Index: core/fxcodec/jbig2/JBig2_GrdProc.cpp |
diff --git a/core/fxcodec/jbig2/JBig2_GrdProc.cpp b/core/fxcodec/jbig2/JBig2_GrdProc.cpp |
index a3fc33c2bed3d06d7173a4062d0af66eb9ffe97c..5f3b47f659758d2be74fc4178cb34c839d863c2a 100644 |
--- a/core/fxcodec/jbig2/JBig2_GrdProc.cpp |
+++ b/core/fxcodec/jbig2/JBig2_GrdProc.cpp |
@@ -66,8 +66,8 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template0_opt3( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
uint8_t *pLine, *pLine1, *pLine2, cVal; |
int32_t nStride, nStride2, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -81,8 +81,8 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template0_opt3( |
nStride2 = nStride << 1; |
nLineBytes = ((GBW + 7) >> 3) - 1; |
nBitsLeft = GBW - (nLineBytes << 3); |
- FX_DWORD height = GBH & 0x7fffffff; |
- for (FX_DWORD h = 0; h < height; h++) { |
+ uint32_t height = GBH & 0x7fffffff; |
+ for (uint32_t h = 0; h < height; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x9b25]); |
LTP = LTP ^ SLTP; |
@@ -156,12 +156,12 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template0_unopt( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2, line3; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2, line3; |
LTP = 0; |
std::unique_ptr<CJBig2_Image> GBREG(new CJBig2_Image(GBW, GBH)); |
GBREG->fill(0); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x9b25]); |
LTP = LTP ^ SLTP; |
@@ -175,7 +175,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template0_unopt( |
line2 |= GBREG->getPixel(1, h - 1) << 1; |
line2 |= GBREG->getPixel(0, h - 1) << 2; |
line3 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, h)) { |
bVal = 0; |
} else { |
@@ -204,8 +204,8 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template1_opt3( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
uint8_t *pLine, *pLine1, *pLine2, cVal; |
int32_t nStride, nStride2, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -219,7 +219,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template1_opt3( |
nStride2 = nStride << 1; |
nLineBytes = ((GBW + 7) >> 3) - 1; |
nBitsLeft = GBW - (nLineBytes << 3); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x0795]); |
LTP = LTP ^ SLTP; |
@@ -293,12 +293,12 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template1_unopt( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2, line3; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2, line3; |
LTP = 0; |
std::unique_ptr<CJBig2_Image> GBREG(new CJBig2_Image(GBW, GBH)); |
GBREG->fill(0); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x0795]); |
LTP = LTP ^ SLTP; |
@@ -313,7 +313,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template1_unopt( |
line2 |= GBREG->getPixel(1, h - 1) << 1; |
line2 |= GBREG->getPixel(0, h - 1) << 2; |
line3 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, h)) { |
bVal = 0; |
} else { |
@@ -338,8 +338,8 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template2_opt3( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
uint8_t *pLine, *pLine1, *pLine2, cVal; |
int32_t nStride, nStride2, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -353,7 +353,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template2_opt3( |
nStride2 = nStride << 1; |
nLineBytes = ((GBW + 7) >> 3) - 1; |
nBitsLeft = GBW - (nLineBytes << 3); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x00e5]); |
LTP = LTP ^ SLTP; |
@@ -427,12 +427,12 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template2_unopt( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2, line3; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2, line3; |
LTP = 0; |
std::unique_ptr<CJBig2_Image> GBREG(new CJBig2_Image(GBW, GBH)); |
GBREG->fill(0); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x00e5]); |
LTP = LTP ^ SLTP; |
@@ -445,7 +445,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template2_unopt( |
line2 = GBREG->getPixel(1, h - 1); |
line2 |= GBREG->getPixel(0, h - 1) << 1; |
line3 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, h)) { |
bVal = 0; |
} else { |
@@ -471,8 +471,8 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template3_opt3( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1; |
+ uint32_t CONTEXT; |
+ uint32_t line1; |
uint8_t *pLine, *pLine1, cVal; |
int32_t nStride, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -485,7 +485,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template3_opt3( |
nStride = GBREG->m_nStride; |
nLineBytes = ((GBW + 7) >> 3) - 1; |
nBitsLeft = GBW - (nLineBytes << 3); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x0195]); |
LTP = LTP ^ SLTP; |
@@ -546,12 +546,12 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template3_unopt( |
CJBig2_ArithDecoder* pArithDecoder, |
JBig2ArithCtx* gbContext) { |
FX_BOOL LTP, SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
LTP = 0; |
