Commit | Line | Data |
---|---|---|
a0d0e21e LW |
1 | package Benchmark; |
2 | ||
f06db76b AD |
3 | =head1 NAME |
4 | ||
8a4f6ac2 | 5 | Benchmark - benchmark running times of Perl code |
431d98c2 | 6 | |
f06db76b AD |
7 | =head1 SYNOPSIS |
8 | ||
9 | timethis ($count, "code"); | |
10 | ||
523cc92b | 11 | # Use Perl code in strings... |
f06db76b AD |
12 | timethese($count, { |
13 | 'Name1' => '...code1...', | |
14 | 'Name2' => '...code2...', | |
15 | }); | |
16 | ||
523cc92b CS |
17 | # ... or use subroutine references. |
18 | timethese($count, { | |
19 | 'Name1' => sub { ...code1... }, | |
20 | 'Name2' => sub { ...code2... }, | |
21 | }); | |
22 | ||
431d98c2 BS |
23 | # cmpthese can be used both ways as well |
24 | cmpthese($count, { | |
25 | 'Name1' => '...code1...', | |
26 | 'Name2' => '...code2...', | |
27 | }); | |
28 | ||
29 | cmpthese($count, { | |
30 | 'Name1' => sub { ...code1... }, | |
31 | 'Name2' => sub { ...code2... }, | |
32 | }); | |
33 | ||
34 | # ...or in two stages | |
35 | $results = timethese($count, | |
36 | { | |
37 | 'Name1' => sub { ...code1... }, | |
38 | 'Name2' => sub { ...code2... }, | |
39 | }, | |
40 | 'none' | |
41 | ); | |
42 | cmpthese( $results ) ; | |
43 | ||
f06db76b AD |
44 | $t = timeit($count, '...other code...') |
45 | print "$count loops of other code took:",timestr($t),"\n"; | |
46 | ||
431d98c2 BS |
47 | $t = countit($time, '...other code...') |
48 | $count = $t->iters ; | |
49 | print "$count loops of other code took:",timestr($t),"\n"; | |
50 | ||
f06db76b AD |
51 | =head1 DESCRIPTION |
52 | ||
53 | The Benchmark module encapsulates a number of routines to help you | |
54 | figure out how long it takes to execute some code. | |
55 | ||
8a4f6ac2 GS |
56 | timethis - run a chunk of code several times |
57 | ||
58 | timethese - run several chunks of code several times | |
59 | ||
60 | cmpthese - print results of timethese as a comparison chart | |
61 | ||
62 | timeit - run a chunk of code and see how long it goes | |
63 | ||
64 | countit - see how many times a chunk of code runs in a given time | |
65 | ||
66 | ||
f06db76b AD |
67 | =head2 Methods |
68 | ||
69 | =over 10 | |
70 | ||
71 | =item new | |
72 | ||
73 | Returns the current time. Example: | |
74 | ||
75 | use Benchmark; | |
76 | $t0 = new Benchmark; | |
77 | # ... your code here ... | |
78 | $t1 = new Benchmark; | |
79 | $td = timediff($t1, $t0); | |
a24a9dfe | 80 | print "the code took:",timestr($td),"\n"; |
f06db76b AD |
81 | |
82 | =item debug | |
83 | ||
84 | Enables or disable debugging by setting the C<$Benchmark::Debug> flag: | |
85 | ||
523cc92b | 86 | debug Benchmark 1; |
f06db76b | 87 | $t = timeit(10, ' 5 ** $Global '); |
523cc92b | 88 | debug Benchmark 0; |
f06db76b | 89 | |
431d98c2 BS |
90 | =item iters |
91 | ||
92 | Returns the number of iterations. | |
93 | ||
f06db76b AD |
94 | =back |
95 | ||
96 | =head2 Standard Exports | |
97 | ||
523cc92b | 98 | The following routines will be exported into your namespace |
f06db76b AD |
99 | if you use the Benchmark module: |
100 | ||
101 | =over 10 | |
102 | ||
103 | =item timeit(COUNT, CODE) | |
104 | ||
523cc92b CS |
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. | |
108 | ||
109 | Returns: a Benchmark object. | |
110 | ||
111 | =item timethis ( COUNT, CODE, [ TITLE, [ STYLE ]] ) | |
112 | ||
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. | |
118 | ||
6ee623d5 GS |
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: | |
122 | ||
123 | timethis(-10, $code) | |
124 | ||
125 | or to run two pieces of code tests for at least 3 seconds: | |
126 | ||
127 | timethese(0, { test1 => '...', test2 => '...'}) | |
128 | ||
