yading@10: /* yading@10: * Simple free lossless/lossy audio codec yading@10: * Copyright (c) 2004 Alex Beregszaszi yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: #include "avcodec.h" yading@10: #include "get_bits.h" yading@10: #include "golomb.h" yading@10: #include "internal.h" yading@10: yading@10: /** yading@10: * @file yading@10: * Simple free lossless/lossy audio codec yading@10: * Based on Paul Francis Harrison's Bonk (http://www.logarithmic.net/pfh/bonk) yading@10: * Written and designed by Alex Beregszaszi yading@10: * yading@10: * TODO: yading@10: * - CABAC put/get_symbol yading@10: * - independent quantizer for channels yading@10: * - >2 channels support yading@10: * - more decorrelation types yading@10: * - more tap_quant tests yading@10: * - selectable intlist writers/readers (bonk-style, golomb, cabac) yading@10: */ yading@10: yading@10: #define MAX_CHANNELS 2 yading@10: yading@10: #define MID_SIDE 0 yading@10: #define LEFT_SIDE 1 yading@10: #define RIGHT_SIDE 2 yading@10: yading@10: typedef struct SonicContext { yading@10: int lossless, decorrelation; yading@10: yading@10: int num_taps, downsampling; yading@10: double quantization; yading@10: yading@10: int channels, samplerate, block_align, frame_size; yading@10: yading@10: int *tap_quant; yading@10: int *int_samples; yading@10: int *coded_samples[MAX_CHANNELS]; yading@10: yading@10: // for encoding yading@10: int *tail; yading@10: int tail_size; yading@10: int *window; yading@10: int window_size; yading@10: yading@10: // for decoding yading@10: int *predictor_k; yading@10: int *predictor_state[MAX_CHANNELS]; yading@10: } SonicContext; yading@10: yading@10: #define LATTICE_SHIFT 10 yading@10: #define SAMPLE_SHIFT 4 yading@10: #define LATTICE_FACTOR (1 << LATTICE_SHIFT) yading@10: #define SAMPLE_FACTOR (1 << SAMPLE_SHIFT) yading@10: yading@10: #define BASE_QUANT 0.6 yading@10: #define RATE_VARIATION 3.0 yading@10: yading@10: static inline int divide(int a, int b) yading@10: { yading@10: if (a < 0) yading@10: return -( (-a + b/2)/b ); yading@10: else yading@10: return (a + b/2)/b; yading@10: } yading@10: yading@10: static inline int shift(int a,int b) yading@10: { yading@10: return (a+(1<<(b-1))) >> b; yading@10: } yading@10: yading@10: static inline int shift_down(int a,int b) yading@10: { yading@10: return (a>>b)+((a<0)?1:0); yading@10: } yading@10: yading@10: #if 1 yading@10: static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part) yading@10: { yading@10: int i; yading@10: yading@10: for (i = 0; i < entries; i++) yading@10: set_se_golomb(pb, buf[i]); yading@10: yading@10: return 1; yading@10: } yading@10: yading@10: static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part) yading@10: { yading@10: int i; yading@10: yading@10: for (i = 0; i < entries; i++) yading@10: buf[i] = get_se_golomb(gb); yading@10: yading@10: return 1; yading@10: } yading@10: yading@10: #else yading@10: yading@10: #define ADAPT_LEVEL 8 yading@10: yading@10: static int bits_to_store(uint64_t x) yading@10: { yading@10: int res = 0; yading@10: yading@10: while(x) yading@10: { yading@10: res++; yading@10: x >>= 1; yading@10: } yading@10: return res; yading@10: } yading@10: yading@10: static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max) yading@10: { yading@10: int i, bits; yading@10: yading@10: if (!max) yading@10: return; yading@10: yading@10: bits = bits_to_store(max); yading@10: yading@10: for (i = 0; i < bits-1; i++) yading@10: put_bits(pb, 1, value & (1 << i)); yading@10: yading@10: if ( (value | (1 << (bits-1))) <= max) yading@10: put_bits(pb, 1, value & (1 << (bits-1))); yading@10: } yading@10: yading@10: static unsigned int read_uint_max(GetBitContext *gb, int max) yading@10: { yading@10: int i, bits, value = 0; yading@10: yading@10: if (!max) yading@10: return 0; yading@10: yading@10: