Sweden-Number/libs/jxr/image/encode/segenc.c

1187 lines
42 KiB
C

//*@@@+++@@@@******************************************************************
//
// Copyright © Microsoft Corp.
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// • Redistributions of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
// • Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
//*@@@---@@@@******************************************************************
#include <stdio.h>
#include <stdlib.h>
#include "strcodec.h"
#include "encode.h"
#ifdef MEM_TRACE
#define TRACE_MALLOC 1
#define TRACE_NEW 0
#define TRACE_HEAP 0
#include "memtrace.h"
#endif
/** local function definitions **/
#ifdef X86OPT_INLINE
__forceinline
#endif
static Int EncodeBlock (Bool bChroma, const Int *aLocalCoef, Int iNumNonzero,
struct CAdaptiveHuffman **pAHexpt,
Int iContextOffset, BitIOInfo* pOut, UInt iLocation);
/*************************************************************************
EncodeSignificantAbsLevel
*************************************************************************/
#ifdef X86OPT_INLINE
//__forceinline
#endif
static Void EncodeSignificantAbsLevel (UInt iAbsLevel, struct CAdaptiveHuffman *pAHexpt, BitIOInfo* pOut)
{
Int iIndex, iFixed, aIndex[] = { 0,1,2,2, 3,3,3,3, 4,4,4,4, 5,5,5,5 };
Int aFixedLength[] = { 0, 0, 1, 2, 2, 2 };
assert(iAbsLevel > 0);
iAbsLevel--;
if (iAbsLevel >= 16) {
Int i = iAbsLevel;
iIndex = 6;
/** find leftmost bit **/
i >>= 5;
iFixed = 4;
while (i) { /** caution - infinite loop if not careful **/
iFixed++;
assert (iFixed < 30);
i >>= 1;
}
pAHexpt->m_iDiscriminant += pAHexpt->m_pDelta[iIndex];
putBit16z(pOut, pAHexpt->m_pTable[iIndex * 2 + 1], pAHexpt->m_pTable[iIndex * 2 + 2]);
if (iFixed > 18) {
putBit16z (pOut, 15, 4);
if (iFixed > 21) {
putBit16z (pOut, 3, 2);
putBit16 (pOut, iFixed - 22, 3); // 22 - 29
}
else
putBit16z (pOut, iFixed - 19, 2); // 19 20 21
}
else {
putBit16z(pOut, (iFixed - 4), 4);
}
putBit32(pOut, iAbsLevel, iFixed);
}
else {
iIndex = aIndex[iAbsLevel];
iFixed = aFixedLength[iIndex];
pAHexpt->m_iDiscriminant += pAHexpt->m_pDelta[iIndex];
putBit16z(pOut, pAHexpt->m_pTable[iIndex * 2 + 1], pAHexpt->m_pTable[iIndex * 2 + 2]);
putBit32(pOut, iAbsLevel, iFixed);
}
}
/*************************************************************************
EncodeMacroblockDC
*************************************************************************/
Void encodeQPIndex(BitIOInfo* pIO, U8 iIndex,U8 cBits)
{
if(iIndex == 0)
putBit16z(pIO, 0, 1);
else{
putBit16z(pIO, 1, 1);
putBit16z(pIO, iIndex - 1, cBits);
}
}
Int EncodeMacroblockDC (CWMImageStrCodec *pSC, CCodingContext *pContext, Int iMBX, Int iMBY)
{
CWMITile * pTile = pSC->pTile + pSC->cTileColumn;
BitIOInfo* pIO = pContext->m_pIODC;
CWMIMBInfo *pMBInfo = &pSC->MBInfo;
Int iIndex, j = 0;
struct CAdaptiveHuffman *pAH;
Int aLaplacianMean[2] = { 0, 0}, *pLM = aLaplacianMean;
Int iModelBits = pContext->m_aModelDC.m_iFlcBits[0];
COLORFORMAT cf = pSC->m_param.cfColorFormat;
const Int iChannels = (Int) pSC->m_param.cNumChannels;
UNREFERENCED_PARAMETER( iMBX );
UNREFERENCED_PARAMETER( iMBY );
writeIS_L1(pSC, pIO);
if(pSC->m_param.bTranscode == FALSE){
pMBInfo->iQIndexLP = (U8)(pTile->cNumQPLP > 1 ? (rand() % pTile->cNumQPLP) : 0);
pMBInfo->iQIndexHP = (U8)(pTile->cNumQPHP > 1 ? (rand() % pTile->cNumQPHP) : 0);
}
if(pTile->cBitsHP == 0 && pTile->cNumQPHP > 1) // use LP QP
pMBInfo->iQIndexHP = pMBInfo->iQIndexLP;
if(pSC->WMISCP.bfBitstreamFormat == SPATIAL && pSC->WMISCP.sbSubband != SB_DC_ONLY){
if(pTile->cBitsLP > 0) // MB-based LP QP index
encodeQPIndex(pIO, pMBInfo->iQIndexLP, pTile->cBitsLP);
if( pSC->WMISCP.sbSubband != SB_NO_HIGHPASS && pTile->cBitsHP > 0) // MB-based HP QP index
encodeQPIndex(pIO, pMBInfo->iQIndexHP, pTile->cBitsHP);
}
if(pSC->m_param.bTranscode == FALSE)
pSC->Quantize(pSC);
predMacroblockEnc(pSC);
/** code path for Y_ONLY, CMYK and N_CHANNEL DC **/
if(cf == Y_ONLY || cf == CMYK || cf == NCOMPONENT) {
Int iQDC, iDC, iSign;
for (j = 0; j < iChannels; j++) {
iDC = pMBInfo->iBlockDC[j][0];
iSign = (iDC < 0);
iDC = abs(iDC);
iQDC = iDC >> iModelBits;
/** send luminance DC **/
if (iQDC) {
putBit16z(pIO, 1, 1);
EncodeSignificantAbsLevel((UInt) iQDC, pContext->m_pAHexpt[3], pIO);
