Code Optimisation

From Bohemia Interactive Community
Jump to navigation Jump to search
This page is about Code Optimisation. For conception optimisation, see Mission Optimisation.

This article will try to be a general guide about improving your code and its performance.

  • The first part (Rules) will focus on having a clean, readable and maintainable code.
  • The second part (Code optimisation) is about improving performance, sometimes trading it against code readability.
  • The third part (Equivalent commands performance) mentions commands that in appearance have identical effects but may differ in terms of performance according to the use you may have of them.
  • The fourth part (Conversion from earlier versions) is a hopefully helpful, short guide about useful new commands or syntaxes to replace the old ways.


Rules

In the domain of development, any rule is a rule of thumb. If a rule states for example that it is better that a line of code doesn't go over 80 characters, it doesn't mean that any line must not go over 80 characters; sometimes, the situation needs it. If you have a good structure, do not change your code to enforce a single arbitrary rule. If you break many of them, you may have to change something. Again, this is according to your judgement.

With that being said, here are the three basic rules to get yourself in the clear:

  1. Make it work
  2. Make it readable
  3. Optimise then

Make It Work

«
« Premature optimization is the root of all evil. » – Donald Knuth

Your first goal when coding is to make your code do what you want it does. A good way to reach this objective is to read and getting inspired by other people's code. If you understand it by reading it once, it is probably a good source of inspiration.

  • When starting from scratch if you know what you want but miss the specific steps to get to your point, it is a good practice to write down in your native language what you want to do. E.g Get all the units near the city, and for each west soldier in them, add 30% damage.
  • Use -showScriptErrors startup parameter and make sure your code doesn't throw errors. Not only will your code run slower but it may also not work at all. Be sure to read the error, isolate the issue and sort it out thanks to this Wiki.
  • Read your Arma RPT (report) to read more details about the error that happened in your code.

Make It Readable

Whether you are cleaning your code or a different person's, you must understand the code without twisting your brain:

  • While SQF is (non-noticeably) impacted by variable name length, this should not take precedence on the fact that code must be readable by a human being. Variables like _u instead of _uniform should not be present.
  • One-lining (putting everything in one statement) memory improvement is most of the time not worth the headache it gives when trying to read it. Don't overuse it.
  • Indentation is important for the human mind, and space is too. Space is free, use it.
  • Same goes for line return; it helps to see a code block wrapping multiple common instructions instead of having to guess where it starts and stops.
  • Do you see the same code multiple times, only with different parameters? Now is the time to write a function!
  • If you have a lot of if..else, you may want to look at a switch condition, or again break your code in smaller functions.
  • Is your function code far too long? Break it in understandable-sized bites for your own sanity.
  • Finally, camel-casing (namingLikeThis) your variables will naturally make the code more readable, especially for long names.
See Code Best Practices for more information.

See the following code:

_w=[]; {_w pushbackunique primaryweapon _x} foreach((allunits+alldeadmen) select{_x call BIS_fnc_objectside==east});

The same example is far more readable with proper spacing, good variable names and intermediate results:

_weaponNames = [];
_allUnitsAliveAndDead = allUnits + allDeadMen;
_allEastAliveAndDead = _allUnitsAliveAndDead select { _x call BIS_fnc_objectSide == east };
{ _weaponNames pushBackUnique primaryWeapon _x } forEach _allEastAliveAndDead;

Constants

Using a hard-coded constant more than once? Use preprocessor directives rather than storing it in memory or cluttering your code with numbers. Such as:

a = _x + 1.053; b = _y + 1.053;

And

_buffer = 1.053; a = _x + _buffer; b = _y + _buffer;

Becomes

#define BUFFER 1.053 // note: no semicolon _a = _x + BUFFER; _b = _y + BUFFER;

Using the #define macro only works within the current SQF file. Such definition will not propagate anywhere else.

