[XviD-devel] [RFC] memcpy thingy

[Michael Militzer]

Hi all,

Quoting Christoph Lampert <chl at math.uni-bonn.de>:

> On Fri, 31 Oct 2003, Edouard Gomez wrote:
> > Or may I ask it this way:
> > "Is every possible improvement commitable ?"
> 
> I'd say: "If it's just replacing a few calls to routines by 
> their optimized versions and function pointers, then yes."

I'd say no. It's not needed to have special AMD optimized memcopy code that
gives just a very minor improvement. It seems more reasonable to optimize
such code for more platforms (so generic mmx code) and instead of a simple
memcpy() replacement there should be a specific mmx replacement for memcpy
using code - this should be even faster. BTW: I'd say that at least the
Intel compiler (maybe gcc also) automatically replaces memcpy calls with
specifically optimized code (depending on the selected platform).

Also I can mainly only think of two code parts where memcpy() /
memset()optimizations might help: setedges and in the decoder part all these
memset() calls to clear out the DCT coeff variables. This really slows
things down a bit but it's a bigger problem on the de- than on the encoder
side. Therefore I don't believe in 10% faster encoding thanks to AMD memcpy,
but some mem[copy,set] optimizations might be helpful, especially for the
decoder.

Yet another note: The fabulous Skal already posted a mmx'ed setedges version
long time ago, if you search the list you should find it - so no need for
a AMD optimized memcpy, rather Skal's code should be integrated...

bye,
Michael