5 Benchmark - benchmark running times of Perl code
9 timethis ($count, "code");
11 # Use Perl code in strings...
13 'Name1' => '...code1...',
14 'Name2' => '...code2...',
17 # ... or use subroutine references.
19 'Name1' => sub { ...code1... },
20 'Name2' => sub { ...code2... },
23 # cmpthese can be used both ways as well
25 'Name1' => '...code1...',
26 'Name2' => '...code2...',
30 'Name1' => sub { ...code1... },
31 'Name2' => sub { ...code2... },
35 $results = timethese($count,
37 'Name1' => sub { ...code1... },
38 'Name2' => sub { ...code2... },
42 cmpthese( $results ) ;
44 $t = timeit($count, '...other code...')
45 print "$count loops of other code took:",timestr($t),"\n";
47 $t = countit($time, '...other code...')
49 print "$count loops of other code took:",timestr($t),"\n";
53 The Benchmark module encapsulates a number of routines to help you
54 figure out how long it takes to execute some code.
56 timethis - run a chunk of code several times
58 timethese - run several chunks of code several times
60 cmpthese - print results of timethese as a comparison chart
62 timeit - run a chunk of code and see how long it goes
64 countit - see how many times a chunk of code runs in a given time
73 Returns the current time. Example:
77 # ... your code here ...
79 $td = timediff($t1, $t0);
80 print "the code took:",timestr($td),"\n";
84 Enables or disable debugging by setting the C<$Benchmark::Debug> flag:
87 $t = timeit(10, ' 5 ** $Global ');
92 Returns the number of iterations.
96 =head2 Standard Exports
98 The following routines will be exported into your namespace
99 if you use the Benchmark module:
103 =item timeit(COUNT, CODE)
105 Arguments: COUNT is the number of times to run the loop, and CODE is
106 the code to run. CODE may be either a code reference or a string to
107 be eval'd; either way it will be run in the caller's package.
109 Returns: a Benchmark object.
111 =item timethis ( COUNT, CODE, [ TITLE, [ STYLE ]] )
113 Time COUNT iterations of CODE. CODE may be a string to eval or a
114 code reference; either way the CODE will run in the caller's package.
115 Results will be printed to STDOUT as TITLE followed by the times.
116 TITLE defaults to "timethis COUNT" if none is provided. STYLE
117 determines the format of the output, as described for timestr() below.
119 The COUNT can be zero or negative: this means the I<minimum number of
120 CPU seconds> to run. A zero signifies the default of 3 seconds. For
121 example to run at least for 10 seconds:
125 or to run two pieces of code tests for at least 3 seconds:
127 timethese(0, { test1 => '...', test2 => '...'})
129 CPU seconds is, in UNIX terms, the user time plus the system time of
130 the process itself, as opposed to the real (wallclock) time and the
131 time spent by the child processes. Less than 0.1 seconds is not
132 accepted (-0.01 as the count, for example, will cause a fatal runtime
135 Note that the CPU seconds is the B<minimum> time: CPU scheduling and
136 other operating system factors may complicate the attempt so that a
137 little bit more time is spent. The benchmark output will, however,
138 also tell the number of C<$code> runs/second, which should be a more
139 interesting number than the actually spent seconds.
141 Returns a Benchmark object.
143 =item timethese ( COUNT, CODEHASHREF, [ STYLE ] )
145 The CODEHASHREF is a reference to a hash containing names as keys
146 and either a string to eval or a code reference for each value.
147 For each (KEY, VALUE) pair in the CODEHASHREF, this routine will
150 timethis(COUNT, VALUE, KEY, STYLE)
152 The routines are called in string comparison order of KEY.
154 The COUNT can be zero or negative, see timethis().
156 Returns a hash of Benchmark objects, keyed by name.
158 =item timediff ( T1, T2 )
160 Returns the difference between two Benchmark times as a Benchmark
161 object suitable for passing to timestr().
163 =item timestr ( TIMEDIFF, [ STYLE, [ FORMAT ] ] )
165 Returns a string that formats the times in the TIMEDIFF object in
166 the requested STYLE. TIMEDIFF is expected to be a Benchmark object
167 similar to that returned by timediff().
169 STYLE can be any of 'all', 'none', 'noc', 'nop' or 'auto'. 'all' shows
170 each of the 5 times available ('wallclock' time, user time, system time,
171 user time of children, and system time of children). 'noc' shows all
172 except the two children times. 'nop' shows only wallclock and the
173 two children times. 'auto' (the default) will act as 'all' unless
174 the children times are both zero, in which case it acts as 'noc'.
