Author: Robert Hyatt
Date: 07:27:15 09/06/01
Go up one level in this thread
On September 05, 2001 at 16:12:52, Francesco Di Tolla wrote: >On September 04, 2001 at 08:32:01, Robert Hyatt wrote: >>Those assumptions are not necessarily valid. IE Cray Blitz used vectors >>quite efficiently. In move generation. In attack detection. In the >>evaluation. It only requires that vectors be included in the design >>process... >> >>If you use them, then a vector machine can scream right along on chess >>problems. We did... > >Vectorial machine usually have separate pipelines for floating point and >integers. Do you mean you used the flaoting point pipeline? > >ciao >Franz Nope. The cray could do either. And the cray is pipelined like no other machine. IE Each _operation_ has a separate pipeline (or more than one in fact). Multiple pipes for scalar operations, for vector operations (whether they are floating point or integer). At execution time, the different pipes dynamically "connect" to each other depending on the order the instructions are issued in, which is why the machine can produce 2 or 4 results every clock cycle, for hours on end. However, we did use some floating point in places. But the reason is not what you would expect. In some cases, floating point math was _faster_ than integer math. We simply did things the fastest way possible. And no, we had _zero_ divides of any kind in the program, floating point or integer.
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.