*pLM += 1;
}
else {
putBit16z(pIO, 0, 1);
}
putBit16(pIO, iDC, iModelBits);
if (iDC) {
putBit16z(pIO, iSign, 1);
}
pLM = aLaplacianMean + 1;
iModelBits = pContext->m_aModelDC.m_iFlcBits[1];
}
}
else { /** code path for YUV DC **/
Int iDCY, iDCU, iDCV, iQDCY, iQDCU, iQDCV;
pAH = pContext->m_pAHexpt[2];
iQDCY = abs(iDCY = pMBInfo->iBlockDC[0][0]);
iQDCU = abs(iDCU = pMBInfo->iBlockDC[1][0]);
iQDCV = abs(iDCV = pMBInfo->iBlockDC[2][0]);
if (iModelBits) {
iQDCY >>= iModelBits;
}
iModelBits = pContext->m_aModelDC.m_iFlcBits[1];
if (iModelBits) {
iQDCU >>= iModelBits;
iQDCV >>= iModelBits;
}
iModelBits = pContext->m_aModelDC.m_iFlcBits[0];
iIndex = (iQDCY != 0) * 4 + (iQDCU != 0) * 2 + (iQDCV != 0);
putBit16z(pIO, pAH->m_pTable[iIndex * 2 + 1], pAH->m_pTable[iIndex * 2 + 2]);
/** send luminance DC **/
if (iQDCY) {
EncodeSignificantAbsLevel((UInt) iQDCY, pContext->m_pAHexpt[3], pIO);
*pLM += 1;
}
putBit16(pIO, abs(iDCY), iModelBits);
if (iDCY) {
putBit16z(pIO, (iDCY < 0), 1);
}
/** send chroma DC **/
pLM = aLaplacianMean + 1;
iModelBits = pContext->m_aModelDC.m_iFlcBits[1];
if (iQDCU) {
EncodeSignificantAbsLevel((UInt) iQDCU, pContext->m_pAHexpt[4], pIO);
*pLM += 1;
}
putBit16(pIO, abs(iDCU), iModelBits);
if (iDCU) {
putBit16z(pIO, (iDCU < 0), 1);
}
if (iQDCV) {
EncodeSignificantAbsLevel((UInt) iQDCV, pContext->m_pAHexpt[4], pIO);
*pLM += 1;
}
putBit16(pIO, abs(iDCV), iModelBits);
if (iDCV) {
putBit16z(pIO, (iDCV < 0), 1);
}
}
UpdateModelMB (cf, iChannels, aLaplacianMean, &(pContext->m_aModelDC));
if (pSC->m_bResetContext && pSC->WMISCP.sbSubband == SB_DC_ONLY) {
AdaptDiscriminant(pContext->m_pAHexpt[2]);
AdaptDiscriminant(pContext->m_pAHexpt[3]);
AdaptDiscriminant(pContext->m_pAHexpt[4]);
}
return ICERR_OK;
}
/*************************************************************************
Scan block with zero model bits
*************************************************************************/
#ifdef X86OPT_INLINE
__forceinline
#endif
static Int AdaptiveScanZero (const PixelI *pCoeffs, CAdaptiveScan *pScan,
Int *pRLCoeffs, const Int iCount)
{
Int k, iRun = 1, iLevel, iNumNonzero = 0;
iLevel = pCoeffs[pScan[1].uScan];
if (iLevel) {
pScan[1].uTotal++;
pRLCoeffs[iNumNonzero * 2] = 0;
pRLCoeffs[iNumNonzero * 2 + 1] = iLevel;
iNumNonzero++;
iRun = 0;
}
for (k = 2; k < iCount; k++) {
iLevel = pCoeffs[pScan[k].uScan];
iRun++;
if (iLevel) {
pScan[k].uTotal++;
if (pScan[k].uTotal > pScan[k - 1].uTotal) {
CAdaptiveScan cTemp = pScan[k];
pScan[k] = pScan[k - 1];
pScan[k - 1] = cTemp;
}
pRLCoeffs[iNumNonzero * 2] = iRun - 1;
pRLCoeffs[iNumNonzero * 2 + 1] = iLevel;
iNumNonzero++;
iRun = 0;
}
}
return iNumNonzero;
}
/*************************************************************************
Scan block with nonzero model bits, all trimmed
*************************************************************************/
#ifdef X86OPT_INLINE
__forceinline
#endif
static Int AdaptiveScanTrim (const PixelI *pCoeffs, CAdaptiveScan *pScan,
const Int iModelBits, Int *pRLCoeffs, const Int iCount)
{
Int k, iRun = 1, iLevel, iNumNonzero = 0;
Int iTemp;
unsigned int iThOff = (1 << iModelBits) - 1, iTh = iThOff * 2 + 1;
iLevel = pCoeffs[pScan[1].uScan];
if ((unsigned int)(iLevel + iThOff) >= iTh) {
iTemp = abs (iLevel) >> iModelBits;
pScan[1].uTotal++;
pRLCoeffs[iNumNonzero * 2] = 0;
pRLCoeffs[iNumNonzero * 2 + 1] = (iLevel < 0) ? -iTemp : iTemp;
iNumNonzero++;
iRun = 0;
}
for (k = 2; k < iCount; k++) {
iRun++;
iLevel = pCoeffs[pScan[k].uScan];
if ((unsigned int)(iLevel + iThOff) >= iTh) {
iTemp = abs (iLevel) >> iModelBits;
pScan[k].uTotal++;
if (pScan[k].uTotal > pScan[k - 1].uTotal) {
CAdaptiveScan cTemp = pScan[k];
pScan[k] = pScan[k - 1];
pScan[k - 1] = cTemp;
}
pRLCoeffs[iNumNonzero * 2] = iRun - 1;
pRLCoeffs[iNumNonzero * 2 + 1] = (iLevel < 0) ? -iTemp : iTemp;
iNumNonzero++;
iRun = 0;
}
}
return iNumNonzero;
}
/*************************************************************************
Scan block with nonzero model bits
*************************************************************************/
/** saves around 1.5% at QP=1 (no SIMD opt) **/
#define USE_GRES_LUT
#ifdef USE_GRES_LUT
static const Int gRes[] = {
65*2+1, 63*2+1, 61*2+1, 59*2+1, 57*2+1, 55*2+1, 53*2+1, 51*2+1, 49*2+1, 47*2+1, 45*2+1, 43*2+1, 41*2+1,
