1 #!/usr/local/bin/perl 2 # * 2016 and later: Unicode, Inc. and others. 3 # * License & terms of use: http://www.unicode.org/copyright.html#License 4 # ******************************************************************************* 5 # * Copyright (C) 2002-2007 International Business Machines Corporation and * 6 # * others. All Rights Reserved. * 7 # ******************************************************************************* 8 9 use strict; 10 11 # Assume we are running within the icu4j root directory 12 use lib 'src/com/ibm/icu/dev/test/perf'; 13 use Dataset; 14 15 #--------------------------------------------------------------------- 16 # Test class 17 my $TESTCLASS = 'com.ibm.icu.dev.test.perf.NormalizerPerformanceTest'; 18 19 # Methods to be tested. Each pair represents a test method and 20 # a baseline method which is used for comparison. 21 my @METHODS = ( 22 ['TestJDK_NFD_NFC_Text', 'TestICU_NFD_NFC_Text'], 23 ['TestJDK_NFC_NFC_Text', 'TestICU_NFC_NFC_Text'], 24 # ['TestJDK_NFC_NFD_Text', 'TestICU_NFC_NFD_Text'], 25 ['TestJDK_NFC_Orig_Text', 'TestICU_NFC_Orig_Text'], 26 ['TestJDK_NFD_NFC_Text', 'TestICU_NFD_NFC_Text'], 27 ['TestJDK_NFD_NFD_Text', 'TestICU_NFD_NFD_Text'], 28 ['TestJDK_NFD_Orig_Text', 'TestICU_NFD_Orig_Text'], 29 ); 30 31 # Patterns which define the set of characters used for testing. 32 33 my $SOURCEDIR ="src/com/ibm/icu/dev/test/perf/data/collation/"; 34 35 my @OPTIONS = ( 36 # src text src encoding mode 37 [ "TestNames_SerbianSH.txt", "UTF-8", "b"], 38 # [ "arabic.txt", "UTF-8", "b"], 39 # [ "french.txt", "UTF-8", "b"], 40 # [ "greek.txt", "UTF-8", "b"], 41 # [ "hebrew.txt", "UTF-8", "b"], 42 # [ "hindi.txt" , "UTF-8", "b"], 43 # [ "japanese.txt", "UTF-8", "b"], 44 # [ "korean.txt", "UTF-8", "b"], 45 # [ "s-chinese.txt", "UTF-8", "b"], 46 # [ "french.txt", "UTF-8", "b"], 47 # [ "greek.txt", "UTF-8", "b"], 48 # [ "hebrew.txt", "UTF-8", "b"], 49 # [ "hindi.txt" , "UTF-8", "b"], 50 # [ "japanese.txt", "UTF-8", "b"], 51 # [ "korean.txt", "UTF-8", "b"], 52 # [ "s-chinese.txt", "UTF-8", "b"], 53 # [ "arabic.html", "UTF-8", "b"], 54 # [ "czech.html", "UTF-8", "b"], 55 # [ "danish.html", "UTF-8", "b"], 56 # [ "english.html", "UTF-8", "b"], 57 # [ "esperanto.html", "UTF-8", "b"], 58 # [ "french.html", "UTF-8", "b"], 59 # [ "georgian.html", "UTF-8", "b"], 60 # [ "german.html", "UTF-8", "b"], 61 # [ "greek.html", "UTF-8", "b"], 62 # [ "hebrew.html", "UTF-8", "b"], 63 # [ "hindi.html", "UTF-8", "b"], 64 # [ "icelandic.html", "UTF-8", "b"], 65 # [ "interlingua.html", "UTF-8", "b"], 66 # [ "italian.html", "UTF-8", "b"], 67 # [ "japanese.html", "UTF-8", "b"], 68 # [ "korean.html", "UTF-8", "b"], 69 # [ "lithuanian.html", "UTF-8", "b"], 70 # [ "maltese.html", "UTF-8", "b"], 71 # [ "persian.html", "UTF-8", "b"], 72 # [ "polish.html", "UTF-8", "b"], 73 # [ "portuguese.html", "UTF-8", "b"], 74 # [ "romanian.html", "UTF-8", "b"], 75 # [ "russian.html", "UTF-8", "b"], 76 # [ "s-chinese.html", "UTF-8", "b"], 77 # [ "spanish.html", "UTF-8", "b"], 78 # [ "swedish.html", "UTF-8", "b"], 79 # [ "t-chinese.html", "UTF-8", "b"], 80 # [ "welsh.html", "UTF-8", "b"], 81 [ "TestNames_Asian.txt", "UTF-8", "l"], 82 [ "TestNames_Chinese.txt", "UTF-8", "l"], 83 [ "TestNames_Japanese.txt", "UTF-8", "l"], 84 [ "TestNames_Japanese_h.txt", "UTF-8", "l"], 85 [ "TestNames_Japanese_k.txt", "UTF-8", "l"], 86 [ "TestNames_Korean.txt", "UTF-8", "l"], 87 [ "TestNames_Latin.txt", "UTF-8", "l"], 88 [ "TestNames_SerbianSH.txt", "UTF-8", "l"], 89 [ "TestNames_SerbianSR.txt", "UTF-8", "l"], 90 [ "TestNames_Thai.txt", "UTF-8", "l"], 91 [ "Testnames_Russian.txt", "UTF-8", "l"], 92 ); 93 94 my $CALIBRATE = 2; # duration in seconds for initial calibration 95 my $DURATION = 10; # duration in seconds for each pass 96 my $NUMPASSES = 4; # number of passes. If > 1 then the first pass 97 # is discarded as a JIT warm-up pass. 98 99 my $TABLEATTR = 'BORDER="1" CELLPADDING="4" CELLSPACING="0"'; 100 101 my $PLUS_MINUS = "±"; 102 103 if ($NUMPASSES < 3) { 104 die "Need at least 3 passes. One is discarded (JIT warmup) and need two to have 1 degree of freedom (t distribution)."; 105 } 106 107 my $OUT; # see out() 108 109 main(); 110 111 #--------------------------------------------------------------------- 112 # ... 113 sub main { 114 my $date = localtime; 115 my $title = "ICU4J Performance Test $date"; 116 117 my $html = $date; 118 $html =~ s/://g; # ':' illegal 119 $html =~ s/\s*\d+$//; # delete year 120 $html =~ s/^\w+\s*//; # delete dow 121 $html = "perf $html.html"; 122 123 open(HTML,">$html") or die "Can't write to $html: $!"; 124 125 print HTML <<EOF; 126 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 127 "http://www.w3.org/TR/html4/strict.dtd"> 128 <HTML> 129 <HEAD> 130 <TITLE>$title</TITLE> 131 </HEAD> 132 <BODY> 133 EOF 134 print HTML "<H1>$title</H1>\n"; 135 136 print HTML "<H2>$TESTCLASS</H2>\n"; 137 138 my $raw = ""; 139 140 for my $methodPair (@METHODS) { 141 142 my $testMethod = $methodPair->[0]; 143 my $baselineMethod = $methodPair->[1]; 144 145 print HTML "<P><TABLE $TABLEATTR><TR><TD>\n"; 146 print HTML "<P><B>$testMethod vs. $baselineMethod</B></P>\n"; 147 148 print HTML "<P><TABLE $TABLEATTR BGCOLOR=\"#CCFFFF\">\n"; 149 print HTML "<TR><TD>Options</TD><TD>$testMethod</TD>"; 150 print HTML "<TD>$baselineMethod</TD><TD>Ratio</TD></TR>\n"; 151 152 $OUT = ''; 153 154 for my $pat (@OPTIONS) { 155 print HTML "<TR><TD>@$pat[0], @$pat[2]</TD>\n"; 156 157 out("<P><TABLE $TABLEATTR WIDTH=\"100%\">"); 158 159 # measure the test method 160 out("<TR><TD>"); 161 print "\n$testMethod [@$pat]\n"; 162 my $t = measure2($testMethod, $pat, -$DURATION); 163 out("</TD></TR>"); 164 print HTML "<TD>", formatSeconds(4, $t->getMean(), $t->getError); 165 print HTML "/event</TD>\n"; 166 167 # measure baseline method 168 out("<TR><TD>"); 169 print "\n$baselineMethod [@$pat]\n"; 170 my $b = measure2($baselineMethod, $pat, -$DURATION); 171 out("</TD></TR>"); 172 print HTML "<TD>", formatSeconds(4, $b->getMean(), $t->getError); 173 print HTML "/event</TD>\n"; 174 175 out("</TABLE></P>"); 176 177 # output ratio 178 my $r = $t->divide($b); 179 my $mean = $r->getMean() - 1; 180 my $color = $mean < 0 ? "RED" : "BLACK"; 181 print HTML "<TD><B><FONT COLOR=\"$color\">", formatPercent(3, $mean, $r->getError); 182 print HTML "</FONT></B></TD></TR>\n"; 183 } 184 185 print HTML "</TABLE></P>\n"; 186 187 print HTML "<P>Raw data:</P>\n"; 188 print HTML $OUT; 189 print HTML "</TABLE></P>\n"; 190 } 191 192 print HTML <<EOF; 193 </BODY> 194 </HTML> 195 EOF 196 close(HTML) or die "Can't close $html: $!"; 197 } 198 199 #--------------------------------------------------------------------- 200 # Append text to the global variable $OUT 201 sub out { 202 $OUT .= join('', @_); 203 } 204 205 #--------------------------------------------------------------------- 206 # Append text to the global variable $OUT 207 sub outln { 208 $OUT .