[XviD-devel] [RFC] AMD64 support

Edouard Gomez ed.gomez at free.fr
Sun Dec 12 12:48:01 CET 2004


Guillaume POIRIER (guillaume.poirier at etudiant.univ-rennes1.fr) wrote:
> AFAIK, XviD already features different SIMD asm for both Athlon and
> P-IV, so it looks to me that being able to squeeze every CPU cycle using
> "proprietary" SIMD like 3D-Now! is nothing new, isn't it?

Well, looking at the active developers, none has an AMD64 computer.
I don't pan on buying such a box until the architecture becomes
more stable hardware wise, i mean there are already 2 sockets in
a 1 year timelife of the CPU, PCI-express is also something being
in transition... all this is too risky to make a choice now
I'm pretty sure if i buy an AMD64 now, i'll regret my choice.

So, considering no developer has an AMD64, maintaining so many lines
of code w/o corresponding hardware is a lot harder than you think.
SSH access is good for checking you don't do things wrong when merging
someone else's work, but it's not ideal.

Moreover, amd64 ships with sse2, which is emulated with mmx opcodes...
so it's slow. amd64 supports mmxext, so why bothering with mmx (if there
is a xmm version) ? Do you get the dilemna now ? It's not like on ia32 CPUs
where, to get maxpeed you had to support tons of sub-intruction sets, amd64
is sort of "standardized" arch at the moment, we can rely on stable instruction
set at least for this hardware generation.

... until intel decides to enter the market with a mostly compatible CPU
but supporting SSE2 and SSE3 natively, because they are intel's technology.


-- 
Edouard Gomez


More information about the XviD-devel mailing list