Re-added the 3DNow! and AltiVec instruction support.

VisualC/SDL/SDL_VS2005.vcproj

@@ -52,7 +52,7 @@
 				Name="VCCLCompilerTool"
 				Optimization="0"
 				AdditionalIncludeDirectories="..\..\include"
-				PreprocessorDefinitions="_DEBUG;_WINDOWS;_WIN32_WINNT=0x0400;__MMX__;__3dNOW__;__SSE__;__SSE2__"
+				PreprocessorDefinitions="_DEBUG;_WINDOWS"
 				RuntimeLibrary="2"
 				BufferSecurityCheck="false"
 				UsePrecompiledHeader="0"
@@ -231,7 +231,7 @@
 				InlineFunctionExpansion="1"
 				EnableIntrinsicFunctions="false"
 				AdditionalIncludeDirectories="..\..\include"
-				PreprocessorDefinitions="NDEBUG;_WINDOWS;_WIN32_WINNT=0x0400;__MMX__;__3dNOW__;__SSE__;__SSE2__"
+				PreprocessorDefinitions="NDEBUG;_WINDOWS"
 				StringPooling="true"
 				RuntimeLibrary="2"
 				BufferSecurityCheck="false"

VisualC/SDL/SDL_VS2008.vcproj

@@ -52,7 +52,7 @@
 				Name="VCCLCompilerTool"
 				Optimization="0"
 				AdditionalIncludeDirectories="..\..\include"
-				PreprocessorDefinitions="_DEBUG;_WINDOWS;_WIN32_WINNT=0x0400;__MMX__;__3dNOW__;__SSE__;__SSE2__"
+				PreprocessorDefinitions="_DEBUG;_WINDOWS"
 				RuntimeLibrary="3"
 				BufferSecurityCheck="false"
 				WarningLevel="3"
@@ -223,7 +223,7 @@
 				InlineFunctionExpansion="1"
 				EnableIntrinsicFunctions="false"
 				AdditionalIncludeDirectories="..\..\include"
-				PreprocessorDefinitions="NDEBUG;_WINDOWS;_WIN32_WINNT=0x0400;__MMX__;__3dNOW__;__SSE__;__SSE2__"
+				PreprocessorDefinitions="NDEBUG;_WINDOWS"
 				StringPooling="true"
 				RuntimeLibrary="2"
 				BufferSecurityCheck="false"

VisualC/SDL/SDL_VS2010.vcxproj

@@ -83,7 +83,7 @@
     <ClCompile>
       <Optimization>Disabled</Optimization>
       <AdditionalIncludeDirectories>..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>_DEBUG;_WINDOWS;_WIN32_WINNT=0x0400;__MMX__;__3dNOW__;__SSE__;__SSE2__;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>_DEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
       <BufferSecurityCheck>false</BufferSecurityCheck>
       <PrecompiledHeader>
@@ -152,7 +152,7 @@
       <InlineFunctionExpansion>OnlyExplicitInline</InlineFunctionExpansion>
       <IntrinsicFunctions>false</IntrinsicFunctions>
       <AdditionalIncludeDirectories>..\..\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>NDEBUG;_WINDOWS;_WIN32_WINNT=0x0400;__MMX__;__3dNOW__;__SSE__;__SSE2__;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>NDEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <StringPooling>true</StringPooling>
       <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
       <BufferSecurityCheck>false</BufferSecurityCheck>
@@ -446,4 +446,4 @@
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
   </ImportGroup>
-</Project>
+</Project>
\ No newline at end of file

configure.in

@@ -501,6 +501,33 @@ AC_HELP_STRING([--enable-mmx], [use MMX assembly routines [[default=yes]]]),
         fi
     fi
 
