void CComplexityAnalysis::AnalyzeGomComplexityViaVar (SPixMap* pSrcPixMap, SPixMap* pRefPixMap) {
int32_t iWidth = pSrcPixMap->sRect.iRectWidth;
int32_t iHeight = pSrcPixMap->sRect.iRectHeight;
int32_t iMbWidth = iWidth >> 4;
int32_t iMbHeight = iHeight >> 4;
int32_t iMbNum = iMbWidth * iMbHeight;
int32_t iMbNumInGom = m_sComplexityAnalysisParam.iMbNumInGom;
int32_t iGomMbNum = (iMbNum + iMbNumInGom - 1) / iMbNumInGom;
int32_t iGomSampleNum = 0;
int32_t iGomMbStartIndex = 0, iGomMbEndIndex = 0, iGomMbRowNum = 0;
int32_t iMbStartIndex = 0, iMbEndIndex = 0;
int32_t iStartSampleIndex = 0;
SVAACalcResult* pVaaCalcResults = m_sComplexityAnalysisParam.pCalcResult;
int32_t* pGomComplexity = (int32_t*)m_sComplexityAnalysisParam.pGomComplexity;
uint8_t* pSrcY = NULL;
int32_t iCurStride = 0;
uint8_t* pCurTmp = NULL;
uint32_t uiSampleSum = 0, uiSquareSum = 0;
pSrcY = (uint8_t*)pSrcPixMap->pPixel[0];
iCurStride = pSrcPixMap->iStride[0];
for (int32_t j = 0; j < iGomMbNum; j ++) {
uiSampleSum = 0;
uiSquareSum = 0;
iGomMbStartIndex = j * iMbNumInGom;
iGomMbEndIndex = WELS_MIN ((j + 1) * iMbNumInGom, iMbNum);
iGomMbRowNum = (iGomMbEndIndex + iMbWidth - 1) / iMbWidth - iGomMbStartIndex / iMbWidth;
iMbStartIndex = iGomMbStartIndex;
iMbEndIndex = WELS_MIN ((iMbStartIndex / iMbWidth + 1) * iMbWidth, iGomMbEndIndex);
iStartSampleIndex = (iMbStartIndex / iMbWidth) * MB_WIDTH_LUMA * iCurStride + (iMbStartIndex % iMbWidth) *
MB_WIDTH_LUMA;
iGomSampleNum = (iMbEndIndex - iMbStartIndex) * MB_WIDTH_LUMA * MB_WIDTH_LUMA;
do {
pCurTmp = pSrcY + iStartSampleIndex;
for (int32_t i = iMbStartIndex; i < iMbEndIndex; i ++) {
uiSampleSum += pVaaCalcResults->pSum16x16[i];
uiSquareSum += pVaaCalcResults->pSumOfSquare16x16[i];
}
iMbStartIndex = iMbEndIndex;
iMbEndIndex = WELS_MIN (iMbEndIndex + iMbWidth, iGomMbEndIndex);
iStartSampleIndex = (iMbStartIndex / iMbWidth) * MB_WIDTH_LUMA * iCurStride + (iMbStartIndex % iMbWidth) *
MB_WIDTH_LUMA;
} while (--iGomMbRowNum);
pGomComplexity[j] = uiSquareSum - (uiSampleSum * uiSampleSum / iGomSampleNum);
}
}
void CComplexityAnalysis::AnalyzeGomComplexityViaSad (SPixMap* pSrcPixMap, SPixMap* pRefPixMap) {
int32_t iWidth = pSrcPixMap->sRect.iRectWidth;
int32_t iHeight = pSrcPixMap->sRect.iRectHeight;
int32_t iMbWidth = iWidth >> 4;
int32_t iMbHeight = iHeight >> 4;
int32_t iMbNum = iMbWidth * iMbHeight;
int32_t iMbNumInGom = m_sComplexityAnalysisParam.iMbNumInGom;
int32_t iGomMbNum = (iMbNum + iMbNumInGom - 1) / iMbNumInGom;
int32_t iGomMbStartIndex = 0, iGomMbEndIndex = 0, iGomMbRowNum = 0;
int32_t iMbStartIndex = 0, iMbEndIndex = 0;
uint8_t* pBackgroundMbFlag = (uint8_t*)m_sComplexityAnalysisParam.pBackgroundMbFlag;
uint32_t* uiRefMbType = (uint32_t*)m_sComplexityAnalysisParam.uiRefMbType;