39*2+1, 37*2+1, 35*2+1, 33*2+1, 31*2+1, 29*2+1, 27*2+1, 25*2+1, 23*2+1, 21*2+1, 19*2+1, 17*2+1, 15*2+1,
13*2+1, 11*2+1, 9*2+1, 7*2+1, 5*2+1, 3*2+1,
0,
2*2+1, 4*2+1, 6*2+1, 8*2+1, 10*2+1, 12*2+1, 14*2+1, 16*2+1, 18*2+1, 20*2+1, 22*2+1, 24*2+1,
26*2+1, 28*2+1, 30*2+1, 32*2+1, 34*2+1, 36*2+1, 38*2+1, 40*2+1, 42*2+1, 44*2+1, 46*2+1, 48*2+1, 50*2+1,
52*2+1, 54*2+1, 56*2+1, 58*2+1, 60*2+1, 62*2+1, 64*2+1 };
#endif // USE_GRES_LUT
#ifdef X86OPT_INLINE
//__forceinline
#endif
static Int AdaptiveScan (const PixelI *pCoeffs, Int *pResidual,
CAdaptiveScan *pScan,
const Int iModelBits, const Int iTrimBits,
Int *pRLCoeffs, const Int iCount)
{
if (iModelBits == 0) {
return AdaptiveScanZero (pCoeffs, pScan, pRLCoeffs, iCount);
}
else if (iModelBits <= iTrimBits) {
return AdaptiveScanTrim (pCoeffs, pScan, iModelBits, pRLCoeffs, iCount);
}
else if (iTrimBits == 0
#ifdef USE_GRES_LUT
&& iModelBits < 6
#endif // USE_GRES_LUT
) {
Int k, iRun = 0, iLevel, iNumNonzero = 0;
Int iTemp, iTemp1;
const unsigned int iThOff = (1 << iModelBits) - 1, iTh = iThOff * 2 + 1;
iLevel = pCoeffs[pScan[1].uScan];
if ((unsigned int)(iLevel + iThOff) >= iTh) {
iTemp1 = abs (iLevel);
iTemp = iTemp1 >> iModelBits;
pResidual[pScan[1].uScan] = (iTemp1 & iThOff) * 2;
pScan[1].uTotal++;
pRLCoeffs[iNumNonzero * 2] = iRun;
pRLCoeffs[iNumNonzero * 2 + 1] = (iLevel < 0) ? -iTemp : iTemp;
iNumNonzero++;
iRun = 0;
}
else {
iRun++;
#ifdef USE_GRES_LUT
pResidual[pScan[1].uScan] = gRes[(iLevel + 32)];
#else // USE_GRES_LUT
iTemp = -(iLevel < 0);
pResidual[pScan[1].uScan] = (iLevel ^ iTemp) * 4 + (6 & iTemp) + (iLevel != 0);
#endif // USE_GRES_LUT
}
for (k = 2; k < iCount; k++) {
const Int sk = pScan[k].uScan;
//pResidual++;
iLevel = pCoeffs[sk];
if ((unsigned int)(iLevel + iThOff) >= iTh) {
const Int iSign = -(iLevel < 0);
iTemp1 = (iSign ^ iLevel) - iSign;
iTemp = iTemp1 >> iModelBits;
pResidual[sk] = (iTemp1 & iThOff) * 2;
pScan[k].uTotal++;
if (pScan[k].uTotal > pScan[k - 1].uTotal) {
CAdaptiveScan cTemp = pScan[k];
pScan[k] = pScan[k - 1];
pScan[k - 1] = cTemp;
}
pRLCoeffs[iNumNonzero * 2] = iRun;
pRLCoeffs[iNumNonzero * 2 + 1] = (iTemp ^ iSign) - iSign;
iNumNonzero++;
iRun = 0;
}
else {
iRun++;
#ifdef USE_GRES_LUT
pResidual[sk] = gRes[(iLevel + 32)];
#else // USE_GRES_LUT
iTemp = -(iLevel < 0);
pResidual[sk] = (iLevel ^ iTemp) * 4 + (6 & iTemp) + (iLevel != 0);
#endif // USE_GRES_LUT
////(abs(iLevel) * 4) + ((iLevel < 0) * 2) + (iLevel != 0);
}
}
return iNumNonzero;
}
else {
Int k, iRun = 0, iLevel, iNumNonzero = 0;
Int iTemp, iTemp1;
const unsigned int iThOff = (1 << iModelBits) - 1, iTh = iThOff * 2 + 1;
iLevel = pCoeffs[pScan[1].uScan];
//pResidual++;
if ((unsigned int)(iLevel + iThOff) >= iTh) {
iTemp1 = abs (iLevel);
iTemp = iTemp1 >> iModelBits;
pResidual[pScan[1].uScan] = ((iTemp1 & iThOff) >> iTrimBits) * 2;
pScan[1].uTotal++;
pRLCoeffs[iNumNonzero * 2] = iRun;
pRLCoeffs[iNumNonzero * 2 + 1] = (iLevel < 0) ? -iTemp : iTemp;
iNumNonzero++;
iRun = 0;
}
else {
iRun++;
iTemp = -(iLevel < 0);
iLevel = ((iLevel + iTemp) >> iTrimBits) - iTemp; // round towards zero
iTemp = -(iLevel < 0);
pResidual[pScan[1].uScan] = (iLevel ^ iTemp) * 4 + (6 & iTemp) + (iLevel != 0);
}
for (k = 2; k < iCount; k++) {
const Int sk = pScan[k].uScan;
//pResidual++;
iLevel = pCoeffs[sk];
if ((unsigned int)(iLevel + iThOff) >= iTh) {
iTemp1 = abs (iLevel);
iTemp = iTemp1 >> iModelBits;
pResidual[sk] = ((iTemp1 & iThOff) >> iTrimBits) * 2;
pScan[k].uTotal++;
if (pScan[k].uTotal > pScan[k - 1].uTotal) {
CAdaptiveScan cTemp = pScan[k];
pScan[k] = pScan[k - 1];
pScan[k - 1] = cTemp;
}
pRLCoeffs[iNumNonzero * 2] = iRun;
pRLCoeffs[iNumNonzero * 2 + 1] = (iLevel < 0) ? -iTemp : iTemp;
iNumNonzero++;
iRun = 0;
}
else {
iRun++;
iTemp = -(iLevel < 0);
iLevel = ((iLevel + iTemp) >> iTrimBits) - iTemp; // round towards zero
iTemp = -(iLevel < 0);
pResidual[sk] = (iLevel ^ iTemp) * 4 + (6 & iTemp) + (iLevel != 0);
}
}
return iNumNonzero;
}
}
/*************************************************************************
EncodeMacroblockLowpass
*************************************************************************/
Int EncodeMacroblockLowpass (CWMImageStrCodec *pSC, CCodingContext *pContext, Int iMBX, Int iMBY)
