ref: b160c198d7b298fb012774dea582cd3ad5edb6e4
parent: fcf345c20eed7664cce50d9fdd03b88ef093b4aa
author: huili2 <huili2@cisco.com>
date: Fri Apr 28 05:36:59 EDT 2017
fix bug with ihadamard to correct place with qmatrix on
--- a/codec/decoder/core/src/decode_slice.cpp
+++ b/codec/decoder/core/src/decode_slice.cpp
@@ -155,9 +155,6 @@
int32_t iMbXy = pCurLayer->iMbXyIndex;
int32_t i, iIndex, iOffset;
- WelsChromaDcIdct (pCurLayer->pScaledTCoeff[iMbXy] + 256); // 256 = 16*16
- WelsChromaDcIdct (pCurLayer->pScaledTCoeff[iMbXy] + 320); // 320 = 16*16 + 16*4
-
if (pCurLayer->pTransformSize8x8Flag[iMbXy]) { for (i = 0; i < 4; i++) {iIndex = g_kuiMbCountScan4Idx[i << 2];
@@ -207,7 +204,7 @@
}
void WelsLumaDcDequantIdct (int16_t* pBlock, int32_t iQp, PWelsDecoderContext pCtx) {- const int32_t kiQMul = pCtx->bUseScalingList ? pCtx->pDequant_coeff4x4[0][iQp][0] >> 4 : g_kuiDequantCoeff[iQp][0];
+ const int32_t kiQMul = pCtx->bUseScalingList ? pCtx->pDequant_coeff4x4[0][iQp][0] : (g_kuiDequantCoeff[iQp][0] << 4);
#define STRIDE 16
int32_t i;
int32_t iTemp[16]; //FIXME check if this is a good idea
@@ -240,10 +237,10 @@
const int32_t kiZ2 = iTemp[kiI4] - iTemp[8 + kiI4];
const int32_t kiZ3 = iTemp[kiI4] + iTemp[8 + kiI4];
- pBlk[kiOffset] = ((kiZ0 + kiZ3) * kiQMul + 2) >> 2; //FIXME think about merging this into decode_resdual
- pBlk[kiYOffset[1] + kiOffset] = ((kiZ1 + kiZ2) * kiQMul + 2) >> 2;
- pBlk[kiYOffset[2] + kiOffset] = ((kiZ1 - kiZ2) * kiQMul + 2) >> 2;
- pBlk[kiYOffset[3] + kiOffset] = ((kiZ0 - kiZ3) * kiQMul + 2) >> 2;
+ pBlk[kiOffset] = ((kiZ0 + kiZ3) * kiQMul + (1 << 5)) >> 6; //FIXME think about merging this into decode_resdual
+ pBlk[kiYOffset[1] + kiOffset] = ((kiZ1 + kiZ2) * kiQMul + (1 << 5)) >> 6;
+ pBlk[kiYOffset[2] + kiOffset] = ((kiZ1 - kiZ2) * kiQMul + (1 << 5)) >> 6;
+ pBlk[kiYOffset[3] + kiOffset] = ((kiZ0 - kiZ3) * kiQMul + (1 << 5)) >> 6;
}
#undef STRIDE
}
@@ -255,19 +252,12 @@
WelsFillRecNeededMbInfo (pCtx, bOutput, pCurLayer);
if (IS_INTRA16x16 (pCurLayer->pMbType[iMbXy])) {- WelsLumaDcDequantIdct (pCurLayer->pScaledTCoeff[iMbXy], pCurLayer->pLumaQp[iMbXy], pCtx);
RecI16x16Mb (iMbXy, pCtx, pCurLayer->pScaledTCoeff[iMbXy], pCurLayer);
-
- return ERR_NONE;
- }
-
- if (IS_INTRA8x8 (pCurLayer->pMbType[iMbXy])) {+ } else if (IS_INTRA8x8 (pCurLayer->pMbType[iMbXy])) {RecI8x8Mb (iMbXy, pCtx, pCurLayer->pScaledTCoeff[iMbXy], pCurLayer);
- }
-
- if (IS_INTRA4x4 (pCurLayer->pMbType[iMbXy]))
+ } else if (IS_INTRA4x4 (pCurLayer->pMbType[iMbXy])) {RecI4x4Mb (iMbXy, pCtx, pCurLayer->pScaledTCoeff[iMbXy], pCurLayer);
-
+ }
return ERR_NONE;
}
