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