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
 | 
