* FLAC audio encoder
* Copyright (c) 2006 Justin Ruggles <jruggle@earthlink.net>
*
- * This library is free software; you can redistribute it and/or
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
*
- * This library is distributed in the hope that it will be useful,
+ * FFmpeg 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 this library; if not, write to the Free Software
+ * License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "avcodec.h"
#include "bitstream.h"
#include "crc.h"
+#include "dsputil.h"
#include "golomb.h"
+#include "lls.h"
#define FLAC_MAX_CH 8
#define FLAC_MIN_BLOCKSIZE 16
#define FLAC_CHMODE_RIGHT_SIDE 9
#define FLAC_CHMODE_MID_SIDE 10
+#define ORDER_METHOD_EST 0
+#define ORDER_METHOD_2LEVEL 1
+#define ORDER_METHOD_4LEVEL 2
+#define ORDER_METHOD_8LEVEL 3
+#define ORDER_METHOD_SEARCH 4
+#define ORDER_METHOD_LOG 5
+
#define FLAC_STREAMINFO_SIZE 34
+#define MIN_LPC_ORDER 1
+#define MAX_LPC_ORDER 32
+#define MAX_FIXED_ORDER 4
+#define MAX_PARTITION_ORDER 8
+#define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER)
+#define MAX_LPC_PRECISION 15
+#define MAX_LPC_SHIFT 15
+#define MAX_RICE_PARAM 14
+
+typedef struct CompressionOptions {
+ int compression_level;
+ int block_time_ms;
+ int use_lpc;
+ int lpc_coeff_precision;
+ int min_prediction_order;
+ int max_prediction_order;
+ int prediction_order_method;
+ int min_partition_order;
+ int max_partition_order;
+} CompressionOptions;
+
typedef struct RiceContext {
int porder;
- int params[256];
+ int params[MAX_PARTITIONS];
} RiceContext;
typedef struct FlacSubframe {
int type_code;
int obits;
int order;
+ int32_t coefs[MAX_LPC_ORDER];
+ int shift;
RiceContext rc;
int32_t samples[FLAC_MAX_BLOCKSIZE];
- int32_t residual[FLAC_MAX_BLOCKSIZE];
+ int32_t residual[FLAC_MAX_BLOCKSIZE+1];
} FlacSubframe;
typedef struct FlacFrame {
int ch_code;
int samplerate;
int sr_code[2];
- int blocksize;
int max_framesize;
uint32_t frame_count;
FlacFrame frame;
+ CompressionOptions options;
AVCodecContext *avctx;
+ DSPContext dsp;
} FlacEncodeContext;
static const int flac_samplerates[16] = {
init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);
/* streaminfo metadata block */
- put_bits(&pb, 16, s->blocksize);
- put_bits(&pb, 16, s->blocksize);
+ put_bits(&pb, 16, s->avctx->frame_size);
+ put_bits(&pb, 16, s->avctx->frame_size);
put_bits(&pb, 24, 0);
put_bits(&pb, 24, s->max_framesize);
put_bits(&pb, 20, s->samplerate);
/* MD5 signature = 0 */
}
-#define BLOCK_TIME_MS 27
-
/**
* Sets blocksize based on samplerate
* Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
*/
-static int select_blocksize(int samplerate)
+static int select_blocksize(int samplerate, int block_time_ms)
{
int i;
int target;
assert(samplerate > 0);
blocksize = flac_blocksizes[1];
- target = (samplerate * BLOCK_TIME_MS) / 1000;
+ target = (samplerate * block_time_ms) / 1000;
for(i=0; i<16; i++) {
if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
blocksize = flac_blocksizes[i];
return blocksize;
}
-static int flac_encode_init(AVCodecContext *avctx)
+static av_cold int flac_encode_init(AVCodecContext *avctx)
{
int freq = avctx->sample_rate;
int channels = avctx->channels;
FlacEncodeContext *s = avctx->priv_data;
- int i;
+ int i, level;
uint8_t *streaminfo;
s->avctx = avctx;
+ dsputil_init(&s->dsp, avctx);
+
if(avctx->sample_fmt != SAMPLE_FMT_S16) {
return -1;
}
s->samplerate = freq;
}
- s->blocksize = select_blocksize(s->samplerate);
- avctx->frame_size = s->blocksize;
+ /* set compression option defaults based on avctx->compression_level */
+ if(avctx->compression_level < 0) {
+ s->options.compression_level = 5;
+ } else {
+ s->options.compression_level = avctx->compression_level;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);
