001/* ***** BEGIN LICENSE BLOCK ***** 002 * Version: MPL 1.1/GPL 2.0/LGPL 2.1 003 * 004 * The contents of this file are subject to the Mozilla Public License Version 005 * 1.1 (the "License"); you may not use this file except in compliance with 006 * the License. You may obtain a copy of the License at 007 * http://www.mozilla.org/MPL/ 008 * 009 * Software distributed under the License is distributed on an "AS IS" basis, 010 * WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License 011 * for the specific language governing rights and limitations under the 012 * License. 013 * 014 * The Original Code is JTransforms. 015 * 016 * The Initial Developer of the Original Code is 017 * Piotr Wendykier, Emory University. 018 * Portions created by the Initial Developer are Copyright (C) 2007-2009 019 * the Initial Developer. All Rights Reserved. 020 * 021 * Alternatively, the contents of this file may be used under the terms of 022 * either the GNU General Public License Version 2 or later (the "GPL"), or 023 * the GNU Lesser General Public License Version 2.1 or later (the "LGPL"), 024 * in which case the provisions of the GPL or the LGPL are applicable instead 025 * of those above. If you wish to allow use of your version of this file only 026 * under the terms of either the GPL or the LGPL, and not to allow others to 027 * use your version of this file under the terms of the MPL, indicate your 028 * decision by deleting the provisions above and replace them with the notice 029 * and other provisions required by the GPL or the LGPL. If you do not delete 030 * the provisions above, a recipient may use your version of this file under 031 * the terms of any one of the MPL, the GPL or the LGPL. 032 * 033 * ***** END LICENSE BLOCK ***** */ 034 035package edu.emory.mathcs.jtransforms.dst; 036 037import java.util.Arrays; 038 039import edu.emory.mathcs.utils.ConcurrencyUtils; 040import edu.emory.mathcs.utils.IOUtils; 041 042/** 043 * Benchmark of single precision DST's 044 * 045 * @author Piotr Wendykier (piotr.wendykier@gmail.com) 046 * 047 */ 048@SuppressWarnings("javadoc") 049public class BenchmarkFloatDST { 050 051 private static int nthread = 8; 052 053 private static int niter = 200; 054 055 private static int nsize = 16; 056 057 private static int threadsBegin2D = 65636; 058 059 private static int threadsBegin3D = 65636; 060 061 private static boolean doWarmup = true; 062 063 private static int[] sizes1D = new int[] { 65536, 131072, 262144, 524288, 1048576, 2097152, 4194304, 8388608, 10368, 064 27000, 75600, 165375, 362880, 1562500, 3211264, 6250000 }; 065 066 private static int[] sizes2D = new int[] { 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 260, 520, 1050, 1458, 1960, 067 2916, 4116, 5832 }; 068 069 private static int[] sizes3D = new int[] { 8, 16, 32, 64, 128, 256, 512, 1024, 5, 17, 30, 95, 180, 270, 324, 420 }; 070 071 private static boolean doScaling = false; 072 073 private BenchmarkFloatDST() { 074 075 } 076 077 public static void parseArguments(String[] args) { 078 if (args.length > 0) { 079 nthread = Integer.parseInt(args[0]); 080 threadsBegin2D = Integer.parseInt(args[1]); 081 threadsBegin3D = Integer.parseInt(args[2]); 082 niter = Integer.parseInt(args[3]); 083 doWarmup = Boolean.parseBoolean(args[4]); 084 doScaling = Boolean.parseBoolean(args[5]); 085 nsize = Integer.parseInt(args[6]); 086 sizes1D = new int[nsize]; 087 sizes2D = new int[nsize]; 088 sizes3D = new int[nsize]; 089 for (int i = 0; i < nsize; i++) { 090 sizes1D[i] = Integer.parseInt(args[7 + i]); 091 } 092 for (int i = 0; i < nsize; i++) { 093 sizes2D[i] = Integer.parseInt(args[7 + nsize + i]); 094 } 095 for (int i = 0; i < nsize; i++) { 096 sizes3D[i] = Integer.parseInt(args[7 + nsize + nsize + i]); 097 } 098 } else { 099 System.out.println("Default settings