5 Benchmark - benchmark running times of code
7 timethis - run a chunk of code several times
9 timethese - run several chunks of code several times
11 cmpthese - print results of timethese as a comparison chart
13 timeit - run a chunk of code and see how long it goes
15 countit - see how many times a chunk of code runs in a given time
19 timethis ($count, "code");
21 # Use Perl code in strings...
23 'Name1' => '...code1...',
24 'Name2' => '...code2...',
27 # ... or use subroutine references.
29 'Name1' => sub { ...code1... },
30 'Name2' => sub { ...code2... },
33 # cmpthese can be used both ways as well
35 'Name1' => '...code1...',
36 'Name2' => '...code2...',
40 'Name1' => sub { ...code1... },
41 'Name2' => sub { ...code2... },
45 $results = timethese($count,
47 'Name1' => sub { ...code1... },
48 'Name2' => sub { ...code2... },
52 cmpthese( $results ) ;
54 $t = timeit($count, '...other code...')
55 print "$count loops of other code took:",timestr($t),"\n";
57 $t = countit($time, '...other code...')
59 print "$count loops of other code took:",timestr($t),"\n";
63 The Benchmark module encapsulates a number of routines to help you
64 figure out how long it takes to execute some code.
72 Returns the current time. Example:
76 # ... your code here ...
78 $td = timediff($t1, $t0);
79 print "the code took:",timestr($td),"\n";
83 Enables or disable debugging by setting the C<$Benchmark::Debug> flag:
86 $t = timeit(10, ' 5 ** $Global ');
91 Returns the number of iterations.
95 =head2 Standard Exports
97 The following routines will be exported into your namespace
98 if you use the Benchmark module:
102 =item cmpthese ( COUT, CODEHASHREF, [ STYLE ] )
104 =item cmpthese ( RESULTSHASHREF )
106 Optionally calls timethese(), then outputs comparison chart. This
107 chart is sorted from slowest to highest, and shows the percent
108 speed difference between each pair of tests. Can also be passed
109 the data structure that timethese() returns:
111 $results = timethese( .... );
112 cmpthese( $results );
114 Returns the data structure returned by timethese().
116 =item countit(TIME, CODE)
118 Arguments: TIME is the minimum length of time to run CODE for, and CODE is
119 the code to run. CODE may be either a code reference or a string to
120 be eval'd; either way it will be run in the caller's package.
122 TIME is I<not> negative. countit() will run the loop many times to
123 calculate the speed of CODE before running it for TIME. The actual
124 time run for will usually be greater than TIME due to system clock
125 resolution, so it's best to look at the number of iterations divided
126 by the times that you are concerned with, not just the iterations.
128 Returns: a Benchmark object.
130 =item timeit(COUNT, CODE)
132 Arguments: COUNT is the number of times to run the loop, and CODE is
133 the code to run. CODE may be either a code reference or a string to
134 be eval'd; either way it will be run in the caller's package.
136 Returns: a Benchmark object.
138 =item timethis ( COUNT, CODE, [ TITLE, [ STYLE ]] )
140 Time COUNT iterations of CODE. CODE may be a string to eval or a
141 code reference; either way the CODE will run in the caller's package.
142 Results will be printed to STDOUT as TITLE followed by the times.
143 TITLE defaults to "timethis COUNT" if none is provided. STYLE
144 determines the format of the output, as described for timestr() below.
146 The COUNT can be zero or negative: this means the I<minimum number of
147 CPU seconds> to run. A zero signifies the default of 3 seconds. For
148 example to run at least for 10 seconds:
152 or to run two pieces of code tests for at least 3 seconds:
154 timethese(0, { test1 => '...', test2 => '...'})
156 CPU seconds is, in UNIX terms, the user time plus the system time of
157 the process itself, as opposed to the real (wallclock) time and the
158 time spent by the child processes. Less than 0.1 seconds is not
159 accepted (-0.01 as the count, for example, will cause a fatal runtime
162 Note that the CPU seconds is the B<minimum> time: CPU scheduling and
163 other operating system factors may complicate the attempt so that a
164 little bit more time is spent. The benchmark output will, however,
165 also tell the number of C<$code> runs/second, which should be a more
166 interesting number than the actually spent seconds.