129 | CPU seconds is, in UNIX terms, the user time plus the system time of | |
130 | the process itself, as opposed to the real (wallclock) time and the | |
131 | time spent by the child processes. Less than 0.1 seconds is not | |
132 | accepted (-0.01 as the count, for example, will cause a fatal runtime | |
133 | exception). | |
134 | ||
135 | Note that the CPU seconds is the B<minimum> time: CPU scheduling and | |
136 | other operating system factors may complicate the attempt so that a | |
137 | little bit more time is spent. The benchmark output will, however, | |
138 | also tell the number of C<$code> runs/second, which should be a more | |
139 | interesting number than the actually spent seconds. | |
140 | ||
141 | Returns a Benchmark object. | |
142 | ||
523cc92b | 143 | =item timethese ( COUNT, CODEHASHREF, [ STYLE ] ) |
f06db76b | 144 | |
523cc92b CS |
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 | |
148 | call | |
f06db76b | 149 | |
523cc92b | 150 | timethis(COUNT, VALUE, KEY, STYLE) |
f06db76b | 151 | |
1d2dff63 GS |
152 | The routines are called in string comparison order of KEY. |
153 | ||
154 | The COUNT can be zero or negative, see timethis(). | |
6ee623d5 | 155 | |
3c6312e9 BS |
156 | Returns a hash of Benchmark objects, keyed by name. |
157 | ||
523cc92b | 158 | =item timediff ( T1, T2 ) |
f06db76b | 159 | |
523cc92b CS |
160 | Returns the difference between two Benchmark times as a Benchmark |
161 | object suitable for passing to timestr(). | |
f06db76b | 162 | |
6ee623d5 | 163 | =item timestr ( TIMEDIFF, [ STYLE, [ FORMAT ] ] ) |
f06db76b | 164 | |
523cc92b CS |
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(). | |
168 | ||
3c6312e9 BS |
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, | |
523cc92b CS |
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'. | |
3c6312e9 | 175 | 'none' prevents output. |
523cc92b CS |
176 | |
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'. | |
f06db76b AD |
179 | |
180 | =back | |
181 | ||
182 | =head2 Optional Exports | |
183 | ||
184 | The following routines will be exported into your namespace | |
185 | if you specifically ask that they be imported: | |
186 | ||
187 | =over 10 | |
188 | ||
523cc92b CS |
189 | =item clearcache ( COUNT ) |
190 | ||
191 | Clear the cached time for COUNT rounds of the null loop. | |
192 | ||
193 | =item clearallcache ( ) | |
f06db76b | 194 | |
523cc92b | 195 | Clear all cached times. |
f06db76b | 196 | |
ac8eabc1 JH |
197 | =item cmpthese ( COUT, CODEHASHREF, [ STYLE ] ) |
198 | ||
199 | =item cmpthese ( RESULTSHASHREF ) | |
200 | ||
201 | Optionally calls timethese(), then outputs comparison chart. This | |
277427cf | 202 | chart is sorted from slowest to fastest, and shows the percent |
ac8eabc1 JH |
203 | speed difference between each pair of tests. Can also be passed |
204 | the data structure that timethese() returns: | |
205 | ||
206 | $results = timethese( .... ); | |
207 | cmpthese( $results ); | |
208 | ||
277427cf | 209 | Returns the data structure returned by timethese() (or passed in). |
ac8eabc1 JH |
210 | |
211 | =item countit(TIME, CODE) | |
212 | ||
213 | Arguments: TIME is the minimum length of time to run CODE for, and CODE is | |
214 | the code to run. CODE may be either a code reference or a string to | |
215 | be eval'd; either way it will be run in the caller's package. | |
216 | ||
217 | TIME is I<not> negative. countit() will run the loop many times to | |
218 | calculate the speed of CODE before running it for TIME. The actual | |
219 | time run for will usually be greater than TIME due to system clock | |
220 | resolution, so it's best to look at the number of iterations divided | |
