SlideShare a Scribd company logo
1 of 18
Download to read offline
Performance tips and tricks for Qt/QML
applications
General

Do not use JavaScript, but rather implement the UI with QML and the
engine with C++
JavaScript should be avoided as much as possible. Do not use it at all.

       JavaScript code should be kept to a minimum. C++ should be used to implement
        application's the business logic.
       JavaScript is especially bad for performance when it is run during animation, e.g. in lists
        the JavaScript monster gets evaluated on every frame of the scrolling animation as many
        times as there are instantiated delegates.
       If you must use JavaScript anyway, simplify your JavaScript expressions as much as
        possible and write them inline if they fit on one line. QML is compiled to an optimized
        bytecode-like stream and the JavaScript expressions pass through an optimized evaluator
        for simple expressions. More
        info: http://doc.qt.nokia.com/latest/qdeclarativejavascript.html#inline-javascript

       It is also recommended to set values for properties using QML property binding
        than JavaScript value assignment
       Wrong:

       itemTitle: {

                 if(targetFeedItem == null){

                      return "";

                 }

                 return targetFeedItem.title;

           }


       Right:

       itemTitle: targetFeedItem == undefined ? "" : targetFeedItem.title;
Use threads or multiprocessing to allow responsive UI
If you don't use threads for long-lasting operations, the UI is blocked. Use threads for the long lasting
operations such as data retrieval to keep UI responsive.

         In QML the WorkerScript element can be used to run JavaScript operations in a new thread
          (you shouldn't use JavaScript in a serious application)
             o   DO NOT use multiple WorkerScripts or you are in trouble.
             o   If you initialize a large list synchronously in a JavaScript block (this is something that
                 a serious application would never do...) it is very likely that UI jamming will occure
                 with high memory consumption. The memory consumption is likely due to the fact
                 that garbage collector does not get a chance to collect any items while a
                 synchronous loop is processing.
             o   More info on WorkerScript: http://doc.qt.nokia.com/latest/qml-workerscript.html
         C++ threads should be used for all serious applications
             o   This can be done by creating a QDeclarativeExtensionPlugin and exposing the
                 needed QObjects to QML
             o   In the QObject one exposes the properties, invokable methods, signals and slots
                 needed in the UI
             o   For any long (>2ms) operations one should use QRunnable with QThreadPool or
                 QFuture see. More info: http://doc.trolltech.com/latest/threads.html
             o   The long lasting operations (> N ms) must be implemented as asyncronous server
                 calls or by using threads so that the UI can update itself while the application is doing
                 long lasting backend operations


Use anchors rather than bindings to position items relative to each other
Wrong way to bind rect2 to rect1:

Rectangle {

         id: rect1

         x: 20

         width: 200; height: 200

 }

 Rectangle {

         id: rect2
x: rect1.x

      y: rect1.y + rect1.height

      width: rect1.width - 20

      height: 200

 }

Right way to do this, anchor rect2 to rect1:

Rectangle {

      id: rect1

      x: 20

      width: 200; height: 200

 }

 Rectangle {

      id: rect2

      height: 200

      anchors.left: rect1.left

      anchors.top: rect1.bottom

      anchors.right: rect1.right

      anchors.rightMargin: 20

 }

More info: http://doc.qt.nokia.com/latest/qdeclarativeperformance.html#anchors-vs-binding
App start

Use Booster
On Symbian and MeeGo it is possible to speed up application start by picking up pre-instantiated
QApplication and QDeclarativeView from cache. Please contact Nokia for details how to do this.


Use loaders to dynamically load/unload QML
QML application starts quite slow if the whole application is implemented in one huge QML file.
Partition our application wisely into logical entities, load minimum QML at start and load more as you
need it using Loaders .

        The Loader item can be used to dynamically load and unload visual QML components
         defined in a QML file or items/components defined within a QML file
        This dynamical behavior allows the developer to control the memory usage and startup
         speed of an application. More info: http://doc.qt.nokia.com/latest/qml-loader.html#details
        Partition your application into several QML files so that each file contains a logical UI entity.
         This way loading and unloading are easier to control. DO NOT have one huge QML file per
         application.
        Load absolutely minimum amount of QML at application start to make your application start
         as quickly as possible. You can connect to network and show spinner etc. after the
         application UI is visible. If your first view is very complex and requires lots of QML to be
         loaded, show a splash screen or something.
        You should load pieces of UI only by demand e.g. when the user navigates to another view,
         but on the other hand it may require more time to navigate between views
        Creating a dialog can take more than 100ms of time. Make sure you use Loader to load
         the dialog.

       import QtQuick 1.0




       Item {

           width: 200; height: 200




           Loader { id: pageLoader }
MouseArea {

    anchors.fill: parent

    onClicked: pageLoader.source = "Page1.qml"

    }

}
Images

Bitmap formats vs. vector graphics formats
Images can be supplied in any of the standard image formats supported by Qt, including bitmap
formats such as PNG and JPEG, and vector graphics formats such as SVG.

However, rendering an SVG is slow compared to a bitmap image.


Load large images asynchronously
If you load images synchronously the UI is blocked. In many cases images are not needed visible
immediately, but they can be lazy-loaded.

      You should load images asynchronously in a separate thread if an image doesn't need to be
       displayed immediately. This can be done by settings QML Image asynchronous to true.
      This way the user interface stays responsive user interface.
      Note that this property is only valid for images read from the local filesystem. Images that are
       loaded via a network resource (e.g. HTTP) are always loaded asynchonously.
      More info: http://doc.qt.nokia.com/latest/qml-image.html#asynchronous-prop


Avoid resizing/scaling
Resizing/scaling is a very heavy operation in QML. Use images that are in their native size rather
than resize/scale large images to correct size.


Use sourceSize with large images
Images are often the greatest user of memory in QML user interfaces.

      sourceSize should be used with large images because property sets the actual number of
       pixels stored for the loaded image
      If you have a HUGE image 3264 x 2448, but you set sourceSize 204x153, then it scaled
       down and will be stored as 204x153 in memory.
      If the image's actual size is larger than the sourceSize, the image is scaled down. This way
       large images do not use more memory than necessary.
      This is especially important for content that is loaded from external sources or provided by
       the user
      Beware that changing this property dynamically causes the image source to be reloaded,
       potentially even from the network, if it is not in the disk cache.
      More info on Image::sourceSize property: http://doc.qt.nokia.com/latest/qml-
       image.html#sourceSize-prop
    Images are cached and shared internally, so if several Image elements have the same
        source, only one copy of the image will be loaded.
       More info: http://doc.qt.nokia.com/latest/qml-image.html#source-prop


Enable Image.smooth only when necessary
Enabling Image.smooth is bad for performance. Use images in their natural size or disable smooth
filtering in animation.

       Image.smooth property causes the image to be smoothly filtered when scaled or transformed
       Smooth filtering gives better visual quality, but is slower
       If the image is displayed at its natural size, Image.smooth has no visual or performance
        effect
       If you really need to enable Image.smooth, disable smooth filtering at the beginning of the
        animation and re-enable it at the end of the animation (scaling artifacts are only visible if the
        image is stationary on the screen)
       More info on Image.smooth property: http://doc.qt.nokia.com/latest/qml-image.html#smooth-
        prop


Avoid composing an image from a number of elements
It is more efficient to compose an image from a single image than from a number of elements.

