[gnugo-devel] Optimization matters!

[bump]

I wrote:

> I'm inclined to use the patch semeai_3_17.2a even though there
> is a net loss of four regressions.

This patch was delayed going into the CVS because I could not
figure out why GNU Go was running so slow. I got times of
13000+ seconds both with and without the semeai patch, and could
not see any recent patches that could account for this. A
proper time for all_batches is about 8100 seconds.

It turns out that to get debug info for arend_3_17.1,
which caused crashes on my machine but not Arend's, I
removed the -O2 flag in engine/Makefile, so reading.c
(as well as semeai.c and maybe other files) was compiled
without optimization.

Naturally you would expect the engine to run slower but
such a slowdown is amazing.

Dan