221 | by the times that you are concerned with, not just the iterations. | |
222 | ||
223 | Returns: a Benchmark object. | |
224 | ||
523cc92b | 225 | =item disablecache ( ) |
f06db76b | 226 | |
523cc92b CS |
227 | Disable caching of timings for the null loop. This will force Benchmark |
228 | to recalculate these timings for each new piece of code timed. | |
229 | ||
230 | =item enablecache ( ) | |
231 | ||
232 | Enable caching of timings for the null loop. The time taken for COUNT | |
233 | rounds of the null loop will be calculated only once for each | |
234 | different COUNT used. | |
f06db76b | 235 | |
ac8eabc1 JH |
236 | =item timesum ( T1, T2 ) |
237 | ||
238 | Returns the sum of two Benchmark times as a Benchmark object suitable | |
239 | for passing to timestr(). | |
240 | ||
f06db76b AD |
241 | =back |
242 | ||
243 | =head1 NOTES | |
244 | ||
245 | The data is stored as a list of values from the time and times | |
523cc92b | 246 | functions: |
f06db76b | 247 | |
431d98c2 | 248 | ($real, $user, $system, $children_user, $children_system, $iters) |
f06db76b AD |
249 | |
250 | in seconds for the whole loop (not divided by the number of rounds). | |
251 | ||
252 | The timing is done using time(3) and times(3). | |
253 | ||
254 | Code is executed in the caller's package. | |
255 | ||
f06db76b AD |
256 | The time of the null loop (a loop with the same |
257 | number of rounds but empty loop body) is subtracted | |
258 | from the time of the real loop. | |
259 | ||
3c6312e9 | 260 | The null loop times can be cached, the key being the |
f06db76b AD |
261 | number of rounds. The caching can be controlled using |
262 | calls like these: | |
263 | ||
523cc92b | 264 | clearcache($key); |
f06db76b AD |
265 | clearallcache(); |
266 | ||
523cc92b | 267 | disablecache(); |
f06db76b AD |
268 | enablecache(); |
269 | ||
3c6312e9 BS |
270 | Caching is off by default, as it can (usually slightly) decrease |
271 | accuracy and does not usually noticably affect runtimes. | |
272 | ||
54e82ce5 GS |
273 | =head1 EXAMPLES |
274 | ||
275 | For example, | |
276 | ||
277 | use Benchmark;$x=3;cmpthese(-5,{a=>sub{$x*$x},b=>sub{$x**2}}) | |
278 | ||
279 | outputs something like this: | |
280 | ||
281 | Benchmark: running a, b, each for at least 5 CPU seconds... | |
282 | a: 10 wallclock secs ( 5.14 usr + 0.13 sys = 5.27 CPU) @ 3835055.60/s (n=20210743) | |
283 | b: 5 wallclock secs ( 5.41 usr + 0.00 sys = 5.41 CPU) @ 1574944.92/s (n=8520452) | |
284 | Rate b a | |
285 | b 1574945/s -- -59% | |
286 | a 3835056/s 144% -- | |
287 | ||
288 | while | |
289 | ||
290 | use Benchmark; | |
291 | $x=3; | |
292 | $r=timethese(-5,{a=>sub{$x*$x},b=>sub{$x**2}},'none'); | |
293 | cmpthese($r); | |
294 | ||
295 | outputs something like this: | |
296 | ||
297 | Rate b a | |
298 | b 1559428/s -- -62% | |
299 | a 4152037/s 166% -- | |
300 | ||
301 | ||
f06db76b AD |
302 | =head1 INHERITANCE |
303 | ||
304 | Benchmark inherits from no other class, except of course | |
305 | for Exporter. | |
306 | ||
307 | =head1 CAVEATS | |
308 | ||
80eab818 | 309 | Comparing eval'd strings with code references will give you |
431d98c2 | 310 | inaccurate results: a code reference will show a slightly slower |
80eab818 CS |
311 | execution time than the equivalent eval'd string. |
312 | ||
f06db76b AD |
313 | The real time timing is done using time(2) and |
314 | the granularity is therefore only one second. | |
315 | ||
316 | Short tests may produce negative figures because perl | |
523cc92b CS |
317 | can appear to take longer to execute the empty loop |
318 | than a short test; try: | |
f06db76b AD |
319 | |
320 | timethis(100,'1'); | |
