avformat/mpeg: demux ivtv captions
[ffmpeg.git] / libavcodec / aptxenc.c
blob5e041b9a1b30209ab8e0b19a9d697f57b6e2c8e3
1 /*
2 * Audio Processing Technology codec for Bluetooth (aptX)
4 * Copyright (C) 2017 Aurelien Jacobs <aurel@gnuage.org>
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
23 #include "config_components.h"
25 #include "libavutil/channel_layout.h"
26 #include "aptx.h"
27 #include "audio_frame_queue.h"
28 #include "codec_internal.h"
29 #include "encode.h"
30 #include "internal.h"
32 typedef struct AptXEncContext {
33 AptXContext common;
34 AudioFrameQueue afq;
35 } AptXEncContext;
38 * Half-band QMF analysis filter realized with a polyphase FIR filter.
39 * Split into 2 subbands and downsample by 2.
40 * So for each pair of samples that goes in, one sample goes out,
41 * split into 2 separate subbands.
43 av_always_inline
44 static void aptx_qmf_polyphase_analysis(FilterSignal signal[NB_FILTERS],
45 const int32_t coeffs[NB_FILTERS][FILTER_TAPS],
46 int shift,
47 int32_t samples[NB_FILTERS],
48 int32_t *low_subband_output,
49 int32_t *high_subband_output)
51 int32_t subbands[NB_FILTERS];
52 int i;
54 for (i = 0; i < NB_FILTERS; i++) {
55 aptx_qmf_filter_signal_push(&signal[i], samples[NB_FILTERS-1-i]);
56 subbands[i] = aptx_qmf_convolution(&signal[i], coeffs[i], shift);
59 *low_subband_output = av_clip_intp2(subbands[0] + subbands[1], 23);
60 *high_subband_output = av_clip_intp2(subbands[0] - subbands[1], 23);
64 * Two stage QMF analysis tree.
65 * Split 4 input samples into 4 subbands and downsample by 4.
66 * So for each group of 4 samples that goes in, one sample goes out,
67 * split into 4 separate subbands.
69 static void aptx_qmf_tree_analysis(QMFAnalysis *qmf,
70 int32_t samples[4],
71 int32_t subband_samples[4])
73 int32_t intermediate_samples[4];
74 int i;
76 /* Split 4 input samples into 2 intermediate subbands downsampled to 2 samples */
77 for (i = 0; i < 2; i++)
78 aptx_qmf_polyphase_analysis(qmf->outer_filter_signal,
79 aptx_qmf_outer_coeffs, 23,
80 &samples[2*i],
81 &intermediate_samples[0+i],
82 &intermediate_samples[2+i]);
84 /* Split 2 intermediate subband samples into 4 final subbands downsampled to 1 sample */
85 for (i = 0; i < 2; i++)
86 aptx_qmf_polyphase_analysis(qmf->inner_filter_signal[i],
87 aptx_qmf_inner_coeffs, 23,
88 &intermediate_samples[2*i],
89 &subband_samples[2*i+0],
90 &subband_samples[2*i+1]);
93 av_always_inline
94 static int32_t aptx_bin_search(int32_t value, int32_t factor,
95 const int32_t *intervals, int32_t nb_intervals)
97 int32_t idx = 0;
98 int i;
100 for (i = nb_intervals >> 1; i > 0; i >>= 1)
101 if (MUL64(factor, intervals[idx + i]) <= ((int64_t)value << 24))
102 idx += i;
104 return idx;
107 static void aptx_quantize_difference(Quantize *quantize,
108 int32_t sample_difference,
109 int32_t dither,
110 int32_t quantization_factor,
111 ConstTables *tables)
113 const int32_t *intervals = tables->quantize_intervals;
114 int32_t quantized_sample, dithered_sample, parity_change;
115 int32_t d, mean, interval, inv, sample_difference_abs;
116 int64_t error;
118 sample_difference_abs = FFABS(sample_difference);
119 sample_difference_abs = FFMIN(sample_difference_abs, (1 << 23) - 1);
121 quantized_sample = aptx_bin_search(sample_difference_abs >> 4,
122 quantization_factor,
123 intervals, tables->tables_size);
125 d = rshift32_clip24(MULH(dither, dither), 7) - (1 << 23);
126 d = rshift64(MUL64(d, tables->quantize_dither_factors[quantized_sample]), 23);
128 intervals += quantized_sample;
129 mean = (intervals[1] + intervals[0]) / 2;
130 interval = (intervals[1] - intervals[0]) * (-(sample_difference < 0) | 1);
132 dithered_sample = rshift64_clip24(MUL64(dither, interval) + ((int64_t)av_clip_intp2(mean + d, 23) << 32), 32);
133 error = ((int64_t)sample_difference_abs << 20) - MUL64(dithered_sample, quantization_factor);
134 quantize->error = FFABS(rshift64(error, 23));
136 parity_change = quantized_sample;
137 if (error < 0)
138 quantized_sample--;
139 else
140 parity_change--;
142 inv = -(sample_difference < 0);
143 quantize->quantized_sample = quantized_sample ^ inv;
144 quantize->quantized_sample_parity_change = parity_change ^ inv;
147 static void aptx_encode_channel(Channel *channel, int32_t samples[4], int hd)
149 int32_t subband_samples[4];
150 int subband;
151 aptx_qmf_tree_analysis(&channel->qmf, samples, subband_samples);
152 ff_aptx_generate_dither(channel);
153 for (subband = 0; subband < NB_SUBBANDS; subband++) {
154 int32_t diff = av_clip_intp2(subband_samples[subband] - channel->prediction[subband].predicted_sample, 23);
155 aptx_quantize_difference(&channel->quantize[subband], diff,
156 channel->dither[subband],
157 channel->invert_quantize[subband].quantization_factor,
158 &ff_aptx_quant_tables[hd][subband]);
162 static void aptx_insert_sync(Channel channels[NB_CHANNELS], int32_t *idx)
164 if (aptx_check_parity(channels, idx)) {
165 int i;
166 Channel *c;
167 static const int map[] = { 1, 2, 0, 3 };
168 Quantize *min = &channels[NB_CHANNELS-1].quantize[map[0]];
169 for (c = &channels[NB_CHANNELS-1]; c >= channels; c--)
170 for (i = 0; i < NB_SUBBANDS; i++)
171 if (c->quantize[map[i]].error < min->error)
172 min = &c->quantize[map[i]];
174 /* Forcing the desired parity is done by offsetting by 1 the quantized
175 * sample from the subband featuring the smallest quantization error. */
176 min->quantized_sample = min->quantized_sample_parity_change;
180 static uint16_t aptx_pack_codeword(Channel *channel)
182 int32_t parity = aptx_quantized_parity(channel);
183 return (((channel->quantize[3].quantized_sample & 0x06) | parity) << 13)
184 | (((channel->quantize[2].quantized_sample & 0x03) ) << 11)
185 | (((channel->quantize[1].quantized_sample & 0x0F) ) << 7)
186 | (((channel->quantize[0].quantized_sample & 0x7F) ) << 0);
189 static uint32_t aptxhd_pack_codeword(Channel *channel)
191 int32_t parity = aptx_quantized_parity(channel);
192 return (((channel->quantize[3].quantized_sample & 0x01E) | parity) << 19)
193 | (((channel->quantize[2].quantized_sample & 0x00F) ) << 15)
194 | (((channel->quantize[1].quantized_sample & 0x03F) ) << 9)
195 | (((channel->quantize[0].quantized_sample & 0x1FF) ) << 0);
198 static void aptx_encode_samples(AptXContext *ctx,
199 int32_t samples[NB_CHANNELS][4],
200 uint8_t *output)
202 int channel;
203 for (channel = 0; channel < NB_CHANNELS; channel++)
204 aptx_encode_channel(&ctx->channels[channel], samples[channel], ctx->hd);
206 aptx_insert_sync(ctx->channels, &ctx->sync_idx);
208 for (channel = 0; channel < NB_CHANNELS; channel++) {
209 ff_aptx_invert_quantize_and_prediction(&ctx->channels[channel], ctx->hd);
210 if (ctx->hd)
211 AV_WB24(output + 3*channel,
212 aptxhd_pack_codeword(&ctx->channels[channel]));
213 else
214 AV_WB16(output + 2*channel,
215 aptx_pack_codeword(&ctx->channels[channel]));
219 static int aptx_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
220 const AVFrame *frame, int *got_packet_ptr)
222 AptXEncContext *const s0 = avctx->priv_data;
223 AptXContext *const s = &s0->common;
224 int pos, ipos, channel, sample, output_size, ret;
226 if ((ret = ff_af_queue_add(&s0->afq, frame)) < 0)
227 return ret;
229 output_size = s->block_size * frame->nb_samples/4;
230 if ((ret = ff_get_encode_buffer(avctx, avpkt, output_size, 0)) < 0)
231 return ret;
233 for (pos = 0, ipos = 0; pos < output_size; pos += s->block_size, ipos += 4) {
234 int32_t samples[NB_CHANNELS][4];
236 for (channel = 0; channel < NB_CHANNELS; channel++)
237 for (sample = 0; sample < 4; sample++)
238 samples[channel][sample] = (int32_t)AV_RN32A(&frame->data[channel][4*(ipos+sample)]) >> 8;
240 aptx_encode_samples(s, samples, avpkt->data + pos);
243 ff_af_queue_remove(&s0->afq, frame->nb_samples, &avpkt->pts, &avpkt->duration);
244 *got_packet_ptr = 1;
245 return 0;
248 static av_cold int aptx_close(AVCodecContext *avctx)
250 AptXEncContext *const s = avctx->priv_data;
251 ff_af_queue_close(&s->afq);
252 return 0;
255 static av_cold int aptx_encode_init(AVCodecContext *avctx)
257 AptXEncContext *const s = avctx->priv_data;
259 ff_af_queue_init(avctx, &s->afq);
261 if (!avctx->frame_size || avctx->frame_size % 4)
262 avctx->frame_size = 1024;
263 avctx->internal->pad_samples = 4;
265 return ff_aptx_init(avctx);
268 #if CONFIG_APTX_ENCODER
269 const FFCodec ff_aptx_encoder = {
270 .p.name = "aptx",
271 CODEC_LONG_NAME("aptX (Audio Processing Technology for Bluetooth)"),
272 .p.type = AVMEDIA_TYPE_AUDIO,
273 .p.id = AV_CODEC_ID_APTX,
274 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE,
275 .priv_data_size = sizeof(AptXEncContext),
276 .init = aptx_encode_init,
277 FF_CODEC_ENCODE_CB(aptx_encode_frame),
278 .close = aptx_close,
279 .p.ch_layouts = (const AVChannelLayout[]) { AV_CHANNEL_LAYOUT_STEREO, { 0 } },
280 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
281 AV_SAMPLE_FMT_NONE },
282 .p.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
284 #endif
286 #if CONFIG_APTX_HD_ENCODER
287 const FFCodec ff_aptx_hd_encoder = {
288 .p.name = "aptx_hd",
289 CODEC_LONG_NAME("aptX HD (Audio Processing Technology for Bluetooth)"),
290 .p.type = AVMEDIA_TYPE_AUDIO,
291 .p.id = AV_CODEC_ID_APTX_HD,
292 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE,
293 .priv_data_size = sizeof(AptXEncContext),
294 .init = aptx_encode_init,
295 FF_CODEC_ENCODE_CB(aptx_encode_frame),
296 .close = aptx_close,
297 .p.ch_layouts = (const AVChannelLayout[]) { AV_CHANNEL_LAYOUT_STEREO, { 0 } },
298 .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
299 AV_SAMPLE_FMT_NONE },
300 .p.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
302 #endif