Hi, This addresses roundup issue #301 : https://roundup.mplayerhq.hu/roundup/ffmpeg/issue301 Summary of changes: - Use of MANGLE when loading into mmx registers some constants - Conversion to non-static for those constants and thus renaming with ff_prefix - Removal of last parameter of MSPEL_FILTER13_CORE (was constant) - Use of "+r" instead of stricter but unnecessary "+g" - Use of REG_c and direct loading of some of the above (not tested No speed regression is noticeable with the new code, but the performance vastly varies depending on the compiler. For a 720p test sequence: 2.95 ~ 4.1s 3.3.6 ~ 3.8s (3.2 in 64b) 4.2.1 ~ 3.4s (3.1 in 64b) Considering those numbers, I think I'll pass next time. (And yeah! for hard-coded limitations) Best regards, -- Christophe GISQUET
Christophe GISQUET a ?crit :
- Use of REG_c and direct loading of some of the above (not tested
Old patch was attached, newer _tested_ with a 64b OS in this mail.
On Fri, Dec 28, 2007 at 07:25:04PM +0100, Christophe GISQUET wrote:
Christophe GISQUET a ?crit :
- Use of REG_c and direct loading of some of the above (not tested
Old patch was attached, newer _tested_ with a 64b OS in this mail.
patch looks ok [...] -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB Opposition brings concord. Out of discord comes the fairest harmony. -- Heraclitus
On Fri, Dec 28, 2007 at 07:48:01PM +0100, Michael Niedermayer wrote:
On Fri, Dec 28, 2007 at 07:25:04PM +0100, Christophe GISQUET wrote:
Christophe GISQUET a ?crit :
- Use of REG_c and direct loading of some of the above (not tested
Old patch was attached, newer _tested_ with a 64b OS in this mail.
patch looks ok
Applied. Diego
participants (3)
-
christophe.gisquet@free.fr -
diego@biurrun.de -
michaelni@gmx.at