Initial commit of the atrac1 decoder, not hooked up yet
[libav.git] / libavcodec / atrac1.c
CommitLineData
dbb0f96f
BL
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 */
57typedef 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 */
73typedef 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
87static float *short_window;
88static float *mid_window;
89DECLARE_ALIGNED_16(static float, long_window[256]);
90static float *window_per_band[3];
91
92/** size of the transform in samples in the long mode for each QMF band */
93static const uint16_t samples_per_band[3] = {128, 128, 256};
94static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
95
96
97static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits, int rev_spec)
98{
99 MDCTContext* mdct_context;
100 int transf_size = 1 << nbits;
101
102 mdct_context = &q->mdct_ctx[nbits - 5 - (nbits>6)];
103
104 if (rev_spec) {
105 int i;
106 for (i=0 ; i<transf_size/2 ; i++)
107 FFSWAP(float, spec[i], spec[transf_size-1-i]);
108 }
109 ff_imdct_half(mdct_context,out,spec);
110}
111
112
113static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
114{
115 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
116 unsigned int start_pos, ref_pos=0, pos = 0;
117
118 for (band_num=0 ; band_num<AT1_QMF_BANDS ; band_num++) {
119 band_samples = samples_per_band[band_num];
120 log2_block_count = su->log2_block_count[band_num];
121
122 /* number of mdct blocks in the current QMF band: 1 - for long mode */
123 /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
124 num_blocks = 1 << log2_block_count;
125
126 /* mdct block size in samples: 128 (long mode, low & mid bands), */
127 /* 256 (long mode, high band) and 32 (short mode, all bands) */
128 block_size = band_samples >> log2_block_count;
129
130 /* calc transform size in bits according to the block_size_mode */
131 nbits = mdct_long_nbits[band_num] - log2_block_count;
132
133 if (nbits!=5 && nbits!=7 && nbits!=8)
134 return -1;
135
136 if (num_blocks == 1) {
137 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos], nbits, band_num);
138 pos += block_size; // move to the next mdct block in the spectrum
139 } else {
140 /* calc start position for the 1st short block: 96(128) or 112(256) */
141 start_pos = (band_samples * (num_blocks - 1)) >> (log2_block_count + 1);
142 memset(&su->spectrum[0][ref_pos], 0, sizeof(float) * (band_samples * 2));
143
144 for (; num_blocks!=0 ; num_blocks--) {
145 /* use hardcoded nbits for the short mode */
146 at1_imdct(q, &q->spec[pos], q->short_buf, 5, band_num);
147
148 /* overlap and window between short blocks */
149 q->dsp.vector_fmul_window(&su->spectrum[0][ref_pos+start_pos],
150 &su->spectrum[0][ref_pos+start_pos],q->short_buf,short_window, 0, 16);
151 start_pos += 32; // use hardcoded block_size
152 pos += 32;
153 }
154 }
155
156 /* overlap and window with the previous frame and output the result */
157 q->dsp.vector_fmul_window(q->bands[band_num], &su->spectrum[1][ref_pos+band_samples/2],
158 &su->spectrum[0][ref_pos], window_per_band[band_num], 0, band_samples/2);
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
169
170static int at1_parse_block_size_mode(GetBitContext* gb, int log2_block_count[AT1_QMF_BANDS])
171{
172 int log2_block_count_tmp, i;
173
174 for(i=0 ; i<2 ; i++) {
175 /* low and mid band */
176 log2_block_count_tmp = get_bits(gb, 2);
177 if (log2_block_count_tmp & 1)
178 return -1;
179 log2_block_count[i] = 2 - log2_block_count_tmp;
180 }
181
182 /* high band */
183 log2_block_count_tmp = get_bits(gb, 2);
184 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
185 return -1;
186 log2_block_count[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
187
188 skip_bits(gb, 2);
189 return 0;
190}
191
192
193static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su, float spec[AT1_SU_SAMPLES])
194{
195 int bits_used, band_num, bfu_num, i;
196
197 /* parse the info byte (2nd byte) telling how much BFUs were coded */
198 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
199
200 /* calc number of consumed bits:
201 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
202 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
203 bits_used = su->num_bfus * 10 + 32 +
204 bfu_amount_tab2[get_bits(gb, 2)] +
205 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
206
207 /* get word length index (idwl) for each BFU */
208 for (i=0 ; i<su->num_bfus ; i++)
209 su->idwls[i] = get_bits(gb, 4);
210
211 /* get scalefactor index (idsf) for each BFU */
212 for (i=0 ; i<su->num_bfus ; i++)
213 su->idsfs[i] = get_bits(gb, 6);
214
215 /* zero idwl/idsf for empty BFUs */
216 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
217 su->idwls[i] = su->idsfs[i] = 0;
218
219 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
220 for (band_num=0 ; band_num<AT1_QMF_BANDS ; band_num++) {
221 for (bfu_num=bfu_bands_t[band_num] ; bfu_num<bfu_bands_t[band_num+1] ; bfu_num++) {
222 int pos;
223
224 int num_specs = specs_per_bfu[bfu_num];
225 int word_len = !!su->idwls[bfu_num] + su->idwls[bfu_num];