Global "constants" can be defined via a Description.ext declaration, though, and accessed using getMissionConfigValue command:

Declaration in Description.ext:

var1 = 123;
var2 = "123";
var3[] = { 1, 2, 3 };
rand = __EVAL(random 999);

Usage in code:

hint str getMissionConfigValue "var1"; // 123 // 0.0007 ms hint str getMissionConfigValue "var2"; // "123" // 0.0008 ms hint str getMissionConfigValue "var3"; // [1,2,3] // 0.0017 ms hint str getMissionConfigValue "rand"; // constant random, for example 935.038 // 0.0007 ms

The getMissionConfigValue command searching Description.ext from top to bottom, it is better for a matter of performance to put all your definitions at the top of the file.

Optimise Then

Once you know what is what, you can understand your code better.

  • Use private variables instead of global variables (preceded by an underscore) as much as possible.
  • You were iterating multiple times on the same array?
    • You should be able to spot your issue now.
  • Are you using execVM on the same file, many times?
  • Is your variable name far too long?


Code Optimisation

Please note: Tests and benchmarks were done with the latest Arma 3 version at the time Arma 3 logo black.png1.82 with Tank DLC. Game engine performance may have changed since.
Benchmark result in milliseconds (ms) is an average for 10000 iterations.
means you must change your ways today, or with us you will ride…
means you may want to look at it if you are targeting pure performance
means the gain is little to insignificant. Going through your code for this replacement is not worth it. You may only consider it for future code.

Scheduled and Unscheduled Environment

There are two code environment types, scheduled and unscheduled.

  • A scheduled script has an execution time limit of 3 ms before being suspended to the benefit of another script until its turn comes back. It is a bit slower than unscheduled, but suspending (sleep, waitUntil) is allowed.
  • An unscheduled script is not watched and will run without limitations. It is recommended for time-critical scripts, but suspending (sleep, waitUntil) is not allowed!
See Scheduler for more information.

Variable Assignment

private _myVar = [33, 66] select false; // 0.0013 ms private _myVar = if (false) then { 33; } else { 66; }; // 0.0020 ms private "_myVar"; if (false) then { _myVar = 33; } else { _myVar = 66; }; // 0.0025 ms

Lazy Evaluation

In SQF the following code will evaluate every single condition, even if one fails:

if (a && b && c) then {};

Even if a returns false (and thus the entire Boolean expression can no longer become true), b and c will still be executed and evaluated regardless.

To avoid this behaviour, one can either imbricate if statements or use lazy evaluation. The latter is done like so:

if (a && { b && { c } }) then {};

In the example above, condition evaluation stops once any condition evaluates to false.

Influence on Semantics

Depending on the arrangement of the curly brackets, lazy evaluation can change the precedence (and therefore the semantics) of a condition. Consider this example:

Expression Condition Value Evaluated Conditions Result
a b c
a && b || c
false any Boolean true a, b, c true
a && {b} || {c}
a, c true
a && {b || {c}}
a false

Performance

Using lazy evaluation is not always the best way as it can both speed up and slow down the code, depending on the current condition being evaluated:

["true || { false || {false}}", nil, 100000] call BIS_fnc_codePerformance; // 0.00080 ms ["true || {false} || {false} ", nil, 100000] call BIS_fnc_codePerformance; // 0.00105 ms ["false || false || false ", nil, 100000] call BIS_fnc_codePerformance; // 0.00123 ms ["true || false || false ", nil, 100000] call BIS_fnc_codePerformance; // 0.00128 ms ["false || {false} || {false} ", nil, 100000] call BIS_fnc_codePerformance; // 0.00200 ms

Concatenating Strings

myString = myString + otherString works fine for small strings, however the bigger the string gets the slower the operation becomes:

myString = ""; for "_i" from 1 to 10000 do { myString = myString + "123" }; // 290 ms

The solution is to use a string array that you will concatenate later:

strings = []; for "_i" from 1 to 10000 do { strings pushBack "123" }; strings = strings joinString ""; // 30 ms

Array Manipulation

Add Elements

New commands append and pushBack hold the best score.

