void update(const std::vector< double >& inputs)
{
optimizer_.setInputValues(inputs);
copy_results();
}
double matrix_multiply(void) {
int oi, oj;
int i, j, k, l, q, p;
double start, end;
// timer for the start of the computation
// Reorganize the data but do not start multiplying elements before
// the timer value is captured.
start = omp_get_wtime();
/* //Single loop - makes no obvious difference
#pragma omp parallel for
for(i=0; i<SIZE_N*SIZE_M; i++){
oi = i / SIZE_M;
oj = i % SIZE_M;
// #pragma omp parallel for
// for(oj=0; oj<SIZE_M; oj++){
//this is where we launch!
//iterate over matrices:
for(k=0; k<SIZE_P; k++){
//submatrix multiply:
for(q=0; q<BLOCK_SIZE; q++){
for(l=0; l<BLOCK_SIZE; l++){
for(p=0; p<BLOCK_SIZE; p++){
SUB_C[oi][oj][q][l] += SUB_A[oi][k][q][p] * SUB_B[k][oj][p][l];
}
}
}
}
//end launch
}
*/
#ifdef PARALLEL
#pragma omp parallel for
#endif
for(oi=0; oi<SIZE_N; oi++){
#ifdef PARALLEL
#pragma omp parallel for
#endif
for(oj=0; oj<SIZE_M; oj++){
//this is where we launch!
//iterate over matrices:
for(k=0; k<SIZE_P; k++){
//submatrix multiply:
for(q=0; q<BLOCK_SIZE; q++){
for(l=0; l<BLOCK_SIZE; l++){
for(p=0; p<BLOCK_SIZE; p++){
SUB_C[oi][oj][q][l] += SUB_A[oi][k][q][p] * SUB_B[k][oj][p][l];
}
}
}
}
//end launch
}
}
// timer for the end of the computation
end = omp_get_wtime();
//copy the results from the streamlined matrix into the old matrix form (for... compatability?)
copy_results();
// return the amount of high resolution time spent
return end - start;
}
void update(const Eigen::VectorXd& inputs)
{
optimizer_.setInputValues(inputs);
copy_results();
}