1 //---------------------------------------------------------------------------//
2 // Copyright (c) 2013-2014 Kyle Lutz <kyle.r.lutz@gmail.com>
3 //
4 // Distributed under the Boost Software License, Version 1.0
5 // See accompanying file LICENSE_1_0.txt or copy at
6 // http://www.boost.org/LICENSE_1_0.txt
7 //
8 // See http://boostorg.github.com/compute for more information.
9 //---------------------------------------------------------------------------//
10 
11 #ifndef BOOST_COMPUTE_DETAIL_WORK_SIZE_HPP
12 #define BOOST_COMPUTE_DETAIL_WORK_SIZE_HPP
13 
14 #include <cmath>
15 
16 namespace boost {
17 namespace compute {
18 namespace detail {
19 
20 // Given a total number of values (count), a number of values to
21 // process per thread (vtp), and a number of threads to execute per
22 // block (tpb), this function returns the global work size to be
23 // passed to clEnqueueNDRangeKernel() for a 1D algorithm.
calculate_work_size(size_t count,size_t vpt,size_t tpb)24 inline size_t calculate_work_size(size_t count, size_t vpt, size_t tpb)
25 {
26     size_t work_size = static_cast<size_t>(std::ceil(float(count) / vpt));
27     if(work_size % tpb != 0){
28         work_size += tpb - work_size % tpb;
29     }
30     return work_size;
31 }
32 
33 } // end detail namespace
34 } // end compute namespace
35 } // end boost namespace
36 
37 #endif // BOOST_COMPUTE_DETAIL_WORK_SIZE_HPP
38