Author: Robert Hyatt
Date: 15:05:10 04/10/02
Go up one level in this thread
On April 10, 2002 at 16:02:50, Tom Kerrigan wrote: >On April 10, 2002 at 00:43:43, Robert Hyatt wrote: > >>Make/Unmake are the procedures that update the bitboard and hash signature >>stuff as moves are made/unmade. Updating hash signature stuff might be a > >The problem is that "updating" bitboards and hash signatures sounds a lot like >memory and control flow operations instead of logical operations. Making FPGA >logic to do this is ugly, slow, and consequently worthless. I don't disagree... I was talking about "what would be idea to accomplish" to make sure no one starts a project that will pay nothing in return. > >You mention in a previous post that it would be useful to rotate bitboards >quickly. How long does it take to rotate a bitboard with an MPU? I don't rotate on the fly. I maintain rotated versions and update all four in Make/Unmake... That causes significant memory traffic. The rotated bitmaps could be cut out completely were I doing an ASIC design, because it is easy to rotate by routing... Here it will take some thought to figure out if there is a good way to eliminate the large arrays necessary for maintaining the rotated bitmaps, which seems to be a critical issue to solve for current FPGAs... A single PCI >cycle is 30 ns, and the cycle time of a 2GHz chip is 0.5 ns, so unless it takes >much more than 60 instructions to do this rotation, using an FPGA will only slow >your program down. > >-Tom It takes _way_ more than 60 instructions. And several of those are memory reads which adds hundreds of cycles to the mix...
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.