Skip to content

F
ffmpeg.wasm-core
Commit 70b8668f authored by Michael Niedermayer's avatar Michael Niedermayer
Browse files
Options

drop LLS1, rename LLS2 to LLS 

Signed-off-by: 's avatarMichael Niedermayer <michaelni@gmx.at>
parent 5e1832c2
Show whitespace changes
Inline Side-by-side
Showing with 32 additions and 268 deletions
libavcodec/lpc.c
View file @ 70b8668f
...@@ -20,7 +20,7 @@ ...@@ -20,7 +20,7 @@
*/ */
#include "libavutil/common.h" #include "libavutil/common.h"
#include "libavutil/lls2.h" #include "libavutil/lls.h"
#define LPC_USE_DOUBLE #define LPC_USE_DOUBLE
#include "lpc.h" #include "lpc.h"
...@@ -208,7 +208,7 @@ int ff_lpc_calc_coefs(LPCContext *s, ...@@ -208,7 +208,7 @@ int ff_lpc_calc_coefs(LPCContext *s,
} }
if (lpc_type == FF_LPC_TYPE_CHOLESKY) { if (lpc_type == FF_LPC_TYPE_CHOLESKY) {
LLSModel2 m[2]; LLSModel m[2];
LOCAL_ALIGNED(32, double, var, [FFALIGN(MAX_LPC_ORDER+1,4)]); LOCAL_ALIGNED(32, double, var, [FFALIGN(MAX_LPC_ORDER+1,4)]);
double av_uninit(weight); double av_uninit(weight);
memset(var, 0, FFALIGN(MAX_LPC_ORDER+1,4)*sizeof(*var)); memset(var, 0, FFALIGN(MAX_LPC_ORDER+1,4)*sizeof(*var));
...@@ -217,7 +217,7 @@ int ff_lpc_calc_coefs(LPCContext *s, ...@@ -217,7 +217,7 @@ int ff_lpc_calc_coefs(LPCContext *s,
m[0].coeff[max_order-1][j] = -lpc[max_order-1][j]; m[0].coeff[max_order-1][j] = -lpc[max_order-1][j];
for(; pass<lpc_passes; pass++){ for(; pass<lpc_passes; pass++){
avpriv_init_lls2(&m[pass&1], max_order); avpriv_init_lls(&m[pass&1], max_order);
weight=0; weight=0;
for(i=max_order; i<blocksize; i++){ for(i=max_order; i<blocksize; i++){
...@@ -238,7 +238,7 @@ int ff_lpc_calc_coefs(LPCContext *s, ...@@ -238,7 +238,7 @@ int ff_lpc_calc_coefs(LPCContext *s,
m[pass&1].update_lls(&m[pass&1], var); m[pass&1].update_lls(&m[pass&1], var);
} }
avpriv_solve_lls2(&m[pass&1], 0.001, 0); avpriv_solve_lls(&m[pass&1], 0.001, 0);
} }
for(i=0; i<max_order; i++){ for(i=0; i<max_order; i++){
......
libavutil/Makefile
View file @ 70b8668f
...@@ -102,8 +102,7 @@ OBJS = adler32.o \ ...@@ -102,8 +102,7 @@ OBJS = adler32.o \
intfloat_readwrite.o \ intfloat_readwrite.o \
intmath.o \ intmath.o \
lfg.o \ lfg.o \
lls1.o \ lls.o \
lls2.o \
log.o \ log.o \
log2_tab.o \ log2_tab.o \
mathematics.o \ mathematics.o \
...@@ -163,8 +162,7 @@ TESTPROGS = adler32 \ ...@@ -163,8 +162,7 @@ TESTPROGS = adler32 \
float_dsp \ float_dsp \
hmac \ hmac \
lfg \ lfg \
lls1 \ lls \
lls2 \
log \ log \
md5 \ md5 \
murmur3 \ murmur3 \
......
