[x265] [PATCH] analysis: fix for rd-0 non-deterministic output

Steve Borho steve at borho.org
Mon Aug 3 17:29:30 CEST 2015


On 07/28, ashok at multicorewareinc.com wrote:
> # HG changeset patch
> # User Ashok Kumar Mishra<ashok at multicorewareinc.com>
> # Date 1438080030 -19800
> #      Tue Jul 28 16:10:30 2015 +0530
> # Node ID dbe8c629ccc726ece98ffbe4af5c78f8945e84a5
> # Parent  e08a245054434090ccb1fc1b985f955a66711157
> analysis: fix for rd-0 non-deterministic output
> 
> diff -r e08a24505443 -r dbe8c629ccc7 source/encoder/analysis.cpp
> --- a/source/encoder/analysis.cpp	Mon Jul 27 15:15:35 2015 +0530
> +++ b/source/encoder/analysis.cpp	Tue Jul 28 16:10:30 2015 +0530
> @@ -1106,7 +1106,8 @@
>      /* Copy best data to encData CTU and recon */
>      X265_CHECK(md.bestMode->ok(), "best mode is not ok");
>      md.bestMode->cu.copyToPic(depth);
> -    md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic, cuAddr, cuGeom.absPartIdx);
> +    if (m_param->rdLevel)
> +        md.bestMode->reconYuv.copyToPicYuv(*m_frame->m_reconPic, cuAddr, cuGeom.absPartIdx);

this looks like an appropriate fix, since at rd-0 we do not generate
recon pixels in the analysis pass, but it also tell me that some piece
of code is incorrectly *using* the recon pixels before they are valid.
This should also be fixed.

-- 
Steve Borho


More information about the x265-devel mailing list