shithub: openh264

--- a/codec/encoder/core/inc/rc.h

+++ b/codec/encoder/core/inc/rc.h

@@ -66,8 +66,6 @@

   VGOP_SIZE             = 8,

   //qp information

-  FIX_MIN_QP_MODE       = 10, //qp <10 will cause level code overflow in cavlc coding which isn't suppored in baseline profile

-  FIX_MAX_QP_MODE       = 51,

   GOM_MIN_QP_MODE       = 12,

   GOM_MAX_QP_MODE       = 36,

   MAX_LOW_BR_QP			= 42,

--- a/codec/encoder/core/inc/set_mb_syn_cavlc.h

+++ b/codec/encoder/core/inc/set_mb_syn_cavlc.h

@@ -79,8 +79,8 @@

 void  InitCavlcTable();

-void  WriteBlockResidualCavlc (int16_t* pCoffLevel, int32_t iEndIdx, int32_t iCalRunLevelFlag,

-                               int32_t iResidualProperty, int8_t iNC, SBitStringAux* pBs);

+int32_t  WriteBlockResidualCavlc (int16_t* pCoffLevel, int32_t iEndIdx, int32_t iCalRunLevelFlag,

+                                  int32_t iResidualProperty, int8_t iNC, SBitStringAux* pBs);

 #if defined(__cplusplus)

 extern "C" {

--- a/codec/encoder/core/inc/svc_enc_slice_segment.h

+++ b/codec/encoder/core/inc/svc_enc_slice_segment.h

@@ -95,6 +95,7 @@

   int32_t		iBsStackLeftBits;

   int32_t		iMbSkipRunStack;

+  uint8_t   uiLastMbQp;

 } SDynamicSlicingStack;

/*!

--- a/codec/encoder/core/inc/svc_set_mb_syn_cavlc.h

+++ b/codec/encoder/core/inc/svc_set_mb_syn_cavlc.h

@@ -50,7 +50,7 @@

 namespace WelsSVCEnc {

-void WelsWriteMbResidual (SMbCache* sMbCacheInfo, SMB* pCurMb, SBitStringAux* pBs);

+int32_t WelsWriteMbResidual (SMbCache* sMbCacheInfo, SMB* pCurMb, SBitStringAux* pBs);

 void WelsSpatialWriteSubMbPred (sWelsEncCtx* pEncCtx, SSlice* pSlice, SMB* pCurMb);

--- a/codec/encoder/core/inc/wels_const.h

+++ b/codec/encoder/core/inc/wels_const.h

@@ -202,6 +202,7 @@

   ENC_RETURN_CORRECTED = 0x08, //unexpected value but corrected by encoder

   ENC_RETURN_INVALIDINPUT = 0x10, //invalid input

   ENC_RETURN_MEMOVERFLOWFOUND = 0x20,

+  ENC_RETURN_VLCOVERFLOWFOUND = 0x40

};

 //TODO: need to complete the return checking in encoder and fill in more types if needed

--- a/codec/encoder/core/src/ratectl.cpp

+++ b/codec/encoder/core/src/ratectl.cpp

@@ -928,7 +928,7 @@

     pEncCtx->iGlobalQp = WELS_CLIP3 (WELS_ROUND (pEncCtx->iGlobalQp -

                                      pEncCtx->pVaa->sAdaptiveQuantParam.dAverMotionTextureIndexToDeltaQp), GOM_MIN_QP_MODE, GOM_MAX_QP_MODE);

   } else {

-    pEncCtx->iGlobalQp = WELS_CLIP3 (pEncCtx->iGlobalQp, FIX_MIN_QP_MODE, FIX_MAX_QP_MODE);

+    pEncCtx->iGlobalQp = WELS_CLIP3 (pEncCtx->iGlobalQp, 0, 51);

   pWelsSvcRc->iAverageFrameQp = pEncCtx->iGlobalQp;

@@ -940,14 +940,16 @@

 void  WelsRcMbInitDisable (void* pCtx, SMB* pCurMb, SSlice* pSlice) {

   sWelsEncCtx* pEncCtx = (sWelsEncCtx*)pCtx;

   int32_t iLumaQp					= pEncCtx->iGlobalQp;

   SDqLayer* pCurLayer				= pEncCtx->pCurDqLayer;

   const uint8_t kuiChromaQpIndexOffset = pCurLayer->sLayerInfo.pPpsP->uiChromaQpIndexOffset;

   if (pEncCtx->pSvcParam->bEnableAdaptiveQuant && (pEncCtx->eSliceType == P_SLICE)) {

     iLumaQp   = (int8_t)WELS_CLIP3 (iLumaQp +

                                     pEncCtx->pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp[pCurMb->iMbXY], GOM_MIN_QP_MODE, 51);

