stepan
/
dynare
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Block kalman filter DLL: various modernizations and simplifications

time-shift
Sébastien Villemot 2019-04-30 12:59:43 +02:00
parent eb975f2488
commit 80267f56e5
No known key found for this signature in database
GPG Key ID: 2CECE9350ECEBE4A
2 changed files with 163 additions and 201 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

338
mex/sources/block_kalman_filter/block_kalman_filter.cc
View File

@ -1,5 +1,5 @@
/* /*
* Copyright © 2007-2017 Dynare Team * Copyright © 2007-2019 Dynare Team
* *
* This file is part of Dynare. * This file is part of Dynare.
* *
@ -16,16 +16,14 @@
* You should have received a copy of the GNU General Public License * You should have received a copy of the GNU General Public License
* along with Dynare. If not, see <http://www.gnu.org/licenses/>. * along with Dynare. If not, see <http://www.gnu.org/licenses/>.
*/ */
#include <cstring>
#include <cmath> #include <cmath>
#include <iostream> #include <algorithm>
#include <fstream>
#include <vector>
#ifdef USE_OMP #ifdef USE_OMP
# include <omp.h> # include <omp.h>
#endif #endif
#include "block_kalman_filter.hh" #include "block_kalman_filter.hh"
using namespace std;
#define BLAS #define BLAS
//#define CUBLAS //#define CUBLAS
@ -33,17 +31,18 @@ using namespace std;
# include <cuda_runtime.h> # include <cuda_runtime.h>
# include <cublas_v2.h> # include <cublas_v2.h>
#endif #endif
void void
mexDisp(mxArray *P) mexDisp(const mxArray *P)
{ {
unsigned int n = mxGetN(P); size_t n = mxGetN(P);
unsigned int m = mxGetM(P); size_t m = mxGetM(P);
double *M = mxGetPr(P); const double *M = mxGetPr(P);
mexPrintf("%d x %d\n", m, n); mexPrintf("%d x %d\n", m, n);
mexEvalString("drawnow;"); mexEvalString("drawnow;");
for (unsigned int i = 0; i < m; i++) for (size_t i = 0; i < m; i++)
{ {
for (unsigned int j = 0; j < n; j++) for (size_t j = 0; j < n; j++)
mexPrintf(" %9.4f", M[i+ j * m]); mexPrintf(" %9.4f", M[i+ j * m]);
mexPrintf("\n"); mexPrintf("\n");
} }
@ -51,7 +50,7 @@ mexDisp(mxArray *P)
} }
void void
mexDisp(double *M, int m, int n) mexDisp(const double *M, int m, int n)
{ {
mexPrintf("%d x %d\n", m, n); mexPrintf("%d x %d\n", m, n);
mexEvalString("drawnow;"); mexEvalString("drawnow;");
@ -63,6 +62,7 @@ mexDisp(double *M, int m, int n)
} }
mexEvalString("drawnow;"); mexEvalString("drawnow;");
} }
/*if block /*if block
%nz_state_var = M_.nz_state_var; %nz_state_var = M_.nz_state_var;
while notsteady && t<smpl while notsteady && t<smpl
@ -118,13 +118,12 @@ mexDisp(double *M, int m, int n)
*/ */
bool bool
not_all_abs_F_bellow_crit(double *F, int size, double crit) not_all_abs_F_bellow_crit(const double *F, int size, double crit)
{ {
int i = 0; int i = 0;
while (i < size && abs(F[i]) < crit) while (i < size && abs(F[i]) < crit)
{ i++;
i++;
}
if (i < size) if (i < size)
return false; return false;
else else
@ -132,7 +131,7 @@ not_all_abs_F_bellow_crit(double *F, int size, double crit)
} }
double double
det(double *F, int dim, lapack_int *ipiv) det(const double *F, int dim, const lapack_int *ipiv)
{ {
double det = 1.0; double det = 1.0;
for (int i = 0; i < dim; i++) for (int i = 0; i < dim; i++)
@ -143,7 +142,7 @@ det(double *F, int dim, lapack_int *ipiv)
return det; return det;
} }
BlockKalmanFilter::BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[], double *P_mf[], double *v_pp[], double *K[], double *v_n[], double *a[], double *K_P[], double *P_t_t1[], double *tmp[], double *P[]) BlockKalmanFilter::BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{ {
if (nlhs > 3) if (nlhs > 3)
DYN_MEX_FUNC_ERR_MSG_TXT("block_kalman_filter provides at most 3 output argument."); DYN_MEX_FUNC_ERR_MSG_TXT("block_kalman_filter provides at most 3 output argument.");
@ -203,7 +202,7 @@ BlockKalmanFilter::BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const
R = mxGetPr(pR); R = mxGetPr(pR);
Q = mxGetPr(pQ); Q = mxGetPr(pQ);
H = mxGetPr(pH); H = mxGetPr(pH);
*P = mxGetPr(pP); P = mxGetPr(pP);
Y = mxGetPr(pY); Y = mxGetPr(pY);
n = mxGetN(pT); // Number of state variables. n = mxGetN(pT); // Number of state variables.