are used."); 100 } 101 ConcurrencyUtils.setNumberOfThreads(nthread); 102 ConcurrencyUtils.setThreadsBeginN_2D(threadsBegin2D); 103 ConcurrencyUtils.setThreadsBeginN_3D(threadsBegin3D); 104 System.out.println("nthred = " + nthread); 105 System.out.println("threadsBegin2D = " + threadsBegin2D); 106 System.out.println("threadsBegin3D = " + threadsBegin3D); 107 System.out.println("niter = " + niter); 108 System.out.println("doWarmup = " + doWarmup); 109 System.out.println("doScaling = " + doScaling); 110 System.out.println("nsize = " + nsize); 111 System.out.println("sizes1D[] = " + Arrays.toString(sizes1D)); 112 System.out.println("sizes2D[] = " + Arrays.toString(sizes2D)); 113 System.out.println("sizes3D[] = " + Arrays.toString(sizes3D)); 114 } 115 116 public static void benchmarkForward_1D() { 117 final double[] times = new double[nsize]; 118 float[] x; 119 for (int i = 0; i < nsize; i++) { 120 System.out.println("Forward DST 1D of size " + sizes1D[i]); 121 FloatDST_1D dst = new FloatDST_1D(sizes1D[i]); 122 x = new float[sizes1D[i]]; 123 if (doWarmup) { // call the transform twice to warm up 124 IOUtils.fillMatrix_1D(sizes1D[i], x); 125 dst.forward(x, doScaling); 126 IOUtils.fillMatrix_1D(sizes1D[i], x); 127 dst.forward(x, doScaling); 128 } 129 float av_time = 0; 130 long elapsedTime = 0; 131 for (int j = 0; j < niter; j++) { 132 IOUtils.fillMatrix_1D(sizes1D[i], x); 133 elapsedTime = System.nanoTime(); 134 dst.forward(x, doScaling); 135 elapsedTime = System.nanoTime() - elapsedTime; 136 av_time = av_time + elapsedTime; 137 } 138 times[i] = av_time / 1000000.0 / niter; 139 System.out.println("Average execution time: " + String.format("%.2f", av_time / 1000000.0 / niter) + " msec"); 140 x = null; 141 dst = null; 142 System.gc(); 143 ConcurrencyUtils.sleep(5000); 144 } 145 IOUtils.writeFFTBenchmarkResultsToFile("benchmarkFloatForwardDST_1D.txt", nthread, niter, doWarmup, doScaling, 146 sizes1D, times); 147 } 148 149 public static void benchmarkForward_2D_input_1D() { 150 final double[] times = new double[nsize]; 151 float[] x; 152 for (int i = 0; i < nsize; i++) { 153 System.out.println("Forward DST 2D (input 1D) of size " + sizes2D[i] + " x " + sizes2D[i]); 154 FloatDST_2D dst2 = new FloatDST_2D(sizes2D[i], sizes2D[i]); 155 x = new float[sizes2D[i] * sizes2D[i]]; 156 if (doWarmup) { // call the transform twice to warm up 157 IOUtils.fillMatrix_2D(sizes2D[i], sizes2D[i], x); 158 dst2.forward(x, doScaling); 159 IOUtils.fillMatrix_2D(sizes2D[i], sizes2D[i], x); 160 dst2.forward(x, doScaling); 161 } 162 float av_time = 0; 163 long elapsedTime = 0; 164 for (int j = 0; j < niter; j++) { 165 IOUtils.fillMatrix_2D(sizes2D[i], sizes2D[i], x); 166 elapsedTime = System.nanoTime(); 167 dst2.forward(x, doScaling); 168 elapsedTime = System.nanoTime() - elapsedTime; 169 av_time = av_time + elapsedTime; 170 } 171 times[i] = av_time / 1000000.0 / niter; 172 System.out.println("Average execution time: " + String.format("%.2f", av_time / 1000000.0 / niter) + " msec"); 173 x = null; 174 dst2 = null; 175 System.gc(); 176 ConcurrencyUtils.sleep(5000); 177 } 178 IOUtils.writeFFTBenchmarkResultsToFile("benchmarkFloatForwardDST_2D_input_1D.txt", nthread, niter, doWarmup, 179 doScaling, sizes2D, times); 180 181 } 182 183 public static void benchmarkForward_2D_input_2D() { 184 final double[] times = new double[nsize]; 185 float[][] x; 186 for (int i = 0; i < nsize; i++) { 187 System.out.println("Forward DST 2D (input 2D) of size " + sizes2D[i] + " x " + sizes2D[i]); 188 FloatDST_2D dst2 = new FloatDST_2D(sizes2D[i], sizes2D[i]); 189 x = new float[sizes2D[i]][sizes2D[i]]; 190 if (doWarmup) { // call the transform twice to warm up 191 IOUtils.fillMatrix_2D(sizes2D[i], sizes2D[i], x); 192 dst2.forward(x, doScaling); 193 IOUtils.fillMatrix_2D(sizes2D[i], sizes2D[i], x); 194 