virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-10 06:10:52 +02:00
Code Issues Releases Activity

g722: split decoder and encoder into separate files

Browse Source
This commit is contained in:
Justin Ruggles
2011-10-22 17:49:50 -04:00
parent b95fbba705
commit 704721bc9c
5 changed files with 542 additions and 421 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/Makefile
View File

@@ -495,8 +495,8 @@ OBJS-$(CONFIG_ADPCM_EA_R1_DECODER) += adpcm.o
OBJS-$(CONFIG_ADPCM_EA_R2_DECODER) += adpcm.o OBJS-$(CONFIG_ADPCM_EA_R2_DECODER) += adpcm.o
OBJS-$(CONFIG_ADPCM_EA_R3_DECODER) += adpcm.o OBJS-$(CONFIG_ADPCM_EA_R3_DECODER) += adpcm.o
OBJS-$(CONFIG_ADPCM_EA_XAS_DECODER) += adpcm.o OBJS-$(CONFIG_ADPCM_EA_XAS_DECODER) += adpcm.o
OBJS-$(CONFIG_ADPCM_G722_DECODER) += g722.o OBJS-$(CONFIG_ADPCM_G722_DECODER) += g722.o g722dec.o
OBJS-$(CONFIG_ADPCM_G722_ENCODER) += g722.o OBJS-$(CONFIG_ADPCM_G722_ENCODER) += g722.o g722enc.o
OBJS-$(CONFIG_ADPCM_G726_DECODER) += g726.o OBJS-$(CONFIG_ADPCM_G726_DECODER) += g726.o
OBJS-$(CONFIG_ADPCM_G726_ENCODER) += g726.o OBJS-$(CONFIG_ADPCM_G726_ENCODER) += g726.o
OBJS-$(CONFIG_ADPCM_IMA_AMV_DECODER) += adpcm.o adpcm_data.o OBJS-$(CONFIG_ADPCM_IMA_AMV_DECODER) += adpcm.o adpcm_data.o

427
libavcodec/g722.c
View File

@@ -36,45 +36,8 @@
* respectively of each byte are ignored. * respectively of each byte are ignored.
*/ */
#include "avcodec.h"
#include "mathops.h" #include "mathops.h"
#include "get_bits.h" #include "g722.h"
#define PREV_SAMPLES_BUF_SIZE 1024
#define FREEZE_INTERVAL 128
typedef struct {
int16_t prev_samples[PREV_SAMPLES_BUF_SIZE]; ///< memory of past decoded samples
int prev_samples_pos; ///< the number of values in prev_samples
/**
* The band[0] and band[1] correspond respectively to the lower band and higher band.
*/
struct G722Band {
int16_t s_predictor; ///< predictor output value
int32_t s_zero; ///< previous output signal from zero predictor
int8_t part_reconst_mem[2]; ///< signs of previous partially reconstructed signals
int16_t prev_qtzd_reconst; ///< previous quantized reconstructed signal (internal value, using low_inv_quant4)
int16_t pole_mem[2]; ///< second-order pole section coefficient buffer
int32_t diff_mem[6]; ///< quantizer difference signal memory
int16_t zero_mem[6]; ///< Seventh-order zero section coefficient buffer
int16_t log_factor; ///< delayed 2-logarithmic quantizer factor
int16_t scale_factor; ///< delayed quantizer scale factor
} band[2];
struct TrellisNode {
struct G722Band state;
uint32_t ssd;
int path;
} *node_buf[2], **nodep_buf[2];
struct TrellisPath {
int value;
int prev;
} *paths[2];
} G722Context;
static const int8_t sign_lookup[2] = { -1, 1 }; static const int8_t sign_lookup[2] = { -1, 1 };
@@ -85,7 +48,7 @@ static const int16_t inv_log2_table[32] = {
3444, 3520, 3597, 3676, 3756, 3838, 3922, 4008 3444, 3520, 3597, 3676, 3756, 3838, 3922, 4008
}; };
static const int16_t high_log_factor_step[2] = { 798, -214 }; static const int16_t high_log_factor_step[2] = { 798, -214 };
static const int16_t high_inv_quant[4] = { -926, -202, 926, 202 }; const int16_t ff_g722_high_inv_quant[4] = { -926, -202, 926, 202 };
/** /**
* low_log_factor_step[index] == wl[rl42[index]] * low_log_factor_step[index] == wl[rl42[index]]
*/ */
@@ -93,11 +56,11 @@ static const int16_t low_log_factor_step[16] = {
-60, 3042, 1198, 538, 334, 172, 58, -30, -60, 3042, 1198, 538, 334, 172, 58, -30,
3042, 1198, 538, 334, 172, 58, -30, -60 3042, 1198, 538, 334, 172, 58, -30, -60
}; };
static const int16_t low_inv_quant4[16] = { const int16_t ff_g722_low_inv_quant4[16] = {
0, -2557, -1612, -1121, -786, -530, -323, -150, 0, -2557, -1612, -1121, -786, -530, -323, -150,
2557, 1612, 1121, 786, 530, 323, 150, 0 2557, 1612, 1121, 786, 530, 323, 150, 0
