inline void loop_compute(double* const target_ptr, double* const density_ptr, boost::shared_array<double*> cell_ptr, int& nbf) {
for (int j = 0; j < nbf; j++) {
for (int i = j; i < nbf; i++) {
double element = 0.0;
double* c_mat_ptr = cell_ptr[CMO2(i, j, nbf)];
/* Build the elements of J or K. */
for (int k = 0; k < nbf * nbf; k++) {
element += c_mat_ptr[k] * density_ptr[k];
}
target_ptr[CMO2(i, j, nbf)] = element;
/* Taking advantage of symmetry. */
if (i != j) {
target_ptr[CMO2(j, i, nbf)] = element;
}
}
}
}
void mexFunction (int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) {
double t1, t2;
double *G, *P, *H2jj, *H2kk;
int i, j, k, nbasis;
mxArray *H2j, *H2k, *mH2jj, *mH2kk;
if (nlhs > 1)
mexErrMsgIdAndTxt("twoElecFock:nlhs",
"Invalid number of output variables for twoElecFock.");
if (nrhs != 3)
mexErrMsgIdAndTxt("twoElecFock:nrhs",
"Invalid number of input variables to twoElecFock.");
/* Pull data out of the input arrays. */
P = mxGetPr(prhs[0]);
H2j = (mxArray*)prhs[1];
H2k = (mxArray*)prhs[2];
nbasis = (int)mxGetM(prhs[0]);
/* Validate input. */
CHECK(nbasis == mxGetM(prhs[0]));
CHECK(nbasis == mxGetN(prhs[0]));
CHECK(nbasis == mxGetM(H2j));
CHECK(nbasis == mxGetN(H2j));
CHECK(nbasis == mxGetM(H2k));
CHECK(nbasis == mxGetN(H2k));
/* Allocate the return array. */
G = mxCalloc(nbasis * nbasis, sizeof(double));
/*
* Theoretically, this order of loop variables is the most cache efficient,
* but testing doesn't show any appreciable difference in practice.
*/
for (j = 0; j < nbasis; j++) {
for (i = j; i < nbasis; i++) {
t1 = 0.0;
t2 = 0.0;
/* Get current arrrays out of the cell arrays and validate input. */
mH2jj = mxGetCell(H2j, CMO2(i, j, nbasis));
mH2kk = mxGetCell(H2k, CMO2(i, j, nbasis));
CHECK(nbasis == mxGetM(mH2jj));
CHECK(nbasis == mxGetN(mH2jj));
CHECK(nbasis == mxGetM(mH2kk));
CHECK(nbasis == mxGetN(mH2kk));
H2jj = mxGetPr(mH2jj);
H2kk = mxGetPr(mH2kk);
/* Build the elements of G. */
for (k = 0; k < nbasis * nbasis; k++) {
t1 += H2jj[k] * P[k];
t2 += H2kk[k] * P[k];
}
G[CMO2(i, j, nbasis)] = t1 - t2 / 2;
/* Taking advantage of symmetry. */
if (i != j)
G[CMO2(j, i, nbasis)] = t1 - t2 / 2;
}
}
/* Prepare the data to hand back to MATLAB. */
plhs[0] = mxCreateNumericMatrix(0, 0, mxDOUBLE_CLASS, mxREAL);
mxSetPr(plhs[0], G);
mxSetM(plhs[0], nbasis);
mxSetN(plhs[0], nbasis);
return;
}
