On Tue, 2009-02-24 at 01:05 +0000, Kai-Martin Knaak wrote: > On Mon, 23 Feb 2009 20:51:31 +0000, Peter Clifton wrote: > > > Long story short.. send slow examples, and they might well get profiled > > and used to help optimise PCB's speed for that kind of case. > > The layout I used to benchmark the GL-enabled pcb version can be > retrieved from > http://lilalaser.de/tmp/lasertreiber.pcb
I've had a quick poke at it. I'm getting about 5fps full screen (zoomed out), with polygon fill. If I turn off all layers but silk, including pads and vias, I'm not getting a huge frame-rate increase, and it would appear that the large number of triangles created by the end-caps of each line making up a piece of text is the killer here. This might somewhat explain why you're not seeing vast improvements in rendering speed when you zoom in. For circular objects (arcs, vias, pins, line-end caps), the number of triangles used increases with the size of the primitive on-screen. As you zoom, there might be less text on-screen, but there is more complexity used in rendering those which are. Once I've deleted text from your silk-screen, I stop being CPU bound, and end up with my profile hot-spots of about 30% copying data to the GPU. (Need to revive my code to use mapped VBOs to try and reduce that overhead). It is possible that doing this would also improve performance in the case which has lots of text on-screen, although it would appear that the graphics card is just not dealing with the triangles particularly fast. I'm getting about 24fps with your silk-screen stripped of (almost) all text. I've love to have found some magic element in the design which was the root-cause of the slowness by some bug which caused it to use far too many triangles, but unfortunately I've not found anything like that. Perhaps I need to start thinking about how the text glyphs could be pre-tessellated into triangles (or textures) and cached. However.. I think that is too much complexity for the moment. Whilst it isn't quite video frame-rate, the speed achieved (on my box anyway) seems quite usable as far as I can tell. You forgot the attachment of glxinfo output when you replied to my original survey on what GL capabilities people have. Both your glxgears benchmarks are faster than mine, so it is interesting to note your PCB frame-rates are lower. (Perhaps the stencil hardware is different). I guess when the Xorg guys keep saying "glxgears is not a benchmark", it is with good reason. If you could send me your glxinfo output, that would be interesting. Best wishes, -- Peter Clifton Electrical Engineering Division, Engineering Department, University of Cambridge, 9, JJ Thomson Avenue, Cambridge CB3 0FA Tel: +44 (0)7729 980173 - (No signal in the lab!) _______________________________________________ geda-user mailing list [email protected] http://www.seul.org/cgi-bin/mailman/listinfo/geda-user