bits = bits_to_store(max); yading@10: yading@10: for (i = 0; i < bits-1; i++) yading@10: if (get_bits1(gb)) yading@10: value += 1 << i; yading@10: yading@10: if ( (value | (1<<(bits-1))) <= max) yading@10: if (get_bits1(gb)) yading@10: value += 1 << (bits-1); yading@10: yading@10: return value; yading@10: } yading@10: yading@10: static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part) yading@10: { yading@10: int i, j, x = 0, low_bits = 0, max = 0; yading@10: int step = 256, pos = 0, dominant = 0, any = 0; yading@10: int *copy, *bits; yading@10: yading@10: copy = av_mallocz(4* entries); yading@10: if (!copy) yading@10: return -1; yading@10: yading@10: if (base_2_part) yading@10: { yading@10: int energy = 0; yading@10: yading@10: for (i = 0; i < entries; i++) yading@10: energy += abs(buf[i]); yading@10: yading@10: low_bits = bits_to_store(energy / (entries * 2)); yading@10: if (low_bits > 15) yading@10: low_bits = 15; yading@10: yading@10: put_bits(pb, 4, low_bits); yading@10: } yading@10: yading@10: for (i = 0; i < entries; i++) yading@10: { yading@10: put_bits(pb, low_bits, abs(buf[i])); yading@10: copy[i] = abs(buf[i]) >> low_bits; yading@10: if (copy[i] > max) yading@10: max = abs(copy[i]); yading@10: } yading@10: yading@10: bits = av_mallocz(4* entries*max); yading@10: if (!bits) yading@10: { yading@10: // av_free(copy); yading@10: return -1; yading@10: } yading@10: yading@10: for (i = 0; i <= max; i++) yading@10: { yading@10: for (j = 0; j < entries; j++) yading@10: if (copy[j] >= i) yading@10: bits[x++] = copy[j] > i; yading@10: } yading@10: yading@10: // store bitstream yading@10: while (pos < x) yading@10: { yading@10: int steplet = step >> 8; yading@10: yading@10: if (pos + steplet > x) yading@10: steplet = x - pos; yading@10: yading@10: for (i = 0; i < steplet; i++) yading@10: if (bits[i+pos] != dominant) yading@10: any = 1; yading@10: yading@10: put_bits(pb, 1, any); yading@10: yading@10: if (!any) yading@10: { yading@10: pos += steplet; yading@10: step += step / ADAPT_LEVEL; yading@10: } yading@10: else yading@10: { yading@10: int interloper = 0; yading@10: yading@10: while (((pos + interloper) < x) && (bits[pos + interloper] == dominant)) yading@10: interloper++; yading@10: yading@10: // note change yading@10: write_uint_max(pb, interloper, (step >> 8) - 1); yading@10: yading@10: pos += interloper + 1; yading@10: step -= step / ADAPT_LEVEL; yading@10: } yading@10: yading@10: if (step < 256) yading@10: { yading@10: step = 65536 / step; yading@10: dominant = !dominant; yading@10: } yading@10: } yading@10: yading@10: // store signs yading@10: for (i = 0; i < entries; i++) yading@10: if (buf[i]) yading@10: put_bits(pb, 1, buf[i] < 0); yading@10: yading@10: // av_free(bits); yading@10: // av_free(copy); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part) yading@10: { yading@10: int i, low_bits = 0, x = 0; yading@10: int n_zeros = 0, step = 256, dominant = 0; yading@10: int pos = 0, level = 0; yading@10: int *bits = av_mallocz(4* entries); yading@10: yading@10: if (!bits) yading@10: return -1; yading@10: yading@10: if (base_2_part) yading@10: { yading@10: low_bits = get_bits(gb, 4); yading@10: yading@10: if (low_bits) yading@10: for (i = 0; i < entries; i++) yading@10: buf[i] = get_bits(gb, low_bits); yading@10: } yading@10: yading@10: // av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits); yading@10: yading@10: while (n_zeros < entries) yading@10: { yading@10: int steplet = step >> 8; yading@10: yading@10: if (!get_bits1(gb)) yading@10: { yading@10: for (i = 0; i < steplet; i++) yading@10: bits[x++] = dominant; yading@10: yading@10: if (!dominant) yading@10: n_zeros += steplet; yading@10: yading@10: step += step / ADAPT_LEVEL; yading@10: } yading@10: else yading@10: { yading@10: int actual_run = read_uint_max(gb, steplet-1); yading@10: yading@10: // av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run); yading@10: yading@10: for (i = 0; i < actual_run; i++) yading@10: bits[x++] = dominant; yading@10: yading@10: bits[x++] = !dominant; yading@10: yading@10: if (!dominant) yading@10: n_zeros += actual_run; yading@10: else yading@10: n_zeros++; yading@10: yading@10: step -= step / ADAPT_LEVEL; yading@10: } yading@10: yading@10: if (step < 256) yading@10: { yading@10: step = 65536 / step; yading@10: dominant = !dominant; yading@10: } yading@10: } yading@10: yading@10: // reconstruct unsigned values yading@10: n_zeros = 0; yading@10: for (i = 0; n_zeros < entries; i++) yading@10: { yading@10: while(1) yading@10: { yading@10: if (pos >= entries) yading@10: { yading@10: pos = 0; yading@10: level += 1 << low_bits; yading@10: } yading@10: yading@10: if (buf[pos] >= level) yading@10: break; yading@10: yading@10: pos++; yading@10: } yading@10: yading@10: if (bits[i]) yading@10: buf[pos] += 1 << low_bits; yading@10: else yading@10: n_zeros++; yading@10: yading@10: pos++; yading@10: } yading@10: // av_free(bits); yading@10: yading@10: // read signs yading@10: for (i = 0; i < entries; i++) yading@10: if (buf[i] && get_bits1(gb)) yading@10: buf[i] = -buf[i]; yading@10: yading@10: // av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos); yading@10: yading@10: return 0; yading@10: } yading@10: #endif yading@10: yading@10: static void predictor_init_state(int *k, int *state, int order) yading@10: { yading@10: int i; yading@10: yading@10: for (i = order-2; i >= 0; i--) yading@10: { yading@10: int j, p, x = state[i]; yading@10: yading@10: for (j = 0, p = i+1; p < order; j++,p++) yading@10: { yading@10: int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT); yading@10: state[p] += shift_down(k[j]*x, LATTICE_SHIFT); yading@10: x = tmp; yading@10: } yading@10: } yading@10: } yading@10: yading@10: static int predictor_calc_error(int *k, int *state, int order, int error) yading@10: { yading@10: int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT); yading@10: yading@10: #if 1 yading@10: int *k_ptr = &(k[order-2]), yading@10: *state_ptr = &(state[order-2]); yading@10: for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--) yading@10: { yading@10: int k_value = *k_ptr, state_value = *state_ptr; yading@10: x -= shift_down(k_value * state_value, LATTICE_SHIFT); yading@10: state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT); yading@10: } yading@10: #else yading@10: for (i = order-2; i >= 0; i--) yading@10: { yading@10: x -= shift_down(k[i] * state[i], LATTICE_SHIFT); yading@10: state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT); yading@10: } yading@10: #endif yading@10: yading@10: // don't drift too far, to avoid overflows yading@10: if (x > (SAMPLE_FACTOR<<16)) x = (SAMPLE_FACTOR<<16); yading@10: if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16); yading@10: yading@10: state[0] = x; yading@10: yading@10: return x; yading@10: } yading@10: yading@10: #if CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER yading@10: // Heavily modified Levinson-Durbin algorithm which yading@10: // copes better with quantization, and calculates the yading@10: // actual whitened result as it goes. yading@10: yading@10: static void modified_levinson_durbin(int *window, int window_entries, yading@10: int *out, int out_entries, int channels, int *tap_quant) yading@10: { yading@10: int i; yading@10: int *state = av_mallocz(4* window_entries); yading@10: yading@10: memcpy(state, window, 4* window_entries); yading@10: yading@10: for (i = 0; i < out_entries; i++) yading@10: { yading@10: int step = (i+1)*channels, k, j; yading@10: double xx = 0.0, xy = 0.0; yading@10: #if 1 yading@10: int *x_ptr = &(window[step]), *state_ptr = &(state[0]); yading@10: j = window_entries - step; yading@10: for (;j>=0;j--,x_ptr++,state_ptr++) yading@10: { yading@10: double x_value = *x_ptr, state_value = *state_ptr; yading@10: xx += state_value*state_value; yading@10: xy += x_value*state_value; yading@10: } yading@10: #else yading@10: for (j = 0; j <= (window_entries - step); j++); yading@10: { yading@10: double stepval = window[step+j], stateval = window[j]; yading@10: // xx += (double)window[j]*(double)window[j]; yading@10: // xy += (double)window[step+j]*(double)window[j]; yading@10: xx += stateval*stateval; yading@10: xy += stepval*stateval; yading@10: } yading@10: #endif yading@10: if (xx == 0.0) yading@10: k = 0; yading@10: else yading@10: k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5)); yading@10: yading@10: if (k > (LATTICE_FACTOR/tap_quant[i])) yading@10: k = LATTICE_FACTOR/tap_quant[i]; yading@10: if (-k > (LATTICE_FACTOR/tap_quant[i])) yading@10: k = -(LATTICE_FACTOR/tap_quant[i]); yading@10: yading@10: out[i] = k; yading@10: k *= tap_quant[i]; yading@10: yading@10: #if 1 yading@10: x_ptr = &(window[step]); yading@10: state_ptr = &(state[0]); yading@10: j = window_entries - step; yading@10: for (;j>=0;j--,x_ptr++,state_ptr++) yading@10: { yading@10: int x_value = *x_ptr, state_value = *state_ptr; yading@10: *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT); yading@10: *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT); yading@10: } yading@10: #else yading@10: for (j=0; j <= (window_entries - step); j++) yading@10: { yading@10: int stepval = window[step+j], stateval=state[j]; yading@10: window[step+j] += shift_down(k * stateval, LATTICE_SHIFT); yading@10: state[j] += shift_down(k * stepval, LATTICE_SHIFT); yading@10: } yading@10: #endif yading@10: } yading@10: yading@10: av_free(state); yading@10: } yading@10: yading@10: static inline int code_samplerate(int samplerate) yading@10: { yading@10: switch (samplerate) yading@10: { yading@10: case 44100: return 0; yading@10: case 22050: return 1; yading@10: case 11025: return 2; yading@10: case 96000: return 3; yading@10: case 48000: return 4; yading@10: case 32000: return 5; yading@10: case 24000: return 6; yading@10: case 16000: return 7; yading@10: case 8000: return 8; yading@10: } yading@10: return -1; yading@10: } yading@10: yading@10: static av_cold int sonic_encode_init(AVCodecContext *avctx) yading@10: { yading@10: SonicContext *s = avctx->priv_data; yading@10: PutBitContext pb; yading@10: int i, version = 0; yading@10: yading@10: if (avctx->channels > MAX_CHANNELS) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n"); yading@10: return -1; /* only stereo or mono for now */ yading@10: } yading@10: yading@10: if (avctx->channels == 2) yading@10: s->decorrelation = MID_SIDE; yading@10: else yading@10: s->decorrelation = 3; yading@10: yading@10: if (avctx->codec->id == AV_CODEC_ID_SONIC_LS) yading@10: { yading@10: s->lossless = 1; yading@10: s->num_taps = 32; yading@10: s->downsampling = 1; yading@10: s->quantization = 0.0; yading@10: } yading@10: else yading@10: { yading@10: s->num_taps = 128; yading@10: s->downsampling = 2; yading@10: s->quantization = 1.0; yading@10: } yading@10: yading@10: // max tap 2048 yading@10: if ((s->num_taps < 32) || (s->num_taps > 1024) || yading@10: ((s->num_taps>>5)<<5 != s->num_taps)) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n"); yading@10: return -1; yading@10: } yading@10: yading@10: // generate taps yading@10: s->tap_quant = av_mallocz(4* s->num_taps); yading@10: for (i = 0; i < s->num_taps; i++) yading@10: s->tap_quant[i] = (int)(sqrt(i+1)); yading@10: yading@10: s->channels = avctx->channels; yading@10: s->samplerate = avctx->sample_rate; yading@10: yading@10: s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling; yading@10: s->frame_size = s->channels*s->block_align*s->downsampling; yading@10: yading@10: s->tail_size = s->num_taps*s->channels; yading@10: s->tail = av_mallocz(4 * s->tail_size); yading@10: if (!s->tail) yading@10: return -1; yading@10: yading@10: s->predictor_k = av_mallocz(4 * s->num_taps); yading@10: if (!s->predictor_k) yading@10: return -1; yading@10: yading@10: for (i = 0; i < s->channels; i++) yading@10: { yading@10: s->coded_samples[i] = av_mallocz(4* s->block_align); yading@10: if (!s->coded_samples[i]) yading@10: return -1; yading@10: } yading@10: yading@10: s->int_samples = av_mallocz(4* s->frame_size); yading@10: yading@10: s->window_size = ((2*s->tail_size)+s->frame_size); yading@10: s->window = av_mallocz(4* s->window_size); yading@10: if (!s->window) yading@10: return -1; yading@10: yading@10: avctx->extradata = av_mallocz(16); yading@10: if (!avctx->extradata) yading@10: return -1; yading@10: init_put_bits(&pb, avctx->extradata, 16*8); yading@10: yading@10: put_bits(&pb, 2, version); // version yading@10: if (version == 1) yading@10: { yading@10: put_bits(&pb, 2, s->channels); yading@10: put_bits(&pb, 4, code_samplerate(s->samplerate)); yading@10: } yading@10: put_bits(&pb, 1, s->lossless); yading@10: if (!s->lossless) yading@10: put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision yading@10: put_bits(&pb, 2, s->decorrelation); yading@10: put_bits(&pb, 2, s->downsampling); yading@10: put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024 yading@10: put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table yading@10: yading@10: flush_put_bits(&pb); yading@10: avctx->extradata_size = put_bits_count(&pb)/8; yading@10: yading@10: av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n", yading@10: version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling); yading@10: yading@10: avctx->coded_frame = avcodec_alloc_frame(); yading@10: if (!avctx->coded_frame) yading@10: return AVERROR(ENOMEM); yading@10: avctx->coded_frame->key_frame = 1; yading@10: avctx->frame_size = s->block_align*s->downsampling; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int sonic_encode_close(AVCodecContext *avctx) yading@10: { yading@10: SonicContext *s = avctx->priv_data; yading@10: int i; yading@10: yading@10: av_freep(&avctx->coded_frame); yading@10: yading@10: for (i = 0; i < s->channels; i++) yading@10: av_free(s->coded_samples[i]); yading@10: yading@10: av_free(s->predictor_k); yading@10: av_free(s->tail); yading@10: av_free(s->tap_quant); yading@10: av_free(s->window); yading@10: av_free(s->int_samples); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int sonic_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, yading@10: const AVFrame *frame, int *got_packet_ptr) yading@10: { yading@10: SonicContext *s = avctx->priv_data; yading@10: PutBitContext pb; yading@10: int i, j, ch, quant = 0, x = 0; yading@10: int ret; yading@10: const short *samples = (const int16_t*)frame->data[0]; yading@10: yading@10: if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size * 5 + 1000)) < 0) yading@10: return ret; yading@10: yading@10: init_put_bits(&pb, avpkt->data, avpkt->size); yading@10: yading@10: // short -> internal yading@10: for (i = 0; i < s->frame_size; i++) yading@10: s->int_samples[i] = samples[i]; yading@10: yading@10: if (!s->lossless) yading@10: for (i = 0; i < s->frame_size; i++) yading@10: s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT; yading@10: yading@10: switch(s->decorrelation) yading@10: { yading@10: case MID_SIDE: yading@10: for (i = 0; i < s->frame_size; i += s->channels) yading@10: { yading@10: s->int_samples[i] += s->int_samples[i+1]; yading@10: s->int_samples[i+1] -= shift(s->int_samples[i], 1); yading@10: } yading@10: break; yading@10: case LEFT_SIDE: yading@10: for (i = 0; i < s->frame_size; i += s->channels) yading@10: s->int_samples[i+1] -= s->int_samples[i]; yading@10: break; yading@10: case