#882

Committed 31 Aug 2023 07:44PM UTC coverage: 41.798% (-44.7%) from 86.546%

Build # #882

Build Type

push

Committed by

Commit Message

Run Details

19442 of 46514 relevant lines covered (41.8%)

126375.38 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

0.0

/examples/transpose/transpose_serial.cpp

//  Copyright (c) 2014 Thomas Heller
//
//  SPDX-License-Identifier: BSL-1.0
//  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)

#include <hpx/chrono.hpp>
#include <hpx/init.hpp>

#include <algorithm>
#include <cstdint>
#include <exception>
#include <iostream>
#include <string>
#include <vector>

#define COL_SHIFT 1000.00    // Constant to shift column index
#define ROW_SHIFT 0.001      // Constant to shift row index

bool verbose = false;

double test_results(std::uint64_t order, std::vector<double> const& trans);

///////////////////////////////////////////////////////////////////////////////
int hpx_main(hpx::program_options::variables_map& vm)
{
    std::uint64_t order = vm["matrix_size"].as<std::uint64_t>();
    std::uint64_t iterations = vm["iterations"].as<std::uint64_t>();
    std::uint64_t tile_size = order;

    if (vm.count("tile_size"))
        tile_size = vm["tile_size"].as<std::uint64_t>();

    verbose = vm.count("verbose") ? true : false;

    std::uint64_t bytes =
        static_cast<std::uint64_t>(2 * sizeof(double) * order * order);

    std::vector<double> A(order * order);
    std::vector<double> B(order * order);

    std::cout << "Serial Matrix transpose: B = A^T\n"
              << "Matrix order          = " << order << "\n";
    if (tile_size < order)
        std::cout << "Tile size             = " << tile_size << "\n";
    else
        std::cout << "Untiled\n";
    std::cout << "Number of iterations  = " << iterations << "\n";

    // Fill the original matrix, set transpose to known garbage value.
    for (std::uint64_t i = 0; i < order; ++i)
    {
        for (std::uint64_t j = 0; j < order; ++j)
        {
            A[i * order + j] = COL_SHIFT * static_cast<double>(j) +
                ROW_SHIFT * static_cast<double>(i);
            B[i * order + j] = -1.0;
        }
    }

    double errsq = 0.0;
    double avgtime = 0.0;
    double maxtime = 0.0;
    double mintime =
        366.0 * 24.0 * 3600.0;    // set the minimum time to a large value;
                                  // one leap year should be enough
    for (std::uint64_t iter = 0; iter < iterations; ++iter)
    {
        hpx::chrono::high_resolution_timer t;

        if (tile_size < order)
        {
            for (std::uint64_t i = 0; i < order; i += tile_size)
            {
                for (std::uint64_t j = 0; j < order; j += tile_size)
                {
                    std::uint64_t i_max = (std::min) (order, i + tile_size);
                    for (std::uint64_t it = i; it < i_max; ++it)
                    {
                        std::uint64_t j_max = (std::min) (order, j + tile_size);
                        for (std::uint64_t jt = j; jt < j_max; ++jt)
                        {
                            B[it + order * jt] = A[jt + order * it];
                        }
                    }
                }
            }
        }
        else
        {
            for (std::uint64_t i = 0; i < order; ++i)
            {
                for (std::uint64_t j = 0; j < order; ++j)
                {
                    B[i + order * j] = A[j + order * i];
                }
            }
        }

        double elapsed = t.elapsed();

        if (iter > 0 || iterations == 1)    // Skip the first iteration
        {
            avgtime = avgtime + elapsed;
            maxtime = (std::max) (maxtime, elapsed);
            mintime = (std::min) (mintime, elapsed);
        }

        errsq += test_results(order, B);
    }    // end of iter loop

    // Analyze and output results

    double epsilon = 1.e-8;
    if (errsq < epsilon)
    {
        std::cout << "Solution validates\n";
        avgtime = avgtime /
            static_cast<double>(
                (std::max) (iterations - 1, static_cast<std::uint64_t>(1)));
        std::cout << "Rate (MB/s): "
                  << 1.e-6 * static_cast<double>(bytes) / mintime << ", "
                  << "Avg time (s): " << avgtime << ", "
                  << "Min time (s): " << mintime << ", "
                  << "Max time (s): " << maxtime << "\n";

