[cairo] Speeding up _cairo_fixed_from_double

[Behdad Esfahbod]

On Sun, 2006-10-29 at 15:56 -0500, Jim Gettys wrote:
> This is at compile time, not at run time.
> 
> Any compiler worth it's salt these days does dead code analysis, and
> drops the code that can never be executed.
> 
> In the sample here, one branch or the other is always dead code, and the
> optimizer should be deleting it.

Right.  But it's compile-time just because of the optimization.  Maybe
what I really mean is that it's compile-time, not preprocessor-time.  So
you cannot define a NATIVE_TO_NET_ORDER(x) macro using it easily.

behdad

>                                    - Jim
> 
> 
> On Sun, 2006-10-29 at 15:48 -0500, Behdad Esfahbod wrote:
> > On Sun, 2006-10-29 at 14:04 -0500, Jim Gettys wrote:
> > > Evil endianness trick from Xlib:
> > > 
> > >         int endian = 1; 
> > > 
> > >         if (*(char *) &endian)
> > >             client.byteOrder = '\154'; /* 'l' */
> > >         else
> > >             client.byteOrder = '\102'; /* 'B' */
> > > 
> > > Even more fun about this, is that the compiler ends up optimizing
> > > exactly what you'd like; everything is determinable at compile time, so
> > > only the right assignment (or branch set of code) is emitted in the
> > > generated code.
> > > 
> > > I'll take this over the typical if-def's configure options, and other
> > > horrors any day.
> > > 
> > > Works fine when cross compiling too.
> > 
> > Right.  This is basically what the configure-time tests do.  The problem
> > with doing it runtime is that we sometimes need to define a macro
> > differently based on the endianness and that's not really easy to
> > implement using this snippet.  It works for the fixed_from_double case
> > at hand though.  Something no one mentioned in this thread is that cairo
> > already has endian-specific code.  So, no new non-portability is being
> > introduced.
> > 
> > 
> > behdad
> > 
> > >                              Regards,
> > >                                       - Jim
> > > 
> > > 
> > > If you don't succeed, try try again...  Was posting from an old address,
> > > got moderated to death, fixed the address, forgot to wait for the
> > > confirm, posted again, got moderated to death....
> > > 
> > > I guess no one deals with posting to the cairo list held for
> > > moderation....
> > > 
> > > 
> > > On Sun, 2006-10-29 at 13:00 -0500, Behdad Esfahbod wrote:
> > > > On Sun, 2006-10-29 at 10:21 -0500, Michael Sweet wrote:
> > > > > Jim Gettys wrote:
> > > > > > Evil endianness trick from Xlib:
> > > > >  > ...
> > > > > 
> > > > > I like it! :)
> > > > 
> > > > Care to share?  I don't think Jim's mail made it to the list.
> > > > 
-- 
behdad
http://behdad.org/

"Commandment Three says Do Not Kill, Amendment Two says Blood Will Spill"
        -- Dan Bern, "New American Language"