38 lines
1.2 KiB
Plaintext
38 lines
1.2 KiB
Plaintext
//---------------------------------------------------------------------------//
|
|
// 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
|