On Samstag 08 August 2009, Andrew Wagner wrote: > Hi guys- > > How many microseconds, ballpark, should I expect to save per kernel by > using prepared kernel calls?
It did make a (small) difference in one of my codes--that's why it got added. I'd say "a couple", but I'd have to measure to give you a precise answer. The prepared call style mainly saves you a Python loop over the arguments and a couple of CUDA setup calls. Andreas
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ PyCUDA mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
