Author: Will Singleton
Date: 08:22:40 10/05/99
Go up one level in this thread
On October 05, 1999 at 10:33:08, William Bryant wrote: >Will, > >There was a nice summary of the AltiVec capabilities in MacTech magazine about 1 >to 2 months ago. It was written by Tom Thompson and probably would have been a >Byte article if that magazine was still in print. I can fax it to you. > I searched the MacTech site and didn't find it, so I'd appreciate the fax. Thanks a lot. (310-278-0710) >Basically, you need to enable the AltiVec instructions in the compiler. >Codewararior 5 supports the AltiVec instruction set and even has support to >emulate AltiVec on older G3 machines (which will be slow) until G4s are >available. > >The internal Bus is 128 bits and it can do a lot of bit mangling as simple >instructions. There is the potential for a lot of complex vector operations as >Bob described if you can a) understand the Altivec instruction set, b) >understand how to add vector operations to create a more complex eval, and >c) take advantage of the very fast internal 128 bit bus. Note, I think it's >still a 32 bit main memory bus at 100 mhz and a 1/2 speed L2 catch at 1 meg. > >BTW, I you do get a G4, allow me to drool a little in your direction. > I'd only get one after the price drops a bit, so perhaps in the Spring. We have an old 6500/250 (real slow 603) that needs replacing. >William >wbryant@ix.netcom.com > > >On October 04, 1999 at 20:03:52, Will Singleton wrote: > >> >>I'm looking at the specs for the 500mhz G4 (available someday), and I'm >>wondering about a couple things. >> >>The AltiVec, or "Velocity Engine", is apparently a vector processing unit for >>which special code must be written to obtain speedups. I wonder if this means >>that the compiler must support those instructions, or can you take advantage of >>vector processing just by rewriting existing code? >> >>Is the vector-processor used mainly in FP operations, or can it be helpful for >>integer-based code? >> >>The specs say that it has data stream prefetching ops supporting 4 simultaneous >>32-bit data streams, as well as a new fpu supporting single-cycle, >>double-precision calcs. Are both of these associated with the vector processing >>unit? >> >>In short, can a chess program take advantage of vector processing without a >>massive rewrite? >> >>Will
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.