MemSubPic refactoring

git-svn-id: https://mpc-hc.svn.sourceforge.net/svnroot/mpc-hc/trunk@1926 10f7b99b-c216-0410-bff0-8a66a9350fd8

3 files changed, 158 insertions, 92 deletions

diff --git a/src/DSUtil/DSUtil.cpp b/src/DSUtil/DSUtil.cpp
index 563dd1ba4..728ac9c39 100644
--- a/src/DSUtil/DSUtil.cpp
+++ b/src/DSUtil/DSUtil.cpp
@@ -976,6 +976,7 @@ void memsetd(void* dst, unsigned int c, size_t nbytes)
 #endif

 	size_t n = nbytes / 4;

 	size_t o = n - (n % 4);

+	ASSERT(n == o);

 

 	__m128i val = _mm_set1_epi32 ( (int)c );

 	if (((uintptr_t)dst & 0x0F) == 0) // 16-byte aligned

diff --git a/src/SubPic/MemSubPic.cpp b/src/SubPic/MemSubPic.cpp
index ec44f8ffd..58c3f1071 100644
--- a/src/SubPic/MemSubPic.cpp
+++ b/src/SubPic/MemSubPic.cpp
@@ -22,6 +22,10 @@
 #include "stdafx.h"

 #include "MemSubPic.h"

 

+// For CPUID usage

+#include "../dsutil/vd.h"

+#include <emmintrin.h>

+

 // color conv

 

 unsigned char Clip_base[256*3];

@@ -155,8 +159,6 @@ STDMETHODIMP CMemSubPic::ClearDirtyRect(DWORD color)
 	BYTE* p = (BYTE*)m_spd.bits + m_spd.pitch*m_rcDirty.top + m_rcDirty.left*(m_spd.bpp>>3);

 	for(ptrdiff_t j = 0, h = m_rcDirty.Height(); j < h; j++, p += m_spd.pitch)

 	{

-//        

-

 		int w = m_rcDirty.Width();

 #ifdef _WIN64

 		memsetd(p, color, w*4); // nya

@@ -292,10 +294,122 @@ STDMETHODIMP CMemSubPic::Unlock(RECT* pDirtyRect)
 }

 

 #ifdef _WIN64

-// For CPUID usage

-#include "../dsutil/vd.h"

-#include <emmintrin.h>

+void AlphaBlt_YUY2_SSE2(int w, int h, BYTE* d, int dstpitch, BYTE* s, int srcpitch)

+{

+	unsigned int ia, c;

+	DWORD* d2 = (DWORD*)d;

+

+	BYTE* s2 = s;

+	BYTE* s2end = s2 + w*4;

+	static const __int64 _8181 = 0x0080001000800010i64;

+

+	for(ptrdiff_t j = 0; j < h; j++, s += srcpitch, d += dstpitch)

+	{	

+		for(; s2 < s2end; s2 += 8, d2++)

+		{

+			ia = (s2[3]+s2[7])>>1;

+			if(ia < 0xff)

+			{

+				c = (s2[4]<<24)|(s2[5]<<16)|(s2[0]<<8)|s2[1]; // (v<<24)|(y2<<16)|(u<<8)|y1;

+

+				ia = (ia<<24)|(s2[7]<<16)|(ia<<8)|s2[3];

+				// SSE2

+				__m128i mm_zero = _mm_setzero_si128();

+				__m128i mm_8181 = _mm_move_epi64(_mm_cvtsi64_si128(_8181));

+				__m128i mm_c = _mm_cvtsi32_si128(c);

+				mm_c = _mm_unpacklo_epi8(mm_c, mm_zero);

+				__m128i mm_d = _mm_cvtsi32_si128(*d2);

+				mm_d = _mm_unpacklo_epi8(mm_d, mm_zero);

+				__m128i mm_a = _mm_cvtsi32_si128(ia);

+				mm_a = _mm_unpacklo_epi8(mm_a, mm_zero);

+				mm_a = _mm_srli_epi16(mm_a,1);

+				mm_d = _mm_sub_epi16(mm_d,mm_8181);