@ -222,13 +221,13 @@ BlockKalmanFilter::BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const
if (mxGetNumberOfElements(pdata_index) != static_cast<unsigned int>(smpl)) if (mxGetNumberOfElements(pdata_index) != static_cast<unsigned int>(smpl))
DYN_MEX_FUNC_ERR_MSG_TXT("the number of element in the cell array passed to block_missing_observation_kalman_filter as first argument has to be equal to the smpl size"); DYN_MEX_FUNC_ERR_MSG_TXT("the number of element in the cell array passed to block_missing_observation_kalman_filter as first argument has to be equal to the smpl size");
i_nz_state_var = static_cast<int *>(mxMalloc(n*sizeof(int))); i_nz_state_var = std::make_unique<int[]>(n);
for (int i = 0; i < n; i++) for (int i = 0; i < n; i++)
i_nz_state_var[i] = nz_state_var[i]; i_nz_state_var[i] = nz_state_var[i];
pa = mxCreateDoubleMatrix(n, 1, mxREAL); // State vector. pa = mxCreateDoubleMatrix(n, 1, mxREAL); // State vector.
*a = mxGetPr(pa); a = mxGetPr(pa);
tmp_a = static_cast<double *>(mxMalloc(n * sizeof(double))); tmp_a = std::make_unique<double[]>(n);
dF = 0.0; // det(F). dF = 0.0; // det(F).
p_tmp1 = mxCreateDoubleMatrix(n, n_shocks, mxREAL); p_tmp1 = mxCreateDoubleMatrix(n, n_shocks, mxREAL);
@ -240,12 +239,11 @@ BlockKalmanFilter::BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const
LIK = 0.0; // Default value of the log likelihood. LIK = 0.0; // Default value of the log likelihood.
notsteady = true; // Steady state flag. notsteady = true; // Steady state flag.
F_singular = true; F_singular = true;
*v_pp = static_cast<double *>(mxMalloc(pp * sizeof(double))); v_pp = std::make_unique<double[]>(pp);
*v_n = static_cast<double *>(mxMalloc(n * sizeof(double))); v_n = std::make_unique<double[]>(n);
mf = static_cast<int *>(mxMalloc(pp * sizeof(int))); mf = std::make_unique<int[]>(pp);
for (int i = 0; i < pp; i++) for (int i = 0; i < pp; i++)
mf[i] = mfd[i] - 1; mf[i] = mfd[i] - 1;
pi = atan2(0.0, -1.0);
/*compute QQ = R*Q*transpose(R)*/ // Variance of R times the vector of structural innovations.; /*compute QQ = R*Q*transpose(R)*/ // Variance of R times the vector of structural innovations.;
// tmp = R * Q; // tmp = R * Q;
@ -278,109 +276,104 @@ BlockKalmanFilter::BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const
piF = mxCreateDoubleMatrix(pp, pp, mxREAL); piF = mxCreateDoubleMatrix(pp, pp, mxREAL);
iF = mxGetPr(piF); iF = mxGetPr(piF);
lw = pp * 4; lw = pp * 4;
w = static_cast<double *>(mxMalloc(lw * sizeof(double))); w = std::make_unique<double[]>(lw);
iw = static_cast<lapack_int *>(mxMalloc(pp * sizeof(lapack_int))); iw = std::make_unique<lapack_int[]>(pp);
ipiv = static_cast<lapack_int *>(mxMalloc(pp * sizeof(lapack_int))); ipiv = std::make_unique<lapack_int[]>(pp);
info = 0; info = 0;
#if defined(BLAS) || defined(CUBLAS) #if defined(BLAS) || defined(CUBLAS)
p_tmp = mxCreateDoubleMatrix(n, n, mxREAL); p_tmp = mxCreateDoubleMatrix(n, n, mxREAL);
*tmp = mxGetPr(p_tmp); tmp = mxGetPr(p_tmp);
p_P_t_t1 = mxCreateDoubleMatrix(n, n, mxREAL); p_P_t_t1 = mxCreateDoubleMatrix(n, n, mxREAL);
*P_t_t1 = mxGetPr(p_P_t_t1); P_t_t1 = mxGetPr(p_P_t_t1);
pK = mxCreateDoubleMatrix(n, n, mxREAL); pK = mxCreateDoubleMatrix(n, n, mxREAL);
*K = mxGetPr(pK); K = mxGetPr(pK);
p_K_P = mxCreateDoubleMatrix(n, n, mxREAL); p_K_P = mxCreateDoubleMatrix(n, n, mxREAL);
*K_P = mxGetPr(p_K_P); K_P = mxGetPr(p_K_P);
oldK = static_cast<double *>(mxMalloc(n * n * sizeof(double))); oldK = std::make_unique<double[]>(n * n);
