[x265] [PATCH] fix bugs in analysis-reuse-level=7 and refine-mv-type=AVC
Ashok Kumar Mishra
ashok at multicorewareinc.com
Thu Dec 21 07:57:05 CET 2017
On Thu, Dec 21, 2017 at 12:24 PM, Santhoshini Sekar <
santhoshini at multicorewareinc.com> wrote:
>
>
> On Thu, Dec 21, 2017 at 12:03 PM, Ashok Kumar Mishra <
> ashok at multicorewareinc.com> wrote:
>
>>
>>
>> On Thu, Dec 21, 2017 at 9:03 AM, <santhoshini at multicorewareinc.com>
>> wrote:
>>
>>> # HG changeset patch
>>> # User Santhoshini Sekar <santhoshini at multicorewareinc.com>
>>> # Date 1513677810 -19800
>>> # Tue Dec 19 15:33:30 2017 +0530
>>> # Node ID 32a2cd5926ed3bc64fc5665f3ecc20e9b371cee2
>>> # Parent 57eaef9abfd8204b568498d4a37a23391e790d44
>>> fix bugs in analysis-reuse-level=7 and refine-mv-type=AVC
>>>
>>> diff --git a/source/encoder/analysis.cpp b/source/encoder/analysis.cpp
>>> --- a/source/encoder/analysis.cpp
>>> +++ b/source/encoder/analysis.cpp
>>> @@ -280,7 +280,7 @@
>>> /* generate residual for entire CTU at once and copy to
>>> reconPic */
>>> encodeResidue(ctu, cuGeom);
>>> }
>>> - else if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD &&
>>> m_param->analysisReuseLevel == 10) || ((m_param->bMVType == AVC_INFO) &&
>>> m_param->analysisReuseLevel >= 7))
>>> + else if ((m_param->analysisReuseMode == X265_ANALYSIS_LOAD &&
>>> m_param->analysisReuseLevel == 10) || ((m_param->bMVType == AVC_INFO) &&
>>> m_param->analysisReuseLevel >= 7 && ctu.m_numPartitions <= 16))
>>> {
>>> analysis_inter_data* interDataCTU =
>>> (analysis_inter_data*)m_frame->m_analysisData.interData;
>>> int posCTU = ctu.m_cuAddr * numPartition;
>>> @@ -461,7 +461,7 @@
>>> int lambdaQP = lqp;
>>>
>>> bool doQPRefine = (bDecidedDepth && depth <=
>>> m_slice->m_pps->maxCuDQPDepth) || (!bDecidedDepth && depth ==
>>> m_slice->m_pps->maxCuDQPDepth);
>>> - if (m_param->analysisReuseLevel == 10)
>>> + if (m_param->analysisReuseLevel >= 7)
>>> doQPRefine = false;
>>>
>>> if (doQPRefine)
>>> @@ -1307,7 +1307,7 @@
>>> }
>>>
>>> /* Step 1. Evaluate Merge/Skip candidates for likely
>>> early-outs, if skip mode was not set above */
>>> - if ((mightNotSplit && depth >= minDepth && !md.bestMode &&
>>> !bCtuInfoCheck) || (m_param->bMVType && (m_modeFlag[0] || m_modeFlag[1])))
>>> /* TODO: Re-evaluate if analysis load/save still works */
>>> + if ((mightNotSplit && depth >= minDepth && !md.bestMode &&
>>> !bCtuInfoCheck) || (m_param->bMVType && m_param->analysisReuseLevel == 7 &&
>>> (m_modeFlag[0] || m_modeFlag[1]))) /* TODO: Re-evaluate if analysis
>>> load/save still works */
>>> {
>>> /* Compute Merge Cost */
>>> md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom, qp);
>>> @@ -1318,7 +1318,7 @@
>>> && md.bestMode && md.bestMode->cu.isSkipped(0); //
>>> TODO: sa8d threshold per depth
>>> }
>>>
>>> - if (md.bestMode && m_param->bEnableRecursionSkip &&
>>> !bCtuInfoCheck && !(m_param->bMVType && (m_modeFlag[0] || m_modeFlag[1])))
>>> + if (md.bestMode && m_param->bEnableRecursionSkip &&
>>> !bCtuInfoCheck && !(m_param->bMVType && m_param->analysisReuseLevel == 7 &&
>>> (m_modeFlag[0] || m_modeFlag[1])))
>>> {
>>> skipRecursion = md.bestMode->cu.isSkipped(0);
>>> if (mightSplit && depth >= minDepth && !skipRecursion)
>>> @@ -1330,7 +1330,7 @@
>>> }
>>> }
>>>
>>> - if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <=
>>> 16)
>>> + if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <=
>>> 16 && m_param->analysisReuseLevel == 7)
>>> skipRecursion = true;
>>>
>>> /* Step 2. Evaluate each of the 4 split sub-blocks in series */
>>> @@ -1389,8 +1389,19 @@
>>> }
>>>
>>> /* If analysis mode is simple do not Evaluate other modes */
>>> - if ((m_param->bMVType && cuGeom.numPartitions <= 16) &&
>>> (m_slice->m_sliceType == P_SLICE || m_slice->m_sliceType == B_SLICE))
>>> - mightNotSplit = !(m_checkMergeAndSkipOnly[0] ||
>>> (m_checkMergeAndSkipOnly[0] && m_checkMergeAndSkipOnly[1]));
>>> + if (m_param->bMVType && m_param->analysisReuseLevel == 7)
>>> + {
>>> + if (m_slice->m_sliceType == P_SLICE)
>>> + {
>>> + if (m_checkMergeAndSkipOnly[0])
>>> + skipModes = true;
>>> + }
>>> + else
>>> + {
>>> + if (m_checkMergeAndSkipOnly[0] &&
>>> m_checkMergeAndSkipOnly[1])
>>> + skipModes = true;
>>> + }
>>> + }
>>>
>>> /* Split CUs
>>> * 0 1
>>> @@ -2001,7 +2012,7 @@
>>>
>>> /* Step 1. Evaluate Merge/Skip candidates for likely early-outs
>>> */
>>> if ((mightNotSplit && !md.bestMode && !bCtuInfoCheck) ||
>>> - (m_param->bMVType && (m_modeFlag[0] || m_modeFlag[1])))
>>> + (m_param->bMVType && m_param->analysisReuseLevel == 7 &&
>>> (m_modeFlag[0] || m_modeFlag[1])))
>>> {
>>> md.pred[PRED_SKIP].cu.initSubCU(parentCTU, cuGeom, qp);
>>> md.pred[PRED_MERGE].cu.initSubCU(parentCTU, cuGeom, qp);
>>> @@ -2017,7 +2028,7 @@
>>> skipRecursion = md.bestMode &&
>>> !md.bestMode->cu.getQtRootCbf(0);
>>> }
>>>
>>> - if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <=
>>> 16)
>>> + if (m_param->bMVType && md.bestMode && cuGeom.numPartitions <=
>>> 16 && m_param->analysisReuseLevel == 7)
>>> skipRecursion = true;
>>>
>>> // estimate split cost
>>> @@ -2073,8 +2084,19 @@
>>> }
>>>
>>> /* If analysis mode is simple do not Evaluate other modes */
>>> - if ((m_param->bMVType && cuGeom.numPartitions <= 16) &&
>>> (m_slice->m_sliceType == P_SLICE || m_slice->m_sliceType == B_SLICE))
>>> - mightNotSplit = !(m_checkMergeAndSkipOnly[0] ||
>>> (m_checkMergeAndSkipOnly[0] && m_checkMergeAndSkipOnly[1]));
>>> + if (m_param->bMVType && m_param->analysisReuseLevel == 7)
>>> + {
>>> + if (m_slice->m_sliceType == P_SLICE)
>>> + {
>>> + if (m_checkMergeAndSkipOnly[0])
>>> + skipModes = true;
>>> + }
>>> + else
>>> + {
>>> + if (m_checkMergeAndSkipOnly[0] &&
>>> m_checkMergeAndSkipOnly[1])
>>> + skipModes = true;
>>> + }
>>> + }
>>>
>>> /* Split CUs
>>> * 0 1
>>> diff --git a/source/encoder/encoder.cpp b/source/encoder/encoder.cpp
>>> --- a/source/encoder/encoder.cpp
>>> +++ b/source/encoder/encoder.cpp
>>> @@ -562,7 +562,7 @@
>>> {
>>> int cuOffset = cuI * bytes + pu;
>>> (interData)->mergeFlag[cuPos + cuOffset] =
>>> (srcInterData)->mergeFlag[(mbIndex * 16) + cuOffset];
>>> -
>>> + (interData)->sadCost[cuPos + cuOffset] =
>>> (srcInterData)->sadCost[(mbIndex * 16) + cuOffset];
>>> (interData)->interDir[cuPos + cuOffset] =
>>> (srcInterData)->interDir[(mbIndex * 16) + cuOffset];
>>> for (uint32_t k = 0; k < numDir; k++)
>>> {
>>> @@ -570,10 +570,10 @@
>>> (interData)->refIdx[k][cuPos + cuOffset] =
>>> (srcInterData)->refIdx[k][(mbIndex * 16) + cuOffset];
>>> memcpy(&(interData)->mv[k][cuPos +
>>> cuOffset], &(srcInterData)->mv[k][(mbIndex * 16) + cuOffset],
>>> sizeof(MV));
>>> if (m_param->analysisReuseLevel == 7 &&
>>> numPU == PU_2Nx2N &&
>>> - ((srcInterData)->depth[cuPos +
>>> cuOffset] == (m_param->maxCUSize >> 5)))
>>> + ((interData)->depth[cuPos + cuOffset]
>>> == (m_param->maxCUSize >> 5)))
>>> {
>>> - int mv_x = ((analysis_inter_data
>>> *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) +
>>> cuOffset].x;
>>> - int mv_y = ((analysis_inter_data
>>> *)curFrame->m_analysisData.interData)->mv[k][(mbIndex * 16) +
>>> cuOffset].y;
>>> + int mv_x = ((analysis_inter_data
>>> *)curFrame->m_analysisData.interData)->mv[k][cuPos + cuOffset].x;
>>> + int mv_y = ((analysis_inter_data
>>> *)curFrame->m_analysisData.interData)->mv[k][cuPos + cuOffset].y;
>>>
>>
>> Always use local pointer analysis_inter_data * interData above.
>>
> ok, I'll change this everywhere.
>
>>
>>
>>> if ((mv_x*mv_x + mv_y*mv_y) <=
>>> MVTHRESHOLD)
>>> memset(&curFrame->m_analysisData.modeFlag[k][cuPos
>>> + cuOffset], 1, bytes);
>>> }
>>> @@ -640,9 +640,10 @@
>>> if (m_param->analysisReuseLevel > 4)
>>> {
>>> memset(&(currInterData)->partSize[count],
>>> (interData)->partSize[d], bytes);
>>> - int numPU = nbPartsTable[(currInterData)->
>>> partSize[d]];
>>> - for (int pu = 0; pu < numPU; pu++, d++)
>>> + int numPU = nbPartsTable[(interData)->part
>>> Size[d]];
>>> + for (int pu = 0; pu < numPU; pu++)
>>> {
>>> + if (pu) d++;
>>>
>>
>> Can be written as
>> for (int pu = 0; pu < numPU; pu++, d++)
>>
> this is logically wrong because 'd' should be incremented only when pu is
> non-zero. But incrementing 'd' in the for loop always increments d.
>
Agreed. I didn't see you are using the d in outer for loop.
>
>>
>>
>>> (currInterData)->mergeFlag[count + pu] =
>>> (interData)->mergeFlag[d];
>>> if (m_param->analysisReuseLevel >= 7)
>>> {
>>> @@ -3061,12 +3062,13 @@
>>> if (m_param->analysisReuseLevel >= 7)
>>> {
>>> CHECKED_MALLOC(interData->interDir, uint8_t,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> + CHECKED_MALLOC(interData->sadCost, int64_t,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> for (int dir = 0; dir < numDir; dir++)
>>> {
>>> CHECKED_MALLOC(interData->mvpIdx[dir], uint8_t,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> CHECKED_MALLOC(interData->refIdx[dir], int8_t,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> CHECKED_MALLOC(interData->mv[dir], MV,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> - CHECKED_MALLOC(analysis->modeFlag[dir], uint8_t,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> + CHECKED_MALLOC_ZERO(analysis->modeFlag[dir], uint8_t,
>>> analysis->numPartitions * analysis->numCUsInFrame);
>>> }
>>>
>>> /* Allocate intra in inter */
>>>
>>> _______________________________________________
>>> x265-devel mailing list
>>> x265-devel at videolan.org
>>> https://mailman.videolan.org/listinfo/x265-devel
>>>
>>>
>>
>> _______________________________________________
>> x265-devel mailing list
>> x265-devel at videolan.org
>> https://mailman.videolan.org/listinfo/x265-devel
>>
>>
>
> _______________________________________________
> 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/20171221/447f4b2e/attachment-0001.html>
More information about the x265-devel
mailing list