}; };
static const int16_t low_inv_quant6[64] = { const int16_t ff_g722_low_inv_quant6[64] = {
-17, -17, -17, -17, -3101, -2738, -2376, -2088, -17, -17, -17, -17, -3101, -2738, -2376, -2088,
-1873, -1689, -1535, -1399, -1279, -1170, -1072, -982, -1873, -1689, -1535, -1399, -1279, -1170, -1072, -982,
-899, -822, -750, -682, -618, -558, -501, -447, -899, -822, -750, -682, -618, -558, -501, -447,
@@ -173,10 +136,10 @@ static int inline linear_scale_factor(const int log_factor)
return shift < 0 ? wd1 >> -shift : wd1 << shift; return shift < 0 ? wd1 >> -shift : wd1 << shift;
} }
static void update_low_predictor(struct G722Band *band, const int ilow) void ff_g722_update_low_predictor(struct G722Band *band, const int ilow)
{ {
do_adaptive_prediction(band, do_adaptive_prediction(band,
band->scale_factor * low_inv_quant4[ilow] >> 10); band->scale_factor * ff_g722_low_inv_quant4[ilow] >> 10);
// quantizer adaptation // quantizer adaptation
band->log_factor = av_clip((band->log_factor * 127 >> 7) + band->log_factor = av_clip((band->log_factor * 127 >> 7) +
@@ -184,7 +147,7 @@ static void update_low_predictor(struct G722Band *band, const int ilow)
band->scale_factor = linear_scale_factor(band->log_factor - (8 << 11)); band->scale_factor = linear_scale_factor(band->log_factor - (8 << 11));
} }
static void update_high_predictor(struct G722Band *band, const int dhigh, void ff_g722_update_high_predictor(struct G722Band *band, const int dhigh,
const int ihigh) const int ihigh)
{ {
do_adaptive_prediction(band, dhigh); do_adaptive_prediction(band, dhigh);
@@ -195,7 +158,7 @@ static void update_high_predictor(struct G722Band *band, const int dhigh,
band->scale_factor = linear_scale_factor(band->log_factor - (10 << 11)); band->scale_factor = linear_scale_factor(band->log_factor - (10 << 11));
} }
static void apply_qmf(const int16_t *prev_samples, int *xout1, int *xout2) void ff_g722_apply_qmf(const int16_t *prev_samples, int *xout1, int *xout2)
{ {
int i; int i;
@@ -206,377 +169,3 @@ static void apply_qmf(const int16_t *prev_samples, int *xout1, int *xout2)
MAC16(*xout1, prev_samples[2*i+1], qmf_coeffs[11-i]); MAC16(*xout1, prev_samples[2*i+1], qmf_coeffs[11-i]);
} }
} }
static av_cold int g722_init(AVCodecContext * avctx)
{
G722Context *c = avctx->priv_data;
if (avctx->channels != 1) {
av_log(avctx, AV_LOG_ERROR, "Only mono tracks are allowed.\n");
return AVERROR_INVALIDDATA;
}
avctx->sample_fmt = AV_SAMPLE_FMT_S16;
switch (avctx->bits_per_coded_sample) {
case 8:
case 7:
case 6:
break;
default:
av_log(avctx, AV_LOG_WARNING, "Unsupported bits_per_coded_sample [%d], "
"assuming 8\n",
avctx->bits_per_coded_sample);
case 0:
avctx->bits_per_coded_sample = 8;
break;
}
c->band[0].scale_factor = 8;
c->band[1].scale_factor = 2;
c->prev_samples_pos = 22;
if (avctx->lowres)
avctx->sample_rate /= 2;
if (avctx->trellis) {
int frontier = 1 << avctx->trellis;
int max_paths = frontier * FREEZE_INTERVAL;
int i;
for (i = 0; i < 2; i++) {
c->paths[i] = av_mallocz(max_paths * sizeof(**c->paths));
c->node_buf[i] = av_mallocz(2 * frontier * sizeof(**c->node_buf));
c->nodep_buf[i] = av_mallocz(2 * frontier * sizeof(**c->nodep_buf));
}
}
return 0;
}
static av_cold int g722_close(AVCodecContext *avctx)
{
G722Context *c = avctx->priv_data;
int i;
for (i = 0; i < 2; i++) {
av_freep(&c->paths[i]);
av_freep(&c->node_buf[i]);
av_freep(&c->nodep_buf[i]);
}
return 0;
}
#if CONFIG_ADPCM_G722_DECODER
static const int16_t low_inv_quant5[32] = {
-35, -35, -2919, -2195, -1765, -1458, -1219, -1023,
-858, -714, -587, -473, -370, -276, -190, -110,
2919, 2195, 1765, 1458, 1219, 1023, 858, 714,
587, 473, 370, 276, 190, 110, 35, -35
};
static const int16_t *low_inv_quants[3] = { low_inv_quant6, low_inv_quant5,
low_inv_quant4 };