libavutil/lls2.c → libavutil/lls.c
View file @ 70b8668f
...@@ -30,9 +30,9 @@ ...@@ -30,9 +30,9 @@
#include "attributes.h" #include "attributes.h"
#include "version.h" #include "version.h"
#include "lls2.h" #include "lls.h"
static void update_lls(LLSModel2 *m, double *var) static void update_lls(LLSModel *m, double *var)
{ {
int i, j; int i, j;
...@@ -43,7 +43,7 @@ static void update_lls(LLSModel2 *m, double *var) ...@@ -43,7 +43,7 @@ static void update_lls(LLSModel2 *m, double *var)
} }
} }
void avpriv_solve_lls2(LLSModel2 *m, double threshold, unsigned short min_order) void avpriv_solve_lls(LLSModel *m, double threshold, unsigned short min_order)
{ {
int i, j, k; int i, j, k;
double (*factor)[MAX_VARS_ALIGN] = (void *) &m->covariance[1][0]; double (*factor)[MAX_VARS_ALIGN] = (void *) &m->covariance[1][0];
...@@ -100,7 +100,7 @@ void avpriv_solve_lls2(LLSModel2 *m, double threshold, unsigned short min_order) ...@@ -100,7 +100,7 @@ void avpriv_solve_lls2(LLSModel2 *m, double threshold, unsigned short min_order)
} }
} }
static double evaluate_lls(LLSModel2 *m, double *param, int order) static double evaluate_lls(LLSModel *m, double *param, int order)
{ {
int i; int i;
double out = 0; double out = 0;
...@@ -111,9 +111,9 @@ static double evaluate_lls(LLSModel2 *m, double *param, int order) ...@@ -111,9 +111,9 @@ static double evaluate_lls(LLSModel2 *m, double *param, int order)
return out; return out;
} }
av_cold void avpriv_init_lls2(LLSModel2 *m, int indep_count) av_cold void avpriv_init_lls(LLSModel *m, int indep_count)
{ {
memset(m, 0, sizeof(LLSModel2)); memset(m, 0, sizeof(LLSModel));
m->indep_count = indep_count; m->indep_count = indep_count;
m->update_lls = update_lls; m->update_lls = update_lls;
m->evaluate_lls = evaluate_lls; m->evaluate_lls = evaluate_lls;
...@@ -129,12 +129,12 @@ av_cold void avpriv_init_lls2(LLSModel2 *m, int indep_count) ...@@ -129,12 +129,12 @@ av_cold void avpriv_init_lls2(LLSModel2 *m, int indep_count)
int main(void) int main(void)
{ {
LLSModel2 m; LLSModel m;
int i, order; int i, order;
AVLFG lfg; AVLFG lfg;
av_lfg_init(&lfg, 1); av_lfg_init(&lfg, 1);
avpriv_init_lls2(&m, 3); avpriv_init_lls(&m, 3);
for (i = 0; i < 100; i++) { for (i = 0; i < 100; i++) {
LOCAL_ALIGNED(32, double, var, [4]); LOCAL_ALIGNED(32, double, var, [4]);
...@@ -145,7 +145,7 @@ int main(void) ...@@ -145,7 +145,7 @@ int main(void)
var[2] = var[1] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5; var[2] = var[1] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5;
var[3] = var[2] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5; var[3] = var[2] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5;
m.update_lls(&m, var); m.update_lls(&m, var);
avpriv_solve_lls2(&m, 0.001, 0); avpriv_solve_lls(&m, 0.001, 0);
for (order = 0; order < 3; order++) { for (order = 0; order < 3; order++) {
eval = m.evaluate_lls(&m, var + 1, order); eval = m.evaluate_lls(&m, var + 1, order);
printf("real:%9f order:%d pred:%9f var:%f coeffs:%f %9f %9f\n", printf("real:%9f order:%d pred:%9f var:%f coeffs:%f %9f %9f\n",
......
libavutil/lls2.h → libavutil/lls.h
View file @ 70b8668f
...@@ -30,12 +30,12 @@ ...@@ -30,12 +30,12 @@
#define MAX_VARS 32 #define MAX_VARS 32
#define MAX_VARS_ALIGN FFALIGN(MAX_VARS+1,4) #define MAX_VARS_ALIGN FFALIGN(MAX_VARS+1,4)
//FIXME avoid direct access to LLSModel2 from outside //FIXME avoid direct access to LLSModel from outside
/** /**
* Linear least squares model. * Linear least squares model.
