DEV Community

Cover image for Swift Memory Management and Performance
Michal Rogowski
Michal Rogowski

Posted on • Updated on • Originally published at rogowski.page

Swift Memory Management and Performance

As part of recruitment processes one of the most common questions is: “What is the difference between struct and class”, we usually can hear that class is copied by reference and struct by value. But what does it even mean? Let’s start with general introduction to Memory Management

What is memory?

We can try to imagine memory as array of cells, each cell contains an information. Typically size of this information is one byte (e.g. bin: 01100101, hex: 0x65), each cell is associated with address which allows us to write or read value from this particular address. Assuming our app is using 20 Megabytes, we have to use 20 000 000 cells, and each cell contains 8 bits.

Memory visualisation

On the picture above, on the left there is the address and on the right is the value associated with this address. But what happens when value we want to store is bigger than size of one cell?

Little-endian vs Big-endian

The purpose of both of those terms is to decide in which order bytes are stored in memory. Big-endian stores most significant value at the lowest storage address, little-endian is doing the opposite. Let’s take a look at this examples:

decimalValue: Int16 = 20306
Enter fullscreen mode Exit fullscreen mode

The binary value of this Integer is 100111101010010
So let’s split those bits by size of our cells which is 1 byte.

firstPart = 01001111
secondPart = 01010010
Enter fullscreen mode Exit fullscreen mode

We need two cells, to allocate this value in memory.

Big-endian

Alt Text

Little-endian

Alt Text

As you can see Big-endian seems more natural for people who read left to right but the problem is with increasing a number, when number growths in binary system, numbers are added to the left side.

bin: 1111 == dec: 15
bin: 0001 1111 == dec: 31
Enter fullscreen mode Exit fullscreen mode

So if our number growths in memory, in Big-endian notation we would have to rewrite all addresses values, in Little-endian we would just append it at the end.

1) decimalValue = 255 //bin: 1111 1111
2) decimalValue += 256 // bin: 0000 0001 1111 1111 dec: 511
Enter fullscreen mode Exit fullscreen mode

1)
Alt Text

Big-endian

2)
Alt Text

As you can see the value from memory address 0x0000000112856f55 was rewritten to 0x0000000112856f56

Little-endian

2)
Alt Text

New value was assigned at the end (new memory address)

Why do I care? Swift vs bits

You will seldom have to know the difference between Big-endian and Little-endian, or even understand how memory looks like. But if you’d like to grow as a developer and understand how things work under the hood - continue reading.

Xcode and Memory Viewer

import Foundation
var helloWorld = Hello, World!
withUnsafePointer(to: &helloWorld) { print($0) }
print(helloWorld)
Enter fullscreen mode Exit fullscreen mode

Let’s put a breakpoint on print, when the breakpoints fires, we can activate memory viewer in Xcode! (Yes it exists ❤️) , when execution is paused on breakpoint just go to Debug -> Debug Workflow -> View Memory.

In our console we can see printed hexadecimal memory address, thanks to it we know address of memory we want to show in our viewer, just paste address and press enter.

Memory Viewer
Now we can check how is it stored on my Mac!
In the left column we can see the addresses of the first cells in each row written in hexadecimal, we can change it to decimal by tapping on left column.

In the center there are cells with their values. Each cell value is one Byte, so value stored under address 0x100002060 is 48 hexadecimal, which means character H, because this string is encoded in UTF-8 whose representation is the same as ASCII for Latin alphabet, here is simple Ascii table - representing values associated with characters, if you’d like to know more about ascii, encoding and decoding please let me know!

In the right column we can see preview of memory Hello, World!í this is our value displayed in preview.

There are few questions that comes to mind:

  • Is it Big-endian or Little-endian?
    It is Little-endian, just replace the string value with 511 from example above and compare results, value is also hexadecimal, we would expect that value is FF 01 It’s hard to debug it on characters, because ascii strings are arrays of single byte values, that’s why endianness does not affect the order.

  • What is this special character, and why is it prefixed with 2 bytes of 0x00?

  • My system is 64 bits so size of this string is not 13 bytes, even if it only contains 13 characters, it’s 2*8 bytes, last byte has a special value to indicate end of string.

We can easily check size of this string by running

print(MemoryLayout.size(ofValue: helloWorld)) //prints 16
Enter fullscreen mode Exit fullscreen mode

Copy by reference

Look at this example:

//
//  Created by Michał Rogowski on 27/09/2019.
//

import Foundation

class TestClassByReference {
    var value1 = test
    var value2 = test2
    var value3 = test3
    var value4 = test4
    var value5 = test5
    var value6 = test6

    var address: UnsafeMutableRawPointer {
        return Unmanaged.passUnretained(self).toOpaque()
    }
}

var testObject = TestClassByReference()
var copyTestObject = testObject

print(testObject:)
print(testObject.address)
print(copyTestObject:)
print(copyTestObject.address)
copyTestObject.value1 = change
print(testObject address after change:)
print(testObject.address)
print(copyTestObject address after change:)
print(copyTestObject.address)

print(testObject)
Enter fullscreen mode Exit fullscreen mode

We should add two break points, on copyTestObject.value1 = “change” and on last print.