@@ -327,10 +317,10 @@
iB = iC - iD;
iC += iD;
- pBlk[0] = (iA + iC) >> 1;
- pBlk[iXStride] = (iE + iB) >> 1;
- pBlk[iStride] = (iA - iC) >> 1;
- pBlk[iStride1] = (iE - iB) >> 1;
+ pBlk[0] = (iA + iC);
+ pBlk[iXStride] = (iE + iB);
+ pBlk[iStride] = (iA - iC);
+ pBlk[iStride1] = (iE - iB);
}
void WelsMapNxNNeighToSampleNormal (PWelsNeighAvail pNeighAvail, int32_t* pSampleAvail) {--- a/codec/decoder/core/src/parse_mb_syn_cabac.cpp
+++ b/codec/decoder/core/src/parse_mb_syn_cabac.cpp
@@ -31,6 +31,7 @@
* parse_mb_syn_cabac.cpp: cabac parse for syntax elements
*/
#include "parse_mb_syn_cabac.h"
+#include "decode_slice.h"
#include "mv_pred.h"
#include "error_code.h"
namespace WelsDec {@@ -927,24 +928,36 @@
int32_t j = 0;
if (iResProperty == I16_LUMA_DC) { do {- if (pSignificantMap[j] != 0)
- sTCoeff[pScanTable[j]] = pSignificantMap[j];
+ sTCoeff[pScanTable[j]] = pSignificantMap[j];
++j;
} while (j < 16);
+ WelsLumaDcDequantIdct (sTCoeff, uiQp, pCtx);
} else if (iResProperty == CHROMA_DC_U || iResProperty == CHROMA_DC_V) { do {- if (pSignificantMap[j] != 0)
- sTCoeff[pScanTable[j]] = pCtx->bUseScalingList ? (int16_t) ((int64_t)pSignificantMap[j] *
- (int64_t)pDeQuantMul[0] >> 4) :
- (pSignificantMap[j] * pDeQuantMul[0]);
+ sTCoeff[pScanTable[j]] = pSignificantMap[j];
++j;
- } while (j < 16);
+ } while (j < 4);
+ //iHadamard2x2
+ WelsChromaDcIdct (sTCoeff);
+ //scaling
+ if (!pCtx->bUseScalingList) {+ for (j = 0; j < 4; ++j) {+ sTCoeff[pScanTable[j]] = (int16_t) ((int64_t)sTCoeff[pScanTable[j]] * (int64_t)pDeQuantMul[0] >> 1);
+ }
+ } else { //with scaling list+ for (j = 0; j < 4; ++j) {+ sTCoeff[pScanTable[j]] = (int16_t) ((int64_t)sTCoeff[pScanTable[j]] * (int64_t)pDeQuantMul[0] >> 5);
+ }
+ }
} else { //luma ac, chroma ac do {- if (pSignificantMap[j] != 0)
- sTCoeff[pScanTable[j]] = pCtx->bUseScalingList ? (int16_t) ((int64_t)pSignificantMap[j] *
- (int64_t)pDeQuantMul[pScanTable[j]] >> 4) :
- pSignificantMap[j] * pDeQuantMul[pScanTable[j] & 0x07];
+ if (pSignificantMap[j] != 0) {+ if (!pCtx->bUseScalingList) {+ sTCoeff[pScanTable[j]] = pSignificantMap[j] * pDeQuantMul[pScanTable[j] & 0x07];
+ } else {+ sTCoeff[pScanTable[j]] = (int16_t) (((int64_t)pSignificantMap[j] * (int64_t)pDeQuantMul[pScanTable[j]] + 8) >> 4);
+ }
+ }
++j;
} while (j < 16);
}
--- a/codec/decoder/core/src/parse_mb_syn_cavlc.cpp
+++ b/codec/decoder/core/src/parse_mb_syn_cavlc.cpp
@@ -40,6 +40,7 @@
#include "parse_mb_syn_cavlc.h"
+#include "decode_slice.h"
#include "error_code.h"
#include "mv_pred.h"
@@ -880,8 +881,19 @@
int32_t j;
iCoeffNum += iRun[i] + 1; //FIXME add 1 earlier ?
