diff --git a/codec/encoder/core/inc/md.h b/codec/encoder/core/inc/md.h
index b346cef9..febd1199 100644
--- a/codec/encoder/core/inc/md.h
+++ b/codec/encoder/core/inc/md.h
@@ -89,6 +89,10 @@ uint16_t		uiReserved;
 int32_t			iCostSkipMb;
 int32_t			iSadPredSkip;
 
+int32_t			iMbPixX;		// pixel position of MB in horizontal axis
+int32_t			iMbPixY;		// pixel position of MB in vertical axis
+int32_t			iBlock8x8StaticIdc[4];
+
 //NO B frame in our Wels, we can ignore list1
 
 struct {
diff --git a/codec/encoder/core/inc/svc_motion_estimate.h b/codec/encoder/core/inc/svc_motion_estimate.h
index 94652cf9..1e02e3cd 100644
--- a/codec/encoder/core/inc/svc_motion_estimate.h
+++ b/codec/encoder/core/inc/svc_motion_estimate.h
@@ -59,14 +59,20 @@ uint16_t*					pMvdCost;
 union SadPredISatdUnit	uSadPredISatd; //reuse the sad_pred as a temp pData
 uint32_t					uiSadCost;  //used by ME and RC //max SAD should be max_delta*size+lambda*mvdsize = 255*256+91*33*2 = 65280 + 6006 = 71286 > (2^16)-1 = 65535
 uint32_t					uiSatdCost; /* satd + lm * nbits */
+uint32_t					uiSadCostThreshold;
+int32_t						iCurMeBlockPixX;
+int32_t						iCurMeBlockPixY;
 uint8_t						uiPixel;   /* PIXEL_WxH */
 uint8_t						uiReserved;
 
 uint8_t*						pEncMb;
 uint8_t*						pRefMb;
+uint8_t*						pColoRefMb;
 
 SMVUnitXY					sMvp;
 SMVUnitXY					sMvBase;
+SMVUnitXY					sDirectionalMv;
+
 /* output */
 SMVUnitXY					sMv;
 } SWelsME;
diff --git a/codec/encoder/core/src/svc_base_layer_md.cpp b/codec/encoder/core/src/svc_base_layer_md.cpp
index 648e73c0..59b00e16 100644
--- a/codec/encoder/core/src/svc_base_layer_md.cpp
+++ b/codec/encoder/core/src/svc_base_layer_md.cpp
@@ -977,6 +977,8 @@ int32_t WelsMdP16x16 (SWelsFuncPtrList* pFunc, SDqLayer* pCurLayer, SWelsMD* pWe
   const int32_t kiMbWidth	= pCurLayer->iMbWidth;	// for assign once
   const int32_t kiMbHeight	= pCurLayer->iMbHeight;
 
+  sMe16x16->iCurMeBlockPixX = pWelsMd->iMbPixX;
+  sMe16x16->iCurMeBlockPixY = pWelsMd->iMbPixY;
   sMe16x16->uiPixel = BLOCK_16x16;
   sMe16x16->pMvdCost = pWelsMd->pMvdCost;
 
@@ -1078,18 +1080,22 @@ int32_t WelsMdP8x8 (SWelsFuncPtrList* pFunc, SDqLayer* pCurDqLayer, SWelsMD* pWe
   int32_t iLineSizeEnc = pCurDqLayer->iEncStride[0];
   int32_t iLineSizeRef = pCurDqLayer->pRefPic->iLineSize[0];
   SWelsME* sMe8x8;
-  int32_t i, iIdxX, iIdxY, iStrideEnc, iStrideRef;
+  int32_t i, iIdxX, iIdxY, iPixelX, iPixelY, iStrideEnc, iStrideRef;
   int32_t iCostP8x8 = 0;
   for (i = 0; i < 4; i++) {
     iIdxX = i & 1;
     iIdxY = i >> 1;
-    iStrideEnc = (iIdxX << 3) + ((iIdxY << 3) * iLineSizeEnc);
-    iStrideRef = (iIdxX << 3) + ((iIdxY << 3) * iLineSizeRef);
+    iPixelX = (iIdxX << 3);
+    iPixelY = (iIdxY << 3);
+    iStrideEnc = iPixelX + ( iPixelY * iLineSizeEnc);
+    iStrideRef = iPixelX + ( iPixelY * iLineSizeRef);
 
     sMe8x8 = &pWelsMd->sMe.sMe8x8[i];
 
