[Freeciv-Dev] Re: (PR#4381) some clean with get_map_tile(...) , get_map_terrain(...), get_map_special(...) ...

["Per I. Mathisen" <per@xxxxxxxxxxx>]

On Fri, 13 Jun 2003, Rafa³ Bursig wrote:
> Could you test this attached patch on your Sun. I fix all bugs and it
> should patch and work fine.

Autogame behaviour was different between patched and unpatched, so
somewhere you changed code that was not merely an optimization. This makes
the tests below almost worthless except for with or without #define TEST.

I did two tests each (first no savegames, second one savegame), with -O3
and -DDEBUG.

Original cvs:
user    0m31.220s
user    0m31.270s

With patch:
user    0m32.230s
user    0m32.270s

With patch + #define TEST:
user    0m32.370s
user    0m32.480s

Then ran the same tests with -O2 and -DNDEBUG:

Original cvs:
user    0m18.160s
user    0m18.140s

With patch:
user    0m19.200s
user    0m19.050s

With patch + #define TEST:
user    0m18.940s
user    0m18.960s

Then -O3 and -DNDEBUG (no savegames):

Original cvs:
user    0m17.500s
user    0m17.410s

With patch:
user    0m18.550s
user    0m18.520s

With patch + #define TEST:
user    0m18.580s
user    0m18.700s

I estimate the error margin to be around +/- 0.2 due to other processes.

Conclusions: -DDEBUG is slow :-) Trying to out-smart gcc in -O3 mode is
usually a waste of time. Making map_get_tile() into a macro may not give
that much increase in speed after all.

> Other problem that I found is in ai/aiair.c ( in
> find_something_to_bomb(...) function )
> line 180 and 186. What x,y cordinates shoud there be : x,y or x1,y1.
> (in map_get_known(...) and map_get_known_and_seen(...) )

Should be x1,y1, AFAIK. Looks like a bug. Not that it will ever trigger,
since the handicap is only present for non-hard AI, and that AI doesn't
use air units... Greg?

  - Per