QwtCurve faster repaint

[baray98]

I have more 4 curves in a plot and each curve have 2000 points in them , these are from the scaled down version from my millions of points in my data, which fed trough the curve by a custom QwtData.

It seems that when i resize my plot it's sluggish. I bring my number of points down to 1000 on each curve and its acceptable but still a bit sluggish.

Now, my question is there a spot on qwtdata where i can i tweak . to note I did not reimplement the boundingRect part of QwtData. Do you guys think this will help if i will reimp this part? How often this is called when i don't do autoscaling.

I really need to go 2000 points on my curve , since i am trying to scale down large sets of data.

I have had some plots where i feed 150 curves and 32000 points each curve. I have connected my data to the curve by QwtPlotCurve::setData (const double *xData, const double *yData, int size) and this is somehow faster than my 4 curve-plot with my custom QwtData.

baray98

[Uwe]

It seems that when i resize my plot it's sluggish. I bring my number of points down to 1000 on each curve and its acceptable but still a bit sluggish.

In SVN trunk you find a refreshtest example, that is intended to play with some settings to check the performance of painting a curve. On my box ( an Atom 330 ) I have more than 150 fps for a curve of 10000 points. When 1000 points make your replots sluggish there is something "suboptimal" in your plot/curve settings.

If the bottleneck is on the painting side I guess you have antialiasing enabled. In combination with a non solid pen you might see a horrible performance.

Now, my question is there a spot on qwtdata where i can i tweak . to note I did not reimplement the boundingRect part of QwtData. Do you guys think this will help if i will reimp this part? How often this is called when i don't do autoscaling.

Never.

... this is somehow faster than my 4 curve-plot with my custom QwtData ...

QwtData::x() and QwtData::y() are called for each point when painting the curve line. The same happens when painting the symbols ( in Qwt 6.x each point is requested at once instead of calling x() + y()).

Without knowing your implementation it is hard to say what could be done faster. But in general you could store the point you have identified for QwtData::x() for the following call of QwtData::y() - or you build an internal cache for your 1000 points, that is recalculated after scale changes only ( not for every resize ).

Uwe

[baray98]

i actually found my bottle neck its in my custom QwtData::size() this method is called everytime .. it happened to be that i always count my buffer (with iterator) for the size. It has improved a lot. the size method is even called even when i am activating my picker ....

thanks uwe