static int g722_decode_frame(AVCodecContext *avctx, void *data,
int *data_size, AVPacket *avpkt)
{
G722Context *c = avctx->priv_data;
int16_t *out_buf = data;
int j, out_len = 0;
const int skip = 8 - avctx->bits_per_coded_sample;
const int16_t *quantizer_table = low_inv_quants[skip];
GetBitContext gb;
init_get_bits(&gb, avpkt->data, avpkt->size * 8);
for (j = 0; j < avpkt->size; j++) {
int ilow, ihigh, rlow;
ihigh = get_bits(&gb, 2);
ilow = get_bits(&gb, 6 - skip);
skip_bits(&gb, skip);
rlow = av_clip((c->band[0].scale_factor * quantizer_table[ilow] >> 10)
+ c->band[0].s_predictor, -16384, 16383);
update_low_predictor(&c->band[0], ilow >> (2 - skip));
if (!avctx->lowres) {
const int dhigh = c->band[1].scale_factor *
high_inv_quant[ihigh] >> 10;
const int rhigh = av_clip(dhigh + c->band[1].s_predictor,
-16384, 16383);
int xout1, xout2;
update_high_predictor(&c->band[1], dhigh, ihigh);
c->prev_samples[c->prev_samples_pos++] = rlow + rhigh;
c->prev_samples[c->prev_samples_pos++] = rlow - rhigh;
apply_qmf(c->prev_samples + c->prev_samples_pos - 24,
&xout1, &xout2);
out_buf[out_len++] = av_clip_int16(xout1 >> 12);
out_buf[out_len++] = av_clip_int16(xout2 >> 12);
if (c->prev_samples_pos >= PREV_SAMPLES_BUF_SIZE) {
memmove(c->prev_samples,
c->prev_samples + c->prev_samples_pos - 22,
22 * sizeof(c->prev_samples[0]));
c->prev_samples_pos = 22;
}
} else
out_buf[out_len++] = rlow;
}
*data_size = out_len << 1;
return avpkt->size;
}
AVCodec ff_adpcm_g722_decoder = {
.name = "g722",
.type = AVMEDIA_TYPE_AUDIO,
.id = CODEC_ID_ADPCM_G722,
.priv_data_size = sizeof(G722Context),
.init = g722_init,
.decode = g722_decode_frame,
.long_name = NULL_IF_CONFIG_SMALL("G.722 ADPCM"),
.max_lowres = 1,
};
#endif
#if CONFIG_ADPCM_G722_ENCODER
static const int16_t low_quant[33] = {
35, 72, 110, 150, 190, 233, 276, 323,
370, 422, 473, 530, 587, 650, 714, 786,
858, 940, 1023, 1121, 1219, 1339, 1458, 1612,
1765, 1980, 2195, 2557, 2919
};
static inline void filter_samples(G722Context *c, const int16_t *samples,
int *xlow, int *xhigh)
{
int xout1, xout2;
c->prev_samples[c->prev_samples_pos++] = samples[0];
c->prev_samples[c->prev_samples_pos++] = samples[1];
apply_qmf(c->prev_samples + c->prev_samples_pos - 24, &xout1, &xout2);
*xlow = xout1 + xout2 >> 13;
*xhigh = xout1 - xout2 >> 13;
if (c->prev_samples_pos >= PREV_SAMPLES_BUF_SIZE) {
memmove(c->prev_samples,
c->prev_samples + c->prev_samples_pos - 22,
22 * sizeof(c->prev_samples[0]));
c->prev_samples_pos = 22;
}
}
static inline int encode_high(const struct G722Band *state, int xhigh)
{
int diff = av_clip_int16(xhigh - state->s_predictor);
int pred = 141 * state->scale_factor >> 8;
/* = diff >= 0 ? (diff < pred) + 2 : diff >= -pred */
return ((diff ^ (diff >> (sizeof(diff)*8-1))) < pred) + 2*(diff >= 0);
}
static inline int encode_low(const struct G722Band* state, int xlow)
{
int diff = av_clip_int16(xlow - state->s_predictor);
/* = diff >= 0 ? diff : -(diff + 1) */
int limit = diff ^ (diff >> (sizeof(diff)*8-1));
int i = 0;
limit = limit + 1 << 10;
if (limit > low_quant[8] * state->scale_factor)
i = 9;
while (i < 29 && limit > low_quant[i] * state->scale_factor)
i++;
return (diff < 0 ? (i < 2 ? 63 : 33) : 61) - i;
}
static int g722_encode_trellis(AVCodecContext *avctx,
uint8_t *dst, int buf_size, void *data)
{
G722Context *c = avctx->priv_data;
const int16_t *samples = data;
int i, j, k;
int frontier = 1 << avctx->trellis;
struct TrellisNode **nodes[2];
struct TrellisNode **nodes_next[2];
int pathn[2] = {0, 0}, froze = -1;
struct TrellisPath *p[2];
for (i = 0; i < 2; i++) {
nodes[i] = c->nodep_buf[i];
nodes_next[i] = c->nodep_buf[i] + frontier;
memset(c->nodep_buf[i], 0, 2 * frontier * sizeof(*c->nodep_buf));
nodes[i][0] = c->node_buf[i] + frontier;
nodes[i][0]->ssd = 0;
nodes[i][0]->path = 0;