+				mm_d = _mm_mullo_epi16(mm_d,mm_a);

+				mm_d = _mm_srai_epi16(mm_d,7);

+				mm_d = _mm_adds_epi16(mm_d,mm_c);

+				mm_d = _mm_packus_epi16(mm_d,mm_d);

+				*d2 = (DWORD)_mm_cvtsi128_si32(mm_d);

+			}

+		}

+	}

+}

 #endif

+

+#ifndef _WIN64

+void AlphaBlt_YUY2_MMX(int w, int h, BYTE* d, int dstpitch, BYTE* s, int srcpitch)

+{

+	unsigned int ia, c;

+	DWORD* d2 = (DWORD*)d;

+

+	BYTE* s2 = s;

+	BYTE* s2end = s2 + w*4;

+	static const __int64 _8181 = 0x0080001000800010i64;

+

+

+	for(ptrdiff_t j = 0; j < h; j++, s += srcpitch, d += dstpitch)

+	{	

+		for(; s2 < s2end; s2 += 8, d2++)

+		{

+			ia = (s2[3]+s2[7])>>1;

+			if(ia < 0xff)

+			{

+				c = (s2[4]<<24)|(s2[5]<<16)|(s2[0]<<8)|s2[1]; // (v<<24)|(y2<<16)|(u<<8)|y1;

+				ia = (ia<<24)|(s2[7]<<16)|(ia<<8)|s2[3];

+				__asm

+				{

+					mov			esi, s2

+					mov			edi, d2

+					pxor		mm0, mm0

+					movq		mm1, _8181

+					movd		mm2, c

+					punpcklbw	mm2, mm0

+					movd		mm3, [edi]

+					punpcklbw	mm3, mm0

+					movd		mm4, ia

+					punpcklbw	mm4, mm0

+					psrlw		mm4, 1

+					psubsw		mm3, mm1

+					pmullw		mm3, mm4

+					psraw		mm3, 7

+					paddsw		mm3, mm2

+					packuswb	mm3, mm3

+					movd		[edi], mm3

+				};

+			}

+		}

+	}

+}

+#endif

+

+void AlphaBlt_YUY2_C(int w, int h, BYTE* d, int dstpitch, BYTE* s, int srcpitch)

+{

+	unsigned int ia, c;

+	DWORD* d2 = (DWORD*)d;

+

+	BYTE* s2 = s;

+	BYTE* s2end = s2 + w*4;

+	static const __int64 _8181 = 0x0080001000800010i64;

+

+	for(ptrdiff_t j = 0; j < h; j++, s += srcpitch, d += dstpitch)

+	{			

+		for(; s2 < s2end; s2 += 8, d2++)

+		{

+			ia = (s2[3]+s2[7])>>1;

+			if(ia < 0xff)

+			{

+				c = (s2[4]<<24)|(s2[5]<<16)|(s2[0]<<8)|s2[1]; // (v<<24)|(y2<<16)|(u<<8)|y1;

+

+				// YUY2 colorspace fix. rewrited from sse2 asm

+				DWORD y1 = (DWORD)(((((*d2&0xff)-0x10)*(s2[3]>>1))>>7)+s2[1])&0xff;			// y1

+				DWORD uu = (DWORD)((((((*d2>>8)&0xff)-0x80)*(ia>>1))>>7)+s2[0])&0xff;		// u

+				DWORD y2 = (DWORD)((((((*d2>>16)&0xff)-0x10)*(s2[7]>>1))>>7)+s2[5])&0xff;	// y2

+				DWORD vv = (DWORD)((((((*d2>>24)&0xff)-0x80)*(ia>>1))>>7)+s2[4])&0xff;		// v

+				*d2 = (y1)|(uu<<8)|(y2<<16)|(vv<<24);

+			}

+		}

+	}

+}

+

 STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)

 {

 	ASSERT(pTarget);

@@ -346,9 +460,10 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 		dst.pitch = -dst.pitch;

 	}

 

-	for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

+	switch(dst.type)

 	{

-		if(dst.type == MSP_RGBA)

+	case MSP_RGBA:

