Increase alignment of av_malloc() as needed by AVX ASM.
[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 *
2912e87a 6 * This file is part of Libav.
dbb0f96f 7 *
2912e87a 8 * Libav is free software; you can redistribute it and/or
dbb0f96f
BL
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 *
2912e87a 13 * Libav is distributed in the hope that it will be useful,
dbb0f96f
BL
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
2912e87a 19 * License along with Libav; if not, write to the Free Software
dbb0f96f
BL
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23/**
ba87f080 24 * @file
dbb0f96f 25 * Atrac 1 compatible decoder.
e704b012 26 * This decoder handles raw ATRAC1 data and probably SDDS data.
dbb0f96f
BL
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"
1429224b 38#include "fft.h"
4538729a 39#include "sinewin.h"
dbb0f96f
BL
40
41#include "atrac.h"
42#include "atrac1data.h"
43
44#define AT1_MAX_BFU 52 ///< max number of block floating units in a sound unit
45#define AT1_SU_SIZE 212 ///< number of bytes in a sound unit
46#define AT1_SU_SAMPLES 512 ///< number of samples in a sound unit
47#define AT1_FRAME_SIZE AT1_SU_SIZE * 2
48#define AT1_SU_MAX_BITS AT1_SU_SIZE * 8
49#define AT1_MAX_CHANNELS 2
50
51#define AT1_QMF_BANDS 3
52#define IDX_LOW_BAND 0
53#define IDX_MID_BAND 1
54#define IDX_HIGH_BAND 2
55
56/**
57 * Sound unit struct, one unit is used per channel
58 */
59typedef struct {
60 int log2_block_count[AT1_QMF_BANDS]; ///< log2 number of blocks in a band
61 int num_bfus; ///< number of Block Floating Units
dbb0f96f 62 float* spectrum[2];
84dc2d8a
MR
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
dbb0f96f
BL
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
84dc2d8a 75 DECLARE_ALIGNED(16, float, spec)[AT1_SU_SAMPLES]; ///< the mdct spectrum buffer
10634c03 76
84dc2d8a
MR
77 DECLARE_ALIGNED(16, float, low)[256];
78 DECLARE_ALIGNED(16, float, mid)[256];
79 DECLARE_ALIGNED(16, float, high)[512];
dbb0f96f 80 float* bands[3];
84dc2d8a 81 DECLARE_ALIGNED(16, float, out_samples)[AT1_MAX_CHANNELS][AT1_SU_SAMPLES];
01b22147 82 FFTContext mdct_ctx[3];
dbb0f96f
BL
83 int channels;
84 DSPContext dsp;
85} AT1Ctx;
86
dbb0f96f
BL
87/** size of the transform in samples in the long mode for each QMF band */
88static const uint16_t samples_per_band[3] = {128, 128, 256};
89static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
90
91
04a6d1b0
BL
92static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
93 int rev_spec)
dbb0f96f 94{
a872e5c1 95 FFTContext* mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];
dbb0f96f
BL
96 int transf_size = 1 << nbits;
97
dbb0f96f
BL
98 if (rev_spec) {
99 int i;
1e1898c0 100 for (i = 0; i < transf_size / 2; i++)
04a6d1b0 101 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
dbb0f96f 102 }
26f548bb 103 mdct_context->imdct_half(mdct_context, out, spec);
dbb0f96f
BL
104}
105
106
107static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
108{
1e1898c0 109 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
ec129499 110 unsigned int start_pos, ref_pos = 0, pos = 0;
dbb0f96f 111
1e1898c0 112 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
b11d40d1
VS
113 float *prev_buf;
114 int j;
115
dbb0f96f
BL
116 band_samples = samples_per_band[band_num];
117 log2_block_count = su->log2_block_count[band_num];
118
119 /* number of mdct blocks in the current QMF band: 1 - for long mode */
120 /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
121 num_blocks = 1 << log2_block_count;
122
b11d40d1 123 if (num_blocks == 1) {
f49bcde6
VS
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;
dbb0f96f 127
f49bcde6
VS
128 /* calc transform size in bits according to the block_size_mode */
129 nbits = mdct_long_nbits[band_num] - log2_block_count;
dbb0f96f 130
f49bcde6
VS
