12345678910111213141516171819202122232425262728293031323334353637 |
- //---------------------------------------------------------------------------//
- // 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
|