*P_mf = static_cast<double *>(mxMalloc(n * n * sizeof(double))); P_mf = std::make_unique<double[]>(n * n);
for (int i = 0; i < n * n; i++) for (int i = 0; i < n * n; i++)
oldK[i] = Inf; oldK[i] = Inf;
#else #else
p_tmp = mxCreateDoubleMatrix(n, n_state, mxREAL); p_tmp = mxCreateDoubleMatrix(n, n_state, mxREAL);
*tmp = mxGetPr(p_tmp); tmp = mxGetPr(p_tmp);
p_P_t_t1 = mxCreateDoubleMatrix(n_state, n_state, mxREAL); p_P_t_t1 = mxCreateDoubleMatrix(n_state, n_state, mxREAL);
*P_t_t1 = mxGetPr(p_P_t_t1); P_t_t1 = mxGetPr(p_P_t_t1);
pK = mxCreateDoubleMatrix(n, pp, mxREAL); pK = mxCreateDoubleMatrix(n, pp, mxREAL);
*K = mxGetPr(pK); K = mxGetPr(pK);
p_K_P = mxCreateDoubleMatrix(n_state, n_state, mxREAL); p_K_P = mxCreateDoubleMatrix(n_state, n_state, mxREAL);
*K_P = mxGetPr(p_K_P); K_P = mxGetPr(p_K_P);
oldK = static_cast<double *>(mxMalloc(n * pp * sizeof(double))); oldK = std::make_unique<double[]>(n * pp);
*P_mf = static_cast<double *>(mxMalloc(n * pp * sizeof(double))); P_mf = std::make_unique<double[]>(n * pp);
for (int i = 0; i < n * pp; i++) for (int i = 0; i < n * pp; i++)
oldK[i] = Inf; oldK[i] = Inf;
#endif #endif
} }
void void
BlockKalmanFilter::block_kalman_filter_ss(double *P_mf, double *v_pp, double *K, double *v_n, double *a, double *K_P, double *P_t_t1, double *tmp, double *P) BlockKalmanFilter::block_kalman_filter_ss()
{ {
if (t+1 < smpl) if (t+1 < smpl)
{ while (t < smpl)
while (t < smpl) {
{ //v = Y(:,t)-a(mf);
//v = Y(:,t)-a(mf); for (int i = 0; i < pp; i++)
for (int i = 0; i < pp; i++) v[i] = Y[i + t * pp] - a[mf[i]];
v[i] = Y[i + t * pp] - a[mf[i]];
//a = T*(a+K*v); //a = T*(a+K*v);
for (int i = pure_obs; i < n; i++) for (int i = pure_obs; i < n; i++)
{ {
double res = 0.0; double res = 0.0;
for (int j = 0; j < pp; j++) for (int j = 0; j < pp; j++)
res += K[j * n + i] * v[j]; res += K[j * n + i] * v[j];
v_n[i] = res + a[i]; v_n[i] = res + a[i];
} }
for (int i = 0; i < n; i++) for (int i = 0; i < n; i++)
{ {
double res = 0.0; double res = 0.0;
for (int j = pure_obs; j < n; j++) for (int j = pure_obs; j < n; j++)
res += T[j * n + i] * v_n[j]; res += T[j * n + i] * v_n[j];
a[i] = res; a[i] = res;
} }
//lik(t) = transpose(v)*iF*v; //lik(t) = transpose(v)*iF*v;
for (int i = 0; i < pp; i++) for (int i = 0; i < pp; i++)
{ {
double res = 0.0; double res = 0.0;
for (int j = 0; j < pp; j++) for (int j = 0; j < pp; j++)
res += v[j] * iF[j * pp + i]; res += v[j] * iF[j * pp + i];
v_pp[i] = res; v_pp[i] = res;
} }
double res = 0.0; double res = 0.0;
for (int i = 0; i < pp; i++) for (int i = 0; i < pp; i++)
res += v_pp[i] * v[i]; res += v_pp[i] * v[i];
lik[t] = (log(dF) + res + pp * log(2.0*pi))/2; lik[t] = (log(dF) + res + pp * log(2.0*M_PI))/2;
if (t + 1 > start) if (t + 1 > start)
LIK += lik[t]; LIK += lik[t];
t++; t++;
} }
}
} }
bool bool
BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf, double *v_pp, double *K, double *v_n, double *a, double *K_P, double *P_t_t1, double *tmp, double *P) BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[])
{ {
while (notsteady && t < smpl) while (notsteady && t < smpl)
{ {
if (missing_observations) if (missing_observations)
{ {
// retrieve the d_index // retrieve the d_index
pd_index = mxGetCell(pdata_index, t); pd_index = mxGetCell(pdata_index, t);
dd_index = static_cast<double *>(mxGetData(pd_index)); dd_index = static_cast<double *>(mxGetData(pd_index));
size_d_index = mxGetM(pd_index); size_d_index = mxGetM(pd_index);
