[Ffmpeg-devel] [RFC] LZO optimization, howto detect builtin memcpy?

Michael Niedermayer michaelni
Wed Jan 31 01:13:53 CET 2007


Hi

On Tue, Jan 30, 2007 at 07:36:46PM +0100, Reimar D?ffinger wrote:
> Hello,
> attached patch provide some LZO decoding speeding if either the compiler
> has a builtin memcpy for fixed-size copies or the architecture can do
> unaligned load/stores.
> Problem is, I don't know how to properly detect either one.
> Any ideas? Or should I just commit attached patch which assumes that a
> memcpy builtin is used? Having a UNALIGNED_LOADSTORE or similar define
> available would be nice for the stuff in intreadwrite.h, too, I guess
> (though there you can at least use the unaligned_16 for gnu compilers,
> that would slow things down in this case if unaligned loads/stores are
> not supported though).

assuming that memcpy(a,b,2 or 4) is inlined is reasonable, so always using
that seems like the obvious solution, alternativly you could

add AV_CP16/32 to intreadwrite.h
add a check to configure which tests if unaligned stored and unaligned loads
are fast
and depending on the result make AV_CP* either use unaligned_* or bytewise

btw do you have a file (url on mphq?) which uses lzo for testing ... ?

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

The misfortune of the wise is better than the prosperity of the fool.
-- Epicurus
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070131/4977ef95/attachment.pgp>



More information about the ffmpeg-devel mailing list