168 Returns a Benchmark object.
170 =item timethese ( COUNT, CODEHASHREF, [ STYLE ] )
172 The CODEHASHREF is a reference to a hash containing names as keys
173 and either a string to eval or a code reference for each value.
174 For each (KEY, VALUE) pair in the CODEHASHREF, this routine will
177 timethis(COUNT, VALUE, KEY, STYLE)
179 The routines are called in string comparison order of KEY.
181 The COUNT can be zero or negative, see timethis().
183 Returns a hash of Benchmark objects, keyed by name.
185 =item timediff ( T1, T2 )
187 Returns the difference between two Benchmark times as a Benchmark
188 object suitable for passing to timestr().
190 =item timesum ( T1, T2 )
192 Returns the sum of two Benchmark times as a Benchmark object suitable
193 for passing to timestr().
195 =item timestr ( TIMEDIFF, [ STYLE, [ FORMAT ] ] )
197 Returns a string that formats the times in the TIMEDIFF object in
198 the requested STYLE. TIMEDIFF is expected to be a Benchmark object
199 similar to that returned by timediff().
201 STYLE can be any of 'all', 'none', 'noc', 'nop' or 'auto'. 'all' shows
202 each of the 5 times available ('wallclock' time, user time, system time,
203 user time of children, and system time of children). 'noc' shows all
204 except the two children times. 'nop' shows only wallclock and the
205 two children times. 'auto' (the default) will act as 'all' unless
206 the children times are both zero, in which case it acts as 'noc'.
207 'none' prevents output.
209 FORMAT is the L<printf(3)>-style format specifier (without the
210 leading '%') to use to print the times. It defaults to '5.2f'.
214 =head2 Optional Exports
216 The following routines will be exported into your namespace
217 if you specifically ask that they be imported:
221 =item clearcache ( COUNT )
223 Clear the cached time for COUNT rounds of the null loop.
225 =item clearallcache ( )
227 Clear all cached times.
229 =item disablecache ( )
231 Disable caching of timings for the null loop. This will force Benchmark
232 to recalculate these timings for each new piece of code timed.
234 =item enablecache ( )
236 Enable caching of timings for the null loop. The time taken for COUNT
237 rounds of the null loop will be calculated only once for each
238 different COUNT used.
244 The data is stored as a list of values from the time and times
247 ($real, $user, $system, $children_user, $children_system, $iters)
249 in seconds for the whole loop (not divided by the number of rounds).
251 The timing is done using time(3) and times(3).
253 Code is executed in the caller's package.
255 The time of the null loop (a loop with the same
256 number of rounds but empty loop body) is subtracted
257 from the time of the real loop.
259 The null loop times can be cached, the key being the
260 number of rounds. The caching can be controlled using
269 Caching is off by default, as it can (usually slightly) decrease
270 accuracy and does not usually noticably affect runtimes.
274 Benchmark inherits from no other class, except of course
279 Comparing eval'd strings with code references will give you
280 inaccurate results: a code reference will show a slightly slower
281 execution time than the equivalent eval'd string.
283 The real time timing is done using time(2) and
284 the granularity is therefore only one second.
286 Short tests may produce negative figures because perl
287 can appear to take longer to execute the empty loop
288 than a short test; try:
292 The system time of the null loop might be slightly
293 more than the system time of the loop with the actual
294 code and therefore the difference might end up being E<lt> 0.
298 Jarkko Hietaniemi <F<jhi@iki.fi>>, Tim Bunce <F<Tim.Bunce@ig.co.uk>>
300 =head1 MODIFICATION HISTORY
302 September 8th, 1994; by Tim Bunce.
304 March 28th, 1997; by Hugo van der Sanden: added support for code
305 references and the already documented 'debug' method; revamped
308 April 04-07th, 1997: by Jarkko Hietaniemi, added the run-for-some-time
311 September, 1999; by Barrie Slaymaker: math fixes and accuracy and
312 efficiency tweaks. Added cmpthese(). A result is now returned from
313 timethese(). Exposed countit() (was runfor()).
