[x265] [PATCH 7 of 7] entropy: after encodeCU, the CU structures need to be reset with the right QP

[Steve Borho]

On 04/18, deepthi at multicorewareinc.com wrote:
> # HG changeset patch
> # User Deepthi Nandakumar <deepthi at multicorewareinc.com>
> # Date 1429334781 -19800
> #      Sat Apr 18 10:56:21 2015 +0530
> # Node ID 313a17ef22f4b82e21098ce9f784d2c6d1e52163
> # Parent  d138bbee3b0155e1433be7b8360afab9d5ccdaf0
> entropy: after encodeCU, the CU structures need to be reset with the right QP
> 
> This is used by the deblocking filter to index into the the beta and tc tables.
> Without deblocking, no-one cares about the QP structures of CU after encode,
> so it wouldnt cause any problems.

series queued for testing with some re-ordering of the patches to ensure
deblocking is only broken on one commit, and a couple of coding style
nits fixed.

> diff -r d138bbee3b01 -r 313a17ef22f4 source/encoder/entropy.cpp
> --- a/source/encoder/entropy.cpp	Thu Apr 09 00:48:54 2015 -0400
> +++ b/source/encoder/entropy.cpp	Sat Apr 18 10:56:21 2015 +0530
> @@ -585,7 +585,7 @@
>          if (ctu.isSkipped(absPartIdx))
>          {
>              codeMergeIndex(ctu, absPartIdx);
> -            finishCU(ctu, absPartIdx, depth);
> +            finishCU(ctu, absPartIdx, depth, bEncodeDQP);
>              return;
>          }
>          codePredMode(ctu.m_predMode[absPartIdx]);
> @@ -606,7 +606,7 @@
>      codeCoeff(ctu, absPartIdx, bEncodeDQP, tuDepthRange);
>  
>      // --- write terminating bit ---
> -    finishCU(ctu, absPartIdx, depth);
> +    finishCU(ctu, absPartIdx, depth, bEncodeDQP);
>  }
>  
>  /* Return bit count of signaling inter mode */
> @@ -658,7 +658,7 @@
>  }
>  
>  /* finish encoding a cu and handle end-of-slice conditions */
> -void Entropy::finishCU(const CUData& ctu, uint32_t absPartIdx, uint32_t depth)
> +void Entropy::finishCU(const CUData& ctu, uint32_t absPartIdx, uint32_t depth, bool bCodeDQP)
>  {
>      const Slice* slice = ctu.m_slice;
>      uint32_t realEndAddress = slice->m_endCUAddr;
> @@ -672,6 +672,10 @@
>      bool granularityBoundary = (((rpelx & granularityMask) == 0 || (rpelx == slice->m_sps->picWidthInLumaSamples )) &&
>                                  ((bpely & granularityMask) == 0 || (bpely == slice->m_sps->picHeightInLumaSamples)));
>  
> +    if (slice->m_pps->bUseDQP)
> +    {
> +        ((CUData&)ctu).setQPSubParts(bCodeDQP ? ctu.getRefQP(absPartIdx) : ctu.m_qp[absPartIdx], absPartIdx, depth);

it's a bit unfortunate having to use a const-cast here. Can deblocking
lookup the ref QP() or is there context missing by then?

> +    }
>      if (granularityBoundary)
>      {
>          // Encode slice finish
> diff -r d138bbee3b01 -r 313a17ef22f4 source/encoder/entropy.h
> --- a/source/encoder/entropy.h	Thu Apr 09 00:48:54 2015 -0400
> +++ b/source/encoder/entropy.h	Sat Apr 18 10:56:21 2015 +0530
> @@ -221,7 +221,7 @@
>      }
>  
>      void encodeCU(const CUData& ctu, const CUGeom &cuGeom, uint32_t absPartIdx, uint32_t depth, bool& bEncodeDQP);
> -    void finishCU(const CUData& ctu, uint32_t absPartIdx, uint32_t depth);
> +    void finishCU(const CUData& ctu, uint32_t absPartIdx, uint32_t depth, bool bEncodeDQP);
>  
>      void writeOut();
>  
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel

-- 
Steve Borho