[Ffmpeg-devel] dsputil h264 ppc patch

[Luca Barbato]

Likai Liu wrote:
> On Mac OS X 10.3, a type error in
> libavcodec/ppc/dsputil_h264_template_altivec.c (line 429) prevents me
> from compiling the file. It is patched below. The rationale of this fix
> is that v16ui looks like it should be unsigned int, so it should be
> constructed using vec_splat_u32 and not vec_splat_u16 (which is unsigned
> short).

you are right, looks like it slipped, 10l of cola, please, I wonder why
it is working even if with a wrong value...

> By the way, why don't we simply say vec_splat_u32(16), but have to use
> vec_sl(vec_splat_u32(1), vec_splat_u32(4)), effectively saying 1 << 4
> for all values in the vector? It doesn't make the code easier to read.
> Same question for v512si.

the vec_splat_u/s won't use any input register and that makes them fast
but at the same time they have a limited range of values ^^;

lu

-- 

Luca Barbato

Gentoo/linux Developer		Gentoo/PPC Operational Leader
http://dev.gentoo.org/~lu_zero