321 | ||
322 | The system time of the null loop might be slightly | |
323 | more than the system time of the loop with the actual | |
a24a9dfe | 324 | code and therefore the difference might end up being E<lt> 0. |
f06db76b | 325 | |
8a4f6ac2 GS |
326 | =head1 SEE ALSO |
327 | ||
328 | L<Devel::DProf> - a Perl code profiler | |
329 | ||
f06db76b AD |
330 | =head1 AUTHORS |
331 | ||
5aabfad6 | 332 | Jarkko Hietaniemi <F<jhi@iki.fi>>, Tim Bunce <F<Tim.Bunce@ig.co.uk>> |
f06db76b AD |
333 | |
334 | =head1 MODIFICATION HISTORY | |
335 | ||
336 | September 8th, 1994; by Tim Bunce. | |
337 | ||
523cc92b CS |
338 | March 28th, 1997; by Hugo van der Sanden: added support for code |
339 | references and the already documented 'debug' method; revamped | |
340 | documentation. | |
f06db76b | 341 | |
6ee623d5 GS |
342 | April 04-07th, 1997: by Jarkko Hietaniemi, added the run-for-some-time |
343 | functionality. | |
344 | ||
3c6312e9 BS |
345 | September, 1999; by Barrie Slaymaker: math fixes and accuracy and |
346 | efficiency tweaks. Added cmpthese(). A result is now returned from | |
431d98c2 | 347 | timethese(). Exposed countit() (was runfor()). |
3c6312e9 | 348 | |
523cc92b | 349 | =cut |
a0d0e21e | 350 | |
3f943bd9 GS |
351 | # evaluate something in a clean lexical environment |
352 | sub _doeval { eval shift } | |
353 | ||
354 | # | |
355 | # put any lexicals at file scope AFTER here | |
356 | # | |
357 | ||
4aa0a1f7 | 358 | use Carp; |
a0d0e21e LW |
359 | use Exporter; |
360 | @ISA=(Exporter); | |
ac8eabc1 JH |
361 | @EXPORT=qw(timeit timethis timethese timediff timestr); |
362 | @EXPORT_OK=qw(timesum cmpthese countit | |
363 | clearcache clearallcache disablecache enablecache); | |
a0d0e21e | 364 | |
d6a466d7 | 365 | $VERSION = 1.01; |
8a4f6ac2 | 366 | |
a0d0e21e LW |
367 | &init; |
368 | ||
369 | sub init { | |
370 | $debug = 0; | |
371 | $min_count = 4; | |
372 | $min_cpu = 0.4; | |
373 | $defaultfmt = '5.2f'; | |
374 | $defaultstyle = 'auto'; | |
375 | # The cache can cause a slight loss of sys time accuracy. If a | |
376 | # user does many tests (>10) with *very* large counts (>10000) | |
377 | # or works on a very slow machine the cache may be useful. | |
378 | &disablecache; | |
379 | &clearallcache; | |
380 | } | |
381 | ||
523cc92b CS |
382 | sub debug { $debug = ($_[1] != 0); } |
383 | ||
bba8fca5 BS |
384 | # The cache needs two branches: 's' for strings and 'c' for code. The |
385 | # emtpy loop is different in these two cases. | |
386 | sub clearcache { delete $cache{"$_[0]c"}; delete $cache{"$_[0]s"}; } | |
a0d0e21e LW |
387 | sub clearallcache { %cache = (); } |
388 | sub enablecache { $cache = 1; } | |
389 | sub disablecache { $cache = 0; } | |
390 | ||
a0d0e21e LW |
391 | # --- Functions to process the 'time' data type |
392 | ||
6ee623d5 GS |
393 | sub new { my @t = (time, times, @_ == 2 ? $_[1] : 0); |
394 | print "new=@t\n" if $debug; | |
395 | bless \@t; } | |
a0d0e21e LW |
396 | |
397 | sub cpu_p { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps ; } | |
398 | sub cpu_c { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $cu+$cs ; } | |
399 | sub cpu_a { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $pu+$ps+$cu+$cs ; } | |
400 | sub real { my($r,$pu,$ps,$cu,$cs) = @{$_[0]}; $r ; } | |
431d98c2 | 401 | sub iters { $_[0]->[5] ; } |
a0d0e21e | 402 | |
523cc92b | 403 | sub timediff { |
a0d0e21e | 404 | my($a, $b) = @_; |
523cc92b | 405 | my @r; |
3f943bd9 | 406 | for (my $i=0; $i < @$a; ++$i) { |
a0d0e21e LW |
407 | push(@r, $a->[$i] - $b->[$i]); |
408 | } | |
409 | bless \@r; | |
410 | } | |
411 | ||
705cc255 TB |
412 | sub timesum { |
413 | my($a, $b) = @_; | |
414 | my @r; | |