_array = [0,1,2,3]; _array append [4,5,6]; // 0.0020 ms _array = [0,1,2,3]; _array = _array + [4,5,6]; // 0.0023 ms _array = [0,1,2,3]; { _array set [count _array, _x]; } forEach [4,5,6]; // 0.0080 ms

_array = [0,1,2,3]; _array pushBack 4; // 0.0016 ms _array = [0,1,2,3]; _array = _array + [4]; // 0.0021 ms _array = [0,1,2,3]; _array set [count _array, _x]; // 0.0022 ms

Iterate Elements

for is twice as fast as forEach and is recommended if _x is not required.

private _array = allUnits; // 64 units 256 units // if the amount of loops is known for "_i" from 0 to 63 do {}; // 0.0120 ms 0.0460 ms for "_i" from 0 to count _array -1 do {}; // 0.0170 ms 0.0480 ms for "_i" from 0 to count _array -1 do { private _x = _array select _i }; // 0.0500 ms 0.1896 ms < for "_i" from 0 to count _array -1 do { private _x = array select _i; _x setDamage 0 }; // 0.107 ms 0.39 ms {} forEach _array; // 0.0250 ms 0.098 ms { _x setDamage 0 } forEach _array; // 0.0770 ms 0.312 ms _array apply {}; // 0.0250 ms 0.098 ms _array apply { _x setDamage 0 }; // 0.0770 ms 0.312 ms private _i = 0; while { _i < 64 } do { _i = _i + 1; }; // 0.048 ms 0.200 ms // counts array every loop private _i = 0; while { _i < count _array } do { _i = _i + 1; }; // 0.062 ms 0.247 ms private _i = 0; while { _i < 64 } do { private _x = _array select _i; _i = _i + 1; }; // 0.086 ms 0.42 ms

Remove Elements

_array = [0,1,2,3]; _array deleteAt 0; // 0.0015 ms _array = [0,1,2,3]; _array set [0, objNull]; _array = _array - [objNull]; // 0.0038 ms

_array = [0,1,2,3]; _array deleteRange [1, 2]; // 0.0018 ms _array = [0,1,2,3]; { _array set [_x, objNull] } forEach [1,2]; _array = _array - [objNull]; // 0.0078 ms

Multiplayer Recommendations

  • Do not saturate the network with information: publicVariable or public setVariable shouldn't be used at high frequency, else everyone's performance experience is at risk!
  • The server is supposed to have a good CPU and a lot of memory, use it: store functions, run them from it, send only the result to the clients
  • publicVariable and setVariable variable name length impacts network, be sure to send well-named, understandable variables
    (and not playerNameBecauseThePlayerIsImportantAndWeNeedToKnowWhoTheyAreAllTheTimeEspeciallyInsideThisImpressiveFunction)
  • Use, use and use remoteExec & remoteExecCall. Ditch BIS_fnc_MP for good!
See Multiplayer Scripting for more information.


Equivalent Commands Performance

call

call without arguments is faster than call with arguments:

call {}; // 0.0007 ms 123 call {}; // 0.0013 ms

Since the variables defined in the parent scope will be available in the called child scope, it could be possible to speed up the code by avoiding passing arguments all together, for example writing:

player addEventHandler ["HandleDamage", { call my_fnc_damage }];

instead of:

player addEventHandler ["HandleDamage", { _this call my_fnc_damage }];

execVM and call

Using execVM multiple times make the game read the file and recompile it every time.
If you use the script more than once, store its code in a variable or better, make it a Function!

