(&gb, &buf[212 * ch], 212 * 8);
|
10157
|
296
|
|
297 /* parse block_size_mode, 1st byte */
|
10170
|
298 ret = at1_parse_bsm(&gb, su->log2_block_count);
|
10157
|
299 if (ret < 0)
|
|
300 return ret;
|
|
301
|
|
302 ret = at1_unpack_dequant(&gb, su, q->spec);
|
|
303 if (ret < 0)
|
|
304 return ret;
|
|
305
|
|
306 ret = at1_imdct_block(su, q);
|
|
307 if (ret < 0)
|
|
308 return ret;
|
|
309 at1_subband_synthesis(q, su, q->out_samples[ch]);
|
|
310 }
|
|
311
|
|
312 /* round, convert to 16bit and interleave */
|
|
313 if (q->channels == 1) {
|
|
314 /* mono */
|
10197
|
315 q->dsp.vector_clipf(samples, q->out_samples[0], -32700.0 / (1 << 15),
|
|
316 32700.0 / (1 << 15), AT1_SU_SAMPLES);
|
10157
|
317 } else {
|
|
318 /* stereo */
|
|
319 for (i = 0; i < AT1_SU_SAMPLES; i++) {
|
10197
|
320 samples[i * 2] = av_clipf(q->out_samples[0][i],
|
|
321 -32700.0 / (1 << 15),
|
|
322 32700.0 / (1 << 15));
|
|
323 samples[i * 2 + 1] = av_clipf(q->out_samples[1][i],
|
|
324 -32700.0 / (1 << 15),
|
|
325 32700.0 / (1 << 15));
|
10157
|
326 }
|
|
327 }
|
|
328
|
|
329 *data_size = q->channels * AT1_SU_SAMPLES * sizeof(*samples);
|
|
330 return avctx->block_align;
|
|
331 }
|
|
332
|
|
333
|
|
334 static av_cold int atrac1_decode_init(AVCodecContext *avctx)
|
|
335 {
|
|
336 AT1Ctx *q = avctx->priv_data;
|
|
337
|
|
338 avctx->sample_fmt = SAMPLE_FMT_FLT;
|
|
339
|
|
340 q->channels = avctx->channels;
|
|
341
|
|
342 /* Init the mdct transforms */
|
10197
|
343 ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15));
|
|
344 ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15));
|
|
345 ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15));
|
10185
|
346
|
|
347 ff_sine_window_init(short_window, 32);
|
10157
|
348
|
|
349 atrac_generate_tables();
|
|
350
|
|
351 dsputil_init(&q->dsp, avctx);
|
|
352
|
|
353 q->bands[0] = q->low;
|
|
354 q->bands[1] = q->mid;
|
|
355 q->bands[2] = q->high;
|
|
356
|
|
357 /* Prepare the mdct overlap buffers */
|
|
358 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
|
|
359 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
|
|
360 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
|
|
361 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
|
|
362
|
|
363 return 0;
|
|
364 }
|
|
365
|
|
366 AVCodec atrac1_decoder = {
|
|
367 .name = "atrac1",
|
|
368 .type = CODEC_TYPE_AUDIO,
|
|
369 .id = CODEC_ID_ATRAC1,
|
|
370 .priv_data_size = sizeof(AT1Ctx),
|
|
371 .init = atrac1_decode_init,
|
|
372 .close = NULL,
|
|
373 .decode = atrac1_decode_frame,
|
|
374 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),
|
|
375 };
|