Speeding up i18n

[Matthias Kretz]

On Tuesday 03 February 2009 09:35:46 Olivier Goffart wrote:
> Le mardi 3 février 2009, John Tapsell a écrit :
> > Hi all,
> >
> >   Has anyone looked into speeding up i18n calls by generating the
> > string hash at compile time rather than at runtime.   Presumably it
> > could be done by a little bit of template magic in the i18n call.
> >
> >   It would be interesting to investigate its effect on compile times
> > and application load times.
>
> I already tried to do operations on strings at compile time but without
> success.
> Using some template, and good optimisation flag, the compiler can optimize
> the hash computation of small strings, but once the string has a reasonable
> size, its not optimized.
>
> My objective was to keep strings simple to type., ie.  STR("Some String") 
> and not
> STR<'S', STR<'o', STR<'m', STR<'e' STR<' ', STR<'S', STR<'t', STR<'r',
> STR<'i', STR<'n', STR<'g'> > > > > > > > > > > >

AFAICS strings are immune to template and macro magic. The problem is that a 
string in C/C++ always needs to point to memory where the string is. And 
constant expressions refuse to work with anything that's a pointer.

The only reliable solution I can think of is to have the hashes computed by 
something other than the compiler.

John, did you profile whether precomputing the hashes might actually do a 
speedup?
Also keep in mind that modern computers often favor computation over memory 
access by such a big factor that lookup-tables and other tricks that were very 
useful some time ago can actually slow things down today.
So I recommend to not look into a solution until we have some idea of whether 
the investment is worth the trouble.


-- 
________________________________________________________
Matthias Kretz (Germany)                            <><
http://Vir.homelinux.org/