From 02fdc8f9fc7583860fc6e4f7d73f8063997af266 Mon Sep 17 00:00:00 2001
From: arcum42 <arcum42@96395faa-99c1-11dd-bbfe-3dabce05a288>
Date: Fri, 26 Mar 2010 09:19:41 +0000
Subject: [PATCH] zzogl-pg: Use the memcpy_fast routines from Utilities in
 Linux. (Right now, Windows is using an old copy of it, and Linux wasn't using
 it at all.)

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@2774 96395faa-99c1-11dd-bbfe-3dabce05a288
---
 plugins/zzogl-pg/opengl/GLWinX11.cpp          |   6 +-
 plugins/zzogl-pg/opengl/GS.h                  |   5 +
 .../opengl/Linux/zzogl-pg/zzogl-pg.cbp        |   6 +-
 plugins/zzogl-pg/opengl/Makefile.am           |  10 +-
 plugins/zzogl-pg/opengl/memcpy_amd.cpp        |  21 +-
 plugins/zzogl-pg/opengl/x86-32.S              | 336 +++++++++---------
 6 files changed, 197 insertions(+), 187 deletions(-)

diff --git a/plugins/zzogl-pg/opengl/GLWinX11.cpp b/plugins/zzogl-pg/opengl/GLWinX11.cpp
index 746be68482..df433c52fc 100644
--- a/plugins/zzogl-pg/opengl/GLWinX11.cpp
+++ b/plugins/zzogl-pg/opengl/GLWinX11.cpp
@@ -62,9 +62,9 @@ void GLWindow::CloseWindow()
 {       
     if ( glDisplay != NULL ) 
     {
-                XCloseDisplay(glDisplay);
-                glDisplay = NULL;
-        }
+		XCloseDisplay(glDisplay);
+		glDisplay = NULL;
+	}
 }
 
 bool GLWindow::DisplayWindow(int _width, int _height)
diff --git a/plugins/zzogl-pg/opengl/GS.h b/plugins/zzogl-pg/opengl/GS.h
index 0de5d18f96..a2a181a0f4 100644
--- a/plugins/zzogl-pg/opengl/GS.h
+++ b/plugins/zzogl-pg/opengl/GS.h
@@ -872,8 +872,13 @@ char *SysLibError();					// Gets previous error loading sysbols
 void SysCloseLibrary(void *lib);		// Closes Library
 void SysMessage(const char *fmt, ...);
 
+#ifdef __LINUX__
+#include "Utilities/MemcpyFast.h"
+#define memcpy_amd memcpy_fast
+#else
 extern "C" void * memcpy_amd(void *dest, const void *src, size_t n);
 extern "C" u8 memcmp_mmx(const void *dest, const void *src, int n);
+#endif
 
 template <typename T>
 class CInterfacePtr
diff --git a/plugins/zzogl-pg/opengl/Linux/zzogl-pg/zzogl-pg.cbp b/plugins/zzogl-pg/opengl/Linux/zzogl-pg/zzogl-pg.cbp
index 97cfe70351..93529022a9 100644
--- a/plugins/zzogl-pg/opengl/Linux/zzogl-pg/zzogl-pg.cbp
+++ b/plugins/zzogl-pg/opengl/Linux/zzogl-pg/zzogl-pg.cbp
@@ -16,6 +16,9 @@
 				<Compiler>
 					<Add option="-g" />
 				</Compiler>
+				<Linker>
+					<Add library="../../../../../deps/debug/libUtilities.a" />
+				</Linker>
 			</Target>
 			<Target title="Release">
 				<Option output="../../../../../bin/plugins/ZZOgl-PG" prefix_auto="1" extension_auto="1" />
@@ -29,6 +32,7 @@
 				</Compiler>
 				<Linker>
 					<Add option="-s" />
+					<Add library="../../../../../deps/release/libUtilities.a" />
 				</Linker>
 			</Target>
 		</Build>
@@ -44,6 +48,7 @@
 			<Add directory="../../../opengl" />
 			<Add directory="../../../../../common/include" />
 			<Add directory="../../../../../3rdparty" />
+			<Add directory="../../../../../common/include/Utilities" />
 		</Compiler>
 		<Linker>
 			<Add option="`pkg-config gtk+-2.0 --libs`" />
@@ -119,7 +124,6 @@
 			<Option compilerVar="CC" />
 		</Unit>
 		<Unit filename="../../glprocs.h" />
-		<Unit filename="../../memcpy_amd.cpp" />
 		<Unit filename="../../rasterfont.cpp" />
 		<Unit filename="../../rasterfont.h" />
 		<Unit filename="../../targets.cpp" />
diff --git a/plugins/zzogl-pg/opengl/Makefile.am b/plugins/zzogl-pg/opengl/Makefile.am
index c055ca4c82..744a9a11bd 100644
--- a/plugins/zzogl-pg/opengl/Makefile.am
+++ b/plugins/zzogl-pg/opengl/Makefile.am
@@ -18,16 +18,16 @@ EXEEXT=$(preext)@so_ext@
 