j = kpZigzagTable[ iCoeffNum ];
- pTCoeff[j] = pCtx->bUseScalingList ? (iLevel[i] * kpDequantCoeff[0]) >> 4 : (iLevel[i] * kpDequantCoeff[0]);
+ pTCoeff[j] = iLevel[i];
}
+ WelsChromaDcIdct (pTCoeff);
+ //scaling
+ if (!pCtx->bUseScalingList) {+ for (int j = 0; j < 4; ++j) {+ pTCoeff[kpZigzagTable[j]] = (pTCoeff[kpZigzagTable[j]] * kpDequantCoeff[0]) >> 1;
+ }
+ } else {+ for (int j = 0; j < 4; ++j) {+ pTCoeff[kpZigzagTable[j]] = ((int64_t) pTCoeff[kpZigzagTable[j]] * (int64_t) kpDequantCoeff[0]) >> 5;
+ }
+ }
} else if (iResidualProperty == I16_LUMA_DC) { //DC coefficent, only call in Intra_16x16, base_mode_flag = 0 for (i = uiTotalCoeff - 1; i >= 0; --i) { //FIXME merge into rundecode?int32_t j;
@@ -889,12 +901,17 @@
j = kpZigzagTable[ iCoeffNum ];
pTCoeff[j] = iLevel[i];
}
+ WelsLumaDcDequantIdct (pTCoeff, uiQp, pCtx);
} else { for (i = uiTotalCoeff - 1; i >= 0; --i) { //FIXME merge into rundecode?int32_t j;
iCoeffNum += iRun[i] + 1; //FIXME add 1 earlier ?
j = kpZigzagTable[ iCoeffNum ];
- pTCoeff[j] = pCtx->bUseScalingList ? (iLevel[i] * kpDequantCoeff[j]) >> 4 : (iLevel[i] * kpDequantCoeff[j & 0x07]);
+ if (!pCtx->bUseScalingList) {+ pTCoeff[j] = (iLevel[i] * kpDequantCoeff[j & 0x07]);
+ } else {+ pTCoeff[j] = (iLevel[i] * kpDequantCoeff[j] + 8) >> 4;
+ }
}
}
--- a/codec/decoder/core/src/rec_mb.cpp
+++ b/codec/decoder/core/src/rec_mb.cpp
@@ -237,8 +237,10 @@
int32_t iBlkWidth, int32_t iBlkHeight, int16_t iMVs[2]) {int32_t iFullMVx = (iXOffset << 2) + iMVs[0]; //quarter pixel
int32_t iFullMVy = (iYOffset << 2) + iMVs[1];
- iFullMVx = WELS_CLIP3 (iFullMVx, ((-PADDING_LENGTH + 2) * (1 << 2)), ((pMCRefMem->iPicWidth + PADDING_LENGTH - 19) * (1 << 2)));
- iFullMVy = WELS_CLIP3 (iFullMVy, ((-PADDING_LENGTH + 2) * (1 << 2)), ((pMCRefMem->iPicHeight + PADDING_LENGTH - 19) * (1 << 2)));
+ iFullMVx = WELS_CLIP3 (iFullMVx, ((-PADDING_LENGTH + 2) * (1 << 2)),
+ ((pMCRefMem->iPicWidth + PADDING_LENGTH - 19) * (1 << 2)));
+ iFullMVy = WELS_CLIP3 (iFullMVy, ((-PADDING_LENGTH + 2) * (1 << 2)),
+ ((pMCRefMem->iPicHeight + PADDING_LENGTH - 19) * (1 << 2)));
int32_t iSrcPixOffsetLuma = (iFullMVx >> 2) + (iFullMVy >> 2) * pMCRefMem->iSrcLineLuma;
int32_t iSrcPixOffsetChroma = (iFullMVx >> 3) + (iFullMVy >> 3) * pMCRefMem->iSrcLineChroma;
@@ -541,8 +543,6 @@
uint8_t uiCbpC = pDqLayer->pCbp[iMBXY] >> 4;
if (1 == uiCbpC || 2 == uiCbpC) {- WelsChromaDcIdct (pScoeffLevel + 256); // 256 = 16*16
- WelsChromaDcIdct (pScoeffLevel + 320); // 256 = 16*16
for (i = 0; i < 2; i++) {int16_t* pRS = pScoeffLevel + 256 + (i << 6);
uint8_t* pPred = pDqLayer->pPred[i + 1];