175 'none' prevents output.
177 FORMAT is the L<printf(3)>-style format specifier (without the
178 leading '%') to use to print the times. It defaults to '5.2f'.
182 =head2 Optional Exports
184 The following routines will be exported into your namespace
185 if you specifically ask that they be imported:
189 =item clearcache ( COUNT )
191 Clear the cached time for COUNT rounds of the null loop.
193 =item clearallcache ( )
195 Clear all cached times.
197 =item cmpthese ( COUT, CODEHASHREF, [ STYLE ] )
199 =item cmpthese ( RESULTSHASHREF, [ STYLE ] )
201 Optionally calls timethese(), then outputs comparison chart. This:
203 cmpthese( -1, { a => "++\$i", b => "\$i *= 2" } ) ;
205 outputs a chart like:
211 This chart is sorted from slowest to fastest, and shows the percent speed
212 difference between each pair of tests.
214 c<cmpthese> can also be passed the data structure that timethese() returns:
216 $results = timethese( -1, { a => "++\$i", b => "\$i *= 2" } ) ;
217 cmpthese( $results );
219 in case you want to see both sets of results.
221 Returns a reference to an ARRAY of rows, each row is an ARRAY of cells from the
222 above chart, including labels. This:
224 my $rows = cmpthese( -1, { a => '++$i', b => '$i *= 2' }, "none" );
226 returns a data structure like:
229 [ '', 'Rate', 'b', 'a' ],
230 [ 'b', '2885232/s', '--', '-59%' ],
231 [ 'a', '7099126/s', '146%', '--' ],
234 B<NOTE>: This result value differs from previous versions, which returned
235 the C<timethese()> result structure. If you want that, just use the two
236 statement C<timethese>...C<cmpthese> idiom shown above.
238 Incidently, note the variance in the result values between the two examples;
239 this is typical of benchmarking. If this were a real benchmark, you would
240 probably want to run a lot more iterations.
242 =item countit(TIME, CODE)
244 Arguments: TIME is the minimum length of time to run CODE for, and CODE is
245 the code to run. CODE may be either a code reference or a string to
246 be eval'd; either way it will be run in the caller's package.
248 TIME is I<not> negative. countit() will run the loop many times to
249 calculate the speed of CODE before running it for TIME. The actual
250 time run for will usually be greater than TIME due to system clock
251 resolution, so it's best to look at the number of iterations divided
252 by the times that you are concerned with, not just the iterations.
254 Returns: a Benchmark object.
256 =item disablecache ( )
258 Disable caching of timings for the null loop. This will force Benchmark
259 to recalculate these timings for each new piece of code timed.
261 =item enablecache ( )
263 Enable caching of timings for the null loop. The time taken for COUNT
264 rounds of the null loop will be calculated only once for each
265 different COUNT used.
267 =item timesum ( T1, T2 )
269 Returns the sum of two Benchmark times as a Benchmark object suitable
270 for passing to timestr().
276 The data is stored as a list of values from the time and times
279 ($real, $user, $system, $children_user, $children_system, $iters)
281 in seconds for the whole loop (not divided by the number of rounds).
283 The timing is done using time(3) and times(3).
285 Code is executed in the caller's package.
287 The time of the null loop (a loop with the same
288 number of rounds but empty loop body) is subtracted
289 from the time of the real loop.
291 The null loop times can be cached, the key being the
292 number of rounds. The caching can be controlled using
301 Caching is off by default, as it can (usually slightly) decrease
302 accuracy and does not usually noticably affect runtimes.
308 use Benchmark;$x=3;cmpthese(-5,{a=>sub{$x*$x},b=>sub{$x**2}})
310 outputs something like this:
312 Benchmark: running a, b, each for at least 5 CPU seconds...
313 a: 10 wallclock secs ( 5.14 usr + 0.13 sys = 5.27 CPU) @ 3835055.60/s (n=20210743)
314 b: 5 wallclock secs ( 5.41 usr + 0.00 sys = 5.41 CPU) @ 1574944.92/s (n=8520452)
323 $r=timethese(-5,{a=>sub{$x*$x},b=>sub{$x**2}},'none');
326 outputs something like this:
335 Benchmark inherits from no other class, except of course
340 Comparing eval'd strings with code references will give you
341 inaccurate results: a code reference will show a slightly slower
342 execution time than the equivalent eval'd string.
344 The real time timing is done using time(2) and
345 the granularity is therefore only one second.