{
const COLORFORMAT cf = pSC->m_param.cfColorFormat;
const Int iChannels = (Int) pSC->m_param.cNumChannels;
Int iFullChannels = (cf == YUV_420 || cf == YUV_422) ? 1 : iChannels;
CWMIMBInfo *pMBInfo = &pSC->MBInfo;
BitIOInfo* pIO = pContext->m_pIOLP;
CAdaptiveScan *pScan = pContext->m_aScanLowpass;
Int k, /*iPrevRun = -1,*/ iRun = 0;// iLastIndex = 0;
Int iModelBits = pContext->m_aModelLP.m_iFlcBits[0];
PixelI aBuf[2][8];
Int aLaplacianMean[2] = {0, 0}, *pLM = aLaplacianMean;
Int iChannel, iVal;
Int aRLCoeffs[MAX_CHANNELS][32], iNumCoeffs[MAX_CHANNELS];
const I32 *aDC[MAX_CHANNELS];
Int aResidual[MAX_CHANNELS][16];
Void (*putBits)(BitIOInfo* pIO, U32 uiBits, U32 cBits) = putBit16;
UNREFERENCED_PARAMETER( iMBX );
UNREFERENCED_PARAMETER( iMBY );
if (iChannels > MAX_CHANNELS)
return ICERR_ERROR;
if((pSC->WMISCP.bfBitstreamFormat != SPATIAL) && (pSC->pTile[pSC->cTileColumn].cBitsLP > 0)) // MB-based LP QP index
encodeQPIndex(pIO, pMBInfo->iQIndexLP, pSC->pTile[pSC->cTileColumn].cBitsLP);
// set arrays
for (k = 0; k < iChannels; k++) {
aDC[k] = pMBInfo->iBlockDC[k];
}
/** reset adaptive scan totals **/
if (pSC->m_bResetRGITotals) {
int iScale = 2;
int iWeight = iScale * 16;
pScan[0].uTotal = MAXTOTAL;
for (k = 1; k < 16; k++) {
pScan[k].uTotal = iWeight;
iWeight -= iScale;
}
}
/** scan 4x4 transform **/
for (iChannel = 0; iChannel < iFullChannels; iChannel++) {
iNumCoeffs[iChannel] = AdaptiveScan (aDC[iChannel], aResidual[iChannel],
pScan, iModelBits, 0, aRLCoeffs[iChannel], 16);
iModelBits = pContext->m_aModelLP.m_iFlcBits[1];
}
if (cf == YUV_420 || cf == YUV_422) { /** interleave U and V **/
static const Int aRemap[] = { 4, 1,2,3, 5,6,7 };
const Int *pRemap = aRemap + (cf == YUV_420);
const Int iCount = (cf == YUV_420) ? 6 : 14;
Int iCoef = 0;
iRun = 0;
iModelBits = pContext->m_aModelLP.m_iFlcBits[1];
for (k = 0; k < iCount; k++) {
Int iIndex = pRemap[k >> 1];
Int iDC = aDC[(k & 1) + 1][iIndex];
aBuf[k & 1][iIndex] = iVal = abs (iDC) >> iModelBits;
if (iVal) {
aRLCoeffs[1][iCoef * 2] = iRun;
aRLCoeffs[1][iCoef * 2 + 1] = (iDC < 0) ? -iVal : iVal;
iCoef++;
iRun = 0;
}
else {
iRun++;
}
}
iNumCoeffs[1] = iCoef;
}
/** in raw mode, this can take 6% of the bits in the extreme low rate case!!! **/
if (cf == YUV_420 || cf == YUV_422)
iFullChannels = 2;
if (cf == YUV_420 || cf == YUV_422 || cf == YUV_444) {
int iCBP, iMax = iFullChannels * 4 - 5; /* actually (1 << iNChannels) - 1 **/
int iCountM = pContext->m_iCBPCountMax, iCountZ = pContext->m_iCBPCountZero;
iCBP = (iNumCoeffs[0] > 0) + (iNumCoeffs[1] > 0) * 2;
if (iFullChannels == 3)
iCBP += (iNumCoeffs[2] > 0) * 4;
if (iCountZ <= 0 || iCountM < 0) {
iVal = iCBP;
if (iCountM < iCountZ) {
iVal = iMax - iCBP;
}
if (iVal == 0)
putBit16z(pIO, 0, 1);
else if (iVal == 1)
putBit16z(pIO, (iFullChannels + 1) & 0x6, iFullChannels); // 2 or 4
else
putBit16z(pIO, iVal + iMax + 1, iFullChannels + 1); // cbp + 4 or cbp + 8
}
else {
putBit16z(pIO, iCBP, iFullChannels);
}
iCountM += 1 - 4 * (iCBP == iMax);//(b + c - 2*a);
iCountZ += 1 - 4 * (iCBP == 0);//(a + b - 2*c);
if (iCountM < -8)
iCountM = -8;
else if (iCountM > 7)
iCountM = 7;
pContext->m_iCBPCountMax = iCountM;
if (iCountZ < -8)
iCountZ = -8;
else if (iCountZ > 7)
iCountZ = 7;
pContext->m_iCBPCountZero = iCountZ;
}
else { /** 1 or N channel **/
for (iChannel = 0; iChannel < iChannels; iChannel++) {
putBit16z(pIO, (iNumCoeffs[iChannel] > 0), 1);
}
}
// set appropriate function pointer
if (pContext->m_aModelLP.m_iFlcBits[0] > 14 || pContext->m_aModelLP.m_iFlcBits[1] > 14) {
putBits = putBit32;
}
iModelBits = pContext->m_aModelLP.m_iFlcBits[0];
for (iChannel = 0; iChannel < iFullChannels; iChannel++) {
const Int *pRL = aRLCoeffs[iChannel];
Int iCoef = iNumCoeffs[iChannel];
if (iCoef) {
(*pLM) += iCoef;
if(EncodeBlock (iChannel > 0, pRL, iCoef, pContext->m_pAHexpt, CTDC,
pIO, 1 + 9 * ((cf == YUV_420) && (iChannel == 1)) + ((cf == YUV_422) && (iChannel == 1))) != ICERR_OK)
return ICERR_ERROR;
}
if (iModelBits) {
if ((cf == YUV_420 || cf == YUV_422) && iChannel) { // 420/422 chroma
for (k = 1; k < ((cf == YUV_420) ? 4 : 8); k++) {
putBits(pIO, abs(aDC[1][k]), iModelBits);
if (aBuf[0][k] == 0 && aDC[1][k]) {
putBit16z(pIO, (aDC[1][k] < 0), 1);