       For example, a frame with a shadow could be created using a Rectangle placed over an
        Image providing the shadow
       It is more efficient to provide an image that includes the frame and the shadow
       More info: http://doc.qt.nokia.com/latest/painting-imagecomposition.html
       More info: http://doc.qt.nokia.com/latest/qdeclarativeperformance.html#image-resources-
        over-composition


Lists

Ensure that your data model is as fast as possible
In many cases the slow model is actually the bottleneck in list scrolling performance. Ensure that the
data model is as fast as possible.

       Delegates must be created quickly as the view is flicked. In order to keep the UI responsive,
        your data() function must handle over 1000 calls in a second.
       Any additional functionality that is only needed when the delegate is clicked, for example,
        should be created by a Loader as it is needed.
    Keep the amount of QML to a minimum in your delegate. The fewer elements in a delegate,
        the faster a view can be scrolled.
       In real-world applicationit is advised to write as much code as possible in C++ and use QML
        only for the user interface. This is especially important for list delegates. More
        info: http://cdumez.blogspot.com/2010/11/how-to-use-c-list-model-in-qml.html


CacheBuffer usage in ListView/GridView
In somes cases, cacheBuffer can be useful in improving your ListView/GridView performance. The
default cacheBuffer is zero.

       cacheBuffer property determines whether delegates are instantiated outside the visible area
        of the view
       Note that cacheBuffer is defined in pixels, e.g. if delegate is 20 pixels high, cacheBuffer is set
        to 40 -> up to 2 delegates above and 2 delegates below the visible area may be retained in
        memory
       Setting this value can improve the smoothness of scrolling behavior at the expense of
        additional memory usage. Data is not cached, but the instantiated delegates that requested
        the data are.
       Define cacheBuffer if you have a short list where every item can be cached
       For longer lists cacheBuffer doesn't bring benefit because items are created at the same rate
        as if there were no cache while scrolling fast. The cacheBuffer just postpones the problem,
        i.e. only pushes the position where delegates are created further above/below the visible part
        of list/grid.
       More info: http://doc.qt.nokia.com/latest/qml-listview.html#cacheBuffer-prop


Avoid useless painting
You should always prevent painting the same area several times. For example, prevent
QDeclarativeView from painting its window background if you provide the background of your
application:

       QDeclarativeView window;

       window.setAttribute(Qt::WA_OpaquePaintEvent);

       window.setAttribute(Qt::WA_NoSystemBackground);

       window.viewport()->setAttribute(Qt::WA_OpaquePaintEvent);

       window.viewport()->setAttribute(Qt::WA_NoSystemBackground);
Also, consider using Item as root element rather than Rectangle to avoid painting the background
several times:

      If your root element is a Rectangle, you’re painting every pixel, maybe even several times
      The system (QDeclarativeView) first paints the background and then paints all the QML
       elements
      You might have a Rectangle as the root element and inside that you have a lot of elements
       without opacity covering most of the Rectangle. In this kind of case the system is doing
       useless painting.
      You could instead use Item as root element because it has no visual appearance
      If you need to paint background, but have static UI elements covering part of the screen, you
       can still use Item as the root element and anchor a Rectangle between those static items.
       This way you don’t do useless painting.

More info: http://doc.qt.nokia.com/latest/qdeclarativeperformance.html#rendering
http://doc.qt.nokia.com/latest/qml-item.html
http://juhaturunen.com/blog/2011/04/a-qml-rendering-performance-trick-so-simple-that-it-hurts/
Other

Fixed length lists with Flickable+Column+Repeater
If you have a simple list with fixed length, you could try to optimize performance using
Flickable+Column+Repeater instead of QML ListView. This way list creation would be slower, but list
scrolling smoother.


Beware of string operations

      Multiple uses of the '+' operator usually means multiple memory allocations
      Use StringBuilder for more efficient strings. QStringBuilder uses expression templates and
       reimplements the '%' operator so that when you use '%' for string concatenation instead of
       '+', multiple substring concatenations will be postponed until the final result is about to be
       assigned to a QString. At this point, the amount of memory required for the final result is
       known. The memory allocator is then called once to get the required space, and the
       substrings are copied into it one by one.
      Define: QT_USE_FAST_CONCATENATION, QT_USE_FAST_OPERATOR_PLUS
      See more: http://doc.trolltech.com/latest/qstring.html#more-efficient-string-construction


Animate as small area of the screen as possible in transition animations
If you need to move 3 elements in a second try moving each at a time in 300ms. The system works
so that it calculates the bounds of items that need repainting and paints everything inside those
bounds. This can be really bad two little things are animating in opposite corners.


Avoid complex clipping
You should enable clipping only when you really need it. The default clip value is false.

      If clipping is enabled, an item will clip its own painting, as well as the painting of its children,
       to its bounding rectangle
      More info on Item.clip: http://doc.qt.nokia.com/latest/qml-item.html#clip-prop


Does it help performance if you strip off comments or white space from
QML files?
Not really. The files are reprocessed into a binary in-memory representation at startup, so by runtime
there should be no performance difference. You might be lucky and get 0.5% improvement, and then
only on start-up, nowhere else.


Avoid unnecessary conversion
QML performs a type conversion if the given value for a property does not match with the type
specified for the property. This conversion consumes additional memory.

      For example, Image and BorderImage have a source, which is of type url. Conversion is
       needed if image source's property is defined as string, it should rather be property url.
      Wrong way:

            property string messageAvatar: ""


      Right way:

            property url messageAvatar: ""


Anchors and memory consumption
Anchoring can be done in two ways:

      The normal way:

        Item {

               width: 800

               height: 400

               Button    {

                    id: button;

                    objectName: "Button";

                    anchors.top: parent.top

                    anchors.left: parent.left

                    anchors.right: parent.right

                    text: "Press me";

               }

        }


      This way consumes slightly less memory:
Item {

        width: 800

        height: 400

        Button   {

             id: button;

             objectName: "Button";

             anchors { top: parent.top; left: parent.left; right: parent.right
}

             text: "Press me";

        }

    }
Qt C++

Controlling QObject Ownership
Qt Script uses garbage collection to reclaim memory used by script objects when they are no longer
needed (only if QObject has QScriptEngine::ScriptOwnership); an object's memory can be
automatically reclaimed when it is no longer referenced anywhere in the scripting environment. Qt
Script lets you control what happens to the underlying C++ QObject when the wrapper object is
reclaimed, i.e. whether the QObject is deleted or not. You can only do this when you create an
object by passing an ownership mode as the second argument to QScriptEngine::newQObject(). By
default QObjects are "Qt Ownership" == objects are never deleted, even if you try do so in
scriptengine side.http://doc.qt.nokia.com/latest/scripting.html#controlling-qobject-ownership


Turn off sample buffers
Turn off sample buffers, check if the graphical results are acceptable to you.

       QGLFormat format = QGLFormat::defaultFormat();

       format.setSampleBuffers(false);

       QGLWidget *glWidget = new QGLWidget(format);

    //### potentially faster, but causes junk to appear if top-level is Item,
not Rectangle

       //glWidget->setAutoFillBackground(false);

       view->setViewport(glWidget);


Integrating with a QWidget-based UI
Creating large numbers of QDeclarativeView objects may lead to performance degradation.