226 float scale_factor = sf_table[su->idsfs[bfu_num]];
227 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
228
229 /* check for bitstream overflow */
230 if (bits_used > AT1_SU_MAX_BITS)
231 return -1;
232
233 /* get the position of the 1st spec according to the block size mode */
234 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
235
236 if (word_len) {
237 float max_quant = 1.0/(float)((1 << (word_len - 1)) - 1);
238
239 for (i=0 ; i<num_specs ; i++) {
240 /* read in a quantized spec and convert it to
241 * signed int and then inverse quantization
242 */
243 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
244 }
245 } else { /* word_len = 0 -> empty BFU, zero all specs in the emty BFU */
246 memset(&spec[pos], 0, num_specs*sizeof(float));
247 }
248 }
249 }
250
251 return 0;
252}
253
254
255void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
256{
257 float temp[256];
258 float iqmf_temp[512 + 46];
259
260 /* combine low and middle bands */
261 atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
262
263 /* delay the signal of the high band by 23 samples */
264 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float)*23);
265 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float)*256);
266
267 /* combine (low + middle) and high bands */
268 atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
269}
270
271
272static int atrac1_decode_frame(AVCodecContext *avctx,
273 void *data, int *data_size,
274 AVPacket *avpkt)
275{
276 const uint8_t *buf = avpkt->data;
277 int buf_size = avpkt->size;
278 AT1Ctx *q = avctx->priv_data;
279 int ch, ret, i;
280 GetBitContext gb;
281 float* samples = data;
282
283
284 if (buf_size < 212 * q->channels) {
285 av_log(q,AV_LOG_ERROR,"Not enought data to decode!\n");
286 return -1;
287 }
288
289 for (ch=0 ; ch<q->channels ; ch++) {
290 AT1SUCtx* su = &q->SUs[ch];
291
292 init_get_bits(&gb, &buf[212*ch], 212*8);
293
294 /* parse block_size_mode, 1st byte */
295 ret = at1_parse_block_size_mode(&gb, su->log2_block_count);
296 if (ret < 0)
297 return ret;
298
299 ret = at1_unpack_dequant(&gb, su, q->spec);
300 if (ret < 0)
301 return ret;
302
303 ret = at1_imdct_block(su, q);
304 if (ret < 0)
305 return ret;
306 at1_subband_synthesis(q, su, q->out_samples[ch]);
307 }
308
309 /* round, convert to 16bit and interleave */
310 if (q->channels == 1) {
311 /* mono */
312 q->dsp.vector_clipf(samples, q->out_samples[0], -32700./(1<<15), 32700./(1<<15), AT1_SU_SAMPLES);
313 } else {
314 /* stereo */
315 for (i = 0; i < AT1_SU_SAMPLES; i++) {
316 samples[i*2] = av_clipf(q->out_samples[0][i], -32700./(1<<15), 32700./(1<<15));
317 samples[i*2+1] = av_clipf(q->out_samples[1][i], -32700./(1<<15), 32700./(1<<15));
318 }
319 }
320
321 *data_size = q->channels * AT1_SU_SAMPLES * sizeof(*samples);
322 return avctx->block_align;
323}
324
325
326static av_cold void init_mdct_windows(void)
327{
328 int i;
329
330 /** The mid and long windows uses the same sine window splitted
331 * in the middle and wrapped into zero/one regions as follows:
332 *
333 * region of "ones"
334 * -------------
335 * /
336 * / 1st half
337 * / of the sine
338 * / window
339 * ---------/
340 * zero region
341 *
342 * The mid and short windows are subsets of the long window.
343 */
344
345 /* Build "zero" region */
346 memset(long_window, 0, sizeof(long_window));
347 /* Build sine window region */
348 short_window = &long_window[112];
349 ff_sine_window_init(short_window,32);
350 /* Build "ones" region */
351 for (i = 0; i < 112; i++)
352 long_window[144 + i] = 1.0f;
353 /* Save the mid window subset start */
354 mid_window = &long_window[64];
355
356 /* Prepare the window table */
357 window_per_band[0] = mid_window;
358 window_per_band[1] = mid_window;
359 window_per_band[2] = long_window;
360}
361
362static av_cold int atrac1_decode_init(AVCodecContext *avctx)
363{
364 AT1Ctx *q = avctx->priv_data;
365
366 avctx->sample_fmt = SAMPLE_FMT_FLT;
367
368 q->channels = avctx->channels;
369
370 /* Init the mdct transforms */
371 ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1<<15));
372 ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1<<15));
373 ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1<<15));
374 init_mdct_windows();
375
376 atrac_generate_tables();
377
378 dsputil_init(&q->dsp, avctx);
379
380 q->bands[0] = q->low;
381 q->bands[1] = q->mid;
382 q->bands[2] = q->high;
383
384 /* Prepare the mdct overlap buffers */
385 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
386 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
387 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
388 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
389
390 return 0;
391}
392
393AVCodec atrac1_decoder = {
394 .name = "atrac1",
395 .type = CODEC_TYPE_AUDIO,
396 .id = CODEC_ID_ATRAC1,
397 .priv_data_size = sizeof(AT1Ctx),
398 .init = atrac1_decode_init,
399 .close = NULL,
400 .decode = atrac1_decode_frame,
401 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),
402};