SVAACalcResult* pVaaCalcResults = m_sComplexityAnalysisParam.pCalcResult;
int32_t* pGomForegroundBlockNum = (int32_t*)m_sComplexityAnalysisParam.pGomForegroundBlockNum;
int32_t* pGomComplexity = (int32_t*)m_sComplexityAnalysisParam.pGomComplexity;
uint32_t uiGomSad = 0, uiFrameSad = 0;
InitGomSadFunc (m_pfGomSad, m_sComplexityAnalysisParam.iCalcBgd);
for (int32_t j = 0; j < iGomMbNum; j ++) {
uiGomSad = 0;
iGomMbStartIndex = j * iMbNumInGom;
iGomMbEndIndex = WELS_MIN ((j + 1) * iMbNumInGom, iMbNum);
iGomMbRowNum = (iGomMbEndIndex + iMbWidth - 1) / iMbWidth - iGomMbStartIndex / iMbWidth;
iMbStartIndex = iGomMbStartIndex;
iMbEndIndex = WELS_MIN ((iMbStartIndex / iMbWidth + 1) * iMbWidth, iGomMbEndIndex);
do {
for (int32_t i = iMbStartIndex; i < iMbEndIndex; i ++) {
m_pfGomSad (&uiGomSad, pGomForegroundBlockNum + j, pVaaCalcResults->pSad8x8[i], pBackgroundMbFlag[i]
&& !IS_INTRA (uiRefMbType[i]));
}
iMbStartIndex = iMbEndIndex;
iMbEndIndex = WELS_MIN (iMbEndIndex + iMbWidth , iGomMbEndIndex);
} while (--iGomMbRowNum);
pGomComplexity[j] = uiGomSad;
uiFrameSad += pGomComplexity[j];
}
m_sComplexityAnalysisParam.iFrameComplexity = uiFrameSad;
}
void CComplexityAnalysis::AnalyzeGomComplexityViaVar (SPixMap* pSrcPixMap, SPixMap* pRefPixMap) {
int32_t iWidth = pSrcPixMap->sRect.iRectWidth;
int32_t iHeight = pSrcPixMap->sRect.iRectHeight;
int32_t iMbWidth = iWidth >> 4;
int32_t iMbHeight = iHeight >> 4;
int32_t iMbNum = iMbWidth * iMbHeight;
int32_t iMbNumInGom = m_sComplexityAnalysisParam.iMbNumInGom;
int32_t iGomMbNum = (iMbNum + iMbNumInGom - 1) / iMbNumInGom;
int32_t iGomSampleNum = 0;
int32_t iGomMbStartIndex = 0, iGomMbEndIndex = 0, iGomMbRowNum = 0;
int32_t iMbStartIndex = 0, iMbEndIndex = 0;
SVAACalcResult* pVaaCalcResults = m_sComplexityAnalysisParam.pCalcResult;
int32_t* pGomComplexity = (int32_t*)m_sComplexityAnalysisParam.pGomComplexity;
uint32_t uiFrameSad = 0;
uint32_t uiSampleSum = 0, uiSquareSum = 0;
for (int32_t j = 0; j < iGomMbNum; j ++) {
uiSampleSum = 0;
uiSquareSum = 0;
iGomMbStartIndex = j * iMbNumInGom;
iGomMbEndIndex = WELS_MIN ((j + 1) * iMbNumInGom, iMbNum);
iGomMbRowNum = (iGomMbEndIndex + iMbWidth - 1) / iMbWidth - iGomMbStartIndex / iMbWidth;
iMbStartIndex = iGomMbStartIndex;
iMbEndIndex = WELS_MIN ((iMbStartIndex / iMbWidth + 1) * iMbWidth, iGomMbEndIndex);
iGomSampleNum = (iMbEndIndex - iMbStartIndex) * MB_WIDTH_LUMA * MB_WIDTH_LUMA;
do {
for (int32_t i = iMbStartIndex; i < iMbEndIndex; i ++) {
uiSampleSum += pVaaCalcResults->pSum16x16[i];
uiSquareSum += pVaaCalcResults->pSumOfSquare16x16[i];
}
iMbStartIndex = iMbEndIndex;
iMbEndIndex = WELS_MIN (iMbEndIndex + iMbWidth, iGomMbEndIndex);
} while (--iGomMbRowNum);
pGomComplexity[j] = uiSquareSum - (uiSampleSum * uiSampleSum / iGomSampleNum);