        if (verbose)
            std::cout << "Squared errors: " << errsq << "\n";
    }
    else
    {
        std::cout << "ERROR: Aggregate squared error " << errsq
                  << " exceeds threshold " << epsilon << "\n";
        std::terminate();
    }

    return hpx::local::finalize();
}

int main(int argc, char* argv[])
{
    using namespace hpx::program_options;

    options_description desc_commandline;
    // clang-format off
    desc_commandline.add_options()
        ("matrix_size", value<std::uint64_t>()->default_value(1024),
         "Matrix Size")
        ("iterations", value<std::uint64_t>()->default_value(10),
         "# iterations")
        ("tile_size", value<std::uint64_t>(),
         "Number of tiles to divide the individual matrix blocks for improved "
         "cache and TLB performance")
        ( "verbose", "Verbose output")
    ;
    // clang-format on

    // Initialize and run HPX, this example is serial and therefore only needs
    // one thread, We just use hpx::init to parse our command line arguments
    std::vector<std::string> const cfg = {"hpx.os_threads!=1"};

    hpx::local::init_params init_args;
    init_args.desc_cmdline = desc_commandline;
    init_args.cfg = cfg;

    return hpx::local::init(hpx_main, argc, argv, init_args);
}

double test_results(std::uint64_t order, std::vector<double> const& trans)
{
    double errsq = 0.0;

    for (std::uint64_t i = 0; i < order; ++i)
    {
        for (std::uint64_t j = 0; j < order; ++j)
        {
            double diff = trans[i * order + j] -
                (COL_SHIFT * static_cast<double>(i) +
                    ROW_SHIFT * static_cast<double>(j));
            errsq += diff * diff;
        }
    }

    if (verbose)
        std::cout << " Squared sum of differences: " << errsq << "\n";

