How can we measure the time elapsed for running a function in Swift? I am trying to display the elapsed time like this: "Elapsed time is .05 seconds". Saw that in Java, we can use System.nanoTime(), is there any equivalent methods are available in Swift to accomplish this?
Please have a look at the sample program:
func isPrime(_ number: Int) -> Bool {
var i = 0;
for i=2; i<number; i++ {
if number % i == 0, i != 0 {
return false
}
}
return true
}
var number = 5915587277
if isPrime(number) {
print("Prime number")
} else {
print("NOT a prime number")
}
sqrt(number)
instead of number
, and you can save a little more time – but there are much more ideas optimising seeking primes.
NSDate
objects and you can measure the difference between them.
Here's a Swift function I wrote to measure Project Euler problems in Swift
As of Swift 3, there is now a version of Grand Central Dispatch that is "swiftified". So the correct answer is probably to use the DispatchTime API.
My function would look something like:
// Swift 3
func evaluateProblem(problemNumber: Int, problemBlock: () -> Int) -> Answer
{
print("Evaluating problem \(problemNumber)")
let start = DispatchTime.now() // <<<<<<<<<< Start time
let myGuess = problemBlock()
let end = DispatchTime.now() // <<<<<<<<<< end time
let theAnswer = self.checkAnswer(answerNum: "\(problemNumber)", guess: myGuess)
let nanoTime = end.uptimeNanoseconds - start.uptimeNanoseconds // <<<<< Difference in nano seconds (UInt64)
let timeInterval = Double(nanoTime) / 1_000_000_000 // Technically could overflow for long running tests
print("Time to evaluate problem \(problemNumber): \(timeInterval) seconds")
return theAnswer
}
Old answer
For Swift 1 and 2, my function uses NSDate:
// Swift 1
func evaluateProblem(problemNumber: Int, problemBlock: () -> Int) -> Answer
{
println("Evaluating problem \(problemNumber)")
let start = NSDate() // <<<<<<<<<< Start time
let myGuess = problemBlock()
let end = NSDate() // <<<<<<<<<< end time
let theAnswer = self.checkAnswer(answerNum: "\(problemNumber)", guess: myGuess)
let timeInterval: Double = end.timeIntervalSinceDate(start) // <<<<< Difference in seconds (double)
println("Time to evaluate problem \(problemNumber): \(timeInterval) seconds")
return theAnswer
}
Note that using NSdate for timing functions is discouraged: "The system time may decrease due to synchronization with external time references or due to an explicit user change of the clock.".
This is a handy timer class based on CoreFoundation
s CFAbsoluteTime
:
import CoreFoundation
class ParkBenchTimer {
let startTime: CFAbsoluteTime
var endTime: CFAbsoluteTime?
init() {
startTime = CFAbsoluteTimeGetCurrent()
}
func stop() -> CFAbsoluteTime {
endTime = CFAbsoluteTimeGetCurrent()
return duration!
}
var duration: CFAbsoluteTime? {
if let endTime = endTime {
return endTime - startTime
} else {
return nil
}
}
}
You can use it like this:
let timer = ParkBenchTimer()
// ... a long runnig task ...
print("The task took \(timer.stop()) seconds.")
mach_absolute_time
which does not suffer from this problem, so I am wondering why that one is not widely known. Maybe just because it is not well documented.
Use clock, ProcessInfo.systemUptime, or DispatchTime for simple start-up time.
There are, as far as I know, at least ten ways to measure elapsed time:
Monotonic Clock based:
ProcessInfo.systemUptime. mach_absolute_time with mach_timebase_info as mentioned in this answer. clock() in POSIX standard. times() in POSIX standard. (Too complicated since we need to consider user-time v.s. system-time, and child processes are involved.) DispatchTime (a wrapper around Mach time API) as mentioned by JeremyP in accepted answer. CACurrentMediaTime().
Wall Clock based:
(never use those for metrics: see below why)
NSDate/Date as mentioned by others. CFAbsoluteTime as mentioned by others. DispatchWallTime. gettimeofday() in POSIX standard.
Option 1, 2 and 3 are elaborated below.
Option 1: Process Info API in Foundation
do {
let info = ProcessInfo.processInfo
let begin = info.systemUptime
// do something
let diff = (info.systemUptime - begin)
}
where diff:NSTimeInterval
is the elapsed time by seconds.