nodes[i][0]->state = c->band[i];
}
for (i = 0; i < buf_size >> 1; i++) {
int xlow, xhigh;
struct TrellisNode *next[2];
int heap_pos[2] = {0, 0};
for (j = 0; j < 2; j++) {
next[j] = c->node_buf[j] + frontier*(i & 1);
memset(nodes_next[j], 0, frontier * sizeof(**nodes_next));
}
filter_samples(c, &samples[2*i], &xlow, &xhigh);
for (j = 0; j < frontier && nodes[0][j]; j++) {
/* Only k >> 2 affects the future adaptive state, therefore testing
* small steps that don't change k >> 2 is useless, the orignal
* value from encode_low is better than them. Since we step k
* in steps of 4, make sure range is a multiple of 4, so that
* we don't miss the original value from encode_low. */
int range = j < frontier/2 ? 4 : 0;
struct TrellisNode *cur_node = nodes[0][j];
int ilow = encode_low(&cur_node->state, xlow);
for (k = ilow - range; k <= ilow + range && k <= 63; k += 4) {
int decoded, dec_diff, pos;
uint32_t ssd;
struct TrellisNode* node;
if (k < 0)
continue;
decoded = av_clip((cur_node->state.scale_factor *
low_inv_quant6[k] >> 10)
+ cur_node->state.s_predictor, -16384, 16383);
dec_diff = xlow - decoded;
#define STORE_NODE(index, UPDATE, VALUE)\
ssd = cur_node->ssd + dec_diff*dec_diff;\
/* Check for wraparound. Using 64 bit ssd counters would \
* be simpler, but is slower on x86 32 bit. */\
if (ssd < cur_node->ssd)\
continue;\
if (heap_pos[index] < frontier) {\
pos = heap_pos[index]++;\
assert(pathn[index] < FREEZE_INTERVAL * frontier);\
node = nodes_next[index][pos] = next[index]++;\
node->path = pathn[index]++;\
} else {\
/* Try to replace one of the leaf nodes with the new \
* one, but not always testing the same leaf position */\
pos = (frontier>>1) + (heap_pos[index] & ((frontier>>1) - 1));\
if (ssd >= nodes_next[index][pos]->ssd)\
continue;\
heap_pos[index]++;\
node = nodes_next[index][pos];\
}\
node->ssd = ssd;\
node->state = cur_node->state;\
UPDATE;\
c->paths[index][node->path].value = VALUE;\
c->paths[index][node->path].prev = cur_node->path;\
/* Sift the newly inserted node up in the heap to restore \
* the heap property */\
while (pos > 0) {\
int parent = (pos - 1) >> 1;\
if (nodes_next[index][parent]->ssd <= ssd)\
break;\
FFSWAP(struct TrellisNode*, nodes_next[index][parent],\
nodes_next[index][pos]);\
pos = parent;\
}
STORE_NODE(0, update_low_predictor(&node->state, k >> 2), k);
}
}
for (j = 0; j < frontier && nodes[1][j]; j++) {
int ihigh;
struct TrellisNode *cur_node = nodes[1][j];
/* We don't try to get any initial guess for ihigh via
* encode_high - since there's only 4 possible values, test
* them all. Testing all of these gives a much, much larger
* gain than testing a larger range around ilow. */
for (ihigh = 0; ihigh < 4; ihigh++) {
int dhigh, decoded, dec_diff, pos;
uint32_t ssd;
struct TrellisNode* node;
dhigh = cur_node->state.scale_factor *
high_inv_quant[ihigh] >> 10;
decoded = av_clip(dhigh + cur_node->state.s_predictor,
-16384, 16383);
dec_diff = xhigh - decoded;
STORE_NODE(1, update_high_predictor(&node->state, dhigh, ihigh), ihigh);
}
}
for (j = 0; j < 2; j++) {
FFSWAP(struct TrellisNode**, nodes[j], nodes_next[j]);
if (nodes[j][0]->ssd > (1 << 16)) {
for (k = 1; k < frontier && nodes[j][k]; k++)
nodes[j][k]->ssd -= nodes[j][0]->ssd;
nodes[j][0]->ssd = 0;
}
}
if (i == froze + FREEZE_INTERVAL) {
p[0] = &c->paths[0][nodes[0][0]->path];
p[1] = &c->paths[1][nodes[1][0]->path];
for (j = i; j > froze; j--) {
dst[j] = p[1]->value << 6 | p[0]->value;
p[0] = &c->paths[0][p[0]->prev];
p[1] = &c->paths[1][p[1]->prev];
}
froze = i;
pathn[0] = pathn[1] = 0;
memset(nodes[0] + 1, 0, (frontier - 1)*sizeof(**nodes));
memset(nodes[1] + 1, 0, (frontier - 1)*sizeof(**nodes));
}
}
p[0] = &c->paths[0][nodes[0][0]->path];
p[1] = &c->paths[1][nodes[1][0]->path];
for (j = i; j > froze; j--) {
