On 2014-02-08 16:12:43 +0100, Diego Biurrun wrote: > On Sat, Feb 08, 2014 at 02:38:53PM +0000, Janne Grunau wrote: > > On 2014-02-06 12:05:16 +0100, Diego Biurrun wrote: > > > On Thu, Feb 06, 2014 at 08:28:44AM +0100, Anton Khirnov wrote: > > > > On Wed, 5 Feb 2014 15:32:59 +0100, Diego Biurrun <[email protected]> > > > > wrote: > > > > > On Wed, Feb 05, 2014 at 09:59:45AM +0100, Anton Khirnov wrote: > > > > > > --- a/libavcodec/ppc/vp8dsp_altivec.c > > > > > > +++ b/libavcodec/ppc/vp8dsp_altivec.c > > > > > > @@ -269,9 +269,44 @@ EPEL_HV(4, 4,6) > > > > > > > > > > > > +static void put_vp8_pixels16_altivec(uint8_t *dst, ptrdiff_t > > > > > > dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) > > > > > > { > > > > > > } > > > > > > > > > > This duplicates the ff_put_pixels16_altivec() function from > > > > > libavcodec/ppc/hpeldsp_altivec.c. > > > > > > > > Yes I know. > > > > Got a better solution? > > > > > > What about replacing the version in hpeldsp_altivec.c with this more > > > flexible version? > > > > Can we please commit this? Complaining over the duplicated ppc/altivec > > version is a little unfair if we already have the same duplication for > > C, x86 and arm. > > Whatever, code duplication seems to be the rage these days ...
it's not a full code duplication, the version in hpeldsp assumes that src stride == dst stride and uses just one stride argument. A function call with one argument less might be a worthwhile optimization if the function which is hte case for put_pixels. It might be worth trying to use macros to expand both functions from the same source lines but I'm not sure if it's worth the trouble. Janne _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
