blob: a8c8c91bccbfc88db4945b4e5b0a7d9ba52c6230
1 | /* |
2 | * ATRAC1 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 |
25 | * ATRAC1 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 "libavutil/float_dsp.h" |
36 | #include "avcodec.h" |
37 | #include "get_bits.h" |
38 | #include "fft.h" |
39 | #include "internal.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 AT1SUCtx { |
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+39]; ///< 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 AT1Ctx { |
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 | FFTContext mdct_ctx[3]; |
83 | AVFloatDSPContext *fdsp; |
84 | } AT1Ctx; |
85 | |
86 | /** size of the transform in samples in the long mode for each QMF band */ |
87 | static const uint16_t samples_per_band[3] = {128, 128, 256}; |
88 | static const uint8_t mdct_long_nbits[3] = {7, 7, 8}; |
89 | |
90 | |
91 | static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits, |
92 | int rev_spec) |
93 | { |
94 | FFTContext* mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)]; |
95 | int transf_size = 1 << nbits; |
96 | |
97 | if (rev_spec) { |
98 | int i; |
99 | for (i = 0; i < transf_size / 2; i++) |
100 | FFSWAP(float, spec[i], spec[transf_size - 1 - i]); |
101 | } |
102 | mdct_context->imdct_half(mdct_context, out, spec); |
103 | } |
104 | |
105 | |
106 | static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q) |
107 | { |
108 | int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size; |
109 | unsigned int start_pos, ref_pos = 0, pos = 0; |
110 | |
111 | for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) { |
112 | float *prev_buf; |
113 | int j; |
114 | |
115 | band_samples = samples_per_band[band_num]; |
116 | log2_block_count = su->log2_block_count[band_num]; |
117 | |
118 | /* number of mdct blocks in the current QMF band: 1 - for long mode */ |
119 | /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/ |
120 | num_blocks = 1 << log2_block_count; |
121 | |
122 | if (num_blocks == 1) { |
123 | /* mdct block size in samples: 128 (long mode, low & mid bands), */ |
124 | /* 256 (long mode, high band) and 32 (short mode, all bands) */ |
125 | block_size = band_samples >> log2_block_count; |
126 | |
127 | /* calc transform size in bits according to the block_size_mode */ |
128 | nbits = mdct_long_nbits[band_num] - log2_block_count; |
129 | |
130 | if (nbits != 5 && nbits != 7 && nbits != 8) |
131 | return AVERROR_INVALIDDATA; |
132 | } else { |
133 | block_size = 32; |
134 | nbits = 5; |
135 | } |
136 | |
137 | start_pos = 0; |
138 | prev_buf = &su->spectrum[1][ref_pos + band_samples - 16]; |
139 | for (j=0; j < num_blocks; j++) { |
140 | at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos + start_pos], nbits, band_num); |
141 | |
142 | /* overlap and window */ |
143 | q->fdsp->vector_fmul_window(&q->bands[band_num][start_pos], prev_buf, |
144 | &su->spectrum[0][ref_pos + start_pos], ff_sine_32, 16); |
145 | |
146 | prev_buf = &su->spectrum[0][ref_pos+start_pos + 16]; |
147 | start_pos += block_size; |
148 | pos += block_size; |
149 | } |
150 | |
151 | if (num_blocks == 1) |
152 | memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float)); |
153 | |
154 | ref_pos += band_samples; |
155 | } |
156 | |
157 | /* Swap buffers so the mdct overlap works */ |
158 | FFSWAP(float*, su->spectrum[0], su->spectrum[1]); |
159 | |
160 | return 0; |
161 | } |
162 | |
163 | /** |
164 | * Parse the block size mode byte |
165 | */ |
166 | |
167 | static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS]) |
168 | { |
169 | int log2_block_count_tmp, i; |
170 | |
171 | for (i = 0; i < 2; i++) { |
172 | /* low and mid band */ |
173 | log2_block_count_tmp = get_bits(gb, 2); |
174 | if (log2_block_count_tmp & 1) |
175 | return AVERROR_INVALIDDATA; |
176 | log2_block_cnt[i] = 2 - log2_block_count_tmp; |
177 | } |
178 | |
179 | /* high band */ |
180 | log2_block_count_tmp = get_bits(gb, 2); |
181 | if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3) |
182 | return AVERROR_INVALIDDATA; |
183 | log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp; |
184 | |
185 | skip_bits(gb, 2); |
186 | return 0; |
187 | } |
188 | |
189 | |
190 | static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su, |
191 | float spec[AT1_SU_SAMPLES]) |
192 | { |
193 | int bits_used, band_num, bfu_num, i; |
194 | uint8_t idwls[AT1_MAX_BFU]; ///< the word length indexes for each BFU |
195 | uint8_t idsfs[AT1_MAX_BFU]; ///< the scalefactor indexes for each BFU |
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 | 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 | 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 | idwls[i] = 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 = !!idwls[bfu_num] + idwls[bfu_num]; |