*/ */
typedef struct LLSModel2 { typedef struct LLSModel {
DECLARE_ALIGNED(32, double, covariance[MAX_VARS_ALIGN][MAX_VARS_ALIGN]); DECLARE_ALIGNED(32, double, covariance[MAX_VARS_ALIGN][MAX_VARS_ALIGN]);
DECLARE_ALIGNED(32, double, coeff[MAX_VARS][MAX_VARS]); DECLARE_ALIGNED(32, double, coeff[MAX_VARS][MAX_VARS]);
double variance[MAX_VARS]; double variance[MAX_VARS];
...@@ -47,18 +47,18 @@ typedef struct LLSModel2 { ...@@ -47,18 +47,18 @@ typedef struct LLSModel2 {
* 32-byte aligned, and any padding elements must be initialized * 32-byte aligned, and any padding elements must be initialized
* (i.e not denormal/nan). * (i.e not denormal/nan).
*/ */
void (*update_lls)(struct LLSModel2 *m, double *var); void (*update_lls)(struct LLSModel *m, double *var);
/** /**
* Inner product of var[] and the LPC coefs. * Inner product of var[] and the LPC coefs.
* @param m this context * @param m this context
* @param var training samples, excluding the value to be predicted. unaligned. * @param var training samples, excluding the value to be predicted. unaligned.
* @param order lpc order * @param order lpc order
*/ */
double (*evaluate_lls)(struct LLSModel2 *m, double *var, int order); double (*evaluate_lls)(struct LLSModel *m, double *var, int order);
} LLSModel2; } LLSModel;
void avpriv_init_lls2(LLSModel2 *m, int indep_count); void avpriv_init_lls(LLSModel *m, int indep_count);
void ff_init_lls_x86(LLSModel2 *m); void ff_init_lls_x86(LLSModel *m);
void avpriv_solve_lls2(LLSModel2 *m, double threshold, unsigned short min_order); void avpriv_solve_lls(LLSModel *m, double threshold, unsigned short min_order);
#endif /* AVUTIL_LLS_H */ #endif /* AVUTIL_LLS_H */
libavutil/lls1.c deleted 100644 → 0
View file @ 5e1832c2
/*
* linear least squares model
*
* Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
*
* 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 2.1 of the License, or (at your option) any later version.
*
* 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 FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* linear least squares model
*/
#include <math.h>
#include <string.h>
#include "attributes.h"
#include "version.h"
#include "lls1.h"
#if FF_API_LLS1
av_cold void avpriv_init_lls(LLSModel *m, int indep_count)
{
memset(m, 0, sizeof(LLSModel));
m->indep_count = indep_count;
}
void avpriv_update_lls(LLSModel *m, double *var, double decay)
{
int i, j;
for (i = 0; i <= m->indep_count; i++) {
for (j = i; j <= m->indep_count; j++) {
m->covariance[i][j] *= decay;
m->covariance[i][j] += var[i] * var[j];
}
}
}
void avpriv_solve_lls(LLSModel *m, double threshold, unsigned short min_order)
{
int i, j, k;
double (*factor)[MAX_VARS + 1] = (void *) &m->covariance[1][0];
double (*covar) [MAX_VARS + 1] = (void *) &m->covariance[1][1];
double *covar_y = m->covariance[0];
int count = m->indep_count;
for (i = 0; i < count; i++) {