       QDeclarativeView is slower to initialize and uses more memory than a QWidget, and creating
        large numbers of QDeclarativeView objects may lead to performance degradation
       Iit may be better to rewrite your widgets in QML, and load the widgets from a main QML
        widget instead of using QDeclarativeView
       QWidgets were designed for a different type of user interface than QML, so it is not always a
        good idea to port a QWidget-based application to QML
       QWidgets are a better choice if your UI is comprised of a small number of complex and static
        elements, and QML is a better choice if your UI is comprised of a large number of simple and
        dynamic elements
    More info: http://doc.qt.nokia.com/latest/qml-integration.html#integrating-with-a-qwidget-
        based-ui


Integrating QML with a QGraphicsView-based UI
If you have an existing UI based on the Graphics View Framework, you can integrate QML widgets
directly into your QGraphicsScene . The following QGraphicsView options are recommended for
optimal performance of QML UIs:

        QGraphicsView ::setOptimizationFlags(QGraphicsView ::DontSavePainterState)
        QGraphicsView ::setViewportUpdateMode(QGraphicsView ::BoundingRectViewportUpdate)
        QGraphicsScene ::setItemIndexMethod(QGraphicsScene ::NoIndex)
       More info: http://doc.qt.nokia.com/latest/qml-integration.html#integrating-with-a-
        qgraphicsview-based-ui


Use showFullScreen()
It is advised to use QDeclarativeView::showFullScreen() instead of QDeclarativeView::show() to
avoid compositing overhead. This should improve the performance a little.


Avoid unnecessary QRegExp construction
QRegExp is faster at matching rather than construction. So if you plan to do a lot of regular
expression matching, construct your QRegExp once, eg as a const static, and use it many times.


Creating QServiceManager instance is costly during application startup
This causes 40-50ms delay for application launch. Create QServiceManager during idle time
instead.


Reduce expense of Qt::SmoothTranformation without reducing image
quality
A combination of Qt::FastTransformation and halfscaling helps to reduce effort for scaling images
without loosing the quality of the resulting image. The trick is to do the scaling in two steps:

   1. Scale down to 2 times of the destination size using Qt::FastTransformation
   2. Scale to the destination size by averaging two pixels in both dimensions using:

       quint32 avg = (((c1 ^ c2) & 0xfefefefeUL) >> 1) + (c1 & c2);

NOTE: Previously we had suggested to scale down to a factor of 1.5 and apply smoothscaling then
however this version should result in much better quality and be much faster. It should be faster as
the new algorithm just requires bit arithmetics and bit shifting. So no floating point operations are
needed anymore and there are no branches anymore that would require branch predictions (as
would be the case for all kinds of bilinear filtering).

Drawback: This optimization does not make any sense for smaller images. Therefore we need a
heuristic to decide whether it should be used.

Simple demo implementation:

       // Smooth scaling is very expensive (size^2). Therefore we reduce the
size

    // to 1.5 of the destination size and using fast transformation
afterwards.

       // Therefore we speed up but don't loose quality..

    if ( canvasPixmap.size().width() > ( 4 * size.toSize().width() ) ) //
Primitive heuristc to decide whether optimization should be used.

          {

            // Improve scaling speed by add an intermediate fast
transformation..

            QSize intermediate_size = QSize( size.toSize().width() * 2,
size.toSize().height() * 2 );

               canvasPixmap = canvasPixmap.scaled( intermediate_size,

                                                              aspectRatioMode,

                                                              Qt::FastTransformation ); //
Cheap operation!

          }

          canvasPixmap = canvasPixmap.scaled( size.toSize(),

                                                         aspectRatioMode,

                                                         Qt::SmoothTransformation ); //
Expensive operation!


Avoid duplicating containers by accident
Qt's container classes like QList, QStringList, QVector, QMap, ... are all implicitly shared. That
means that a simple copy from one to the other (e.g., as a function return value) doesn't cost much:
Only a shallow copy is made, i.e. just the container's management structure is really duplicated; the
items are not copied, just a reference counter is incremented.

As long as you don't try to modify the copy, that is. And even then a deep copy (i.e., all items are
copied, too) is only made if the reference counter is greater than 1.

This is why the coding conventions insist that const_iterators should be used whenever possible: A
non-const iterator is a write access to the container, causing a deep copy. begin() and end() on a
non-const (!) container are what get the deep copy started.

Sometimes you get lucky. This code, for example, is harmless:

      QDir dir(myPath);

      QStringList files=dir.entryList();

      for ( QStringList::iterator it=files.begin(); it != files.end(); ++files
) {

          myWidget->addItem(*it);

      }

But this is just by dumb luck. The call to files.begin() would normally initiate copying
the files QStringList that was returned by dir.entryList(). But since its reference count is back
to 1 after returning from that function, there is no reason for a deep copy; files now holds the only
reference to the items.

Better than relying on dumb luck would be to be explicit:

      QDir dir(myPath);

      QStringList files=dir.entryList();

    for ( QStringList::const_iterator it=files.constBegin(); it !=
files.constEnd(); ++files ) {

          myWidget->addItem(*it);

      }


Avoid implicit copies caused by operator[]
Not so harmless, however, is this piece of code:

      QStringList nameList=myClass.nameList();

      if ( nameList[0]==searchName ) { // deep copy of nameList
...

     }

This seemingly harmless nameList[0] will cause a deep copy of nameList, duplicating each string
it contains in the process. From this copy, element #0 is taken and compared with searchName.
After the if() statement this temporary copy of nameList is immediately destroyed again.

A lot of operations. And so useless.

Why is a temporary object created? Because nameList[] is an LValue - a value that can be used
on the left side of an assignment - like this:

     nameList[0]=someValue;

So this operator[] is a write access on its argument, and Qt's container classes initiate a deep
copy on a write access.

You can get lucky here again: QList etc. have overloaded versions
of operator[]() for const containers. In that case, no deep copy will occur. So this is safe:

     void myFunc( const QStringList &nameList )

     {

           if ( nameList[0]==searchedName) { // no deep copy: nameList is const

                 ...

           }

     }

But again, this is just dumb luck. Better not rely on it. Better be explicit.

Use at() instead of operator[] if you don't want to assign a value to that element.

Avoid a deep copy with code like that:

     QStringList nameList=myClass.nameList();

     if ( nameList.at(0)==searchName ) { // no deep copy

           ...

     }
In the special case of at(0) it is even a little more efficient and and considerably better readable to
use first() :

     QStringList nameList=myClass.nameList();

     if ( nameList.first()==searchName ) {

          ...

     }

Prefer first() to at(0) .

Most Qt containers have first(). Regrettably, QString or QByteArray don't.