= join('', @_) . "\n"; 209 } 210 211 #--------------------------------------------------------------------- 212 # Measure a given test method with a give test pattern using the 213 # global run parameters. 214 # 215 # @param the method to run 216 # @param the pattern defining characters to test 217 # @param if >0 then the number of iterations per pass. If <0 then 218 # (negative of) the number of seconds per pass. 219 # 220 # @return a Dataset object, scaled by iterations per pass and 221 # events per iteration, to give time per event 222 # 223 sub measure2 { 224 my @data = measure1(@_); 225 my $iterPerPass = shift(@data); 226 my $eventPerIter = shift(@data); 227 228 shift(@data) if (@data > 1); # discard first run 229 230 my $ds = Dataset->new(@data); 231 $ds->setScale(1.0e-3 / ($iterPerPass * $eventPerIter)); 232 $ds; 233 } 234 235 #--------------------------------------------------------------------- 236 # Measure a given test method with a give test pattern using the 237 # global run parameters. 238 # 239 # @param the method to run 240 # @param the pattern defining characters to test 241 # @param if >0 then the number of iterations per pass. If <0 then 242 # (negative of) the number of seconds per pass. 243 # 244 # @return array of: 245 # [0] iterations per pass 246 # [1] events per iteration 247 # [2..] ms reported for each pass, in order 248 # 249 sub measure1 { 250 my $method = shift; 251 my $pat = shift; 252 my $iterCount = shift; # actually might be -seconds/pass 253 254 out("<P>Measuring $method for input file @$pat[0] in @$pat[2] , "); 255 if ($iterCount > 0) { 256 out("$iterCount iterations/pass, $NUMPASSES passes</P>\n"); 257 } else { 258 out(-$iterCount, " seconds/pass, $NUMPASSES passes</P>\n"); 259 } 260 261 # is $iterCount actually -seconds/pass? 262 if ($iterCount < 0) { 263 264 # calibrate: estimate ms/iteration 265 print "Calibrating..."; 266 my @t = callJava($method, $pat, -$CALIBRATE, 1); 267 print "done.\n"; 268 269 my @data = split(/\s+/, $t[0]->[2]); 270 $data[0] *= 1.0e+3; 271 272 my $timePerIter = 1.0e-3 * $data[0] / $data[1]; 273 274 # determine iterations/pass 275 $iterCount = int(-$iterCount / $timePerIter + 0.5); 276 277 out("<P>Calibration pass ($CALIBRATE sec): "); 278 out("$data[0] ms, "); 279 out("$data[1] iterations = "); 280 out(formatSeconds(4, $timePerIter), "/iteration<BR>\n"); 281 } 282 283 # run passes 284 print "Measuring $iterCount iterations x $NUMPASSES passes..."; 285 my @t = callJava($method, $pat, $iterCount, $NUMPASSES); 286 print "done.\n"; 287 my @ms = (); 288 my @b; # scratch 289 for my $a (@t) { 290 # $a->[0]: method name, corresponds to $method 291 # $a->[1]: 'begin' data, == $iterCount 292 # $a->[2]: 'end' data, of the form <ms> <loops> <eventsPerIter> 293 # $a->[3...]: gc messages from JVM during pass 294 @b = split(/\s+/, $a->[2]); 295 push(@ms, $b[0] * 1.0e+3); 296 } 297 my $eventsPerIter = $b[2]; 298 299 out("Iterations per pass: $iterCount<BR>\n"); 300 out("Events per iteration: $eventsPerIter<BR>\n"); 301 302 my @ms_str = @ms; 303 $ms_str[0] .= " (discarded)" if (@ms_str > 1); 304 out("Raw times (ms/pass): ", join(", ", @ms_str), "<BR>\n"); 305 306 ($iterCount, $eventsPerIter, @ms); 307 } 308 309 #--------------------------------------------------------------------- 310 # Invoke java to run $TESTCLASS, passing it the given parameters. 