347 Short tests may produce negative figures because perl
348 can appear to take longer to execute the empty loop
349 than a short test; try:
353 The system time of the null loop might be slightly
354 more than the system time of the loop with the actual
355 code and therefore the difference might end up being E<lt> 0.
359 L<Devel::DProf> - a Perl code profiler
363 Jarkko Hietaniemi <F<jhi@iki.fi>>, Tim Bunce <F<Tim.Bunce@ig.co.uk>>
365 =head1 MODIFICATION HISTORY
367 September 8th, 1994; by Tim Bunce.
369 March 28th, 1997; by Hugo van der Sanden: added support for code
370 references and the already documented 'debug' method; revamped
373 April 04-07th, 1997: by Jarkko Hietaniemi, added the run-for-some-time
376 September, 1999; by Barrie Slaymaker: math fixes and accuracy and
377 efficiency tweaks. Added cmpthese(). A result is now returned from
378 timethese(). Exposed countit() (was runfor()).
382 # evaluate something in a clean lexical environment
383 sub _doeval { eval shift }
386 # put any lexicals at file scope AFTER here
392 @EXPORT=qw(timeit timethis timethese timediff timestr);
393 @EXPORT_OK=qw(timesum cmpthese countit
394 clearcache clearallcache disablecache enablecache);
404 $defaultfmt = '5.2f';
405 $defaultstyle = 'auto';
406 # The cache can cause a slight loss of sys time accuracy. If a
407 # user does many tests (>10) with *very* large counts (>10000)
408 # or works on a very slow machine the cache may be useful.
413 sub debug { $debug = ($_[1] != 0); }
415 # The cache needs two branches: 's' for strings and 'c' for code. The
416 # emtpy loop is different in these two cases.
417 sub clearcache { delete $cache{"$_[0]c"}; delete $cache{"$_[0]s"}; }
418 sub clearallcache { %cache = (); }
419 sub enablecache { $cache = 1; }
420 sub disablecache { $cache = 0; }
422 # --- Functions to process the 'time' data type
424 sub new { my @t = (time, times, @_ == 2 ? $_[1] : 0);
425 print "new=@t\n" if $debug;
428 sub cpu_p { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps ; }
429 sub cpu_c { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $cu+$cs ; }
430 sub cpu_a { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps+$cu+$cs ; }
431 sub real { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $r ; }
432 sub iters { $_[0]->[5] ; }
437 for (my $i=0; $i < @$a; ++$i) {
438 push(@r, $a->[$i] - $b->[$i]);
446 for (my $i=0; $i < @$a; ++$i) {
447 push(@r, $a->[$i] + $b->[$i]);
453 my($tr, $style, $f) = @_;
455 warn "bad time value (@t)" unless @t==6;
456 my($r, $pu, $ps, $cu, $cs, $n) = @t;
457 my($pt, $ct, $tt) = ($tr->cpu_p, $tr->cpu_c, $tr->cpu_a);
458 $f = $defaultfmt unless defined $f;
459 # format a time in the required style, other formats may be added here
460 $style ||= $defaultstyle;
461 $style = ($ct>0) ? 'all' : 'noc' if $style eq 'auto';
462 my $s = "@t $style"; # default for unknown style
463 $s=sprintf("%2d wallclock secs (%$f usr %$f sys + %$f cusr %$f csys = %$f CPU)",
464 $r,$pu,$ps,$cu,$cs,$tt) if $style eq 'all';
465 $s=sprintf("%2d wallclock secs (%$f usr + %$f sys = %$f CPU)",
466 $r,$pu,$ps,$pt) if $style eq 'noc';
467 $s=sprintf("%2d wallclock secs (%$f cusr + %$f csys = %$f CPU)",
468 $r,$cu,$cs,$ct) if $style eq 'nop';
469 $s .= sprintf(" @ %$f/s (n=$n)", $n / ( $pu + $ps )) if $n && $pu+$ps;
475 print STDERR "$msg",timestr($t),"\n" if $debug;
478 # --- Functions implementing low-level support for timing loops
483 $n+=0; # force numeric now, so garbage won't creep into the eval
484 croak "negative loopcount $n" if $n<0;
485 confess "Usage: runloop(number, [string | coderef])" unless defined $c;
486 my($t0, $t1, $td); # before, after, difference
488 # find package of caller so we can execute code there
489 my($curpack) = caller(0);
491 while (($pack) = caller(++$i)) {
492 last if $pack ne $curpack;
495 my ($subcode, $subref);
496 if (ref $c eq 'CODE') {
497 $subcode = "sub { for (1 .. $n) { local \$_; package $pack; &\$c; } }";
498 $subref = eval $subcode;
501 $subcode = "sub { for (1 .. $n) { local \$_; package $pack; $c;} }";
502 $subref = _doeval($subcode);
504 croak "runloop unable to compile '$c': $@\ncode: $subcode\n" if $@;
505 print STDERR "runloop $n '$subcode'\n" if $debug;
507 # Wait for the user timer to tick. This makes the error range more like
508 # -0.01, +0. If we don't wait, then it's more like -0.01, +0.01. This
509 # may not seem important, but it significantly reduces the chances of
510 # getting a too low initial $n in the initial, 'find the minimum' loop
511 # in &countit. This, in turn, can reduce the number of calls to
512 # &runloop a lot, and thus reduce additive errors.