dst[j] = p[1]->value << 6 | p[0]->value;
p[0] = &c->paths[0][p[0]->prev];
p[1] = &c->paths[1][p[1]->prev];
}
c->band[0] = nodes[0][0]->state;
c->band[1] = nodes[1][0]->state;
return i;
}
static int g722_encode_frame(AVCodecContext *avctx,
uint8_t *dst, int buf_size, void *data)
{
G722Context *c = avctx->priv_data;
const int16_t *samples = data;
int i;
if (avctx->trellis)
return g722_encode_trellis(avctx, dst, buf_size, data);
for (i = 0; i < buf_size >> 1; i++) {
int xlow, xhigh, ihigh, ilow;
filter_samples(c, &samples[2*i], &xlow, &xhigh);
ihigh = encode_high(&c->band[1], xhigh);
ilow = encode_low(&c->band[0], xlow);
update_high_predictor(&c->band[1], c->band[1].scale_factor *
high_inv_quant[ihigh] >> 10, ihigh);
update_low_predictor(&c->band[0], ilow >> 2);
*dst++ = ihigh << 6 | ilow;
}
return i;
}
AVCodec ff_adpcm_g722_encoder = {
.name = "g722",
.type = AVMEDIA_TYPE_AUDIO,
.id = CODEC_ID_ADPCM_G722,
.priv_data_size = sizeof(G722Context),
.init = g722_init,
.close = g722_close,
.encode = g722_encode_frame,
.long_name = NULL_IF_CONFIG_SMALL("G.722 ADPCM"),
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
};
#endif

74
libavcodec/g722.h Normal file
View File

@@ -0,0 +1,74 @@
/*
* Copyright (c) CMU 1993 Computer Science, Speech Group
* Chengxiang Lu and Alex Hauptmann
* Copyright (c) 2005 Steve Underwood <steveu at coppice.org>
* Copyright (c) 2009 Kenan Gillet
* Copyright (c) 2010 Martin Storsjo
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVCODEC_G722_H
#define AVCODEC_G722_H
#include <stdint.h>
#define PREV_SAMPLES_BUF_SIZE 1024
typedef struct {
int16_t prev_samples[PREV_SAMPLES_BUF_SIZE]; ///< memory of past decoded samples
int prev_samples_pos; ///< the number of values in prev_samples
/**
* The band[0] and band[1] correspond respectively to the lower band and higher band.
*/
struct G722Band {
int16_t s_predictor; ///< predictor output value
int32_t s_zero; ///< previous output signal from zero predictor
int8_t part_reconst_mem[2]; ///< signs of previous partially reconstructed signals
int16_t prev_qtzd_reconst; ///< previous quantized reconstructed signal (internal value, using low_inv_quant4)
int16_t pole_mem[2]; ///< second-order pole section coefficient buffer
int32_t diff_mem[6]; ///< quantizer difference signal memory
int16_t zero_mem[6]; ///< Seventh-order zero section coefficient buffer
int16_t log_factor; ///< delayed 2-logarithmic quantizer factor
int16_t scale_factor; ///< delayed quantizer scale factor
} band[2];
struct TrellisNode {
struct G722Band state;
uint32_t ssd;
int path;
} *node_buf[2], **nodep_buf[2];
struct TrellisPath {
int value;
int prev;
} *paths[2];
} G722Context;
extern const int16_t ff_g722_high_inv_quant[4];
extern const int16_t ff_g722_low_inv_quant4[16];
extern const int16_t ff_g722_low_inv_quant6[64];
void ff_g722_update_low_predictor(struct G722Band *band, const int ilow);
void ff_g722_update_high_predictor(struct G722Band *band, const int dhigh,
const int ihigh);
void ff_g722_apply_qmf(const int16_t *prev_samples, int *xout1, int *xout2);
#endif /* AVCODEC_G722_H */

147
libavcodec/g722dec.c Normal file
View File

@@ -0,0 +1,147 @@
/*
* Copyright (c) CMU 1993 Computer Science, Speech Group
* Chengxiang Lu and Alex Hauptmann
* Copyright (c) 2005 Steve Underwood <steveu at coppice.org>
* Copyright (c) 2009 Kenan Gillet
* Copyright (c) 2010 Martin Storsjo
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* G.722 ADPCM audio decoder
*
* This G.722 decoder is a bit-exact implementation of the ITU G.722
* specification for all three specified bitrates - 64000bps, 56000bps
* and 48000bps. It passes the ITU tests.
*
* @note For the 56000bps and 48000bps bitrates, the lowest 1 or 2 bits
* respectively of each byte are ignored.
