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];
|
|
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
|
|
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
|
|
75 DECLARE_ALIGNED_16(float,spec[AT1_SU_SAMPLES]); ///< the mdct spectrum buffer
|
|
76 DECLARE_ALIGNED_16(float,short_buf[64]); ///< buffer for the short mode
|
|
77 DECLARE_ALIGNED_16(float, low[256]);
|
|
78 DECLARE_ALIGNED_16(float, mid[256]);
|
|
79 DECLARE_ALIGNED_16(float,high[512]);
|
|
80 float* bands[3];
|
|
81 float out_samples[AT1_MAX_CHANNELS][AT1_SU_SAMPLES];
|
|
82 MDCTContext mdct_ctx[3];
|
|
83 int channels;
|
|
84 DSPContext dsp;
|
|
85 } AT1Ctx;
|
|
86
|
|
87 static float *short_window;
|
|
88 static float *mid_window;
|
|
89 DECLARE_ALIGNED_16(static float, long_window[256]);
|
|
90 static float *window_per_band[3];
|
|
91
|
|
92 /** size of the transform in samples in the long mode for each QMF band */
|
|
93 static const uint16_t samples_per_band[3] = {128, 128, 256};
|
|
94 static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
|
|
95
|
|
96
|
10170
|
97 static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
|
|
98 int rev_spec)
|
10157
|
99 {
|
|
100 MDCTContext* mdct_context;
|
|
101 int transf_size = 1 << nbits;
|
|
102
|
|
103 mdct_context = &q->mdct_ctx[nbits - 5 - (nbits>6)];
|
|
104
|
|
105 if (rev_spec) {
|
|
106 int i;
|
|
107 for (i=0 ; i<transf_size/2 ; i++)
|
10170
|
108 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
|
10157
|
109 }
|
10170
|
110 ff_imdct_half(mdct_context, out, spec);
|
10157
|
111 }
|
|
112
|
|
113
|
|
114 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
|
|
115 {
|
|
116 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
|
|
117 unsigned int start_pos, ref_pos=0, pos = 0;
|
|
118
|
|
119 for (band_num=0 ; band_num<AT1_QMF_BANDS ; band_num++) {
|
|
120 band_samples = samples_per_band[band_num];
|
|
121 log2_block_count = su->log2_block_count[band_num];
|
|
122
|
|
123 /* number of mdct blocks in the current QMF band: 1 - for long mode */
|
|
124 /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
|
|
125 num_blocks = 1 << log2_block_count;
|
|
126
|
|
127 /* mdct block size in samples: 128 (long mode, low & mid bands), */
|
|
128 /* 256 (long mode, high band) and 32 (short mode, all bands) */
|
|
129 block_size = band_samples >> log2_block_count;
|
|
130
|
|
131 /* calc transform size in bits according to the block_size_mode */
|
|
132 nbits = mdct_long_nbits[band_num] - log2_block_count;
|
|
133
|
|
134 if (nbits!=5 && nbits!=7 && nbits!=8)
|
|
135 return -1;
|
|
136
|
|
137 if (num_blocks == 1) {
|
|
138 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos], nbits, band_num);
|
|
139 pos += block_size; // move to the next mdct block in the spectrum
|
|
140 } else {
|
|
141 /* calc start position for the 1st short block: 96(128) or 112(256) */
|
|
142 start_pos = (band_samples * (num_blocks - 1)) >> (log2_block_count + 1);
|
|
143 memset(&su->spectrum[0][ref_pos], 0, sizeof(float) * (band_samples * 2));
|
|
144
|
|
145 for (; num_blocks!=0 ; num_blocks--) {
|
|
146 /* use hardcoded nbits for the short mode */
|
|
147 at1_imdct(q, &q->spec[pos], q->short_buf, 5, band_num);
|
|
148
|
|
149 /* overlap and window between short blocks */
|
|
150 q->dsp.vector_fmul_window(&su->spectrum[0][ref_pos+start_pos],
|
10170
|
151 &su->spectrum[0][ref_pos+start_pos],
|
|
152 q->short_buf,short_window, 0, 16);
|
10157
|
153 start_pos += 32; // use hardcoded block_size
|
|
154 pos += 32;
|
|
155 }
|
|
156 }
|
|
157
|
|
158 /* overlap and window with the previous frame and output the result */
|
|
159 q->dsp.vector_fmul_window(q->bands[band_num], &su->spectrum[1][ref_pos+band_samples/2],
|
|
160 &su->spectrum[0][ref_pos], window_per_band[band_num], 0, band_samples/2);
|
|
161
|
|
162 ref_pos += band_samples;
|
|
163 }
|
|
164
|
|
165 /* Swap buffers so the mdct overlap works */
|
|
166 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