+    AC_ARG_ENABLE(3dnow,
+AC_HELP_STRING([--enable-3dnow], [use MMX assembly routines [[default=yes]]]),
+                  , enable_3dnow=yes)
+    if test x$enable_3dnow = xyes; then
+        save_CFLAGS="$CFLAGS"
+        have_gcc_3dnow=no
+        AC_MSG_CHECKING(for GCC -m3dnow option)
+        amd3dnow_CFLAGS="-m3dnow"
+        CFLAGS="$save_CFLAGS $amd3dnow_CFLAGS"
+
+        AC_TRY_COMPILE([
+        #include <mm3dnow.h>
+        #ifndef __3dNOW__
+        #error Assembler CPP flag not enabled
+        #endif
+        ],[
+        ],[
+        have_gcc_3dnow=yes
+        ])
+        AC_MSG_RESULT($have_gcc_3dnow)
+        CFLAGS="$save_CFLAGS"
+
+        if test x$have_gcc_3dnow = xyes; then
+            EXTRA_CFLAGS="$EXTRA_CFLAGS $amd3dnow_CFLAGS"
+        fi
+    fi
+
     AC_ARG_ENABLE(sse,
 AC_HELP_STRING([--enable-sse], [use SSE assembly routines [[default=yes]]]),
                   , enable_sse=yes)
@@ -572,6 +599,82 @@ AC_HELP_STRING([--enable-sse2], [use SSE2 assembly routines [[default=no]]]),
             EXTRA_CFLAGS="$EXTRA_CFLAGS $sse2_CFLAGS"
         fi
     fi
+
+    AC_ARG_ENABLE(altivec,
+AC_HELP_STRING([--enable-altivec], [use Altivec assembly routines [[default=yes]]]),
+                  , enable_altivec=yes)
+    if test x$enable_altivec = xyes; then
+        save_CFLAGS="$CFLAGS"
+        have_gcc_altivec=no
+        have_altivec_h_hdr=no
+        altivec_CFLAGS="-maltivec"
+        CFLAGS="$save_CFLAGS $altivec_CFLAGS"
+
+        AC_MSG_CHECKING(for Altivec with GCC altivec.h and -maltivec option)
+        AC_TRY_COMPILE([
+        #include <altivec.h>
+        vector unsigned int vzero() {
+            return vec_splat_u32(0);
+        }
+        ],[
+        ],[
+        have_gcc_altivec=yes
+        have_altivec_h_hdr=yes
+        ])
+        AC_MSG_RESULT($have_gcc_altivec)
+
+        if test x$have_gcc_altivec = xno; then
+            AC_MSG_CHECKING(for Altivec with GCC -maltivec option)
+            AC_TRY_COMPILE([
+            vector unsigned int vzero() {
+                return vec_splat_u32(0);
+            }
+            ],[
+            ],[
+            have_gcc_altivec=yes
+            ])
+            AC_MSG_RESULT($have_gcc_altivec)
+        fi
+
+        if test x$have_gcc_altivec = xno; then
+            AC_MSG_CHECKING(for Altivec with GCC altivec.h and -faltivec option)
+            altivec_CFLAGS="-faltivec"
+            CFLAGS="$save_CFLAGS $altivec_CFLAGS"
+            AC_TRY_COMPILE([
+            #include <altivec.h>
+            vector unsigned int vzero() {
+                return vec_splat_u32(0);
+            }
+            ],[
+            ],[
+            have_gcc_altivec=yes
+            have_altivec_h_hdr=yes
+            ])
+            AC_MSG_RESULT($have_gcc_altivec)
+        fi
+
+        if test x$have_gcc_altivec = xno; then
+            AC_MSG_CHECKING(for Altivec with GCC -faltivec option)
+            AC_TRY_COMPILE([
+            vector unsigned int vzero() {
+                return vec_splat_u32(0);
+            }
+            ],[
+            ],[
+            have_gcc_altivec=yes
+            ])
+            AC_MSG_RESULT($have_gcc_altivec)
+        fi
+        CFLAGS="$save_CFLAGS"
+
+        if test x$have_gcc_altivec = xyes; then
+            AC_DEFINE(SDL_ALTIVEC_BLITTERS)
+            if test x$have_altivec_h_hdr = xyes; then
+              AC_DEFINE(HAVE_ALTIVEC_H)
+            fi
+            EXTRA_CFLAGS="$EXTRA_CFLAGS $altivec_CFLAGS"
+        fi
+    fi
 fi
 
 dnl See if the OSS audio interface is supported

include/SDL_config.h.in

@@ -34,7 +34,7 @@
 
 /* Make sure that this isn't included by Visual C++ */
 #ifdef _MSC_VER
-#error You should copy include/SDL_config.h.default to include/SDL_config.h
+#error You should run hg revert SDL_config.h 
 #endif
 
