[FFmpeg-devel] ASM support for MACOSX 10.5: really?

[Reimar Döffinger]

On Tue, Mar 25, 2008 at 10:00:49PM +0000, M?ns Rullg?rd wrote:
> Michael Niedermayer <michaelni at gmx.at> writes:
> > On Tue, Mar 25, 2008 at 09:48:46PM +0100, matthieu castet wrote:
> >> Michael Niedermayer wrote:
> >> > On Tue, Mar 25, 2008 at 07:04:10PM +0000, Robert Swain wrote:
> >> > 
> >> > (all with -fPIC -DPIC)
> >> > 2.95        works
> >> > 3.3         works
> >> > 3.4         works
> >> > 4.0         works
> >> > 4.1         works
> >> > 4.2         5 "error: 'asm' operand has impossible constraints" nonsense
> >> > 4.3         3 "error: 'asm' operand has impossible constraints" nonsense
> >> > 4.4         3 "error: 'asm' operand has impossible constraints" nonsense
> >> > 
> >> > so, if someone wants to use latestest and "greatest" gcc go report it.
> >> > 
> >> > Also if someone wants to workaround it in ffmpeg, no problem, the code
> >> > though must not be slower!
> >> > The whole loop of that code should be written in a asm( ) anyway, no
> >> > stincky for(){asm()}.
> >> > 
> >> Even better the code is most assembler and should be easily written in 
> >> pure assembler (*.S). The tricky part could be to access global variable.
> >> That way no dependence on gcc.
> >
> > You can write the whole function in an asm() and use MANGLE to access globals
> 
> True.
> 
> > This would be easier than .S
> 
> I doubt that.

A separate .S file means adding loads of #ifdefs because the functions
headers will look completely different and it will make switching
between inlining/non-inlining a major pain, and certainly not a matter
of choosing the right -O on configuring as it is at least partially
the case currently, as would switching between omitting/not omitting the
frame pointer be a big pain.
So if using a .S file is much harder depends on how much functionality
you are ready to drop...

Greetings,
Reimar D?ffinger