for (j = i; j < count; j++) {
double sum = covar[i][j];
for (k = i - 1; k >= 0; k--)
sum -= factor[i][k] * factor[j][k];
if (i == j) {
if (sum < threshold)
sum = 1.0;
factor[i][i] = sqrt(sum);
} else {
factor[j][i] = sum / factor[i][i];
}
}
}
for (i = 0; i < count; i++) {
double sum = covar_y[i + 1];
for (k = i - 1; k >= 0; k--)
sum -= factor[i][k] * m->coeff[0][k];
m->coeff[0][i] = sum / factor[i][i];
}
for (j = count - 1; j >= min_order; j--) {
for (i = j; i >= 0; i--) {
double sum = m->coeff[0][i];
for (k = i + 1; k <= j; k++)
sum -= factor[k][i] * m->coeff[j][k];
m->coeff[j][i] = sum / factor[i][i];
}
m->variance[j] = covar_y[0];
for (i = 0; i <= j; i++) {
double sum = m->coeff[j][i] * covar[i][i] - 2 * covar_y[i + 1];
for (k = 0; k < i; k++)
sum += 2 * m->coeff[j][k] * covar[k][i];
m->variance[j] += m->coeff[j][i] * sum;
}
}
}
double avpriv_evaluate_lls(LLSModel *m, double *param, int order)
{
int i;
double out = 0;
for (i = 0; i <= order; i++)
out += param[i] * m->coeff[order][i];
return out;
}
#if FF_API_LLS_PRIVATE
av_cold void av_init_lls(LLSModel *m, int indep_count)
{
avpriv_init_lls(m, indep_count);
}
void av_update_lls(LLSModel *m, double *param, double decay)
{
avpriv_update_lls(m, param, decay);
}
void av_solve_lls(LLSModel *m, double threshold, int min_order)
{
avpriv_solve_lls(m, threshold, min_order);
}
double av_evaluate_lls(LLSModel *m, double *param, int order)
{
return avpriv_evaluate_lls(m, param, order);
}
#endif /* FF_API_LLS_PRIVATE */
#endif /* FF_API_LLS1 */
#ifdef TEST
#include <stdio.h>
#include <limits.h>
#include "lfg.h"
int main(void)
{
LLSModel m;
int i, order;
AVLFG lfg;
av_lfg_init(&lfg, 1);
avpriv_init_lls(&m, 3);
for (i = 0; i < 100; i++) {
double var[4];
double eval;
var[0] = (av_lfg_get(&lfg) / (double) UINT_MAX - 0.5) * 2;
var[1] = var[0] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5;
var[2] = var[1] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5;
var[3] = var[2] + av_lfg_get(&lfg) / (double) UINT_MAX - 0.5;
avpriv_update_lls(&m, var, 0.99);
avpriv_solve_lls(&m, 0.001, 0);
for (order = 0; order < 3; order++) {
eval = avpriv_evaluate_lls(&m, var + 1, order);
printf("real:%9f order:%d pred:%9f var:%f coeffs:%f %9f %9f\n",
var[0], order, eval, sqrt(m.variance[order] / (i + 1)),
m.coeff[order][0], m.coeff[order][1],
m.coeff[order][2]);
}
}
return 0;
}
#endif
libavutil/lls1.h deleted 100644 → 0
View file @ 5e1832c2
/*
* linear least squares model
*
* Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
*
* 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 2.1 of the License, or (at your option) any later version.
*
* 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 FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVUTIL_LLS_H
#define AVUTIL_LLS_H
#include "version.h"
#define MAX_VARS 32
//FIXME avoid direct access to LLSModel from outside
/**
* Linear least squares model.