131 if (nbits != 5 && nbits != 7 && nbits != 8)
132 return -1;
dbb0f96f 133 } else {
b11d40d1
VS
134 block_size = 32;
135 nbits = 5;
136 }
137
f49bcde6
VS
138 start_pos = 0;
139 prev_buf = &su->spectrum[1][ref_pos + band_samples - 16];
140 for (j=0; j < num_blocks; j++) {
141 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos + start_pos], nbits, band_num);
dbb0f96f 142
f49bcde6
VS
143 /* overlap and window */
144 q->dsp.vector_fmul_window(&q->bands[band_num][start_pos], prev_buf,
80ba1ddb 145 &su->spectrum[0][ref_pos + start_pos], ff_sine_32, 16);
10634c03 146
f49bcde6
VS
147 prev_buf = &su->spectrum[0][ref_pos+start_pos + 16];
148 start_pos += block_size;
149 pos += block_size;
150 }
b11d40d1
VS
151
152 if (num_blocks == 1)
153 memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float));
154
dbb0f96f
BL
155 ref_pos += band_samples;
156 }
157
158 /* Swap buffers so the mdct overlap works */
159 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
160
161 return 0;
162}
163
04a6d1b0
BL
164/**
165 * Parse the block size mode byte
166 */
dbb0f96f 167
04a6d1b0 168static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
dbb0f96f
BL
169{
170 int log2_block_count_tmp, i;
171
1e1898c0 172 for (i = 0; i < 2; i++) {
dbb0f96f
BL
173 /* low and mid band */
174 log2_block_count_tmp = get_bits(gb, 2);
175 if (log2_block_count_tmp & 1)
176 return -1;
04a6d1b0 177 log2_block_cnt[i] = 2 - log2_block_count_tmp;
dbb0f96f
BL
178 }
179
180 /* high band */
181 log2_block_count_tmp = get_bits(gb, 2);
182 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
183 return -1;
04a6d1b0 184 log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
dbb0f96f
BL
185
186 skip_bits(gb, 2);
187 return 0;
188}
189
190
04a6d1b0
BL
191static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
192 float spec[AT1_SU_SAMPLES])
dbb0f96f
BL
193{
194 int bits_used, band_num, bfu_num, i;
b6a23702
BL
195 uint8_t idwls[AT1_MAX_BFU]; ///< the word length indexes for each BFU
196 uint8_t idsfs[AT1_MAX_BFU]; ///< the scalefactor indexes for each BFU
dbb0f96f
BL
197
198 /* parse the info byte (2nd byte) telling how much BFUs were coded */
199 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
200
201 /* calc number of consumed bits:
202 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
203 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
204 bits_used = su->num_bfus * 10 + 32 +
205 bfu_amount_tab2[get_bits(gb, 2)] +
206 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
207
208 /* get word length index (idwl) for each BFU */
1e1898c0 209 for (i = 0; i < su->num_bfus; i++)
b6a23702 210 idwls[i] = get_bits(gb, 4);
dbb0f96f
BL
211
212 /* get scalefactor index (idsf) for each BFU */
1e1898c0 213 for (i = 0; i < su->num_bfus; i++)
b6a23702 214 idsfs[i] = get_bits(gb, 6);
dbb0f96f
BL
215
216 /* zero idwl/idsf for empty BFUs */
217 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
b6a23702 218 idwls[i] = idsfs[i] = 0;
dbb0f96f
BL
219
220 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
1e1898c0
DB
221 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
222 for (bfu_num = bfu_bands_t[band_num]; bfu_num < bfu_bands_t[band_num+1]; bfu_num++) {
dbb0f96f
BL
223 int pos;
224
225 int num_specs = specs_per_bfu[bfu_num];
b6a23702 226 int word_len = !!idwls[bfu_num] + idwls[bfu_num];
82e1f217 227 float scale_factor = ff_atrac_sf_table[idsfs[bfu_num]];
78b3a12d 228 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
dbb0f96f
BL
229
230 /* check for bitstream overflow */
231 if (bits_used > AT1_SU_MAX_BITS)
232 return -1;
233
234 /* get the position of the 1st spec according to the block size mode */
235 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
236
237 if (word_len) {
04a6d1b0 238 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
dbb0f96f 239
1e1898c0 240 for (i = 0; i < num_specs; i++) {