+		for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

 		{

 			BYTE* s2 = s;

 			BYTE* s2end = s2 + w*4;

@@ -366,11 +481,14 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 				}

 			}

 		}

-		else if(dst.type == MSP_RGB32 || dst.type == MSP_AYUV)

+		break;

+	case MSP_RGB32:

+	case MSP_AYUV:

+		for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

 		{

 			BYTE* s2 = s;

 			BYTE* s2end = s2 + w*4;

-			

+

 			DWORD* d2 = (DWORD*)d;

 			for(; s2 < s2end; s2 += 4, d2++)

 			{

@@ -390,7 +508,9 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 #endif

 			}

 		}

-		else if(dst.type == MSP_RGB24)

+		break;

+	case MSP_RGB24:

+		for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

 		{

 			BYTE* s2 = s;

 			BYTE* s2end = s2 + w*4;

@@ -405,7 +525,9 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 				}

 			}

 		}

-		else if(dst.type == MSP_RGB16)

+		break;

+	case MSP_RGB16:

+		for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

 		{

 			BYTE* s2 = s;

 			BYTE* s2end = s2 + w*4;

@@ -414,13 +536,15 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 			{

 				if(s2[3] < 0x1f)

 				{

-					

+

 					*d2 = (WORD)((((((*d2&0xf81f)*s2[3])>>5) + (*(DWORD*)s2&0xf81f))&0xf81f)

-								| (((((*d2&0x07e0)*s2[3])>>5) + (*(DWORD*)s2&0x07e0))&0x07e0));

+						| (((((*d2&0x07e0)*s2[3])>>5) + (*(DWORD*)s2&0x07e0))&0x07e0));

 				}

 			}

 		}

-		else if(dst.type == MSP_RGB15)

+		break;

+	case MSP_RGB15:

+		for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

 		{

 			BYTE* s2 = s;

 			BYTE* s2end = s2 + w*4;

@@ -430,87 +554,27 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 				if(s2[3] < 0x1f)

 				{

 					*d2 = (WORD)((((((*d2&0x7c1f)*s2[3])>>5) + (*(DWORD*)s2&0x7c1f))&0x7c1f)

-								| (((((*d2&0x03e0)*s2[3])>>5) + (*(DWORD*)s2&0x03e0))&0x03e0));

+						| (((((*d2&0x03e0)*s2[3])>>5) + (*(DWORD*)s2&0x03e0))&0x03e0));

 				}

 			}

 		}

-		else if(dst.type == MSP_YUY2)

-		{	

-			unsigned int ia, c;

-#ifdef _WIN64

-			// CPUID from VDub

-			bool fSSE2 = !!(g_cpuid.m_flags & CCpuID::sse2);

-#endif

-			DWORD* d2 = (DWORD*)d;

-

-			BYTE* s2 = s;

-			BYTE* s2end = s2 + w*4;

-			static const __int64 _8181 = 0x0080001000800010i64;

-

-			for(; s2 < s2end; s2 += 8, d2++)

-			{

-				ia = (s2[3]+s2[7])>>1;

-				if(ia < 0xff)

-				{

-					c = (s2[4]<<24)|(s2[5]<<16)|(s2[0]<<8)|s2[1]; // (v<<24)|(y2<<16)|(u<<8)|y1;

+		break;

+	case MSP_YUY2:

+		{

+			void (*alphablt_func)(int w, int h, BYTE* d, int dstpitch, BYTE* s, int srcpitch);

 #ifdef _WIN64

-					if(fSSE2) 

-					{

-						ia = (ia<<24)|(s2[7]<<16)|(ia<<8)|s2[3];

-						// SSE2

-						__m128i mm_zero = _mm_setzero_si128();

-						__m128i mm_8181 = _mm_move_epi64(_mm_cvtsi64_si128(_8181));

-						__m128i mm_c = _mm_cvtsi32_si128(c);

-						mm_c = _mm_unpacklo_epi8(mm_c, mm_zero);

-						__m128i mm_d = _mm_cvtsi32_si128(*d2);

-						mm_d = _mm_unpacklo_epi8(mm_d, mm_zero);