415 | for (my $i=0; $i < @$a; ++$i) { | |
416 | push(@r, $a->[$i] + $b->[$i]); | |
417 | } | |
418 | bless \@r; | |
419 | } | |
420 | ||
523cc92b | 421 | sub timestr { |
a0d0e21e | 422 | my($tr, $style, $f) = @_; |
523cc92b | 423 | my @t = @$tr; |
6ee623d5 GS |
424 | warn "bad time value (@t)" unless @t==6; |
425 | my($r, $pu, $ps, $cu, $cs, $n) = @t; | |
ce9550df | 426 | my($pt, $ct, $tt) = ($tr->cpu_p, $tr->cpu_c, $tr->cpu_a); |
523cc92b | 427 | $f = $defaultfmt unless defined $f; |
a0d0e21e | 428 | # format a time in the required style, other formats may be added here |
80eab818 | 429 | $style ||= $defaultstyle; |
523cc92b CS |
430 | $style = ($ct>0) ? 'all' : 'noc' if $style eq 'auto'; |
431 | my $s = "@t $style"; # default for unknown style | |
7be077a2 | 432 | $s=sprintf("%2d wallclock secs (%$f usr %$f sys + %$f cusr %$f csys = %$f CPU)", |
ce9550df | 433 | $r,$pu,$ps,$cu,$cs,$tt) if $style eq 'all'; |
7be077a2 GS |
434 | $s=sprintf("%2d wallclock secs (%$f usr + %$f sys = %$f CPU)", |
435 | $r,$pu,$ps,$pt) if $style eq 'noc'; | |
436 | $s=sprintf("%2d wallclock secs (%$f cusr + %$f csys = %$f CPU)", | |
437 | $r,$cu,$cs,$ct) if $style eq 'nop'; | |
cc31225e | 438 | $s .= sprintf(" @ %$f/s (n=$n)", $n / ( $pu + $ps )) if $n && $pu+$ps; |
a0d0e21e LW |
439 | $s; |
440 | } | |
523cc92b CS |
441 | |
442 | sub timedebug { | |
a0d0e21e | 443 | my($msg, $t) = @_; |
523cc92b | 444 | print STDERR "$msg",timestr($t),"\n" if $debug; |
a0d0e21e LW |
445 | } |
446 | ||
a0d0e21e LW |
447 | # --- Functions implementing low-level support for timing loops |
448 | ||
449 | sub runloop { | |
450 | my($n, $c) = @_; | |
4aa0a1f7 AD |
451 | |
452 | $n+=0; # force numeric now, so garbage won't creep into the eval | |
523cc92b CS |
453 | croak "negative loopcount $n" if $n<0; |
454 | confess "Usage: runloop(number, [string | coderef])" unless defined $c; | |
a0d0e21e LW |
455 | my($t0, $t1, $td); # before, after, difference |
456 | ||
457 | # find package of caller so we can execute code there | |
523cc92b CS |
458 | my($curpack) = caller(0); |
459 | my($i, $pack)= 0; | |
a0d0e21e LW |
460 | while (($pack) = caller(++$i)) { |
461 | last if $pack ne $curpack; | |
462 | } | |
463 | ||
3f943bd9 GS |
464 | my ($subcode, $subref); |
465 | if (ref $c eq 'CODE') { | |
466 | $subcode = "sub { for (1 .. $n) { local \$_; package $pack; &\$c; } }"; | |
467 | $subref = eval $subcode; | |
468 | } | |
469 | else { | |
470 | $subcode = "sub { for (1 .. $n) { local \$_; package $pack; $c;} }"; | |
471 | $subref = _doeval($subcode); | |
472 | } | |
4aa0a1f7 | 473 | croak "runloop unable to compile '$c': $@\ncode: $subcode\n" if $@; |
523cc92b | 474 | print STDERR "runloop $n '$subcode'\n" if $debug; |
a0d0e21e | 475 | |
3c6312e9 BS |
476 | # Wait for the user timer to tick. This makes the error range more like |
477 | # -0.01, +0. If we don't wait, then it's more like -0.01, +0.01. This | |
478 | # may not seem important, but it significantly reduces the chances of | |
479 | # getting a too low initial $n in the initial, 'find the minimum' loop | |
431d98c2 | 480 | # in &countit. This, in turn, can reduce the number of calls to |
bba8fca5 BS |
481 | # &runloop a lot, and thus reduce additive errors. |
482 | my $tbase = Benchmark->new(0)->[1]; | |
277427cf | 483 | while ( ( $t0 = Benchmark->new(0) )->[1] == $tbase ) {} ; |
a0d0e21e | 484 | &$subref; |
6ee623d5 | 485 | $t1 = Benchmark->new($n); |
a0d0e21e | 486 | $td = &timediff($t1, $t0); |
a0d0e21e LW |
487 | timedebug("runloop:",$td); |
488 | $td; | |
489 | } | |
490 | ||
491 | ||
492 | sub timeit { | |
493 | my($n, $code) = @_; | |
494 | my($wn, $wc, $wd); | |
495 | ||
496 | printf STDERR "timeit $n $code\n" if $debug; | |
3c6312e9 | 497 | my $cache_key = $n . ( ref( $code ) ? 'c' : 's' ); |
bba8fca5 BS |
498 | if ($cache && exists $cache{$cache_key} ) { |
499 | $wn = $cache{$cache_key}; | |
523cc92b | 500 | } else { |
bba8fca5 | 501 | $wn = &runloop($n, ref( $code ) ? sub { undef } : '' ); |
3c6312e9 BS |
502 | # Can't let our baseline have any iterations, or they get subtracted |
503 | # out of the result. | |
504 | $wn->[5] = 0; | |
bba8fca5 | 505 | $cache{$cache_key} = $wn; |
a0d0e21e LW |
506 | } |
507 | ||
508 | $wc = &runloop($n, $code); | |
509 | ||
510 | $wd = timediff($wc, $wn); | |
a0d0e21e LW |
511 | timedebug("timeit: ",$wc); |
512 | timedebug(" - ",$wn); | |
513 | timedebug(" = ",$wd); | |
514 | ||
515 | $wd; | |
516 | } | |
517 | ||
6ee623d5 GS |
518 | |
519 | my $default_for = 3; | |
520 | my $min_for = 0.1; | |
521 | ||
3c6312e9 | 522 | |
431d98c2 BS |
523 | sub countit { |
524 | my ( $tmax, $code ) = @_; | |
6ee623d5 GS |
525 | |
526 | if ( not defined $tmax or $tmax == 0 ) { | |
527 | $tmax = $default_for; | |
528 | } elsif ( $tmax < 0 ) { | |
529 | $tmax = -$tmax; | |
530 | } | |
531 | ||
431d98c2 | 532 | die "countit($tmax, ...): timelimit cannot be less than $min_for.\n" |
6ee623d5 GS |
533 | if $tmax < $min_for; |
534 | ||
3c6312e9 | 535 | my ($n, $tc); |
6ee623d5 | 536 | |
bba8fca5 | 537 | # First find the minimum $n that gives a significant timing. |
3c6312e9 BS |
538 | for ($n = 1; ; $n *= 2 ) { |
539 | my $td = timeit($n, $code); | |
540 | $tc = $td->[1] + $td->[2]; | |
541 | last if $tc > 0.1; | |
542 | } | |
6ee623d5 | 543 | |
3c6312e9 BS |
544 | my $nmin = $n; |
545 | ||
546 | # Get $n high enough that we can guess the final $n with some accuracy. | |
547 | my $tpra = 0.1 * $tmax; # Target/time practice. | |
548 | while ( $tc < $tpra ) { | |
549 | # The 5% fudge is to keep us from iterating again all | |
550 | # that often (this speeds overall responsiveness when $tmax is big | |
551 | # and we guess a little low). This does not noticably affect | |
552 | # accuracy since we're not couting these times. | |
553 | $n = int( $tpra * 1.05 * $n / $tc ); # Linear approximation. | |
554 | my $td = timeit($n, $code); | |
c5d57293 A |
555 | my $new_tc = $td->[1] + $td->[2]; |
556 | # Make sure we are making progress. | |
557 | $tc = $new_tc > 1.2 * $tc ? $new_tc : 1.2 * $tc; | |
6ee623d5 GS |
558 | } |
559 | ||
3c6312e9 BS |
560 | # Now, do the 'for real' timing(s), repeating until we exceed |
561 | # the max. | |
562 | my $ntot = 0; | |
563 | my $rtot = 0; | |
564 | my $utot = 0.0; | |
565 | my $stot = 0.0; | |
566 | my $cutot = 0.0; | |
567 | my $cstot = 0.0; | |
568 | my $ttot = 0.0; | |
569 | ||
570 | # The 5% fudge is because $n is often a few % low even for routines | |
571 | # with stable times and avoiding extra timeit()s is nice for | |
572 | # accuracy's sake. | |
573 | $n = int( $n * ( 1.05 * $tmax / $tc ) ); | |
574 | ||
575 | while () { | |
576 | my $td = timeit($n, $code); | |
577 | $ntot += $n; | |
578 | $rtot += $td->[0]; | |
579 | $utot += $td->[1]; | |
580 | $stot += $td->[2]; | |
6ee623d5 GS |
581 | $cutot += $td->[3]; |
582 | $cstot += $td->[4]; | |
3c6312e9 BS |
583 | $ttot = $utot + $stot; |
584 | last if $ttot >= $tmax; | |
6ee623d5 | 585 | |
c5d57293 | 586 | $ttot = 0.01 if $ttot < 0.01; |
3c6312e9 | 587 | my $r = $tmax / $ttot - 1; # Linear approximation. |
bba8fca5 | 588 | $n = int( $r * $ntot ); |
6ee623d5 | 589 | $n = $nmin if $n < $nmin; |
6ee623d5 GS |
590 | } |
591 | ||
592 | return bless [ $rtot, $utot, $stot, $cutot, $cstot, $ntot ]; | |
593 | } | |
594 | ||
a0d0e21e LW |
595 | # --- Functions implementing high-level time-then-print utilities |
596 | ||
6ee623d5 GS |
597 | sub n_to_for { |
598 | my $n = shift; | |
599 | return $n == 0 ? $default_for : $n < 0 ? -$n : undef; | |
600 | } | |
601 | ||
a0d0e21e LW |
602 | sub timethis{ |
603 | my($n, $code, $title, $style) = @_; | |
6ee623d5 GS |
604 | my($t, $for, $forn); |
605 | ||
606 | if ( $n > 0 ) { | |
607 | croak "non-integer loopcount $n, stopped" if int($n)<$n; | |
608 | $t = timeit($n, $code); | |
609 | $title = "timethis $n" unless defined $title; | |
610 | } else { | |
611 | $fort = n_to_for( $n ); | |
431d98c2 | 612 | $t = countit( $fort, $code ); |
6ee623d5 GS |
613 | $title = "timethis for $fort" unless defined $title; |
614 | $forn = $t->[-1]; | |
615 | } | |
523cc92b | 616 | local $| = 1; |
523cc92b | 617 | $style = "" unless defined $style; |
3c6312e9 BS |
618 | printf("%10s: ", $title) unless $style eq 'none'; |
619 | print timestr($t, $style, $defaultfmt),"\n" unless $style eq 'none'; | |
6ee623d5 GS |
620 | |
621 | $n = $forn if defined $forn; | |
523cc92b | 622 | |
a0d0e21e LW |
623 | # A conservative warning to spot very silly tests. |
624 | # Don't assume that your benchmark is ok simply because | |
625 | # you don't get this warning! | |
626 | print " (warning: too few iterations for a reliable count)\n" | |
523cc92b | 627 | if $n < $min_count |
a0d0e21e | 628 | || ($t->real < 1 && $n < 1000) |
523cc92b | 629 | || $t->cpu_a < $min_cpu; |
a0d0e21e LW |
630 | $t; |
631 | } | |
632 | ||
a0d0e21e LW |
633 | sub timethese{ |
634 | my($n, $alt, $style) = @_; | |
635 | die "usage: timethese(count, { 'Name1'=>'code1', ... }\n" | |
636 | unless ref $alt eq HASH; | |
523cc92b CS |
637 | my @names = sort keys %$alt; |
638 | $style = "" unless defined $style; | |
3c6312e9 | 639 | print "Benchmark: " unless $style eq 'none'; |
6ee623d5 GS |
640 | if ( $n > 0 ) { |
641 | croak "non-integer loopcount $n, stopped" if int($n)<$n; | |
3c6312e9 | 642 | print "timing $n iterations of" unless $style eq 'none'; |
6ee623d5 | 643 | } else { |
3c6312e9 | 644 | print "running" unless $style eq 'none'; |
6ee623d5 | 645 | } |
3c6312e9 | 646 | print " ", join(', ',@names) unless $style eq 'none'; |
6ee623d5 GS |
647 | unless ( $n > 0 ) { |
648 | my $for = n_to_for( $n ); | |
3c6312e9 | 649 | print ", each for at least $for CPU seconds" unless $style eq 'none'; |
6ee623d5 | 650 | } |
3c6312e9 | 651 | print "...\n" unless $style eq 'none'; |
523cc92b CS |
652 | |
653 | # we could save the results in an array and produce a summary here | |
a0d0e21e | 654 | # sum, min, max, avg etc etc |
3c6312e9 | 655 | my %results; |
4dbb2df9 | 656 | foreach my $name (@names) { |
3c6312e9 | 657 | $results{$name} = timethis ($n, $alt -> {$name}, $name, $style); |
4dbb2df9 | 658 | } |
3c6312e9 BS |
659 | |
660 | return \%results; | |
a0d0e21e LW |
661 | } |
662 | ||
3c6312e9 BS |
663 | sub cmpthese{ |
664 | my $results = ref $_[0] ? $_[0] : timethese( @_ ); | |
665 | ||
666 | return $results | |
667 | if defined $_[2] && $_[2] eq 'none'; | |
668 | ||
669 | # Flatten in to an array of arrays with the name as the first field | |
670 | my @vals = map{ [ $_, @{$results->{$_}} ] } keys %$results; | |
671 | ||
672 | for (@vals) { | |
673 | # The epsilon fudge here is to prevent div by 0. Since clock | |
674 | # resolutions are much larger, it's below the noise floor. | |
675 | my $rate = $_->[6] / ( $_->[2] + $_->[3] + 0.000000000000001 ); | |
676 | $_->[7] = $rate; | |
677 | } | |
678 | ||
679 | # Sort by rate | |
680 | @vals = sort { $a->[7] <=> $b->[7] } @vals; | |
681 | ||
682 | # If more than half of the rates are greater than one... | |
683 | my $display_as_rate = $vals[$#vals>>1]->[7] > 1; | |