+
+ level= s->options.compression_level;
+ if(level > 12) {
+ av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
+ s->options.compression_level);
+ return -1;
+ }
+
+ s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
+ s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
+ s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level];
+ s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level];
+ s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
+ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST,
+ ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL,
+ ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
+ ORDER_METHOD_SEARCH})[level];
+ s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level];
+ s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level];
+
+ /* set compression option overrides from AVCodecContext */
+ if(avctx->use_lpc >= 0) {
+ s->options.use_lpc = av_clip(avctx->use_lpc, 0, 11);
+ }
+ if(s->options.use_lpc == 1)
+ av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n");
+ else if(s->options.use_lpc > 1)
+ av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n");
+
+ if(avctx->min_prediction_order >= 0) {
+ if(s->options.use_lpc) {
+ if(avctx->min_prediction_order < MIN_LPC_ORDER ||
+ avctx->min_prediction_order > MAX_LPC_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
+ avctx->min_prediction_order);
+ return -1;
+ }
+ } else {
+ if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
+ avctx->min_prediction_order);
+ return -1;
+ }
+ }
+ s->options.min_prediction_order = avctx->min_prediction_order;
+ }
+ if(avctx->max_prediction_order >= 0) {
+ if(s->options.use_lpc) {
+ if(avctx->max_prediction_order < MIN_LPC_ORDER ||
+ avctx->max_prediction_order > MAX_LPC_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
+ avctx->max_prediction_order);
+ return -1;
+ }
+ } else {
+ if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
+ avctx->max_prediction_order);
+ return -1;
+ }
+ }
+ s->options.max_prediction_order = avctx->max_prediction_order;
+ }
+ if(s->options.max_prediction_order < s->options.min_prediction_order) {
+ av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
+ s->options.min_prediction_order, s->options.max_prediction_order);
+ return -1;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
+ s->options.min_prediction_order, s->options.max_prediction_order);
+
+ if(avctx->prediction_order_method >= 0) {
+ if(avctx->prediction_order_method > ORDER_METHOD_LOG) {
+ av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
+ avctx->prediction_order_method);
+ return -1;
+ }
+ s->options.prediction_order_method = avctx->prediction_order_method;
+ }
+ switch(s->options.prediction_order_method) {
+ case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "estimate"); break;
+ case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "2-level"); break;
+ case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "4-level"); break;
+ case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "8-level"); break;
+ case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "full search"); break;
+ case ORDER_METHOD_LOG: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
+ "log search"); break;
+ }
+
+ if(avctx->min_partition_order >= 0) {
+ if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
+ avctx->min_partition_order);
+ return -1;
+ }
+ s->options.min_partition_order = avctx->min_partition_order;
+ }
+ if(avctx->max_partition_order >= 0) {
+ if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
+ av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
+ avctx->max_partition_order);
+ return -1;
+ }
+ s->options.max_partition_order = avctx->max_partition_order;