 traplib_PROGRAMS=libzzoglpg
 libzzoglpg_SOURCES=
-libzzoglpg_DEPENDENCIES = libzzoglpg.a
+libzzoglpg_DEPENDENCIES = libzzoglpg.a ../../../deps/release/libUtilities.a
 libzzoglpg_LDFLAGS= @SHARED_LDFLAGS@
 libzzoglpg_LDFLAGS+=-Wl,-soname,@ZEROGS_SONAME@
 libzzoglpg_LDADD=$(libzzoglpg_a_OBJECTS)
 
 libzzoglpg_a_SOURCES = \
-GSmain.cpp  GifTransfer.cpp memcpy_amd.cpp  Regs.cpp     x86.cpp     zpipe.cpp 		Mem.cpp     \
+GSmain.cpp GLWinX11.cpp GifTransfer.cpp memcpy_amd.cpp  Regs.cpp     x86.cpp     zpipe.cpp 		Mem.cpp     \
 rasterfont.cpp  targets.cpp  zerogs.cpp  ZZoglVB.cpp  ZZoglShoots.cpp	ZZoglCreate.cpp	\
 ZZoglShaders.cpp	ZZoglCRTC.cpp	ZZoglSave.cpp	ZZoglFlush.cpp \
-Mem_Swizzle.h Mem_Tables.cpp Mem_Transmit.h
+Mem_Swizzle.h Mem_Tables.cpp Mem_Transmit.h Mem_Swizzle.cpp
 
 
 libzzoglpg_a_SOURCES += x86-32.S
@@ -36,7 +36,7 @@ if SSE2
 CCASFLAGS+= -DZEROGS_SSE2
 endif
 
-libzzoglpg_a_DEPENDENCIES = Linux/libZeroGSLinux.a
-libzzoglpg_LDADD += Linux/libZeroGSLinux.a
+libzzoglpg_a_DEPENDENCIES = Linux/libZeroGSLinux.a ../../../deps/release/libUtilities.a
+libzzoglpg_LDADD += Linux/libZeroGSLinux.a ../../../deps/release/libUtilities.a
 
 SUBDIRS = Linux ZeroGSShaders .
diff --git a/plugins/zzogl-pg/opengl/memcpy_amd.cpp b/plugins/zzogl-pg/opengl/memcpy_amd.cpp
index dfd1119330..2477078f6b 100644
--- a/plugins/zzogl-pg/opengl/memcpy_amd.cpp
+++ b/plugins/zzogl-pg/opengl/memcpy_amd.cpp
@@ -75,11 +75,10 @@ MEMCPY_AMD.CPP
 #include <windows.h>
 #endif
 
+#if defined(_MSC_VER) && !defined(__x86_64__)
 extern "C" {
 #include "PS2Etypes.h"
 
