static void rowSums(TMat<TNumMat> mat, Vec<TNumVec> vec, int nthreads){
if (mat.nrow != vec.len) throw std::invalid_argument("provided vector has invalid length");
int nrow = mat.nrow;
int ncol = mat.ncol;
#pragma omp parallel num_threads(std::max(1, nthreads))
{
std::vector<TNumVec> acc(nrow, 0);
TNumVec* accBegin = acc.data();
#pragma omp for schedule(static) nowait
for (int col = 0; col < ncol; ++col){
TNumMat* matCol = mat.colptr(col);
TNumVec* accIter = accBegin;
for (int row = 0; row < nrow; ++row){//this loop should be unrolled...
*accIter++ += *matCol++;
}
}
#pragma omp critical
{
for (int row = 0; row < nrow; ++row){
vec[row] += acc[row];
}
}
}
}
static void colSums(TMat<TNumMat> mat, Vec<TNumVec> vec, int nthreads){
if (mat.ncol != vec.len) throw std::invalid_argument("provided vector has invalid length");
TNumVec* cs = vec.ptr;
int nrow = mat.nrow;
int ncol = mat.ncol;
#pragma omp parallel for schedule(static) num_threads(std::max(1, nthreads))
for (int col = 0; col < ncol; ++col){
TNumMat* ptr = mat.colptr(col);
TNumMat tmp = 0;
for (int row = 0; row < nrow; ++row){
tmp += *ptr++;
}
cs[col] = tmp;
}
}