|
|
167
|
|
168 return 0;
|
|
169 }
|
|
170
|
10170
|
171 /**
|
|
172 * Parse the block size mode byte
|
|
173 */
|
10157
|
174
|
10170
|
175 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
|
10157
|
176 {
|
|
177 int log2_block_count_tmp, i;
|
|
178
|
|
179 for(i=0 ; i<2 ; i++) {
|
|
180 /* low and mid band */
|
|
181 log2_block_count_tmp = get_bits(gb, 2);
|
|
182 if (log2_block_count_tmp & 1)
|
|
183 return -1;
|
10170
|
184 log2_block_cnt[i] = 2 - log2_block_count_tmp;
|
10157
|
185 }
|
|
186
|
|
187 /* high band */
|
|
188 log2_block_count_tmp = get_bits(gb, 2);
|
|
189 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
|
|
190 return -1;
|
10170
|
191 log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
|
10157
|
192
|
|
193 skip_bits(gb, 2);
|
|
194 return 0;
|
|
195 }
|
|
196
|
|
197
|
10170
|
198 static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
|
|
199 float spec[AT1_SU_SAMPLES])
|
10157
|
200 {
|
|
201 int bits_used, band_num, bfu_num, i;
|
|
202
|
|
203 /* parse the info byte (2nd byte) telling how much BFUs were coded */
|
|
204 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
|
|
205
|
|
206 /* calc number of consumed bits:
|
|
207 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
|
|
208 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
|
|
209 bits_used = su->num_bfus * 10 + 32 +
|
|
210 bfu_amount_tab2[get_bits(gb, 2)] +
|
|
211 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
|
|
212
|
|
213 /* get word length index (idwl) for each BFU */
|
|
214 for (i=0 ; i<su->num_bfus ; i++)
|
|
215 su->idwls[i] = get_bits(gb, 4);
|
|
216
|
|
217 /* get scalefactor index (idsf) for each BFU */
|
|
218 for (i=0 ; i<su->num_bfus ; i++)
|
|
219 su->idsfs[i] = get_bits(gb, 6);
|
|
220
|
|
221 /* zero idwl/idsf for empty BFUs */
|
|
222 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
|
|
223 su->idwls[i] = su->idsfs[i] = 0;
|
|
224
|
|
225 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
|
|
226 for (band_num=0 ; band_num<AT1_QMF_BANDS ; band_num++) {
|
|
227 for (bfu_num=bfu_bands_t[band_num] ; bfu_num<bfu_bands_t[band_num+1] ; bfu_num++) {
|
|
228 int pos;
|
|
229
|
|
230 int num_specs = specs_per_bfu[bfu_num];
|
|
231 int word_len = !!su->idwls[bfu_num] + su->idwls[bfu_num];
|
|
232 float scale_factor = sf_table[su->idsfs[bfu_num]];
|
|
233 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
|
|
234
|
|
235 /* check for bitstream overflow */
|
|
236 if (bits_used > AT1_SU_MAX_BITS)
|
|
237 return -1;
|
|
238
|
|
239 /* get the position of the 1st spec according to the block size mode */
|
|
240 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
|
|
241
|
|
242 if (word_len) {
|
10170
|
243 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
|
10157
|
244
|
|
245 for (i=0 ; i<num_specs ; i++) {
|
|
246 /* read in a quantized spec and convert it to
|
|
247 * signed int and then inverse quantization
|
|
248 */
|
|
249 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
|
|
250 }
|
|
251 } else { /* word_len = 0 -> empty BFU, zero all specs in the emty BFU */
|
|
252 memset(&spec[pos], 0, num_specs*sizeof(float));
|
|
253 }
|
|
254 }
|
|
255 }
|
|
256
|
|
257 return 0;
|
|
258 }
|
|
259
|
|
260
|
|
261 void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
|
|
262 {
|
|
263 float temp[256];
|
|
264 float iqmf_temp[512 + 46];
|
|
265
|
|
266 /* combine low and middle bands */
|
|
267 atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
|
|
268
|
|
269 /* delay the signal of the high band by 23 samples */
|
10170
|
270 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float)*23);
|
|
271 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float)*256);
|
10157
|
272
|
|
273 /* combine (low + middle) and high bands */
|
|
274 atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
|
|
275 }
|
|
276
|
|
277
|
10170
|
278 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
|
|
279 int *data_size, AVPacket *avpkt)
