On Mon, 2005-05-16 at 15:02 +1000, Nigel Pearson wrote:
> Problems:
> 
> 1) A lot of the filters have common stuff that should be abstracted
>     (e.g. mm_support() or emms() for non-Intel compilation,
>           content of .pro files are nearly identical)
> 2) Several dummy functions are needlessly defined for non-MMX
> 3) Altivec accelleration is not determined at run time
> 4) Using i386 instead of MMX.
1/3) Why not have mm_support() return MM_ALTIVEC if altivec is available?
This is already implemented in libs/libavcodec/ppc/dsputil_ppc.c
2) ok, get rid of these.
4) Can't do it. i386 != MMX. MMX is defined when the processor is any
MMX supporting processor, while i386 is defined when the processor is a
IA32 processor supporting MMX. The problem is that the instruction
mnemonics have different meanings on a x86-64, popl still pops a long
off the stack, but now a long is 64 bits not 32 bits, so it won't fit in
eax... Anyway, some algorithms have been ported to work on a x86-64, and
others haven't so we need the two defines.

>       Better names (or other criticism) gladly accepted/debated,
> otherwise I will commit in a few days?
Thanks for not committing right away, I already did the 
i386 -> MMX change a few weeks ago, and had to revert...

-- Daniel

_______________________________________________
mythtv-dev mailing list
[email protected]
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-dev

Reply via email to