[gst-devel] Configuration (fwd)

[Wim Taymans]

Erik Walthinsen wrote:

> > We need to clean this up. I would prefer one binary with MMX support
> > activated when the processor is capable. This is always good from the
> > distributers perspective.
> Agreed.  Check the design of getbits.c.

Almost, it's activated at compile time. If you compile this on an MMX
capable compiler, it will crash on a regular pentium.
> 
> > I checked in a mmx.h, removing the dependency for libmmx. I think we
> > should rename HAVE_LIBMMX to something like USE_MMX or so.
> With mmx.h in include/, yes, we should remove that.  Not sure if USE_MMX
> makes sense, if we're going to build a binary with it anyway.  I also
> can't see anyone trying to build this on an x86 machine with a compiler
> that didn't understand MMX (besides, I think mmx.h uses opcodes instead
> of inline-asm).
> 
I'll add a check in configure to detect if the compiler can compile MMX.

> > Maybe we should set a global flag (or have a method) to query the
> > MMX capabilities of the CPU. All codecs could then set the pointers
> > to their different functions as needed.
> Yup.  Wonder how that could be done generically, maybe part of gstutil.c
> with a global _gst_processor_caps struct and query functions?
That sounds good.

Wim

-- 
No line available at 300 baud.