d_index.resize(size_d_index); d_index.resize(size_d_index);
for (int i = 0; i < size_d_index; i++) for (int i = 0; i < size_d_index; i++)
{ d_index[i] = ceil(dd_index[i]) - 1;
d_index[i] = ceil(dd_index[i]) - 1;
}
//v = Y(:,t) - a(mf) //v = Y(:,t) - a(mf)
int i_i = 0; int i_i = 0;
//#pragma omp parallel for shared(v, i_i, d_index) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) //#pragma omp parallel for shared(v, i_i, d_index) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
for (vector<int>::const_iterator i = d_index.begin(); i != d_index.end(); i++) for (auto i = d_index.begin(); i != d_index.end(); i++)
{ {
//mexPrintf("i_i=%d, omp_get_max_threads()=%d\n",i_i,omp_get_max_threads()); //mexPrintf("i_i=%d, omp_get_max_threads()=%d\n",i_i,omp_get_max_threads());
v[i_i] = Y[*i + t * pp] - a[mf[*i]]; v[i_i] = Y[*i + t * pp] - a[mf[*i]];
@ -390,26 +383,21 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
//F = P(mf,mf) + H; //F = P(mf,mf) + H;
i_i = 0; i_i = 0;
if (H_size == 1) if (H_size == 1)
{ //#pragma omp parallel for shared(iF, F, i_i) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
//#pragma omp parallel for shared(iF, F, i_i) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) for (auto i = d_index.begin(); i != d_index.end(); i++, i_i++)
for (vector<int>::const_iterator i = d_index.begin(); i != d_index.end(); i++, i_i++) {
{ int j_j = 0;
int j_j = 0; for (auto j = d_index.begin(); j != d_index.end(); j++, j_j++)
for (vector<int>::const_iterator j = d_index.begin(); j != d_index.end(); j++, j_j++) iF[i_i + j_j * size_d_index] = F[i_i + j_j * size_d_index] = P[mf[*i] + mf[*j] * n] + H[0];
iF[i_i + j_j * size_d_index] = F[i_i + j_j * size_d_index] = P[mf[*i] + mf[*j] * n] + H[0]; }
}
}
else else
{ //#pragma omp parallel for shared(iF, F, P, H, mf, i_i) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
//#pragma omp parallel for shared(iF, F, P, H, mf, i_i) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) for (auto i = d_index.begin(); i != d_index.end(); i++, i_i++)
for (vector<int>::const_iterator i = d_index.begin(); i != d_index.end(); i++, i_i++) {
{ int j_j = 0;
int j_j = 0; for (auto j = d_index.begin(); j != d_index.end(); j++, j_j++)
for (vector<int>::const_iterator j = d_index.begin(); j != d_index.end(); j++, j_j++) iF[i_i + j_j * size_d_index] = F[i_i + j_j * size_d_index] = P[mf[*i] + mf[*j] * n] + H[*i + *j * pp];
iF[i_i + j_j * size_d_index] = F[i_i + j_j * size_d_index] = P[mf[*i] + mf[*j] * n] + H[*i + *j * pp]; }
}
}
} }
else else
{ {
@ -421,30 +409,26 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
//F = P(mf,mf) + H; //F = P(mf,mf) + H;
if (H_size == 1) if (H_size == 1)
{ for (int i = 0; i < pp; i++)
for (int i = 0; i < pp; i++) for (int j = 0; j < pp; j++)
for (int j = 0; j < pp; j++) iF[i + j * pp] = F[i + j * pp] = P[mf[i] + mf[j] * n] + H[0];
iF[i + j * pp] = F[i + j * pp] = P[mf[i] + mf[j] * n] + H[0];
}
else else
{ for (int i = 0; i < pp; i++)
for (int i = 0; i < pp; i++) for (int j = 0; j < pp; j++)
for (int j = 0; j < pp; j++) iF[i + j * pp] = F[i + j * pp] = P[mf[i] + mf[j] * n] + H[i + j * pp];
iF[i + j * pp] = F[i + j * pp] = P[mf[i] + mf[j] * n] + H[i + j * pp];
}
} }
/* Computes the norm of iF */ /* Computes the norm of iF */
double anorm = dlange("1", &size_d_index, &size_d_index, iF, &size_d_index, w); double anorm = dlange("1", &size_d_index, &size_d_index, iF, &size_d_index, w.get());
