1 //=================================================================================================
2 /*!
3 //  \file src/main/DVecSVecMult.cpp
4 //  \brief Source file for the dense vector/sparse vector multiplication benchmark
5 //
6 //  Copyright (C) 2012-2020 Klaus Iglberger - All Rights Reserved
7 //
8 //  This file is part of the Blaze library. You can redistribute it and/or modify it under
9 //  the terms of the New (Revised) BSD License. Redistribution and use in source and binary
10 //  forms, with or without modification, are permitted provided that the following conditions
11 //  are met:
12 //
13 //  1. Redistributions of source code must retain the above copyright notice, this list of
14 //     conditions and the following disclaimer.
15 //  2. Redistributions in binary form must reproduce the above copyright notice, this list
16 //     of conditions and the following disclaimer in the documentation and/or other materials
17 //     provided with the distribution.
18 //  3. Neither the names of the Blaze development group nor the names of its contributors
19 //     may be used to endorse or promote products derived from this software without specific
20 //     prior written permission.
21 //
22 //  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
23 //  EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 //  OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT
25 //  SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 //  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
27 //  TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 //  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29 //  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30 //  ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
31 //  DAMAGE.
32 */
33 //=================================================================================================
34 
35 
36 //*************************************************************************************************
37 // Includes
38 //*************************************************************************************************
39 
40 #include <algorithm>
41 #include <cstdlib>
42 #include <iostream>
43 #include <stdexcept>
44 #include <string>
45 #include <vector>
46 #include <blaze/math/CompressedVector.h>
47 #include <blaze/math/DynamicVector.h>
48 #include <blaze/math/Infinity.h>
49 #include <blaze/util/algorithms/Max.h>
50 #include <blaze/util/Random.h>
51 #include <blaze/util/Timing.h>
52 #include <blazemark/blaze/DVecSVecMult.h>
53 #include <blazemark/blaze/init/CompressedVector.h>
54 #include <blazemark/blaze/init/DynamicVector.h>
55 #include <blazemark/boost/DVecSVecMult.h>
56 #include <blazemark/system/Boost.h>
57 #include <blazemark/system/Config.h>
58 #include <blazemark/system/Types.h>
59 #include <blazemark/util/Benchmarks.h>
60 #include <blazemark/util/DynamicSparseRun.h>
61 #include <blazemark/util/Parser.h>
62 
63 #ifdef BLAZE_USE_HPX_THREADS
64 #  include <hpx/hpx_main.hpp>
65 #endif
66 
67 
68 //*************************************************************************************************
69 // Using declarations
70 //*************************************************************************************************
71 
72 using blazemark::Benchmarks;
73 using blazemark::DynamicSparseRun;
74 using blazemark::Parser;
75 
76 
77 
78 
79 //=================================================================================================
80 //
81 //  TYPE DEFINITIONS
82 //
83 //=================================================================================================
84 
85 //*************************************************************************************************
86 /*!\brief Type of a benchmark run.
87 //
88 // This type definition specifies the type of a single benchmark run for the dense vector/sparse
89 // vector multiplication benchmark.
90 */
91 using Run = DynamicSparseRun;
92 //*************************************************************************************************
93 
94 
95 
96 
97 //=================================================================================================
98 //
99 //  UTILITY FUNCTIONS
100 //
101 //=================================================================================================
102 
103 //*************************************************************************************************
104 /*!\brief Estimating the necessary number of steps for each benchmark.
105 //
106 // \param run The parameters for the benchmark run.
107 // \return void
108 //
109 // This function estimates the necessary number of steps for the given benchmark based on the
110 // performance of the Blaze library.
111 */
estimateSteps(Run & run)112 void estimateSteps( Run& run )
113 {
114    using blazemark::element_t;
115    using blaze::columnVector;
116 
117    ::blaze::setSeed( ::blazemark::seed );
118 
119    const size_t N( run.getSize() );
120    const size_t F( run.getNonZeros() );
121 
122    blaze::DynamicVector<element_t,columnVector> a( N );
123    blaze::CompressedVector<element_t,columnVector> b( N, F ), c( N );
124    blaze::timing::WcTimer timer;
125    double wct( 0.0 );
126    size_t steps( 1UL );
127 
128    blazemark::blaze::init( a );
129    blazemark::blaze::init( b, F );
130 
131    while( true ) {
132       timer.start();
133       for( size_t i=0UL; i<steps; ++i ) {
134          c = a * b;
135       }
136       timer.end();
137       wct = timer.last();
138       if( wct >= 0.2 ) break;
139       steps *= 2UL;
140    }
141 
142    if( c.size() != N )
143       std::cerr << " Line " << __LINE__ << ": ERROR detected!!!\n";
144 
145    const size_t estimatedSteps( ( blazemark::runtime * steps ) / timer.last() );
146    run.setSteps( blaze::max( 1UL, estimatedSteps ) );
147 }
148 //*************************************************************************************************
149 
150 
151 //*************************************************************************************************
152 /*!\brief Estimating the necessary number of floating point operations.
153 //
154 // \param run The parameters for the benchmark run.
