atrac1: check for ff_mdct_init() failure
[libav.git] / libavcodec / atrac1.c
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 Libav.
7 *
8 * Libav 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 * Libav 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 Libav; 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 * Atrac 1 compatible decoder.
26 * This decoder handles raw ATRAC1 data and probably SDDS 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 #include "fft.h"
39 #include "fmtconvert.h"
40 #include "sinewin.h"
41
42 #include "atrac.h"
43 #include "atrac1data.h"
44
45 #define AT1_MAX_BFU 52 ///< max number of block floating units in a sound unit
46 #define AT1_SU_SIZE 212 ///< number of bytes in a sound unit
47 #define AT1_SU_SAMPLES 512 ///< number of samples in a sound unit
48 #define AT1_FRAME_SIZE AT1_SU_SIZE * 2
49 #define AT1_SU_MAX_BITS AT1_SU_SIZE * 8
50 #define AT1_MAX_CHANNELS 2
51
52 #define AT1_QMF_BANDS 3
53 #define IDX_LOW_BAND 0
54 #define IDX_MID_BAND 1
55 #define IDX_HIGH_BAND 2
56
57 /**
58 * Sound unit struct, one unit is used per channel
59 */
60 typedef struct {
61 int log2_block_count[AT1_QMF_BANDS]; ///< log2 number of blocks in a band
62 int num_bfus; ///< number of Block Floating Units
63 float* spectrum[2];
64 DECLARE_ALIGNED(32, float, spec1)[AT1_SU_SAMPLES]; ///< mdct buffer
65 DECLARE_ALIGNED(32, float, spec2)[AT1_SU_SAMPLES]; ///< mdct buffer
66 DECLARE_ALIGNED(32, float, fst_qmf_delay)[46]; ///< delay line for the 1st stacked QMF filter
67 DECLARE_ALIGNED(32, float, snd_qmf_delay)[46]; ///< delay line for the 2nd stacked QMF filter
68 DECLARE_ALIGNED(32, float, last_qmf_delay)[256+23]; ///< delay line for the last stacked QMF filter
69 } AT1SUCtx;
70
71 /**
72 * The atrac1 context, holds all needed parameters for decoding
73 */
74 typedef struct {
75 AT1SUCtx SUs[AT1_MAX_CHANNELS]; ///< channel sound unit
76 DECLARE_ALIGNED(32, float, spec)[AT1_SU_SAMPLES]; ///< the mdct spectrum buffer
77
78 DECLARE_ALIGNED(32, float, low)[256];
79 DECLARE_ALIGNED(32, float, mid)[256];
80 DECLARE_ALIGNED(32, float, high)[512];
81 float* bands[3];
82 float *out_samples[AT1_MAX_CHANNELS];
83 FFTContext mdct_ctx[3];
84 int channels;
85 DSPContext dsp;
86 FmtConvertContext fmt_conv;
87 } AT1Ctx;
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
94 static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
95 int rev_spec)
96 {
97 FFTContext* mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];
98 int transf_size = 1 << nbits;
99
100 if (rev_spec) {
101 int i;
102 for (i = 0; i < transf_size / 2; i++)
103 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
104 }
105 mdct_context->imdct_half(mdct_context, out, spec);
106 }
107
108
109 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
110 {
111 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
112 unsigned int start_pos, ref_pos = 0, pos = 0;
113
114 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
115 float *prev_buf;
116 int j;
117
118 band_samples = samples_per_band[band_num];
119 log2_block_count = su->log2_block_count[band_num];
120
121 /* number of mdct blocks in the current QMF band: 1 - for long mode */
122 /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
123 num_blocks = 1 << log2_block_count;
124
125 if (num_blocks == 1) {
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 } else {
136 block_size = 32;
137 nbits = 5;
138 }
139
140 start_pos = 0;
141 prev_buf = &su->spectrum[1][ref_pos + band_samples - 16];
142 for (j=0; j < num_blocks; j++) {
143 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos + start_pos], nbits, band_num);
144
145 /* overlap and window */
146 q->dsp.vector_fmul_window(&q->bands[band_num][start_pos], prev_buf,
147 &su->spectrum[0][ref_pos + start_pos], ff_sine_32, 16);
148
149 prev_buf = &su->spectrum[0][ref_pos+start_pos + 16];
150 start_pos += block_size;
151 pos += block_size;
152 }
153
154 if (num_blocks == 1)
155 memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float));
156
157 ref_pos += band_samples;
158 }
159
160 /* Swap buffers so the mdct overlap works */
161 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
162
163 return 0;
164 }
165
166 /**
167 * Parse the block size mode byte
168 */
169
170 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[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_cnt[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_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
187
188 skip_bits(gb, 2);
189 return 0;
190 }
191
192
193 static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
194 float spec[AT1_SU_SAMPLES])
195 {
196 int bits_used, band_num, bfu_num, i;
197 uint8_t idwls[AT1_MAX_BFU]; ///< the word length indexes for each BFU
198 uint8_t idsfs[AT1_MAX_BFU]; ///< the scalefactor indexes for each BFU
199
200 /* parse the info byte (2nd byte) telling how much BFUs were coded */
201 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
202
203 /* calc number of consumed bits:
204 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
205 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
206 bits_used = su->num_bfus * 10 + 32 +
207 bfu_amount_tab2[get_bits(gb, 2)] +
208 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
209
210 /* get word length index (idwl) for each BFU */
211 for (i = 0; i < su->num_bfus; i++)
212 idwls[i] = get_bits(gb, 4);
213
214 /* get scalefactor index (idsf) for each BFU */
215 for (i = 0; i < su->num_bfus; i++)
