hi wini ...
well, after my last mail, i was looking at both code and instructions in more details...
it seems that the bottleneck is the memory mapped transfer from pd to the audio device ... i'm more or less clueless how to improve this ... possibly by using movaps or at least movups instructions to reduce the number of memory transfers ... basically the float / int conversion shouldn't be a big problem any more (thanks to the sse)
cheers ... tim