From 483eb2f56657e8e7f419ab1a4fab8dce9ade8609 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sat, 27 Apr 2024 20:24:20 +0200 Subject: Adding upstream version 14.2.21. Signed-off-by: Daniel Baumann --- .../libs/compute/perf/perf_set_intersection.cpp | 75 ++++++++++++++++++++++ 1 file changed, 75 insertions(+) create mode 100644 src/boost/libs/compute/perf/perf_set_intersection.cpp (limited to 'src/boost/libs/compute/perf/perf_set_intersection.cpp') diff --git a/src/boost/libs/compute/perf/perf_set_intersection.cpp b/src/boost/libs/compute/perf/perf_set_intersection.cpp new file mode 100644 index 00000000..dbfeb42f --- /dev/null +++ b/src/boost/libs/compute/perf/perf_set_intersection.cpp @@ -0,0 +1,75 @@ +//---------------------------------------------------------------------------// +// Copyright (c) 2014 Roshan +// +// 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. +//---------------------------------------------------------------------------// + +#include +#include +#include +#include + +#include +#include +#include + +#include "perf.hpp" + +int rand_int() +{ + return static_cast((rand() / double(RAND_MAX)) * 25.0); +} + +int main(int argc, char *argv[]) +{ + perf_parse_args(argc, argv); + std::cout << "size: " << PERF_N << std::endl; + + // setup context and queue for the default device + boost::compute::device device = boost::compute::system::default_device(); + boost::compute::context context(device); + boost::compute::command_queue queue(context, device); + std::cout << "device: " << device.name() << std::endl; + + // create vectors of random numbers on the host + std::vector v1(std::floor(PERF_N / 2.0)); + std::vector v2(std::ceil(PERF_N / 2.0)); + std::generate(v1.begin(), v1.end(), rand_int); + std::generate(v2.begin(), v2.end(), rand_int); + std::sort(v1.begin(), v1.end()); + std::sort(v2.begin(), v2.end()); + + // create vectors on the device and copy the data + boost::compute::vector gpu_v1(std::floor(PERF_N / 2.0), context); + boost::compute::vector gpu_v2(std::ceil(PERF_N / 2.0), context); + + boost::compute::copy( + v1.begin(), v1.end(), gpu_v1.begin(), queue + ); + boost::compute::copy( + v2.begin(), v2.end(), gpu_v2.begin(), queue + ); + + boost::compute::vector gpu_v3(PERF_N, context); + boost::compute::vector::iterator gpu_v3_end; + + perf_timer t; + for(size_t trial = 0; trial < PERF_TRIALS; trial++){ + t.start(); + gpu_v3_end = boost::compute::set_intersection( + gpu_v1.begin(), gpu_v1.end(), + gpu_v2.begin(), gpu_v2.end(), + gpu_v3.begin(), queue + ); + queue.finish(); + t.stop(); + } + std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl; + std::cout << "size: " << std::distance(gpu_v3.begin(), gpu_v3_end) << std::endl; + + return 0; +} -- cgit v1.2.3