Avoid connecting to QGraphicsScene ::changed() signal
To avoid rather expensive change area calculations, do not connect to QGraphicsScene ::changed()
signal.
More info: http://doc.qt.nokia.com/latest/qgraphicsscene.html#changed

More Related Content

What's hot

Qt Memory Management & Signal and Slots
Qt Memory Management & Signal and SlotsQt Memory Management & Signal and Slots
Qt Memory Management & Signal and Slots
Jussi Pohjolainen
 

What's hot (20)

Best Practices in Qt Quick/QML - Part II
Best Practices in Qt Quick/QML - Part IIBest Practices in Qt Quick/QML - Part II
Best Practices in Qt Quick/QML - Part II
 
Qt Internationalization
Qt InternationalizationQt Internationalization
Qt Internationalization
 
Basics of Model/View Qt programming
Basics of Model/View Qt programmingBasics of Model/View Qt programming
Basics of Model/View Qt programming
 
Best Practices in Qt Quick/QML - Part 3
Best Practices in Qt Quick/QML - Part 3Best Practices in Qt Quick/QML - Part 3
Best Practices in Qt Quick/QML - Part 3
 
Qt multi threads
Qt multi threadsQt multi threads
Qt multi threads
 
IPC with Qt
IPC with QtIPC with Qt
IPC with Qt
 
모바일 메신저 아키텍쳐 소개
모바일 메신저 아키텍쳐 소개모바일 메신저 아키텍쳐 소개
모바일 메신저 아키텍쳐 소개
 
(알도개) GraalVM – 자바를 넘어선 새로운 시작의 서막
(알도개) GraalVM – 자바를 넘어선 새로운 시작의 서막(알도개) GraalVM – 자바를 넘어선 새로운 시작의 서막
(알도개) GraalVM – 자바를 넘어선 새로운 시작의 서막
 
Qt for Beginners Part 3 - QML and Qt Quick
Qt for Beginners Part 3 - QML and Qt QuickQt for Beginners Part 3 - QML and Qt Quick
Qt for Beginners Part 3 - QML and Qt Quick
 
Introduction to GraphQL
Introduction to GraphQLIntroduction to GraphQL
Introduction to GraphQL
 
Spring boot
Spring bootSpring boot
Spring boot
 
PlantUML
PlantUMLPlantUML
PlantUML
 
JCR, Sling or AEM? Which API should I use and when?
JCR, Sling or AEM? Which API should I use and when?JCR, Sling or AEM? Which API should I use and when?
JCR, Sling or AEM? Which API should I use and when?
 
Introduction to Qt
Introduction to QtIntroduction to Qt
Introduction to Qt
 
Introduction to the Qt Quick Scene Graph
Introduction to the Qt Quick Scene GraphIntroduction to the Qt Quick Scene Graph
Introduction to the Qt Quick Scene Graph
 
Qt Memory Management & Signal and Slots
Qt Memory Management & Signal and SlotsQt Memory Management & Signal and Slots
Qt Memory Management & Signal and Slots
 
[143] Modern C++ 무조건 써야 해?
[143] Modern C++ 무조건 써야 해?[143] Modern C++ 무조건 써야 해?
[143] Modern C++ 무조건 써야 해?
 
Introduction to QML
Introduction to QMLIntroduction to QML
Introduction to QML
 
Setting up your development environment
Setting up your development environmentSetting up your development environment
Setting up your development environment
 
Concurrency & Parallel Programming
Concurrency & Parallel ProgrammingConcurrency & Parallel Programming
Concurrency & Parallel Programming
 

Viewers also liked

Efficient Graphics with Qt
Efficient Graphics with QtEfficient Graphics with Qt
Efficient Graphics with Qt
Ariya Hidayat
 
"How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use...
"How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use..."How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use...
"How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use...
FELGO SDK
 

Viewers also liked (14)

Intro to QML / Declarative UI
Intro to QML / Declarative UIIntro to QML / Declarative UI
Intro to QML / Declarative UI
 
Optimizing Performance in Qt-Based Applications
Optimizing Performance in Qt-Based ApplicationsOptimizing Performance in Qt-Based Applications
Optimizing Performance in Qt-Based Applications
 
Qt Widget In-Depth
Qt Widget In-DepthQt Widget In-Depth
Qt Widget In-Depth
 
Case Study: Using Qt to Develop Advanced GUIs & Advanced Visualization Software
Case Study: Using Qt to Develop Advanced GUIs & Advanced Visualization SoftwareCase Study: Using Qt to Develop Advanced GUIs & Advanced Visualization Software
Case Study: Using Qt to Develop Advanced GUIs & Advanced Visualization Software
 
Efficient Graphics with Qt
Efficient Graphics with QtEfficient Graphics with Qt
Efficient Graphics with Qt
 
Qt for beginners part 2 widgets
Qt for beginners part 2   widgetsQt for beginners part 2   widgets
Qt for beginners part 2 widgets
 
Qt for beginners part 5 ask the experts
Qt for beginners part 5   ask the expertsQt for beginners part 5   ask the experts
Qt for beginners part 5 ask the experts
 
Qt for beginners part 4 doing more
Qt for beginners part 4   doing moreQt for beginners part 4   doing more
Qt for beginners part 4 doing more
 
How to Make Your Qt App Look Native
How to Make Your Qt App Look NativeHow to Make Your Qt App Look Native
How to Make Your Qt App Look Native
 
Qt for beginners part 1 overview and key concepts
Qt for beginners part 1   overview and key conceptsQt for beginners part 1   overview and key concepts
Qt for beginners part 1 overview and key concepts
 
[Webinar] 10 Keys to Ensuring Success for Your Next Qt Project
[Webinar] 10 Keys to Ensuring Success for Your Next Qt Project[Webinar] 10 Keys to Ensuring Success for Your Next Qt Project
[Webinar] 10 Keys to Ensuring Success for Your Next Qt Project
 
"How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use...
"How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use..."How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use...
"How to Develop with Qt for Multiple Screen Resolutions and Increase Your Use...
 
QtとC++でGUIプログラミング
QtとC++でGUIプログラミングQtとC++でGUIプログラミング
QtとC++でGUIプログラミング
 
Why Zsh is Cooler than Your Shell
Why Zsh is Cooler than Your ShellWhy Zsh is Cooler than Your Shell
Why Zsh is Cooler than Your Shell
 

Similar to Qt and QML performance tips & tricks for Qt 4.7

High Performance Ajax Applications 1197671494632682 2
High Performance Ajax Applications 1197671494632682 2High Performance Ajax Applications 1197671494632682 2
High Performance Ajax Applications 1197671494632682 2
Niti Chotkaew
 
High Performance Ajax Applications
High Performance Ajax ApplicationsHigh Performance Ajax Applications
High Performance Ajax Applications
Julien Lecomte
 

Similar to Qt and QML performance tips & tricks for Qt 4.7 (20)

Introduction to Functional Reactive Web with Clojurescript
Introduction to Functional Reactive Web with ClojurescriptIntroduction to Functional Reactive Web with Clojurescript
Introduction to Functional Reactive Web with Clojurescript
 
Ask the expert - App performance on Series 40 phones
Ask the expert - App performance on Series 40 phonesAsk the expert - App performance on Series 40 phones
Ask the expert - App performance on Series 40 phones
 
High Performance Ajax Applications 1197671494632682 2
High Performance Ajax Applications 1197671494632682 2High Performance Ajax Applications 1197671494632682 2
High Performance Ajax Applications 1197671494632682 2
 