|
10157
|
280 {
|
|
281 const uint8_t *buf = avpkt->data;
|
10170
|
282 int buf_size = avpkt->size;
|
|
283 AT1Ctx *q = avctx->priv_data;
|
10157
|
284 int ch, ret, i;
|
|
285 GetBitContext gb;
|
|
286 float* samples = data;
|
|
287
|
|
288
|
|
289 if (buf_size < 212 * q->channels) {
|
|
290 av_log(q,AV_LOG_ERROR,"Not enought data to decode!\n");
|
|
291 return -1;
|
|
292 }
|
|
293
|
|
294 for (ch=0 ; ch<q->channels ; ch++) {
|
|
295 AT1SUCtx* su = &q->SUs[ch];
|
|
296
|
|
297 init_get_bits(&gb, &buf[212*ch], 212*8);
|
|
298
|
|
299 /* parse block_size_mode, 1st byte */
|
10170
|
300 ret = at1_parse_bsm(&gb, su->log2_block_count);
|
10157
|
301 if (ret < 0)
|
|
302 return ret;
|
|
303
|
|
304 ret = at1_unpack_dequant(&gb, su, q->spec);
|
|
305 if (ret < 0)
|
|
306 return ret;
|
|
307
|
|
308 ret = at1_imdct_block(su, q);
|
|
309 if (ret < 0)
|
|
310 return ret;
|
|
311 at1_subband_synthesis(q, su, q->out_samples[ch]);
|
|
312 }
|
|
313
|
|
314 /* round, convert to 16bit and interleave */
|
|
315 if (q->channels == 1) {
|
|
316 /* mono */
|
10170
|
317 q->dsp.vector_clipf(samples, q->out_samples[0], -32700.0 / (1<<15),
|
|
318 32700.0 / (1<<15), AT1_SU_SAMPLES);
|
10157
|
319 } else {
|
|
320 /* stereo */
|
|
321 for (i = 0; i < AT1_SU_SAMPLES; i++) {
|
10170
|
322 samples[i*2] = av_clipf(q->out_samples[0][i], -32700.0 / (1<<15),
|
|
323 32700.0 / (1<<15));
|
|
324 samples[i*2+1] = av_clipf(q->out_samples[1][i], -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 void init_mdct_windows(void)
|
|
335 {
|
|
336 int i;
|
|
337
|
|
338 /** The mid and long windows uses the same sine window splitted
|
|
339 * in the middle and wrapped into zero/one regions as follows:
|
|
340 *
|
|
341 * region of "ones"
|
|
342 * -------------
|
|
343 * /
|
|
344 * / 1st half
|
|
345 * / of the sine
|
|
346 * / window
|
|
347 * ---------/
|
|
348 * zero region
|
|
349 *
|
|
350 * The mid and short windows are subsets of the long window.
|
|
351 */
|
|
352
|
|
353 /* Build "zero" region */
|
|
354 memset(long_window, 0, sizeof(long_window));
|
|
355 /* Build sine window region */
|
|
356 short_window = &long_window[112];
|
|
357 ff_sine_window_init(short_window,32);
|
|
358 /* Build "ones" region */
|
|
359 for (i = 0; i < 112; i++)
|
|
360 long_window[144 + i] = 1.0f;
|
|
361 /* Save the mid window subset start */
|
|
362 mid_window = &long_window[64];
|
|
363
|
|
364 /* Prepare the window table */
|
|
365 window_per_band[0] = mid_window;
|
|
366 window_per_band[1] = mid_window;
|
|
367 window_per_band[2] = long_window;
|
|
368 }
|
|
369
|
|
370 static av_cold int atrac1_decode_init(AVCodecContext *avctx)
|
|
371 {
|
|
372 AT1Ctx *q = avctx->priv_data;
|
|
373
|
|
374 avctx->sample_fmt = SAMPLE_FMT_FLT;
|
|
375
|
|
376 q->channels = avctx->channels;
|
|
377
|
|
378 /* Init the mdct transforms */
|
|
379 ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1<<15));
|
|
380 ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1<<15));
|
|
381 ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1<<15));
|
|
382 init_mdct_windows();
|
|
383
|
|
384 atrac_generate_tables();
|
|
385
|
|
386 dsputil_init(&q->dsp, avctx);
|
|
387
|
|
388 q->bands[0] = q->low;
|
|
389 q->bands[1] = q->mid;
|
|
390 q->bands[2] = q->high;
|
|
391
|
|
392 /* Prepare the mdct overlap buffers */
|
|
393 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
|
|
394 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
|
|
395 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
|
|
396 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
|
|
397
|
|
398 return 0;
|
|
399 }
|
|
400
|
|
401 AVCodec atrac1_decoder = {
|
|
402 .name = "atrac1",
|
|
403 .type = CODEC_TYPE_AUDIO,
|
|
404 .id = CODEC_ID_ATRAC1,
|
|
405 .priv_data_size = sizeof(AT1Ctx),
|
|
406 .init = atrac1_decode_init,
|
|
407 .close = NULL,
|
|
408 .decode = atrac1_decode_frame,
|
|
409 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),
|
|
410 };
|