 /* C language features */
@@ -82,6 +82,7 @@
 #undef HAVE_MATH_H
 #undef HAVE_ICONV_H
 #undef HAVE_SIGNAL_H
+#undef HAVE_ALTIVEC_H
 
 /* C library functions */
 #undef HAVE_MALLOC
@@ -302,5 +303,6 @@
 
 /* Enable assembly routines */
 #undef SDL_ASSEMBLY_ROUTINES
+#undef SDL_ALTIVEC_BLITTERS
 
 #endif /* _SDL_config_h */

include/SDL_config_macosx.h

@@ -168,5 +168,8 @@
 
 /* Enable assembly routines */
 #define SDL_ASSEMBLY_ROUTINES	1
+#ifdef __ppc__
+#define SDL_ALTIVEC_BLITTERS	1
+#endif
 
 #endif /* _SDL_config_macosx_h */

include/SDL_cpuinfo.h

@@ -31,6 +31,34 @@
 
 #include "SDL_stdinc.h"
 
+/* Need to do this here because intrin.h has C++ code in it */
+/* Visual Studio 2005 has a bug where intrin.h conflicts with winnt.h */
+#if defined(_MSC_VER) && (_MSC_VER >= 1500) && !defined(_WIN32_WCE)
+#include <intrin.h>
+#define __MMX__
+#define __3dNOW__
+#define __SSE__
+#define __SSE2__
+#elif defined(__MINGW64_VERSION_MAJOR)
+#include <intrin.h>
+#else
+#ifdef __MMX__
+#include <mmintrin.h>
+#endif
+#ifdef __3dNOW__
+#include <mm3dnow.h>
+#endif
+#ifdef __SSE__
+#include <xmmintrin.h>
+#endif
+#ifdef __SSE2__
+#include <emmintrin.h>
+#endif
+#ifdef HAVE_ALTIVEC_H
+#include <altivec.h>
+#endif
+#endif
+
 #include "begin_code.h"
 /* Set up for C function definitions, even when using C++ */
 #ifdef __cplusplus
@@ -64,11 +92,21 @@ extern DECLSPEC int SDLCALL SDL_GetCPUCacheLineSize(void);
  */
 extern DECLSPEC SDL_bool SDLCALL SDL_HasRDTSC(void);
 
+/**
+ *  This function returns true if the CPU has AltiVec features.
+ */
+extern DECLSPEC SDL_bool SDLCALL SDL_HasAltiVec(void);
+
 /**
  *  This function returns true if the CPU has MMX features.
  */
 extern DECLSPEC SDL_bool SDLCALL SDL_HasMMX(void);
 
+/**
+ *  This function returns true if the CPU has 3DNow! features.
+ */
+extern DECLSPEC SDL_bool SDLCALL SDL_Has3DNow(void);
+
 /**
  *  This function returns true if the CPU has SSE features.
  */

src/cpuinfo/SDL_cpuinfo.c

@@ -32,18 +32,37 @@
 #include <sys/types.h>
 #include <sys/sysctl.h>
 #endif
+#if defined(__MACOSX__) && (defined(__ppc__) || defined(__ppc64__))
+#include <sys/sysctl.h>         /* For AltiVec check */
+#elif SDL_ALTIVEC_BLITTERS && HAVE_SETJMP
+#include <signal.h>
+#include <setjmp.h>
+#endif
 #ifdef __WIN32__
 #include "../core/windows/SDL_windows.h"
 #endif
 