-#if defined(_MSC_VER) && !defined(__x86_64__)
-
 void * memcpy_amd(void *dest, const void *src, size_t n)
 {
     __asm {
@@ -460,19 +459,21 @@ End:
 	}
 }
 
+}
 #else // _MSC_VER
 // assume gcc or mingw or win x64
 
 #include <memory.h>
 #include <string.h>
-
-void * memcpy_amd(void *dest, const void *src, size_t n)
-{
-memcpy(dest, src, n);
-return dest;
-}
-
+#include "PS2Etypes.h"
+#include "Utilities/MemcpyFast.h"
+//void * memcpy_amd(void *dest, const void *src, size_t n)
+//{
+//	//memcpy(dest, src, n);
+//	memcpy_fast(dest, src, n);
+//	return dest;
+//}
+#define memcpy_amd memcpy_fast
 
 #endif
 
-}
diff --git a/plugins/zzogl-pg/opengl/x86-32.S b/plugins/zzogl-pg/opengl/x86-32.S
index 73b11f11d6..8b3752d545 100644
--- a/plugins/zzogl-pg/opengl/x86-32.S
+++ b/plugins/zzogl-pg/opengl/x86-32.S
@@ -23,174 +23,174 @@
 ## ~10 times faster than standard memcmp
 ## (zerofrog)
 #u8 memcmp_mmx(const void* src1, const void* src2, int cmpsize)
-.globl memcmp_mmx
-	.type memcmp_mmx, @function
-memcmp_mmx:
-        push %esi
-        mov %ecx, dword ptr [%esp+16]
-		mov %edx, dword ptr [%esp+8]
-		mov %esi, dword ptr [%esp+12]
-
-		cmp %ecx, 32
-		jl Done4
-
-		// custom test first 8 to make sure things are ok
-		movq %mm0, [%esi]
-		movq %mm1, [%esi+8]
-		pcmpeqd %mm0, [%edx]
-		pcmpeqd %mm1, [%edx+8]
-		pand %mm0, %mm1
-		movq %mm2, [%esi+16]
-		pmovmskb %eax, %mm0
-		movq %mm3, [%esi+24]
-
-		// check if eq
-		cmp %eax, 0xff
-		je NextComp
-		mov %eax, 1
-		jmp End
-
-NextComp:
-		pcmpeqd %mm2, [%edx+16]
-		pcmpeqd %mm3, [%edx+24]
-		pand %mm2, %mm3
-		pmovmskb %eax, %mm2
-
-		sub %ecx, 32
-		add %esi, 32
-		add %edx, 32
-
-		// check if eq
-		cmp %eax, 0xff
-		je ContinueTest
-		mov %eax, 1
-		jmp End
-
-		cmp %ecx, 64
-		jl Done8
-
-Cmp8:
-		movq %mm0, [%esi]
-		movq %mm1, [%esi+8]
-		movq %mm2, [%esi+16]
-		movq %mm3, [%esi+24]
-		movq %mm4, [%esi+32]
-		movq %mm5, [%esi+40]
-		movq %mm6, [%esi+48]
-		movq %mm7, [%esi+56]
-		pcmpeqd %mm0, [%edx]
-		pcmpeqd %mm1, [%edx+8]
-		pcmpeqd %mm2, [%edx+16]
-		pcmpeqd %mm3, [%edx+24]
-		pand %mm0, %mm1
-		pcmpeqd %mm4, [%edx+32]
-		pand %mm0, %mm2
-		pcmpeqd %mm5, [%edx+40]
-		pand %mm0, %mm3
-		pcmpeqd %mm6, [%edx+48]
-		pand %mm0, %mm4
-		pcmpeqd %mm7, [%edx+56]
