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

Fix integer types in Kronecker DLLs

time-shift
Sébastien Villemot 2011-01-13 19:34:49 +01:00
parent 23d797d214
commit ffc53ca910
2 changed files with 56 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
mex/sources/kronecker/A_times_B_kronecker_C.cc
View File

@ -1,5 +1,5 @@
/* /*
* Copyright (C) 2007-2010 Dynare Team * Copyright (C) 2007-2011 Dynare Team
* *
* This file is part of Dynare. * This file is part of Dynare.
* *
@ -39,36 +39,36 @@ full_A_times_kronecker_B_C(double *A, double *B, double *C, double *D,
{ {
#if USE_OMP #if USE_OMP
# pragma omp parallel for num_threads(number_of_threads) # pragma omp parallel for num_threads(number_of_threads)
for (long int colD = 0; colD < nB*nC; colD++) for (blas_int colD = 0; colD < nB*nC; colD++)
{ {
# if DEBUG_OMP # if DEBUG_OMP
mexPrintf("%d thread number is %d (%d).\n", colD, omp_get_thread_num(), omp_get_num_threads()); mexPrintf("%d thread number is %d (%d).\n", colD, omp_get_thread_num(), omp_get_num_threads());
# endif # endif
unsigned int colB = colD/nC; blas_int colB = colD/nC;
unsigned int colC = colD%nC; blas_int colC = colD%nC;
for (int colA = 0; colA < nA; colA++) for (blas_int colA = 0; colA < nA; colA++)
{ {
unsigned int rowB = colA/mC; blas_int rowB = colA/mC;
unsigned int rowC = colA%mC; blas_int rowC = colA%mC;
unsigned int idxA = colA*mA; blas_int idxA = colA*mA;
unsigned int idxD = colD*mA; blas_int idxD = colD*mA;
double BC = B[colB*mB+rowB]*C[colC*mC+rowC]; double BC = B[colB*mB+rowB]*C[colC*mC+rowC];
for (int rowD = 0; rowD < mA; rowD++) for (blas_int rowD = 0; rowD < mA; rowD++)
{ {
D[idxD+rowD] += A[idxA+rowD]*BC; D[idxD+rowD] += A[idxA+rowD]*BC;
} }
} }
} }
#else #else
const unsigned long shiftA = mA*mC; const blas_int shiftA = mA*mC;
const unsigned long shiftD = mA*nC; const blas_int shiftD = mA*nC;
unsigned long int kd = 0, ka = 0; blas_int kd = 0, ka = 0;
char transpose[2] = "N"; char transpose[2] = "N";
double one = 1.0; double one = 1.0;
for (unsigned long int col = 0; col < nB; col++) for (blas_int col = 0; col < nB; col++)
{ {
ka = 0; ka = 0;
for (unsigned long int row = 0; row < mB; row++) for (blas_int row = 0; row < mB; row++)
{ {
dgemm(transpose, transpose, &mA, &nC, &mC, &B[mB*col+row], &A[ka], &mA, &C[0], &mC, &one, &D[kd], &mA); dgemm(transpose, transpose, &mA, &nC, &mC, &B[mB*col+row], &A[ka], &mA, &C[0], &mC, &one, &D[kd], &mA);
ka += shiftA; ka += shiftA;
@ -83,36 +83,36 @@ full_A_times_kronecker_B_B(double *A, double *B, double *D, blas_int mA, blas_in
