From 2270ba4eeeb6d71d3525475dd3586d2ea6d5289e Mon Sep 17 00:00:00 2001
From: arcum42 <arcum42@96395faa-99c1-11dd-bbfe-3dabce05a288>
Date: Sun, 1 Mar 2009 08:21:14 +0000
Subject: [PATCH] Finish the Linux implementation of memcpy_fast_. I've
 disabled it by default until I'm sure it's working right, but it can easily
 be enabled in build.sh. Should be a speed boost in Linux (which Windows
 already had), but I haven't tested it enough to be able to tell yet.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@643 96395faa-99c1-11dd-bbfe-3dabce05a288
---
 build.sh                  |  4 ++++
 pcsx2/MemcpyFast.h        | 15 ++++++++-------
 pcsx2/configure.ac        | 10 ++++++++++
 pcsx2/x86/fast_routines.S |  6 +++---
 4 files changed, 25 insertions(+), 10 deletions(-)

diff --git a/build.sh b/build.sh
index e409a5c185..966770bf59 100644
--- a/build.sh
+++ b/build.sh
@@ -11,6 +11,10 @@
 #Optimized, but a devbuild
 export PCSX2OPTIONS="--enable-sse3 --enable-sse4 --enable-devbuild --prefix `pwd`"
 
+
+#Optimized, but a devbuild - with memcpy_fast_ enabled.
+#export PCSX2OPTIONS="--enable-sse3 --enable-sse4 --enable-devbuild --enable-memcpyfast --prefix `pwd`"
+
 #Debug / Devbuild version
 #export PCSX2OPTIONS="--enable-debug --enable-devbuild --enable-sse3 --prefix `pwd`"
 
diff --git a/pcsx2/MemcpyFast.h b/pcsx2/MemcpyFast.h
index 8e2ef8c90a..844de2ff62 100644
--- a/pcsx2/MemcpyFast.h
+++ b/pcsx2/MemcpyFast.h
@@ -50,14 +50,15 @@ void _memset16_unaligned( void* dest, u16 data, size_t size );
 	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
 
-	// Currently broken.
-//#	define memcpy_fast memcpy_amd_
-//#	define memcpy_aligned memcpy_amd_
-//	extern "C" void __fastcall memcpy_amd_(void *dest, const void *src, size_t bytes);
+#endif // WIN32
 
-#endif
-
-#endif
+#endif //Header
diff --git a/pcsx2/configure.ac b/pcsx2/configure.ac
index 4e28aa7368..acd3813e59 100644
--- a/pcsx2/configure.ac
+++ b/pcsx2/configure.ac
@@ -63,6 +63,15 @@ AC_MSG_RESULT($debug)
 
 AC_CHECK_FUNCS([ _aligned_malloc _aligned_free ], AC_DEFINE(HAVE_ALIGNED_MALLOC))
 
+AC_MSG_CHECKING(turn on memcpy_fast_)
+AC_ARG_ENABLE(memcpyfast, AC_HELP_STRING([--enable-memcpyfast], [Turns on memcpy_fast - EXPERIMENTAL]),
+ memcpyfast=$enableval,memcpyfast=no)
+if test "x$memcpyfast" == xyes
+then
+ AC_DEFINE(LINUX_USE_FAST_MEMORY,1,[LINUX_USE_FAST_MEMORY])
+fi
+AC_MSG_RESULT($memcpyfast)
+
 dnl Check for dev build
 AC_MSG_CHECKING(for development build)
 AC_ARG_ENABLE(devbuild, AC_HELP_STRING([--enable-devbuild], [Special Build for developers that simplifies testing and adds extra checks]),
@@ -138,3 +147,4 @@ echo "  Force sse3?	           $sse3"
 echo " nls support?   $nls"
 echo " local plugin inis? $localinis"
 echo " custom cflags? $customcflags"
+echo " memcpy_fast? $memcpyfast"
diff --git a/pcsx2/x86/fast_routines.S b/pcsx2/x86/fast_routines.S
index a06f3749ec..be2b118af8 100644
--- a/pcsx2/x86/fast_routines.S
+++ b/pcsx2/x86/fast_routines.S
@@ -359,7 +359,7 @@ memcpy_amd_:
 $memcpy_do_align:
 	mov		%eax, 8			// a trick that's faster than rep movsb...
 	sub		%eax, %edi		// align destination to qword
-	and		%eax, 0x111b		// get the low bits
+	andb	%eax, 111		// get the low bits
 	sub		%ecx, %eax		// update copy count
 	neg		%eax				// set up to jump into the array
 	add		%eax, offset $memcpy_align_done
@@ -427,7 +427,7 @@ $memcpy_ic_2:
 	mov		%eax, %ecx		// has valid low 6 bits of the byte count
 $memcpy_ic_3:
 	shr		%eax, 2			// dword count
-	and		%eax, 0x1111b		// only look at the "remainder" bits
+	andb	%eax, 1111		// only look at the "remainder" bits
 	neg		%eax				// set up to jump into the array
 	add		%eax, offset $memcpy_last_few
 	jmp		%eax				// jump to array of movsd's
@@ -512,7 +512,7 @@ $memcpy_uc_1:				// 64-byte blocks, uncached copy
 
 $memcpy_last_few:		// dword aligned from before movsd's
 	mov		%eax, %ecx	// has valid low 2 bits of the byte count
-	and		%eax, 0x11b	// the last few cows must come home
+	andb	%eax, 11	// the last few cows must come home
 	jz		$memcpy_final	// no more, let's leave
 	rep		movsb		// the last 1, 2, or 3 bytes