std::unique_ptr<CJBig2_Image> GBREG(new CJBig2_Image(GBW, GBH)); |
GBREG->fill(0); |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x0195]); |
LTP = LTP ^ SLTP; |
@@ -562,7 +562,7 @@ CJBig2_Image* CJBig2_GRDProc::decode_Arith_Template3_unopt( |
line1 = GBREG->getPixel(1, h - 1); |
line1 |= GBREG->getPixel(0, h - 1) << 1; |
line2 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, h)) { |
bVal = 0; |
} else { |
@@ -673,7 +673,7 @@ FXCODEC_STATUS CJBig2_GRDProc::Start_decode_MMR(CJBig2_Image** pImage, |
FaxG4Decode(pStream->getBuf(), pStream->getLength(), &bitpos, |
(*pImage)->m_pData, GBW, GBH, (*pImage)->m_nStride); |
pStream->setBitPos(bitpos); |
- for (i = 0; (FX_DWORD)i < (*pImage)->m_nStride * GBH; i++) { |
+ for (i = 0; (uint32_t)i < (*pImage)->m_nStride * GBH; i++) { |
(*pImage)->m_pData[i] = ~(*pImage)->m_pData[i]; |
} |
m_ProssiveStatus = FXCODEC_STATUS_DECODE_FINISH; |
@@ -698,8 +698,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template0_opt3( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
uint8_t *pLine1, *pLine2, cVal; |
int32_t nStride, nStride2, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -710,7 +710,7 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template0_opt3( |
nStride2 = nStride << 1; |
nLineBytes = ((GBW + 7) >> 3) - 1; |
nBitsLeft = GBW - (nLineBytes << 3); |
- FX_DWORD height = GBH & 0x7fffffff; |
+ uint32_t height = GBH & 0x7fffffff; |
for (; m_loopIndex < height; m_loopIndex++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x9b25]); |
@@ -793,8 +793,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template0_unopt( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2, line3; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2, line3; |
for (; m_loopIndex < GBH; m_loopIndex++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x9b25]); |
@@ -809,7 +809,7 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template0_unopt( |
line2 |= pImage->getPixel(1, m_loopIndex - 1) << 1; |
line2 |= pImage->getPixel(0, m_loopIndex - 1) << 2; |
line3 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, m_loopIndex)) { |
bVal = 0; |
} else { |
@@ -848,8 +848,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template1_opt3( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
uint8_t *pLine1, *pLine2, cVal; |
int32_t nStride, nStride2, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -942,9 +942,9 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template1_unopt( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2, line3; |
- for (FX_DWORD h = 0; h < GBH; h++) { |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2, line3; |
+ for (uint32_t h = 0; h < GBH; h++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x0795]); |
LTP = LTP ^ SLTP; |
@@ -959,7 +959,7 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template1_unopt( |
line2 |= pImage->getPixel(1, h - 1) << 1; |
line2 |= pImage->getPixel(0, h - 1) << 2; |
line3 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, h)) { |
bVal = 0; |
} else { |
@@ -993,8 +993,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template2_opt3( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
uint8_t *pLine1, *pLine2, cVal; |
int32_t nStride, nStride2, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -1087,8 +1087,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template2_unopt( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2, line3; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2, line3; |
for (; m_loopIndex < GBH; m_loopIndex++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x00e5]); |
@@ -1102,7 +1102,7 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template2_unopt( |
line2 = pImage->getPixel(1, m_loopIndex - 1); |
line2 |= pImage->getPixel(0, m_loopIndex - 1) << 1; |
line3 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, m_loopIndex)) { |
bVal = 0; |
} else { |
@@ -1138,8 +1138,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template3_opt3( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1; |
+ uint32_t CONTEXT; |
+ uint32_t line1; |
uint8_t *pLine1, cVal; |
int32_t nStride, k; |
int32_t nLineBytes, nBitsLeft, cc; |
@@ -1218,8 +1218,8 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template3_unopt( |
JBig2ArithCtx* gbContext, |
IFX_Pause* pPause) { |
FX_BOOL SLTP, bVal; |
- FX_DWORD CONTEXT; |
- FX_DWORD line1, line2; |
+ uint32_t CONTEXT; |
+ uint32_t line1, line2; |
for (; m_loopIndex < GBH; m_loopIndex++) { |
if (TPGDON) { |
SLTP = pArithDecoder->DECODE(&gbContext[0x0195]); |
@@ -1231,7 +1231,7 @@ FXCODEC_STATUS CJBig2_GRDProc::decode_Arith_Template3_unopt( |
line1 = pImage->getPixel(1, m_loopIndex - 1); |
line1 |= pImage->getPixel(0, m_loopIndex - 1) << 1; |
line2 = 0; |
- for (FX_DWORD w = 0; w < GBW; w++) { |
+ for (uint32_t w = 0; w < GBW; w++) { |
if (USESKIP && SKIP->getPixel(w, m_loopIndex)) { |
bVal = 0; |
} else { |