{ {
#if USE_OMP #if USE_OMP
# pragma omp parallel for num_threads(number_of_threads) # pragma omp parallel for num_threads(number_of_threads)
for (long int colD = 0; colD < nB*nB; colD++) for (blas_int colD = 0; colD < nB*nB; colD++)
{ {
# if DEBUG_OMP # if DEBUG_OMP
mexPrintf("%d thread number is %d (%d).\n", colD, omp_get_thread_num(), omp_get_num_threads()); mexPrintf("%d thread number is %d (%d).\n", colD, omp_get_thread_num(), omp_get_num_threads());
# endif # endif
unsigned int j1B = colD/nB; blas_int j1B = colD/nB;
unsigned int j2B = colD%nB; blas_int j2B = colD%nB;
for (int colA = 0; colA < nA; colA++) for (blas_int colA = 0; colA < nA; colA++)
{ {
unsigned int i1B = colA/mB; blas_int i1B = colA/mB;
unsigned int i2B = colA%mB; blas_int i2B = colA%mB;
unsigned int idxA = colA*mA; blas_int idxA = colA*mA;
unsigned int idxD = colD*mA; blas_int idxD = colD*mA;
double BB = B[j1B*mB+i1B]*B[j2B*mB+i2B]; double BB = B[j1B*mB+i1B]*B[j2B*mB+i2B];
for (int rowD = 0; rowD < mA; rowD++) for (blas_int rowD = 0; rowD < mA; rowD++)
{ {
D[idxD+rowD] += A[idxA+rowD]*BB; D[idxD+rowD] += A[idxA+rowD]*BB;
} }
} }
} }
#else #else
const unsigned long int shiftA = mA*mB; const blas_int shiftA = mA*mB;
const unsigned long int shiftD = mA*nB; const blas_int shiftD = mA*nB;
unsigned long int kd = 0, ka = 0; blas_int kd = 0, ka = 0;
char transpose[2] = "N"; char transpose[2] = "N";
double one = 1.0; double one = 1.0;
for (unsigned long int col = 0; col < nB; col++) for (blas_int col = 0; col < nB; col++)
{ {
ka = 0; ka = 0;
for (unsigned long int row = 0; row < mB; row++) for (blas_int row = 0; row < mB; row++)
{ {
dgemm(transpose, transpose, &mA, &nB, &mB, &B[mB*col+row], &A[ka], &mA, &B[0], &mB, &one, &D[kd], &mA); dgemm(transpose, transpose, &mA, &nB, &mB, &B[mB*col+row], &A[ka], &mA, &B[0], &mB, &one, &D[kd], &mA);
ka += shiftA; ka += shiftA;
@ -174,11 +174,11 @@ mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
// Computational part: // Computational part:
if (nrhs == 3) if (nrhs == 3)
{ {
full_A_times_kronecker_B_B(A, B, &D[0], (int) mA, (int) nA, (int) mB, (int) nB, numthreads); full_A_times_kronecker_B_B(A, B, &D[0], mA, nA, mB, nB, numthreads);
} }
else else
{ {
full_A_times_kronecker_B_C(A, B, C, &D[0], (int) mA, (int) nA, (int) mB, (int) nB, (int) mC, (int) nC, numthreads); full_A_times_kronecker_B_C(A, B, C, &D[0], mA, nA, mB, nB, mC, nC, numthreads);
} }
plhs[0] = mxCreateDoubleScalar(0); plhs[0] = mxCreateDoubleScalar(0);
} }