-		pand %mm0, %mm5
-		pand %mm0, %mm6
-		pand %mm0, %mm7
-		pmovmskb %eax, %mm0
-		
-		// check if eq
-		cmp %eax, 0xff
-		je Continue
-		mov %eax, 1
-		jmp End
-
-Continue:
-		sub %ecx, 64
-		add %esi, 64
-		add %edx, 64
-ContinueTest:
-		cmp %ecx, 64
-		jge Cmp8
-
-Done8:
-		test %ecx, 0x20
-		jz Done4
-		movq %mm0, [%esi]
-		movq %mm1, [%esi+8]
-		movq %mm2, [%esi+16]
-		movq %mm3, [%esi+24]
-		pcmpeqd %mm0, [%edx]
-		pcmpeqd %mm1, [%edx+8]
-		pcmpeqd %mm2, [%edx+16]
-		pcmpeqd %mm3, [%edx+24]
-		pand %mm0, %mm1
-		pand %mm0, %mm2
-		pand %mm0, %mm3
-		pmovmskb %eax, %mm0
-		sub %ecx, 32
-		add %esi, 32
-		add %edx, 32
-
-		// check if eq
-		cmp %eax, 0xff
-		je Done4
-		mov %eax, 1
-		jmp End
-
-Done4:
-		cmp %ecx, 24
-		jne Done2
-		movq %mm0, [%esi]
-		movq %mm1, [%esi+8]
-		movq %mm2, [%esi+16]
-		pcmpeqd %mm0, [%edx]
-		pcmpeqd %mm1, [%edx+8]
-		pcmpeqd %mm2, [%edx+16]
-		pand %mm0, %mm1
-		pand %mm0, %mm2
-		pmovmskb %eax, %mm0
-
-		// check if eq
-		cmp %eax, 0xff
-		setne %al
-		jmp End
-
-Done2:
-		cmp %ecx, 16
-		jne Done1
-
-		movq %mm0, [%esi]
-		movq %mm1, [%esi+8]
-		pcmpeqd %mm0, [%edx]
-		pcmpeqd %mm1, [%edx+8]
-		pand %mm0, %mm1
-		pmovmskb %eax, %mm0
-
-		// check if eq
-		cmp %eax, 0xff
-		setne %al
-		jmp End
-
-Done1:
-		cmp %ecx, 8
-		jne Done
-
-		mov %eax, [%esi]
-		mov %esi, [%esi+4]
-		cmp %eax, [%edx]
-		je Next
-		mov %eax, 1
-		jmp End
-
-Next:
-		cmp %esi, [%edx+4]
-		setne %al
-		jmp End
-
-Done:
-		xor %eax, %eax
-
-End:
-		pop %esi
-		emms
-        ret
+#.globl memcmp_mmx
+#	.type memcmp_mmx, @function
+#memcmp_mmx:
+#        push %esi
+#        mov %ecx, dword ptr [%esp+16]
+#		mov %edx, dword ptr [%esp+8]
+#		mov %esi, dword ptr [%esp+12]
+#
+#		cmp %ecx, 32
+#		jl Done4
+#
+#		// custom test first 8 to make sure things are ok
+#		movq %mm0, [%esi]
+#		movq %mm1, [%esi+8]
+#		pcmpeqd %mm0, [%edx]
+#		pcmpeqd %mm1, [%edx+8]
+#		pand %mm0, %mm1
+#		movq %mm2, [%esi+16]
+#		pmovmskb %eax, %mm0
+#		movq %mm3, [%esi+24]
+#
+#		// check if eq
+#		cmp %eax, 0xff
+#		je NextComp
+#		mov %eax, 1
+#		jmp End
+#
+#NextComp:
+#		pcmpeqd %mm2, [%edx+16]
+#		pcmpeqd %mm3, [%edx+24]
+#		pand %mm2, %mm3
+#		pmovmskb %eax, %mm2
+#
+#		sub %ecx, 32
+#		add %esi, 32
+#		add %edx, 32
+#
+#		// check if eq
