Update of /cvsroot/pure-data/externals/iem/iemmatrix/src In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv14656
Added Files: mtx_cholesky.c Log Message: cholesky decomposition
--- NEW FILE: mtx_cholesky.c --- /* * iemmatrix * * objects for manipulating simple matrices * mostly refering to matlab/octave matrix functions * * Copyright (c) IOhannes m zmölnig, forum::für::umläute * IEM, Graz, Austria * * For information on usage and redistribution, and for a DISCLAIMER OF ALL * WARRANTIES, see the file, "LICENSE.txt," in this distribution. * */ #include "iemmatrix.h"
/* mtx_cholesky */
/* * calculate the "Cholesky Decomposition" of a "symmetric and positive definite matrix " * no check is done, whether the input matrix is really symmetric and positive definite. */
static t_class *mtx_cholesky_class;
static void mtx_cholesky_matrix(t_matrix *x, t_symbol *s, int argc, t_atom *argv) { /* maybe we should do this in double or long double ? */ int row=atom_getfloat(argv); int col=atom_getfloat(argv+1); int i, j, k, row2=row*row;
t_matrixfloat *original, *cholesky; t_matrixfloat *a1, *b1; // dummy pointers
int ok = 0;
if(row*col+2>argc){ post("mtx_print : sparse matrices not yet supported : use "mtx_check""); return; } if (row!=col){ post("mtx_cholesky: only symmetric and positive definite matrices can be cholesky-decomposed"); return; }
// reserve memory for outputting afterwards adjustsize(x, row, row); // 1. get the 2 matrices : orig; invert (create as eye, but will be orig^(-1)) cholesky = (t_matrixfloat *)getbytes(sizeof(t_matrixfloat)*row2); // 1a extract values of A to float-buf original=matrix2float(argv);
// 2 set the cholesky matrix to zero for(i=0; i<row2; i++)cholesky[i]=0.;
// 3 do the cholesky decomposition for(i=0; i<col; i++){ // 3a get the diagonal element // l_ii=sqrt(a_ii-sum(k=1..i-1)((l_ik)^2)) t_matrixfloat sum=0.; t_matrixfloat result=0.f;
for(k=0; k<i; k++){ t_matrixfloat lik=cholesky[k*col+i]; sum+=lik*lik; } if((result=original[i*(col+1)]-sum)<0){ post("[mtx_cholesky]: only symmetric and positive definite matrices can be cholesky-decomposed"); return; } result=sqrtf(result); // LATER check whether this is real cholesky[i*(col+1)]=result; // 3b get the other elements within this row/col // l_ji=(a_ji-sum(k=1..i-1)(l_jk*l_ik))/l_ii for(j=i+1; j<row; j++){ sum=0.; for(k=0; k<i; k++){ t_matrixfloat ljk=cholesky[k*col+j]; t_matrixfloat lik=cholesky[k*col+i];
sum+=ljk*lik; } cholesky[i*row+j]=(original[i*col+j]-sum)/result; } }
// 4. output the matrix // 4a convert the floatbuf to an atombuf; float2matrix(x->atombuffer, cholesky); // 4b destroy the buffers freebytes(original, sizeof(t_matrixfloat)*row2);
// 4c output the atombuf; matrix_bang(x); }
static void *mtx_cholesky_new(t_symbol *s, int argc, t_atom *argv) { t_matrix *x = (t_matrix *)pd_new(mtx_cholesky_class); outlet_new(&x->x_obj, 0); x->col=x->row=0; x->atombuffer=0;
return (x); } void mtx_cholesky_setup(void) { mtx_cholesky_class = class_new(gensym("mtx_cholesky"), (t_newmethod)mtx_cholesky_new, (t_method)matrix_free, sizeof(t_matrix), 0, A_GIMME, 0); class_addbang (mtx_cholesky_class, matrix_bang); class_addmethod(mtx_cholesky_class, (t_method)mtx_cholesky_matrix, gensym("matrix"), A_GIMME, 0); class_sethelpsymbol(mtx_cholesky_class, gensym("iemmatrix/mtx_cholesky")); }
void iemtx_cholesky_setup(void){ mtx_cholesky_setup(); }