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
25 * Atrac 1 compatible decoder.
26 * This decoder handles raw ATRAC1 data and probably SDDS data.
29 /* Many thanks to Tim Craig for all the help! */
39 #include "fmtconvert.h"
43 #include "atrac1data.h"
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
52 #define AT1_QMF_BANDS 3
53 #define IDX_LOW_BAND 0
54 #define IDX_MID_BAND 1
55 #define IDX_HIGH_BAND 2
58 * Sound unit struct, one unit is used per channel
61 int log2_block_count[AT1_QMF_BANDS]; ///< log2 number of blocks in a band
62 int num_bfus; ///< number of Block Floating Units
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
72 * The atrac1 context, holds all needed parameters for decoding
76 AT1SUCtx SUs[AT1_MAX_CHANNELS]; ///< channel sound unit
77 DECLARE_ALIGNED(32, float, spec)[AT1_SU_SAMPLES]; ///< the mdct spectrum buffer
79 DECLARE_ALIGNED(32, float, low)[256];
80 DECLARE_ALIGNED(32, float, mid)[256];
81 DECLARE_ALIGNED(32, float, high)[512];
83 float *out_samples[AT1_MAX_CHANNELS];
84 FFTContext mdct_ctx[3];
87 FmtConvertContext fmt_conv;
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 FFTContext* mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];
99 int transf_size = 1 << nbits;
103 for (i = 0; i < transf_size / 2; i++)
104 FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
106 mdct_context->imdct_half(mdct_context, out, spec);
110 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
112 int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
113 unsigned int start_pos, ref_pos = 0, pos = 0;
115 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
119 band_samples = samples_per_band[band_num];
120 log2_block_count = su->log2_block_count[band_num];
122 /* number of mdct blocks in the current QMF band: 1 - for long mode */
123 /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
124 num_blocks = 1 << log2_block_count;
126 if (num_blocks == 1) {
127 /* mdct block size in samples: 128 (long mode, low & mid bands), */
128 /* 256 (long mode, high band) and 32 (short mode, all bands) */
129 block_size = band_samples >> log2_block_count;
131 /* calc transform size in bits according to the block_size_mode */
132 nbits = mdct_long_nbits[band_num] - log2_block_count;
134 if (nbits != 5 && nbits != 7 && nbits != 8)
135 return AVERROR_INVALIDDATA;
142 prev_buf = &su->spectrum[1][ref_pos + band_samples - 16];
143 for (j=0; j < num_blocks; j++) {
144 at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos + start_pos], nbits, band_num);
146 /* overlap and window */
147 q->dsp.vector_fmul_window(&q->bands[band_num][start_pos], prev_buf,
148 &su->spectrum[0][ref_pos + start_pos], ff_sine_32, 16);
150 prev_buf = &su->spectrum[0][ref_pos+start_pos + 16];
151 start_pos += block_size;
156 memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float));
158 ref_pos += band_samples;
161 /* Swap buffers so the mdct overlap works */
162 FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
168 * Parse the block size mode byte
171 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
173 int log2_block_count_tmp, i;
175 for (i = 0; i < 2; i++) {
176 /* low and mid band */
177 log2_block_count_tmp = get_bits(gb, 2);
178 if (log2_block_count_tmp & 1)
179 return AVERROR_INVALIDDATA;
180 log2_block_cnt[i] = 2 - log2_block_count_tmp;
184 log2_block_count_tmp = get_bits(gb, 2);
185 if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
186 return AVERROR_INVALIDDATA;
187 log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
194 static int at1_unpack_dequant(GetBitContext* gb, AT1SUCtx* su,
195 float spec[AT1_SU_SAMPLES])
197 int bits_used, band_num, bfu_num, i;
198 uint8_t idwls[AT1_MAX_BFU]; ///< the word length indexes for each BFU
199 uint8_t idsfs[AT1_MAX_BFU]; ///< the scalefactor indexes for each BFU
201 /* parse the info byte (2nd byte) telling how much BFUs were coded */
202 su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
204 /* calc number of consumed bits:
205 num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
206 + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
207 bits_used = su->num_bfus * 10 + 32 +
208 bfu_amount_tab2[get_bits(gb, 2)] +
209 (bfu_amount_tab3[get_bits(gb, 3)] << 1);
211 /* get word length index (idwl) for each BFU */
212 for (i = 0; i < su->num_bfus; i++)
213 idwls[i] = get_bits(gb, 4);
215 /* get scalefactor index (idsf) for each BFU */
216 for (i = 0; i < su->num_bfus; i++)
217 idsfs[i] = get_bits(gb, 6);
219 /* zero idwl/idsf for empty BFUs */
220 for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
221 idwls[i] = idsfs[i] = 0;
223 /* read in the spectral data and reconstruct MDCT spectrum of this channel */
224 for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
225 for (bfu_num = bfu_bands_t[band_num]; bfu_num < bfu_bands_t[band_num+1]; bfu_num++) {
228 int num_specs = specs_per_bfu[bfu_num];
229 int word_len = !!idwls[bfu_num] + idwls[bfu_num];
230 float scale_factor = ff_atrac_sf_table[idsfs[bfu_num]];
231 bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
233 /* check for bitstream overflow */
234 if (bits_used > AT1_SU_MAX_BITS)
235 return AVERROR_INVALIDDATA;
237 /* get the position of the 1st spec according to the block size mode */
238 pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
241 float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
243 for (i = 0; i < num_specs; i++) {
244 /* read in a quantized spec and convert it to
245 * signed int and then inverse quantization
247 spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
249 } else { /* word_len = 0 -> empty BFU, zero all specs in the empty BFU */
250 memset(&spec[pos], 0, num_specs * sizeof(float));
259 static void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
262 float iqmf_temp[512 + 46];
264 /* combine low and middle bands */
265 ff_atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
267 /* delay the signal of the high band by 23 samples */
268 memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 23);
269 memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float) * 256);
271 /* combine (low + middle) and high bands */
272 ff_atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
276 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
277 int *got_frame_ptr, AVPacket *avpkt)
279 const uint8_t *buf = avpkt->data;
280 int buf_size = avpkt->size;
281 AT1Ctx *q = avctx->priv_data;
287 if (buf_size < 212 * q->channels) {
288 av_log(avctx, AV_LOG_ERROR, "Not enough data to decode!\n");
289 return AVERROR_INVALIDDATA;
292 /* get output buffer */
293 q->frame.nb_samples = AT1_SU_SAMPLES;
294 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
295 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
298 samples = (float *)q->frame.data[0];
300 for (ch = 0; ch < q->channels; ch++) {
301 AT1SUCtx* su = &q->SUs[ch];
303 init_get_bits(&gb, &buf[212 * ch], 212 * 8);
305 /* parse block_size_mode, 1st byte */
306 ret = at1_parse_bsm(&gb, su->log2_block_count);
310 ret = at1_unpack_dequant(&gb, su, q->spec);
314 ret = at1_imdct_block(su, q);
317 at1_subband_synthesis(q, su, q->channels == 1 ? samples : q->out_samples[ch]);
321 if (q->channels == 2) {
322 q->fmt_conv.float_interleave(samples, (const float **)q->out_samples,
327 *(AVFrame *)data = q->frame;
329 return avctx->block_align;
333 static av_cold int atrac1_decode_end(AVCodecContext * avctx)
335 AT1Ctx *q = avctx->priv_data;
337 av_freep(&q->out_samples[0]);
339 ff_mdct_end(&q->mdct_ctx[0]);
340 ff_mdct_end(&q->mdct_ctx[1]);
341 ff_mdct_end(&q->mdct_ctx[2]);
347 static av_cold int atrac1_decode_init(AVCodecContext *avctx)
349 AT1Ctx *q = avctx->priv_data;
352 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
354 if (avctx->channels < 1 || avctx->channels > AT1_MAX_CHANNELS) {
355 av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n",
357 return AVERROR(EINVAL);
359 q->channels = avctx->channels;
361 if (avctx->channels == 2) {
362 q->out_samples[0] = av_malloc(2 * AT1_SU_SAMPLES * sizeof(*q->out_samples[0]));
363 q->out_samples[1] = q->out_samples[0] + AT1_SU_SAMPLES;
364 if (!q->out_samples[0]) {
365 av_freep(&q->out_samples[0]);
366 return AVERROR(ENOMEM);
370 /* Init the mdct transforms */
371 if ((ret = ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15))) ||
372 (ret = ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15))) ||
373 (ret = ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15)))) {
374 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
375 atrac1_decode_end(avctx);
379 ff_init_ff_sine_windows(5);
381 ff_atrac_generate_tables();
383 ff_dsputil_init(&q->dsp, avctx);
384 ff_fmt_convert_init(&q->fmt_conv, avctx);
386 q->bands[0] = q->low;
387 q->bands[1] = q->mid;
388 q->bands[2] = q->high;
390 /* Prepare the mdct overlap buffers */
391 q->SUs[0].spectrum[0] = q->SUs[0].spec1;
392 q->SUs[0].spectrum[1] = q->SUs[0].spec2;
393 q->SUs[1].spectrum[0] = q->SUs[1].spec1;
394 q->SUs[1].spectrum[1] = q->SUs[1].spec2;
396 avcodec_get_frame_defaults(&q->frame);
397 avctx->coded_frame = &q->frame;
403 AVCodec ff_atrac1_decoder = {
405 .type = AVMEDIA_TYPE_AUDIO,
406 .id = CODEC_ID_ATRAC1,
407 .priv_data_size = sizeof(AT1Ctx),
408 .init = atrac1_decode_init,
409 .close = atrac1_decode_end,
410 .decode = atrac1_decode_frame,
411 .capabilities = CODEC_CAP_DR1,
412 .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),