2 * Atrac 1 compatible decoder
3 * Copyright (c) 2009 Maxim Poliakovski
4 * Copyright (c) 2009 Benjamin Larsson
6 * This file is part of FFmpeg.
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.
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.
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
24 * @file libavcodec/atrac1.c
25 * Atrac 1 compatible decoder.
26 * This decoder handles raw ATRAC1 data.
29 /* Many thanks to Tim Craig for all the help! */
40 #include "atrac1data.h"
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
49 #define AT1_QMF_BANDS 3
50 #define IDX_LOW_BAND 0
51 #define IDX_MID_BAND 1
52 #define IDX_HIGH_BAND 2
55 * Sound unit struct, one unit is used per channel
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
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
71 * The atrac1 context, holds all needed parameters for decoding
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[512]); ///< buffer for the short mode
78 DECLARE_ALIGNED_16(float, low[256]);
79 DECLARE_ALIGNED_16(float, mid[256]);
80 DECLARE_ALIGNED_16(float,high[512]);
82 float out_samples[AT1_MAX_CHANNELS][AT1_SU_SAMPLES];
83 MDCTContext mdct_ctx[3];
88 DECLARE_ALIGNED_16(static float, short_window[32]);
90 /** size of the transform in samples in the long mode for each QMF band */
91 static const uint16_t samples_per_band[3] = {128, 128, 256};
92 static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
95 static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
98 MDCTContext* mdct_context;
99 int transf_size = 1 << nbits;
101 mdct_context = &q->mdct_ctx[nbits - 5 - (nbits>6)];
105 for (i=0 ; i<transf_size/2 ; i++)
106 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
108 ff_imdct_half(mdct_context, out, spec);
112 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
114 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
115 unsigned int start_pos, ref_pos=0, pos = 0;
117 for (band_num=0 ; band_num<AT1_QMF_BANDS ; band_num++) {
118 band_samples = samples_per_band[band_num];
119 log2_block_count = su->log2_block_count[band_num];
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;
125 /* mdct block size in samples: 128 (long mode, low & mid bands), */
126 /* 256 (long mode, high band) and 32 (short mode, all bands) */
127 block_size = band_samples >> log2_block_count;
129 /* calc transform size in bits according to the block_size_mode */
130 nbits = mdct_long_nbits[band_num] - log2_block_count;
132 if (nbits!=5 && nbits!=7 && nbits!=8)
135 if (num_blocks == 1) {
136 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos], nbits, band_num);
137 pos += block_size; // move to the next mdct block in the spectrum
139 /* overlap and window long blocks */
140 q->dsp.vector_fmul_window(q->bands[band_num], &su->spectrum[1][ref_pos+band_samples-16],
141 &su->spectrum[0][ref_pos], short_window, 0, 16);
142 memcpy(q->bands[band_num]+32, &su->spectrum[0][ref_pos+16], 240 * sizeof(float));
145 /* calc start position for the 1st short block: 96(128) or 112(256) */
148 start_pos = (band_samples * (num_blocks - 1)) >> (log2_block_count + 1);
149 memset(&su->spectrum[0][ref_pos], 0, sizeof(float) * (band_samples * 2));
151 prev_buf = &su->spectrum[1][ref_pos+band_samples-16];
152 for (; num_blocks!=0 ; num_blocks--) {
153 /* use hardcoded nbits for the short mode */
154 at1_imdct(q, &q->spec[pos], &q->short_buf[short_pos], 5, band_num);
156 /* overlap and window between short blocks */
157 q->dsp.vector_fmul_window(&su->spectrum[0][ref_pos+start_pos],
158 &q->short_buf[short_pos-16],
159 &q->short_buf[short_pos],short_window, 0, 16);
161 prev_buf = &q->short_buf[short_pos+16];
163 start_pos += 32; // use hardcoded block_size
167 memcpy(q->bands[band_num], &su->spectrum[0][ref_pos], band_samples*sizeof(float));
169 ref_pos += band_samples;
172 /* Swap buffers so the mdct overlap works */
173 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
179 * Parse the block size mode byte
182 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
184 int log2_block_count_tmp, i;
186 for(i=0 ; i<2 ; i++) {
187 /* low and mid band */
188 log2_block_count_tmp = get_bits(gb, 2);
189 if (log2_block_count_tmp & 1)
191 log2_block_cnt[i] = 2 - log2_block_count_tmp;
195 log2_block_count_tmp = get_bits(gb, 2);
196 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
198 log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
205 static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
206 float spec[AT1_SU_SAMPLES])
208 int bits_used, band_num, bfu_num, i;