 #define CPU_HAS_RDTSC   0x00000001
-#define CPU_HAS_MMX     0x00000002
+#define CPU_HAS_ALTIVEC 0x00000002
+#define CPU_HAS_MMX     0x00000004
+#define CPU_HAS_3DNOW   0x00000008
 #define CPU_HAS_SSE     0x00000010
 #define CPU_HAS_SSE2    0x00000020
 #define CPU_HAS_SSE3    0x00000040
-#define CPU_HAS_SSE41   0x00000080
-#define CPU_HAS_SSE42   0x00000100
-
+#define CPU_HAS_SSE41   0x00000100
+#define CPU_HAS_SSE42   0x00000200
+
+#if SDL_ALTIVEC_BLITTERS && HAVE_SETJMP && !__MACOSX__
+/* This is the brute force way of detecting instruction sets...
+   the idea is borrowed from the libmpeg2 library - thanks!
+ */
+static jmp_buf jmpbuf;
+static void
+illegal_instruction(int sig)
+{
+    longjmp(jmpbuf, 1);
+}
+#endif /* HAVE_SETJMP */
 
 static __inline__ int
 CPU_haveCPUID(void)
@@ -192,6 +211,29 @@ CPU_haveRDTSC(void)
     return 0;
 }
 
+static __inline__ int
+CPU_haveAltiVec(void)
+{
+    volatile int altivec = 0;
+#if defined(__MACOSX__) && (defined(__ppc__) || defined(__ppc64__))
+    int selectors[2] = { CTL_HW, HW_VECTORUNIT };
+    int hasVectorUnit = 0;
+    size_t length = sizeof(hasVectorUnit);
+    int error = sysctl(selectors, 2, &hasVectorUnit, &length, NULL, 0);
+    if (0 == error)
+        altivec = (hasVectorUnit != 0);
+#elif SDL_ALTIVEC_BLITTERS && HAVE_SETJMP
+    void (*handler) (int sig);
+    handler = signal(SIGILL, illegal_instruction);
+    if (setjmp(jmpbuf) == 0) {
+        asm volatile ("mtspr 256, %0\n\t" "vand %%v0, %%v0, %%v0"::"r" (-1));
+        altivec = 1;
+    }
+    signal(SIGILL, handler);
+#endif
+    return altivec;
+}
+
 static __inline__ int
 CPU_haveMMX(void)
 {
@@ -201,6 +243,21 @@ CPU_haveMMX(void)
     return 0;
 }
 
+static __inline__ int
+CPU_have3DNow(void)
+{
+    if (CPU_haveCPUID()) {
+        int a, b, c, d;
+
+        cpuid(0x80000000, a, b, c, d);
+        if (a >= 0x80000001) {
+            cpuid(0x80000001, a, b, c, d);
+            return (d & 0x80000000);
+        }
+    }
+    return 0;
+}
+
 static __inline__ int
 CPU_haveSSE(void)
 {
@@ -431,9 +488,15 @@ SDL_GetCPUFeatures(void)
         if (CPU_haveRDTSC()) {
             SDL_CPUFeatures |= CPU_HAS_RDTSC;
         }
+        if (CPU_haveAltiVec()) {
+            SDL_CPUFeatures |= CPU_HAS_ALTIVEC;
+        }
         if (CPU_haveMMX()) {
             SDL_CPUFeatures |= CPU_HAS_MMX;
         }
+        if (CPU_have3DNow()) {
+            SDL_CPUFeatures |= CPU_HAS_3DNOW;
+        }
         if (CPU_haveSSE()) {
             SDL_CPUFeatures |= CPU_HAS_SSE;
         }
@@ -462,6 +525,15 @@ SDL_HasRDTSC(void)
     return SDL_FALSE;
 }
 
+SDL_bool
+SDL_HasAltiVec(void)
+{
+    if (SDL_GetCPUFeatures() & CPU_HAS_ALTIVEC) {
+        return SDL_TRUE;
+    }
+    return SDL_FALSE;
+}
+
 SDL_bool
 SDL_HasMMX(void)
 {
@@ -471,6 +543,15 @@ SDL_HasMMX(void)
     return SDL_FALSE;
 }
 