513 my $tbase = Benchmark->new(0)->[1];
514 while ( ( $t0 = Benchmark->new(0) )->[1] == $tbase ) {} ;
516 $t1 = Benchmark->new($n);
517 $td = &timediff($t1, $t0);
518 timedebug("runloop:",$td);
527 printf STDERR "timeit $n $code\n" if $debug;
528 my $cache_key = $n . ( ref( $code ) ? 'c' : 's' );
529 if ($cache && exists $cache{$cache_key} ) {
530 $wn = $cache{$cache_key};
532 $wn = &runloop($n, ref( $code ) ? sub { undef } : '' );
533 # Can't let our baseline have any iterations, or they get subtracted
536 $cache{$cache_key} = $wn;
539 $wc = &runloop($n, $code);
541 $wd = timediff($wc, $wn);
542 timedebug("timeit: ",$wc);
543 timedebug(" - ",$wn);
544 timedebug(" = ",$wd);
555 my ( $tmax, $code ) = @_;
557 if ( not defined $tmax or $tmax == 0 ) {
558 $tmax = $default_for;
559 } elsif ( $tmax < 0 ) {
563 die "countit($tmax, ...): timelimit cannot be less than $min_for.\n"
568 # First find the minimum $n that gives a significant timing.
569 for ($n = 1; ; $n *= 2 ) {
570 my $td = timeit($n, $code);
571 $tc = $td->[1] + $td->[2];
577 # Get $n high enough that we can guess the final $n with some accuracy.
578 my $tpra = 0.1 * $tmax; # Target/time practice.
579 while ( $tc < $tpra ) {
580 # The 5% fudge is to keep us from iterating again all
581 # that often (this speeds overall responsiveness when $tmax is big
582 # and we guess a little low). This does not noticably affect
583 # accuracy since we're not couting these times.
584 $n = int( $tpra * 1.05 * $n / $tc ); # Linear approximation.
585 my $td = timeit($n, $code);
586 my $new_tc = $td->[1] + $td->[2];
587 # Make sure we are making progress.
588 $tc = $new_tc > 1.2 * $tc ? $new_tc : 1.2 * $tc;
591 # Now, do the 'for real' timing(s), repeating until we exceed
601 # The 5% fudge is because $n is often a few % low even for routines
602 # with stable times and avoiding extra timeit()s is nice for
604 $n = int( $n * ( 1.05 * $tmax / $tc ) );
607 my $td = timeit($n, $code);
614 $ttot = $utot + $stot;
615 last if $ttot >= $tmax;
617 $ttot = 0.01 if $ttot < 0.01;
618 my $r = $tmax / $ttot - 1; # Linear approximation.
619 $n = int( $r * $ntot );
620 $n = $nmin if $n < $nmin;
623 return bless [ $rtot, $utot, $stot, $cutot, $cstot, $ntot ];
626 # --- Functions implementing high-level time-then-print utilities
630 return $n == 0 ? $default_for : $n < 0 ? -$n : undef;
634 my($n, $code, $title, $style) = @_;
638 croak "non-integer loopcount $n, stopped" if int($n)<$n;
639 $t = timeit($n, $code);
640 $title = "timethis $n" unless defined $title;
642 $fort = n_to_for( $n );
643 $t = countit( $fort, $code );
644 $title = "timethis for $fort" unless defined $title;
648 $style = "" unless defined $style;
649 printf("%10s: ", $title) unless $style eq 'none';
650 print timestr($t, $style, $defaultfmt),"\n" unless $style eq 'none';
652 $n = $forn if defined $forn;
654 # A conservative warning to spot very silly tests.