210 /* parse the info byte (2nd byte) telling how much BFUs were coded */
211 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
213 /* calc number of consumed bits:
214 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
215 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
216 bits_used = su->num_bfus * 10 + 32 +
217 bfu_amount_tab2[get_bits(gb, 2)] +
218 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
220 /* get word length index (idwl) for each BFU */
221 for (i=0 ; i<su->num_bfus ; i++)
222 su->idwls[i] = get_bits(gb, 4);
224 /* get scalefactor index (idsf) for each BFU */
225 for (i=0 ; i<su->num_bfus ; i++)
226 su->idsfs[i] = get_bits(gb, 6);
228 /* zero idwl/idsf for empty BFUs */
229 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
230 su->idwls[i] = su->idsfs[i] = 0;
232 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
233 for (band_num=0 ; band_num<AT1_QMF_BANDS ; band_num++) {
234 for (bfu_num=bfu_bands_t[band_num] ; bfu_num<bfu_bands_t[band_num+1] ; bfu_num++) {
237 int num_specs = specs_per_bfu[bfu_num];
238 int word_len = !!su->idwls[bfu_num] + su->idwls[bfu_num];
239 float scale_factor = sf_table[su->idsfs[bfu_num]];
240 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
242 /* check for bitstream overflow */
243 if (bits_used > AT1_SU_MAX_BITS)
246 /* get the position of the 1st spec according to the block size mode */
247 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
250 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
252 for (i=0 ; i<num_specs ; i++) {
253 /* read in a quantized spec and convert it to
254 * signed int and then inverse quantization
256 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
258 } else { /* word_len = 0 -> empty BFU, zero all specs in the emty BFU */
259 memset(&spec[pos], 0, num_specs*sizeof(float));
268 void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
271 float iqmf_temp[512 + 46];
273 /* combine low and middle bands */
274 atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
276 /* delay the signal of the high band by 23 samples */
277 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float)*23);
278 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float)*256);
280 /* combine (low + middle) and high bands */
281 atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
285 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
286 int *data_size, AVPacket *avpkt)
288 const uint8_t *buf = avpkt->data;
289 int buf_size = avpkt->size;
290 AT1Ctx *q = avctx->priv_data;
293 float* samples = data;
296 if (buf_size < 212 * q->channels) {
297 av_log(q,AV_LOG_ERROR,"Not enought data to decode!\n");
301 for (ch=0 ; ch<q->channels ; ch++) {
302 AT1SUCtx* su = &q->SUs[ch];
304 init_get_bits(&gb, &buf[212*ch], 212*8);
306 /* parse block_size_mode, 1st byte */
307 ret = at1_parse_bsm(&gb, su->log2_block_count);
311 ret = at1_unpack_dequant(&gb, su, q->spec);
315 ret = at1_imdct_block(su, q);
318 at1_subband_synthesis(q, su, q->out_samples[ch]);
321 /* round, convert to 16bit and interleave */
322 if (q->channels == 1) {
324 q->dsp.vector_clipf(samples, q->out_samples[0], -32700.0 / (1<<15),
325 32700.0 / (1<<15), AT1_SU_SAMPLES);
328 for (i = 0; i < AT1_SU_SAMPLES; i++) {
329 samples[i*2] = av_clipf(q->out_samples[0][i], -32700.0 / (1<<15),
331 samples[i*2+1] = av_clipf(q->out_samples[1][i], -32700.0 / (1<<15),
336 *data_size = q->channels * AT1_SU_SAMPLES * sizeof(*samples);
337 return avctx->block_align;
341 static av_cold int atrac1_decode_init(AVCodecContext *avctx)
343 AT1Ctx *q = avctx->priv_data;
345 avctx->sample_fmt = SAMPLE_FMT_FLT;
347 q->channels = avctx->channels;
349 /* Init the mdct transforms */
350 ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1<<15));
351 ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1<<15));
352 ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1<<15));
354 ff_sine_window_init(short_window, 32);
356 atrac_generate_tables();
358 dsputil_init(&q->dsp, avctx);
360 q->bands[0] = q->low;
361 q->bands[1] = q->mid;
362 q->bands[2] = q->high;
364 /* Prepare the mdct overlap buffers */
365 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
366 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
367 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
368 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
373 AVCodec atrac1_decoder = {
375 .type = CODEC_TYPE_AUDIO,
376 .id = CODEC_ID_ATRAC1,
377 .priv_data_size = sizeof(AT1Ctx),
378 .init = atrac1_decode_init,
380 .decode = atrac1_decode_frame,
381 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),