High Performance Ajax Applications
High Performance Ajax ApplicationsHigh Performance Ajax Applications
High Performance Ajax Applications
 
DDD Framework for Java: JdonFramework
DDD Framework for Java: JdonFrameworkDDD Framework for Java: JdonFramework
DDD Framework for Java: JdonFramework
 
Ultimate Guide to Microservice Architecture on Kubernetes
Ultimate Guide to Microservice Architecture on KubernetesUltimate Guide to Microservice Architecture on Kubernetes
Ultimate Guide to Microservice Architecture on Kubernetes
 
Top 7 Angular Best Practices to Organize Your Angular App
Top 7 Angular Best Practices to Organize Your Angular AppTop 7 Angular Best Practices to Organize Your Angular App
Top 7 Angular Best Practices to Organize Your Angular App
 
Developing high performance and responsive web apps using web worker
Developing high performance and responsive web apps using web workerDeveloping high performance and responsive web apps using web worker
Developing high performance and responsive web apps using web worker
 
Developing Series 40 web apps with Nokia Web Tools 2.0
Developing Series 40 web apps with Nokia Web Tools 2.0Developing Series 40 web apps with Nokia Web Tools 2.0
Developing Series 40 web apps with Nokia Web Tools 2.0
 
The Theory Of The Dom
The Theory Of The DomThe Theory Of The Dom
The Theory Of The Dom
 
Large-scaled Deploy Over 100 Servers in 3 Minutes
Large-scaled Deploy Over 100 Servers in 3 MinutesLarge-scaled Deploy Over 100 Servers in 3 Minutes
Large-scaled Deploy Over 100 Servers in 3 Minutes
 
Architecting Microservices in .Net
Architecting Microservices in .NetArchitecting Microservices in .Net
Architecting Microservices in .Net
 
Spring, Functions, Serverless and You
Spring, Functions, Serverless and YouSpring, Functions, Serverless and You
Spring, Functions, Serverless and You
 
Google I/O 2019 - what's new in Android Q and Jetpack
Google I/O 2019 - what's new in Android Q and JetpackGoogle I/O 2019 - what's new in Android Q and Jetpack
Google I/O 2019 - what's new in Android Q and Jetpack
 
NodeJS : Communication and Round Robin Way
NodeJS : Communication and Round Robin WayNodeJS : Communication and Round Robin Way
NodeJS : Communication and Round Robin Way
 
Dark launching with Consul at Hootsuite - Bill Monkman
Dark launching with Consul at Hootsuite - Bill MonkmanDark launching with Consul at Hootsuite - Bill Monkman
Dark launching with Consul at Hootsuite - Bill Monkman
 
SCM Puppet: from an intro to the scaling
SCM Puppet: from an intro to the scalingSCM Puppet: from an intro to the scaling
SCM Puppet: from an intro to the scaling
 
five Sling features you should know
five Sling features you should knowfive Sling features you should know
five Sling features you should know
 
Active web page chapter for reading purpose
Active web page chapter for reading purposeActive web page chapter for reading purpose
Active web page chapter for reading purpose
 
Building reusable components as micro frontends with glimmer js and webcompo...
Building reusable components as micro frontends  with glimmer js and webcompo...Building reusable components as micro frontends  with glimmer js and webcompo...
Building reusable components as micro frontends with glimmer js and webcompo...
 

Recently uploaded

CORS (Kitworks Team Study 양다윗 발표자료 240510)
CORS (Kitworks Team Study 양다윗 발표자료 240510)CORS (Kitworks Team Study 양다윗 발표자료 240510)
CORS (Kitworks Team Study 양다윗 발표자료 240510)
Wonjun Hwang
 
Harnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptx
Harnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptxHarnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptx
Harnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptx
FIDO Alliance
 

Recently uploaded (20)

ADP Passwordless Journey Case Study.pptx
ADP Passwordless Journey Case Study.pptxADP Passwordless Journey Case Study.pptx
ADP Passwordless Journey Case Study.pptx
 
ERP Contender Series: Acumatica vs. Sage Intacct
ERP Contender Series: Acumatica vs. Sage IntacctERP Contender Series: Acumatica vs. Sage Intacct
ERP Contender Series: Acumatica vs. Sage Intacct
 
Observability Concepts EVERY Developer Should Know (DevOpsDays Seattle)
Observability Concepts EVERY Developer Should Know (DevOpsDays Seattle)Observability Concepts EVERY Developer Should Know (DevOpsDays Seattle)
Observability Concepts EVERY Developer Should Know (DevOpsDays Seattle)
 
The Ultimate Prompt Engineering Guide for Generative AI: Get the Most Out of ...
The Ultimate Prompt Engineering Guide for Generative AI: Get the Most Out of ...The Ultimate Prompt Engineering Guide for Generative AI: Get the Most Out of ...
The Ultimate Prompt Engineering Guide for Generative AI: Get the Most Out of ...
 
Event-Driven Architecture Masterclass: Engineering a Robust, High-performance...
Event-Driven Architecture Masterclass: Engineering a Robust, High-performance...Event-Driven Architecture Masterclass: Engineering a Robust, High-performance...
Event-Driven Architecture Masterclass: Engineering a Robust, High-performance...
 
CORS (Kitworks Team Study 양다윗 발표자료 240510)
CORS (Kitworks Team Study 양다윗 발표자료 240510)CORS (Kitworks Team Study 양다윗 발표자료 240510)
CORS (Kitworks Team Study 양다윗 발표자료 240510)
 
Design Guidelines for Passkeys 2024.pptx
Design Guidelines for Passkeys 2024.pptxDesign Guidelines for Passkeys 2024.pptx
Design Guidelines for Passkeys 2024.pptx
 
Introduction to use of FHIR Documents in ABDM
Introduction to use of FHIR Documents in ABDMIntroduction to use of FHIR Documents in ABDM
Introduction to use of FHIR Documents in ABDM
 
Overview of Hyperledger Foundation
Overview of Hyperledger FoundationOverview of Hyperledger Foundation
Overview of Hyperledger Foundation
 
The Zero-ETL Approach: Enhancing Data Agility and Insight
The Zero-ETL Approach: Enhancing Data Agility and InsightThe Zero-ETL Approach: Enhancing Data Agility and Insight
The Zero-ETL Approach: Enhancing Data Agility and Insight
 
AI mind or machine power point presentation
AI mind or machine power point presentationAI mind or machine power point presentation
AI mind or machine power point presentation
 
ChatGPT and Beyond - Elevating DevOps Productivity
ChatGPT and Beyond - Elevating DevOps ProductivityChatGPT and Beyond - Elevating DevOps Productivity
ChatGPT and Beyond - Elevating DevOps Productivity
 
AI in Action: Real World Use Cases by Anitaraj
AI in Action: Real World Use Cases by AnitarajAI in Action: Real World Use Cases by Anitaraj
AI in Action: Real World Use Cases by Anitaraj
 
UiPath manufacturing technology benefits and AI overview
UiPath manufacturing technology benefits and AI overviewUiPath manufacturing technology benefits and AI overview
UiPath manufacturing technology benefits and AI overview
 
Harnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptx
Harnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptxHarnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptx
Harnessing Passkeys in the Battle Against AI-Powered Cyber Threats.pptx
 
الأمن السيبراني - ما لا يسع للمستخدم جهله
الأمن السيبراني - ما لا يسع للمستخدم جهلهالأمن السيبراني - ما لا يسع للمستخدم جهله
الأمن السيبراني - ما لا يسع للمستخدم جهله
 
2024 May Patch Tuesday
2024 May Patch Tuesday2024 May Patch Tuesday
2024 May Patch Tuesday
 
(Explainable) Data-Centric AI: what are you explaininhg, and to whom?
(Explainable) Data-Centric AI: what are you explaininhg, and to whom?(Explainable) Data-Centric AI: what are you explaininhg, and to whom?
(Explainable) Data-Centric AI: what are you explaininhg, and to whom?
 
