timeit
Time a snippet.
Usage
var timeit = require( '@stdlib/utils/timeit' );
timeit( code, [options,] clbk )
Times a snippet.
var code = 'var x = Math.pow( Math.random(), 3 );';
code += 'if ( x !== x ) {';
code += 'throw new Error( \'Something went wrong.\' );';
code += '}';
timeit( code, done );
function done( error, results ) {
if ( error ) {
throw error;
}
console.dir( results );
/* e.g., =>
{
'iterations': 1000000,
'repeats': 3,
'min': [0,135734733], // [seconds,nanoseconds]
'elapsed': 0.135734733, // seconds
'rate': 7367311.062526641, // iterations/second
'times': [ // raw timing results
[0,145641393],
[0,135734733],
[0,140462721]
]
}
*/
}
The function supports the following options
:
- before: setup code. Default:
""
. - after: cleanup code. Default:
""
. - iterations: number of iterations. If
null
, the number of iterations is determined by trying successive powers of10
until the total time is at least0.1
seconds. Default:1e6
. - repeats: number of repeats. Default:
3
. - asynchronous:
boolean
indicating whether a snippet is asynchronous. Default:false
.
To perform any setup or initialization, provide setup code.
var setup = 'var randu = require( \'@stdlib/random/base/randu\' );';
setup += 'var pow = require( \'@stdlib/math/base/special/pow\' );';
var code = 'var x = pow( randu(), 3 );';
code += 'if ( x !== x ) {';
code += 'throw new Error( \'Something went wrong.\' );';
code += '}';
var opts = {
'before': setup
};
timeit( code, opts, done );
function done( error, results ) {
if ( error ) {
throw error;
}
console.dir( results );
}
To perform any cleanup, provide cleanup code.
var setup = 'var randu = require( \'@stdlib/random/base/randu\' );';
setup += 'var hypot = require( \'@stdlib/math/base/special/hypot\' );';
var code = 'var h = hypot( randu()*10, randu()*10 );';
code += 'if ( h < 0 || h > 200 ) {';
code += 'throw new Error( \'Something went wrong.\' );';
code += '}';
var cleanup = 'if ( h !== h ) {';
cleanup += 'throw new Error( \'Something went wrong.\' );';
cleanup += '}';
var opts = {
'before': setup,
'after': cleanup
};
timeit( code, opts, done );
function done( error, results ) {
if ( error ) {
throw error;
}
console.dir( results );
}
To time an asynchronous snippet, set the asynchronous
option to true
.
var code = 'var x = Math.pow( Math.random(), 3 );';
code += 'if ( x !== x ) {';
code += 'var err = new Error( \'Something went wrong.\' );';
code += 'next( err );';
code += '}';
code += 'process.nextTick( next );';
var opts = {
'iterations': 1e2,
'asynchronous': true
};
timeit( code, opts, done );
function done( error, results ) {
if ( error ) {
throw error;
}
console.dir( results );
}
If asynchronous
is true
, the implementation assumes that before
, after
, and code
snippets are all asynchronous. Accordingly, these snippets should invoke a next( [error] )
callback once complete. For example, given the following snippet,
setTimeout( done, 0 );
function done( error ) {
if ( error ) {
return next( error );
}
next();
}
the implementation wraps the snippet within a function having the following signature
function wrapped( state, next ) {
setTimeout( done, 0 );
function done( error ) {
if ( error ) {
return next( error );
}
next();
}
}
The state
parameter is simply an empty {}
which allows the before
, after
, and code
snippets to share state.
function before( state, next ) {
state.counter = 0;
}
function code( state, next ) {
setTimeout( done, 0 );
function done( error ) {
if ( error ) {
return next( error );
}
state.counter += 1;
next();
}
}
function after( state, next ) {
var err;
if ( state.counter !== state.counter ) {
err = new Error( 'Something went wrong!' );
return next( err );
}
next();
}
Notes
- Snippets always run in strict mode.
- Always verify results. Doing so prevents the compiler from performing dead code elimination and other optimization techniques, which would render timing results meaningless.
- Executed code is not sandboxed and has access to the global state. You are strongly advised against timing untrusted code. To time untrusted code, do so in an isolated environment (e.g., a separate process with restricted access to both global state and the host environment).
- Wrapping asynchronous code does add overhead, but, in most cases, the overhead should be negligible compared to the execution cost of the timed snippet.
- Ensure that, when
asynchronous
istrue
, the maincode
snippet is actually asynchronous. If a snippet releases the zalgo, an error complaining about exceeding the maximum call stack size is highly likely. - While many benchmark frameworks calculate various statistics over raw timing results (e.g., mean and standard deviation), do not do this. Instead, consider the fastest time an approximate lower bound for how fast an environment can execute a snippet. Slower times are more likely attributable to other processes interfering with timing accuracy rather than attributable to variability in JavaScript's speed. In which case, the minimum time is most likely the only result of interest. When considering all raw timing results, apply common sense rather than statistics.
Examples
var join = require( 'path' ).join;
var readFileSync = require( '@stdlib/fs/read-file' ).sync;
var timeit = require( '@stdlib/utils/timeit' );
var before = readFileSync( join( __dirname, 'examples', 'before.txt' ), 'utf8' );
var code = readFileSync( join( __dirname, 'examples', 'code.txt' ), 'utf8' );
var opts = {
'iterations': 1e6,
'repeats': 5,
'before': before
};
timeit( code, opts, done );
function done( error, results ) {
if ( error ) {
throw error;
}
console.dir( results );
}
CLI
Usage
Usage: timeit [options] [<code>]
Options:
-h, --help Print this message.
-V, --version Print the package version.
--iterations iter Number of iterations.
--repeats repeats Number of repeats. Default: 3.
--before setup Setup code.
--after cleanup Cleanup code.
--async Time asynchronous code.
--format fmt Output format: pretty, csv, json. Default: pretty.
Notes
- When the output format is
csv
, the output consists of only raw timing results. - If not explicitly provided
--iterations
, the implementation tries successive powers of10
until the total time is at least0.1
seconds.
Examples
$ timeit "$(cat ./examples/code.txt)" --before "$(cat ./examples/before.txt)" --iterations 1000000
iterations: 1000000
repeats: 3
iterations/s: 7261975.851461222
elapsed time: 0.13770357 sec
lower bound: 0.13770357 usec/iteration
To output results as JSON,
$ timeit "$(cat ./examples/code.txt)" --before "$(cat ./examples/before.txt)" --iterations 1000000 --format json
{"iterations":1000000,"repeats":3,"min":[0,132431806],"elapsed":0.132431806,"rate":7551056.1261997735,"times":[[0,142115140],[0,132431806],[0,134808376]]}
To output results as comma-separated values (CSV),
$ timeit "$(cat ./examples/code.txt)" --before "$(cat ./examples/before.txt)" --iterations 1000000 --format csv
seconds,nanoseconds
0,139365407
0,138033545
0,135175834
To use as part of a pipeline,
$ cat ./examples/code.txt | timeit --before "$(cat ./examples/before.txt)" --iterations 1000000
iterations: 1000000
repeats: 3
iterations/s: 7433536.674260073
elapsed time: 0.134525468 sec
lower bound: 0.134525468 usec/iteration
References
- Chen, Jiahao, and Jarrett Revels. 2016. "Robust benchmarking in noisy environments." CoRR abs/1608.04295 (August). http://arxiv.org/abs/1608.04295.