*/
#include "avcodec.h"
#include "get_bits.h"
#include "g722.h"
static av_cold int g722_decode_init(AVCodecContext * avctx)
{
G722Context *c = avctx->priv_data;
if (avctx->channels != 1) {
av_log(avctx, AV_LOG_ERROR, "Only mono tracks are allowed.\n");
return AVERROR_INVALIDDATA;
}
avctx->sample_fmt = AV_SAMPLE_FMT_S16;
switch (avctx->bits_per_coded_sample) {
case 8:
case 7:
case 6:
break;
default:
av_log(avctx, AV_LOG_WARNING, "Unsupported bits_per_coded_sample [%d], "
"assuming 8\n",
avctx->bits_per_coded_sample);
case 0:
avctx->bits_per_coded_sample = 8;
break;
}
c->band[0].scale_factor = 8;
c->band[1].scale_factor = 2;
c->prev_samples_pos = 22;
if (avctx->lowres)
avctx->sample_rate /= 2;
return 0;
}
static const int16_t low_inv_quant5[32] = {
-35, -35, -2919, -2195, -1765, -1458, -1219, -1023,
-858, -714, -587, -473, -370, -276, -190, -110,
2919, 2195, 1765, 1458, 1219, 1023, 858, 714,
587, 473, 370, 276, 190, 110, 35, -35
};
static const int16_t *low_inv_quants[3] = { ff_g722_low_inv_quant6,
low_inv_quant5,
ff_g722_low_inv_quant4 };
static int g722_decode_frame(AVCodecContext *avctx, void *data,
int *data_size, AVPacket *avpkt)
{
G722Context *c = avctx->priv_data;
int16_t *out_buf = data;
int j, out_len = 0;
const int skip = 8 - avctx->bits_per_coded_sample;
const int16_t *quantizer_table = low_inv_quants[skip];
GetBitContext gb;
init_get_bits(&gb, avpkt->data, avpkt->size * 8);
for (j = 0; j < avpkt->size; j++) {
int ilow, ihigh, rlow;
ihigh = get_bits(&gb, 2);
ilow = get_bits(&gb, 6 - skip);
skip_bits(&gb, skip);
rlow = av_clip((c->band[0].scale_factor * quantizer_table[ilow] >> 10)
+ c->band[0].s_predictor, -16384, 16383);
ff_g722_update_low_predictor(&c->band[0], ilow >> (2 - skip));
if (!avctx->lowres) {
const int dhigh = c->band[1].scale_factor *
ff_g722_high_inv_quant[ihigh] >> 10;
const int rhigh = av_clip(dhigh + c->band[1].s_predictor,
-16384, 16383);
int xout1, xout2;
ff_g722_update_high_predictor(&c->band[1], dhigh, ihigh);
c->prev_samples[c->prev_samples_pos++] = rlow + rhigh;
c->prev_samples[c->prev_samples_pos++] = rlow - rhigh;
ff_g722_apply_qmf(c->prev_samples + c->prev_samples_pos - 24,
&xout1, &xout2);
out_buf[out_len++] = av_clip_int16(xout1 >> 12);
out_buf[out_len++] = av_clip_int16(xout2 >> 12);
if (c->prev_samples_pos >= PREV_SAMPLES_BUF_SIZE) {
memmove(c->prev_samples,
c->prev_samples + c->prev_samples_pos - 22,
22 * sizeof(c->prev_samples[0]));
c->prev_samples_pos = 22;
}
} else
out_buf[out_len++] = rlow;
}
*data_size = out_len << 1;
return avpkt->size;
}
AVCodec ff_adpcm_g722_decoder = {
.name = "g722",
.type = AVMEDIA_TYPE_AUDIO,
.id = CODEC_ID_ADPCM_G722,
.priv_data_size = sizeof(G722Context),
.init = g722_decode_init,
.decode = g722_decode_frame,
.long_name = NULL_IF_CONFIG_SMALL("G.722 ADPCM"),
.max_lowres = 1,
};

311
libavcodec/g722enc.c Normal file
View File

@@ -0,0 +1,311 @@
/*
* Copyright (c) CMU 1993 Computer Science, Speech Group
* Chengxiang Lu and Alex Hauptmann
* Copyright (c) 2005 Steve Underwood <steveu at coppice.org>
* Copyright (c) 2009 Kenan Gillet
* Copyright (c) 2010 Martin Storsjo
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* G.722 ADPCM audio encoder
*/
#include "avcodec.h"
#include "g722.h"
#define FREEZE_INTERVAL 128
static av_cold int g722_encode_init(AVCodecContext * avctx)
{
G722Context *c = avctx->priv_data;
if (avctx->channels != 1) {
av_log(avctx, AV_LOG_ERROR, "Only mono tracks are allowed.\n");
return AVERROR_INVALIDDATA;
}
c->band[0].scale_factor = 8;
c->band[1].scale_factor = 2;
c->prev_samples_pos = 22;
if (avctx->trellis) {
int frontier = 1 << avctx->trellis;
int max_paths = frontier * FREEZE_INTERVAL;
int i;
for (i = 0; i < 2; i++) {