+ }
+ if(s->options.max_partition_order < s->options.min_partition_order) {
+ av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
+ s->options.min_partition_order, s->options.max_partition_order);
+ return -1;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
+ s->options.min_partition_order, s->options.max_partition_order);
+
+ if(avctx->frame_size > 0) {
+ if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
+ avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
+ av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
+ avctx->frame_size);
+ return -1;
+ }
+ } else {
+ s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms);
+ }
+ av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->avctx->frame_size);
+
+ /* set LPC precision */
+ if(avctx->lpc_coeff_precision > 0) {
+ if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
+ av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
+ avctx->lpc_coeff_precision);
+ return -1;
+ }
+ s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
+ } else {
+ /* default LPC precision */
+ s->options.lpc_coeff_precision = 15;
+ }
+ av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
+ s->options.lpc_coeff_precision);
/* set maximum encoded frame size in verbatim mode */
if(s->channels == 2) {
- s->max_framesize = 14 + ((s->blocksize * 33 + 7) >> 3);
+ s->max_framesize = 14 + ((s->avctx->frame_size * 33 + 7) >> 3);
} else {
- s->max_framesize = 14 + (s->blocksize * s->channels * 2);
+ s->max_framesize = 14 + (s->avctx->frame_size * s->channels * 2);
}
streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
frame = &s->frame;
for(i=0; i<16; i++) {
- if(s->blocksize == flac_blocksizes[i]) {
+ if(s->avctx->frame_size == flac_blocksizes[i]) {
frame->blocksize = flac_blocksizes[i];
frame->bs_code[0] = i;
frame->bs_code[1] = 0;
}
}
if(i == 16) {
- frame->blocksize = s->blocksize;
+ frame->blocksize = s->avctx->frame_size;
if(frame->blocksize <= 256) {
frame->bs_code[0] = 6;
frame->bs_code[1] = frame->blocksize-1;
#define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))
+/**
+ * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
+ */
static int find_optimal_param(uint32_t sum, int n)
{
- int k, k_opt;
- uint32_t nbits, nbits_opt;
+ int k;
+ uint32_t sum2;
- k_opt = 0;
- nbits_opt = rice_encode_count(sum, n, 0);
- for(k=1; k<=14; k++) {
- nbits = rice_encode_count(sum, n, k);
- if(nbits < nbits_opt) {
- nbits_opt = nbits;
- k_opt = k;
- }
- }
- return k_opt;
+ if(sum <= n>>1)
+ return 0;
+ sum2 = sum-(n>>1);
+ k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
+ return FFMIN(k, MAX_RICE_PARAM);
}
static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
uint32_t all_bits;
part = (1 << porder);
- all_bits = 0;
+ all_bits = 4 * part;
cnt = (n >> porder) - pred_order;
for(i=0; i<part; i++) {
- if(i == 1) cnt = (n >> porder);
k = find_optimal_param(sums[i], cnt);
rc->params[i] = k;
all_bits += rice_encode_count(sums[i], cnt, k);
+ cnt = n >> porder;
}
- all_bits += (4 * part);
rc->porder = porder;
return all_bits;
}
-static void calc_sums(int pmax, uint32_t *data, int n, int pred_order,
- uint32_t sums[][256])
+static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
+ uint32_t sums[][MAX_PARTITIONS])
{
int i, j;
int parts;
res = &data[pred_order];
res_end = &data[n >> pmax];
for(i=0; i<parts; i++) {
- sums[pmax][i] = 0;
+ uint32_t sum = 0;
while(res < res_end){
- sums[pmax][i] += *(res++);
+ sum += *(res++);
}
+ sums[pmax][i] = sum;
res_end+= n >> pmax;
}
/* sums for lower levels */
- for(i=pmax-1; i>=0; i--) {
+ for(i=pmax-1; i>=pmin; i--) {
parts = (1 << i);
for(j=0; j<parts; j++) {
sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
}
}
-static uint32_t calc_rice_params(RiceContext *rc, int pmax, int32_t *data,
- int n, int pred_order)
+static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
+ int32_t *data, int n, int pred_order)
{
int i;
- uint32_t bits, opt_bits;
+ uint32_t bits[MAX_PARTITION_ORDER+1];
int opt_porder;
- RiceContext opt_rc;
+ RiceContext tmp_rc;
uint32_t *udata;
- uint32_t sums[9][256];
+ uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
- assert(pmax >= 0 && pmax <= 8);
+ assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
+ assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
+ assert(pmin <= pmax);
udata = av_malloc(n * sizeof(uint32_t));
for(i=0; i<n; i++) {
udata[i] = (2*data[i]) ^ (data[i]>>31);
}
- calc_sums(pmax, udata, n, pred_order, sums);
+ calc_sums(pmin, pmax, udata, n, pred_order, sums);
- opt_porder = 0;
- opt_bits = UINT32_MAX;
- for(i=0; i<=pmax; i++) {
- bits = calc_optimal_rice_params(rc, i, sums[i], n, pred_order);
- if(bits < opt_bits) {
- opt_bits = bits;
+ opt_porder = pmin;
+ bits[pmin] = UINT32_MAX;
+ for(i=pmin; i<=pmax; i++) {
+ bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
+ if(bits[i] <= bits[opt_porder]) {
opt_porder = i;
- memcpy(&opt_rc, rc, sizeof(RiceContext));
+ *rc= tmp_rc;
}
}
- if(opt_porder != pmax) {
- memcpy(rc, &opt_rc, sizeof(RiceContext));
- }
av_freep(&udata);
- return opt_bits;
+ return bits[opt_porder];
+}
+
+static int get_max_p_order(int max_porder, int n, int order)
+{
+ int porder = FFMIN(max_porder, av_log2(n^(n-1)));
+ if(order > 0)
+ porder = FFMIN(porder, av_log2(n/order));
+ return porder;
}
-static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmax, int32_t *data,
- int n, int pred_order, int bps)
+static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
+ int32_t *data, int n, int pred_order,
+ int bps)
{
uint32_t bits;
+ pmin = get_max_p_order(pmin, n, pred_order);
+ pmax = get_max_p_order(pmax, n, pred_order);
bits = pred_order*bps + 6;
- bits += calc_rice_params(rc, pmax, data, n, pred_order);
+ bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
+ return bits;
+}
+
+static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
+ int32_t *data, int n, int pred_order,
+ int bps, int precision)
+{
+ uint32_t bits;
+ pmin = get_max_p_order(pmin, n, pred_order);
+ pmax = get_max_p_order(pmax, n, pred_order);
+ bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
+ bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
return bits;
}
+/**
+ * Apply Welch window function to audio block
+ */
+static void apply_welch_window(const int32_t *data, int len, double *w_data)
+{
+ int i, n2;
+ double w;
+ double c;
+
+ assert(!(len&1)); //the optimization in r11881 does not support odd len
+ //if someone wants odd len extend the change in r11881
+
+ n2 = (len >> 1);
+ c = 2.0 / (len - 1.0);
+
+ w_data+=n2;
+ data+=n2;
+ for(i=0; i<n2; i++) {
+ w = c - n2 + i;
+ w = 1.0 - (w * w);
+ w_data[-i-1] = data[-i-1] * w;
+ w_data[+i ] = data[+i ] * w;
+ }
+}
+
+/**
+ * Calculates autocorrelation data from audio samples
+ * A Welch window function is applied before calculation.
+ */
+void ff_flac_compute_autocorr(const int32_t *data, int len, int lag,
+ double *autoc)
+{
+ int i, j;
+ double tmp[len + lag + 1];
+ double *data1= tmp + lag;
+
+ apply_welch_window(data, len, data1);
+
+ for(j=0; j<lag; j++)
+ data1[j-lag]= 0.0;
+ data1[len] = 0.0;
+
+ for(j=0; j<lag; j+=2){
+ double sum0 = 1.0, sum1 = 1.0;
+ for(i=0; i<len; i++){
+ sum0 += data1[i] * data1[i-j];
+ sum1 += data1[i] * data1[i-j-1];
+ }
+ autoc[j ] = sum0;
+ autoc[j+1] = sum1;
+ }
+
+ if(j==lag){
+ double sum = 1.0;
+ for(i=0; i<len; i+=2){
+ sum += data1[i ] * data1[i-j ]
+ + data1[i+1] * data1[i-j+1];
+ }
+ autoc[j] = sum;
+ }
+}
+
+/**