How to Check GPS Location with a Live Tracker in Pakistan
How to Check GPS Location with a Live Tracker in PakistanHow to Check GPS Location with a Live Tracker in Pakistan
How to Check GPS Location with a Live Tracker in Pakistan
 
Frisco Automating Purchase Orders with MuleSoft IDP- May 10th, 2024.pptx.pdf
Frisco Automating Purchase Orders with MuleSoft IDP- May 10th, 2024.pptx.pdfFrisco Automating Purchase Orders with MuleSoft IDP- May 10th, 2024.pptx.pdf
Frisco Automating Purchase Orders with MuleSoft IDP- May 10th, 2024.pptx.pdf
 

Qt and QML performance tips & tricks for Qt 4.7

  • 1. Performance tips and tricks for Qt/QML applications General Do not use JavaScript, but rather implement the UI with QML and the engine with C++ JavaScript should be avoided as much as possible. Do not use it at all.  JavaScript code should be kept to a minimum. C++ should be used to implement application's the business logic.  JavaScript is especially bad for performance when it is run during animation, e.g. in lists the JavaScript monster gets evaluated on every frame of the scrolling animation as many times as there are instantiated delegates.  If you must use JavaScript anyway, simplify your JavaScript expressions as much as possible and write them inline if they fit on one line. QML is compiled to an optimized bytecode-like stream and the JavaScript expressions pass through an optimized evaluator for simple expressions. More info: http://doc.qt.nokia.com/latest/qdeclarativejavascript.html#inline-javascript  It is also recommended to set values for properties using QML property binding than JavaScript value assignment  Wrong: itemTitle: { if(targetFeedItem == null){ return ""; } return targetFeedItem.title; }  Right: itemTitle: targetFeedItem == undefined ? "" : targetFeedItem.title;
  • 2. Use threads or multiprocessing to allow responsive UI If you don't use threads for long-lasting operations, the UI is blocked. Use threads for the long lasting operations such as data retrieval to keep UI responsive.  In QML the WorkerScript element can be used to run JavaScript operations in a new thread (you shouldn't use JavaScript in a serious application) o DO NOT use multiple WorkerScripts or you are in trouble. o If you initialize a large list synchronously in a JavaScript block (this is something that a serious application would never do...) it is very likely that UI jamming will occure with high memory consumption. The memory consumption is likely due to the fact that garbage collector does not get a chance to collect any items while a synchronous loop is processing. o More info on WorkerScript: http://doc.qt.nokia.com/latest/qml-workerscript.html  C++ threads should be used for all serious applications o This can be done by creating a QDeclarativeExtensionPlugin and exposing the needed QObjects to QML o In the QObject one exposes the properties, invokable methods, signals and slots needed in the UI o For any long (>2ms) operations one should use QRunnable with QThreadPool or QFuture see. More info: http://doc.trolltech.com/latest/threads.html o The long lasting operations (> N ms) must be implemented as asyncronous server calls or by using threads so that the UI can update itself while the application is doing long lasting backend operations Use anchors rather than bindings to position items relative to each other Wrong way to bind rect2 to rect1: Rectangle { id: rect1 x: 20 width: 200; height: 200 } Rectangle { id: rect2
  • 3. x: rect1.x y: rect1.y + rect1.height width: rect1.width - 20 height: 200 } Right way to do this, anchor rect2 to rect1: Rectangle { id: rect1 x: 20 width: 200; height: 200 } Rectangle { id: rect2 height: 200 anchors.left: rect1.left anchors.top: rect1.bottom anchors.right: rect1.right anchors.rightMargin: 20 } More info: http://doc.qt.nokia.com/latest/qdeclarativeperformance.html#anchors-vs-binding
  • 4. App start Use Booster On Symbian and MeeGo it is possible to speed up application start by picking up pre-instantiated QApplication and QDeclarativeView from cache. Please contact Nokia for details how to do this. Use loaders to dynamically load/unload QML QML application starts quite slow if the whole application is implemented in one huge QML file. Partition our application wisely into logical entities, load minimum QML at start and load more as you need it using Loaders .  The Loader item can be used to dynamically load and unload visual QML components defined in a QML file or items/components defined within a QML file  This dynamical behavior allows the developer to control the memory usage and startup speed of an application. More info: http://doc.qt.nokia.com/latest/qml-loader.html#details  Partition your application into several QML files so that each file contains a logical UI entity. This way loading and unloading are easier to control. DO NOT have one huge QML file per application.  Load absolutely minimum amount of QML at application start to make your application start as quickly as possible. You can connect to network and show spinner etc. after the application UI is visible. If your first view is very complex and requires lots of QML to be loaded, show a splash screen or something.  You should load pieces of UI only by demand e.g. when the user navigates to another view, but on the other hand it may require more time to navigate between views  Creating a dialog can take more than 100ms of time. Make sure you use Loader to load the dialog. import QtQuick 1.0 Item { width: 200; height: 200 Loader { id: pageLoader }
  • 5. MouseArea { anchors.fill: parent onClicked: pageLoader.source = "Page1.qml" } }
  • 6. Images Bitmap formats vs. vector graphics formats Images can be supplied in any of the standard image formats supported by Qt, including bitmap formats such as PNG and JPEG, and vector graphics formats such as SVG. However, rendering an SVG is slow compared to a bitmap image. Load large images asynchronously If you load images synchronously the UI is blocked. In many cases images are not needed visible immediately, but they can be lazy-loaded.  You should load images asynchronously in a separate thread if an image doesn't need to be displayed immediately. This can be done by settings QML Image asynchronous to true.  This way the user interface stays responsive user interface.  Note that this property is only valid for images read from the local filesystem. Images that are loaded via a network resource (e.g. HTTP) are always loaded asynchonously.  More info: http://doc.qt.nokia.com/latest/qml-image.html#asynchronous-prop Avoid resizing/scaling Resizing/scaling is a very heavy operation in QML. Use images that are in their native size rather than resize/scale large images to correct size. Use sourceSize with large images Images are often the greatest user of memory in QML user interfaces.  sourceSize should be used with large images because property sets the actual number of pixels stored for the loaded image  If you have a HUGE image 3264 x 2448, but you set sourceSize 204x153, then it scaled down and will be stored as 204x153 in memory.  If the image's actual size is larger than the sourceSize, the image is scaled down. This way large images do not use more memory than necessary.  This is especially important for content that is loaded from external sources or provided by the user  Beware that changing this property dynamically causes the image source to be reloaded, potentially even from the network, if it is not in the disk cache.  More info on Image::sourceSize property: http://doc.qt.nokia.com/latest/qml- image.html#sourceSize-prop
