boost/compute/detail/work_size.hpp
//---------------------------------------------------------------------------//
// Copyright (c) 2013-2014 Kyle Lutz <kyle.r.lutz@gmail.com>
//
// Distributed under the Boost Software License, Version 1.0
// See accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt
//
// See http://boostorg.github.com/compute for more information.
//---------------------------------------------------------------------------//
#ifndef BOOST_COMPUTE_DETAIL_WORK_SIZE_HPP
#define BOOST_COMPUTE_DETAIL_WORK_SIZE_HPP
#include <cmath>
namespace boost {
namespace compute {
namespace detail {
// Given a total number of values (count), a number of values to
// process per thread (vtp), and a number of threads to execute per
// block (tpb), this function returns the global work size to be
// passed to clEnqueueNDRangeKernel() for a 1D algorithm.
inline size_t calculate_work_size(size_t count, size_t vpt, size_t tpb)
{
size_t work_size = static_cast<size_t>(std::ceil(float(count) / vpt));
if(work_size % tpb != 0){
work_size += tpb - work_size % tpb;
}
return work_size;
}
} // end detail namespace
} // end compute namespace
} // end boost namespace
#endif // BOOST_COMPUTE_DETAIL_WORK_SIZE_HPP