uiFrameSad += pGomComplexity[j];
}
m_sComplexityAnalysisParam.iFrameComplexity = uiFrameSad;
}
int32_t CComplexityAnalysis::GetFrameSadExcludeBackground (SPixMap* pSrcPixMap, SPixMap* pRefPixMap) {
int32_t iWidth = pSrcPixMap->sRect.iRectWidth;
int32_t iHeight = pSrcPixMap->sRect.iRectHeight;
int32_t iMbWidth = iWidth >> 4;
int32_t iMbHeight = iHeight >> 4;
int32_t iMbNum = iMbWidth * iMbHeight;
int32_t iMbNumInGom = m_sComplexityAnalysisParam.iMbNumInGom;
int32_t iGomMbNum = (iMbNum + iMbNumInGom - 1) / iMbNumInGom;
int32_t iGomMbStartIndex = 0, iGomMbEndIndex = 0;
uint8_t* pBackgroundMbFlag = (uint8_t*)m_sComplexityAnalysisParam.pBackgroundMbFlag;
uint32_t* uiRefMbType = (uint32_t*)m_sComplexityAnalysisParam.uiRefMbType;
SVAACalcResult* pVaaCalcResults = m_sComplexityAnalysisParam.pCalcResult;
int32_t* pGomForegroundBlockNum = m_sComplexityAnalysisParam.pGomForegroundBlockNum;
uint32_t uiFrameSad = 0;
for (int32_t j = 0; j < iGomMbNum; j ++) {
iGomMbStartIndex = j * iMbNumInGom;
iGomMbEndIndex = WELS_MIN ((j + 1) * iMbNumInGom, iMbNum);
for (int32_t i = iGomMbStartIndex; i < iGomMbEndIndex; i ++) {
if (pBackgroundMbFlag[i] == 0 || IS_INTRA (uiRefMbType[i])) {
pGomForegroundBlockNum[j]++;
uiFrameSad += pVaaCalcResults->pSad8x8[i][0];
uiFrameSad += pVaaCalcResults->pSad8x8[i][1];
uiFrameSad += pVaaCalcResults->pSad8x8[i][2];
uiFrameSad += pVaaCalcResults->pSad8x8[i][3];
}
}
}
return (uiFrameSad);
}
void CComplexityAnalysisScreen::GomComplexityAnalysisIntra (SPixMap* pSrc) {
int32_t iWidth = pSrc->sRect.iRectWidth;
int32_t iHeight = pSrc->sRect.iRectHeight;
int32_t iBlockWidth = iWidth >> 4;
int32_t iBlockHeight = iHeight >> 4;
int32_t iBlockSadH, iBlockSadV, iGomSad = 0;
int32_t iIdx = 0;
uint8_t* pPtrY = NULL;
int32_t iStrideY = 0;
int32_t iRowStrideY = 0;
uint8_t* pTmpCur = NULL;
ENFORCE_STACK_ALIGN_1D (uint8_t, iMemPredMb, 256, 16)
pPtrY = (uint8_t*)pSrc->pPixel[0];
iStrideY = pSrc->iStride[0];
iRowStrideY = iStrideY << 4;
m_ComplexityAnalysisParam.iFrameComplexity = 0;
for (int32_t j = 0; j < iBlockHeight; j ++) {
pTmpCur = pPtrY;
for (int32_t i = 0; i < iBlockWidth; i++) {
iBlockSadH = iBlockSadV = 0x7fffffff; // INT_MAX
if (j > 0) {
m_pIntraFunc[0] (iMemPredMb, pTmpCur, iStrideY);
iBlockSadH = m_pSadFunc (pTmpCur, iStrideY, iMemPredMb, 16);
}
if (i > 0) {
m_pIntraFunc[1] (iMemPredMb, pTmpCur, iStrideY);
iBlockSadV = m_pSadFunc (pTmpCur, iStrideY, iMemPredMb, 16);
}
if (i || j)
iGomSad += WELS_MIN (iBlockSadH, iBlockSadV);
pTmpCur += 16;
if (i == iBlockWidth - 1 && ((j + 1) % m_ComplexityAnalysisParam.iMbRowInGom == 0 || j == iBlockHeight - 1)) {
m_ComplexityAnalysisParam.pGomComplexity[iIdx] = iGomSad;
m_ComplexityAnalysisParam.iFrameComplexity += iGomSad;
iIdx++;
iGomSad = 0;