317 # evaluate something in a clean lexical environment
318 sub _doeval { eval shift }
321 # put any lexicals at file scope AFTER here
327 @EXPORT=qw(cmpthese countit timeit timethis timethese timediff timestr);
328 @EXPORT_OK=qw(clearcache clearallcache disablecache enablecache);
336 $defaultfmt = '5.2f';
337 $defaultstyle = 'auto';
338 # The cache can cause a slight loss of sys time accuracy. If a
339 # user does many tests (>10) with *very* large counts (>10000)
340 # or works on a very slow machine the cache may be useful.
345 sub debug { $debug = ($_[1] != 0); }
347 # The cache needs two branches: 's' for strings and 'c' for code. The
348 # emtpy loop is different in these two cases.
349 sub clearcache { delete $cache{"$_[0]c"}; delete $cache{"$_[0]s"}; }
350 sub clearallcache { %cache = (); }
351 sub enablecache { $cache = 1; }
352 sub disablecache { $cache = 0; }
354 # --- Functions to process the 'time' data type
356 sub new { my @t = (time, times, @_ == 2 ? $_[1] : 0);
357 print "new=@t\n" if $debug;
360 sub cpu_p { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps ; }
361 sub cpu_c { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $cu+$cs ; }
362 sub cpu_a { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps+$cu+$cs ; }
363 sub real { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $r ; }
364 sub iters { $_[0]->[5] ; }
369 for (my $i=0; $i < @$a; ++$i) {
370 push(@r, $a->[$i] - $b->[$i]);
378 for (my $i=0; $i < @$a; ++$i) {
379 push(@r, $a->[$i] + $b->[$i]);
385 my($tr, $style, $f) = @_;
387 warn "bad time value (@t)" unless @t==6;
388 my($r, $pu, $ps, $cu, $cs, $n) = @t;
389 my($pt, $ct, $t) = ($tr->cpu_p, $tr->cpu_c, $tr->cpu_a);
390 $f = $defaultfmt unless defined $f;
391 # format a time in the required style, other formats may be added here
392 $style ||= $defaultstyle;
393 $style = ($ct>0) ? 'all' : 'noc' if $style eq 'auto';
394 my $s = "@t $style"; # default for unknown style
395 $s=sprintf("%2d wallclock secs (%$f usr %$f sys + %$f cusr %$f csys = %$f CPU)",
396 @t,$t) if $style eq 'all';
397 $s=sprintf("%2d wallclock secs (%$f usr + %$f sys = %$f CPU)",
398 $r,$pu,$ps,$pt) if $style eq 'noc';
399 $s=sprintf("%2d wallclock secs (%$f cusr + %$f csys = %$f CPU)",
400 $r,$cu,$cs,$ct) if $style eq 'nop';
401 $s .= sprintf(" @ %$f/s (n=$n)", $n / ( $pu + $ps )) if $n;
407 print STDERR "$msg",timestr($t),"\n" if $debug;
410 # --- Functions implementing low-level support for timing loops
415 $n+=0; # force numeric now, so garbage won't creep into the eval
416 croak "negative loopcount $n" if $n<0;
417 confess "Usage: runloop(number, [string | coderef])" unless defined $c;
418 my($t0, $t1, $td); # before, after, difference
420 # find package of caller so we can execute code there
421 my($curpack) = caller(0);
423 while (($pack) = caller(++$i)) {
424 last if $pack ne $curpack;
427 my ($subcode, $subref);
428 if (ref $c eq 'CODE') {
429 $subcode = "sub { for (1 .. $n) { local \$_; package $pack; &\$c; } }";
430 $subref = eval $subcode;
433 $subcode = "sub { for (1 .. $n) { local \$_; package $pack; $c;} }";
434 $subref = _doeval($subcode);
436 croak "runloop unable to compile '$c': $@\ncode: $subcode\n" if $@;
437 print STDERR "runloop $n '$subcode'\n" if $debug;
439 # Wait for the user timer to tick. This makes the error range more like
440 # -0.01, +0. If we don't wait, then it's more like -0.01, +0.01. This
441 # may not seem important, but it significantly reduces the chances of
442 # getting a too low initial $n in the initial, 'find the minimum' loop
443 # in &countit. This, in turn, can reduce the number of calls to
444 # &runloop a lot, and thus reduce additive errors.
