[Freeciv-Dev] Re: [Patch] [RFC] Path finding version 14

[Raimar Falke <rf13@xxxxxxxxxxxxxxxxx>]

On Mon, Sep 16, 2002 at 07:18:57PM +0100, Gregory Berkolaiko wrote:
> On Mon, 16 Sep 2002, Raimar Falke wrote:
> 
> > On Mon, Sep 16, 2002 at 04:55:48PM +0100, Gregory Berkolaiko wrote:
> > > On Mon, 16 Sep 2002, Raimar Falke wrote:
> > > 
> > > > On Sun, Sep 15, 2002 at 07:21:58PM +0100, Gregory Berkolaiko wrote:
> > > > > On Sun, 15 Sep 2002, Gregory Berkolaiko wrote:
> > > > > 
> > > > > 
> > > > > And the last version of the patch.  First it makes one test run for 
> > > > > all
> > > > > units, both PF algorithms are used and if a difference in results is
> > > > > detected, the whole thing aborts.  Then 100 runs of each algorithm for
> > > > > each unit are done and the times are printed.
> > > > > 
> > > > > Usage: 
> > > > > 1. Apply the patch to latest CVS.
> > > > > 2. Load the attached game, connect, start and press Turn Done.
> > > > > 3. Watch the output in the server screen.
> > > > 
> > > > Patch from x2 to x3 attached.
> > > 
> > > Patch from x3 to x4 is attached.  Several mistakes fixed.  Added support 
> > > for sea-going (non-sinking) and IGTER units.
> > > 
> > > Savegame runs smoothly, apart from occasional non-unique path with a 
> > > given 
> > > COP hiccup.  
> > 
> > > GB_pf is at least 2 times faster than RF_pf.  Hopefully this settles
> > > the choice of implementation.
> > 
> > Mhhh. This is odd. Both use exactly the same amount of external
> > functions. Lets see if you have implemented all functionality
> > (move_backward, ignore_enemy, goto_move_restriction and
> > is_position_dangerous). This should at least add some ifs in the hot
> > path (or an extra call in the is_position_dangerous case). We should
> > maybe also run some tests if asserts disabled.
> 
> move_backward and goto_move_straightest will be implemented through COSTFN 
> callbacks.  

> I thought we agreed that ignore_enemy is the default behaviour and not 
> ignoring enemy will be done through tile_behaviour.

Ok but then we need this behaviour in form of a defined callback in
path_finding_tools.

> I don't think one call is so important.  In fact I just added an
> extra wrapper on top of pf_next and removed some macro-associated
> cruft from your code.  The results change but not by much, the
> speedup is still greater than 2.  ZOC will bring it down a bit, but
> again not significantly.

So why do you think that you are faster by a factor of two?

> Compiling without asserts will change it much more, in all
> likelyhood.  Can you try it please?

Sorry not much time till I moved.

        Raimar

-- 
 email: rf13@xxxxxxxxxxxxxxxxx
  "With a PC, I always felt limited by the software available.
   On Unix, I am limited by my knowledge."
    -- Peter J. Schoenster <pschon@xxxxxxxxxxxxxxxxx>