// myFile.sqf is an EMPTY file private _myFunction = compile preprocessFileLineNumbers "myFile.sqf"; // compile time is done only once call _myFunction; // 0.0009 ms execVM "myFile.sqf"; // 0.275 ms // myFile.sqf is BIS_fnc_showRespawnMenu private _myFunction = compile preprocessFileLineNumbers "myFile.sqf"; // compile time is done only once ["close"] call _myFunction; // 0.0056 ms ["close"] execVM "myFile.sqf"; // 0.506 ms

loadFile, preprocessFile and preprocessFileLineNumbers

// myFile.sqf is an empty file loadFile "myFile.sqf"; // 0.219 ms preprocessFile "myFile.sqf"; // 0.353 ms preprocessFileLineNumbers "myFile.sqf"; // 0.355 ms

// myFile.sqf is BIS_fnc_showRespawnMenu loadFile "myFile.sqf"; // 0.3516 ms preprocessFile "myFile.sqf"; // 2.75 ms preprocessFileLineNumbers "myFile.sqf"; // 2.73 ms

// myFile.sqf is a missing file loadFile "myFile.sqf"; // 0.692 ms preprocessFile "myFile.sqf"; // 0.6225 ms preprocessFileLineNumbers "myFile.sqf"; // 0.6225 ms

The comparison of loadFile with preprocessFile* is not exactly fair as loadFile doesn't preprocess the file's content.
On the other hand, the loaded file cannot contain any //  or /* */ comments nor any preprocessor instructions (including debug information like line numbers or file information).

if

if (condition) then { /* thenCode */ }; // 0.0011 ms if (condition) exitWith { /* exitCode */ }; // 0.0014 ms if (condition) then { /* thenCode */ } else { /* elseCode */ }; // 0.0015 ms if (condition) then [{ /* thenCode */ }, { /* elseCode */ }]; // 0.0016 ms

if and select

Use [array] select boolean instead of the lazy-evaluated if.

_result = ["false result", "true result"] select true; // 0.0011 ms _result = if (true) then { "true result"; } else { "false result"; }; // 0.0017 ms

if and switch

_result = call { if (false) exitWith {}; if (false) exitWith {}; if (true) exitWith {}; if (false) exitWith {}; if (false) exitWith {}; }; // 0.0032 ms

_result = switch (true) do { case (false): {}; case (false): {}; case (true) : {}; case (false): {}; case (false): {}; }; // 0.0047 ms

in vs find

// String search "bar" in "foobar" // 0.0008 ms "foobar" find "bar" > -1 // 0.0012 ms

// array search - case-sensitive "bar" in ["foo", "Bar", "bar", "BAR"]; // 0.0012 ms ["foo", "Bar", "bar", "BAR"] find "bar" > -1; // 0.0016 ms

for

The for..from..to..do is twice as fast as its alternative syntax, for..do.

for "_i" from 0 to 10 do { /* forCode */ }; // 0.015 ms for [{ _i = 0 }, { _i < 100 }, { _i = _i + 1 }] do { /* forCode */ }; // 0.030 ms

forEach vs count vs findIf

Both forEach and count commands will step through all the array elements and both commands will contain reference to current element with the _x variable. However, count loop is a little faster than forEach loop, but it does not benefit from the _forEachIndex variable.
Also, there is a limitation as the code inside count expects Boolean or Nothing while the command itself returns Number. This limitation is very important if you try to replace your forEach by count. If you have to add an extra true/false/nil at the end to make count work, it will be slower than the forEach equivalent.

{ diag_log _x } count [1,2,3,4,5]; // 0.082 ms { diag_log _x } forEach [1,2,3,4,5]; // 0.083 ms

// with an empty array _someoneIsNear = (allUnits findIf { _x distance [0,0,0] < 1000 }) != -1; // 0.0046 ms _someoneIsNear = { _x distance [0,0,0] < 1000 } count allUnits > 0; // 0.0047 ms < _someoneIsNear = { if (_x distance [0,0,0] < 1000) exitWith { true }; false } forEach allUnits; // 0.0060 ms

// with a 30 items array _someoneIsNear = (allUnits findIf { _x distance [0,0,0] < 1000 }) != -1; // 0.0275 ms _someoneIsNear = { _x distance [0,0,0] < 1000 } count allUnits > 0; // 0.0645 ms < _someoneIsNear = { if (_x distance [0,0,0] < 1000) exitWith { true }; false } forEach allUnits; // 0.0390 ms

findIf

findIf stops array iteration as soon as the condition is met.

