2 * ATRAC1 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
25 * ATRAC1 compatible decoder.
26 * This decoder handles raw ATRAC1 data and probably SDDS data.
29 /* Many thanks to Tim Craig for all the help! */
35 #include "libavutil/float_dsp.h"
36 #include "libavutil/mem_internal.h"
45 #include "atrac1data.h"
47 #define AT1_MAX_BFU 52 ///< max number of block floating units in a sound unit
48 #define AT1_SU_SIZE 212 ///< number of bytes in a sound unit
49 #define AT1_SU_SAMPLES 512 ///< number of samples in a sound unit
50 #define AT1_FRAME_SIZE AT1_SU_SIZE * 2
51 #define AT1_SU_MAX_BITS AT1_SU_SIZE * 8
52 #define AT1_MAX_CHANNELS 2
54 #define AT1_QMF_BANDS 3
55 #define IDX_LOW_BAND 0
56 #define IDX_MID_BAND 1
57 #define IDX_HIGH_BAND 2
60 * Sound unit struct, one unit is used per channel
62 typedef struct AT1SUCtx {
63 int log2_block_count[AT1_QMF_BANDS]; ///< log2 number of blocks in a band
64 int num_bfus; ///< number of Block Floating Units
66 DECLARE_ALIGNED(32, float, spec1)[AT1_SU_SAMPLES]; ///< mdct buffer
67 DECLARE_ALIGNED(32, float, spec2)[AT1_SU_SAMPLES]; ///< mdct buffer
68 DECLARE_ALIGNED(32, float, fst_qmf_delay)[46]; ///< delay line for the 1st stacked QMF filter
69 DECLARE_ALIGNED(32, float, snd_qmf_delay)[46]; ///< delay line for the 2nd stacked QMF filter
70 DECLARE_ALIGNED(32, float, last_qmf_delay)[256+39]; ///< delay line for the last stacked QMF filter
74 * The atrac1 context, holds all needed parameters for decoding
76 typedef struct AT1Ctx {
77 AT1SUCtx SUs[AT1_MAX_CHANNELS]; ///< channel sound unit
78 DECLARE_ALIGNED(32, float, spec)[AT1_SU_SAMPLES]; ///< the mdct spectrum buffer
80 DECLARE_ALIGNED(32, float, low)[256];
81 DECLARE_ALIGNED(32, float, mid)[256];
82 DECLARE_ALIGNED(32, float, high)[512];
84 FFTContext mdct_ctx[3];
85 void (*vector_fmul_window)(float *dst, const float *src0,
86 const float *src1, const float *win, int len);
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};
94 static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
97 FFTContext* mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];
98 int transf_size = 1 << nbits;
102 for (i = 0; i < transf_size / 2; i++)
103 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
105 mdct_context->imdct_half(mdct_context, out, spec);
109 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
111 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
112 unsigned int start_pos, ref_pos = 0, pos = 0;
114 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 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;
130 /* calc transform size in bits according to the block_size_mode */
131 nbits = mdct_long_nbits[band_num] - log2_block_count;
133 if (nbits != 5 && nbits != 7 && nbits != 8)
134 return AVERROR_INVALIDDATA;
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);
145 /* overlap and window */
146 q->vector_fmul_window(&q->bands[band_num][start_pos], prev_buf,
147 &su->spectrum[0][ref_pos + start_pos], ff_sine_32, 16);
149 prev_buf = &su->spectrum[0][ref_pos+start_pos + 16];
150 start_pos += block_size;
155 memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float));
157 ref_pos += band_samples;
160 /* Swap buffers so the mdct overlap works */
161 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
167 * Parse the block size mode byte
170 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
172 int log2_block_count_tmp, i;
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 AVERROR_INVALIDDATA;
179 log2_block_cnt[i] = 2 - log2_block_count_tmp;
183 log2_block_count_tmp = get_bits(gb, 2);
184 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
185 return AVERROR_INVALIDDATA;
186 log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
193 static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
194 float spec[AT1_SU_SAMPLES])
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
200 /* parse the info byte (2nd byte) telling how much BFUs were coded */
201 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
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);
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);
214 /* get scalefactor index (idsf) for each BFU */
215 for (i = 0; i < su->num_bfus; i++)
