Author: h.g.muller
Date: 00:41:41 02/24/06
Go up one level in this thread
Thank for pointing me to the dependency-chain issue! I had not realized that there also was a dependency on the operand that is not picked, and this saves me finding it out the hard way: there are two basic choices in my combined loop, for the two loop-counters, (incrementing/reinitializing, or freezing/incrementing) that depend on the same condition (namely if the 'inner' loop terminated). In the design I posted these were taken in series, adding their dependency changes, while the might have been taken in parallel (since they in the end depend on the same condition). This reduces the depedency chain spanning the iteration from 8 to 5, and the number of operations is such that especially an Athlon should have no problems executing them in 5 CPU clocks (even with pawn double-move test). We'll see...
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.