-						__m128i mm_a = _mm_cvtsi32_si128(ia);

-						mm_a = _mm_unpacklo_epi8(mm_a, mm_zero);

-						mm_a = _mm_srli_epi16(mm_a,1);

-						mm_d = _mm_sub_epi16(mm_d,mm_8181);

-						mm_d = _mm_mullo_epi16(mm_d,mm_a);

-						mm_d = _mm_srai_epi16(mm_d,7);

-						mm_d = _mm_adds_epi16(mm_d,mm_c);

-						mm_d = _mm_packus_epi16(mm_d,mm_d);

-						*d2 = (DWORD)_mm_cvtsi128_si32(mm_d);

-					}

-					else

-					{

-						// YUY2 colorspace fix. rewrited from sse2 asm

-						DWORD y1 = (DWORD)(((((*d2&0xff)-0x10)*(s2[3]>>1))>>7)+s2[1])&0xff;			// y1

-						DWORD uu = (DWORD)((((((*d2>>8)&0xff)-0x80)*(ia>>1))>>7)+s2[0])&0xff;		// u

-						DWORD y2 = (DWORD)((((((*d2>>16)&0xff)-0x10)*(s2[7]>>1))>>7)+s2[5])&0xff;	// y2

-						DWORD vv = (DWORD)((((((*d2>>24)&0xff)-0x80)*(ia>>1))>>7)+s2[4])&0xff;		// v

-						*d2 = (y1)|(uu<<8)|(y2<<16)|(vv<<24);

-					}

-		

+			alphablt_func = AlphaBlt_YUY2_SSE2;

 #else

-					ia = (ia<<24)|(s2[7]<<16)|(ia<<8)|s2[3];

-					__asm

-					{

-						mov			esi, s2

-						mov			edi, d2

-						pxor		mm0, mm0

-						movq		mm1, _8181

-						movd		mm2, c

-						punpcklbw	mm2, mm0

-						movd		mm3, [edi]

-						punpcklbw	mm3, mm0

-						movd		mm4, ia

-						punpcklbw	mm4, mm0

-						psrlw		mm4, 1

-						psubsw		mm3, mm1

-						pmullw		mm3, mm4

-						psraw		mm3, 7

-						paddsw		mm3, mm2

-						packuswb	mm3, mm3

-						movd		[edi], mm3

-					};

+			alphablt_func = AlphaBlt_YUY2_MMX;

 #endif

-				}

-			}

+			//alphablt_func = AlphaBlt_YUY2_C;

+

+			alphablt_func(w, h, d, dst.pitch, s, src.pitch);

 		}

-		else if(dst.type == MSP_YV12 || dst.type == MSP_IYUV)

+		break;

+	case MSP_YV12:

+	case MSP_IYUV:

+		for(ptrdiff_t j = 0; j < h; j++, s += src.pitch, d += dst.pitch)

 		{

 			BYTE* s2 = s;

 			BYTE* s2end = s2 + w*4;

@@ -523,10 +587,9 @@ STDMETHODIMP CMemSubPic::AlphaBlt(RECT* pSrc, RECT* pDst, SubPicDesc* pTarget)
 				}

 			}

 		}

-		else

-		{

-			return E_NOTIMPL;

-		}

+		break;

+	default:

+		return E_NOTIMPL;

 	}

 

 	dst.pitch = abs(dst.pitch);

diff --git a/src/subtitles/RTS.cpp b/src/subtitles/RTS.cpp
index d5be4c52f..c6e81ce4c 100644
--- a/src/subtitles/RTS.cpp
+++ b/src/subtitles/RTS.cpp
@@ -146,12 +146,14 @@ void CWord::Paint(CPoint p, CPoint org)
 

 void CWord::Transform(CPoint org)

 {

+#ifdef _VSMOD

     // CPUID from VDub

     bool fSSE2 = !!(g_cpuid.m_flags & CCpuID::sse2);

 

     if(fSSE2)	// SSE code

 		Transform_SSE2(org);

 	else		// C-code

+#endif

 		Transform_C(org);

 }