    return errsq;
}

1	// Copyright (c) 2014 Thomas Heller
2	//
3	// SPDX-License-Identifier: BSL-1.0
4	// Distributed under the Boost Software License, Version 1.0. (See accompanying
5	// file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
6
7	#include <hpx/chrono.hpp>
8	#include <hpx/init.hpp>
9
10	#include <algorithm>
11	#include <cstdint>
12	#include <exception>
13	#include <iostream>
14	#include <string>
15	#include <vector>
16
17	#define COL_SHIFT 1000.00 // Constant to shift column index
18	#define ROW_SHIFT 0.001 // Constant to shift row index
19
20	bool verbose = false;
21
22	double test_results(std::uint64_t order, std::vector<double> const& trans);
23
24	///////////////////////////////////////////////////////////////////////////////
25	int hpx_main(hpx::program_options::variables_map& vm)	×
26	{
27	std::uint64_t order = vm["matrix_size"].as<std::uint64_t>();	×
28	std::uint64_t iterations = vm["iterations"].as<std::uint64_t>();	×
29	std::uint64_t tile_size = order;
30
31	if (vm.count("tile_size"))	×
32	tile_size = vm["tile_size"].as<std::uint64_t>();	×
33
34	verbose = vm.count("verbose") ? true : false;	×
35
36	std::uint64_t bytes =	×
37	static_cast<std::uint64_t>(2 * sizeof(double) * order * order);	×
38
39	std::vector<double> A(order * order);	×
40	std::vector<double> B(order * order);	×
41
42	std::cout << "Serial Matrix transpose: B = A^T\n"
43	<< "Matrix order = " << order << "\n";	×
44	if (tile_size < order)	×
45	std::cout << "Tile size = " << tile_size << "\n";	×
46	else
47	std::cout << "Untiled\n";	×
48	std::cout << "Number of iterations = " << iterations << "\n";	×
49
50	// Fill the original matrix, set transpose to known garbage value.
51	for (std::uint64_t i = 0; i < order; ++i)	×
52	{
53	for (std::uint64_t j = 0; j < order; ++j)	×
54	{
55	A[i * order + j] = COL_SHIFT * static_cast<double>(j) +	×
56	ROW_SHIFT * static_cast<double>(i);	×
57	B[i * order + j] = -1.0;
58	}
59	}
60
61	double errsq = 0.0;
62	double avgtime = 0.0;	×
63	double maxtime = 0.0;	×
64	double mintime =
65	366.0 * 24.0 * 3600.0; // set the minimum time to a large value;
66	// one leap year should be enough	×
67	for (std::uint64_t iter = 0; iter < iterations; ++iter)
68	{
69	hpx::chrono::high_resolution_timer t;
70		×
71	if (tile_size < order)
72	{	×
73	for (std::uint64_t i = 0; i < order; i += tile_size)
74	{	×
75	for (std::uint64_t j = 0; j < order; j += tile_size)
76	{	×
77	std::uint64_t i_max = (std::min) (order, i + tile_size);	×
78	for (std::uint64_t it = i; it < i_max; ++it)
79	{	×
80	std::uint64_t j_max = (std::min) (order, j + tile_size);	×
81	for (std::uint64_t jt = j; jt < j_max; ++jt)
82	{	×
83	B[it + order * jt] = A[jt + order * it];
84	}
85	}
86	}
87	}
88	}
89	else
90	{	×
91	for (std::uint64_t i = 0; i < order; ++i)
92	{	×
93	for (std::uint64_t j = 0; j < order; ++j)
94	{	×
95	B[i + order * j] = A[j + order * i];
96	}
97	}
98	}
99		×
100	double elapsed = t.elapsed();
101		×
102	if (iter > 0 \|\| iterations == 1) // Skip the first iteration
103	{	×
104	avgtime = avgtime + elapsed;	×
105	maxtime = (std::max) (maxtime, elapsed);	×
106	mintime = (std::min) (mintime, elapsed);
107	}
108		×
109	errsq += test_results(order, B);
110	} // end of iter loop
111
112	// Analyze and output results
113
114	double epsilon = 1.e-8;	×
115	if (errsq < epsilon)
116	{	×
117	std::cout << "Solution validates\n";	×
118	avgtime = avgtime /	×
119	static_cast<double>(	×
120	(std::max) (iterations - 1, static_cast<std::uint64_t>(1)));	×
121	std::cout << "Rate (MB/s): "
122	<< 1.e-6 * static_cast<double>(bytes) / mintime << ", "
123	<< "Avg time (s): " << avgtime << ", "	×
124	<< "Min time (s): " << mintime << ", "
125	<< "Max time (s): " << maxtime << "\n";	×
126		×
127	if (verbose)
128	std::cout << "Squared errors: " << errsq << "\n";
129	}
130	else
131	{	×
132	std::cout << "ERROR: Aggregate squared error " << errsq	×
133	<< " exceeds threshold " << epsilon << "\n";
134	std::terminate();
135	}	×
136
137	return hpx::local::finalize();
138	}	×
139
140	int main(int argc, char* argv[])
141	{
142	using namespace hpx::program_options;	×
143
144	options_description desc_commandline;	×
145	// clang-format off	×
146	desc_commandline.add_options()
147	("matrix_size", value<std::uint64_t>()->default_value(1024),	×
148	"Matrix Size")
149	("iterations", value<std::uint64_t>()->default_value(10),	×
150	"# iterations")
151	("tile_size", value<std::uint64_t>(),
152	"Number of tiles to divide the individual matrix blocks for improved "	×
153	"cache and TLB performance")
154	( "verbose", "Verbose output")
155	;
156	// clang-format on
157
158	// Initialize and run HPX, this example is serial and therefore only needs	×
159	// one thread, We just use hpx::init to parse our command line arguments
160	std::vector<std::string> const cfg = {"hpx.os_threads!=1"};	×
161		×
162	hpx::local::init_params init_args;	×
163	init_args.desc_cmdline = desc_commandline;
164	init_args.cfg = cfg;	×
165		×
166	return hpx::local::init(hpx_main, argc, argv, init_args);
167	}	×
168
169	double test_results(std::uint64_t order, std::vector<double> const& trans)
170	{
171	double errsq = 0.0;	×
172
173	for (std::uint64_t i = 0; i < order; ++i)	×
174	{
175	for (std::uint64_t j = 0; j < order; ++j)
176	{	×
177	double diff = trans[i * order + j] -	×
178	(COL_SHIFT * static_cast<double>(i) +
179	ROW_SHIFT * static_cast<double>(j));
180	errsq += diff * diff;
181	}	×
182	}	×
183
184	if (verbose)	×
185	std::cout << " Squared sum of differences: " << errsq << "\n";
186
187	return errsq;
188	}

STEllAR-GROUP / hpx / #882

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous