1Benchmark(3pm) Perl Programmers Reference Guide Benchmark(3pm)
2
6 Benchmark - benchmark running times of Perl code
7
9 use Benchmark qw(:all) ;
10 timethis ($count, "code");
12 # Use Perl code in strings...
14 timethese($count, {
15 'Name1' => '...code1...',
16 'Name2' => '...code2...',
17 });
18 # ... or use subroutine references.
20 timethese($count, {
21 'Name1' => sub { ...code1... },
22 'Name2' => sub { ...code2... },
23 });
24 # cmpthese can be used both ways as well
26 cmpthese($count, {
27 'Name1' => '...code1...',
28 'Name2' => '...code2...',
29 });
30 cmpthese($count, {
32 'Name1' => sub { ...code1... },
33 'Name2' => sub { ...code2... },
34 });
35 # ...or in two stages
37 $results = timethese($count,
38 {
39 'Name1' => sub { ...code1... },
40 'Name2' => sub { ...code2... },
41 },
42 'none'
43 );
44 cmpthese( $results ) ;
45 $t = timeit($count, '...other code...')
47 print "$count loops of other code took:",timestr($t),"\n";
48 $t = countit($time, '...other code...')
50 $count = $t->iters ;
51 print "$count loops of other code took:",timestr($t),"\n";
52 # enable hires wallclock timing if possible
54 use Benchmark ':hireswallclock';
55
57 The Benchmark module encapsulates a number of routines to help you fig‐
58 ure out how long it takes to execute some code.
59 timethis - run a chunk of code several times
61 timethese - run several chunks of code several times
63 cmpthese - print results of timethese as a comparison chart
65 timeit - run a chunk of code and see how long it goes
67 countit - see how many times a chunk of code runs in a given time
69 Methods
71 new Returns the current time. Example:
73 use Benchmark;
75 $t0 = new Benchmark;
76 # ... your code here ...
77 $t1 = new Benchmark;
78 $td = timediff($t1, $t0);
79 print "the code took:",timestr($td),"\n";
80 debug Enables or disable debugging by setting the $Benchmark::Debug
82 flag:
83 debug Benchmark 1;
85 $t = timeit(10, ' 5 ** $Global ');
86 debug Benchmark 0;
87 iters Returns the number of iterations.
89 Standard Exports
91 The following routines will be exported into your namespace if you use
93 the Benchmark module:
94 timeit(COUNT, CODE)
96 Arguments: COUNT is the number of times to run the loop, and
97 CODE is the code to run. CODE may be either a code reference
98 or a string to be eval'd; either way it will be run in the
99 caller's package.
100 Returns: a Benchmark object.
102 timethis ( COUNT, CODE, [ TITLE, [ STYLE ]] )
104 Time COUNT iterations of CODE. CODE may be a string to eval
105 or a code reference; either way the CODE will run in the
106 caller's package. Results will be printed to STDOUT as TITLE
107 followed by the times. TITLE defaults to "timethis COUNT" if
108 none is provided. STYLE determines the format of the output,
109 as described for timestr() below.
110 The COUNT can be zero or negative: this means the minimum
112 number of CPU seconds to run. A zero signifies the default
113 of 3 seconds. For example to run at least for 10 seconds:
114 timethis(-10, $code)
116 or to run two pieces of code tests for at least 3 seconds:
118 timethese(0, { test1 => '...', test2 => '...'})
120 CPU seconds is, in UNIX terms, the user time plus the system
122 time of the process itself, as opposed to the real (wall‐
123 clock) time and the time spent by the child processes. Less
124 than 0.1 seconds is not accepted (-0.01 as the count, for
125 example, will cause a fatal runtime exception).
126 Note that the CPU seconds is the minimum time: CPU scheduling
128 and other operating system factors may complicate the attempt
129 so that a little bit more time is spent. The benchmark out‐
130 put will, however, also tell the number of $code runs/second,
131 which should be a more interesting number than the actually
132 spent seconds.