c->paths[i] = av_mallocz(max_paths * sizeof(**c->paths));
c->node_buf[i] = av_mallocz(2 * frontier * sizeof(**c->node_buf));
c->nodep_buf[i] = av_mallocz(2 * frontier * sizeof(**c->nodep_buf));
}
}
return 0;
}
static av_cold int g722_encode_close(AVCodecContext *avctx)
{
G722Context *c = avctx->priv_data;
int i;
for (i = 0; i < 2; i++) {
av_freep(&c->paths[i]);
av_freep(&c->node_buf[i]);
av_freep(&c->nodep_buf[i]);
}
return 0;
}
static const int16_t low_quant[33] = {
35, 72, 110, 150, 190, 233, 276, 323,
370, 422, 473, 530, 587, 650, 714, 786,
858, 940, 1023, 1121, 1219, 1339, 1458, 1612,
1765, 1980, 2195, 2557, 2919
};
static inline void filter_samples(G722Context *c, const int16_t *samples,
int *xlow, int *xhigh)
{
int xout1, xout2;
c->prev_samples[c->prev_samples_pos++] = samples[0];
c->prev_samples[c->prev_samples_pos++] = samples[1];
ff_g722_apply_qmf(c->prev_samples + c->prev_samples_pos - 24, &xout1, &xout2);
*xlow = xout1 + xout2 >> 13;
*xhigh = xout1 - xout2 >> 13;
if (c->prev_samples_pos >= PREV_SAMPLES_BUF_SIZE) {
memmove(c->prev_samples,
c->prev_samples + c->prev_samples_pos - 22,
22 * sizeof(c->prev_samples[0]));
c->prev_samples_pos = 22;
}
}
static inline int encode_high(const struct G722Band *state, int xhigh)
{
int diff = av_clip_int16(xhigh - state->s_predictor);
int pred = 141 * state->scale_factor >> 8;
/* = diff >= 0 ? (diff < pred) + 2 : diff >= -pred */
return ((diff ^ (diff >> (sizeof(diff)*8-1))) < pred) + 2*(diff >= 0);
}
static inline int encode_low(const struct G722Band* state, int xlow)
{
int diff = av_clip_int16(xlow - state->s_predictor);
/* = diff >= 0 ? diff : -(diff + 1) */
int limit = diff ^ (diff >> (sizeof(diff)*8-1));
int i = 0;
limit = limit + 1 << 10;
if (limit > low_quant[8] * state->scale_factor)
i = 9;
while (i < 29 && limit > low_quant[i] * state->scale_factor)
i++;
return (diff < 0 ? (i < 2 ? 63 : 33) : 61) - i;
}
static int g722_encode_trellis(AVCodecContext *avctx,
uint8_t *dst, int buf_size, void *data)
{
G722Context *c = avctx->priv_data;
const int16_t *samples = data;
int i, j, k;
int frontier = 1 << avctx->trellis;
struct TrellisNode **nodes[2];
struct TrellisNode **nodes_next[2];
int pathn[2] = {0, 0}, froze = -1;
struct TrellisPath *p[2];
for (i = 0; i < 2; i++) {
nodes[i] = c->nodep_buf[i];
nodes_next[i] = c->nodep_buf[i] + frontier;
memset(c->nodep_buf[i], 0, 2 * frontier * sizeof(*c->nodep_buf));
nodes[i][0] = c->node_buf[i] + frontier;
nodes[i][0]->ssd = 0;
nodes[i][0]->path = 0;
nodes[i][0]->state = c->band[i];
}
for (i = 0; i < buf_size >> 1; i++) {
int xlow, xhigh;
struct TrellisNode *next[2];
int heap_pos[2] = {0, 0};
for (j = 0; j < 2; j++) {
next[j] = c->node_buf[j] + frontier*(i & 1);
memset(nodes_next[j], 0, frontier * sizeof(**nodes_next));
}
filter_samples(c, &samples[2*i], &xlow, &xhigh);
for (j = 0; j < frontier && nodes[0][j]; j++) {
/* Only k >> 2 affects the future adaptive state, therefore testing
* small steps that don't change k >> 2 is useless, the orignal
* value from encode_low is better than them. Since we step k
* in steps of 4, make sure range is a multiple of 4, so that
* we don't miss the original value from encode_low. */
int range = j < frontier/2 ? 4 : 0;
struct TrellisNode *cur_node = nodes[0][j];
int ilow = encode_low(&cur_node->state, xlow);
for (k = ilow - range; k <= ilow + range && k <= 63; k += 4) {
int decoded, dec_diff, pos;
uint32_t ssd;
struct TrellisNode* node;
if (k < 0)
continue;
decoded = av_clip((cur_node->state.scale_factor *
ff_g722_low_inv_quant6[k] >> 10)
+ cur_node->state.s_predictor, -16384, 16383);
dec_diff = xlow - decoded;
#define STORE_NODE(index, UPDATE, VALUE)\
ssd = cur_node->ssd + dec_diff*dec_diff;\
/* Check for wraparound. Using 64 bit ssd counters would \
