[FFmpeg-devel] [PATCH 2/3] x86/hevcdsp: add ff_hevc_sao_edge_filter_8_{ssse3, avx2}

[Hendrik Leppkes]

On Thu, Feb 5, 2015 at 7:07 PM, James Almer <jamrial at gmail.com> wrote:
>
> On 05/02/15 12:49 PM, Christophe Gisquet wrote:
> > Hi,
> >
> > 2015-02-05 5:18 GMT+01:00 James Almer <jamrial at gmail.com>:
> >> Original x86 intrinsics code and initial yasm port by Pierre-Edouard Lepere.
> >> Refactoring and optimizations by James Almer.
> >
> > No further comment from me.
>
> Pushed, thanks.
>

I looked into the MSVC 64bit failure from this patch, and from what I
can tell doing this doesn't work:
movsx      a_strideq, byte [pb_eo+eoq*4+1]

I'm not entirely sure on the specifics why it breaks however..
But all I could find suggests that you should load the table address
into a reg first, and then use that reg for the address computation.

For some reason, the non-WIN64 version does just this, but the WIN64
version does not.
Any particular reason for this difference?

- Hendrik