Linux: memcpy_fast seems stable enough on Linux, so I'm removing the switch, and turning it on by default.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@759 96395faa-99c1-11dd-bbfe-3dabce05a288

build.sh

@@ -9,14 +9,11 @@
 #export PCSX2OPTIONS="--enable-sse3 --enable-sse4 --prefix `pwd`"
 
 #Optimized, but a devbuild
-#export PCSX2OPTIONS="--enable-sse3 --enable-sse4 --enable-devbuild --prefix `pwd`"
+export PCSX2OPTIONS="--enable-sse3 --enable-sse4 --enable-devbuild --prefix `pwd`"
 
 #Debug / Devbuild version
 #export PCSX2OPTIONS="--enable-debug --enable-devbuild --enable-sse3 --prefix `pwd`"
 
-#Optimized, but a devbuild - with memcpy_fast_ enabled. - EXPERIMENTAL
-export PCSX2OPTIONS="--enable-sse3 --enable-sse4 --enable-devbuild --enable-memcpyfast --prefix `pwd`"
-
 #ZeroGS Normal mode
 export ZEROGSOPTIONS="--enable-sse2"
 

pcsx2/IopDma.cpp

@@ -153,7 +153,7 @@ void psxDma10(u32 madr, u32 bcr, u32 chcr) {
 		psHu32(0x1000F240) &= ~0x40;
 		psHu32(0x1000F240) &= ~0x100;
 		psHu32(0x1000F240) &= ~0x4000;
-		FreezeXMMRegs(0)
+		FreezeXMMRegs(0);
 	}
 }
 

pcsx2/MemcpyFast.h

@@ -19,46 +19,29 @@
 #ifndef __MEMCPY_FAST_H__
 #define __MEMCPY_FAST_H__
 
-void _memset16_unaligned( void* dest, u16 data, size_t size );
+#if defined(_WIN32)
+	#include "windows/memzero.h"
+#else
+	#include "Linux/memzero.h"
+#endif // WIN32
 
-#if defined(_WIN32) && !defined(__x86_64__)
+	void _memset16_unaligned( void* dest, u16 data, size_t size );
 
 	// The new simplified memcpy_amd_ is now faster than memcpy_raz_.
 	// memcpy_amd_ also does mmx register saving, negating the need for freezeregs (code cleanup!)
 	// Additionally, using one single memcpy implementation keeps the code cache cleaner.
 
-	//extern void __fastcall memcpy_raz_udst(void *dest, const void *src, size_t bytes);
+#ifdef __LINUX__
-	//extern void __fastcall memcpy_raz_usrc(void *dest, const void *src, size_t bytes);
+	extern "C" void __fastcall memcpy_amd_(void *dest, const void *src, size_t bytes);
-	//extern void __fastcall memcpy_raz_(void *dest, const void *src, size_t bytes);
+	extern "C" u8 memcmp_mmx(const void* src1, const void* src2, int cmpsize);
+	extern "C" void memxor_mmx(void* dst, const void* src1, int cmpsize);
+#else
 	extern void __fastcall memcpy_amd_(void *dest, const void *src, size_t bytes);
 	extern u8 memcmp_mmx(const void* src1, const void* src2, int cmpsize);
 	extern void memxor_mmx(void* dst, const void* src1, int cmpsize);
+#endif
 
-#	include "windows/memzero.h"
+	#define memcpy_fast memcpy_amd_
-#	define memcpy_fast memcpy_amd_
+	#define memcpy_aligned memcpy_amd_
-#	define memcpy_aligned memcpy_amd_
+		
-
-#else
-
-	// for now linux uses the GCC memcpy/memset implementations.
-	//#define memcpy_raz_udst memcpy
-	//#define memcpy_raz_usrc memcpy
-	//#define memcpy_raz_ memcpy
-	
-	// fast_routines.S
-	extern "C" u8 memcmp_mmx(const void* src1, const void* src2, int cmpsize);
-	extern "C" void memxor_mmx(void* dst, const void* src1, int cmpsize);
-
-#	include "Linux/memzero.h"
-#if defined(LINUX_USE_FAST_MEMORY)
-#	define memcpy_fast memcpy_amd_
-#	define memcpy_aligned memcpy_amd_
-	extern "C" void __fastcall memcpy_amd_(void *dest, const void *src, size_t bytes);
-#else
-#	define memcpy_fast memcpy
-#	define memcpy_aligned memcpy
-#endif // LINUX_USE_FAST_MEMORY
-
-#endif // WIN32
-
 #endif //Header

pcsx2/configure.ac

@@ -63,14 +63,14 @@ AC_MSG_RESULT($debug)
 
