Re: [Bug-gnubg] Attempt to speed up single threaded use in multi threaded build

[Michael Petch]

I have my tests running for the next few hours. But here is one thing I can say (preliminary data) .I have an 800 move match being analysed 10 times under each scenario (and the results being md5 checksummed) . New and old builds have SSE2 on fully optimized with –O3, GUI Build running with –t option to suppress the graphics. . I’m using 32 Bit Debian Lenny Stable on a dual quad core system. With the multithreaded new and old builds) the new code in CVS is ~2.5% faster when Threads=Cores vs old code. I’m using a 168MB cache and I set the priority to –19 (Highest priority).

I’ll post my full results when complete, but thought I’d point this out now. As well the md5 checksums of the files generated on each version so far all match (no anomalies)

My old release is an Aug23rd CVS checkout vs your new Aug30th CVS changes.

On 30/08/09 3:07 PM, "Jonathan Kinsey" <address@hidden> wrote:

> old multi threaded build with threads set to number of cores.
> new multi threaded build with threads set to number of cores.