*/
typedef struct LLSModel {
double covariance[MAX_VARS + 1][MAX_VARS + 1];
double coeff[MAX_VARS][MAX_VARS];
double variance[MAX_VARS];
int indep_count;
} LLSModel;
void avpriv_init_lls(LLSModel *m, int indep_count);
void avpriv_update_lls(LLSModel *m, double *param, double decay);
void avpriv_solve_lls(LLSModel *m, double threshold, unsigned short min_order);
double avpriv_evaluate_lls(LLSModel *m, double *param, int order);
#if FF_API_LLS_PRIVATE
void av_init_lls(LLSModel *m, int indep_count);
void av_update_lls(LLSModel *m, double *param, double decay);
void av_solve_lls(LLSModel *m, double threshold, int min_order);
double av_evaluate_lls(LLSModel *m, double *param, int order);
#endif /* FF_API_LLS_PRIVATE */
#endif /* AVUTIL_LLS_H */
libavutil/x86/lls.asm
View file @ 70b8668f
...@@ -29,7 +29,7 @@ SECTION .text ...@@ -29,7 +29,7 @@ SECTION .text
%define COVAR_STRIDE MAX_VARS_ALIGN*8 %define COVAR_STRIDE MAX_VARS_ALIGN*8
%define COVAR(x,y) [covarq + (x)*8 + (y)*COVAR_STRIDE] %define COVAR(x,y) [covarq + (x)*8 + (y)*COVAR_STRIDE]
struc LLSModel2 struc LLSModel
.covariance: resq MAX_VARS_ALIGN*MAX_VARS_ALIGN .covariance: resq MAX_VARS_ALIGN*MAX_VARS_ALIGN
.coeff: resq MAX_VARS*MAX_VARS .coeff: resq MAX_VARS*MAX_VARS
.variance: resq MAX_VARS .variance: resq MAX_VARS
...@@ -49,7 +49,7 @@ INIT_XMM sse2 ...@@ -49,7 +49,7 @@ INIT_XMM sse2
%define movdqa movaps %define movdqa movaps
cglobal update_lls, 2,5,8, ctx, var, i, j, covar2 cglobal update_lls, 2,5,8, ctx, var, i, j, covar2
%define covarq ctxq %define covarq ctxq
mov id, [ctxq + LLSModel2.indep_count] mov id, [ctxq + LLSModel.indep_count]
lea varq, [varq + iq*8] lea varq, [varq + iq*8]
neg iq neg iq
mov covar2q, covarq mov covar2q, covarq
...@@ -129,7 +129,7 @@ cglobal update_lls, 2,5,8, ctx, var, i, j, covar2 ...@@ -129,7 +129,7 @@ cglobal update_lls, 2,5,8, ctx, var, i, j, covar2
INIT_YMM avx INIT_YMM avx
cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2
%define covarq ctxq %define covarq ctxq
mov countd, [ctxq + LLSModel2.indep_count] mov countd, [ctxq + LLSModel.indep_count]
lea count2d, [countq-2] lea count2d, [countq-2]
xor id, id xor id, id
.loopi: .loopi:
...@@ -206,7 +206,7 @@ cglobal evaluate_lls, 3,4,2, ctx, var, order, i ...@@ -206,7 +206,7 @@ cglobal evaluate_lls, 3,4,2, ctx, var, order, i
%define coefsq ctxq %define coefsq ctxq
mov id, orderd mov id, orderd
imul orderd, MAX_VARS imul orderd, MAX_VARS
lea coefsq, [ctxq + LLSModel2.coeff + orderq*8] lea coefsq, [ctxq + LLSModel.coeff + orderq*8]
movsd m0, [varq] movsd m0, [varq]
movhpd m0, [varq + 8] movhpd m0, [varq + 8]
mulpd m0, [coefsq] mulpd m0, [coefsq]
......
libavutil/x86/lls_init.c
View file @ 70b8668f
...@@ -20,14 +20,14 @@ ...@@ -20,14 +20,14 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/ */
#include "libavutil/lls2.h" #include "libavutil/lls.h"
#include "libavutil/x86/cpu.h" #include "libavutil/x86/cpu.h"
void ff_update_lls_sse2(LLSModel2 *m, double *var); void ff_update_lls_sse2(LLSModel *m, double *var);
void ff_update_lls_avx(LLSModel2 *m, double *var); void ff_update_lls_avx(LLSModel *m, double *var);
double ff_evaluate_lls_sse2(LLSModel2 *m, double *var, int order); double ff_evaluate_lls_sse2(LLSModel *m, double *var, int order);
av_cold void ff_init_lls_x86(LLSModel2 *m) av_cold void ff_init_lls_x86(LLSModel *m)
{ {
int cpu_flags = av_get_cpu_flags(); int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_SSE2(cpu_flags)) { if (EXTERNAL_SSE2(cpu_flags)) {
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment