annotate lls.c @ 999:effd4ae1769d libavutil

Make ff_inverse stay with libavutil, and optional copy it to libavcodec. The ff_inverse table is used by FASTDIV macro, defined in libavutil, but up to now the table was defined only in libavcodec. After this change, the main copy of ff_inverse is part of libavutil (just like FASTDIV), but if CONFIG_SMALL is unset, then a different copy is made available to libavcodec, to avoid the performance penalty of using an external look up table. Dynamic linking works, because the libraries are linked with -Bsymbolic, so the local copy of the symbol has priority over the external; static linking works because the table is on a standalone object file in both libraries, so the linker is able to discard one of the two. Tested on Linux/x86-64 and Mac OS X/x86-64.
author flameeyes
date Wed, 21 Jul 2010 12:37:37 +0000
parents 0795a743bda1
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
1 /*
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
2 * linear least squares model
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
3 *
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
4 * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
5 *
116
d76a36742464 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 90
diff changeset
6 * This file is part of FFmpeg.
d76a36742464 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 90
diff changeset
7 *
d76a36742464 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 90
diff changeset
8 * FFmpeg is free software; you can redistribute it and/or
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
9 * modify it under the terms of the GNU Lesser General Public
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
10 * License as published by the Free Software Foundation; either
116
d76a36742464 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 90
diff changeset
11 * version 2.1 of the License, or (at your option) any later version.
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
12 *
116
d76a36742464 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 90
diff changeset
13 * FFmpeg is distributed in the hope that it will be useful,
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
16 * Lesser General Public License for more details.
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
17 *
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
18 * You should have received a copy of the GNU Lesser General Public
116
d76a36742464 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 90
diff changeset
19 * License along with FFmpeg; if not, write to the Free Software
358
f13e5473611e license header consistency cosmetics
diego
parents: 116
diff changeset
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
21 */
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
22
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
23 /**
899
0795a743bda1 Remove explicit filename from Doxygen @file commands.
diego
parents: 763
diff changeset
24 * @file
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
25 * linear least squares model
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
26 */
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
27
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
28 #include <math.h>
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
29 #include <string.h>
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
30
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
31 #include "lls.h"
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
32
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
33 void av_init_lls(LLSModel *m, int indep_count){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
34 memset(m, 0, sizeof(LLSModel));
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
35
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
36 m->indep_count= indep_count;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
37 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
38
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
39 void av_update_lls(LLSModel *m, double *var, double decay){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
40 int i,j;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
41
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
42 for(i=0; i<=m->indep_count; i++){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
43 for(j=i; j<=m->indep_count; j++){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
44 m->covariance[i][j] *= decay;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
45 m->covariance[i][j] += var[i]*var[j];
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
46 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
47 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
48 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
49
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
50 void av_solve_lls(LLSModel *m, double threshold, int min_order){
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
51 int i,j,k;
441
6335e6e63d91 Fix the following using void* casts, proper casts are less readable and
michael
parents: 424
diff changeset
52 double (*factor)[MAX_VARS+1]= (void*)&m->covariance[1][0];
6335e6e63d91 Fix the following using void* casts, proper casts are less readable and
michael
parents: 424
diff changeset
53 double (*covar )[MAX_VARS+1]= (void*)&m->covariance[1][1];
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
54 double *covar_y = m->covariance[0];
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
55 int count= m->indep_count;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
56
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
57 for(i=0; i<count; i++){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
58 for(j=i; j<count; j++){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
59 double sum= covar[i][j];
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
60
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
61 for(k=i-1; k>=0; k--)
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
62 sum -= factor[i][k]*factor[j][k];
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
63
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
64 if(i==j){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
65 if(sum < threshold)
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
66 sum= 1.0;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
67 factor[i][i]= sqrt(sum);
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
68 }else
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
69 factor[j][i]= sum / factor[i][i];
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
70 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
71 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
72 for(i=0; i<count; i++){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
73 double sum= covar_y[i+1];
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
74 for(k=i-1; k>=0; k--)
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
75 sum -= factor[i][k]*m->coeff[0][k];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
76 m->coeff[0][i]= sum / factor[i][i];
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
77 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
78
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
79 for(j=count-1; j>=min_order; j--){
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
80 for(i=j; i>=0; i--){
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
81 double sum= m->coeff[0][i];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
82 for(k=i+1; k<=j; k++)
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
83 sum -= factor[k][i]*m->coeff[j][k];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
84 m->coeff[j][i]= sum / factor[i][i];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
85 }
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
86
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
87 m->variance[j]= covar_y[0];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
88 for(i=0; i<=j; i++){
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
89 double sum= m->coeff[j][i]*covar[i][i] - 2*covar_y[i+1];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
90 for(k=0; k<i; k++)
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
91 sum += 2*m->coeff[j][k]*covar[k][i];
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
92 m->variance[j] += m->coeff[j][i]*sum;
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
93 }
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
94 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
95 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
96
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
97 double av_evaluate_lls(LLSModel *m, double *param, int order){
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
98 int i;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
99 double out= 0;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
100
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
101 for(i=0; i<=order; i++)
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
102 out+= param[i]*m->coeff[order][i];
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
103
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
104 return out;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
105 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
106
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
107 #ifdef TEST
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
108
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
109 #include <stdlib.h>
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
110 #include <stdio.h>
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
111
404
f9a4c04ebb0e main() --> main(void)
diego
parents: 358
diff changeset
112 int main(void){
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
113 LLSModel m;
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
114 int i, order;
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
115
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
116 av_init_lls(&m, 3);
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
117
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
118 for(i=0; i<100; i++){
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
119 double var[4];
424
1cdbf12cb116 Remove unused variable variance.
diego
parents: 404
diff changeset
120 double eval;
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
121 var[0] = (rand() / (double)RAND_MAX - 0.5)*2;
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
122 var[1] = var[0] + rand() / (double)RAND_MAX - 0.5;
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
123 var[2] = var[1] + rand() / (double)RAND_MAX - 0.5;
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
124 var[3] = var[2] + rand() / (double)RAND_MAX - 0.5;
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
125 av_update_lls(&m, var, 0.99);
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
126 av_solve_lls(&m, 0.001, 0);
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
127 for(order=0; order<3; order++){
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
128 eval= av_evaluate_lls(&m, var+1, order);
700
e110508543ac Align test program output columns.
diego
parents: 642
diff changeset
129 printf("real:%9f order:%d pred:%9f var:%f coeffs:%f %9f %9f\n",
79
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
130 var[0], order, eval, sqrt(m.variance[order] / (i+1)),
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
131 m.coeff[order][0], m.coeff[order][1], m.coeff[order][2]);
adbb5540fa47 calculate all coefficients for several orders during cholesky factorization, the resulting coefficients are not strictly optimal though as there is a small difference in the autocorrelation matrixes which is ignored for the smaller orders
michael
parents: 78
diff changeset
132 }
76
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
133 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
134 return 0;
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
135 }
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
136
8c75234388b5 linear least squares solver using cholesky factorization
michael
parents:
diff changeset
137 #endif