[FFmpeg-devel] [PATCH] x86: hevc: adding transform_add

Ronald S. Bultje rsbultje at gmail.com
Wed Aug 20 02:06:36 CEST 2014


Hi,

On Mon, Aug 18, 2014 at 2:28 PM, James Almer <jamrial at gmail.com> wrote:

> On 18/08/14 5:01 AM, Pierre Edouard Lepere wrote:
> > Hi,
> > here's the new version of the patch. Sorry for the delay.
> > James, I have not done 8-bit AVX versions because it requires unpacks
> that are done differently in AVX.
>
> Aren't you thinking of AVX2 with 256bits wide registers? With AVX i mean
> an xmm version of
> the function that takes advantage of the three-operand instruction format,
> which should be
> essentially the same as the SSE2 version except for the many saved movdqa
> between registers.
>
> Anyway, LGTM and this can be committed as is. There's some of room for
> improvements (AVX
> version, or making all the functions use 8 or less xmm registers) but that
> can be done later
> instead of sending the same patch a fourth time.
>
> Not sure if someone else wants to comment (Ronald?).


Basically agree with what you just wrote. Everything can be improved,
perhaps this also, but this is great as-is.

Ronald


More information about the ffmpeg-devel mailing list