[x265] [PATCH] analysis: bidir/inter choice using sa8dCost

[Deepthi Nandakumar]

I wanted to check the performance gain, I measured about 10% improvement
for slightly less bitrate, slightly less ssim.

But yes, psy-rd might suffer due to the averaging effect of bidir. For rd
2, certainly makes sense.

On Wed, Mar 11, 2015 at 7:20 PM, Steve Borho <steve at borho.org> wrote:

> On 03/11, deepthi at multicorewareinc.com wrote:
> > # HG changeset patch
> > # User Deepthi Nandakumar <deepthi at multicorewareinc.com>
> > # Date 1426065176 -19800
> > #      Wed Mar 11 14:42:56 2015 +0530
> > # Node ID 7d8f7aadfb6d0af071d804a3eec1511cd00818c5
> > # Parent  37ec8002eba15ec3b7bf2b16a792af572b099078
> > analysis: bidir/inter choice using sa8dCost
> >
> > diff -r 37ec8002eba1 -r 7d8f7aadfb6d source/encoder/analysis.cpp
> > --- a/source/encoder/analysis.cpp     Wed Mar 11 14:31:57 2015 +0530
> > +++ b/source/encoder/analysis.cpp     Wed Mar 11 14:42:56 2015 +0530
> > @@ -749,7 +749,7 @@
> >          /* Compute Merge Cost */
> >          md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom);
> >          md.pred[PRED_SKIP].cu.initSubCU(parentCTU, cuGeom);
> > -        checkMerge2Nx2N_rd0_4(md.pred[PRED_SKIP], md.pred[PRED_MERGE],
> cuGeom);
> > +        checkMerge2Nx2N_rd0_4(md.pred[PRED_SKIP], md.pred[PRED_MERGE],
> cuGeom); /* writes into md.bestMode */
> >
> >          bool earlyskip = false;
> >          if (m_param->rdLevel)
> > @@ -822,24 +822,26 @@
> >
> >              if (m_param->rdLevel >= 3)
> >              {
> > -                /* Calculate RD cost of best inter option */
> > -                if (!m_bChromaSa8d) /* When m_bChromaSa8d is enabled,
> chroma MC has already been done */
> > -                {
> > -                    for (uint32_t puIdx = 0; puIdx <
> bestInter->cu.getNumPartInter(); puIdx++)
> > -                    {
> > -                        PredictionUnit pu(bestInter->cu, cuGeom, puIdx);
> > -                        motionCompensation(bestInter->cu, pu,
> bestInter->predYuv, false, true);
> > -                    }
> > -                }
> > -                encodeResAndCalcRdInterCU(*bestInter, cuGeom);
> > -                checkBestMode(*bestInter, depth);
> > -
> > -                /* If BIDIR is available and within 17/16 of best inter
> option, choose by RDO */
> > +                /* If BIDIR is available and within 17/16 of best inter
> option, choose bidir */
> >                  if (m_slice->m_sliceType == B_SLICE &&
> md.pred[PRED_BIDIR].sa8dCost != MAX_INT64 &&
> >                      md.pred[PRED_BIDIR].sa8dCost * 16 <=
> bestInter->sa8dCost * 17)
> >                  {
> >                      encodeResAndCalcRdInterCU(md.pred[PRED_BIDIR],
> cuGeom);
> > -                    checkBestMode(md.pred[PRED_BIDIR], depth);
> > +                    checkBestMode(md.pred[PRED_BIDIR], depth); /*
> compare against Merge by RDO*/
> > +                }
> > +                else
> > +                {
> > +                    /* Calculate RD cost of best inter option */
> > +                    if (!m_bChromaSa8d) /* When m_bChromaSa8d is
> enabled, chroma MC has already been done */
> > +                    {
> > +                        for (uint32_t puIdx = 0; puIdx <
> bestInter->cu.getNumPartInter(); puIdx++)
> > +                        {
> > +                            PredictionUnit pu(bestInter->cu, cuGeom,
> puIdx);
> > +                            motionCompensation(bestInter->cu, pu,
> bestInter->predYuv, false, true);
> > +                        }
> > +                    }
> > +                    encodeResAndCalcRdInterCU(*bestInter, cuGeom);
> > +                    checkBestMode(*bestInter, depth);
> >                  }
>
> If bidir sa8d is within 17/16 of skip/merge sa8d, then compare bidir RD
> cost with merge/skip RD cost, else compare merge/skip RD with best inter
> RD cost?
>
> It's giving up the chance to compare inter (psy-)RD cost against bidir
> (psy-)RD cost. It might be appropriate for RD levels 0..2, but the case
> for 3 and 4 is less solid.
>
> --
> Steve Borho
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20150311/237fc407/attachment.html>