RIGHT_SIDE: yading@10: for (i = 0; i < s->frame_size; i += s->channels) yading@10: s->int_samples[i] -= s->int_samples[i+1]; yading@10: break; yading@10: } yading@10: yading@10: memset(s->window, 0, 4* s->window_size); yading@10: yading@10: for (i = 0; i < s->tail_size; i++) yading@10: s->window[x++] = s->tail[i]; yading@10: yading@10: for (i = 0; i < s->frame_size; i++) yading@10: s->window[x++] = s->int_samples[i]; yading@10: yading@10: for (i = 0; i < s->tail_size; i++) yading@10: s->window[x++] = 0; yading@10: yading@10: for (i = 0; i < s->tail_size; i++) yading@10: s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i]; yading@10: yading@10: // generate taps yading@10: modified_levinson_durbin(s->window, s->window_size, yading@10: s->predictor_k, s->num_taps, s->channels, s->tap_quant); yading@10: if (intlist_write(&pb, s->predictor_k, s->num_taps, 0) < 0) yading@10: return -1; yading@10: yading@10: for (ch = 0; ch < s->channels; ch++) yading@10: { yading@10: x = s->tail_size+ch; yading@10: for (i = 0; i < s->block_align; i++) yading@10: { yading@10: int sum = 0; yading@10: for (j = 0; j < s->downsampling; j++, x += s->channels) yading@10: sum += s->window[x]; yading@10: s->coded_samples[ch][i] = sum; yading@10: } yading@10: } yading@10: yading@10: // simple rate control code yading@10: if (!s->lossless) yading@10: { yading@10: double energy1 = 0.0, energy2 = 0.0; yading@10: for (ch = 0; ch < s->channels; ch++) yading@10: { yading@10: for (i = 0; i < s->block_align; i++) yading@10: { yading@10: double sample = s->coded_samples[ch][i]; yading@10: energy2 += sample*sample; yading@10: energy1 += fabs(sample); yading@10: } yading@10: } yading@10: yading@10: energy2 = sqrt(energy2/(s->channels*s->block_align)); yading@10: energy1 = sqrt(2.0)*energy1/(s->channels*s->block_align); yading@10: yading@10: // increase bitrate when samples are like a gaussian distribution yading@10: // reduce bitrate when samples are like a two-tailed exponential distribution yading@10: yading@10: if (energy2 > energy1) yading@10: energy2 += (energy2-energy1)*RATE_VARIATION; yading@10: yading@10: quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR); yading@10: // av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2); yading@10: yading@10: if (quant < 1) yading@10: quant = 1; yading@10: if (quant > 65534) yading@10: quant = 65534; yading@10: yading@10: set_ue_golomb(&pb, quant); yading@10: yading@10: quant *= SAMPLE_FACTOR; yading@10: } yading@10: yading@10: // write out coded samples yading@10: for (ch = 0; ch < s->channels; ch++) yading@10: { yading@10: if (!s->lossless) yading@10: for (i = 0; i < s->block_align; i++) yading@10: s->coded_samples[ch][i] = divide(s->coded_samples[ch][i], quant); yading@10: yading@10: if (intlist_write(&pb, s->coded_samples[ch], s->block_align, 1) < 0) yading@10: return -1; yading@10: } yading@10: yading@10: // av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8); yading@10: yading@10: flush_put_bits(&pb); yading@10: avpkt->size = (put_bits_count(&pb)+7)/8; yading@10: *got_packet_ptr = 1; yading@10: return 0; yading@10: } yading@10: #endif /* CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER */ yading@10: yading@10: #if CONFIG_SONIC_DECODER yading@10: static const int samplerate_table[] = yading@10: { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 }; yading@10: yading@10: static av_cold int sonic_decode_init(AVCodecContext *avctx) yading@10: { yading@10: SonicContext *s = avctx->priv_data; yading@10: GetBitContext gb; yading@10: int i, version; yading@10: yading@10: s->channels = avctx->channels; yading@10: s->samplerate = avctx->sample_rate; yading@10: yading@10: if (!avctx->extradata) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n"); yading@10: return -1; yading@10: } yading@10: yading@10: init_get_bits(&gb, avctx->extradata, avctx->extradata_size); yading@10: yading@10: version = get_bits(&gb, 2); yading@10: if (version > 1) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n"); yading@10: return -1; yading@10: } yading@10: yading@10: if (version == 1) yading@10: { yading@10: s->channels = get_bits(&gb, 2); yading@10: s->samplerate = samplerate_table[get_bits(&gb, 4)]; yading@10: av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n", yading@10: s->channels, s->samplerate); yading@10: } yading@10: yading@10: if (s->channels > MAX_CHANNELS) yading@10: { yading@10: av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n"); yading@10: return -1; yading@10: } yading@10: yading@10: s->lossless = get_bits1(&gb); yading@10: if (!s->lossless) yading@10: skip_bits(&gb, 3); // XXX FIXME yading@10: s->decorrelation = get_bits(&gb, 2); yading@10: if (s->decorrelation != 3 && s->channels != 2) { yading@10: av_log(avctx, AV_LOG_ERROR, "invalid decorrelation %d\n", s->decorrelation); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: s->downsampling = get_bits(&gb, 2); yading@10: if (!s->downsampling) { yading@10: av_log(avctx, AV_LOG_ERROR, "invalid downsampling value\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: s->num_taps = (get_bits(&gb, 5)+1)<<5; yading@10: if (get_bits1(&gb)) // XXX FIXME yading@10: av_log(avctx, AV_LOG_INFO, "Custom quant table\n"); yading@10: yading@10: s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling; yading@10: s->frame_size = s->channels*s->block_align*s->downsampling; yading@10: // avctx->frame_size = s->block_align; yading@10: yading@10: av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n", yading@10: version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling); yading@10: yading@10: // generate taps yading@10: s->tap_quant = av_mallocz(4* s->num_taps); yading@10: for (i = 0; i < s->num_taps; i++) yading@10: s->tap_quant[i] = (int)(sqrt(i+1)); yading@10: yading@10: s->predictor_k = av_mallocz(4* s->num_taps); yading@10: yading@10: for (i = 0; i < s->channels; i++) yading@10: { yading@10: s->predictor_state[i] = av_mallocz(4* s->num_taps); yading@10: if (!s->predictor_state[i]) yading@10: return -1; yading@10: } yading@10: yading@10: for (i = 0; i < s->channels; i++) yading@10: { yading@10: s->coded_samples[i] = av_mallocz(4* s->block_align); yading@10: if (!s->coded_samples[i]) yading@10: return -1; yading@10: } yading@10: s->int_samples = av_mallocz(4* s->frame_size); yading@10: yading@10: avctx->sample_fmt = AV_SAMPLE_FMT_S16; yading@10: return 0; yading@10: } yading@10: yading@10: static av_cold int sonic_decode_close(AVCodecContext *avctx) yading@10: { yading@10: SonicContext *s = avctx->priv_data; yading@10: int i; yading@10: yading@10: av_free(s->int_samples); yading@10: av_free(s->tap_quant); yading@10: av_free(s->predictor_k); yading@10: yading@10: for (i = 0; i < s->channels; i++) yading@10: { yading@10: av_free(s->predictor_state[i]); yading@10: av_free(s->coded_samples[i]); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int sonic_decode_frame(AVCodecContext *avctx, yading@10: void *data, int *got_frame_ptr, yading@10: AVPacket *avpkt) yading@10: { yading@10: const uint8_t *buf = avpkt->data; yading@10: int buf_size = avpkt->size; yading@10: SonicContext *s = avctx->priv_data; yading@10: GetBitContext gb; yading@10: int i, quant, ch, j, ret; yading@10: int16_t *samples; yading@10: AVFrame *frame = data; yading@10: yading@10: if (buf_size == 0) return 0; yading@10: yading@10: frame->nb_samples = s->frame_size / avctx->channels; yading@10: if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) yading@10: return ret; yading@10: samples = (int16_t *)frame->data[0]; yading@10: yading@10: // av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size); yading@10: yading@10: init_get_bits(&gb, buf, buf_size*8); yading@10: yading@10: intlist_read(&gb, s->predictor_k, s->num_taps, 0); yading@10: yading@10: // dequantize yading@10: for (i = 0; i < s->num_taps; i++) yading@10: s->predictor_k[i] *= s->tap_quant[i]; yading@10: yading@10: if (s->lossless) yading@10: quant = 1; yading@10: else yading@10: quant = get_ue_golomb(&gb) * SAMPLE_FACTOR; yading@10: yading@10: // av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant); yading@10: yading@10: for (ch = 0; ch < s->channels; ch++) yading@10: { yading@10: int x = ch; yading@10: yading@10: predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps); yading@10: yading@10: intlist_read(&gb, s->coded_samples[ch], s->block_align, 1); yading@10: yading@10: for (i = 0; i < s->block_align; i++) yading@10: { yading@10: for (j = 0; j < s->downsampling - 1; j++) yading@10: { yading@10: s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0); yading@10: x += s->channels; yading@10: } yading@10: yading@10: s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant); yading@10: x += s->channels; yading@10: } yading@10: yading@10: for (i = 0; i < s->num_taps; i++) yading@10: s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels]; yading@10: } yading@10: yading@10: switch(s->decorrelation) yading@10: { yading@10: case MID_SIDE: yading@10: for (i = 0; i < s->frame_size; i += s->channels) yading@10: { yading@10: s->int_samples[i+1] += shift(s->int_samples[i], 1); yading@10: s->int_samples[i] -= s->int_samples[i+1]; yading@10: } yading@10: break; yading@10: case LEFT_SIDE: yading@10: for (i = 0; i < s->frame_size; i += s->channels) yading@10: s->int_samples[i+1] += s->int_samples[i]; yading@10: break; yading@10: case RIGHT_SIDE: yading@10: for (i = 0; i < s->frame_size; i += s->channels) yading@10: s->int_samples[i] += s->int_samples[i+1]; yading@10: break; yading@10: } yading@10: yading@10: if (!s->lossless) yading@10: for (i = 0; i < s->frame_size; i++) yading@10: s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT); yading@10: yading@10: // internal -> short yading@10: for (i = 0; i < s->frame_size; i++) yading@10: samples[i] = av_clip_int16(s->int_samples[i]); yading@10: yading@10: align_get_bits(&gb); yading@10: yading@10: *got_frame_ptr = 1; yading@10: yading@10: return (get_bits_count(&gb)+7)/8; yading@10: } yading@10: yading@10: AVCodec ff_sonic_decoder = { yading@10: .name = "sonic", yading@10: .type = AVMEDIA_TYPE_AUDIO, yading@10: .id = AV_CODEC_ID_SONIC, yading@10: .priv_data_size = sizeof(SonicContext), yading@10: .init = sonic_decode_init, yading@10: .close = sonic_decode_close, yading@10: .decode = sonic_decode_frame, yading@10: .capabilities = CODEC_CAP_DR1 | CODEC_CAP_EXPERIMENTAL, yading@10: .long_name = NULL_IF_CONFIG_SMALL("Sonic"), yading@10: }; yading@10: #endif /* CONFIG_SONIC_DECODER */ yading@10: yading@10: #if CONFIG_SONIC_ENCODER yading@10: AVCodec ff_sonic_encoder = { yading@10: .name = "sonic", yading@10: .type = AVMEDIA_TYPE_AUDIO, yading@10: .id = AV_CODEC_ID_SONIC, yading@10: .priv_data_size = sizeof(SonicContext), yading@10: .init = sonic_encode_init, yading@10: .encode2 = sonic_encode_frame, yading@10: .capabilities = CODEC_CAP_EXPERIMENTAL, yading@10: .close = sonic_encode_close, yading@10: .long_name = NULL_IF_CONFIG_SMALL("Sonic"), yading@10: }; yading@10: #endif yading@10: yading@10: #if CONFIG_SONIC_LS_ENCODER yading@10: AVCodec ff_sonic_ls_encoder = { yading@10: .name = "sonicls", yading@10: .type = AVMEDIA_TYPE_AUDIO, yading@10: .id = AV_CODEC_ID_SONIC_LS, yading@10: .priv_data_size = sizeof(SonicContext), yading@10: .init = sonic_encode_init, yading@10: .encode2 = sonic_encode_frame, yading@10: .capabilities = CODEC_CAP_EXPERIMENTAL, yading@10: .close = sonic_encode_close, yading@10: .long_name = NULL_IF_CONFIG_SMALL("Sonic lossless"), yading@10: }; yading@10: #endif