216 idsfs[i] = get_bits(gb, 6);
218 /* zero idwl/idsf for empty BFUs */
219 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
220 idwls[i] = idsfs[i] = 0;
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++) {
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 */
232 /* check for bitstream overflow */
233 if (bits_used > AT1_SU_MAX_BITS)
234 return AVERROR_INVALIDDATA;
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];
240 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
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
246 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
248 } else { /* word_len = 0 -> empty BFU, zero all specs in the empty BFU */
249 memset(&spec[pos], 0, num_specs * sizeof(float));
258 static void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
261 float iqmf_temp[512 + 46];
263 /* combine low and middle bands */
264 ff_atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
266 /* delay the signal of the high band by 39 samples */
267 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 39);
268 memcpy(&su->last_qmf_delay[39], q->bands[2], sizeof(float) * 256);
270 /* combine (low + middle) and high bands */
271 ff_atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
275 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
276 int *got_frame_ptr, AVPacket *avpkt)
278 AVFrame *frame = data;
279 const uint8_t *buf = avpkt->data;
280 int buf_size = avpkt->size;
281 AT1Ctx *q = avctx->priv_data;
286 if (buf_size < 212 * avctx->channels) {
287 av_log(avctx, AV_LOG_ERROR, "Not enough data to decode!\n");
288 return AVERROR_INVALIDDATA;
291 /* get output buffer */
292 frame->nb_samples = AT1_SU_SAMPLES;
293 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
296 for (ch = 0; ch < avctx->channels; ch++) {
297 AT1SUCtx* su = &q->SUs[ch];
299 init_get_bits(&gb, &buf[212 * ch], 212 * 8);
301 /* parse block_size_mode, 1st byte */
302 ret = at1_parse_bsm(&gb, su->log2_block_count);
306 ret = at1_unpack_dequant(&gb, su, q->spec);
310 ret = at1_imdct_block(su, q);
313 at1_subband_synthesis(q, su, (float *)frame->extended_data[ch]);
318 return avctx->block_align;
322 static av_cold int atrac1_decode_end(AVCodecContext * avctx)
324 AT1Ctx *q = avctx->priv_data;
326 ff_mdct_end(&q->mdct_ctx[0]);
327 ff_mdct_end(&q->mdct_ctx[1]);
328 ff_mdct_end(&q->mdct_ctx[2]);
334 static av_cold int atrac1_decode_init(AVCodecContext *avctx)
336 AT1Ctx *q = avctx->priv_data;
337 AVFloatDSPContext *fdsp;
340 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
342 if (avctx->channels < 1 || avctx->channels > AT1_MAX_CHANNELS) {
343 av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n",
345 return AVERROR(EINVAL);
348 if (avctx->block_align <= 0) {
349 av_log(avctx, AV_LOG_ERROR, "Unsupported block align.");
350 return AVERROR_PATCHWELCOME;
353 /* Init the mdct transforms */
354 if ((ret = ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15))) ||
355 (ret = ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15))) ||
356 (ret = ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15)))) {
357 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
361 ff_init_ff_sine_windows(5);
363 ff_atrac_generate_tables();
365 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
367 return AVERROR(ENOMEM);
368 q->vector_fmul_window = fdsp->vector_fmul_window;
371 q->bands[0] = q->low;
372 q->bands[1] = q->mid;
373 q->bands[2] = q->high;
375 /* Prepare the mdct overlap buffers */
376 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
377 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
378 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
379 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
385 AVCodec ff_atrac1_decoder = {
387 .long_name = NULL_IF_CONFIG_SMALL("ATRAC1 (Adaptive TRansform Acoustic Coding)"),
388 .type = AVMEDIA_TYPE_AUDIO,
389 .id = AV_CODEC_ID_ATRAC1,
390 .priv_data_size = sizeof(AT1Ctx),
391 .init = atrac1_decode_init,
392 .close = atrac1_decode_end,
393 .decode = atrac1_decode_frame,
394 .capabilities = AV_CODEC_CAP_DR1,
395 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
396 AV_SAMPLE_FMT_NONE },
397 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,