* be simpler, but is slower on x86 32 bit. */\
if (ssd < cur_node->ssd)\
continue;\
if (heap_pos[index] < frontier) {\
pos = heap_pos[index]++;\
assert(pathn[index] < FREEZE_INTERVAL * frontier);\
node = nodes_next[index][pos] = next[index]++;\
node->path = pathn[index]++;\
} else {\
/* Try to replace one of the leaf nodes with the new \
* one, but not always testing the same leaf position */\
pos = (frontier>>1) + (heap_pos[index] & ((frontier>>1) - 1));\
if (ssd >= nodes_next[index][pos]->ssd)\
continue;\
heap_pos[index]++;\
node = nodes_next[index][pos];\
}\
node->ssd = ssd;\
node->state = cur_node->state;\
UPDATE;\
c->paths[index][node->path].value = VALUE;\
c->paths[index][node->path].prev = cur_node->path;\
/* Sift the newly inserted node up in the heap to restore \
* the heap property */\
while (pos > 0) {\
int parent = (pos - 1) >> 1;\
if (nodes_next[index][parent]->ssd <= ssd)\
break;\
FFSWAP(struct TrellisNode*, nodes_next[index][parent],\
nodes_next[index][pos]);\
pos = parent;\
}
STORE_NODE(0, ff_g722_update_low_predictor(&node->state, k >> 2), k);
}
}
for (j = 0; j < frontier && nodes[1][j]; j++) {
int ihigh;
struct TrellisNode *cur_node = nodes[1][j];
/* We don't try to get any initial guess for ihigh via
* encode_high - since there's only 4 possible values, test
* them all. Testing all of these gives a much, much larger
* gain than testing a larger range around ilow. */
for (ihigh = 0; ihigh < 4; ihigh++) {
int dhigh, decoded, dec_diff, pos;
uint32_t ssd;
struct TrellisNode* node;
dhigh = cur_node->state.scale_factor *
ff_g722_high_inv_quant[ihigh] >> 10;
decoded = av_clip(dhigh + cur_node->state.s_predictor,
-16384, 16383);
dec_diff = xhigh - decoded;
STORE_NODE(1, ff_g722_update_high_predictor(&node->state, dhigh, ihigh), ihigh);
}
}
for (j = 0; j < 2; j++) {
FFSWAP(struct TrellisNode**, nodes[j], nodes_next[j]);
if (nodes[j][0]->ssd > (1 << 16)) {
for (k = 1; k < frontier && nodes[j][k]; k++)
nodes[j][k]->ssd -= nodes[j][0]->ssd;
nodes[j][0]->ssd = 0;
}
}
if (i == froze + FREEZE_INTERVAL) {
p[0] = &c->paths[0][nodes[0][0]->path];
p[1] = &c->paths[1][nodes[1][0]->path];
for (j = i; j > froze; j--) {
dst[j] = p[1]->value << 6 | p[0]->value;
p[0] = &c->paths[0][p[0]->prev];
p[1] = &c->paths[1][p[1]->prev];
}
froze = i;
pathn[0] = pathn[1] = 0;
memset(nodes[0] + 1, 0, (frontier - 1)*sizeof(**nodes));
memset(nodes[1] + 1, 0, (frontier - 1)*sizeof(**nodes));
}
}
p[0] = &c->paths[0][nodes[0][0]->path];
p[1] = &c->paths[1][nodes[1][0]->path];
for (j = i; j > froze; j--) {
dst[j] = p[1]->value << 6 | p[0]->value;
p[0] = &c->paths[0][p[0]->prev];
p[1] = &c->paths[1][p[1]->prev];
}
c->band[0] = nodes[0][0]->state;
c->band[1] = nodes[1][0]->state;
return i;
}
static int g722_encode_frame(AVCodecContext *avctx,
uint8_t *dst, int buf_size, void *data)
{
G722Context *c = avctx->priv_data;
const int16_t *samples = data;
int i;
if (avctx->trellis)
return g722_encode_trellis(avctx, dst, buf_size, data);
for (i = 0; i < buf_size >> 1; i++) {
int xlow, xhigh, ihigh, ilow;
filter_samples(c, &samples[2*i], &xlow, &xhigh);
ihigh = encode_high(&c->band[1], xhigh);
ilow = encode_low(&c->band[0], xlow);
ff_g722_update_high_predictor(&c->band[1], c->band[1].scale_factor *
ff_g722_high_inv_quant[ihigh] >> 10, ihigh);
ff_g722_update_low_predictor(&c->band[0], ilow >> 2);
*dst++ = ihigh << 6 | ilow;
}
return i;
}
AVCodec ff_adpcm_g722_encoder = {
.name = "g722",
.type = AVMEDIA_TYPE_AUDIO,
.id = CODEC_ID_ADPCM_G722,
.priv_data_size = sizeof(G722Context),
.init = g722_encode_init,
.close = g722_encode_close,
.encode = g722_encode_frame,
.long_name = NULL_IF_CONFIG_SMALL("G.722 ADPCM"),
.sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
};