445 my $tbase = Benchmark->new(0)->[1];
447 $t0 = Benchmark->new(0);
448 } while ( $t0->[1] == $tbase );
450 $t1 = Benchmark->new($n);
451 $td = &timediff($t1, $t0);
452 timedebug("runloop:",$td);
461 printf STDERR "timeit $n $code\n" if $debug;
462 my $cache_key = $n . ( ref( $code ) ? 'c' : 's' );
463 if ($cache && exists $cache{$cache_key} ) {
464 $wn = $cache{$cache_key};
466 $wn = &runloop($n, ref( $code ) ? sub { undef } : '' );
467 # Can't let our baseline have any iterations, or they get subtracted
470 $cache{$cache_key} = $wn;
473 $wc = &runloop($n, $code);
475 $wd = timediff($wc, $wn);
476 timedebug("timeit: ",$wc);
477 timedebug(" - ",$wn);
478 timedebug(" = ",$wd);
489 my ( $tmax, $code ) = @_;
491 if ( not defined $tmax or $tmax == 0 ) {
492 $tmax = $default_for;
493 } elsif ( $tmax < 0 ) {
497 die "countit($tmax, ...): timelimit cannot be less than $min_for.\n"
502 # First find the minimum $n that gives a significant timing.
503 for ($n = 1; ; $n *= 2 ) {
504 my $td = timeit($n, $code);
505 $tc = $td->[1] + $td->[2];
511 # Get $n high enough that we can guess the final $n with some accuracy.
512 my $tpra = 0.1 * $tmax; # Target/time practice.
513 while ( $tc < $tpra ) {
514 # The 5% fudge is to keep us from iterating again all
515 # that often (this speeds overall responsiveness when $tmax is big
516 # and we guess a little low). This does not noticably affect
517 # accuracy since we're not couting these times.
518 $n = int( $tpra * 1.05 * $n / $tc ); # Linear approximation.
519 my $td = timeit($n, $code);
520 $tc = $td->[1] + $td->[2];
523 # Now, do the 'for real' timing(s), repeating until we exceed
533 # The 5% fudge is because $n is often a few % low even for routines
534 # with stable times and avoiding extra timeit()s is nice for
536 $n = int( $n * ( 1.05 * $tmax / $tc ) );
539 my $td = timeit($n, $code);
546 $ttot = $utot + $stot;
547 last if $ttot >= $tmax;
549 my $r = $tmax / $ttot - 1; # Linear approximation.
550 $n = int( $r * $ntot );
551 $n = $nmin if $n < $nmin;
554 return bless [ $rtot, $utot, $stot, $cutot, $cstot, $ntot ];
557 # --- Functions implementing high-level time-then-print utilities
561 return $n == 0 ? $default_for : $n < 0 ? -$n : undef;
565 my($n, $code, $title, $style) = @_;
569 croak "non-integer loopcount $n, stopped" if int($n)<$n;
570 $t = timeit($n, $code);
571 $title = "timethis $n" unless defined $title;
573 $fort = n_to_for( $n );
574 $t = countit( $fort, $code );
575 $title = "timethis for $fort" unless defined $title;
579 $style = "" unless defined $style;
580 printf("%10s: ", $title) unless $style eq 'none';
581 print timestr($t, $style, $defaultfmt),"\n" unless $style eq 'none';
583 $n = $forn if defined $forn;
585 # A conservative warning to spot very silly tests.
586 # Don't assume that your benchmark is ok simply because
587 # you don't get this warning!
