Benchmarker module

Benchmarker

A roblox lua module that allows you to benchmark functions, it benchmarks through two ways, the average time it took to call the function for a given amount of cycles. And the total amount times it can call the function for a given duration. You can also directly compare two functions.

Example

code

local ServerScriptService = game:GetService('ServerScriptService')

local Benchmarker = require(ServerScriptService:WaitForChild("Benchmarker"))

-- test setup for ipairs vs pairs

local tbl = table.create(1e6,"t") -- create a table with 1 million entries of "t"

local function testIpairs()
    for _ in ipairs(tbl) do

    end
end

local function testPairs()
    for _ in pairs(tbl) do

    end
end

Benchmarker:compare(testIpairs, testPairs)

output

image.png

Customization

cycles

You can set the amount times the function will be called for calculating the average time per call. By default it does a 1000 cycles.

duration

You can set the duration (in seconds). This is used for calculating the amount of cycles for a given duration. By default it has a duration of 1 second.

showProgress

This boolean controls if it should print the current progress on the benchmark. It will perodically print the current completion. It is off by default.

showFullInfo

This boolean controls if it should print the full info. It is on by default and turning it off will only show the minimal amount of prints.

ReadableNumbers

Here you can pass a ReadableNumbers object. Which converts numbers into a human readable format. For example 4.14001e-08 s → 41.4 ns. You can turn this feature off by passing a false. You can customize it by passing your own ReadableNumbers object. see ReadableNumbers on the specifics.

noYieldTime

The time in seconds in which it wont yield. By default it is 0.1 s. Which means it will only yield every 0.1 s. You can increase this number to have faster benchmarks but if you are getting “Script timeout: exhausted allowed execution time” errors, lower this number.

Code examples

local ServerScriptService = game:GetService('ServerScriptService')

local Benchmarker = require(ServerScriptService:WaitForChild("Benchmarker"))

Benchmarker:benchmark(math.floor, 10.4144) --  will output the results for doing 1000 cycles and amount of cycles it did in 1s (default settings)

Benchmarker:compare(math.floor, math.round, 1, 10.4144, 10.4144 ) -- will print the results of each benchmark and print how much faster/slower function1 is compared to function2

local benchmark = Benchmarker.new(1e4, 10, true) -- create a benchmark object with our own configuration, showProgress has been set to to true so that you can see how long it takes for it to complete

-- test setup for ipairs vs pairs

local tbl = table.create(1e6,"t") -- create a table with 1 million entries of "t"

local function testIpairs()
    for _ in ipairs(tbl) do

    end
end

local function testPairs()
    for _ in pairs(tbl) do

    end
end

benchmark:compare(testIpairs, testPairs) -- Ipairs has always been faster in my comparisons

benchmark.cycles = 10 -- you can also change the settings directly through the properties

API

Is in the form of returnType function(argumentName:type, argumentName:type defaultValue)

Constructors

Benchmarker Benchmarker.new(cycles: int 1e3, duration: number 1, showProgress: bool false, showFullInfo: bool true, ReadableNumbers: false|ReadableNumbers ReadableNumbers.new(), noYieldTime: number 0.1)

It returns a Benchmarker instance, all properties can be changed by indexing the argument name. See customization on what each argument does.

Methods

void Benchmarker:compare(func1: function, func2: function, func1AmountArgs: number, …)

This method does a comparison of the functions on the two modes using the default settings. func1AmountArgs splits the vararg on that index. For example: compare(math.floor, math.round, 1, 10, 15), math.floor gets 10 and math.round gets 15, Another example compare(select, math.round, 4, 2, “A” , “B”, “C”, 15) which 2, "A" , "B", "C" go to select and math.round gets 15.

The reason for the func1AmountArgs is that it needs to split the varags between the two given functions. It prints the results to the output.

void Benchmarker:benchmark(func: function, …)

This method prints the result of both modes (cycles for given amount of duration, average tim per cycle of given amount of cycles). The vararg will be passed as the arguments to the function.

tuple avgTimePerCycleInSec: number, totalTime: number Benchmarker:getAvg(func: function, …)

This method calculates the average time a cycle took for the set amount of cycles. This method is internally used by compare and benchmark. It returns a tuple with the first value the average time per cycle in seconds and the second value the total time it took for all the cycles in seconds.

tuple amount: number, amountPerSec: number, totalTime: number Benchmarker:getCycles(func: function, …)

This method calcualtes how many times the function can be called for a set duration (in seconds). his method is internally used by compare and benchmark. It returns a tuple with the first value the total amount times the function has been called for the set duration. The second value is the amount per second. The third value is the actual amount of duration it took, this should always the be same as the set duration.

Links

Feel free to ask any questions! Feedback is always welcome.

8 Likes

Basically function performance testing, right?

2 Likes

Yes, it tests how long it takes to call the functions

Sounds very useful, especially for micro-optimizing.
Does it use time() or tick()?

1 Like

It uses os.clock() which is made for benchmarking

1 Like

Oh, interesting.
I haven’t used it before, so I might have to check it out.

1 Like

Here is a nice overview on all the time functions


from

btw you shouldnt use tick() anymore because

tick() sounds perfect - it has a high resolution (usually around 1 microsecond), and a well-defined baseline - it counts since UNIX epoch! Or, well, it actually doesn’t. On Windows, it returns you a variant of the UNIX timestamp in local time zone. In addition, it can be off by 1 second from the actual, real UNIX timestamp, and might have other idiosyncrasies on non-Windows platforms. We’re going to deprecate this in the future.

from

Oh btw they also added a new datatype made for time, which is called Datetime

3 Likes