216 idsfs[i] = get_bits(gb, 6);
217
218 /* zero idwl/idsf for empty BFUs */
219 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
220 idwls[i] = idsfs[i] = 0;
221
222 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
223 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
224 for (bfu_num = bfu_bands_t[band_num]; bfu_num < bfu_bands_t[band_num+1]; bfu_num++) {
225 int pos;
226
227 int num_specs = specs_per_bfu[bfu_num];
228 int word_len = !!idwls[bfu_num] + idwls[bfu_num];
229 float scale_factor = ff_atrac_sf_table[idsfs[bfu_num]];
230 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
231
232 /* check for bitstream overflow */
233 if (bits_used > AT1_SU_MAX_BITS)
234 return -1;
235
236 /* get the position of the 1st spec according to the block size mode */
237 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
238
239 if (word_len) {
240 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
241
242 for (i = 0; i < num_specs; i++) {
243 /* read in a quantized spec and convert it to
244 * signed int and then inverse quantization
245 */
246 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
247 }
248 } else { /* word_len = 0 -> empty BFU, zero all specs in the emty BFU */
249 memset(&spec[pos], 0, num_specs * sizeof(float));
250 }
251 }
252 }
253
254 return 0;
255 }
256
257
258 static void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
259 {
260 float temp[256];
261 float iqmf_temp[512 + 46];
262
263 /* combine low and middle bands */
264 atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
265
266 /* delay the signal of the high band by 23 samples */
267 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 23);
268 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float) * 256);
269
270 /* combine (low + middle) and high bands */
271 atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
272 }
273
274
275 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
276 int *data_size, AVPacket *avpkt)
277 {
278 const uint8_t *buf = avpkt->data;
279 int buf_size = avpkt->size;
280 AT1Ctx *q = avctx->priv_data;
281 int ch, ret, out_size;
282 GetBitContext gb;
283 float* samples = data;
284
285
286 if (buf_size < 212 * q->channels) {
287 av_log(q,AV_LOG_ERROR,"Not enough data to decode!\n");
288 return -1;
289 }
290
291 out_size = q->channels * AT1_SU_SAMPLES *
292 av_get_bytes_per_sample(avctx->sample_fmt);
293 if (*data_size < out_size) {
294 av_log(avctx, AV_LOG_ERROR, "Output buffer is too small\n");
295 return AVERROR(EINVAL);
296 }
297
298 for (ch = 0; ch < q->channels; ch++) {
299 AT1SUCtx* su = &q->SUs[ch];
300
301 init_get_bits(&gb, &buf[212 * ch], 212 * 8);
302
303 /* parse block_size_mode, 1st byte */
304 ret = at1_parse_bsm(&gb, su->log2_block_count);
305 if (ret < 0)
306 return ret;
307
308 ret = at1_unpack_dequant(&gb, su, q->spec);
309 if (ret < 0)
310 return ret;
311
312 ret = at1_imdct_block(su, q);
313 if (ret < 0)
314 return ret;
315 at1_subband_synthesis(q, su, q->channels == 1 ? samples : q->out_samples[ch]);
316 }
317
318 /* interleave */
319 if (q->channels == 2) {
320 q->fmt_conv.float_interleave(samples, (const float **)q->out_samples,
321 AT1_SU_SAMPLES, 2);
322 }
323
324 *data_size = out_size;
325 return avctx->block_align;
326 }
327
328
329 static av_cold int atrac1_decode_end(AVCodecContext * avctx)
330 {
331 AT1Ctx *q = avctx->priv_data;
332
333 av_freep(&q->out_samples[0]);
334
335 ff_mdct_end(&q->mdct_ctx[0]);
336 ff_mdct_end(&q->mdct_ctx[1]);
337 ff_mdct_end(&q->mdct_ctx[2]);
338
339 return 0;
340 }
341
342
343 static av_cold int atrac1_decode_init(AVCodecContext *avctx)
344 {
345 AT1Ctx *q = avctx->priv_data;
346 int ret;
347
348 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
349
350 if (avctx->channels < 1 || avctx->channels > AT1_MAX_CHANNELS) {
351 av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n",
352 avctx->channels);
353 return AVERROR(EINVAL);
354 }
355 q->channels = avctx->channels;
356
357 if (avctx->channels == 2) {
358 q->out_samples[0] = av_malloc(2 * AT1_SU_SAMPLES * sizeof(*q->out_samples[0]));
359 q->out_samples[1] = q->out_samples[0] + AT1_SU_SAMPLES;
360 if (!q->out_samples[0]) {
361 av_freep(&q->out_samples[0]);
362 return AVERROR(ENOMEM);
363 }
364 }
365
366 /* Init the mdct transforms */
367 if ((ret = ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15))) ||
368 (ret = ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15))) ||
369 (ret = ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15)))) {
370 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
371 atrac1_decode_end(avctx);
372 return ret;
373 }
374
375 ff_init_ff_sine_windows(5);
376
377 atrac_generate_tables();
378
379 dsputil_init(&q->dsp, avctx);
380 ff_fmt_convert_init(&q->fmt_conv, avctx);
381
382 q->bands[0] = q->low;
383 q->bands[1] = q->mid;
384 q->bands[2] = q->high;
385
386 /* Prepare the mdct overlap buffers */
387 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
388 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
389 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
390 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
391
392 return 0;
393 }
394
395
396 AVCodec ff_atrac1_decoder = {
397 .name = "atrac1",
398 .type = AVMEDIA_TYPE_AUDIO,
399 .id = CODEC_ID_ATRAC1,
400 .priv_data_size = sizeof(AT1Ctx),
401 .init = atrac1_decode_init,
402 .close = atrac1_decode_end,
403 .decode = atrac1_decode_frame,
404 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),
405 };