On first breakpoint we can notice that two addresses are the same, so jump to memory viewer!
Memory View reference.1
On the right column, our properties values are visible (test, test2, test3 etc.), now we can continue program execution, address of variables did not change, so we can move to memory viewer again.
Memory View reference.2
Value test was changed and the memory reflects that by presenting a new string value in a place of previous one! Both instances point to same address - share a single copy of data. This explains the first category of instances types supported by Swift which is Reference Type.

Copy-on-Write

//
//  Created by Michał Rogowski on 27/09/2019.
//

import Foundation

extension Array where Element == String {
    var address: UnsafePointer<String>? {
        withUnsafeBufferPointer { $0.baseAddress }
    }
}


var testObject = [test, test2, test3, test4, test5, test6]
var copyTestObject = testObject

print(testObject:)
print(testObject.address)
print(copyTestObject:)
print(copyTestObject.address)
copyTestObject[0] = change
print(testObject address after change:)
print(testObject.address)
print(copyTestObject address after change:)
print(copyTestObject.address)

print(testObject)
Enter fullscreen mode Exit fullscreen mode

Now we can debug it step by step, compare the addresses of both arrays before changing value in our copy. Addresses are the same, but what happens after our change.
Copy on write before
Copy on write after
Address was changed and whole array was rewritten.
We have to notice that Copy-on-Write is only implemented for certain types, for example Arrays. If you’d like to use it for your structures-to optimise memory, you would have to create value wrapper, as described by Apple Here

Copy by value

Second category that is supported by Swift are Value Types. That means each instance keeps unique memory location for its data. Take a look

//
//  Created by Michał Rogowski on 27/09/2019.
//

import Foundation

struct MyStruct {
    let value = 5
}

var firstStruct = MyStruct()
var copyStruct = firstStruct

withUnsafePointer(to: &firstStruct) { print($0) } //print 0x00000001000040b0
withUnsafePointer(to: &copyStruct) { print($0) } //print 0x00000001000040b8
Enter fullscreen mode Exit fullscreen mode

Both properties have different addresses since the begging, in this case the memory addresses are allocated by 8 Bytes (Because of 64-bit system).
As you can read in Swift Documentation:

/// A signed integer value type.
///
/// On 32-bit platforms, `Int` is the same size as `Int32`, and
/// on 64-bit platforms, `Int` is the same size as `Int64`.
Enter fullscreen mode Exit fullscreen mode

Why should we waste 8 bytes for variable that needs only 1 byte? What would happen if we change default Int in our structure to Int8? Change declaration of Int in MyStruct

struct MyStruct {
    let value: Int8 = 5
}
Enter fullscreen mode Exit fullscreen mode

And those are allocated addresses for this two structs:

0x0000000100004088
0x0000000100004089
Enter fullscreen mode Exit fullscreen mode

Now the difference between addresses is only 1 Byte. We were able to free 7 bytes of memory per instance of MyStruct just by changing declaration to Int8, on the other hand you have to be careful to choose correct size because integer overflow may cause unintended behaviour.

Value Types are usually defines as a struct, enum, or tuple (Int in Swift is also a Struct).

In-Out Parameters

We can think about memory efficiency not only by allocating one Byte integers instead of eight Byte, but there is also In-Out Parameter
As Swift documentations states Here:
In-out parameters are passed as follows:
1. When the function is called, the value of the argument is copied.
2. In the body of the function, the copy is modified.
3. When the function returns, the copy’s value is assigned to the original argument.
This behaviour is known as copy-in copy-out or call by value result.

struct InoutStruct {
    var value = test

    mutating func printAddress() {
        withUnsafePointer(to: &self) { print(address = \($0)) }
    }
}

func addTestInout(to inoutStruct: inout InoutStruct) {
    inoutStruct.value +=  test
    inoutStruct.printAddress()
}


func addTest(to normalStruct: InoutStruct) -> InoutStruct {
    var inoutStructCopy = normalStruct
    inoutStructCopy.value +=  test
    inoutStructCopy.printAddress()
    return inoutStructCopy
}

var inoutStruct = InoutStruct()
var normalStruct = InoutStruct()

inoutStruct.printAddress() //0x0000000100003090
normalStruct.printAddress() //0x00000001000030a0

addTestInout(to: &inoutStruct) //0x0000000100003090
normalStruct = addTest(to: normalStruct) //0x00007ffeefbff270

inoutStruct.printAddress() //0x0000000100003090
normalStruct.printAddress() //0x00000001000030a0
Enter fullscreen mode Exit fullscreen mode

We’ve done the “same” operation twice to separate instances of InoutStruct, if you go through Memory View, you can check that, in the memory we have 3 instances of InoutStruct. One of those is variable inoutStructCopy. It’s clear now that, inout parameter is memory efficient if we want to change instance that is Value Type in function. There is no extra rewrite, which means you wan’t have to waste time and memory when changing really big Value Type instances.

Thanks!

Top comments (3)

Collapse
 
gdahboy profile image
gdahboy

great one thank you

Collapse
 
bananowazyrafa profile image
bananowaZyrafa

Very interesting and insightful content, thanks for writing it! However, your english level in this article is quite low, please work on it cause it's a bit hard to consume :)
Best regards

Collapse
 
michalrogowski profile image
Michal Rogowski

Thanks! I know you’re right, I hope next article will be more consumable :)