void jacobi(
crs_matrix<double> const & A
, std::vector<double> const & b
, std::size_t iterations
, std::size_t block_size
)
{
typedef std::vector<double> vector_type;
std::shared_ptr<vector_type> dst(new vector_type(b));
std::shared_ptr<vector_type> src(new vector_type(b));
hpx::util::high_resolution_timer t;
for(std::size_t i = 0; i < iterations; ++i)
{
// MSVC is unhappy if the OMP loop variable is unsigned
#pragma omp parallel for schedule(JACOBI_SMP_OMP_SCHEDULE)
for(boost::int64_t row = 0; row < boost::int64_t(b.size()); ++row)
{
jacobi_kernel_nonuniform(
A
, *dst
, *src
, b
, row
);
}
std::swap(dst, src);
}
double time_elapsed = t.elapsed();
std::cout << dst->size() << " "
<< ((double(dst->size() * iterations)/1e6)/time_elapsed) << " MLUPS/s\n"
<< std::flush;
}
void jacobi_kernel_wrap(
range const & r,
crs_matrix<double> const & A,
std::vector<double> & x_dst, std::vector<double> const & x_src, std::vector<double> const & b)
{
for(std::size_t row = r.begin(); row < r.end(); ++row)
{
jacobi_kernel_nonuniform(A, x_dst, x_src, b, row);
}
}