   } else {

-    iLumaQp = WELS_CLIP3 (iLumaQp, FIX_MIN_QP_MODE, FIX_MAX_QP_MODE);

+    iLumaQp = WELS_CLIP3 (iLumaQp, 0, 51);

   pCurMb->uiChromaQp = g_kuiChromaQpTable[CLIP3_QP_0_51 (iLumaQp + kuiChromaQpIndexOffset)];

   pCurMb->uiLumaQp = iLumaQp;

--- a/codec/encoder/core/src/set_mb_syn_cavlc.cpp

+++ b/codec/encoder/core/src/set_mb_syn_cavlc.cpp

@@ -41,6 +41,7 @@

 #include "set_mb_syn_cavlc.h"

 #include "vlc_encoder.h"

 #include "cpu_core.h"

+#include "wels_const.h"

 namespace WelsSVCEnc {

 SCoeffFunc    sCoeffFunc;

@@ -77,8 +78,8 @@

   return iTotalZeros;

-void  WriteBlockResidualCavlc (int16_t* pCoffLevel, int32_t iEndIdx, int32_t iCalRunLevelFlag,

-                               int32_t iResidualProperty, int8_t iNC, SBitStringAux* pBs) {

+int32_t  WriteBlockResidualCavlc (int16_t* pCoffLevel, int32_t iEndIdx, int32_t iCalRunLevelFlag,

+                                  int32_t iResidualProperty, int8_t iNC, SBitStringAux* pBs) {

   ENFORCE_STACK_ALIGN_1D (int16_t, iLevel, 16, 16)

   ENFORCE_STACK_ALIGN_1D (uint8_t, uiRun, 16, 16)

@@ -121,7 +122,7 @@

     CAVLC_BS_WRITE (n, iValue);

     CAVLC_BS_UNINIT (pBs);

-    return;

+    return ENC_RETURN_SUCCESS;

   /* Step 4: */

@@ -152,7 +153,9 @@

     } else if (iLevelPrefix >= 15) {

       iLevelPrefix = 15;

       iLevelSuffix = iLevelCode - (iLevelPrefix << uiSuffixLength);

+      //for baseline profile,overflow when the length of iLevelSuffix is larger than 11.

+      if (iLevelSuffix >> 11)

+        return ENC_RETURN_VLCOVERFLOWFOUND;

       if (uiSuffixLength == 0) {

         iLevelSuffix -= 15;

@@ -197,6 +200,7 @@

   CAVLC_BS_UNINIT (pBs);

+  return ENC_RETURN_SUCCESS;

--- a/codec/encoder/core/src/svc_encode_slice.cpp

+++ b/codec/encoder/core/src/svc_encode_slice.cpp

@@ -460,7 +460,25 @@

     pfIdctFour4x4 (pDecV, kiDecStrideChroma, pDecV, kiDecStrideChroma, pScaledTcoeff + 320);

+inline void StashMBStatus (SDynamicSlicingStack* pDss, SBitStringAux* pBs, SSlice* pSlice, int32_t iMbSkipRun = 0) {

+  pDss->pBsStackBufPtr	= pBs->pBufPtr;

+  pDss->uiBsStackCurBits	= pBs->uiCurBits;

+  pDss->iBsStackLeftBits	= pBs->iLeftBits;

+  pDss->uiLastMbQp =  pSlice->uiLastMbQp;

+  pDss->iMbSkipRunStack = iMbSkipRun;

+}

+void StashPopMBStatus (SDynamicSlicingStack* pDss, SBitStringAux* pBs, SSlice* pSlice, int32_t* pMbSkipRun = 0) {

+  pBs->pBufPtr		= pDss->pBsStackBufPtr;

+  pBs->uiCurBits	= pDss->uiBsStackCurBits;

+  pBs->iLeftBits	= pDss->iBsStackLeftBits;

+  pSlice->uiLastMbQp = pDss->uiLastMbQp;

+  if(pMbSkipRun)

+    *pMbSkipRun = pDss->iMbSkipRunStack;

+}

+void UpdateQpForOverflow (SMB* pCurMb, uint8_t kuiChromaQpIndexOffset) {

+  pCurMb->uiLumaQp += DELTA_QP;

+  pCurMb->uiChromaQp = g_kuiChromaQpTable[CLIP3_QP_0_51 (pCurMb->uiLumaQp + kuiChromaQpIndexOffset)];

+}

 // for intra non-dynamic pSlice

 //encapsulate two kinds of reconstruction:

 //first. store base or highest Dependency Layer with only one quality (without CS RS reconstruction)

@@ -477,22 +495,31 @@

   const int32_t kiTotalNumMb		= pCurLayer->iMbWidth * pCurLayer->iMbHeight;

   int32_t iCurMbIdx				= 0, iNumMbCoded = 0;

   const int32_t kiSliceIdx			= pSlice->uiSliceIdx;

+  const uint8_t kuiChromaQpIndexOffset = pCurLayer->sLayerInfo.pPpsP->uiChromaQpIndexOffset;

   SWelsMD sMd;

   int32_t iEncReturn = ENC_RETURN_SUCCESS;

+  SBitStringAux* pBs = pSlice->pSliceBsa;

+  SDynamicSlicingStack sDss;

   for (; ;) {

+    StashMBStatus (&sDss, pBs, pSlice);

     iCurMbIdx	= iNextMbIdx;

     pCurMb = &pMbList[ iCurMbIdx ];

     pEncCtx->pFuncList->pfRc.pfWelsRcMbInit (pEncCtx, pCurMb, pSlice);

+    WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);

+TRY_REENCODING:

     sMd.iLambda = g_kiQpCostTable[pCurMb->uiLumaQp];

-    WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);

     WelsMdIntraMb (pEncCtx, &sMd, pCurMb, pMbCache);

     UpdateNonZeroCountCache (pCurMb, pMbCache);

     iEncReturn = WelsSpatialWriteMbSyn (pEncCtx, pSlice, pCurMb);

+    if (iEncReturn == ENC_RETURN_VLCOVERFLOWFOUND) {

+      StashPopMBStatus (&sDss, pBs, pSlice);

+      UpdateQpForOverflow (pCurMb, kuiChromaQpIndexOffset);

+      goto TRY_REENCODING;

+    }

     if (ENC_RETURN_SUCCESS != iEncReturn)

       return iEncReturn;

@@ -505,7 +532,6 @@

     pEncCtx->pFuncList->pfRc.pfWelsRcMbInfoUpdate (pEncCtx, pCurMb, sMd.iCostLuma, pSlice);

     ++iNumMbCoded;

     iNextMbIdx = WelsGetNextMbOfSlice (pSliceCtx, iCurMbIdx);

     if (iNextMbIdx == -1 || iNextMbIdx >= kiTotalNumMb || iNumMbCoded >= kiTotalNumMb) {

       break;

@@ -540,6 +566,8 @@

   for (; ;) {

     iCurMbIdx	= iNextMbIdx;

     pCurMb = &pMbList[ iCurMbIdx ];

+    StashMBStatus (&sDss, pBs, pSlice);

     pEncCtx->pFuncList->pfRc.pfWelsRcMbInit (pEncCtx, pCurMb, pSlice);

     // if already reaches the largest number of slices, set QPs to the upper bound

     if (pSlice->bDynamicSlicingSliceSizeCtrlFlag) {

@@ -546,18 +574,19 @@

       pCurMb->uiLumaQp = pEncCtx->pWelsSvcRc[pEncCtx->uiDependencyId].iMaxQp;

       pCurMb->uiChromaQp = g_kuiChromaQpTable[CLIP3_QP_0_51 (pCurMb->uiLumaQp + kuiChromaQpIndexOffset)];

+    WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);

+TRY_REENCODING:

     sMd.iLambda = g_kiQpCostTable[pCurMb->uiLumaQp];

-    WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);

     WelsMdIntraMb (pEncCtx, &sMd, pCurMb, pMbCache);

     UpdateNonZeroCountCache (pCurMb, pMbCache);

-    //stack pBs pointer

-    sDss.pBsStackBufPtr	= pBs->pBufPtr;

-    sDss.uiBsStackCurBits	= pBs->uiCurBits;

-    sDss.iBsStackLeftBits	= pBs->iLeftBits;

     iEncReturn = WelsSpatialWriteMbSyn (pEncCtx, pSlice, pCurMb);

+    if (iEncReturn == ENC_RETURN_VLCOVERFLOWFOUND) {

+      StashPopMBStatus (&sDss, pBs, pSlice);

+      UpdateQpForOverflow (pCurMb, kuiChromaQpIndexOffset);