}
putBits(pIO, abs(aDC[2][k]), iModelBits);
if (aBuf[1][k] == 0 && aDC[2][k]) {
putBit16z(pIO, (aDC[2][k] < 0), 1);
}
}
}
else { // normal case
for (k = 1; k < 16; k++) {
putBit16z(pIO, aResidual[iChannel][k] >> 1, iModelBits + (aResidual[iChannel][k] & 1));
}
}
}
pLM = aLaplacianMean + 1;
iModelBits = pContext->m_aModelLP.m_iFlcBits[1];
}
writeIS_L1(pSC, pIO);
UpdateModelMB (cf, iChannels, aLaplacianMean, &pContext->m_aModelLP);
if (pSC->m_bResetContext) {
AdaptLowpassEnc(pContext);
}
return ICERR_OK;
}
/*************************************************************************
Adapt
*************************************************************************/
Void AdaptLowpassEnc(CCodingContext *pSC)
{
Int kk;
for (kk = 0; kk < CONTEXTX + CTDC; kk++) { /** adapt fixed code (index 0 and 1) as well **/
AdaptDiscriminant (pSC->m_pAHexpt[kk]);
}
}
Void AdaptHighpassEnc(CCodingContext *pSC)
{
Int kk;
//Adapt (pSC->m_pAdaptHuffCBPCY, FALSE);
AdaptDiscriminant (pSC->m_pAdaptHuffCBPCY);
AdaptDiscriminant (pSC->m_pAdaptHuffCBPCY1);
for (kk = 0; kk < CONTEXTX; kk++) { /** adapt fixed code **/
AdaptDiscriminant (pSC->m_pAHexpt[kk + CONTEXTX + CTDC]);
}
}
/*************************************************************************
Experimental code -- encodeBlock
SR = <0 1 2> == <last, nonsignificant, significant run>
alphabet 12:
pAHexpt[0] == <SR', SL, SR | first symbol>
alphabet 6:
pAHexpt[1] == <SR', SL | continuous>
pAHexpt[2] == <SR', SL | continuous>
alphabet 4:
pAHexpt[3] == <SR', SL | 1 free slot> (SR may be last or insignificant only)
alphabet f(run) (this can be extended to 6 contexts - SL and SR')
pAHexpt[4] == <run | continuous>
alphabet f(lev) (this can be extended to 9 contexts)
pAHexpt[5-6] == <lev | continuous> first symbol
pAHexpt[7-8] == <lev | continuous> condition on SRn no use
*************************************************************************/
#ifdef X86OPT_INLINE
__forceinline
#endif
static Void EncodeSignificantRun (Int iRun, Int iMaxRun, struct CAdaptiveHuffman *pAHexpt, BitIOInfo* pOut)
{
Int iIndex, iFLC, iBin;
static const Int aIndex[] = {
0,1,2,2,3,3,4,4,4,4,4,4,4,4,
0,1,2,2,3,3,4,4,4,4,0,0,0,0,
0,1,2,3,4,4
};
if (iMaxRun < 5) {
//if (iMaxRun == 4) {
//static const Int gCode[] = { 0, 1, 1, 1 };
static const Int gLen[] = { 3, 3, 2, 1 };
if (iMaxRun > 1)
putBit16z(pOut, (iMaxRun != iRun), gLen[iMaxRun - iRun] - (4 - iMaxRun));
//}
//else if (iMaxRun == 3) {
// if (iRun == 1) {
// putBit16z(pOut, 1, 1);
// }
// else {
// putBit16z(pOut, 3 ^ iRun, 2);
// }
//}
//else if (iMaxRun == 2) {
// putBit16z(pOut, 2 - iRun, 1);
//}
return;
}
iBin = gSignificantRunBin[iMaxRun];
iIndex = aIndex[iRun + iBin * 14 - 1];
iFLC = gSignificantRunFixedLength[iIndex + iBin * 5];
putBit16z(pOut, pAHexpt->m_pTable[iIndex * 2 + 1], pAHexpt->m_pTable[iIndex * 2 + 2]);
//this always uses table 0
//pAHexpt->m_iDiscriminant += pAHexpt->m_pDelta[iIndex];
putBit16(pOut, iRun + 1, iFLC);
}
#ifdef X86OPT_INLINE
__forceinline
#endif
static Void EncodeFirstIndex (Bool bChroma, Int iLoc, Int iCont, Int iIndex, Int iSign,
struct CAdaptiveHuffman **ppAHexpt, BitIOInfo* pOut)
{
// Int iContext = iCont + 1 + bChroma * 3;
struct CAdaptiveHuffman *pAHexpt = ppAHexpt[bChroma * 3];
UNREFERENCED_PARAMETER( iLoc );
UNREFERENCED_PARAMETER( iCont );
pAHexpt->m_iDiscriminant += pAHexpt->m_pDelta[iIndex];
pAHexpt->m_iDiscriminant1 += pAHexpt->m_pDelta1[iIndex];
putBit16z(pOut, pAHexpt->m_pTable[iIndex * 2 + 1] * 2 + iSign, pAHexpt->m_pTable[iIndex * 2 + 2] + 1);
return;
}
#ifdef X86OPT_INLINE
__forceinline
#endif
static Void EncodeIndex (Bool bChroma, Int iLoc, Int iCont, Int iIndex, Int iSign,
struct CAdaptiveHuffman **ppAHexpt, BitIOInfo* pOut)
{
Int iContext = iCont + 1 + bChroma * 3;
if (iLoc < 15) {
struct CAdaptiveHuffman *pAHexpt = ppAHexpt[iContext];
pAHexpt->m_iDiscriminant += pAHexpt->m_pDelta[iIndex];
pAHexpt->m_iDiscriminant1 += pAHexpt->m_pDelta1[iIndex];
putBit16z(pOut, pAHexpt->m_pTable[iIndex * 2 + 1] * 2 + iSign, pAHexpt->m_pTable[iIndex * 2 + 2] + 1);
}
else if (iLoc == 15) {
static const U32 gCode[] = { 0, 6, 2, 7 };
static const U32 gLen[] = { 1, 3, 2, 3 };
putBit16z(pOut, gCode[iIndex] * 2 + iSign, gLen[iIndex] + 1);
return;
}