+ * Levinson-Durbin recursion.
+ * Produces LPC coefficients from autocorrelation data.
+ */
+static void compute_lpc_coefs(const double *autoc, int max_order,
+ double lpc[][MAX_LPC_ORDER], double *ref)
+{
+ int i, j, i2;
+ double r, err, tmp;
+ double lpc_tmp[MAX_LPC_ORDER];
+
+ for(i=0; i<max_order; i++) lpc_tmp[i] = 0;
+ err = autoc[0];
+
+ for(i=0; i<max_order; i++) {
+ r = -autoc[i+1];
+ for(j=0; j<i; j++) {
+ r -= lpc_tmp[j] * autoc[i-j];
+ }
+ r /= err;
+ ref[i] = fabs(r);
+
+ err *= 1.0 - (r * r);
+
+ i2 = (i >> 1);
+ lpc_tmp[i] = r;
+ for(j=0; j<i2; j++) {
+ tmp = lpc_tmp[j];
+ lpc_tmp[j] += r * lpc_tmp[i-1-j];
+ lpc_tmp[i-1-j] += r * tmp;
+ }
+ if(i & 1) {
+ lpc_tmp[j] += lpc_tmp[j] * r;
+ }
+
+ for(j=0; j<=i; j++) {
+ lpc[i][j] = -lpc_tmp[j];
+ }
+ }
+}
+
+/**
+ * Quantize LPC coefficients
+ */
+static void quantize_lpc_coefs(double *lpc_in, int order, int precision,
+ int32_t *lpc_out, int *shift)
+{
+ int i;
+ double cmax, error;
+ int32_t qmax;
+ int sh;
+
+ /* define maximum levels */
+ qmax = (1 << (precision - 1)) - 1;
+
+ /* find maximum coefficient value */
+ cmax = 0.0;
+ for(i=0; i<order; i++) {
+ cmax= FFMAX(cmax, fabs(lpc_in[i]));
+ }
+
+ /* if maximum value quantizes to zero, return all zeros */
+ if(cmax * (1 << MAX_LPC_SHIFT) < 1.0) {
+ *shift = 0;
+ memset(lpc_out, 0, sizeof(int32_t) * order);
+ return;
+ }
+
+ /* calculate level shift which scales max coeff to available bits */
+ sh = MAX_LPC_SHIFT;
+ while((cmax * (1 << sh) > qmax) && (sh > 0)) {
+ sh--;
+ }
+
+ /* since negative shift values are unsupported in decoder, scale down
+ coefficients instead */
+ if(sh == 0 && cmax > qmax) {
+ double scale = ((double)qmax) / cmax;
+ for(i=0; i<order; i++) {
+ lpc_in[i] *= scale;
+ }
+ }
+
+ /* output quantized coefficients and level shift */
+ error=0;
+ for(i=0; i<order; i++) {
+ error += lpc_in[i] * (1 << sh);
+ lpc_out[i] = av_clip(lrintf(error), -qmax, qmax);
+ error -= lpc_out[i];
+ }
+ *shift = sh;
+}
+
+static int estimate_best_order(double *ref, int max_order)
+{
+ int i, est;
+
+ est = 1;
+ for(i=max_order-1; i>=0; i--) {
+ if(ref[i] > 0.10) {
+ est = i+1;
+ break;
+ }
+ }
+ return est;
+}
+
+/**
+ * Calculate LPC coefficients for multiple orders
+ */
+static int lpc_calc_coefs(FlacEncodeContext *s,
+ const int32_t *samples, int blocksize, int max_order,
+ int precision, int32_t coefs[][MAX_LPC_ORDER],
+ int *shift, int use_lpc, int omethod)
+{
+ double autoc[MAX_LPC_ORDER+1];
+ double ref[MAX_LPC_ORDER];
+ double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER];
+ int i, j, pass;
+ int opt_order;
+
+ assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER);
+
+ if(use_lpc == 1){
+ s->dsp.flac_compute_autocorr(samples, blocksize, max_order, autoc);
+
+ compute_lpc_coefs(autoc, max_order, lpc, ref);
+ }else{
+ LLSModel m[2];
+ double var[MAX_LPC_ORDER+1], weight;
+
+ for(pass=0; pass<use_lpc-1; pass++){
+ av_init_lls(&m[pass&1], max_order);
+
+ weight=0;
+ for(i=max_order; i<blocksize; i++){
+ for(j=0; j<=max_order; j++)
+ var[j]= samples[i-j];
+
+ if(pass){
+ double eval, inv, rinv;
+ eval= av_evaluate_lls(&m[(pass-1)&1], var+1, max_order-1);
+ eval= (512>>pass) + fabs(eval - var[0]);
+ inv = 1/eval;
+ rinv = sqrt(inv);
+ for(j=0; j<=max_order; j++)
+ var[j] *= rinv;
+ weight += inv;
+ }else
+ weight++;
+
+ av_update_lls(&m[pass&1], var, 1.0);
+ }
+ av_solve_lls(&m[pass&1], 0.001, 0);
+ }
+
+ for(i=0; i<max_order; i++){
+ for(j=0; j<max_order; j++)
+ lpc[i][j]= m[(pass-1)&1].coeff[i][j];
+ ref[i]= sqrt(m[(pass-1)&1].variance[i] / weight) * (blocksize - max_order) / 4000;
+ }
+ for(i=max_order-1; i>0; i--)
+ ref[i] = ref[i-1] - ref[i];
+ }
+ opt_order = max_order;
+
+ if(omethod == ORDER_METHOD_EST) {