+#		cmp %eax, 0xff
+#		je ContinueTest
+#		mov %eax, 1
+#		jmp End
+#
+#		cmp %ecx, 64
+#		jl Done8
+#
+#Cmp8:
+#		movq %mm0, [%esi]
+#		movq %mm1, [%esi+8]
+#		movq %mm2, [%esi+16]
+#		movq %mm3, [%esi+24]
+#		movq %mm4, [%esi+32]
+#		movq %mm5, [%esi+40]
+#		movq %mm6, [%esi+48]
+#		movq %mm7, [%esi+56]
+#		pcmpeqd %mm0, [%edx]
+#		pcmpeqd %mm1, [%edx+8]
+#		pcmpeqd %mm2, [%edx+16]
+#		pcmpeqd %mm3, [%edx+24]
+#		pand %mm0, %mm1
+#		pcmpeqd %mm4, [%edx+32]
+#		pand %mm0, %mm2
+#		pcmpeqd %mm5, [%edx+40]
+#		pand %mm0, %mm3
+#		pcmpeqd %mm6, [%edx+48]
+#		pand %mm0, %mm4
+#		pcmpeqd %mm7, [%edx+56]
+#		pand %mm0, %mm5
+#		pand %mm0, %mm6
+#		pand %mm0, %mm7
+#		pmovmskb %eax, %mm0
+#		
+#		// check if eq
+#		cmp %eax, 0xff
+#		je Continue
+#		mov %eax, 1
+#		jmp End
+#
+#Continue:
+#		sub %ecx, 64
+#		add %esi, 64
+#		add %edx, 64
+#ContinueTest:
+#		cmp %ecx, 64
+#		jge Cmp8
+#
+#Done8:
+#		test %ecx, 0x20
+#		jz Done4
+#		movq %mm0, [%esi]
+#		movq %mm1, [%esi+8]
+#		movq %mm2, [%esi+16]
+#		movq %mm3, [%esi+24]
+#		pcmpeqd %mm0, [%edx]
+#		pcmpeqd %mm1, [%edx+8]
+#		pcmpeqd %mm2, [%edx+16]
+#		pcmpeqd %mm3, [%edx+24]
+#		pand %mm0, %mm1
+#		pand %mm0, %mm2
+#		pand %mm0, %mm3
+#		pmovmskb %eax, %mm0
+#		sub %ecx, 32
+#		add %esi, 32
+#		add %edx, 32
+#
+#		// check if eq
+#		cmp %eax, 0xff
+#		je Done4
+#		mov %eax, 1
+#		jmp End
+#
+#Done4:
+#		cmp %ecx, 24
+#		jne Done2
+#		movq %mm0, [%esi]
+#		movq %mm1, [%esi+8]
+#		movq %mm2, [%esi+16]
+#		pcmpeqd %mm0, [%edx]
+#		pcmpeqd %mm1, [%edx+8]
+#		pcmpeqd %mm2, [%edx+16]
+#		pand %mm0, %mm1
+#		pand %mm0, %mm2
+#		pmovmskb %eax, %mm0
+#
+#		// check if eq
+#		cmp %eax, 0xff
+#		setne %al
+#		jmp End
+#
+#Done2:
+#		cmp %ecx, 16
+#		jne Done1
+#
+#		movq %mm0, [%esi]
+#		movq %mm1, [%esi+8]
+#		pcmpeqd %mm0, [%edx]
+#		pcmpeqd %mm1, [%edx+8]
+#		pand %mm0, %mm1
+#		pmovmskb %eax, %mm0
+#
+#		// check if eq
+#		cmp %eax, 0xff
+#		setne %al
+#		jmp End
+#
+#Done1:
+#		cmp %ecx, 8
+#		jne Done
+#
+#		mov %eax, [%esi]
+#		mov %esi, [%esi+4]
+#		cmp %eax, [%edx]
+#		je Next
+#		mov %eax, 1
+#		jmp End
+#
+#Next:
+#		cmp %esi, [%edx+4]
+#		setne %al
+#		jmp End
+#
+#Done:
+#		xor %eax, %eax
+#
+#End:
+#		pop %esi
+#		emms
+#        ret
 
 
 #ifdef ZEROGS_SSE2