133 Returns a Benchmark object.
135 timethese ( COUNT, CODEHASHREF, [ STYLE ] )
137 The CODEHASHREF is a reference to a hash containing names as
138 keys and either a string to eval or a code reference for each
139 value. For each (KEY, VALUE) pair in the CODEHASHREF, this
140 routine will call
141 timethis(COUNT, VALUE, KEY, STYLE)
143 The routines are called in string comparison order of KEY.
145 The COUNT can be zero or negative, see timethis().
147 Returns a hash of Benchmark objects, keyed by name.
149 timediff ( T1, T2 )
151 Returns the difference between two Benchmark times as a
152 Benchmark object suitable for passing to timestr().
153 timestr ( TIMEDIFF, [ STYLE, [ FORMAT ] ] )
155 Returns a string that formats the times in the TIMEDIFF
156 object in the requested STYLE. TIMEDIFF is expected to be a
157 Benchmark object similar to that returned by timediff().
158 STYLE can be any of 'all', 'none', 'noc', 'nop' or 'auto'.
160 'all' shows each of the 5 times available ('wallclock' time,
161 user time, system time, user time of children, and system
162 time of children). 'noc' shows all except the two children
163 times. 'nop' shows only wallclock and the two children times.
164 'auto' (the default) will act as 'all' unless the children
165 times are both zero, in which case it acts as 'noc'. 'none'
166 prevents output.
167 FORMAT is the printf(3)-style format specifier (without the
169 leading '%') to use to print the times. It defaults to
170 '5.2f'.
171 Optional Exports
173 The following routines will be exported into your namespace if you
175 specifically ask that they be imported:
176 clearcache ( COUNT )
178 Clear the cached time for COUNT rounds of the null loop.
179 clearallcache ( )
181 Clear all cached times.
182 cmpthese ( COUNT, CODEHASHREF, [ STYLE ] )
184 cmpthese ( RESULTSHASHREF, [ STYLE ] )
185 Optionally calls timethese(), then outputs comparison chart.
186 This:
187 cmpthese( -1, { a => "++\$i", b => "\$i *= 2" } ) ;
189 outputs a chart like:
191 Rate b a
193 b 2831802/s -- -61%
194 a 7208959/s 155% --
195 This chart is sorted from slowest to fastest, and shows the
197 percent speed difference between each pair of tests.
198 c<cmpthese> can also be passed the data structure that
200 timethese() returns:
201 $results = timethese( -1, { a => "++\$i", b => "\$i *= 2" } ) ;
203 cmpthese( $results );
204 in case you want to see both sets of results.
206 Returns a reference to an ARRAY of rows, each row is an ARRAY
208 of cells from the above chart, including labels. This:
209 my $rows = cmpthese( -1, { a => '++$i', b => '$i *= 2' }, "none" );
211 returns a data structure like:
213 [
215 [ '', 'Rate', 'b', 'a' ],
216 [ 'b', '2885232/s', '--', '-59%' ],
217 [ 'a', '7099126/s', '146%', '--' ],
218 ]
219 NOTE: This result value differs from previous versions, which
221 returned the "timethese()" result structure. If you want
222 that, just use the two statement "timethese"..."cmpthese"
223 idiom shown above.
224 Incidently, note the variance in the result values between
226 the two examples; this is typical of benchmarking. If this
227 were a real benchmark, you would probably want to run a lot
228 more iterations.
229 countit(TIME, CODE)
231 Arguments: TIME is the minimum length of time to run CODE
232 for, and CODE is the code to run. CODE may be either a code
233 reference or a string to be eval'd; either way it will be run
234 in the caller's package.
235 TIME is not negative. countit() will run the loop many times
237 to calculate the speed of CODE before running it for TIME.
238 The actual time run for will usually be greater than TIME due
239 to system clock resolution, so it's best to look at the num‐
240 ber of iterations divided by the times that you are concerned
241 with, not just the iterations.
242 Returns: a Benchmark object.
244 disablecache ( )
246 Disable caching of timings for the null loop. This will force
247 Benchmark to recalculate these timings for each new piece of
248 code timed.
249 enablecache ( )
251 Enable caching of timings for the null loop. The time taken
252 for COUNT rounds of the null loop will be calculated only
253 once for each different COUNT used.
254 timesum ( T1, T2 )
256 Returns the sum of two Benchmark times as a Benchmark object
257 suitable for passing to timestr().