  • 7. Images are cached and shared internally, so if several Image elements have the same source, only one copy of the image will be loaded.  More info: http://doc.qt.nokia.com/latest/qml-image.html#source-prop Enable Image.smooth only when necessary Enabling Image.smooth is bad for performance. Use images in their natural size or disable smooth filtering in animation.  Image.smooth property causes the image to be smoothly filtered when scaled or transformed  Smooth filtering gives better visual quality, but is slower  If the image is displayed at its natural size, Image.smooth has no visual or performance effect  If you really need to enable Image.smooth, disable smooth filtering at the beginning of the animation and re-enable it at the end of the animation (scaling artifacts are only visible if the image is stationary on the screen)  More info on Image.smooth property: http://doc.qt.nokia.com/latest/qml-image.html#smooth- prop Avoid composing an image from a number of elements It is more efficient to compose an image from a single image than from a number of elements.  For example, a frame with a shadow could be created using a Rectangle placed over an Image providing the shadow  It is more efficient to provide an image that includes the frame and the shadow  More info: http://doc.qt.nokia.com/latest/painting-imagecomposition.html  More info: http://doc.qt.nokia.com/latest/qdeclarativeperformance.html#image-resources- over-composition Lists Ensure that your data model is as fast as possible In many cases the slow model is actually the bottleneck in list scrolling performance. Ensure that the data model is as fast as possible.  Delegates must be created quickly as the view is flicked. In order to keep the UI responsive, your data() function must handle over 1000 calls in a second.  Any additional functionality that is only needed when the delegate is clicked, for example, should be created by a Loader as it is needed.
  • 8. Keep the amount of QML to a minimum in your delegate. The fewer elements in a delegate, the faster a view can be scrolled.  In real-world applicationit is advised to write as much code as possible in C++ and use QML only for the user interface. This is especially important for list delegates. More info: http://cdumez.blogspot.com/2010/11/how-to-use-c-list-model-in-qml.html CacheBuffer usage in ListView/GridView In somes cases, cacheBuffer can be useful in improving your ListView/GridView performance. The default cacheBuffer is zero.  cacheBuffer property determines whether delegates are instantiated outside the visible area of the view  Note that cacheBuffer is defined in pixels, e.g. if delegate is 20 pixels high, cacheBuffer is set to 40 -> up to 2 delegates above and 2 delegates below the visible area may be retained in memory  Setting this value can improve the smoothness of scrolling behavior at the expense of additional memory usage. Data is not cached, but the instantiated delegates that requested the data are.  Define cacheBuffer if you have a short list where every item can be cached  For longer lists cacheBuffer doesn't bring benefit because items are created at the same rate as if there were no cache while scrolling fast. The cacheBuffer just postpones the problem, i.e. only pushes the position where delegates are created further above/below the visible part of list/grid.  More info: http://doc.qt.nokia.com/latest/qml-listview.html#cacheBuffer-prop Avoid useless painting You should always prevent painting the same area several times. For example, prevent QDeclarativeView from painting its window background if you provide the background of your application: QDeclarativeView window; window.setAttribute(Qt::WA_OpaquePaintEvent); window.setAttribute(Qt::WA_NoSystemBackground); window.viewport()->setAttribute(Qt::WA_OpaquePaintEvent); window.viewport()->setAttribute(Qt::WA_NoSystemBackground);
  • 9. Also, consider using Item as root element rather than Rectangle to avoid painting the background several times:  If your root element is a Rectangle, you’re painting every pixel, maybe even several times  The system (QDeclarativeView) first paints the background and then paints all the QML elements  You might have a Rectangle as the root element and inside that you have a lot of elements without opacity covering most of the Rectangle. In this kind of case the system is doing useless painting.  You could instead use Item as root element because it has no visual appearance  If you need to paint background, but have static UI elements covering part of the screen, you can still use Item as the root element and anchor a Rectangle between those static items. This way you don’t do useless painting. More info: http://doc.qt.nokia.com/latest/qdeclarativeperformance.html#rendering http://doc.qt.nokia.com/latest/qml-item.html http://juhaturunen.com/blog/2011/04/a-qml-rendering-performance-trick-so-simple-that-it-hurts/
  • 10. Other Fixed length lists with Flickable+Column+Repeater If you have a simple list with fixed length, you could try to optimize performance using Flickable+Column+Repeater instead of QML ListView. This way list creation would be slower, but list scrolling smoother. Beware of string operations  Multiple uses of the '+' operator usually means multiple memory allocations  Use StringBuilder for more efficient strings. QStringBuilder uses expression templates and reimplements the '%' operator so that when you use '%' for string concatenation instead of '+', multiple substring concatenations will be postponed until the final result is about to be assigned to a QString. At this point, the amount of memory required for the final result is known. The memory allocator is then called once to get the required space, and the substrings are copied into it one by one.  Define: QT_USE_FAST_CONCATENATION, QT_USE_FAST_OPERATOR_PLUS  See more: http://doc.trolltech.com/latest/qstring.html#more-efficient-string-construction Animate as small area of the screen as possible in transition animations If you need to move 3 elements in a second try moving each at a time in 300ms. The system works so that it calculates the bounds of items that need repainting and paints everything inside those bounds. This can be really bad two little things are animating in opposite corners. Avoid complex clipping You should enable clipping only when you really need it. The default clip value is false.  If clipping is enabled, an item will clip its own painting, as well as the painting of its children, to its bounding rectangle  More info on Item.clip: http://doc.qt.nokia.com/latest/qml-item.html#clip-prop Does it help performance if you strip off comments or white space from QML files? Not really. The files are reprocessed into a binary in-memory representation at startup, so by runtime there should be no performance difference. You might be lucky and get 0.5% improvement, and then only on start-up, nowhere else. Avoid unnecessary conversion
  • 11. QML performs a type conversion if the given value for a property does not match with the type specified for the property. This conversion consumes additional memory.  For example, Image and BorderImage have a source, which is of type url. Conversion is needed if image source's property is defined as string, it should rather be property url.  Wrong way: property string messageAvatar: ""  Right way: property url messageAvatar: "" Anchors and memory consumption Anchoring can be done in two ways:  The normal way: Item { width: 800 height: 400 Button { id: button; objectName: "Button"; anchors.top: parent.top anchors.left: parent.left anchors.right: parent.right text: "Press me"; } }  This way consumes slightly less memory:
  • 12. Item { width: 800 height: 400 Button { id: button; objectName: "Button"; anchors { top: parent.top; left: parent.left; right: parent.right } text: "Press me"; } }