Option 2: Mach C API
do {
var info = mach_timebase_info(numer: 0, denom: 0)
mach_timebase_info(&info)
let begin = mach_absolute_time()
// do something
let diff = Double(mach_absolute_time() - begin) * Double(info.numer) / Double(info.denom)
}
where diff:Double
is the elapsed time by nano-seconds.
Option 3: POSIX clock API
do {
let begin = clock()
// do something
let diff = Double(clock() - begin) / Double(CLOCKS_PER_SEC)
}
where diff:Double
is the elapsed time by seconds.
Why Not Wall-Clock Time for Elapsed Time?
In documentation of CFAbsoluteTimeGetCurrent
:
Repeated calls to this function do not guarantee monotonically increasing results.
Reason is similar to currentTimeMillis
vs nanoTime
in Java:
You can't use the one for the other purpose. The reason is that no computer's clock is perfect; it always drifts and occasionally needs to be corrected. This correction might either happen manually, or in the case of most machines, there's a process that runs and continually issues small corrections to the system clock ("wall clock"). These tend to happen often. Another such correction happens whenever there is a leap second.
Here CFAbsoluteTime
provides wall clock time instead of start-up time. NSDate
is wall clock time as well.
clock()
function.
ProcessInfo.processInfo.systemUptime
, mach_absolute_time()
, DispatchTime.now()
and CACurrentMediaTime()
. But the solution with clock()
was not converting to seconds properly. So I would advise to update your first sentence to: "Use ProcessInfo.systemUptime or DispatchTime for accurate start-up time."
clock_gettime_nsec_np(CLOCK_UPTIME_RAW)
which does the time base conversion for you and provides nanosec value (see manpagez.com/man/3/clock_gettime_nsec_np)
Swift 4 shortest answer:
let startingPoint = Date()
// ... intensive task
print("\(startingPoint.timeIntervalSinceNow * -1) seconds elapsed")
It will print you something like 1.02107906341553 seconds elapsed (time of course will vary depending on the task, I'm just showing this for you guys to see the decimal precision level for this measurement).
Hope it helps someone in Swift 4 from now on!
Update
If you want to have a generic way of testing portions of code, I'd suggest the next snippet:
func measureTime(for closure: @autoclosure () -> Any) {
let start = CFAbsoluteTimeGetCurrent()
closure()
let diff = CFAbsoluteTimeGetCurrent() - start
print("Took \(diff) seconds")
}
Usage
measureTime(for: <insert method signature here>)
Console log
Took xx.xxxxx seconds
Just Copy and Paste this function. Written in swift 5. Copying JeremyP here.
func calculateTime(block : (() -> Void)) {
let start = DispatchTime.now()
block()
let end = DispatchTime.now()
let nanoTime = end.uptimeNanoseconds - start.uptimeNanoseconds
let timeInterval = Double(nanoTime) / 1_000_000_000
print("Time: \(timeInterval) seconds")
}
Use it like
calculateTime {
exampleFunc()// function whose execution time to be calculated
}
let start = NSDate()
for index in 1...10000 {
// do nothing
}
let elapsed = start.timeIntervalSinceNow
// elapsed is a negative value.
You could create a time
function for measuring you calls. I am inspired by Klaas' answer.
func time <A> (f: @autoclosure () -> A) -> (result:A, duration: String) {
let startTime = CFAbsoluteTimeGetCurrent()
let result = f()
let endTime = CFAbsoluteTimeGetCurrent()
return (result, "Elapsed time is \(endTime - startTime) seconds.")
}
This function would allow you to call it like this time (isPrime(7))
which would return a tuple containing the result and a string description of the elapsed time.
If you only wish the elapsed time you can do this time (isPrime(7)).duration
It looks like iOS 13 introduced a new API to use with DispatchTime
that removes a need to calculate the difference between two timestamps manually.
let start: DispatchTime = .now()
heavyTaskToMeasure()
let duration = start.distance(to: .now())
print(duration)
// prints: nanoseconds(NUMBER_OF_NANOSECONDS_BETWEEN_TWO_TIMESTAMPS)
Sadly the documentation is not provided, but after doing some tests it looks like the .nanoseconds
case is always returned.
