[Qt-jambi-interest] QPixmap erratic performance

[Gunnar Sletta]

Bart van Deenen wrote:
> On Monday 17 November 2008 10:09:51 Bart van Deenen wrote:
> 
>  > On Monday 17 November 2008 09:54:38 Gunnar Sletta wrote:
> 
>  > > > Ok. I'm starting to get somewhere. I've added instrumentation to the
> 
>  > > > code (not shown here) which uses System.currentTimeMillis() to 
> measure
> 
>  > > > intervals, and it turns out that the interval between two consecutive
> 
>  > > > calls to timerEvent is very erratic! My 'render' function (currently
> 
>  > > > with a much simplified QPixmap tree under it) only takes a few 
> ms. The
> 
>  > > > QTimer is supposed to run at 40 ms interval, but the actual intervals
> 
>  > > > are anywhere between 15 and 300 ms!
> 
>  > >
> 
>  > > A timer tries to fire as close to, but at least 40 ms after the last
> 
>  > > one, but there are no guarantees. I've never seen such behaviour, so I
> 
>  > > would have to see your code to know more
> 
>  >
> 
>  > I'm simplifying the code to narrow it down to where the problem 
> occurs, and
> 
>  > then I'll post it. Thanks for thinking along.
> 
>  >
> 
>  > Bart
> 
> I've narrowed down my source code to something simple. Unfortunately, 
> now the erraticness is gone! It seems to have to do with the number or 
> size of QPixmaps I'm using to draw my main QPixmap :-(
> 
> Here's my non-erratic code. The only difference with my problem code is 
> that the 'render' function here consists of one QPixmap being drawn into 
> my root object, and that with the trouble code there's a lot of QPixmaps 
> rendering into each other, and finally being combined into the root 
> QPixmap. The problem is not the overal time it takes to do the 
> rendering; with a simplified tree of objects that is consistently below 
> 5 ms. I don't understand where the problem is starting to occur :-(. 
> Next I'll add some more of my 'sprite_object's to see if it's the number 
> or size of them that matters.

There is also the fact that the precision of System.currentTimeMillis() 
is system dependent. On windows its precision is ~17ms on Linux I 
believe it is between 1 and 5ms, so you also have to factor that into 
benchmarks and run enough iterations for the precision to be of no 
consequence.

best regards,
Gunnar