}
}
pPtrY += iRowStrideY;
}
m_ComplexityAnalysisParam.iGomNumInFrame = iIdx;
}
int32_t PredIntra4x4Mode (int8_t* pIntraPredMode, int32_t iIdx4) {
int8_t iTopMode = pIntraPredMode[g_kuiScan8[iIdx4] - 8];
int8_t iLeftMode = pIntraPredMode[g_kuiScan8[iIdx4] - 1];
int8_t iBestMode;
if (-1 == iLeftMode || -1 == iTopMode) {
iBestMode = 2;
} else {
iBestMode = WELS_MIN (iLeftMode, iTopMode);
}
return iBestMode;
}
void ScrollDetectionCore (SPixMap* pSrcPixMap, SPixMap* pRefPixMap, int32_t iWidth, int32_t iHeight,
int32_t iOffsetX, int32_t iOffsetY, SScrollDetectionParam& sScrollDetectionParam) {
bool bScrollDetected = 0;
uint8_t* pYLine;
uint8_t* pYTmp;
int32_t iTestPos, iSearchPos = 0, iOffsetAbs, iMaxAbs;
int32_t iPicHeight = pRefPixMap->sRect.iRectHeight;
int32_t iMinHeight = WELS_MAX (iOffsetY, 0);
int32_t iMaxHeight = WELS_MIN (iOffsetY + iHeight - 1, iPicHeight - 1) ; //offset_y + height - 1;//
uint8_t* pYRef, *pYSrc;
int32_t iYStride;
pYRef = (uint8_t*)pRefPixMap->pPixel[0];
pYSrc = (uint8_t*)pSrcPixMap->pPixel[0];
iYStride = pRefPixMap->iStride[0];
iTestPos = SelectTestLine (pYSrc, iWidth, iHeight, iPicHeight, iYStride, iOffsetX, iOffsetY);
if (iTestPos == -1) {
sScrollDetectionParam.bScrollDetectFlag = 0;
return;
}
pYLine = pYSrc + iYStride * iTestPos + iOffsetX;
iMaxAbs = WELS_MIN (WELS_MAX (iTestPos - iMinHeight - 1, iMaxHeight - iTestPos), MAX_SCROLL_MV_Y);
iSearchPos = iTestPos;
for (iOffsetAbs = 0; iOffsetAbs <= iMaxAbs; iOffsetAbs++) {
iSearchPos = iTestPos + iOffsetAbs;
if (iSearchPos <= iMaxHeight) {
pYTmp = pYRef + iSearchPos * iYStride + iOffsetX;
if (!CompareLine (pYLine, pYTmp, iWidth)) {
uint8_t* pYUpper, *pYLineUpper;
int32_t iCheckedLines;
int32_t iLowOffset = WELS_MIN (iMaxHeight - iSearchPos, CHECK_OFFSET);
int32_t i;
iCheckedLines = WELS_MIN (iTestPos - iMinHeight + iLowOffset, 2 * CHECK_OFFSET);
pYUpper = pYTmp - (iCheckedLines - iLowOffset) * iYStride;
pYLineUpper = pYLine - (iCheckedLines - iLowOffset) * iYStride;
for (i = 0; i < iCheckedLines; i ++) {
if (CompareLine (pYLineUpper, pYUpper, iWidth)) {
break;
}
pYUpper += iYStride;
pYLineUpper += iYStride;
}
if (i == iCheckedLines) {
bScrollDetected = 1;
break;
}
}
}
iSearchPos = iTestPos - iOffsetAbs - 1;
if (iSearchPos >= iMinHeight) {
pYTmp = pYRef + iSearchPos * iYStride + iOffsetX;
if (!CompareLine (pYLine, pYTmp, iWidth)) {
uint8_t* pYUpper, *pYLineUpper;
int32_t iCheckedLines;
int32_t iUpOffset = WELS_MIN (iSearchPos - iMinHeight, CHECK_OFFSET);
int32_t i;
pYUpper = pYTmp - iUpOffset * iYStride;
pYLineUpper = pYLine - iUpOffset * iYStride;
iCheckedLines = WELS_MIN (iMaxHeight - iTestPos + iUpOffset, 2 * CHECK_OFFSET);
for (i = 0; i < iCheckedLines; i ++) {
if (CompareLine (pYLineUpper, pYUpper, iWidth)) {
break;