//mexPrintf("anorm = %f\n",anorm); //mexPrintf("anorm = %f\n",anorm);
/* Modifies F in place with a LU decomposition */ /* Modifies F in place with a LU decomposition */
dgetrf(&size_d_index, &size_d_index, iF, &size_d_index, ipiv, &info); dgetrf(&size_d_index, &size_d_index, iF, &size_d_index, ipiv.get(), &info);
if (info != 0) if (info != 0)
mexPrintf("dgetrf failure with error %d\n", static_cast<int>(info)); mexPrintf("dgetrf failure with error %d\n", static_cast<int>(info));
/* Computes the reciprocal norm */ /* Computes the reciprocal norm */
dgecon("1", &size_d_index, iF, &size_d_index, &anorm, &rcond, w, iw, &info); dgecon("1", &size_d_index, iF, &size_d_index, &anorm, &rcond, w.get(), iw.get(), &info);
if (info != 0) if (info != 0)
mexPrintf("dgecon failure with error %d\n", static_cast<int>(info)); mexPrintf("dgecon failure with error %d\n", static_cast<int>(info));
@ -454,12 +438,10 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
mexPrintf("error: F singular\n"); mexPrintf("error: F singular\n");
LIK = Inf; LIK = Inf;
if (nlhs == 3) if (nlhs == 3)
{ for (int i = t; i < smpl; i++)
for (int i = t; i < smpl; i++) lik[i] = Inf;
lik[i] = Inf;
}
// info = 0 // info = 0
return_results_and_clean(nlhs, plhs, &P_mf, &v_pp, &K, &K_P, &a, &K_P, &P_t_t1, &tmp, &P); return_results_and_clean(nlhs, plhs);
return false; return false;
} }
else else
@ -474,10 +456,10 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
res += T[i + j *n] * a[j]; res += T[i + j *n] * a[j];
tmp_a[i] = res; tmp_a[i] = res;
} }
memcpy(a, tmp_a, n * sizeof(double)); std::copy_n(tmp_a.get(), n, a);
//P = T*P*transpose(T)+QQ; //P = T*P*transpose(T)+QQ;
memset(tmp, 0, n * n_state * sizeof(double)); std::fill_n(tmp, 0, n * n_state);
for (int i = 0; i < n; i++) for (int i = 0; i < n; i++)
for (int j = pure_obs; j < n; j++) for (int j = pure_obs; j < n; j++)
@ -488,17 +470,15 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
tmp[i + j1 * n ] += T[i + k * n] * P[k + j1_n_state]; tmp[i + j1 * n ] += T[i + k * n] * P[k + j1_n_state];
} }
memset(P, 0, n * n * sizeof(double)); std::fill_n(P, 0, n * n);
int n_n_obs = n * pure_obs; int n_n_obs = n * pure_obs;
for (int i = 0; i < n; i++) for (int i = 0; i < n; i++)
for (int j = i; j < n; j++) for (int j = i; j < n; j++)
{ for (int k = pure_obs; k < i_nz_state_var[j]; k++)
for (int k = pure_obs; k < i_nz_state_var[j]; k++) {
{ int k_n = k * n;
int k_n = k * n; P[i * n + j] += tmp[i + k_n - n_n_obs] * T[j + k_n];
P[i * n + j] += tmp[i + k_n - n_n_obs] * T[j + k_n]; }
}
}
for (int i = 0; i < n; i++) for (int i = 0; i < n; i++)
{ {
@ -513,11 +493,11 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
F_singular = false; F_singular = false;
//dF = det(F); //dF = det(F);
dF = det(iF, size_d_index, ipiv); dF = det(iF, size_d_index, ipiv.get());
//iF = inv(F); //iF = inv(F);
//int lwork = 4/*2*/* pp; //int lwork = 4/*2*/* pp;
dgetri(&size_d_index, iF, &size_d_index, ipiv, w, &lw, &info); dgetri(&size_d_index, iF, &size_d_index, ipiv.get(), w.get(), &lw, &info);
if (info != 0) if (info != 0)
mexPrintf("dgetri failure with error %d\n", static_cast<int>(info)); mexPrintf("dgetri failure with error %d\n", static_cast<int>(info));
@ -536,31 +516,27 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
for (int i = 0; i < size_d_index; i++) for (int i = 0; i < size_d_index; i++)
res += v_pp[i] * v[i]; res += v_pp[i] * v[i];
