Qt/Embedded Performance Tuning

General programming style

The following guidelines will improve CPU performance:

• Create dialogs and widgets once, then QWidget::hide() and QWidget::show() them, rather than creating them and deleting them every time they are needed. This will use a little more memory, but will be much faster. Try to create them the first time "lazily" to avoid slow startup (e.g. only create a Find dialog the first time the user invokes it).

Static vs. Dynamic linking

A lot of CPU and memory is used by the ELF linking process. You can make significant savings by using a static build of your application suite. This means that rather than having a dynamic library (libqte.so) and a collection of executables which link dynamically to that library, you build all the applications into a single executable and statically link that with a static library (libqt.a). This improves start-up time, and reduces memory usage, at the expense of flexibility (to add a new application, you must recompile the single executable) and robustness (if one application has a bug, it might harm other applications). If you need to install end-user applications, this may not be an option, but if you are building a single application suite for a device with limited CPU power and memory, this option could be very beneficial.

To compile Qt as a static library, add the -static options when you run configure.

To build your application suite as an all-in-one application, design each application as a stand-alone widget or set of widgets, with only minimal code in the main() function. Then, write an application that gives some way to switch between the applications (e.g. a QIconView). The QPE is an example of this. It can be built either as a set of dynamically-linked executables, or as a single static application.

Note that you should generally still link dynamically against the standard C library and any other libraries which might be used by other applications on your device.

Alternative memory allocation

We have found that the libraries shipped with some C++ compilers on some platforms have poor performance in the built-in "new" and "delete" operators. You might gain performance by re-implementing these functions. For example, you can switch to the plain C allocators by adding the following to your code:

    void* operator new[]( size_t size )
    {
        return malloc( size );
    }

    void* operator new( size_t size )
    {
        return malloc( size );
    }

    void operator delete[]( void *p )
    {
        free( p );
    }

    void operator delete[]( void *p, size_t size )
    {
        free( p );
    }

    void operator delete( void *p )
    {
        free( p );
    }

    void operator delete( void *p, size_t size )
    {
        free( p );
    }