155 // \return void
156 //
157 // This function estimates the number of floating point operations required for a single
158 // computation of the (composite) arithmetic operation.
159 */
estimateFlops(Run & run)160 void estimateFlops( Run& run )
161 {
162    const size_t F( run.getNonZeros() );
163 
164    run.setFlops( F );
165 }
166 //*************************************************************************************************
167 
168 
169 
170 
171 //=================================================================================================
172 //
173 //  BENCHMARK FUNCTIONS
174 //
175 //=================================================================================================
176 
177 //*************************************************************************************************
178 /*!\brief Dense vector/sparse vector multiplication benchmark function.
179 //
180 // \param runs The specified benchmark runs.
181 // \param benchmarks The selection of benchmarks.
182 // \return void
183 */
dvecsvecmult(std::vector<Run> & runs,Benchmarks benchmarks)184 void dvecsvecmult( std::vector<Run>& runs, Benchmarks benchmarks )
185 {
186    std::cout << std::left;
187 
188    std::sort( runs.begin(), runs.end() );
189 
190    size_t slowSize( blaze::inf );
191    for( std::vector<Run>::iterator run=runs.begin(); run!=runs.end(); ++run )
192    {
193       estimateFlops( *run );
194 
195       if( run->getSteps() == 0UL ) {
196          if( run->getSize() < slowSize ) {
197             estimateSteps( *run );
198             if( run->getSteps() == 1UL )
199                slowSize = run->getSize();
200          }
201          else run->setSteps( 1UL );
202       }
203    }
204 
205    if( benchmarks.runBlaze ) {
206       std::vector<Run>::iterator run=runs.begin();
207       while( run != runs.end() ) {
208          const float fill( run->getFillingDegree() );
209          std::cout << "   Blaze (" << fill << "% filled) [MFlop/s]:\n";
210          for( ; run!=runs.end(); ++run ) {
211             if( run->getFillingDegree() != fill ) break;
212             const size_t N    ( run->getSize()     );
213             const size_t F    ( run->getNonZeros() );
214             const size_t steps( run->getSteps()    );
215             run->setBlazeResult( blazemark::blaze::dvecsvecmult( N, F, steps ) );
216             const double mflops( run->getFlops() * steps / run->getBlazeResult() / 1E6 );
217             std::cout << "     " << std::setw(12) << N << mflops << std::endl;
218          }
219       }
220    }
221 
222 #if BLAZEMARK_BOOST_MODE
223    if( benchmarks.runBoost ) {
224       std::vector<Run>::iterator run=runs.begin();
225       while( run != runs.end() ) {
226          const float fill( run->getFillingDegree() );
227          std::cout << "   Boost uBLAS (" << fill << "% filled) [MFlop/s]:\n";
228          for( ; run!=runs.end(); ++run ) {
229             if( run->getFillingDegree() != fill ) break;
230             const size_t N    ( run->getSize()     );
231             const size_t F    ( run->getNonZeros() );
232             const size_t steps( run->getSteps()    );
233             run->setBoostResult( blazemark::boost::dvecsvecmult( N, F, steps ) );
234             const double mflops( run->getFlops() * steps / run->getBoostResult() / 1E6 );
235             std::cout << "     " << std::setw(12) << N << mflops << std::endl;
236          }
237       }
238    }
239 #endif
240 
241    for( std::vector<Run>::iterator run=runs.begin(); run!=runs.end(); ++run ) {
242       std::cout << *run;
243    }
244 }
245 //*************************************************************************************************
246 
247 
248 
249 
250 //=================================================================================================
251 //
252 //  MAIN FUNCTION
253 //
254 //=================================================================================================
255 
256 //*************************************************************************************************
257 /*!\brief The main function for the dense vector/sparse vector multiplication benchmark.
258 //
259 // \param argc The total number of command line arguments.
260 // \param argv The array of command line arguments.
261 // \return void
262 */
main(int argc,char ** argv)263 int main( int argc, char** argv )
264 {
265    std::cout << "\n Dense Vector/Sparse Vector Multiplication:\n";
266 
267    Benchmarks benchmarks;
268 
269    try {
270       parseCommandLineArguments( argc, argv, benchmarks );
271    }
272    catch( std::exception& ex ) {
273       std::cerr << "   " << ex.what() << "\n";
274       return EXIT_FAILURE;
275    }
276 
277    const std::string installPath( INSTALL_PATH );
278    const std::string parameterFile( installPath + "/params/dvecsvecmult.prm" );
279    Parser<Run> parser;
280    std::vector<Run> runs;
281 
282    try {
283       parser.parse( parameterFile.c_str(), runs );
284    }
285    catch( std::exception& ex ) {
286       std::cerr << "   Error during parameter extraction: " << ex.what() << "\n";
287       return EXIT_FAILURE;
288    }
289 
290    try {
291       dvecsvecmult( runs, benchmarks );
292    }
293    catch( std::exception& ex ) {
294       std::cerr << "   Error during benchmark execution: " << ex.what() << "\n";
295       return EXIT_FAILURE;
296    }
297 
298    return EXIT_SUCCESS;
299 }
300 //*************************************************************************************************
301