lik[t] = (log(dF) + res + size_d_index * log(2.0*pi))/2; lik[t] = (log(dF) + res + size_d_index * log(2.0*M_PI))/2;
if (t + 1 >= start) if (t + 1 >= start)
LIK += lik[t]; LIK += lik[t];
if (missing_observations) if (missing_observations)
{
//K = P(:,mf)*iF; //K = P(:,mf)*iF;
#ifdef USE_OMP #ifdef USE_OMP
# pragma omp parallel for shared(P_mf) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) # pragma omp parallel for shared(P_mf) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
#endif #endif
for (int i = 0; i < n; i++) for (int i = 0; i < n; i++)
{ {
int j_j = 0; int j_j = 0;
//for (int j = 0; j < pp; j++) //for (int j = 0; j < pp; j++)
for (vector<int>::const_iterator j = d_index.begin(); j != d_index.end(); j++, j_j++) for (auto j = d_index.begin(); j != d_index.end(); j++, j_j++)
P_mf[i + j_j * n] = P[i + mf[*j] * n]; P_mf[i + j_j * n] = P[i + mf[*j] * n];
} }
}
else else
{ //K = P(:,mf)*iF;
//K = P(:,mf)*iF; for (int i = 0; i < n; i++)
for (int i = 0; i < n; i++) for (int j = 0; j < pp; j++)
for (int j = 0; j < pp; j++) P_mf[i + j * n] = P[i + mf[j] * n];
P_mf[i + j * n] = P[i + mf[j] * n];
}
#ifdef USE_OMP #ifdef USE_OMP
# pragma omp parallel for shared(K) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) # pragma omp parallel for shared(K) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
@ -603,20 +579,18 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
//P = T*(P-K*P(mf,:))*transpose(T)+QQ; //P = T*(P-K*P(mf,:))*transpose(T)+QQ;
int i_i = 0; int i_i = 0;
//#pragma omp parallel for shared(P_mf) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) //#pragma omp parallel for shared(P_mf) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
for (vector<int>::const_iterator i = d_index.begin(); i != d_index.end(); i++, i_i++) for (auto i = d_index.begin(); i != d_index.end(); i++, i_i++)
for (int j = pure_obs; j < n; j++) for (int j = pure_obs; j < n; j++)
P_mf[i_i + j * size_d_index] = P[mf[*i] + j * n]; P_mf[i_i + j * size_d_index] = P[mf[*i] + j * n];
} }
else else
{ //P = T*(P-K*P(mf,:))*transpose(T)+QQ;
//P = T*(P-K*P(mf,:))*transpose(T)+QQ;
#ifdef USE_OMP #ifdef USE_OMP
# pragma omp parallel for shared(P_mf) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) # pragma omp parallel for shared(P_mf) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
#endif #endif
for (int i = 0; i < pp; i++) for (int i = 0; i < pp; i++)
for (int j = pure_obs; j < n; j++) for (int j = pure_obs; j < n; j++)
P_mf[i + j * pp] = P[mf[i] + j * n]; P_mf[i + j * pp] = P[mf[i] + j * n];
}
#ifdef BLAS #ifdef BLAS
# ifdef USE_OMP # ifdef USE_OMP
@ -640,7 +614,7 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
} }
double one = 1.0; double one = 1.0;
double zero = 0.0; double zero = 0.0;
memcpy(P, QQ, n * n *sizeof(double)); std::copy_n(QQ, n * n, P);
blas_int n_b = n; blas_int n_b = n;
/*mexPrintf("sizeof(n_b)=%d, n_b=%d, sizeof(n)=%d, n=%d\n",sizeof(n_b),n_b,sizeof(n),n); /*mexPrintf("sizeof(n_b)=%d, n_b=%d, sizeof(n)=%d, n=%d\n",sizeof(n_b),n_b,sizeof(n),n);
mexEvalString("drawnow;");*/ mexEvalString("drawnow;");*/
@ -685,27 +659,24 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
/*int device; /*int device;
cudaGetDevice(&device);*/ cudaGetDevice(&device);*/
int n2 = n * n; int n2 = n * n;
double *d_A = 0; double *d_A = nullptr, *d_B = nullptr, *d_C = nullptr, *d_D = nullptr;
double *d_B = 0;
double *d_C = 0;
double *d_D = 0;
// Allocate device memory for the matrices // Allocate device memory for the matrices
if (cudaMalloc((void **) &d_A, n2 * sizeof(double)) != cudaSuccess) if (cudaMalloc(static_cast<void **>(&d_A), n2 * sizeof(double)) != cudaSuccess)
{ {
mexPrintf("!!!! device memory allocation error (allocate A)\n"); mexPrintf("!!!! device memory allocation error (allocate A)\n");
return false; return false;
} }
if (cudaMalloc((void **) &d_B, n2 * sizeof(d_B[0])) != cudaSuccess) if (cudaMalloc(static_cast<void **>(&d_B), n2 * sizeof(d_B[0])) != cudaSuccess)
{ {
mexPrintf("!!!! device memory allocation error (allocate B)\n"); mexPrintf("!!!! device memory allocation error (allocate B)\n");
return false; return false;
} }
if (cudaMalloc((void **) &d_C, n2 * sizeof(d_C[0])) != cudaSuccess) if (cudaMalloc(static_cast<void **>(&d_C), n2 * sizeof(d_C[0])) != cudaSuccess)
{ {
mexPrintf("!!!! device memory allocation error (allocate C)\n"); mexPrintf("!!!! device memory allocation error (allocate C)\n");
return false; return false;
} }
if (cudaMalloc((void **) &d_D, n2 * sizeof(d_D[0])) != cudaSuccess) if (cudaMalloc(static_cast<void **>(&d_D), n2 * sizeof(d_D[0])) != cudaSuccess)
{ {
mexPrintf("!!!! device memory allocation error (allocate D)\n"); mexPrintf("!!!! device memory allocation error (allocate D)\n");
return false; return false;
@ -793,7 +764,7 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
} }
} }
memset(tmp, 0, n * n_state * sizeof(double)); fill_n(tmp, 0, n * n_state);
# ifdef USE_OMP # ifdef USE_OMP
# pragma omp parallel for shared(tmp) num_threads(atoi(getenv("DYNARE_NUM_THREADS"))) # pragma omp parallel for shared(tmp) num_threads(atoi(getenv("DYNARE_NUM_THREADS")))
@ -811,7 +782,7 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
} }
} }
memset(P, 0, n * n * sizeof(double)); fill_n(P, 0, n * n);
int n_n_obs = -n * pure_obs; int n_n_obs = -n * pure_obs;
# ifdef USE_OMP # ifdef USE_OMP
@ -849,14 +820,13 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
for (int i = 0; i < n * size_d_index; i++) for (int i = 0; i < n * size_d_index; i++)
{ {
double res = abs(K[i] - oldK[i]); double res = abs(K[i] - oldK[i]);
if (res > max_abs) max_abs = std::max(res, max_abs);
max_abs = res;
} }
notsteady = max_abs > riccati_tol; notsteady = max_abs > riccati_tol;
//oldK = K(:); //oldK = K(:);
memcpy(oldK, K, n * pp * sizeof(double)); std::copy_n(K, n * pp, oldK.get());
} }
} }
t++; t++;
@ -865,12 +835,12 @@ BlockKalmanFilter::block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf,
if (F_singular) if (F_singular)
mexErrMsgTxt("The variance of the forecast error remains singular until the end of the sample\n"); mexErrMsgTxt("The variance of the forecast error remains singular until the end of the sample\n");
if (t < smpl) if (t < smpl)
block_kalman_filter_ss(P_mf, v_pp, K, K_P, a, K_P, P_t_t1, tmp, P); block_kalman_filter_ss();
return true; return true;
} }
void void
BlockKalmanFilter::return_results_and_clean(int nlhs, mxArray *plhs[], double *P_mf[], double *v_pp[], double *K[], double *v_n[], double *a[], double *K_P[], double *P_t_t1[], double *tmp[], double *P[]) BlockKalmanFilter::return_results_and_clean(int nlhs, mxArray *plhs[])
{ {
plhs[0] = mxCreateDoubleScalar(0); plhs[0] = mxCreateDoubleScalar(0);
@ -886,17 +856,6 @@ BlockKalmanFilter::return_results_and_clean(int nlhs, mxArray *plhs[], double *P
else else
mxDestroyArray(plik); mxDestroyArray(plik);
mxFree(w);
mxFree(i_nz_state_var);
mxFree(tmp_a);
//mxFree(v_pp);
//mxFree(v_n);
mxFree(mf);
mxFree(iw);
mxFree(ipiv);