+ opt_order = estimate_best_order(ref, max_order);
+ i = opt_order-1;
+ quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
+ } else {
+ for(i=0; i<max_order; i++) {
+ quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
+ }
+ }
+
+ return opt_order;
+}
+
+
static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
{
assert(n > 0);
memcpy(res, smp, n * sizeof(int32_t));
}
-static void encode_residual_fixed(int32_t *res, int32_t *smp, int n, int order)
+static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
+ int order)
{
int i;
for(i=order; i<n; i++)
res[i]= smp[i] - smp[i-1];
}else if(order==2){
- for(i=order; i<n; i++)
- res[i]= smp[i] - 2*smp[i-1] + smp[i-2];
+ int a = smp[order-1] - smp[order-2];
+ for(i=order; i<n; i+=2) {
+ int b = smp[i] - smp[i-1];
+ res[i]= b - a;
+ a = smp[i+1] - smp[i];
+ res[i+1]= a - b;
+ }
}else if(order==3){
- for(i=order; i<n; i++)
- res[i]= smp[i] - 3*smp[i-1] + 3*smp[i-2] - smp[i-3];
+ int a = smp[order-1] - smp[order-2];
+ int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
+ for(i=order; i<n; i+=2) {
+ int b = smp[i] - smp[i-1];
+ int d = b - a;
+ res[i]= d - c;
+ a = smp[i+1] - smp[i];
+ c = a - b;
+ res[i+1]= c - d;
+ }
}else{
- for(i=order; i<n; i++)
- res[i]= smp[i] - 4*smp[i-1] + 6*smp[i-2] - 4*smp[i-3] + smp[i-4];
+ int a = smp[order-1] - smp[order-2];
+ int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
+ int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
+ for(i=order; i<n; i+=2) {
+ int b = smp[i] - smp[i-1];
+ int d = b - a;
+ int f = d - c;
+ res[i]= f - e;
+ a = smp[i+1] - smp[i];
+ c = a - b;
+ e = c - d;
+ res[i+1]= e - f;
+ }
}
}
-static int get_max_p_order(int max_porder, int n, int order)
+#define LPC1(x) {\
+ int c = coefs[(x)-1];\
+ p0 += c*s;\
+ s = smp[i-(x)+1];\
+ p1 += c*s;\
+}
+
+static av_always_inline void encode_residual_lpc_unrolled(
+ int32_t *res, const int32_t *smp, int n,
+ int order, const int32_t *coefs, int shift, int big)
{
- int porder, max_parts;
+ int i;
+ for(i=order; i<n; i+=2) {
+ int s = smp[i-order];
+ int p0 = 0, p1 = 0;
+ if(big) {
+ switch(order) {
+ case 32: LPC1(32)
+ case 31: LPC1(31)
+ case 30: LPC1(30)
+ case 29: LPC1(29)
+ case 28: LPC1(28)
+ case 27: LPC1(27)
+ case 26: LPC1(26)
+ case 25: LPC1(25)
+ case 24: LPC1(24)
+ case 23: LPC1(23)
+ case 22: LPC1(22)
+ case 21: LPC1(21)
+ case 20: LPC1(20)
+ case 19: LPC1(19)
+ case 18: LPC1(18)
+ case 17: LPC1(17)
+ case 16: LPC1(16)
+ case 15: LPC1(15)
+ case 14: LPC1(14)
+ case 13: LPC1(13)
+ case 12: LPC1(12)
+ case 11: LPC1(11)
+ case 10: LPC1(10)
+ case 9: LPC1( 9)
+ LPC1( 8)
+ LPC1( 7)
+ LPC1( 6)
+ LPC1( 5)
+ LPC1( 4)
+ LPC1( 3)
+ LPC1( 2)
+ LPC1( 1)
+ }
+ } else {
+ switch(order) {
+ case 8: LPC1( 8)
+ case 7: LPC1( 7)
+ case 6: LPC1( 6)
+ case 5: LPC1( 5)
+ case 4: LPC1( 4)
+ case 3: LPC1( 3)
+ case 2: LPC1( 2)
+ case 1: LPC1( 1)
+ }
+ }
+ res[i ] = smp[i ] - (p0 >> shift);
+ res[i+1] = smp[i+1] - (p1 >> shift);
+ }
+}
- porder = max_porder;
- while(porder > 0) {
- max_parts = (1 << porder);
- if(!(n % max_parts) && (n > max_parts*order)) {
- break;
+static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
+ int order, const int32_t *coefs, int shift)
+{
+ int i;
+ for(i=0; i<order; i++) {
+ res[i] = smp[i];
+ }
+#ifdef CONFIG_SMALL
+ for(i=order; i<n; i+=2) {
+ int j;
+ int s = smp[i];
+ int p0 = 0, p1 = 0;
+ for(j=0; j<order; j++) {
+ int c = coefs[j];
+ p1 += c*s;
+ s = smp[i-j-1];
+ p0 += c*s;
}
- porder--;
+ res[i ] = smp[i ] - (p0 >> shift);
+ res[i+1] = smp[i+1] - (p1 >> shift);
}
- return porder;
+#else
+ switch(order) {
+ case 1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
+ case 2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
+ case 3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
