if anybody can get the best out of Cell its some of the best coders we have in this community who still remember how to get the best out of strange and challenging hardware.
It's not that strange or challenging, to program an SPE it's basically double buffering and AltiVec. Then you just unroll, unroll, unroll and the performance goes up and up. IIRC I got a 20X speed boost that way - it wasn't terribly difficult, apart from the framework to get it working and the double buffering, the main loop was pretty much cut & paste from a Mac.
Here´s what Dean Calver of Ninja Theory says about programming Cell:
"I can’t exactly describe what I’ve been doing, but its extremely low level, what good console programming is all about. So thats what I’ve been doing, munging some data into formats that the different bits of the system can consume and then writing lots of code to manipulate it all as fast as possible. This is where the Cell architecture really starts to pay off, the SPUs are very powerful when programmed at a low level, with an MFC to move data and a decent integer ISA (as well as the float unit that everybody goes on about) they can do this kind of thing with ease. And because they are completely indepedent, you can use them to really leave the PU to other stuff.
Of course the fun bits is that its just like programming an Amiga’s Copper but N times more complex, just like the good ol’ days"
15 crawler(s) on-line.
67 guest(s) on-line.
1 member(s) on-line.
minator 