[0,1,2,3,4,5,6,7,8,9] findIf { _x == 2 }; // 0.0050 ms { if (_x == 2) exitWith { _forEachIndex; }; } forEach [0,1,2,3,4,5,6,7,8,9]; // 0.0078 ms _quantity = { _x == 2 } count [0,1,2,3,4,5,6,7,8,9]; // 0.0114 ms

format vs str

str 33; // 0.0016 ms format ["%1", 33]; // 0.0022 ms

+ vs format vs joinString

non-String data:

[33, 45, 78] joinString ""; // 0.0052 ms - no length limit format ["%1%2%3", 33, 45, 78]; // 0.0054 ms - limited to ~8Kb str 33 + str 45 + str 78; // 0.0059 ms - no length limit

String data:

["str1", "str2", "str3"] joinString ""; // 0.0015 ms - no length limit format ["%1%2%3", "str1", "str2", "str3"]; // 0.0015 ms - limited to ~8Kb "str1" + "str2" + "str3"; // 0.0012 ms - no length limit

private

Direct declaration (private _var = value) is faster than declaring then assigning the variable.

private _a = 1; private _b = 2; private _c = 3; private _d = 4; // 0.0023 ms

private ["_a", "_b", "_c", "_d"]; _a = 1; _b = 2; _c = 3; _d = 4; // 0.0040 ms

However, if you have to reuse the same variable in a loop, external declaration is faster.
The reason behind this is that a declaration in the loop will create, assign and delete the variable in each loop.
An external declaration creates the variable only once and the loop only assigns the value.

private ["_a", "_b", "_c", "_d"]; for "_i" from 1 to 10 do { _a = 1; _b = 2; _c = 3; _d = 4; }; // 0.0195 ms

for "_i" from 1 to 10 do { private _a = 1; private _b = 2; private _c = 3; private _d = 4; }; // 0.0235 ms

isNil

isNil "varName"; // 0.0007 ms isNil { varName }; // 0.0012 ms

isEqualType and typeName

isEqualType is much faster than typeName

"string" isEqualType 33; // 0.0006 ms typeName "string" == typeName 33; // 0.0018 ms

isEqualTo and count

// with a items array allUnits isEqualTo []; // 0.0040 ms count allUnits == 0; // 0.0043 ms

select and param

[1,2,3] select 0; // 0.0008 ms [1,2,3] param [0]; // 0.0011 ms

createSimpleObject vs createVehicle

// createSimpleObject is over 43× faster than createVehicle! deleteVehicle createSimpleObject ["a3\structures_f_mark\vr\shapes\vr_shape_01_cube_1m_f.p3d", [0,0,0]]; // ~0.08 ms deleteVehicle createSimpleObject ["Land_VR_Shape_01_cube_1m_F", [0,0,0]]; // ~3.2 ms deleteVehicle createVehicle ["Land_VR_Shape_01_cube_1m_F", [0,0,0], [], 0, "CAN_COLLIDE"]; // ~2.7 ms deleteVehicle createVehicle ["Land_VR_Shape_01_cube_1m_F", [0,0,0], [], 0, "NONE"]; // ~78 ms

objectParent and vehicle

isNull objectParent player; // 0.0013 ms vehicle player == player; // 0.0022 ms

nearEntities vs nearestObjects

nearEntities and nearestObjects

nearEntities is much faster than nearestObjects given on range and amount of objects within the given range. If range is over 100 meters it is highly recommended to use nearEntities over nearestObjects.

