Re: Ann: SWI-Prolog 7.1.9

Hi Jan,

Thanks for the very constructive answer, it really helped and
eye with --plugin is now loading at a speed of 125 Ktripes/sec
which is 3 times faster than yesterday :-)

> Hi Jos,
> 
> On 02/27/2014 12:38 AM, Jos De Roo wrote:
>> Hi Jan,
>>
>> With the attached file graph.n3p we get
>> $ swipl -f graph.n3p -g halt
>> % graph.n3p compiled 2.21 sec, 100,103 clauses
>>
>> Compared to version 6.6.1 this is a speed drop of 38%
>> $ swipl -f graph.n3p -g halt
>> % graph.n3p compiled 1.60 sec, 100,103 clauses
> 
> The original file didn't make it to the list because of
> the long attachment, but that doesn't matter too much.
> 
> Yes, compilation in 7.x is slower.  This is because there
> is more support for term/goal expansion/4, which includes
> translation the source location information.  The second
> reason is that the functional notation support requires
> analyzing the source in more depth.
> 
> If you want best speed for loading big Prolog data files,
> load them using a read/assert loop and finally call
> compile_predicates/1 to make the code static.  So,
> 
> ==
> load_file(File, Module) :-
>                setup_call_cleanup(
>                    open(File, read, In),
>                    read_file(In, Module),
>                    close(In)),
>                findall(Module:Name/Arity,
>                                ( predicate_property(Module:Head, 
dynamic),
>                                  functor(Head, Name, Arity)
>                                ), Preds),
>                compile_predicates(Preds).
> 
> read_file(In, Module) :-
>                read_term(In, T0, []),
>                read_file(T0, In, Module).
> 
> read_file(end_of_file, _, _) :- !.
> read_file(Term, In, Module) :-
>                assert_clause(Term, Module),
>                read_term(In, T2, []),
>                read_file(T2, In, Module).
> 
> assert_clause((:-_), _) :- !.
> assert_clause(Term, Module) :-
>                assertz(Module:Term).
> ==
> 
> 1 ?- time(load_file('graph.n3p', test)).
> % 401,925 inferences, 0.449 CPU in 0.449 seconds (100% CPU, 895476 Lips)
> 
> (vs. 1.5. seconds for consult).  The time is now 92% read_term and 7%
> assertz and 1% for the loop above.
> 
> And of course, there is the RDF database :-)  Did you compare the
> two in terms of loading time, storage space and query time?

No and I am feeling sorry about that ..

The thing is that we typically use graph literals (N3 formulae)
esp. in RESTdesc http://restdesc.org/about/concept proofs like
http://eulersharp.sourceforge.net/2007/07test/resto-proof.n3

Just to get an idea, loading the 100000 triples from graph.n3p
in EYE takes 0.809 sec and requires 500297 VM-codes.

>                Cheers --- Jan

Kind Regards,

Jos De Roo | Agfa HealthCare
Senior Researcher | HE/Advanced Clinical Applications Research
T  +32 3444 7618 | F  +32 3 444 8211 | M  +32 494 56 03 95

http://www.agfahealthcare.com
http://blog.agfahealthcare.com
R.O.: Septestraat 27, B-2640 Mortsel, Belgium | RLE Antwerp | VAT BE 
0403.003.524 | IBAN Operational Account BE81363012356224 | IBAN Customer 
Account BE20375104592856 | ING Belgium NV, B-1000 Brussels
Click on link to read important disclaimer: 
http://www.agfahealthcare.com/maildisclaimer 
-------------- next part --------------
HTML attachment scrubbed and removed