311 # 312 # @param the method to run 313 # @param the number of iterations, or if negative, the duration 314 # in seconds. If more than on pass is desired, pass in 315 # a string, e.g., "100 100 100". 316 # @param the pattern defining characters to test 317 # 318 # @return an array of results. Each result is an array REF 319 # describing one pass. The array REF contains: 320 # ->[0]: The method name as reported 321 # ->[1]: The params on the '= <meth> begin ...' line 322 # ->[2]: The params on the '= <meth> end ...' line 323 # ->[3..]: GC messages from the JVM, if any 324 # 325 sub callJava { 326 my $method = shift; 327 my $pat = shift; 328 my $n = shift; 329 my $passes = shift; 330 331 my $fileName = $SOURCEDIR . @$pat[0] ; 332 my $n = ($n < 0) ? "-t ".(-$n) : "-i ".$n; 333 334 my $cmd = "java -classpath classes $TESTCLASS $method $n -p $passes -f $fileName -e @$pat[1] -@$pat[2]"; 335 print "[$cmd]\n"; # for debugging 336 open(PIPE, "$cmd|") or die "Can't run \"$cmd\""; 337 my @out; 338 while (<PIPE>) { 339 push(@out, $_); 340 } 341 close(PIPE) or die "Java failed: \"$cmd\""; 342 343 @out = grep(!/^\#/, @out); # filter out comments 344 345 #print "[", join("\n", @out), "]\n"; 346 347 my @results; 348 my $method = ''; 349 my $data = []; 350 foreach (@out) { 351 next unless (/\S/); 352 353 if (/^=\s*(\w+)\s*(\w+)\s*(.*)/) { 354 my ($m, $state, $d) = ($1, $2, $3); 355 #print "$_ => [[$m $state $data]]\n"; 356 if ($state eq 'begin') { 357 die "$method was begun but not finished" if ($method); 358 $method = $m; 359 push(@$data, $d); 360 push(@$data, ''); # placeholder for end data 361 } elsif ($state eq 'end') { 362 if ($m ne $method) { 363 die "$method end does not match: $_"; 364 } 365 $data->[1] = $d; # insert end data at [1] 366 #print "#$method:", join(";",@$data), "\n"; 367 unshift(@$data, $method); # add method to start 368 369 push(@results, $data); 370 $method = ''; 371 $data = []; 372 } else { 373 die "Can't parse: $_"; 374 } 375 } 376 377 elsif (/^\[/) { 378 if ($method) { 379 push(@$data, $_); 380 } else { 381 # ignore extraneous GC notices 382 } 383 } 384 385 else { 386 die "Can't parse: $_"; 387 } 388 } 389 390 die "$method was begun but not finished" if ($method); 391 392 @results; 393 } 394 395 #|#--------------------------------------------------------------------- 396 #|# Format a confidence interval, as given by a Dataset. Output is as 397 #|# as follows: 398 #|# 241.23 - 241.98 => 241.5 +/- 0.3 399 #|# 241.2 - 243.8 => 242 +/- 1 400 #|# 211.0 - 241.0 => 226 +/- 15 or? 230 +/- 20 401 #|# 220.3 - 234.3 => 227 +/- 7 402 #|# 220.3 - 300.3 => 260 +/- 40 403 #|# 220.3 - 1000 => 610 +/- 390 or? 600 +/- 400 404 #|# 0.022 - 0.024 => 0.023 +/- 0.001 405 #|# 0.022 - 0.032 => 0.027 +/- 0.005 406 #|# 0.022 - 1.000 => 0.5 +/- 0.5 407 #|# In other words, take one significant digit of the error value and 408 #|# display the mean to the same precision. 409 #|sub formatDataset { 410 #| my $ds = shift; 411 #| my $lower = $ds->getMean() - $ds->getError(); 412 #| my $upper = $ds->getMean() + $ds->getError(); 413 #| my $scale = 0; 414 #| # Find how many initial digits are the same 415 #| while ($lower < 1 || 416 #| int($lower) == int($upper)) { 417 #| $lower *= 10; 418 #| $upper *= 10; 419 #| $scale++; 420 #| } 421 #| while ($lower >= 10 && 422 #| int($lower) == int($upper)) { 423 #| $lower /= 10; 424 #| $upper /= 10; 425 #| $scale--; 426 #| } 427 #|} 428 429 #--------------------------------------------------------------------- 430 # Format a number, optionally with a +/- delta, to n significant 431 # digits. 