226 | float scale_factor = ff_atrac_sf_table[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 AVERROR_INVALIDDATA; |
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 empty BFU */ |
246 | memset(&spec[pos], 0, num_specs * sizeof(float)); |
247 | } |
248 | } |
249 | } |
250 | |
251 | return 0; |
252 | } |
253 | |
254 | |
255 | static void 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 | ff_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 39 samples */ |
264 | memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 39); |
265 | memcpy(&su->last_qmf_delay[39], q->bands[2], sizeof(float) * 256); |
266 | |
267 | /* combine (low + middle) and high bands */ |
268 | ff_atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp); |
269 | } |
270 | |
271 | |
272 | static int atrac1_decode_frame(AVCodecContext *avctx, void *data, |
273 | int *got_frame_ptr, AVPacket *avpkt) |
274 | { |
275 | AVFrame *frame = data; |
276 | const uint8_t *buf = avpkt->data; |
277 | int buf_size = avpkt->size; |
278 | AT1Ctx *q = avctx->priv_data; |
279 | int ch, ret; |
280 | GetBitContext gb; |
281 | |
282 | |
283 | if (buf_size < 212 * avctx->channels) { |
284 | av_log(avctx, AV_LOG_ERROR, "Not enough data to decode!\n"); |
285 | return AVERROR_INVALIDDATA; |
286 | } |
287 | |
288 | /* get output buffer */ |
289 | frame->nb_samples = AT1_SU_SAMPLES; |
290 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
291 | return ret; |
292 | |
293 | for (ch = 0; ch < avctx->channels; ch++) { |
294 | AT1SUCtx* su = &q->SUs[ch]; |
295 | |
296 | init_get_bits(&gb, &buf[212 * ch], 212 * 8); |
297 | |
298 | /* parse block_size_mode, 1st byte */ |
299 | ret = at1_parse_bsm(&gb, su->log2_block_count); |
300 | if (ret < 0) |
301 | return ret; |
302 | |
303 | ret = at1_unpack_dequant(&gb, su, q->spec); |
304 | if (ret < 0) |
305 | return ret; |
306 | |
307 | ret = at1_imdct_block(su, q); |
308 | if (ret < 0) |
309 | return ret; |
310 | at1_subband_synthesis(q, su, (float *)frame->extended_data[ch]); |
311 | } |
312 | |
313 | *got_frame_ptr = 1; |
314 | |
315 | return avctx->block_align; |
316 | } |
317 | |
318 | |
319 | static av_cold int atrac1_decode_end(AVCodecContext * avctx) |
320 | { |
321 | AT1Ctx *q = avctx->priv_data; |
322 | |
323 | ff_mdct_end(&q->mdct_ctx[0]); |
324 | ff_mdct_end(&q->mdct_ctx[1]); |
325 | ff_mdct_end(&q->mdct_ctx[2]); |
326 | |
327 | av_freep(&q->fdsp); |
328 | |
329 | return 0; |
330 | } |
331 | |
332 | |
333 | static av_cold int atrac1_decode_init(AVCodecContext *avctx) |
334 | { |
335 | AT1Ctx *q = avctx->priv_data; |
336 | int ret; |
337 | |
338 | avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; |
339 | |
340 | if (avctx->channels < 1 || avctx->channels > AT1_MAX_CHANNELS) { |
341 | av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n", |
342 | avctx->channels); |
343 | return AVERROR(EINVAL); |
344 | } |
345 | |
346 | if (avctx->block_align <= 0) { |
347 | av_log(avctx, AV_LOG_ERROR, "Unsupported block align."); |
348 | return AVERROR_PATCHWELCOME; |
349 | } |
350 | |
351 | /* Init the mdct transforms */ |
352 | if ((ret = ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15))) || |
353 | (ret = ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15))) || |
354 | (ret = ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15)))) { |
355 | av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n"); |
356 | atrac1_decode_end(avctx); |
357 | return ret; |
358 | } |
359 | |
360 | ff_init_ff_sine_windows(5); |
361 | |
362 | ff_atrac_generate_tables(); |
363 | |
364 | q->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT); |
365 | |
366 | q->bands[0] = q->low; |
367 | q->bands[1] = q->mid; |
368 | q->bands[2] = q->high; |
369 | |
370 | /* Prepare the mdct overlap buffers */ |
371 | q->SUs[0].spectrum[0] = q->SUs[0].spec1; |
372 | q->SUs[0].spectrum[1] = q->SUs[0].spec2; |
373 | q->SUs[1].spectrum[0] = q->SUs[1].spec1; |
374 | q->SUs[1].spectrum[1] = q->SUs[1].spec2; |
375 | |
376 | return 0; |
377 | } |
378 | |
379 | |
380 | AVCodec ff_atrac1_decoder = { |
381 | .name = "atrac1", |
382 | .long_name = NULL_IF_CONFIG_SMALL("ATRAC1 (Adaptive TRansform Acoustic Coding)"), |
383 | .type = AVMEDIA_TYPE_AUDIO, |
384 | .id = AV_CODEC_ID_ATRAC1, |
385 | .priv_data_size = sizeof(AT1Ctx), |
386 | .init = atrac1_decode_init, |
387 | .close = atrac1_decode_end, |
388 | .decode = atrac1_decode_frame, |
389 | .capabilities = AV_CODEC_CAP_DR1, |
390 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
391 | AV_SAMPLE_FMT_NONE }, |
392 | }; |
393 |