mxFree(oldK);
//mxFree(P_mf);
mxDestroyArray(pa); mxDestroyArray(pa);
mxDestroyArray(p_tmp); mxDestroyArray(p_tmp);
mxDestroyArray(pQQ); mxDestroyArray(pQQ);
@ -911,8 +870,7 @@ BlockKalmanFilter::return_results_and_clean(int nlhs, mxArray *plhs[], double *P
void void
mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{ {
double *P_mf, *v_pp, *v_n, *a, *K, *K_P, *P_t_t1, *tmp, *P; BlockKalmanFilter block_kalman_filter(nlhs, plhs, nrhs, prhs);
BlockKalmanFilter block_kalman_filter(nlhs, plhs, nrhs, prhs, &P_mf, &v_pp, &K, &v_n, &a, &K_P, &P_t_t1, &tmp, &P); if (block_kalman_filter.block_kalman_filter(nlhs, plhs))
if (block_kalman_filter.block_kalman_filter(nlhs, plhs, P_mf, v_pp, K, K_P, a, K_P, P_t_t1, tmp, P)) block_kalman_filter.return_results_and_clean(nlhs, plhs);
block_kalman_filter.return_results_and_clean(nlhs, plhs, &P_mf, &v_pp, &K, &K_P, &a, &K_P, &P_t_t1, &tmp, &P);
} }

26
mex/sources/block_kalman_filter/block_kalman_filter.hh
View File

@ -1,5 +1,5 @@
/* /*
* Copyright © 2007-2017 Dynare Team * Copyright © 2007-2019 Dynare Team
* *
* This file is part of Dynare. * This file is part of Dynare.
* *
@ -26,9 +26,11 @@
# include "mex_interface.hh" # include "mex_interface.hh"
#endif #endif
#include <vector>
#include <memory>
#include <dynblas.h> #include <dynblas.h>
#include <dynlapack.h> #include <dynlapack.h>
using namespace std;
class BlockKalmanFilter class BlockKalmanFilter
{ {
@ -40,25 +42,27 @@ public:
lapack_int pp, lw, info; lapack_int pp, lw, info;
double *nz_state_var; double *nz_state_var;
int *i_nz_state_var, *mf; std::unique_ptr<int[]> i_nz_state_var, mf;
int n_diag, t; int n_diag, t;
mxArray *M_; mxArray *M_;
mxArray *pa, *p_tmp, *p_tmp1, *plik; mxArray *pa, *p_tmp, *p_tmp1, *plik;
double *tmp_a, *tmp1, *lik, *v_n, *w, *oldK; std::unique_ptr<double[]> tmp_a;
double *tmp1, *lik;
std::unique_ptr<double[]> v_n, v_pp, w, oldK, P_mf;
bool notsteady, F_singular, missing_observations; bool notsteady, F_singular, missing_observations;
lapack_int *iw, *ipiv; std::unique_ptr<lapack_int[]> iw, ipiv;
double anorm, rcond; double anorm, rcond;
lapack_int size_d_index; lapack_int size_d_index;
int no_more_missing_observations, number_of_observations; int no_more_missing_observations, number_of_observations;
const mxArray *pdata_index; const mxArray *pdata_index;
vector<int> d_index; std::vector<int> d_index;
const mxArray *pd_index; const mxArray *pd_index;
double *dd_index; double *dd_index;
double *K, *a, *K_P, *P_t_t1, *tmp, *P;
public: public:
BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[], double *P_mf[], double *v_pp[], double *K[], double *v_n[], double *a[], double *K_P[], double *P_t_t1[], double *tmp[], double *P[]); BlockKalmanFilter(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]);
bool block_kalman_filter(int nlhs, mxArray *plhs[], double *P_mf, double *v_pp, double *K, double *v_n, double *a, double *K_P, double *P_t_t1, double *tmp, double *P); bool block_kalman_filter(int nlhs, mxArray *plhs[]);
void block_kalman_filter_ss(double *P_mf, double *v_pp, double *K, double *v_n, double *a, double *K_P, double *P_t_t1, double *tmp, double *P); void block_kalman_filter_ss();
void return_results_and_clean(int nlhs, mxArray *plhs[], double *P_mf[], double *v_pp[], double *K[], double *v_n[], double *a[], double *K_P[], double *P_t_t1[], double *tmp[], double *P[]); void return_results_and_clean(int nlhs, mxArray *plhs[]);
}; };
#endif #endif