3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5 * Copyright (c) 2008-2013 Alex Converse <alex.converse@gmail.com>
8 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
11 * This file is part of FFmpeg.
13 * FFmpeg is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU Lesser General Public
15 * License as published by the Free Software Foundation; either
16 * version 2.1 of the License, or (at your option) any later version.
18 * FFmpeg is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * Lesser General Public License for more details.
23 * You should have received a copy of the GNU Lesser General Public
24 * License along with FFmpeg; if not, write to the Free Software
25 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
31 * @author Oded Shimon ( ods15 ods15 dyndns org )
32 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
36 #define FFT_FIXED_32 0
39 #include "libavutil/float_dsp.h"
40 #include "libavutil/opt.h"
52 #include "aacdectab.h"
53 #include "adts_header.h"
54 #include "cbrt_data.h"
57 #include "mpeg4audio.h"
59 #include "libavutil/intfloat.h"
69 # include "mips/aacdec_mips.h"
72 DECLARE_ALIGNED(32, static INTFLOAT, AAC_RENAME(sine_120))[120];
73 DECLARE_ALIGNED(32, static INTFLOAT, AAC_RENAME(sine_960))[960];
75 static av_always_inline void reset_predict_state(PredictorState *ps)
86 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
90 *dst++ = v[idx & 15] * s;
91 *dst++ = v[idx>>4 & 15] * s;
97 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
101 *dst++ = v[idx & 3] * s;
102 *dst++ = v[idx>>2 & 3] * s;
103 *dst++ = v[idx>>4 & 3] * s;
104 *dst++ = v[idx>>6 & 3] * s;
110 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
111 unsigned sign, const float *scale)
113 union av_intfloat32 s0, s1;
115 s0.f = s1.f = *scale;
116 s0.i ^= sign >> 1 << 31;
119 *dst++ = v[idx & 15] * s0.f;
120 *dst++ = v[idx>>4 & 15] * s1.f;
127 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
128 unsigned sign, const float *scale)
130 unsigned nz = idx >> 12;
131 union av_intfloat32 s = { .f = *scale };
132 union av_intfloat32 t;
134 t.i = s.i ^ (sign & 1U<<31);
135 *dst++ = v[idx & 3] * t.f;
137 sign <<= nz & 1; nz >>= 1;
138 t.i = s.i ^ (sign & 1U<<31);
139 *dst++ = v[idx>>2 & 3] * t.f;
141 sign <<= nz & 1; nz >>= 1;
142 t.i = s.i ^ (sign & 1U<<31);
143 *dst++ = v[idx>>4 & 3] * t.f;
146 t.i = s.i ^ (sign & 1U<<31);
147 *dst++ = v[idx>>6 & 3] * t.f;
153 static av_always_inline float flt16_round(float pf)
155 union av_intfloat32 tmp;
157 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
161 static av_always_inline float flt16_even(float pf)
163 union av_intfloat32 tmp;
165 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
169 static av_always_inline float flt16_trunc(float pf)
171 union av_intfloat32 pun;
173 pun.i &= 0xFFFF0000U;
177 static av_always_inline void predict(PredictorState *ps, float *coef,
180 const float a = 0.953125; // 61.0 / 64
181 const float alpha = 0.90625; // 29.0 / 32
185 float r0 = ps->r0, r1 = ps->r1;
186 float cor0 = ps->cor0, cor1 = ps->cor1;
187 float var0 = ps->var0, var1 = ps->var1;
189 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
190 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
192 pv = flt16_round(k1 * r0 + k2 * r1);
199 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
200 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
201 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
202 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
204 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
205 ps->r0 = flt16_trunc(a * e0);
209 * Apply dependent channel coupling (applied before IMDCT).
211 * @param index index into coupling gain array
213 static void apply_dependent_coupling(AACContext *ac,
214 SingleChannelElement *target,
215 ChannelElement *cce, int index)
217 IndividualChannelStream *ics = &cce->ch[0].ics;
218 const uint16_t *offsets = ics->swb_offset;
219 float *dest = target->coeffs;
220 const float *src = cce->ch[0].coeffs;
221 int g, i, group, k, idx = 0;
222 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
223 av_log(ac->avctx, AV_LOG_ERROR,
224 "Dependent coupling is not supported together with LTP\n");
227 for (g = 0; g < ics->num_window_groups; g++) {
228 for (i = 0; i < ics->max_sfb; i++, idx++) {
229 if (cce->ch[0].band_type[idx] != ZERO_BT) {
230 const float gain = cce->coup.gain[index][idx];
231 for (group = 0; group < ics->group_len[g]; group++) {
232 for (k = offsets[i]; k < offsets[i + 1]; k++) {
234 dest[group * 128 + k] += gain * src[group * 128 + k];
239 dest += ics->group_len[g] * 128;
240 src += ics->group_len[g] * 128;
245 * Apply independent channel coupling (applied after IMDCT).