}
pYUpper += iYStride;
pYLineUpper += iYStride;
}
if (i == iCheckedLines) {
bScrollDetected = 1;
break;
}
}
}
}
if (!bScrollDetected) {
sScrollDetectionParam.bScrollDetectFlag = 0;
} else {
sScrollDetectionParam.bScrollDetectFlag = 1;
sScrollDetectionParam.iScrollMvY = iSearchPos - iTestPos; // pre_pos - cur_pos, change to mv
sScrollDetectionParam.iScrollMvX = 0;
}
}
void CComplexityAnalysisScreen::GomComplexityAnalysisInter (SPixMap* pSrc, SPixMap* pRef, bool bScrollFlag) {
int32_t iWidth = pSrc->sRect.iRectWidth;
int32_t iHeight = pSrc->sRect.iRectHeight;
int32_t iBlockWidth = iWidth >> 4;
int32_t iBlockHeight = iHeight >> 4;
int32_t iInterSad, iScrollSad, iBlockSadH, iBlockSadV, iGomSad = 0;
int32_t iIdx = 0;
int32_t iScrollMvX = m_ComplexityAnalysisParam.sScrollResult.iScrollMvX;
int32_t iScrollMvY = m_ComplexityAnalysisParam.sScrollResult.iScrollMvY;
uint8_t* pPtrX = NULL, *pPtrY = NULL;
int32_t iStrideX = 0, iStrideY = 0;
int32_t iRowStrideX = 0, iRowStrideY = 0;
uint8_t* pTmpRef = NULL, *pTmpCur = NULL, *pTmpRefScroll = NULL;
ENFORCE_STACK_ALIGN_1D (uint8_t, iMemPredMb, 256, 16)
pPtrX = (uint8_t*)pRef->pPixel[0];
pPtrY = (uint8_t*)pSrc->pPixel[0];
iStrideX = pRef->iStride[0];
iStrideY = pSrc->iStride[0];
iRowStrideX = pRef->iStride[0] << 4;
iRowStrideY = pSrc->iStride[0] << 4;
m_ComplexityAnalysisParam.iFrameComplexity = 0;
for (int32_t j = 0; j < iBlockHeight; j ++) {
pTmpRef = pPtrX;
pTmpCur = pPtrY;
for (int32_t i = 0; i < iBlockWidth; i++) {
int32_t iBlockPointX = i << 4;
int32_t iBlockPointY = j << 4;
iInterSad = m_pSadFunc (pTmpCur, iStrideY, pTmpRef, iStrideX);
if (bScrollFlag) {
if ((iInterSad != 0) &&
(iBlockPointX + iScrollMvX >= 0) && (iBlockPointX + iScrollMvX <= iWidth - 8) &&
(iBlockPointY + iScrollMvY >= 0) && (iBlockPointY + iScrollMvY <= iHeight - 8)) {
pTmpRefScroll = pTmpRef - iScrollMvY * iStrideX + iScrollMvX;
iScrollSad = m_pSadFunc (pTmpCur, iStrideY, pTmpRefScroll, iStrideX);
if (iScrollSad < iInterSad) {
iInterSad = iScrollSad;
}
}
}
iBlockSadH = iBlockSadV = 0x7fffffff; // INT_MAX
if (j > 0) {
m_pIntraFunc[0] (iMemPredMb, pTmpCur, iStrideY);
iBlockSadH = m_pSadFunc (pTmpCur, iStrideY, iMemPredMb, 16);
}
if (i > 0) {
m_pIntraFunc[1] (iMemPredMb, pTmpCur, iStrideY);
iBlockSadV = m_pSadFunc (pTmpCur, iStrideY, iMemPredMb, 16);
}
iGomSad += WELS_MIN (WELS_MIN (iBlockSadH, iBlockSadV), iInterSad);
if (i == iBlockWidth - 1 && ((j + 1) % m_ComplexityAnalysisParam.iMbRowInGom == 0 || j == iBlockHeight - 1)) {
m_ComplexityAnalysisParam.pGomComplexity[iIdx] = iGomSad;
m_ComplexityAnalysisParam.iFrameComplexity += iGomSad;
iIdx++;
iGomSad = 0;
}
pTmpRef += 16;
pTmpCur += 16;
}
pPtrX += iRowStrideX;
pPtrY += iRowStrideY;
}
m_ComplexityAnalysisParam.iGomNumInFrame = iIdx;
}