588 print " (warning: too few iterations for a reliable count)\n"
590 || ($t->real < 1 && $n < 1000)
591 || $t->cpu_a < $min_cpu;
596 my($n, $alt, $style) = @_;
597 die "usage: timethese(count, { 'Name1'=>'code1', ... }\n"
598 unless ref $alt eq HASH;
599 my @names = sort keys %$alt;
600 $style = "" unless defined $style;
601 print "Benchmark: " unless $style eq 'none';
603 croak "non-integer loopcount $n, stopped" if int($n)<$n;
604 print "timing $n iterations of" unless $style eq 'none';
606 print "running" unless $style eq 'none';
608 print " ", join(', ',@names) unless $style eq 'none';
610 my $for = n_to_for( $n );
611 print ", each for at least $for CPU seconds" unless $style eq 'none';
613 print "...\n" unless $style eq 'none';
615 # we could save the results in an array and produce a summary here
616 # sum, min, max, avg etc etc
618 foreach my $name (@names) {
619 $results{$name} = timethis ($n, $alt -> {$name}, $name, $style);
626 my $results = ref $_[0] ? $_[0] : timethese( @_ );
629 if defined $_[2] && $_[2] eq 'none';
631 # Flatten in to an array of arrays with the name as the first field
632 my @vals = map{ [ $_, @{$results->{$_}} ] } keys %$results;
635 # The epsilon fudge here is to prevent div by 0. Since clock
636 # resolutions are much larger, it's below the noise floor.
637 my $rate = $_->[6] / ( $_->[2] + $_->[3] + 0.000000000000001 );
642 @vals = sort { $a->[7] <=> $b->[7] } @vals;
644 # If more than half of the rates are greater than one...
645 my $display_as_rate = $vals[$#vals>>1]->[7] > 1;
652 $display_as_rate ? 'Rate' : 's/iter',
653 map { $_->[0] } @vals
656 push @rows, \@top_row;
657 @col_widths = map { length( $_ ) } @top_row;
659 # Build the data rows
660 # We leave the last column in even though it never has any data. Perhaps
661 # it should go away. Also, perhaps a style for a single column of
662 # percentages might be nice.
663 for my $row_val ( @vals ) {
666 # Column 0 = test name
667 push @row, $row_val->[0];
668 $col_widths[0] = length( $row_val->[0] )
669 if length( $row_val->[0] ) > $col_widths[0];
671 # Column 1 = performance
672 my $row_rate = $row_val->[7];
674 # We assume that we'll never get a 0 rate.
675 my $a = $display_as_rate ? $row_rate : 1 / $row_rate;
677 # Only give a few decimal places before switching to sci. notation,
678 # since the results aren't usually that accurate anyway.
692 # Using $b here due to optimizing bug in _58 through _61
693 my $b = sprintf( $format, $a );
695 $col_widths[1] = length( $b )
696 if length( $b ) > $col_widths[1];
698 # Columns 2..N = performance ratios
700 for ( my $col_num = 0 ; $col_num < @vals ; ++$col_num ) {
701 my $col_val = $vals[$col_num];
706 elsif ( $col_val->[0] eq $row_val->[0] ) {
711 my $col_rate = $col_val->[7];
712 $out = sprintf( "%.0f%%", 100*$row_rate/$col_rate - 100 );
715 $col_widths[$col_num+2] = length( $out )
716 if length( $out ) > $col_widths[$col_num+2];
718 # A little wierdness to set the first column width properly
719 $col_widths[$col_num+2] = length( $col_val->[0] )
720 if length( $col_val->[0] ) > $col_widths[$col_num+2];
725 # Equalize column widths in the chart as much as possible without
726 # exceeding 80 characters. This does not use or affect cols 0 or 1.
727 my @sorted_width_refs =
728 sort { $$a <=> $$b } map { \$_ } @col_widths[2..$#col_widths];
729 my $max_width = ${$sorted_width_refs[-1]};
732 for ( @col_widths ) { $total += $_ }
735 while ( $total < 80 ) {
736 my $min_width = ${$sorted_width_refs[0]};
738 if $min_width == $max_width;
739 for ( @sorted_width_refs ) {
750 my $format = join( ' ', map { "%${_}s" } @col_widths ) . "\n";
751 substr( $format, 1, 0 ) = '-';