+      goto TRY_REENCODING;

+    }

     if (ENC_RETURN_SUCCESS != iEncReturn)

       return iEncReturn;

@@ -926,12 +955,14 @@

   const int32_t kiSliceIdx				= pSlice->uiSliceIdx;

   const uint8_t kuiChromaQpIndexOffset = pCurLayer->sLayerInfo.pPpsP->uiChromaQpIndexOffset;

   int32_t iEncReturn = ENC_RETURN_SUCCESS;

+  SDynamicSlicingStack sDss;

   for (;;) {

+    StashMBStatus (&sDss, pBs, pSlice, iMbSkipRun);

     //point to current pMb

     iCurMbIdx	= iNextMbIdx;

     pCurMb = &pMbList[ iCurMbIdx ];

     //step(1): set QP for the current MB

     pEncCtx->pFuncList->pfRc.pfWelsRcMbInit (pEncCtx, pCurMb, pSlice);

@@ -938,6 +969,8 @@

     //step (2). save some vale for future use, initial pWelsMd

     WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);

     WelsMdInterInit (pEncCtx, pSlice, pCurMb, kiSliceFirstMbXY);

+TRY_REENCODING:

     WelsInitInterMDStruc (pCurMb, pMvdCostTableInter, kiMvdInterTableStride, pMd);

     pEncCtx->pFuncList->pfInterMd (pEncCtx, pMd, pSlice, pCurMb, pMbCache);

     //mb_qp

@@ -961,6 +994,11 @@

       BsWriteUE (pBs, iMbSkipRun);

       iMbSkipRun = 0;

       iEncReturn = WelsSpatialWriteMbSyn (pEncCtx, pSlice, pCurMb);

+      if (iEncReturn == ENC_RETURN_VLCOVERFLOWFOUND) {

+        StashPopMBStatus (&sDss, pBs, pSlice, &iMbSkipRun);

+        UpdateQpForOverflow (pCurMb, kuiChromaQpIndexOffset);

+        goto TRY_REENCODING;

+      }

       if (ENC_RETURN_SUCCESS != iEncReturn)

         return iEncReturn;

@@ -1018,6 +1056,10 @@

   SDynamicSlicingStack sDss;

   sDss.iStartPos = BsGetBitsPos (pBs);

   for (;;) {

+    //DYNAMIC_SLICING_ONE_THREAD - MultiD

+    //stack pBs pointer

+    StashMBStatus (&sDss, pBs, pSlice, iMbSkipRun);

     //point to current pMb

     iCurMbIdx	= iNextMbIdx;

     pCurMb = &pMbList[ iCurMbIdx ];

@@ -1036,6 +1078,8 @@

     //step (2). save some vale for future use, initial pWelsMd

     WelsMdIntraInit (pEncCtx, pCurMb, pMbCache, kiSliceFirstMbXY);

     WelsMdInterInit (pEncCtx, pSlice, pCurMb, kiSliceFirstMbXY);

+TRY_REENCODING:

     WelsInitInterMDStruc (pCurMb, pMvdCostTableInter, kiMvdInterTableStride, pMd);

     pEncCtx->pFuncList->pfInterMd (pEncCtx, pMd, pSlice, pCurMb, pMbCache);

     //mb_qp

@@ -1051,15 +1095,6 @@

     //step (6): begin to write bit stream; if the pSlice size is controlled, the writing may be skipped

-    //DYNAMIC_SLICING_ONE_THREAD - MultiD

-    //stack pBs pointer

-    sDss.pBsStackBufPtr	= pBs->pBufPtr;

-    sDss.uiBsStackCurBits	= pBs->uiCurBits;

-    sDss.iBsStackLeftBits	= pBs->iLeftBits;

-    //stack Pskip status

-    sDss.iMbSkipRunStack = iMbSkipRun;

-    //DYNAMIC_SLICING_ONE_THREAD - MultiD

     if (IS_SKIP (pCurMb->uiMbType)) {

       pCurMb->uiLumaQp	= pSlice->uiLastMbQp;

       pCurMb->uiChromaQp = g_kuiChromaQpTable[CLIP3_QP_0_51 (pCurMb->uiLumaQp + kuiChromaQpIndexOffset)];

@@ -1069,6 +1104,11 @@

       BsWriteUE (pBs, iMbSkipRun);

       iMbSkipRun = 0;

       iEncReturn = WelsSpatialWriteMbSyn (pEncCtx, pSlice, pCurMb);