else {//if (iLoc == 16) {
putBit16z(pOut, iIndex * 2 + iSign, 1 + 1);
return;
}
}
#ifdef X86OPT_INLINE
__forceinline
#endif
static Int EncodeBlock (Bool bChroma, const Int *aLocalCoef, Int iNumNonzero,
struct CAdaptiveHuffman **pAHexpt, Int iContextOffset,
BitIOInfo* pOut, UInt iLocation)
{
Int iSR, iSL, iSRn, iIndex, k, iCont, iLev;
/** first symbol **/
iLev = aLocalCoef[1];
iSR = (aLocalCoef[0] == 0);
iSL = ((unsigned int) (iLev + 1) > 2U);
iSRn = 1;
if (iNumNonzero == 1) {
iSRn = 0;
}
else if (aLocalCoef[2] > 0) {
iSRn = 2;
}
iIndex = iSRn * 4 + iSL * 2 + iSR;
EncodeFirstIndex (bChroma, iLocation, 0, iIndex, (iLev < 0), pAHexpt + iContextOffset, pOut);
iCont = iSR & iSRn;
if (iSL) {
EncodeSignificantAbsLevel ((UInt)(abs(iLev) - 1), pAHexpt[6 + iContextOffset + iCont], pOut);
}
if (iSR == 0) {
EncodeSignificantRun (aLocalCoef[0], 15 - iLocation, pAHexpt[0], pOut);
}
iLocation += aLocalCoef[0] + 1;
for (k = 1; k < iNumNonzero; k++) {
if (iSRn == 2) {
EncodeSignificantRun (aLocalCoef[k * 2], 15 - iLocation, pAHexpt[0], pOut);
}
iLocation += aLocalCoef[k * 2] + 1;
iSRn = 1;
if (k == iNumNonzero - 1) {
iSRn = 0;
}
else if (aLocalCoef[k * 2 + 2] > 0) {
iSRn = 2;
}
//iSL = (abs(aLocalCoef[k * 2 + 1]) > 1);
iLev = aLocalCoef[k * 2 + 1];
iSL = ((unsigned int) (iLev + 1) > 2U);
iIndex = iSRn * 2 + iSL;
EncodeIndex (bChroma, iLocation, iCont, iIndex, (iLev < 0), pAHexpt + iContextOffset, pOut);
iCont &= iSRn; /** big difference! **/
if (iSL) {
EncodeSignificantAbsLevel ((UInt)(abs(iLev) - 1), pAHexpt[6 + iContextOffset + iCont], pOut);
}
//else {
// putBit16z(pOut, (iLev < 0), 1);
//}
}
return ICERR_OK;
}
/*************************************************************************
CodeCoeffs
*************************************************************************/
#ifdef X86OPT_INLINE
__forceinline
#endif
static Int CodeCoeffs (CWMImageStrCodec * pSC, CCodingContext *pContext,
Int iMBX, Int iMBY, BitIOInfo* pIO, BitIOInfo* pIOFL)
{
const COLORFORMAT cf = pSC->m_param.cfColorFormat;
const Int iChannels = (Int) pSC->m_param.cNumChannels;
const Int iPlanes = (cf == YUV_420 || cf == YUV_422) ? 1 : iChannels;
CWMIMBInfo * pMBInfo = &pSC->MBInfo;
CAdaptiveScan *pScan;
Int iBlock, iNBlocks = 4;
Int iSubblock, iIndex = 0;
Int i, k;
const Int iNumCoeffs = 16;
Int iModelBits = pContext->m_aModelAC.m_iFlcBits[0], iFlex = 0, iTrim = 0, iMask = 0;
Int aLaplacianMean[2] = { 0, 0}, *pLM = aLaplacianMean;
Bool bChroma = FALSE;
UNREFERENCED_PARAMETER( iMBX );
UNREFERENCED_PARAMETER( iMBY );
assert (iModelBits < 16);
if (pContext->m_iTrimFlexBits <= iModelBits && pSC->WMISCP.sbSubband != SB_NO_FLEXBITS) {
iTrim = pContext->m_iTrimFlexBits;
iFlex = iModelBits - pContext->m_iTrimFlexBits;
iMask = (1 << iFlex) - 1;
}
if(pSC->WMISCP.sbSubband != SB_NO_FLEXBITS)
writeIS_L1(pSC, pIOFL);
/** set scan arrays **/
if (pMBInfo->iOrientation == 1) {
pScan = pContext->m_aScanVert;
}
else {
pScan = pContext->m_aScanHoriz;
}
/** write out coefficients **/
for (i = 0; i < iPlanes; i++) {
Int iPattern = pMBInfo->iCBP[i];
if (cf == YUV_420) {
iNBlocks = 6;
iPattern += (pMBInfo->iCBP[1] << 16) + (pMBInfo->iCBP[2] << 20);
}
else if (cf == YUV_422) {
iNBlocks = 8;
iPattern += (pMBInfo->iCBP[1] << 16) + (pMBInfo->iCBP[2] << 24);
}
for (iBlock = iIndex = 0; iBlock < iNBlocks; iBlock++) {
writeIS_L2(pSC, pIO);
if (pIO != pIOFL)
writeIS_L2(pSC, pIOFL);
for (iSubblock = 0; iSubblock < 4; iSubblock++, iPattern >>= 1, iIndex ++) {
const PixelI *pCoeffs = NULL;
if(iBlock < 4){
pCoeffs = pSC->pPlane[i] + blkOffset[iIndex];
}
else if(cf == YUV_420){
pCoeffs = pSC->pPlane[iBlock - 3] + blkOffsetUV[iSubblock];
}
else if(cf == YUV_422){
pCoeffs = pSC->pPlane[1 + ((iBlock - 4) >> 1)] + blkOffsetUV_422[(iBlock & 1) * 4 + iSubblock];
}
/** put AC bits **/
if ((iPattern & 1) == 0) {
if (iFlex) {
/** FLC only, all else is skipped **/
for (k = 1; k < iNumCoeffs; k++) {
Int data = pCoeffs[dctIndex[0][k]];
Int atdata = (abs(data) >> iTrim);
Int word = atdata & iMask, len = iFlex;
if (atdata) {
word += word + (data < 0);
len++;
}
putBit16z(pIOFL, word, len);
}
}
}
else {
// WARNING!!! interaction between lowpass coefficients and highpass scan ordering - may lead to break in decoding when model bits is nonzero!