247 * @param index index into coupling gain array
249 static void apply_independent_coupling(AACContext *ac,
250 SingleChannelElement *target,
251 ChannelElement *cce, int index)
253 const float gain = cce->coup.gain[index][0];
254 const float *src = cce->ch[0].ret;
255 float *dest = target->ret;
256 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
258 ac->fdsp->vector_fmac_scalar(dest, src, gain, len);
261 #include "aacdec_template.c"
263 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
266 AACContext aac_ctx; ///< containing AACContext
267 int initialized; ///< initialized after a valid extradata was seen
270 int audio_mux_version_A; ///< LATM syntax version
271 int frame_length_type; ///< 0/1 variable/fixed frame length
272 int frame_length; ///< frame length for fixed frame length
275 static inline uint32_t latm_get_value(GetBitContext *b)
277 int length = get_bits(b, 2);
279 return get_bits_long(b, (length+1)*8);
282 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
283 GetBitContext *gb, int asclen)
285 AACContext *ac = &latmctx->aac_ctx;
286 AVCodecContext *avctx = ac->avctx;
287 MPEG4AudioConfig m4ac = { 0 };
289 int config_start_bit = get_bits_count(gb);
290 int sync_extension = 0;
291 int bits_consumed, esize, i;
295 asclen = FFMIN(asclen, get_bits_left(gb));
296 init_get_bits(&gbc, gb->buffer, config_start_bit + asclen);
297 skip_bits_long(&gbc, config_start_bit);
298 } else if (asclen == 0) {
301 return AVERROR_INVALIDDATA;
304 if (get_bits_left(gb) <= 0)
305 return AVERROR_INVALIDDATA;
307 bits_consumed = decode_audio_specific_config_gb(NULL, avctx, &m4ac,
308 &gbc, config_start_bit,
311 if (bits_consumed < config_start_bit)
312 return AVERROR_INVALIDDATA;
313 bits_consumed -= config_start_bit;
316 asclen = bits_consumed;
318 if (!latmctx->initialized ||
319 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
320 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
322 if (latmctx->initialized) {
323 av_log(avctx, AV_LOG_INFO, "audio config changed (sample_rate=%d, chan_config=%d)\n", m4ac.sample_rate, m4ac.chan_config);
325 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
327 latmctx->initialized = 0;
329 esize = (asclen + 7) / 8;
331 if (avctx->extradata_size < esize) {
332 av_free(avctx->extradata);
333 avctx->extradata = av_malloc(esize + AV_INPUT_BUFFER_PADDING_SIZE);
334 if (!avctx->extradata)
335 return AVERROR(ENOMEM);
338 avctx->extradata_size = esize;
340 for (i = 0; i < esize; i++) {
341 avctx->extradata[i] = get_bits(&gbc, 8);
343 memset(avctx->extradata+esize, 0, AV_INPUT_BUFFER_PADDING_SIZE);
345 skip_bits_long(gb, asclen);
350 static int read_stream_mux_config(struct LATMContext *latmctx,
353 int ret, audio_mux_version = get_bits(gb, 1);
355 latmctx->audio_mux_version_A = 0;
356 if (audio_mux_version)
357 latmctx->audio_mux_version_A = get_bits(gb, 1);
359 if (!latmctx->audio_mux_version_A) {
361 if (audio_mux_version)
362 latm_get_value(gb); // taraFullness
364 skip_bits(gb, 1); // allStreamSameTimeFraming
365 skip_bits(gb, 6); // numSubFrames
367 if (get_bits(gb, 4)) { // numPrograms
368 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
369 return AVERROR_PATCHWELCOME;
372 // for each program (which there is only one in DVB)
374 // for each layer (which there is only one in DVB)
375 if (get_bits(gb, 3)) { // numLayer
376 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
377 return AVERROR_PATCHWELCOME;
380 // for all but first stream: use_same_config = get_bits(gb, 1);
381 if (!audio_mux_version) {
382 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
385 int ascLen = latm_get_value(gb);
386 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
390 latmctx->frame_length_type = get_bits(gb, 3);
391 switch (latmctx->frame_length_type) {
393 skip_bits(gb, 8); // latmBufferFullness
396 latmctx->frame_length = get_bits(gb, 9);
401 skip_bits(gb, 6); // CELP frame length table index
405 skip_bits(gb, 1); // HVXC frame length table index
409 if (get_bits(gb, 1)) { // other data
410 if (audio_mux_version) {
411 latm_get_value(gb); // other_data_bits
415 if (get_bits_left(gb) < 9)
416 return AVERROR_INVALIDDATA;
417 esc = get_bits(gb, 1);
423 if (get_bits(gb, 1)) // crc present
424 skip_bits(gb, 8); // config_crc
430 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
434 if (ctx->frame_length_type == 0) {