+ case 4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
+ case 5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
+ case 6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
+ case 7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
+ case 8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
+ default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
+ }
+#endif
}
static int encode_residual(FlacEncodeContext *ctx, int ch)
{
- int i, opt_order, porder, max_porder, n;
+ int i, n;
+ int min_order, max_order, opt_order, precision, omethod;
+ int min_porder, max_porder;
FlacFrame *frame;
FlacSubframe *sub;
- uint32_t bits[5];
+ int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
+ int shift[MAX_LPC_ORDER];
int32_t *res, *smp;
frame = &ctx->frame;
return sub->obits * n;
}
- max_porder = 3;
+ min_order = ctx->options.min_prediction_order;
+ max_order = ctx->options.max_prediction_order;
+ min_porder = ctx->options.min_partition_order;
+ max_porder = ctx->options.max_partition_order;
+ precision = ctx->options.lpc_coeff_precision;
+ omethod = ctx->options.prediction_order_method;
/* FIXED */
- opt_order = 0;
- bits[0] = UINT32_MAX;
- for(i=0; i<=4; i++) {
- encode_residual_fixed(res, smp, n, i);
- porder = get_max_p_order(max_porder, n, i);
- bits[i] = calc_rice_params_fixed(&sub->rc, porder, res, n, i, sub->obits);
- if(bits[i] < bits[opt_order]) {
- opt_order = i;
+ if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
+ uint32_t bits[MAX_FIXED_ORDER+1];
+ if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
+ opt_order = 0;
+ bits[0] = UINT32_MAX;
+ for(i=min_order; i<=max_order; i++) {
+ encode_residual_fixed(res, smp, n, i);
+ bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
+ n, i, sub->obits);
+ if(bits[i] < bits[opt_order]) {
+ opt_order = i;
+ }
+ }
+ sub->order = opt_order;
+ sub->type = FLAC_SUBFRAME_FIXED;
+ sub->type_code = sub->type | sub->order;
+ if(sub->order != max_order) {
+ encode_residual_fixed(res, smp, n, sub->order);
+ return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
+ sub->order, sub->obits);
+ }
+ return bits[sub->order];
+ }
+
+ /* LPC */
+ opt_order = lpc_calc_coefs(ctx, smp, n, max_order, precision, coefs, shift, ctx->options.use_lpc, omethod);
+
+ if(omethod == ORDER_METHOD_2LEVEL ||
+ omethod == ORDER_METHOD_4LEVEL ||
+ omethod == ORDER_METHOD_8LEVEL) {
+ int levels = 1 << omethod;
+ uint32_t bits[levels];
+ int order;
+ int opt_index = levels-1;
+ opt_order = max_order-1;
+ bits[opt_index] = UINT32_MAX;
+ for(i=levels-1; i>=0; i--) {
+ order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
+ if(order < 0) order = 0;
+ encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
+ bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
+ res, n, order+1, sub->obits, precision);
+ if(bits[i] < bits[opt_index]) {
+ opt_index = i;
+ opt_order = order;
+ }
+ }
+ opt_order++;
+ } else if(omethod == ORDER_METHOD_SEARCH) {
+ // brute-force optimal order search
+ uint32_t bits[MAX_LPC_ORDER];
+ opt_order = 0;
+ bits[0] = UINT32_MAX;
+ for(i=min_order-1; i<max_order; i++) {
+ encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
+ bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
+ res, n, i+1, sub->obits, precision);
+ if(bits[i] < bits[opt_order]) {
+ opt_order = i;
+ }
+ }
+ opt_order++;
+ } else if(omethod == ORDER_METHOD_LOG) {
+ uint32_t bits[MAX_LPC_ORDER];
+ int step;
+
+ opt_order= min_order - 1 + (max_order-min_order)/3;
+ memset(bits, -1, sizeof(bits));
+
+ for(step=16 ;step; step>>=1){
+ int last= opt_order;
+ for(i=last-step; i<=last+step; i+= step){
+ if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
+ continue;
+ encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
+ bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
+ res, n, i+1, sub->obits, precision);
+ if(bits[i] < bits[opt_order])
+ opt_order= i;
+ }
}
+ opt_order++;
}
+
sub->order = opt_order;
- sub->type = FLAC_SUBFRAME_FIXED;
- sub->type_code = sub->type | sub->order;
- if(sub->order != 4) {
- encode_residual_fixed(res, smp, n, sub->order);
- porder = get_max_p_order(max_porder, n, sub->order);
- calc_rice_params_fixed(&sub->rc, porder, res, n, sub->order, sub->obits);
+ sub->type = FLAC_SUBFRAME_LPC;
+ sub->type_code = sub->type | (sub->order-1);
+ sub->shift = shift[sub->order-1];
+ for(i=0; i<sub->order; i++) {
+ sub->coefs[i] = coefs[sub->order-1][i];
}
- return bits[sub->order];
+ encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
+ return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order,
+ sub->obits, precision);
}
static int encode_residual_v(FlacEncodeContext *ctx, int ch)
uint64_t score[4];
int k;
- /* calculate sum of squares for each channel */
+ /* calculate sum of 2nd order residual for each channel */
sum[0] = sum[1] = sum[2] = sum[3] = 0;
for(i=2; i<n; i++) {
lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
- sum[2] += ABS((lt + rt) >> 1);
- sum[3] += ABS(lt - rt);
- sum[0] += ABS(lt);
- sum[1] += ABS(rt);
+ sum[2] += FFABS((lt + rt) >> 1);
+ sum[3] += FFABS(lt - rt);
+ sum[0] += FFABS(lt);
+ sum[1] += FFABS(rt);
}
+ /* estimate bit counts */
for(i=0; i<4; i++) {
k = find_optimal_param(2*sum[i], n);
sum[i] = rice_encode_count(2*sum[i], n, k);
static void write_utf8(PutBitContext *pb, uint32_t val)
{
- int bytes, shift;
-
- if(val < 0x80){
- put_bits(pb, 8, val);
- return;
- }
-
- bytes= (av_log2(val)+4) / 5;
- shift = (bytes - 1) * 6;
- put_bits(pb, 8, (256 - (256>>bytes)) | (val >> shift));
- while(shift >= 6){
- shift -= 6;
- put_bits(pb, 8, 0x80 | ((val >> shift) & 0x3F));
- }
+ uint8_t tmp;
+ PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
}
static void output_frame_header(FlacEncodeContext *s)
put_bits(&s->pb, 16, s->sr_code[1]);
}
flush_put_bits(&s->pb);
- crc = av_crc(av_crc07, 0, s->pb.buf, put_bits_count(&s->pb)>>3);
+ crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
+ s->pb.buf, put_bits_count(&s->pb)>>3);
put_bits(&s->pb, 8, crc);
}
output_residual(ctx, ch);
}
+static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
+{
+ int i, cbits;
+ FlacFrame *frame;
+ FlacSubframe *sub;
+
+ frame = &ctx->frame;
+ sub = &frame->subframes[ch];
+
+ /* warm-up samples */
+ for(i=0; i<sub->order; i++) {
+ put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
+ }
+
+ /* LPC coefficients */
+ cbits = ctx->options.lpc_coeff_precision;
+ put_bits(&ctx->pb, 4, cbits-1);
+ put_sbits(&ctx->pb, 5, sub->shift);
+ for(i=0; i<sub->order; i++) {
+ put_sbits(&ctx->pb, cbits, sub->coefs[i]);
+ }
+
+ /* residual */
+ output_residual(ctx, ch);
+}
+
static void output_subframes(FlacEncodeContext *s)
{
FlacFrame *frame;
output_subframe_verbatim(s, ch);
} else if(sub->type == FLAC_SUBFRAME_FIXED) {
output_subframe_fixed(s, ch);
+ } else if(sub->type == FLAC_SUBFRAME_LPC) {
+ output_subframe_lpc(s, ch);
}
}
}
{
int crc;
flush_put_bits(&s->pb);
- crc = bswap_16(av_crc(av_crc8005, 0, s->pb.buf, put_bits_count(&s->pb)>>3));
+ crc = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
+ s->pb.buf, put_bits_count(&s->pb)>>3));
put_bits(&s->pb, 16, crc);
flush_put_bits(&s->pb);
}
s = avctx->priv_data;
- s->blocksize = avctx->frame_size;
init_frame(s);
copy_samples(s, samples);
return out_bytes;
}
-static int flac_encode_close(AVCodecContext *avctx)
+static av_cold int flac_encode_close(AVCodecContext *avctx)
{
av_freep(&avctx->extradata);
avctx->extradata_size = 0;
projects / ffmpeg / blobdiff