// Fix is to use same scan order in model bits transmission, and defer update of scan order to end of block
/** collect coefficients **/
Int aLocalCoef[32], iNumNonzero = 0;
Int aResidual[16];
iNumNonzero = AdaptiveScan (pCoeffs, aResidual,
pScan, iModelBits, iTrim, aLocalCoef, 16);
(*pLM) += iNumNonzero;
EncodeBlock (bChroma, aLocalCoef, iNumNonzero, pContext->m_pAHexpt, CTDC + CONTEXTX, pIO, 1);
if (iFlex) {
for (k = 1; k < iNumCoeffs; k++) {
putBit16z(pIOFL, aResidual[dctIndex[0][k]] >> 1, iFlex + (aResidual[dctIndex[0][k]] & 1));
}
}
}
}
if (iBlock == 3) {
iModelBits = pContext->m_aModelAC.m_iFlcBits[1];
assert (iModelBits < 16);
pLM = aLaplacianMean + 1;
bChroma = TRUE;
iTrim = iFlex = iMask = 0;
if (pContext->m_iTrimFlexBits <= iModelBits && pSC->WMISCP.sbSubband != SB_NO_FLEXBITS) {
iTrim = pContext->m_iTrimFlexBits;
iFlex = iModelBits - iTrim;
iMask = (1 << iFlex) - 1;
}
}
}
}
/** update model at end of MB **/
UpdateModelMB (cf, iChannels, aLaplacianMean, &pContext->m_aModelAC);
return ICERR_OK;
}
/*************************************************************************
CodeCBP
*************************************************************************/
static Void CodeCBP (CWMImageStrCodec * pSC, CCodingContext *pContext,
Int iMBX, Int iMBY, BitIOInfo *pIO)
{
const COLORFORMAT cf = pSC->m_param.cfColorFormat;
const Int iChannel = (cf == NCOMPONENT || cf == CMYK) ? (Int) pSC->m_param.cNumChannels : 1;
Int iDiffCBPCY, iDiffCBPCU = 0, iDiffCBPCV = 0, iDY;
Int iBlock, i, k;
static const Int aNumOnes[] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4 };
static const Int aTabLen[] = { 0, 2, 2, 2, 2, 2, 3, 2, 2, 3, 3, 2, 3, 2, 2, 0 };
static const Int aTabCode[] = { 0, 0, 1, 0, 2, 1, 4, 3, 3, 5, 6, 2, 7, 1, 0, 0 };
CAdaptiveHuffman *pAH;
Int iCount, iPattern, iCode, iCodeU = 0, iCodeV = 0;
UNREFERENCED_PARAMETER( iMBX );
UNREFERENCED_PARAMETER( iMBY );
predCBPEnc(pSC, pContext);
writeIS_L1(pSC, pIO);
iDiffCBPCU = pSC->MBInfo.iDiffCBP[1];
iDiffCBPCV = pSC->MBInfo.iDiffCBP[2];
for (i = 0; i < iChannel; i++) {
iDiffCBPCY = pSC->MBInfo.iDiffCBP[i];
if(cf == YUV_420){ // PackCBP420
iDiffCBPCY = (iDiffCBPCY & 0xf) + ((iDiffCBPCU & 1) << 4) + ((iDiffCBPCV & 1) << 5) +
((iDiffCBPCY & 0x00f0) << 2) + ((iDiffCBPCU & 2) << 9) + ((iDiffCBPCV & 2) << 10) +
((iDiffCBPCY & 0x0f00) << 4) + ((iDiffCBPCU & 4) << 14) + ((iDiffCBPCV & 4) << 15) +
((iDiffCBPCY & 0xf000) << 6) + ((iDiffCBPCU & 8) << 19) + ((iDiffCBPCV & 8) << 20);
}
else if(cf == YUV_422){// PackCBP422
iDiffCBPCY = (iDiffCBPCY & 0xf) + ((iDiffCBPCU & 1) << 4) + ((iDiffCBPCU & 4) << 3) +
((iDiffCBPCV & 1) << 6) + ((iDiffCBPCV & 4) << 5) +
((iDiffCBPCY & 0x00f0) << 4) + ((iDiffCBPCU & 2) << 11) + ((iDiffCBPCU & 8) << 10) +
((iDiffCBPCV & 2) << 13) + ((iDiffCBPCV & 8) << 12) +
((iDiffCBPCY & 0x0f00) << 8) + ((iDiffCBPCU & 16) << 16) + ((iDiffCBPCU & 64) << 15) +
((iDiffCBPCV & 16) << 18) + ((iDiffCBPCV & 64) << 17) +
((iDiffCBPCY & 0xf000) << 12) + ((iDiffCBPCU & 32) << 23) + ((iDiffCBPCU & 128) << 22) +
((iDiffCBPCV & 32) << 25) + ((iDiffCBPCV & 128) << 24);
}
/** send CBPCY **/
iPattern = 0;
iDY = iDiffCBPCY;
if (cf == YUV_444) {
iDY |= (iDiffCBPCU | iDiffCBPCV);
}
for (iBlock = 0; iBlock < 4; iBlock++) {
if(cf == YUV_422) {
iPattern |= ((iDY & 0xff) != 0) * 0x10;
iDY >>= 8;
}
else if (cf == YUV_420) {
iPattern |= ((iDY & 0x3f) != 0) * 0x10;
iDY >>= 6;
}
else {