50
mex/sources/kronecker/sparse_hessian_times_B_kronecker_C.cc
View File

@ -1,5 +1,5 @@
/* /*
* Copyright (C) 2007-2010 Dynare Team * Copyright (C) 2007-2011 Dynare Team
* *
* This file is part of Dynare. * This file is part of Dynare.
* *
@ -45,37 +45,37 @@ sparse_hessian_times_B_kronecker_B(mwIndex *isparseA, mwIndex *jsparseA, double
#if USE_OMP #if USE_OMP
# pragma omp parallel for num_threads(number_of_threads) # pragma omp parallel for num_threads(number_of_threads)
#endif #endif
for (int j1B = 0; j1B < nB; j1B++) for (mwIndex j1B = 0; j1B < nB; j1B++)
{ {
#if DEBUG_OMP #if DEBUG_OMP
mexPrintf("%d thread number is %d (%d).\n", j1B, omp_get_thread_num(), omp_get_num_threads()); mexPrintf("%d thread number is %d (%d).\n", j1B, omp_get_thread_num(), omp_get_num_threads());
#endif #endif
for (unsigned int j2B = j1B; j2B < nB; j2B++) for (mwIndex j2B = j1B; j2B < nB; j2B++)
{ {
unsigned long int jj = j1B*nB+j2B; // column of kron(B,B) index. mwIndex jj = j1B*nB+j2B; // column of kron(B,B) index.
unsigned long int iv = 0; mwIndex iv = 0;
unsigned int nz_in_column_ii_of_A = 0; int nz_in_column_ii_of_A = 0;
unsigned int k1 = 0; mwIndex k1 = 0;
unsigned int k2 = 0; mwIndex k2 = 0;
/* /*
** Loop over the rows of kron(B,B) (column jj). ** Loop over the rows of kron(B,B) (column jj).
*/ */
for (unsigned long int ii = 0; ii < nA; ii++) for (mwIndex ii = 0; ii < nA; ii++)
{ {
k1 = jsparseA[ii]; k1 = jsparseA[ii];
k2 = jsparseA[ii+1]; k2 = jsparseA[ii+1];
if (k1 < k2) // otherwise column ii of A does not have non zero elements (and there is nothing to compute). if (k1 < k2) // otherwise column ii of A does not have non zero elements (and there is nothing to compute).
{ {
++nz_in_column_ii_of_A; ++nz_in_column_ii_of_A;
unsigned int i1B = (ii/mB); mwIndex i1B = (ii/mB);
unsigned int i2B = (ii%mB); mwIndex i2B = (ii%mB);
double bb = B[j1B*mB+i1B]*B[j2B*mB+i2B]; double bb = B[j1B*mB+i1B]*B[j2B*mB+i2B];
/* /*
** Loop over the non zero entries of A(:,ii). ** Loop over the non zero entries of A(:,ii).
*/ */
for (unsigned int k = k1; k < k2; k++) for (mwIndex k = k1; k < k2; k++)
{ {
unsigned int kk = isparseA[k]; mwIndex kk = isparseA[k];
D[jj*mA+kk] = D[jj*mA+kk] + bb*vsparseA[iv]; D[jj*mA+kk] = D[jj*mA+kk] + bb*vsparseA[iv];
iv++; iv++;
} }
@ -100,37 +100,37 @@ sparse_hessian_times_B_kronecker_C(mwIndex *isparseA, mwIndex *jsparseA, double
#if USE_OMP #if USE_OMP
# pragma omp parallel for num_threads(number_of_threads) # pragma omp parallel for num_threads(number_of_threads)
#endif #endif
for (long int jj = 0; jj < nB*nC; jj++) // column of kron(B,C) index. for (mwIndex jj = 0; jj < nB*nC; jj++) // column of kron(B,C) index.
{ {
// Uncomment the following line to check if all processors are used. // Uncomment the following line to check if all processors are used.
#if DEBUG_OMP #if DEBUG_OMP
mexPrintf("%d thread number is %d (%d).\n", jj, omp_get_thread_num(), omp_get_num_threads()); mexPrintf("%d thread number is %d (%d).\n", jj, omp_get_thread_num(), omp_get_num_threads());
#endif #endif
unsigned int jB = jj/nC; mwIndex jB = jj/nC;
unsigned int jC = jj%nC; mwIndex jC = jj%nC;
unsigned int k1 = 0; mwIndex k1 = 0;
unsigned int k2 = 0; mwIndex k2 = 0;
unsigned long int iv = 0; mwIndex iv = 0;
unsigned int nz_in_column_ii_of_A = 0; int nz_in_column_ii_of_A = 0;
/* /*
** Loop over the rows of kron(B,C) (column jj). ** Loop over the rows of kron(B,C) (column jj).
*/ */
for (unsigned long int ii = 0; ii < nA; ii++) for (mwIndex ii = 0; ii < nA; ii++)
{ {
k1 = jsparseA[ii]; k1 = jsparseA[ii];
k2 = jsparseA[ii+1]; k2 = jsparseA[ii+1];
if (k1 < k2) // otherwise column ii of A does not have non zero elements (and there is nothing to compute). if (k1 < k2) // otherwise column ii of A does not have non zero elements (and there is nothing to compute).
{ {
++nz_in_column_ii_of_A; ++nz_in_column_ii_of_A;
unsigned int iC = (ii%mB); mwIndex iC = (ii%mB);
unsigned int iB = (ii/mB); mwIndex iB = (ii/mB);
double cb = C[jC*mC+iC]*B[jB*mB+iB]; double cb = C[jC*mC+iC]*B[jB*mB+iB];
/* /*
** Loop over the non zero entries of A(:,ii). ** Loop over the non zero entries of A(:,ii).
*/ */
for (unsigned int k = k1; k < k2; k++) for (mwIndex k = k1; k < k2; k++)
{ {
unsigned int kk = isparseA[k]; mwIndex kk = isparseA[k];
D[jj*mA+kk] = D[jj*mA+kk] + cb*vsparseA[iv]; D[jj*mA+kk] = D[jj*mA+kk] + cb*vsparseA[iv];
iv++; iv++;
} }