yade-dev team mailing list archive

[Vaclav SMILAUER <eudoxos@xxxxxxxx>]

We will have to do some benchmarks on rendering, but my video card is 
fairly crappy onboard i945 (DRI works, acceleration also), hence my case 
may not be relevant. For synchronized display, I frequently got crashes 
so I don't use it.

Since there were no changes introduced to drawing during the transition 
to official wm3, I would think there might be other things that caused it.

For wm3, I found out that it doesn't make use of SEE2 vector 
instructions (which may give easily 50% sppedup for vector calculations) 
- the only library that does that is atlas (tuned specifically for 
different instruction sets - like package atlas-sse2), but it is fairly 
low-level (blas). If there were a library that does link to atlas and 
has nice c++ interface, I would perhaps use it instead of wm3? Don't 
know what the speedup would be. Before you ask - gcc will not vectorize 
the code during optimizations, at least not the one I have here (checked 
with objdump).

Bruno: for IRC, I will pass by.

Vaclav
_______________________________________________
yade-dev mailing list
yade-dev@xxxxxxxxxxxxxxxx
https://lists.berlios.de/mailman/listinfo/yade-dev