dst2.forward(x, doScaling); 195 } 196 float av_time = 0; 197 long elapsedTime = 0; 198 for (int j = 0; j < niter; j++) { 199 IOUtils.fillMatrix_2D(sizes2D[i], sizes2D[i], x); 200 elapsedTime = System.nanoTime(); 201 dst2.forward(x, doScaling); 202 elapsedTime = System.nanoTime() - elapsedTime; 203 av_time = av_time + elapsedTime; 204 } 205 times[i] = av_time / 1000000.0 / niter; 206 System.out.println("Average execution time: " + String.format("%.2f", av_time / 1000000.0 / niter) + " msec"); 207 x = null; 208 dst2 = null; 209 System.gc(); 210 ConcurrencyUtils.sleep(5000); 211 } 212 IOUtils.writeFFTBenchmarkResultsToFile("benchmarkFloatForwardDST_2D_input_2D.txt", nthread, niter, doWarmup, 213 doScaling, sizes2D, times); 214 215 } 216 217 public static void benchmarkForward_3D_input_1D() { 218 final double[] times = new double[nsize]; 219 float[] x; 220 for (int i = 0; i < nsize; i++) { 221 System.out.println("Forward DST 3D (input 1D) of size " + sizes3D[i] + " x " + sizes3D[i] + " x " 222 + sizes3D[i]); 223 FloatDST_3D dst3 = new FloatDST_3D(sizes3D[i], sizes3D[i], sizes3D[i]); 224 x = new float[sizes3D[i] * sizes3D[i] * sizes3D[i]]; 225 if (doWarmup) { // call the transform twice to warm up 226 IOUtils.fillMatrix_3D(sizes3D[i], sizes3D[i], sizes3D[i], x); 227 dst3.forward(x, doScaling); 228 IOUtils.fillMatrix_3D(sizes3D[i], sizes3D[i], sizes3D[i], x); 229 dst3.forward(x, doScaling); 230 } 231 float av_time = 0; 232 long elapsedTime = 0; 233 for (int j = 0; j < niter; j++) { 234 IOUtils.fillMatrix_3D(sizes3D[i], sizes3D[i], sizes3D[i], x); 235 elapsedTime = System.nanoTime(); 236 dst3.forward(x, doScaling); 237 elapsedTime = System.nanoTime() - elapsedTime; 238 av_time = av_time + elapsedTime; 239 } 240 times[i] = av_time / 1000000.0 / niter; 241 System.out.println("Average execution time: " + String.format("%.2f", av_time / 1000000.0 / niter) + " msec"); 242 x = null; 243 dst3 = null; 244 System.gc(); 245 ConcurrencyUtils.sleep(5000); 246 } 247 IOUtils.writeFFTBenchmarkResultsToFile("benchmarkFloatForwardDST_3D_input_1D.txt", nthread, niter, doWarmup, 248 doScaling, sizes3D, times); 249 250 } 251 252 public static void benchmarkForward_3D_input_3D() { 253 final double[] times = new double[nsize]; 254 float[][][] x; 255 for (int i = 0; i < nsize; i++) { 256 System.out.println("Forward DST 3D (input 3D) of size " + sizes3D[i] + " x " + sizes3D[i] + " x " 257 + sizes3D[i]); 258 FloatDST_3D dst3 = new FloatDST_3D(sizes3D[i], sizes3D[i], sizes3D[i]); 259 x = new float[sizes3D[i]][sizes3D[i]][sizes3D[i]]; 260 if (doWarmup) { // call the transform twice to warm up 261 IOUtils.fillMatrix_3D(sizes3D[i], sizes3D[i], sizes3D[i], x); 262 dst3.forward(x, doScaling); 263 IOUtils.fillMatrix_3D(sizes3D[i], sizes3D[i], sizes3D[i], x); 264 dst3.forward(x, doScaling); 265 } 266 float av_time = 0; 267 long elapsedTime = 0; 268 for (int j = 0; j < niter; j++) { 269 IOUtils.fillMatrix_3D(sizes3D[i], sizes3D[i], sizes3D[i], x); 270 elapsedTime = System.nanoTime(); 271 dst3.forward(x, doScaling); 272 elapsedTime = System.nanoTime() - elapsedTime; 273 av_time = av_time + elapsedTime; 274 } 275 times[i] = av_time / 1000000.0 / (niter); 276 System.out.println("Average execution time: " + String.format("%.2f", av_time / 1000000.0 / (niter)) 277 + " msec"); 278 x = null; 279 dst3 = null; 280 System.gc(); 281 ConcurrencyUtils.sleep(5000); 282 } 283 IOUtils.writeFFTBenchmarkResultsToFile("benchmarkFloatForwardDST_3D_input_3D.txt", nthread, niter, doWarmup, 284 doScaling, sizes3D, times); 285 } 286 287 public static void main(String[] args) { 288 parseArguments(args); 289 benchmarkForward_1D(); 290 benchmarkForward_2D_input_1D(); 291 benchmarkForward_2D_input_2D(); 292 benchmarkForward_3D_input_1D(); 293 benchmarkForward_3D_input_3D(); 294 System.exit(0); 295 296 } 297}