// tested with a NATO rifle squad amongst solar power plant panels on Altis at coordinates [20762,15837] getPosATL player nearEntities [["Man"], 50]; // 0.0075 ms nearestObjects [getPosATL player, ["Man"], 50]; // 0.0145 ms

nearEntities only searches for alive objects and on-foot soldiers.
In-vehicle units, killed units, destroyed vehicles, static objects and buildings will be ignored.

Global Variables vs Local Variables

If you need to use global variable repeatedly in a loop, copy its value to local variable and use local variable instead:

SomeGlobalVariable = [123]; for "_i" from 1 to 100 do { SomeGlobalVariable select 0; }; // 0.13 ms

is noticeably slower than

SomeGlobalVariable = [123]; private _var = SomeGlobalVariable; for "_i" from 1 to 100 do { _var select 0; }; // 0.08 ms

Config path delimiter

>> is slightly faster than / when used in config path with configFile or missionConfigFile.

configFile >> "CfgVehicles"; // 0.0019 ms configFile / "CfgVehicles"; // 0.0023 ms

A config path can be stored in a variable for later use, saving CPU time: _cfgVehicles = configFile >> "CfgVehicles".

getPos* and setPos*

getPosWorld // 0.0015 ms getPosASL // 0.0016 ms getPosATL // 0.0016 ms getPosASLW // 0.0023 ms getPos // 0.0030-0.0300 ms; performance depends on where this command is used - see its documentation position // same as getPos getPosVisual // same as getPos visiblePosition // same as getPos

setPosWorld // 0.0060 ms setPosASL // 0.0060 ms setPosATL // 0.0060 ms setPos // 0.0063 ms setPosASLW // 0.0068 ms setVehiclePosition // 0.0077 ms with "CAN_COLLIDE" // 0.0390 ms with "NONE"

toLower/toUpper vs toLowerANSI/toUpperANSI

// _myString is a 100 chars "aAaAaA(…)" string toLowerANSI _myString; // 0.0006 ms toLower _myString; // 0.0016 ms toUpperANSI _myString; // 0.0006 ms toUpper _myString; // 0.0016 ms


Equivalent Data Structures Performance

Key-Value Data Structures

private _hashMap = createHashMapFromArray [["id", 123], ["name", "player name"], ["unit", player]]; // since Arma 3 v2.02 private _goodFormat = [["id", "name", "unit"], [123, "player name", player]]; private _slowFormat = [["id", 123], ["name", "player name"], ["unit", player]];

private _name = _hashMap get "name"; // 0.0018ms // this takes 0.0038ms: private _index = _goodFormat select 0 find "name"; // loop in engine private _name = _goodFormat select 1 select _index; // this takes 0.0116ms: private _index = _slowFormat findIf { _x select 0 == "name" }; // loop in script private _name = _slowFormat select _index select 1;

Database Functions use a slow format.


Conversion From Earlier Versions

Each iteration of Bohemia games (Operation Flashpoint, Armed Assault, Arma 2, Take On Helicopters, Arma 3) brought their own new commands, especially Arma 2 and Arma 3.
For that, if you are converting scripts from older versions of the engine, the following aspects should be reviewed.

Loops

Array Operations

result = (arrayOfNumbers select { _x % 2 == 0 }); // 1.55 ms
result = []; { if (_x % 2 == 0) then { result pushBack _x; }; } forEach arrayOfNumbers; // 2.57 ms

Vector Operations

private _vector = [102, 687, 1543]; private _factor = 53; _vector vectorMultiply _factor; // 0.0028 ms [_vector, _factor] call BIS_fnc_vectorMultiply; // 0.0145 ms _vector vectorMultiply (1 / _factor); // 0.003 ms - but beware of 0 divisor [_vector, _factor] call BIS_fnc_vectorDivide; // 0.017 ms

String Operations

String manipulation has been simplified with the following commands:

Number Operations

Type Comparison

Multiplayer

Parameters