[MPlayer-dev-eng] [PATCH] support for libXvMCW wrapper library

[Ivan Kalvachev]

2006/1/16, Dominik 'Rathann' Mierzejewski <dominik at rangers.eu.org>:
> On Sunday, 15 January 2006 at 23:09, Ivan Kalvachev wrote:
> > 2006/1/15, Dominik 'Rathann' Mierzejewski <dominik at rangers.eu.org>:
> > > On Monday, 12 September 2005 at 01:14, Dominik 'Rathann' Mierzejewski wrote:
> > > > On Monday, 12 September 2005 at 00:56, Guillaume POIRIER wrote:
> > > > > Hi,
> > > > >
> > > > > On 8/21/05, Wes Shull <wes.shull at gmail.com> wrote:
> > > > > > libXvMCW is a wrapper which dlopens a proper libXvMC at runtime (which
> > > > > > one it loads is configurable); it eliminates the need to link directly
> > > > > > against a hardware-specific xvmc lib to support it.
> > > > > >
> > > > > > The library originated in the Unichrome driver project (
> > > > > > http://unichrome.sourceforge.net/ ) but is useful for nvidia, etc.
> > > > > > too.  It's in ATrpms, probably other places too.
> > > > > >
> > > > > > Attached is a pretty simple mostly cut&paste patch to enable its use
> > > > > > with mplayer.  It adds an --enable-xvmcw (and --disable-xvmcw) option,
> > > > > > which must be used in conjunction with --enable-xvmc.
> > > > > >
> > > > > > The patch is directly against configure, from today's (anon) cvs,
> > > > > > which feels very wrong to me--it should be against configure.in.
> > > > > > Except I can't find that.
> > > > > >
> > > > > > This is my first patch submission to mplayer.  Flame away ;-)
> > > > >
> > > > > Do you have a new patch available that features Dominik's suggestions?
> > > >
> > > > And Ivan's.
> > >
> > > As it turns out, there's no need for your patch, Wes. All the support is
> > > already in place, we just need to make it autodetected and set the
> > > defaults, i.e. like the attached patch.
> > >
> > > I've tested it as far as the compilation goes, because I have no
> > > XvMC-capable cards at home, but it should work, as long as XvMCW works.
> > >
> > > Regards,
> > > R.
> >
> > I'm against enabling xvmc by default. it is not generic usable yet. It
> > also slows down (by a tinny bit) the ffmpeg12 decoder.
> >
> > Also, when linking statically we get benefit and shared memory
> > subpicture transfer.  As NVidia library is never installed as part of
> > Xorg it is quite safe to assume we want it. Same applies for some of
> > the other.
> >
> > Please make it as hinted above, (with <for> loop and test for libraries).
>
> <for> loop makes no sense, as it's common to have more than one of these
> libs installed, so which are we to choose? And no, I'm not doing hardware
> detection in ./configure.
>
> IMHO the sane solution is the one I've proposed, with XvMCW getting
> detected by default (if present) while still allowing to link directly
> with a chosen library using --with-xvmclib=XvMCNVIDIA for example.

Read again my previous mail.
XvMCNVidia is never installed if there is no nvidia hardware. Same
applies to few others drivers. Only i810/5 comes with X. It is safe to
assume that user have installed them because he have the hardware.

Anyway, I agree that XvMCW is safer default, but what I have in mind
is something like:
-_xvmclib="XvMCNVIDIA"
@
+for _temp in $_xvmclib XvMCNVIDIA XvMCW ; do
@
+done