435 int mux_slot_length = 0;
437 if (get_bits_left(gb) < 8)
438 return AVERROR_INVALIDDATA;
439 tmp = get_bits(gb, 8);
440 mux_slot_length += tmp;
441 } while (tmp == 255);
442 return mux_slot_length;
443 } else if (ctx->frame_length_type == 1) {
444 return ctx->frame_length;
445 } else if (ctx->frame_length_type == 3 ||
446 ctx->frame_length_type == 5 ||
447 ctx->frame_length_type == 7) {
448 skip_bits(gb, 2); // mux_slot_length_coded
453 static int read_audio_mux_element(struct LATMContext *latmctx,
457 uint8_t use_same_mux = get_bits(gb, 1);
459 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
461 } else if (!latmctx->aac_ctx.avctx->extradata) {
462 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
463 "no decoder config found\n");
466 if (latmctx->audio_mux_version_A == 0) {
467 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
468 if (mux_slot_length_bytes < 0 || mux_slot_length_bytes * 8LL > get_bits_left(gb)) {
469 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
470 return AVERROR_INVALIDDATA;
471 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
472 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
473 "frame length mismatch %d << %d\n",
474 mux_slot_length_bytes * 8, get_bits_left(gb));
475 return AVERROR_INVALIDDATA;
482 static int latm_decode_frame(AVCodecContext *avctx, void *out,
483 int *got_frame_ptr, AVPacket *avpkt)
485 struct LATMContext *latmctx = avctx->priv_data;
489 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
492 // check for LOAS sync word
493 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
494 return AVERROR_INVALIDDATA;
496 muxlength = get_bits(&gb, 13) + 3;
497 // not enough data, the parser should have sorted this out
498 if (muxlength > avpkt->size)
499 return AVERROR_INVALIDDATA;
501 if ((err = read_audio_mux_element(latmctx, &gb)))
502 return (err < 0) ? err : avpkt->size;
504 if (!latmctx->initialized) {
505 if (!avctx->extradata) {
509 push_output_configuration(&latmctx->aac_ctx);
510 if ((err = decode_audio_specific_config(
511 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
512 avctx->extradata, avctx->extradata_size*8LL, 1)) < 0) {
513 pop_output_configuration(&latmctx->aac_ctx);
516 latmctx->initialized = 1;
520 if (show_bits(&gb, 12) == 0xfff) {
521 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
522 "ADTS header detected, probably as result of configuration "
524 return AVERROR_INVALIDDATA;
527 switch (latmctx->aac_ctx.oc[1].m4ac.object_type) {
532 err = aac_decode_er_frame(avctx, out, got_frame_ptr, &gb);
535 err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt);
543 static av_cold int latm_decode_init(AVCodecContext *avctx)
545 struct LATMContext *latmctx = avctx->priv_data;
546 int ret = aac_decode_init(avctx);
548 if (avctx->extradata_size > 0)
549 latmctx->initialized = !ret;
554 AVCodec ff_aac_decoder = {
556 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
557 .type = AVMEDIA_TYPE_AUDIO,
558 .id = AV_CODEC_ID_AAC,
559 .priv_data_size = sizeof(AACContext),
560 .init = aac_decode_init,
561 .close = aac_decode_close,
562 .decode = aac_decode_frame,
563 .sample_fmts = (const enum AVSampleFormat[]) {
564 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
566 .capabilities = AV_CODEC_CAP_CHANNEL_CONF | AV_CODEC_CAP_DR1,
567 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
568 .channel_layouts = aac_channel_layout,
570 .priv_class = &aac_decoder_class,
571 .profiles = NULL_IF_CONFIG_SMALL(ff_aac_profiles),
575 Note: This decoder filter is intended to decode LATM streams transferred
576 in MPEG transport streams which only contain one program.
577 To do a more complex LATM demuxing a separate LATM demuxer should be used.
579 AVCodec ff_aac_latm_decoder = {
581 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
582 .type = AVMEDIA_TYPE_AUDIO,
583 .id = AV_CODEC_ID_AAC_LATM,
584 .priv_data_size = sizeof(struct LATMContext),
585 .init = latm_decode_init,
586 .close = aac_decode_close,
587 .decode = latm_decode_frame,
588 .sample_fmts = (const enum AVSampleFormat[]) {
589 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
591 .capabilities = AV_CODEC_CAP_CHANNEL_CONF | AV_CODEC_CAP_DR1,
592 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
593 .channel_layouts = aac_channel_layout,
595 .profiles = NULL_IF_CONFIG_SMALL(ff_aac_profiles),