684 | ||
685 | my @rows; | |
686 | my @col_widths; | |
687 | ||
688 | my @top_row = ( | |
689 | '', | |
690 | $display_as_rate ? 'Rate' : 's/iter', | |
691 | map { $_->[0] } @vals | |
692 | ); | |
693 | ||
694 | push @rows, \@top_row; | |
695 | @col_widths = map { length( $_ ) } @top_row; | |
696 | ||
697 | # Build the data rows | |
698 | # We leave the last column in even though it never has any data. Perhaps | |
699 | # it should go away. Also, perhaps a style for a single column of | |
700 | # percentages might be nice. | |
701 | for my $row_val ( @vals ) { | |
702 | my @row; | |
703 | ||
704 | # Column 0 = test name | |
705 | push @row, $row_val->[0]; | |
706 | $col_widths[0] = length( $row_val->[0] ) | |
707 | if length( $row_val->[0] ) > $col_widths[0]; | |
708 | ||
709 | # Column 1 = performance | |
710 | my $row_rate = $row_val->[7]; | |
711 | ||
712 | # We assume that we'll never get a 0 rate. | |
713 | my $a = $display_as_rate ? $row_rate : 1 / $row_rate; | |
714 | ||
715 | # Only give a few decimal places before switching to sci. notation, | |
716 | # since the results aren't usually that accurate anyway. | |
717 | my $format = | |
718 | $a >= 100 ? | |
719 | "%0.0f" : | |
720 | $a >= 10 ? | |
721 | "%0.1f" : | |
722 | $a >= 1 ? | |
723 | "%0.2f" : | |
724 | $a >= 0.1 ? | |
725 | "%0.3f" : | |
726 | "%0.2e"; | |
727 | ||
728 | $format .= "/s" | |
729 | if $display_as_rate; | |
730 | # Using $b here due to optimizing bug in _58 through _61 | |
731 | my $b = sprintf( $format, $a ); | |
732 | push @row, $b; | |
733 | $col_widths[1] = length( $b ) | |
734 | if length( $b ) > $col_widths[1]; | |
735 | ||
736 | # Columns 2..N = performance ratios | |
737 | my $skip_rest = 0; | |
738 | for ( my $col_num = 0 ; $col_num < @vals ; ++$col_num ) { | |
739 | my $col_val = $vals[$col_num]; | |
740 | my $out; | |
741 | if ( $skip_rest ) { | |
742 | $out = ''; | |
743 | } | |
744 | elsif ( $col_val->[0] eq $row_val->[0] ) { | |
745 | $out = "--"; | |
746 | # $skip_rest = 1; | |
747 | } | |
748 | else { | |
749 | my $col_rate = $col_val->[7]; | |
750 | $out = sprintf( "%.0f%%", 100*$row_rate/$col_rate - 100 ); | |
751 | } | |
752 | push @row, $out; | |
753 | $col_widths[$col_num+2] = length( $out ) | |
754 | if length( $out ) > $col_widths[$col_num+2]; | |
755 | ||
756 | # A little wierdness to set the first column width properly | |
757 | $col_widths[$col_num+2] = length( $col_val->[0] ) | |
758 | if length( $col_val->[0] ) > $col_widths[$col_num+2]; | |
759 | } | |
760 | push @rows, \@row; | |
761 | } | |
762 | ||
763 | # Equalize column widths in the chart as much as possible without | |
764 | # exceeding 80 characters. This does not use or affect cols 0 or 1. | |
765 | my @sorted_width_refs = | |
766 | sort { $$a <=> $$b } map { \$_ } @col_widths[2..$#col_widths]; | |
767 | my $max_width = ${$sorted_width_refs[-1]}; | |
768 | ||
277427cf | 769 | my $total = @col_widths - 1 ; |
3c6312e9 BS |
770 | for ( @col_widths ) { $total += $_ } |
771 | ||
772 | STRETCHER: | |
773 | while ( $total < 80 ) { | |
774 | my $min_width = ${$sorted_width_refs[0]}; | |
775 | last | |
776 | if $min_width == $max_width; | |
777 | for ( @sorted_width_refs ) { | |
778 | last | |
779 | if $$_ > $min_width; | |
780 | ++$$_; | |
781 | ++$total; | |
782 | last STRETCHER | |
783 | if $total >= 80; | |
784 | } | |
785 | } | |
786 | ||
787 | # Dump the output | |
788 | my $format = join( ' ', map { "%${_}s" } @col_widths ) . "\n"; | |
789 | substr( $format, 1, 0 ) = '-'; | |
790 | for ( @rows ) { | |
791 | printf $format, @$_; | |
792 | } | |
793 | ||
794 | return $results; | |
795 | } | |
796 | ||
797 | ||
a0d0e21e | 798 | 1; |