With a simple extension you could convert the DispatchTimeInterval
to TimeInterval
. credit
extension TimeInterval {
init?(dispatchTimeInterval: DispatchTimeInterval) {
switch dispatchTimeInterval {
case .seconds(let value):
self = Double(value)
case .milliseconds(let value):
self = Double(value) / 1_000
case .microseconds(let value):
self = Double(value) / 1_000_000
case .nanoseconds(let value):
self = Double(value) / 1_000_000_000
case .never:
return nil
}
}
}
Simple helper function for measuring execution time with closure.
func printExecutionTime(withTag tag: String, of closure: () -> ()) {
let start = CACurrentMediaTime()
closure()
print("#\(tag) - execution took \(CACurrentMediaTime() - start) seconds")
}
Usage:
printExecutionTime(withTag: "Init") {
// Do your work here
}
Result: #Init - execution took 1.00104497105349 seconds
you can measure the nanoseconds like e.g. this:
let startDate: NSDate = NSDate()
// your long procedure
let endDate: NSDate = NSDate()
let dateComponents: NSDateComponents = NSCalendar(calendarIdentifier: NSCalendarIdentifierGregorian).components(NSCalendarUnit.CalendarUnitNanosecond, fromDate: startDate, toDate: endDate, options: NSCalendarOptions(0))
println("runtime is nanosecs : \(dateComponents.nanosecond)")
NSCalendar
is expensive procedure, but you can do it before you start to run your procedure, so it won't be added to your long procedure's runtime... bear in mind calling creating an NSDate
instance and calling –components(_:, fromDate:)
method still takes time, so probably you will never get 0.0
nanoseconds.
6.9
microseconds on my device.
I use this:
public class Stopwatch {
public init() { }
private var start_: NSTimeInterval = 0.0;
private var end_: NSTimeInterval = 0.0;
public func start() {
start_ = NSDate().timeIntervalSince1970;
}
public func stop() {
end_ = NSDate().timeIntervalSince1970;
}
public func durationSeconds() -> NSTimeInterval {
return end_ - start_;
}
}
I don't know if it's more or less accurate than previously posted. But the seconds have a lot of decimals and seem to catch small code changes in algorithms like QuickSort using swap() vs. implementing swap urself etc.
Remember to crank up your build optimizations when testing performance:
https://i.stack.imgur.com/u5hw3.png
Here is my try for the simplest answer:
let startTime = Date().timeIntervalSince1970 // 1512538946.5705 seconds
// time passes (about 10 seconds)
let endTime = Date().timeIntervalSince1970 // 1512538956.57195 seconds
let elapsedTime = endTime - startTime // 10.0014500617981 seconds
Notes
startTime and endTime are of the type TimeInterval, which is just a typealias for Double, so it is easy to convert it to an Int or whatever. Time is measured in seconds with sub-millisecond precision.
See also DateInterval, which includes an actual start and end time.
Using the time since 1970 is similar to Java timestamps.
I have borrowed the idea from Klaas to create a lightweight struct to measure running and interval time:
Code Usage:
var timer = RunningTimer.init()
// Code to be timed
print("Running: \(timer) ") // Gives time interval
// Second code to be timed
print("Running: \(timer) ") // Gives final time
The stop function does not have to be called, as the print function will give the time lapsed. It may be called repeatedly to get the time lapsed. But to stop the timer at certain point in the code use timer.stop()
it may also be used to return the time in seconds: let seconds = timer.stop()
After the timer is stopped the interval timer will not, so the print("Running: \(timer) ")
will give the correct time even after a few lines of code.