iPattern |= ((iDY & 0xf) != 0) * 0x10;
iDY >>= 4;
}
iPattern >>= 1;
}
pAH = pContext->m_pAdaptHuffCBPCY1;
iCount = aNumOnes[iPattern];
putBit16z(pIO, pAH->m_pTable[iCount * 2 + 1], pAH->m_pTable[iCount * 2 + 2]);
pAH->m_iDiscriminant += pAH->m_pDelta[iCount];
if (aTabLen[iPattern]) {
putBit16z(pIO, aTabCode[iPattern], aTabLen[iPattern]);
}
for (iBlock = 0; iBlock < 4; iBlock++) {
switch (cf) {
case YUV_444:
iCode = iDiffCBPCY & 0xf;
iCodeU = iDiffCBPCU & 0xf;
iCodeV = iDiffCBPCV & 0xf;
iCode |= ((iCodeU != 0) << 4);
iCode |= ((iCodeV != 0) << 5);
iDiffCBPCY >>= 4;
iDiffCBPCU >>= 4;
iDiffCBPCV >>= 4;
break;
case YUV_422:
iCode = iDiffCBPCY & 0xff;
iDiffCBPCY >>= 8;
break;
case YUV_420:
iCode = iDiffCBPCY & 0x3f;
iDiffCBPCY >>= 6;
break;
default:
iCode = iDiffCBPCY & 0xf;
iDiffCBPCY >>= 4;
}
if (iCode) {
static const Int gTab0[16] = { 0,1,1,2, 1,3,3,4, 1,3,3,4, 2,4,4,5 };
static const Int gFL0[16] = { 0,2,2,1, 2,2,2,2, 2,2,2,2, 1,2,2,0 };
static const Int gCode0[16] = { 0,0,1,0, 2,0,1,0, 3,2,3,1, 1,2,3,0 };
int val, iChroma = (iCode >> 4);
iCode &= 0xf;
if(cf == YUV_422) {
iCodeU = (iChroma & 3);
iCodeV = ((iChroma >> 2) & 3);
iChroma = (iCodeU == 0 ? 0 : 1);
if(iCodeV != 0) {
iChroma += 2;
}
}
if (iChroma) {
if (gTab0[iCode] > 2) {
val = 8;
}
else {
val = gTab0[iCode] + 6 - 1;
}
}
else {
val = gTab0[iCode] - 1;
}
pAH = pContext->m_pAdaptHuffCBPCY;
putBit16z(pIO, pAH->m_pTable[val * 2 + 1], pAH->m_pTable[val * 2 + 2]);
pAH->m_iDiscriminant += pAH->m_pDelta[val];
if (iChroma) {
if (iChroma == 1)
putBit16z(pIO, 1, 1);
else
putBit16z(pIO, 3 - iChroma, 2);
}
if (val == 8) {
if (gTab0[iCode] == 3) {
putBit16z(pIO, 1, 1);
}
else {
putBit16z(pIO, 5 - gTab0[iCode], 2);
}
}
if (gFL0[iCode]) {
putBit16z(pIO, gCode0[iCode], gFL0[iCode]);
}
if (cf == YUV_444) {
pAH = pContext->m_pAHexpt[1];
iPattern = iCodeU;
for (k = 0; k < 2; k++) {
if (iPattern) {
iCount = aNumOnes[iPattern];
iCount--;
putBit16z(pIO, pAH->m_pTable[iCount * 2 + 1], pAH->m_pTable[iCount * 2 + 2]);
if (aTabLen[iPattern]) {
putBit16z(pIO, aTabCode[iPattern], aTabLen[iPattern]);
}
}
iPattern = iCodeV;
}
}
else if (cf == YUV_422){
iPattern = iCodeU;
for(k = 0; k < 2; k ++) {
if(iPattern) {
if (iPattern == 1)
putBit16z(pIO, 1, 1);
else {
putBit16z(pIO, 3 - iPattern, 2);
}
}
iPattern = iCodeV;
}
}
}
}
}
}
/*************************************************************************
macroblock encode function using 4x4 transforms
*************************************************************************/
Int EncodeMacroblockHighpass(CWMImageStrCodec * pSC, CCodingContext *pContext, Int iMBX, Int iMBY)
{
BitIOInfo* pIO = pContext->m_pIOAC;
BitIOInfo* pIOFL = pContext->m_pIOFL;
if((pSC->WMISCP.bfBitstreamFormat != SPATIAL) && (pSC->pTile[pSC->cTileColumn].cBitsHP > 0)) // MB-based HP QP index
encodeQPIndex(pIO, pSC->MBInfo.iQIndexHP, pSC->pTile[pSC->cTileColumn].cBitsHP);
/** reset adaptive scan totals **/
if (pSC->m_bResetRGITotals) {
Int iScale = 2;
Int iWeight = iScale * 16;
Int k;
pContext->m_aScanHoriz[0].uTotal = pContext->m_aScanVert[0].uTotal = MAXTOTAL;
for (k = 1; k < 16; k++) {
pContext->m_aScanHoriz[k].uTotal = pContext->m_aScanVert[k].uTotal = iWeight;
iWeight -= iScale;
}
}
CodeCBP(pSC, pContext, iMBX, iMBY, pIO);
if(CodeCoeffs(pSC, pContext, iMBX, iMBY, pIO, pIOFL) != ICERR_OK)
return ICERR_ERROR;
if (pSC->m_bResetContext) {
AdaptHighpassEnc(pContext);
}
return ICERR_OK;
}