+    sMe8x8->iCurMeBlockPixX = pWelsMd->iMbPixX + iPixelX;
+    sMe8x8->iCurMeBlockPixY = pWelsMd->iMbPixY + iPixelY;
     sMe8x8->uiPixel = BLOCK_8x8;
-    sMe8x8->pMvdCost     = pWelsMd->pMvdCost;
+    sMe8x8->pMvdCost = pWelsMd->pMvdCost;
 
     sMe8x8->pEncMb       = pMbCache->SPicData.pEncMb[0] + iStrideEnc;
     sMe8x8->pRefMb       = pMbCache->SPicData.pRefMb[0] + iStrideRef;
diff --git a/codec/encoder/core/src/svc_encode_slice.cpp b/codec/encoder/core/src/svc_encode_slice.cpp
index 3f7bc30a..395f70a2 100644
--- a/codec/encoder/core/src/svc_encode_slice.cpp
+++ b/codec/encoder/core/src/svc_encode_slice.cpp
@@ -918,6 +918,14 @@ bool DynSlcJudgeSliceBoundaryStepBack (void* pCtx, void* pSlice, SSliceCtx* pSli
 ///////////////
 //  pMb loop
 ///////////////
+inline void WelsInitInterMDStruc(const SMB* pCurMb, uint16_t *pMvdCostTableInter, const int32_t kiMvdInterTableStride, SWelsMD* pMd )
+{
+  pMd->iLambda = g_kiQpCostTable[pCurMb->uiLumaQp];
+  pMd->pMvdCost = &pMvdCostTableInter[pCurMb->uiLumaQp * kiMvdInterTableStride];
+  pMd->	iMbPixX = (pCurMb->iMbX<<4);
+  pMd->	iMbPixY = (pCurMb->iMbY<<4);
+  memset( &pMd->iBlock8x8StaticIdc[0], 0, sizeof(pMd->iBlock8x8StaticIdc) );
+}
 // for inter non-dynamic pSlice
 int32_t WelsMdInterMbLoop (sWelsEncCtx* pEncCtx, SSlice* pSlice, void* pWelsMd, const int32_t kiSliceFirstMbXY) {
   SWelsMD* pMd					= (SWelsMD*)pWelsMd;
@@ -948,10 +956,9 @@ int32_t WelsMdInterMbLoop (sWelsEncCtx* pEncCtx, SSlice* pSlice, void* pWelsMd,
     pEncCtx->pFuncList->pfRc.pfWelsRcMbInit (pEncCtx, pCurMb, pSlice);
 
     //step (2). save some vale for future use, initial pWelsMd
-    pMd->iLambda = g_kiQpCostTable[pCurMb->uiLumaQp];
-    pMd->pMvdCost = &pMvdCostTableInter[pCurMb->uiLumaQp * kiMvdInterTableStride];
     WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);
     WelsMdInterInit (pEncCtx, pSlice, pCurMb, kiSliceFirstMbXY);
+    WelsInitInterMDStruc(pCurMb, pMvdCostTableInter, kiMvdInterTableStride, pMd );
     pEncCtx->pFuncList->pfInterMd (pEncCtx, pMd, pSlice, pCurMb, pMbCache);
     //mb_qp
 
@@ -1047,11 +1054,9 @@ int32_t WelsMdInterMbLoopOverDynamicSlice (sWelsEncCtx* pEncCtx, SSlice* pSlice,
     }
 
     //step (2). save some vale for future use, initial pWelsMd
-    pMd->iLambda = g_kiQpCostTable[pCurMb->uiLumaQp];
-    pMd->pMvdCost = &pMvdCostTableInter[pCurMb->uiLumaQp * kiMvdInterTableStride];
-
     WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);
     WelsMdInterInit (pEncCtx, pSlice, pCurMb, kiSliceFirstMbXY);
+    WelsInitInterMDStruc(pCurMb, pMvdCostTableInter, kiMvdInterTableStride, pMd );
     pEncCtx->pFuncList->pfInterMd (pEncCtx, pMd, pSlice, pCurMb, pMbCache);
     //mb_qp