Following is the code for RunningTimer. It is tested for Swift 2.1:
import CoreFoundation
// Usage: var timer = RunningTimer.init()
// Start: timer.start() to restart the timer
// Stop: timer.stop() returns the time and stops the timer
// Duration: timer.duration returns the time
// May also be used with print(" \(timer) ")
struct RunningTimer: CustomStringConvertible {
var begin:CFAbsoluteTime
var end:CFAbsoluteTime
init() {
begin = CFAbsoluteTimeGetCurrent()
end = 0
}
mutating func start() {
begin = CFAbsoluteTimeGetCurrent()
end = 0
}
mutating func stop() -> Double {
if (end == 0) { end = CFAbsoluteTimeGetCurrent() }
return Double(end - begin)
}
var duration:CFAbsoluteTime {
get {
if (end == 0) { return CFAbsoluteTimeGetCurrent() - begin }
else { return end - begin }
}
}
var description:String {
let time = duration
if (time > 100) {return " \(time/60) min"}
else if (time < 1e-6) {return " \(time*1e9) ns"}
else if (time < 1e-3) {return " \(time*1e6) µs"}
else if (time < 1) {return " \(time*1000) ms"}
else {return " \(time) s"}
}
}
Wrap it up in a completion block for easy use.
public class func secElapsed(completion: () -> Void) {
let startDate: NSDate = NSDate()
completion()
let endDate: NSDate = NSDate()
let timeInterval: Double = endDate.timeIntervalSinceDate(startDate)
println("seconds: \(timeInterval)")
}
The recommend way to check elapsed time/performance is using the measure
function that is available in XCTest
s.
It isn't reliable to write your own measure blocks, since the performance (and therefore execution/elapsed time) of a block of code is influenced by e.g. CPU caches.
The second time a function is invoked, would likely be quicker than the first time it is invoked, although it can vary a few %. Therefore 'benchmarking' by using your own closures (given all over the place here) by executing it once, can give different results than your code being executed in production by real users.
The measure
function invokes your block of code several times, mimicking the performance/elapsed time of your code like it is used in production (at least gives more accurate results).
This is the snippet I came up with and it seems to work for me on my Macbook with Swift 4.
Never tested on other systems, but I thought it's worth sharing anyway.
typealias MonotonicTS = UInt64
let monotonic_now: () -> MonotonicTS = mach_absolute_time
let time_numer: UInt64
let time_denom: UInt64
do {
var time_info = mach_timebase_info(numer: 0, denom: 0)
mach_timebase_info(&time_info)
time_numer = UInt64(time_info.numer)
time_denom = UInt64(time_info.denom)
}
// returns time interval in seconds
func monotonic_diff(from: MonotonicTS, to: MonotonicTS) -> TimeInterval {
let diff = (to - from)
let nanos = Double(diff * time_numer / time_denom)
return nanos / 1_000_000_000
}
func seconds_elapsed(since: MonotonicTS) -> TimeInterval {
return monotonic_diff(from: since, to:monotonic_now())
}
Here's an example of how to use it:
let t1 = monotonic_now()
// .. some code to run ..
let elapsed = seconds_elapsed(since: t1)
print("Time elapsed: \(elapsed*1000)ms")
Another way is to do it more explicitly:
let t1 = monotonic_now()
// .. some code to run ..
let t2 = monotonic_now()
let elapsed = monotonic_diff(from: t1, to: t2)
print("Time elapsed: \(elapsed*1000)ms")
This is how I wrote it.
func measure<T>(task: () -> T) -> Double {
let startTime = CFAbsoluteTimeGetCurrent()
task()
let endTime = CFAbsoluteTimeGetCurrent()
let result = endTime - startTime
return result
}
To measure a algorithm use it like that.
let time = measure {
var array = [2,4,5,2,5,7,3,123,213,12]
array.sorted()
}
print("Block is running \(time) seconds.")
Static Swift3 class for basic function timing. It will keep track of each timer by name. Call it like this at the point you want to start measuring:
Stopwatch.start(name: "PhotoCapture")
Call this to capture and print the time elapsed:
Stopwatch.timeElapsed(name: "PhotoCapture")
This is the output: *** PhotoCapture elapsed ms: 1402.415125 There is a "useNanos" parameter if you want to use nanos. Please feel free to change as needed.