 AC_CHECK_FUNCS([ _aligned_malloc _aligned_free ], AC_DEFINE(HAVE_ALIGNED_MALLOC))
 
-AC_MSG_CHECKING(turn on memcpy_fast_)
+#AC_MSG_CHECKING(turn on memcpy_fast_)
-AC_ARG_ENABLE(memcpyfast, AC_HELP_STRING([--enable-memcpyfast], [Turns on memcpy_fast - EXPERIMENTAL]),
+#AC_ARG_ENABLE(memcpyfast, AC_HELP_STRING([--enable-memcpyfast], [Turns on memcpy_fast - EXPERIMENTAL]),
- memcpyfast=$enableval,memcpyfast=no)
+#memcpyfast=$enableval,memcpyfast=no)
-if test "x$memcpyfast" == xyes
+#if test "x$memcpyfast" == xyes
-then
+#then
- AC_DEFINE(LINUX_USE_FAST_MEMORY,1,[LINUX_USE_FAST_MEMORY])
+# AC_DEFINE(LINUX_USE_FAST_MEMORY,1,[LINUX_USE_FAST_MEMORY])
-fi
+#fi
-AC_MSG_RESULT($memcpyfast)
+#AC_MSG_RESULT($memcpyfast)
 
 #AC_MSG_CHECKING(turn on microVU)
 #AC_ARG_ENABLE(microVU, AC_HELP_STRING([--enable-microVU], [Turns on the currently incomplete microVU files - Not a good idea]),

pcsx2/x86/iVUzerorec.cpp

@@ -698,19 +698,10 @@ void* SuperVUGetProgram(u32 startpc, int vuindex)
 bool VuFunctionHeader::IsSame(void* pmem)
 {
 #ifdef SUPERVU_CACHING
-	//u32 checksum[2];
 	vector<RANGE>::iterator it;
-	FORIT(it, ranges) {
+	FORIT(it, ranges) 
-		//memxor_mmx(checksum, (u8*)pmem+it->start, it->size);
+	{
-		//if( checksum[0] != it->checksum[0] || checksum[1] != it->checksum[1] )
+		if( memcmp_mmx((u8*)pmem+it->start, it->pmem, it->size) )
-		//	return false;
-        // memcmp_mmx doesn't work on x86-64 machines
-	// and neither does pcsx2. 
-//#if defined(_MSC_VER)
-	if( memcmp_mmx((u8*)pmem+it->start, it->pmem, it->size) )
-//#else
-//         if( memcmp((u8*)pmem+it->start, it->pmem, it->size) )
-//#endif
 			return false;
 	}
 #endif

pcsx2/x86/ix86-32/iR5900AritImm.cpp

@@ -486,19 +486,6 @@ void recADDIU( void )
 ////////////////////////////////////////////////////
 void recDADDI( void ) 
 {
-#ifdef __x86_64_
-   if ( ! _Rt_ )
-   {
-      return;
-   }
-
-	MOV64MtoR( RAX, (int)&cpuRegs.GPR.r[ _Rs_ ].UL[ 0 ] );
-	if ( _Imm_ != 0 )
-	   {
-		   ADD64ItoR( EAX, _Imm_ );
-	   }
-	MOV64RtoM( (int)&cpuRegs.GPR.r[ _Rt_ ].UL[ 0 ], RAX );
-#else
    if ( ! _Rt_ )
    {
       return;
@@ -520,7 +507,6 @@ void recDADDI( void )
 	   }
 	   MOV32RtoM( (int)&cpuRegs.GPR.r[ _Rt_ ].UL[ 0 ], EAX );
 	   MOV32RtoM( (int)&cpuRegs.GPR.r[ _Rt_ ].UL[ 1 ], EDX );
-#endif  
 }
 
 ////////////////////////////////////////////////////