258 :hireswallclock
260 If the Time::HiRes module has been installed, you can specify the spe‐
262 cial tag ":hireswallclock" for Benchmark (if Time::HiRes is not avail‐
263 able, the tag will be silently ignored). This tag will cause the wall‐
264 clock time to be measured in microseconds, instead of integer seconds.
265 Note though that the speed computations are still conducted in CPU
266 time, not wallclock time.
267
269 The data is stored as a list of values from the time and times func‐
270 tions:
271 ($real, $user, $system, $children_user, $children_system, $iters)
273 in seconds for the whole loop (not divided by the number of rounds).
275 The timing is done using time(3) and times(3).
277 Code is executed in the caller's package.
279 The time of the null loop (a loop with the same number of rounds but
281 empty loop body) is subtracted from the time of the real loop.
282 The null loop times can be cached, the key being the number of rounds.
284 The caching can be controlled using calls like these:
285 clearcache($key);
287 clearallcache();
288 disablecache();
290 enablecache();
291 Caching is off by default, as it can (usually slightly) decrease accu‐
293 racy and does not usually noticably affect runtimes.
294
296 For example,
297 use Benchmark qw( cmpthese ) ;
299 $x = 3;
300 cmpthese( -5, {
301 a => sub{$x*$x},
302 b => sub{$x**2},
303 } );
304 outputs something like this:
306 Benchmark: running a, b, each for at least 5 CPU seconds...
308 Rate b a
309 b 1559428/s -- -62%
310 a 4152037/s 166% --
311 while
313 use Benchmark qw( timethese cmpthese ) ;
315 $x = 3;
316 $r = timethese( -5, {
317 a => sub{$x*$x},
318 b => sub{$x**2},
319 } );
320 cmpthese $r;
321 outputs something like this:
323 Benchmark: running a, b, each for at least 5 CPU seconds...
325 a: 10 wallclock secs ( 5.14 usr + 0.13 sys = 5.27 CPU) @ 3835055.60/s (n=20210743)
326 b: 5 wallclock secs ( 5.41 usr + 0.00 sys = 5.41 CPU) @ 1574944.92/s (n=8520452)
327 Rate b a
328 b 1574945/s -- -59%
329 a 3835056/s 144% --
330
332 Benchmark inherits from no other class, except of course for Exporter.
333
335 Comparing eval'd strings with code references will give you inaccurate
336 results: a code reference will show a slightly slower execution time
337 than the equivalent eval'd string.
338 The real time timing is done using time(2) and the granularity is
340 therefore only one second.
341 Short tests may produce negative figures because perl can appear to
343 take longer to execute the empty loop than a short test; try:
344 timethis(100,'1');
346 The system time of the null loop might be slightly more than the system
348 time of the loop with the actual code and therefore the difference
349 might end up being < 0.
350
352 Devel::DProf - a Perl code profiler
353
355 Jarkko Hietaniemi <jhi@iki.fi>, Tim Bunce <Tim.Bunce@ig.co.uk>
356
358 September 8th, 1994; by Tim Bunce.
359 March 28th, 1997; by Hugo van der Sanden: added support for code refer‐
361 ences and the already documented 'debug' method; revamped documenta‐
362 tion.
363 April 04-07th, 1997: by Jarkko Hietaniemi, added the run-for-some-time
365 functionality.
366 September, 1999; by Barrie Slaymaker: math fixes and accuracy and effi‐
368 ciency tweaks. Added cmpthese(). A result is now returned from
369 timethese(). Exposed countit() (was runfor()).
370 December, 2001; by Nicholas Clark: make timestr() recognise the style
372 'none' and return an empty string. If cmpthese is calling timethese,
373 make it pass the style in. (so that 'none' will suppress output). Make
374 sub new dump its debugging output to STDERR, to be consistent with
375 everything else. All bugs found while writing a regression test.
376 September, 2002; by Jarkko Hietaniemi: add ':hireswallclock' special
378 tag.
379 February, 2004; by Chia-liang Kao: make cmpthese and timestr use time
381 statistics for children instead of parent when the style is 'nop'.
382perl v5.8.8 2001-09-21 Benchmark(3pm)