dbb0f96f
BL
241 /* read in a quantized spec and convert it to
242 * signed int and then inverse quantization
243 */
244 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
245 }
246 } else { /* word_len = 0 -> empty BFU, zero all specs in the emty BFU */
1e1898c0 247 memset(&spec[pos], 0, num_specs * sizeof(float));
dbb0f96f
BL
248 }
249 }
250 }
251
252 return 0;
253}
254
255
da0ac0ee 256static void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
dbb0f96f 257{
1e1898c0
DB
258 float temp[256];
259 float iqmf_temp[512 + 46];
dbb0f96f
BL
260
261 /* combine low and middle bands */
262 atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
263
264 /* delay the signal of the high band by 23 samples */
1e1898c0
DB
265 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 23);
266 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float) * 256);
dbb0f96f
BL
267
268 /* combine (low + middle) and high bands */
269 atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
270}
271
272
04a6d1b0
BL
273static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
274 int *data_size, AVPacket *avpkt)
dbb0f96f
BL
275{
276 const uint8_t *buf = avpkt->data;
04a6d1b0
BL
277 int buf_size = avpkt->size;
278 AT1Ctx *q = avctx->priv_data;
dbb0f96f
BL
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
1e1898c0 289 for (ch = 0; ch < q->channels; ch++) {
dbb0f96f
BL
290 AT1SUCtx* su = &q->SUs[ch];
291
1e1898c0 292 init_get_bits(&gb, &buf[212 * ch], 212 * 8);
dbb0f96f
BL
293
294 /* parse block_size_mode, 1st byte */
04a6d1b0 295 ret = at1_parse_bsm(&gb, su->log2_block_count);
dbb0f96f
BL
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
b1078e9f 309 /* interleave; FIXME, should create/use a DSP function */
dbb0f96f
BL
310 if (q->channels == 1) {
311 /* mono */
b1078e9f 312 memcpy(samples, q->out_samples[0], AT1_SU_SAMPLES * 4);
dbb0f96f
BL
313 } else {
314 /* stereo */
315 for (i = 0; i < AT1_SU_SAMPLES; i++) {
b1078e9f
RB
316 samples[i * 2] = q->out_samples[0][i];
317 samples[i * 2 + 1] = q->out_samples[1][i];
dbb0f96f
BL
318 }
319 }
320
321 *data_size = q->channels * AT1_SU_SAMPLES * sizeof(*samples);
322 return avctx->block_align;
323}
324
325
dbb0f96f
BL
326static av_cold int atrac1_decode_init(AVCodecContext *avctx)
327{
328 AT1Ctx *q = avctx->priv_data;
329
5d6e4c16 330 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
dbb0f96f
BL
331
332 q->channels = avctx->channels;
333
334 /* Init the mdct transforms */
1e1898c0
DB
335 ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15));
336 ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15));
337 ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15));
10634c03 338
14b86070 339 ff_init_ff_sine_windows(5);
dbb0f96f
BL
340
341 atrac_generate_tables();
342
343 dsputil_init(&q->dsp, avctx);
344
345 q->bands[0] = q->low;
346 q->bands[1] = q->mid;
347 q->bands[2] = q->high;
348
349 /* Prepare the mdct overlap buffers */
350 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
351 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
352 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
353 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
354
355 return 0;
356}
357
9caab878
BL
358
359static av_cold int atrac1_decode_end(AVCodecContext * avctx) {
360 AT1Ctx *q = avctx->priv_data;
361
362 ff_mdct_end(&q->mdct_ctx[0]);
363 ff_mdct_end(&q->mdct_ctx[1]);
364 ff_mdct_end(&q->mdct_ctx[2]);
365 return 0;
366}
367
368
d36beb3f 369AVCodec ff_atrac1_decoder = {
dbb0f96f 370 .name = "atrac1",
72415b2a 371 .type = AVMEDIA_TYPE_AUDIO,
dbb0f96f
BL
372 .id = CODEC_ID_ATRAC1,
373 .priv_data_size = sizeof(AT1Ctx),
374 .init = atrac1_decode_init,
9caab878 375 .close = atrac1_decode_end,
dbb0f96f
BL
376 .decode = atrac1_decode_frame,
377 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),
378};