655 # Don't assume that your benchmark is ok simply because
656 # you don't get this warning!
657 print " (warning: too few iterations for a reliable count)\n"
659 || ($t->real < 1 && $n < 1000)
660 || $t->cpu_a < $min_cpu;
665 my($n, $alt, $style) = @_;
666 die "usage: timethese(count, { 'Name1'=>'code1', ... }\n"
667 unless ref $alt eq HASH;
668 my @names = sort keys %$alt;
669 $style = "" unless defined $style;
670 print "Benchmark: " unless $style eq 'none';
672 croak "non-integer loopcount $n, stopped" if int($n)<$n;
673 print "timing $n iterations of" unless $style eq 'none';
675 print "running" unless $style eq 'none';
677 print " ", join(', ',@names) unless $style eq 'none';
679 my $for = n_to_for( $n );
680 print ", each for at least $for CPU seconds" unless $style eq 'none';
682 print "...\n" unless $style eq 'none';
684 # we could save the results in an array and produce a summary here
685 # sum, min, max, avg etc etc
687 foreach my $name (@names) {
688 $results{$name} = timethis ($n, $alt -> {$name}, $name, $style);
695 my ($results, $style) = ref $_[0] ? @_ : ( timethese( @_[0,1] ), $_[2] ) ;
697 $style = "" unless defined $style;
699 # Flatten in to an array of arrays with the name as the first field
700 my @vals = map{ [ $_, @{$results->{$_}} ] } keys %$results;
703 # The epsilon fudge here is to prevent div by 0. Since clock
704 # resolutions are much larger, it's below the noise floor.
705 my $rate = $_->[6] / ( $_->[2] + $_->[3] + 0.000000000000001 );
710 @vals = sort { $a->[7] <=> $b->[7] } @vals;
712 # If more than half of the rates are greater than one...
713 my $display_as_rate = $vals[$#vals>>1]->[7] > 1;
720 $display_as_rate ? 'Rate' : 's/iter',
721 map { $_->[0] } @vals
724 push @rows, \@top_row;
725 @col_widths = map { length( $_ ) } @top_row;
727 # Build the data rows
728 # We leave the last column in even though it never has any data. Perhaps
729 # it should go away. Also, perhaps a style for a single column of
730 # percentages might be nice.
731 for my $row_val ( @vals ) {
734 # Column 0 = test name
735 push @row, $row_val->[0];
736 $col_widths[0] = length( $row_val->[0] )
737 if length( $row_val->[0] ) > $col_widths[0];
739 # Column 1 = performance
740 my $row_rate = $row_val->[7];
742 # We assume that we'll never get a 0 rate.
743 my $a = $display_as_rate ? $row_rate : 1 / $row_rate;
745 # Only give a few decimal places before switching to sci. notation,
746 # since the results aren't usually that accurate anyway.
760 # Using $b here due to optimizing bug in _58 through _61
761 my $b = sprintf( $format, $a );
763 $col_widths[1] = length( $b )
764 if length( $b ) > $col_widths[1];
766 # Columns 2..N = performance ratios
768 for ( my $col_num = 0 ; $col_num < @vals ; ++$col_num ) {
769 my $col_val = $vals[$col_num];
774 elsif ( $col_val->[0] eq $row_val->[0] ) {
779 my $col_rate = $col_val->[7];
780 $out = sprintf( "%.0f%%", 100*$row_rate/$col_rate - 100 );
783 $col_widths[$col_num+2] = length( $out )
784 if length( $out ) > $col_widths[$col_num+2];
786 # A little wierdness to set the first column width properly
787 $col_widths[$col_num+2] = length( $col_val->[0] )
788 if length( $col_val->[0] ) > $col_widths[$col_num+2];
793 return \@rows if $style eq "none";
795 # Equalize column widths in the chart as much as possible without
796 # exceeding 80 characters. This does not use or affect cols 0 or 1.
797 my @sorted_width_refs =
798 sort { $$a <=> $$b } map { \$_ } @col_widths[2..$#col_widths];
799 my $max_width = ${$sorted_width_refs[-1]};
801 my $total = @col_widths - 1 ;
802 for ( @col_widths ) { $total += $_ }
805 while ( $total < 80 ) {
806 my $min_width = ${$sorted_width_refs[0]};
808 if $min_width == $max_width;
809 for ( @sorted_width_refs ) {
820 my $format = join( ' ', map { "%${_}s" } @col_widths ) . "\n";
821 substr( $format, 1, 0 ) = '-';