+      if (iEncReturn == ENC_RETURN_VLCOVERFLOWFOUND) {

+        StashPopMBStatus (&sDss, pBs, pSlice, &iMbSkipRun);

+        UpdateQpForOverflow (pCurMb, kuiChromaQpIndexOffset);

+        goto TRY_REENCODING;

+      }

       if (ENC_RETURN_SUCCESS != iEncReturn)

         return iEncReturn;

--- a/codec/encoder/core/src/svc_set_mb_syn_cavlc.cpp

+++ b/codec/encoder/core/src/svc_set_mb_syn_cavlc.cpp

@@ -244,7 +244,8 @@

     pSlice->uiLastMbQp = pCurMb->uiLumaQp;

     BsWriteSE (pBs, kiDeltaQp);

-    WelsWriteMbResidual (pMbCache, pCurMb, pBs);

+    if (WelsWriteMbResidual (pMbCache, pCurMb, pBs))

+      return ENC_RETURN_VLCOVERFLOWFOUND;

   } else {

     pCurMb->uiLumaQp = pSlice->uiLastMbQp;

     pCurMb->uiChromaQp = g_kuiChromaQpTable[CLIP3_QP_0_51 (pCurMb->uiLumaQp +

@@ -255,7 +256,7 @@

   return CheckBitstreamBuffer (pSlice->uiSliceIdx, pEncCtx, pBs);

-void WelsWriteMbResidual (SMbCache* sMbCacheInfo, SMB* pCurMb, SBitStringAux* pBs) {

+int32_t WelsWriteMbResidual (SMbCache* sMbCacheInfo, SMB* pCurMb, SBitStringAux* pBs) {

   int32_t i;

   Mb_Type uiMbType					= pCurMb->uiMbType;

   const int32_t kiCbpChroma		= pCurMb->uiCbp >> 4;

@@ -269,7 +270,8 @@

     iA = pNonZeroCoeffCount[8];

     iB = pNonZeroCoeffCount[ 1];

     WELS_NON_ZERO_COUNT_AVERAGE (iC, iA, iB);

-    WriteBlockResidualCavlc (sMbCacheInfo->pDct->iLumaI16x16Dc, 15, 1, LUMA_4x4, iC, pBs);

+    if (WriteBlockResidualCavlc (sMbCacheInfo->pDct->iLumaI16x16Dc, 15, 1, LUMA_4x4, iC, pBs))

+      return ENC_RETURN_VLCOVERFLOWFOUND;

     /* AC Luma */

     if (kiCbpLuma) {

@@ -324,10 +326,12 @@

   if (kiCbpChroma) {

     /* Chroma DC residual present */

     pBlock = sMbCacheInfo->pDct->iChromaDc[0]; // Cb

-    WriteBlockResidualCavlc (pBlock, 3, 1, CHROMA_DC, CHROMA_DC_NC_OFFSET, pBs);

+    if (WriteBlockResidualCavlc (pBlock, 3, 1, CHROMA_DC, CHROMA_DC_NC_OFFSET, pBs))

+      return ENC_RETURN_VLCOVERFLOWFOUND;

     pBlock += 4; // Cr

-    WriteBlockResidualCavlc (pBlock, 3, 1, CHROMA_DC, CHROMA_DC_NC_OFFSET, pBs);

+    if (WriteBlockResidualCavlc (pBlock, 3, 1, CHROMA_DC, CHROMA_DC_NC_OFFSET, pBs))

+      return ENC_RETURN_VLCOVERFLOWFOUND;

     /* Chroma AC residual present */

     if (kiCbpChroma & 0x02) {

@@ -355,6 +359,7 @@

+  return 0;

 } // namespace WelsSVCEnc

--- a/codec/processing/src/common/util.h

+++ b/codec/processing/src/common/util.h

@@ -67,8 +67,15 @@

 #define WELS_MAX(x, y)	((x) > (y) ? (x) : (y))

 #define WELS_MIN(x, y)	((x) < (y) ? (x) : (y))

+#ifndef WELS_SIGN

 #define WELS_SIGN(a)	((int32_t)(a) >> 31)

+#endif

+#ifndef WELS_ABS

 #define WELS_ABS(a)		((WELS_SIGN(a) ^ (int32_t)(a)) - WELS_SIGN(a))

+#endif

 #define WELS_CLAMP(x, minv, maxv)  WELS_MIN(WELS_MAX(x, minv), maxv)

 #define ALIGNBYTES         (16)       /* Worst case is requiring alignment to an 16 byte boundary */