API for optimized predicates (was Re: [postgis-devel] 1.3.3)

[Martin Davis]

Nice...

5x speed is good to have!.  From this test it doesn't seem like there's 
too much penalty for using the simple 2-param  MEMCMP version.

I wonder if there is a situation where the IDCMP pulls significantly 
ahead of MEMCMP - say for huge input geometries?  But it could well be 
that in all realistic situations the time is dominated by the time to 
actually compute the PIP result anyway, so you would never notice that 
much of a difference.

If it was my call I would just provide the 2-param MEMCMP version, which 
has little or no requirement for new documentation and low potential for 
confusion.   The 3-param version could be documented somewhere as "could 
be provided if we really need it".

Paul Ramsey wrote:
> MEMCMP:
>
> select e.id, count(*) from ed2000 e, vada2005 v where
> st_contains(e.the_geom, v.centroid, NULL) group by e.id;
>
> Time: 6518.143 ms
>
> UNPREPARED:
>
> select e.id, count(*) from ed2000 e, vada2005 v where
> st_contains(e.the_geom, v.centroid) group by e.id;
>
> Time: 34362.747 ms
>
> IDCMP:
>
> select e.id, count(*) from ed2000 e, vada2005 v where
> st_contains(e.the_geom, v.centroid, e.gid) group by e.id;
>
> Time: 6204.834 ms
>
> DATA:
>
> # select count(*) from vada2005;
>  count
> -------
>   8792
>
> # select count(*) from ed2000;
>  count
> -------
>     79
>
> # select 100*h, count(*) from (select
> npoints(geometryn(the_geom,1))/100 as h from ed2000) as foo group by
> h;
>  ?column? | count
> ----------+-------
>       100 |    25
>       200 |    11
>       300 |     6
>       400 |     1
>       500 |     2
>       600 |     4
>       700 |     1
>       800 |     2
>       900 |     1
>      1200 |     2
>      1300 |     1
>      1400 |     2
>      1500 |     2
>      1600 |     1
>      1700 |     1
>      2200 |     2
>      2300 |     2
>      2500 |     1
>      2700 |     1
>      2800 |     2
>      3100 |     1
>      3600 |     1
>      3700 |     1
>      3900 |     1
>      4200 |     1
>      5000 |     1
>      5600 |     1
>      6400 |     1
>     10600 |     1
>
> All polygons have > 100 points, about 30% have more than 1000.
> _______________________________________________
> postgis-devel mailing list
> postgis-devel at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-devel
>
>   

-- 
Martin Davis
Senior Technical Architect
Refractions Research, Inc.
(250) 383-3022