Vector<dataType> mvmul(const MatrixExpression<MT,dataType> &A,
const VectorExpression<VT,dataType> &x) {
#ifndef _OPTIMIZED
//ensure proper dimensions
if(A.cols() != x.size()) {
throw std::length_error("mvmul dimension mismatch");
}
#endif
Vector<dataType> b(A.rows());
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(std::size_t i=0; i<A.rows(); ++i) {
dataType sum(0);
auto N = A.cols()/4;
auto NN = 4*N;
for(std::size_t j=0; j<NN; j+= 4) {
sum += A(i,j+0)*x(j+0);
sum += A(i,j+1)*x(j+1);
sum += A(i,j+2)*x(j+2);
sum += A(i,j+3)*x(j+3);
}
for(std::size_t j=NN; j<A.cols(); ++j) {
sum += A(i,j)*x(j);
}
b(i) = sum;
}
return b;
}
inline const typename A::BASE_TYPE Sum(const VectorExpression<T2, A> &expr, bool concurrent = false) {
typename A::BASE_TYPE ret; // = TT(0.0);
if (concurrent) {
int range = expr.Size(0) / std::thread::hardware_concurrency();
std::vector<std::thread> pool;
std::vector<typename A::BASE_TYPE > temp(std::thread::hardware_concurrency());
for (int i = 0; i < std::thread::hardware_concurrency(); i++) {
if (i < (std::thread::hardware_concurrency() - 1)) {
// std::cout << i*range << " - " << (i + 1) * range << "\n";
pool.push_back(std::thread(SumThread<T2, A>, std::ref(temp[i]), i*range, (i + 1) * range, std::ref(expr)));
} else {
pool.push_back(std::thread(SumThread<T2, A>, std::ref(temp[i]), i*range, expr.Size(0), std::ref(expr)));
// std::cout << i*range << " - " << expr.Size(0) << "\n";
}
}
for (int i = 0; i < pool.size(); i++) {
pool[i].join();
}
for (int i = 0; i < std::thread::hardware_concurrency(); i++) {
ret += temp[i];
}
} else {
size_t s = expr.Size(0);
size_t end = (((s - 1UL) & size_t(-2)) + 1UL);
ret = expr(0);
for (size_t i = 1UL; i < end; i += 2UL) {
ret += expr(i) + expr(i + 1);
}
if (end < s) {
ret += expr(end);
}
}
return ret;
}
inline const typename atl::PromoteType<typename LHS::RET_TYPE, typename RHS::RET_TYPE >::return_type Dot(const VectorExpression<typename LHS::RET_TYPE, LHS>& a,
const VectorExpression<typename RHS::RET_TYPE, RHS>& b) {
typedef typename atl::PromoteType<typename LHS::BASE_TYPE, typename RHS::BASE_TYPE>::return_type R_TYPE;
R_TYPE sum;
int lmin = a.IndexMin();
int lmax = a.IndexMax();
int rmin = b.IndexMin();
int rmax = b.IndexMax();
int min = std::max(lmin, rmin);
int max = std::min(lmax, rmax);
for (int i = min; i <= max; i++) {
sum += a(i) * b(i);
}
return sum;
}
Vector(const VectorExpression<T2, A> &expr)
: isize(0) {
isize = expr.Size(0);
data_m.resize(isize);
for (int i = 0; i < isize; i++) {
data_m[i] = expr(i);
}
}
Vector(VectorExpression<E> const& other) : values(other.size()) {
for (std::size_t i=0; i<other.size(); i++) {
values[i] = other[i];
}
}