class Stopwatch: NSObject {
private static var watches = [String:TimeInterval]()
private static func intervalFromMachTime(time: TimeInterval, useNanos: Bool) -> TimeInterval {
var info = mach_timebase_info()
guard mach_timebase_info(&info) == KERN_SUCCESS else { return -1 }
let currentTime = mach_absolute_time()
let nanos = currentTime * UInt64(info.numer) / UInt64(info.denom)
if useNanos {
return (TimeInterval(nanos) - time)
}
else {
return (TimeInterval(nanos) - time) / TimeInterval(NSEC_PER_MSEC)
}
}
static func start(name: String) {
var info = mach_timebase_info()
guard mach_timebase_info(&info) == KERN_SUCCESS else { return }
let currentTime = mach_absolute_time()
let nanos = currentTime * UInt64(info.numer) / UInt64(info.denom)
watches[name] = TimeInterval(nanos)
}
static func timeElapsed(name: String) {
return timeElapsed(name: name, useNanos: false)
}
static func timeElapsed(name: String, useNanos: Bool) {
if let start = watches[name] {
let unit = useNanos ? "nanos" : "ms"
print("*** \(name) elapsed \(unit): \(intervalFromMachTime(time: start, useNanos: useNanos))")
}
}
}
mach_timebase_info
is already implemented better than you did in the source code of ProcessInfo.processInfo.systemUptime
. So simply do watches[name] = ProcessInfo.processInfo.systemUptime
. And * TimeInterval(NSEC_PER_MSEC)
if you want nanos.
Based on Franklin Yu answer and Cœur comments
Details
Xcode 10.1 (10B61)
Swift 4.2
Solution 1
Solution 2
import Foundation
class Measurer<T: Numeric> {
private let startClosure: ()->(T)
private let endClosure: (_ beginningTime: T)->(T)
init (startClosure: @escaping ()->(T), endClosure: @escaping (_ beginningTime: T)->(T)) {
self.startClosure = startClosure
self.endClosure = endClosure
}
init (getCurrentTimeClosure: @escaping ()->(T)) {
startClosure = getCurrentTimeClosure
endClosure = { beginningTime in
return getCurrentTimeClosure() - beginningTime
}
}
func measure(closure: ()->()) -> T {
let value = startClosure()
closure()
return endClosure(value)
}
}
Usage of solution 2
// Sample with ProcessInfo class
m = Measurer { ProcessInfo.processInfo.systemUptime }
time = m.measure {
_ = (1...1000).map{_ in Int(arc4random()%100)}
}
print("ProcessInfo: \(time)")
// Sample with Posix clock API
m = Measurer(startClosure: {Double(clock())}) { (Double(clock()) - $0 ) / Double(CLOCKS_PER_SEC) }
time = m.measure {
_ = (1...1000).map{_ in Int(arc4random()%100)}
}
print("POSIX: \(time)")
Date
to measure anything is extremely discouraged (but systemUptime
or clock()
are OK). As for tests, we have measure(_:)
already.
Solution similar to C#'s Stopwatch class.
Usage:
let s = Stopwatch()
s.start()
for _ in 1..<5 {
Thread.sleep(forTimeInterval: 5)
print("Sleeped \(s.elapsedS) S")
}
s.stop()
print("Sleeped \(s.elapsedS) s")
result:
Sleeped 5.001374235 s
Sleeped 10.002803108 s
Sleeped 15.003942841 s
Sleeped 20.004250347000003 s
Sleeped 20.004299962 s
Code:
public class Stopwatch {
public var elapsedS : Double {
return elapsedMs/1000
}
public var elapsedMs: Double {
guard let startTime = startTime else { return 0 }
if isGoing {
let end = DispatchTime.now()
let nanoTime = end.uptimeNanoseconds - startTime.uptimeNanoseconds
return Double(nanoTime) / 1000000
}
if let endTime = endTime {
let nanoTime = endTime.uptimeNanoseconds - startTime.uptimeNanoseconds
return Double(nanoTime) / 1000000
}
return 0
}
private var startTime : DispatchTime? = nil
private var endTime : DispatchTime? = nil
private var isGoing = false
public init () { }
public func start() -> Stopwatch {
isGoing = true
startTime = DispatchTime.now()
return self
}
public func stop() {
endTime = DispatchTime.now()
isGoing = false
}
public func printS(_ str : String? = nil){
if let str = str {
print("\(str): \(elapsedS)")
}
else {
print("\(elapsedS)")
}
}
public func printMs(_ str : String? = nil){
if let str = str {
print("\(str): \(elapsedMs)")
}
else {
print("\(elapsedMs)")
}
}
}
Success story sharing
Date
alongside, and sure enough that reports 41.87 seconds. I don't know whatuptimeNanoseconds
is doing, but it's not reporting the correct duration.