432 # 433 # @param significant digit, a value >= 1 434 # @param multiplier 435 # @param time in seconds to be formatted 436 # @optional delta in seconds 437 # 438 # @return string of the form "23" or "23 +/- 10". 439 # 440 sub formatNumber { 441 my $sigdig = shift; 442 my $mult = shift; 443 my $a = shift; 444 my $delta = shift; # may be undef 445 446 my $result = formatSigDig($sigdig, $a*$mult); 447 if (defined($delta)) { 448 my $d = formatSigDig($sigdig, $delta*$mult); 449 # restrict PRECISION of delta to that of main number 450 if ($result =~ /\.(\d+)/) { 451 # TODO make this work for values with all significant 452 # digits to the left of the decimal, e.g., 1234000. 453 454 # TODO the other thing wrong with this is that it 455 # isn't rounding the $delta properly. Have to put 456 # this logic into formatSigDig(). 457 my $x = length($1); 458 $d =~ s/\.(\d{$x})\d+/.$1/; 459 } 460 $result .= " $PLUS_MINUS " . $d; 461 } 462 $result; 463 } 464 465 #--------------------------------------------------------------------- 466 # Format a time, optionally with a +/- delta, to n significant 467 # digits. 468 # 469 # @param significant digit, a value >= 1 470 # @param time in seconds to be formatted 471 # @optional delta in seconds 472 # 473 # @return string of the form "23 ms" or "23 +/- 10 ms". 474 # 475 sub formatSeconds { 476 my $sigdig = shift; 477 my $a = shift; 478 my $delta = shift; # may be undef 479 480 my @MULT = (1 , 1e3, 1e6, 1e9); 481 my @SUFF = ('s' , 'ms', 'us', 'ns'); 482 483 # Determine our scale 484 my $i = 0; 485 ++$i while ($a*$MULT[$i] < 1 && $i < @MULT); 486 487 formatNumber($sigdig, $MULT[$i], $a, $delta) . ' ' . $SUFF[$i]; 488 } 489 490 #--------------------------------------------------------------------- 491 # Format a percentage, optionally with a +/- delta, to n significant 492 # digits. 493 # 494 # @param significant digit, a value >= 1 495 # @param value to be formatted, as a fraction, e.g. 0.5 for 50% 496 # @optional delta, as a fraction 497 # 498 # @return string of the form "23 %" or "23 +/- 10 %". 499 # 500 sub formatPercent { 501 my $sigdig = shift; 502 my $a = shift; 503 my $delta = shift; # may be undef 504 505 formatNumber($sigdig, 100, $a, $delta) . ' %'; 506 } 507 508 #--------------------------------------------------------------------- 509 # Format a number to n significant digits without using exponential 510 # notation. 511 # 512 # @param significant digit, a value >= 1 513 # @param number to be formatted 514 # 515 # @return string of the form "1234" "12.34" or "0.001234". If 516 # number was negative, prefixed by '-'. 517 # 518 sub formatSigDig { 519 my $n = shift() - 1; 520 my $a = shift; 521 522 local $_ = sprintf("%.${n}e", $a); 523 my $sign = (s/^-//) ? '-' : ''; 524 525 my $a_e; 526 my $result; 527 if (/^(\d)\.(\d+)e([-+]\d+)$/) { 528 my ($d, $dn, $e) = ($1, $2, $3); 529 $a_e = $e; 530 $d .= $dn; 531 $e++; 532 $d .= '0' while ($e > length($d)); 533 while ($e < 1) { 534 $e++; 535 $d = '0' . $d; 536 } 537 if ($e == length($d)) { 538 $result = $sign . $d; 539 } else { 540 $result = $sign . substr($d, 0, $e) . '.' . substr($d, $e); 541 } 542 } else { 543 die "Can't parse $_"; 544 } 545 $result; 546 } 547 548 #eof 549
