[FFmpeg-devel] [PATCH] x86: hevc_mc: remove non necessary moves

[James Almer]

On 03/02/15 4:04 PM, Christophe Gisquet wrote:
> ​​Hi,
> 
> Le 3 févr. 2015 18:47, "James Almer" <jamrial at gmail.com> a écrit :
>>
>> On 02/02/15 2:11 PM, Christophe Gisquet wrote:
>>> @@ -87,11 +95,22 @@ QPEL_TABLE 12, 4, w, sse4
>>>  %elif %1 <= 8
>>>      movdqa            %3,
> [%2]                                              ; load data from source2
>>>  %elif %1 <= 12
>>> +%if avx_enabled
>>
>> If this is meant for avx2, then the correct check is "cpuflag(avx2)" or
> "mmsize > 16",
> 
> I initially chose the former but since Michael's comment, I try avoiding
> anything but strictly necessar​y​. Not sure what to do then? Do another
> "cleanup" commit?

If you really want to keep the functional parts as close as the original as 
possible then just ignore this part. It can be replaced once and if an avx
version is ever added.

> 
>> because if at some point we add an avx version (just to take advantage of
> the
>> non-destructive three operand format compared to sse4), they will have to
> be replaced.
> 
> We​ ​can ​probably bench mark this now.
> 
>> Indentation for all the stuff bellow is wrong.
> 
> Nearly same reason. The original code doesn't check x86_64, which I added.
> I didn't want to change the indentation. But it's rather ridiculous here.

Fixing indentation doesn't change a functional part of the commit, so you can do 
it just fine. It will still be Mickaël's code as he intended it.
You already had to adapt other parts because of the differences between out tree 
and OpenHEVC's after all.

> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>