INFO: Tips for Improving Performance of VCMac Applications (132874)

MORE INFORMATION

Tips for Improving Performance

• To determine which portions of the application need to be optimized, use the Profiler, which is documented in Appendix C of the 68K Programmer's Guide.
• The process of creating a memory DC, creating a bitmap, and selecting the bitmap into the DC is expensive when performed on the Macintosh. It is faster to create the DC and cache it instead of recreating it every time.
• Copying DIBs to the screen or to other DCs is expensive because the DIB color table must be converted and the DIB data must be reoriented. If a DIB will be rendered frequently, copy the DIB to a memory DC. For each screen update, copy the DC to the screen instead of the DIB.
• Currently PeekMessage and GetMessage call WaitNextEvent if there aren't any messages waiting in the internal queue. WaitNextEvent on the Macintosh is considerably slower than PeekMessage and GetMessage are under Windows. You may want to wrap calls to PeekMessage or GetMessage in functions that limit the frequency by which PeekMessage or GetMessage are called to every two ticks. For more information on ticks, please see TickCount in the Inside Macintosh documentation.
• Applications frequently perform background processing that is terminated as soon as there are messages to process. PeekMessage(PM_NOREMOVE) is often used to test for the existence of messages in the queue. To avoid the overhead associated with calling PeekMessage(), use GetInputState() instead. GetInputState doesn[ASCII 146]t call WaitNextEvent.
• If you allocate a lot of memory in fixed blocks (GlobalAlloc(GPTR) or LocalAlloc(LPTR)), switch to using relocatable blocks instead, and lock these blocks of memory while dereferencing them. This will reduce heap fragmentation and make memory allocations faster.
• Increase your partition size to reduce the amount of code swapping and to prevent thrashing. The partition size is determined by the SIZE resource defined in a project[ASCII 146]s .R file. The Generic.Mac sample contains an example .R file with a SIZE resource.
• Segment the application so that code segments can be swapped out efficiently. Use consistent segment sizes such as 30k to 50k. Additionally, use the Profiler to optimize segmentation.
• Use Compiler (speed or size) and Linker (opt:ref) optimizations available in the Project Settings. The debug Windows Portability Libraries perform a lot of parameter checking, so comparing speed and size with the debug libraries is not a valid exercise.
• To reduce the size of the application, remove MacsBug symbols from shipping code.
• To reduce the application size further, remove segment names from the CODE resources, and delete the MSCV resource used for debugging. Use an MRC script to do this.
• Use special case drawing and screen updating associated with user feedback. For example, if you want to drag a shape across the screen, the Windows way of invalidating the old location (updating data structures and waiting for the WM_PAINT message to draw the update) does not work well on the Macintosh. A more efficient method follows in pseudo code:

     hdc = GetDC(hwnd);
     do
     {
        msg = PeekMsg (mouse msgs only)
  
        switch (msg)
        {
           case WM_MOUSEMOVE:
              erase old one
              update data structure
              draw new location
              break;
           case ...
  
       } while msg != WM_MOUSEUP
  						

• The startup and shutdown times can be negatively impacted by reading or writing a lot of preferences with ProfileString APIs. Try to limit the number of preferences to be written out.
• If the Profiler indicates that a specific WLM API is consuming too much time, use the wrapper functions to obtain the related data structure, and then call the appropriate Macintosh API directly.
• Property sheets and property pages implemented using MFC would be slow on the Mac. To workaround, use your own implementation of property pages. You could use native Mac calls to create property pages.