  • 13. Qt C++ Controlling QObject Ownership Qt Script uses garbage collection to reclaim memory used by script objects when they are no longer needed (only if QObject has QScriptEngine::ScriptOwnership); an object's memory can be automatically reclaimed when it is no longer referenced anywhere in the scripting environment. Qt Script lets you control what happens to the underlying C++ QObject when the wrapper object is reclaimed, i.e. whether the QObject is deleted or not. You can only do this when you create an object by passing an ownership mode as the second argument to QScriptEngine::newQObject(). By default QObjects are "Qt Ownership" == objects are never deleted, even if you try do so in scriptengine side.http://doc.qt.nokia.com/latest/scripting.html#controlling-qobject-ownership Turn off sample buffers Turn off sample buffers, check if the graphical results are acceptable to you. QGLFormat format = QGLFormat::defaultFormat(); format.setSampleBuffers(false); QGLWidget *glWidget = new QGLWidget(format); //### potentially faster, but causes junk to appear if top-level is Item, not Rectangle //glWidget->setAutoFillBackground(false); view->setViewport(glWidget); Integrating with a QWidget-based UI Creating large numbers of QDeclarativeView objects may lead to performance degradation.  QDeclarativeView is slower to initialize and uses more memory than a QWidget, and creating large numbers of QDeclarativeView objects may lead to performance degradation  Iit may be better to rewrite your widgets in QML, and load the widgets from a main QML widget instead of using QDeclarativeView  QWidgets were designed for a different type of user interface than QML, so it is not always a good idea to port a QWidget-based application to QML  QWidgets are a better choice if your UI is comprised of a small number of complex and static elements, and QML is a better choice if your UI is comprised of a large number of simple and dynamic elements
  • 14. More info: http://doc.qt.nokia.com/latest/qml-integration.html#integrating-with-a-qwidget- based-ui Integrating QML with a QGraphicsView-based UI If you have an existing UI based on the Graphics View Framework, you can integrate QML widgets directly into your QGraphicsScene . The following QGraphicsView options are recommended for optimal performance of QML UIs:  QGraphicsView ::setOptimizationFlags(QGraphicsView ::DontSavePainterState)  QGraphicsView ::setViewportUpdateMode(QGraphicsView ::BoundingRectViewportUpdate)  QGraphicsScene ::setItemIndexMethod(QGraphicsScene ::NoIndex)  More info: http://doc.qt.nokia.com/latest/qml-integration.html#integrating-with-a- qgraphicsview-based-ui Use showFullScreen() It is advised to use QDeclarativeView::showFullScreen() instead of QDeclarativeView::show() to avoid compositing overhead. This should improve the performance a little. Avoid unnecessary QRegExp construction QRegExp is faster at matching rather than construction. So if you plan to do a lot of regular expression matching, construct your QRegExp once, eg as a const static, and use it many times. Creating QServiceManager instance is costly during application startup This causes 40-50ms delay for application launch. Create QServiceManager during idle time instead. Reduce expense of Qt::SmoothTranformation without reducing image quality A combination of Qt::FastTransformation and halfscaling helps to reduce effort for scaling images without loosing the quality of the resulting image. The trick is to do the scaling in two steps: 1. Scale down to 2 times of the destination size using Qt::FastTransformation 2. Scale to the destination size by averaging two pixels in both dimensions using: quint32 avg = (((c1 ^ c2) & 0xfefefefeUL) >> 1) + (c1 & c2); NOTE: Previously we had suggested to scale down to a factor of 1.5 and apply smoothscaling then however this version should result in much better quality and be much faster. It should be faster as the new algorithm just requires bit arithmetics and bit shifting. So no floating point operations are
  • 15. needed anymore and there are no branches anymore that would require branch predictions (as would be the case for all kinds of bilinear filtering). Drawback: This optimization does not make any sense for smaller images. Therefore we need a heuristic to decide whether it should be used. Simple demo implementation: // Smooth scaling is very expensive (size^2). Therefore we reduce the size // to 1.5 of the destination size and using fast transformation afterwards. // Therefore we speed up but don't loose quality.. if ( canvasPixmap.size().width() > ( 4 * size.toSize().width() ) ) // Primitive heuristc to decide whether optimization should be used. { // Improve scaling speed by add an intermediate fast transformation.. QSize intermediate_size = QSize( size.toSize().width() * 2, size.toSize().height() * 2 ); canvasPixmap = canvasPixmap.scaled( intermediate_size, aspectRatioMode, Qt::FastTransformation ); // Cheap operation! } canvasPixmap = canvasPixmap.scaled( size.toSize(), aspectRatioMode, Qt::SmoothTransformation ); // Expensive operation! Avoid duplicating containers by accident Qt's container classes like QList, QStringList, QVector, QMap, ... are all implicitly shared. That means that a simple copy from one to the other (e.g., as a function return value) doesn't cost much:
  • 16. Only a shallow copy is made, i.e. just the container's management structure is really duplicated; the items are not copied, just a reference counter is incremented. As long as you don't try to modify the copy, that is. And even then a deep copy (i.e., all items are copied, too) is only made if the reference counter is greater than 1. This is why the coding conventions insist that const_iterators should be used whenever possible: A non-const iterator is a write access to the container, causing a deep copy. begin() and end() on a non-const (!) container are what get the deep copy started. Sometimes you get lucky. This code, for example, is harmless: QDir dir(myPath); QStringList files=dir.entryList(); for ( QStringList::iterator it=files.begin(); it != files.end(); ++files ) { myWidget->addItem(*it); } But this is just by dumb luck. The call to files.begin() would normally initiate copying the files QStringList that was returned by dir.entryList(). But since its reference count is back to 1 after returning from that function, there is no reason for a deep copy; files now holds the only reference to the items. Better than relying on dumb luck would be to be explicit: QDir dir(myPath); QStringList files=dir.entryList(); for ( QStringList::const_iterator it=files.constBegin(); it != files.constEnd(); ++files ) { myWidget->addItem(*it); } Avoid implicit copies caused by operator[] Not so harmless, however, is this piece of code: QStringList nameList=myClass.nameList(); if ( nameList[0]==searchName ) { // deep copy of nameList
  • 17. ... } This seemingly harmless nameList[0] will cause a deep copy of nameList, duplicating each string it contains in the process. From this copy, element #0 is taken and compared with searchName. After the if() statement this temporary copy of nameList is immediately destroyed again. A lot of operations. And so useless. Why is a temporary object created? Because nameList[] is an LValue - a value that can be used on the left side of an assignment - like this: nameList[0]=someValue; So this operator[] is a write access on its argument, and Qt's container classes initiate a deep copy on a write access. You can get lucky here again: QList etc. have overloaded versions of operator[]() for const containers. In that case, no deep copy will occur. So this is safe: void myFunc( const QStringList &nameList ) { if ( nameList[0]==searchedName) { // no deep copy: nameList is const ... } } But again, this is just dumb luck. Better not rely on it. Better be explicit. Use at() instead of operator[] if you don't want to assign a value to that element. Avoid a deep copy with code like that: QStringList nameList=myClass.nameList(); if ( nameList.at(0)==searchName ) { // no deep copy ... }
  • 18. In the special case of at(0) it is even a little more efficient and and considerably better readable to use first() : QStringList nameList=myClass.nameList(); if ( nameList.first()==searchName ) { ... } Prefer first() to at(0) . Most Qt containers have first(). Regrettably, QString or QByteArray don't. Avoid connecting to QGraphicsScene ::changed() signal To avoid rather expensive change area calculations, do not connect to QGraphicsScene ::changed() signal. More info: http://doc.qt.nokia.com/latest/qgraphicsscene.html#changed