[MPlayer-dev-eng] Non-inline asm code

[Alex Beregszaszi]

Hi,

On Sat, Dec 22, 2001 at 08:23:19PM +0100, G?bor L?n?rt wrote:
> On Sat, Dec 22, 2001 at 07:32:02PM +0200, Arpi wrote:
> > > > It's no big problem - I'll do it soon (during jan-feb-2002).
> > > > I don't like function inlining and prefer call them through pointer
> > > > (for runtime support of each x86 clone - due unnecesary cache pollution).
> > > 
> > > Agreed. However it WOULD be possible to use "plugins" or whatever you wanted
> > > to call ... We would be able to produce .a files for every CPUs, maybe
> > > separated demuxer, native codecs and so on. This technique is used by XFree
> > > for example: it CAN load .a files runtime! This is quite good thing, since
> > did you already implemented .a loader/linker? :)
> 
> No. But it seems XFree86 does ;-) Just see modules of XFree. AFAIK it was
> donated by MetroXY (well I don't remeber their name ...) to XFree project.

MetroLink, they're writing AcceleratedX or something like that. An another
X11R6 compat. commercial X server.

>  
> > > we don't need use dl_open() and other functions (refeering functions with
> > > pointers, slowing down because of used register for GOT in shared libraries).
> > > It would be simply as fast as if mplayer would be statically linked together.
> > > Nice, isn't it? 
> > > 
> > 
> > .so does the same if compiled without -fPIC option. it's nice feature of
> > linux, and a few other system (-fPIC means pointers for everythink, but
> > without fPIC it will be relocated runtime so no pointers needed)
> 
> IMHO, pointers ARE needed anyway! This is a must because of you won't know
> DATA offsets used in a shared library (data segment was created by the
> application ...).
> 
> - Gabor
> _______________________________________________
> MPlayer-dev-eng mailing list
> MPlayer-dev-eng at mplayerhq.hu
> http://mplayerhq.hu/mailman/listinfo/mplayer-dev-eng

--
alex