+SDL_bool
+SDL_Has3DNow(void)
+{
+    if (SDL_GetCPUFeatures() & CPU_HAS_3DNOW) {
+        return SDL_TRUE;
+    }
+    return SDL_FALSE;
+}
+
 SDL_bool
 SDL_HasSSE(void)
 {
@@ -528,7 +609,9 @@ main()
     printf("CPU name: %s\n", SDL_GetCPUName());
     printf("CacheLine size: %d\n", SDL_GetCPUCacheLineSize());
     printf("RDTSC: %d\n", SDL_HasRDTSC());
+    printf("Altivec: %d\n", SDL_HasAltiVec());
     printf("MMX: %d\n", SDL_HasMMX());
+    printf("3DNow: %d\n", SDL_Has3DNow());
     printf("SSE: %d\n", SDL_HasSSE());
     printf("SSE2: %d\n", SDL_HasSSE2());
     printf("SSE3: %d\n", SDL_HasSSE3());

src/video/SDL_blit.c

@@ -100,6 +100,30 @@ SDL_SoftBlit(SDL_Surface * src, SDL_Rect * srcrect,
     return (okay ? 0 : -1);
 }
 
+#ifdef __MACOSX__
+#include <sys/sysctl.h>
+
+static SDL_bool
+SDL_UseAltivecPrefetch()
+{
+    const char key[] = "hw.l3cachesize";
+    u_int64_t result = 0;
+    size_t typeSize = sizeof(result);
+
+    if (sysctlbyname(key, &result, &typeSize, NULL, 0) == 0 && result > 0) {
+        return SDL_TRUE;
+    } else {
+        return SDL_FALSE;
+    }
+}
+#else
+static SDL_bool
+SDL_UseAltivecPrefetch()
+{
+    /* Just guess G4 */
+    return SDL_TRUE;
+}
+#endif /* __MACOSX__ */
 
 static SDL_BlitFunc
 SDL_ChooseBlitFunc(Uint32 src_format, Uint32 dst_format, int flags,
@@ -121,12 +145,22 @@ SDL_ChooseBlitFunc(Uint32 src_format, Uint32 dst_format, int flags,
             if (SDL_HasMMX()) {
                 features |= SDL_CPU_MMX;
             }
+            if (SDL_Has3DNow()) {
+                features |= SDL_CPU_3DNOW;
+            }
             if (SDL_HasSSE()) {
                 features |= SDL_CPU_SSE;
             }
             if (SDL_HasSSE2()) {
                 features |= SDL_CPU_SSE2;
             }
+            if (SDL_HasAltiVec()) {
+                if (SDL_UseAltivecPrefetch()) {
+                    features |= SDL_CPU_ALTIVEC_PREFETCH;
+                } else {
+                    features |= SDL_CPU_ALTIVEC_NOPREFETCH;
+                }
+            }
         }
     }
 

src/video/SDL_blit.h

@@ -24,24 +24,6 @@
 #ifndef _SDL_blit_h
 #define _SDL_blit_h
 
-#ifdef __MINGW32__
-#include <_mingw.h>
-#endif
-
-#if defined(__MINGW32__) && defined(__MINGW64_VERSION_MAJOR)
-#include <intrin.h>
-#else
-#ifdef __MMX__
-#include <mmintrin.h>
-#endif
-#ifdef __SSE__
-#include <xmmintrin.h>
-#endif
-#ifdef __SSE2__
-#include <emmintrin.h>
-#endif
-#endif
-
 #include "SDL_cpuinfo.h"
 #include "SDL_endian.h"
 #include "SDL_surface.h"
@@ -62,8 +44,11 @@
 /* SDL blit CPU flags */
 #define SDL_CPU_ANY                 0x00000000
 #define SDL_CPU_MMX                 0x00000001
+#define SDL_CPU_3DNOW               0x00000002
 #define SDL_CPU_SSE                 0x00000004
 #define SDL_CPU_SSE2                0x00000008
+#define SDL_CPU_ALTIVEC_PREFETCH    0x00000010
+#define SDL_CPU_ALTIVEC_NOPREFETCH  0x00000020
 
 typedef struct
 {

test/automated/platform/platform.c

@@ -158,7 +158,9 @@ int test_platform (void)
    SDL_ATprintVerbose( 1, "CPU count: %d\n", SDL_GetCPUCount());
    SDL_ATprintVerbose( 1, "Available extensions:\n" );
    SDL_ATprintVerbose( 1, "   RDTSC %s\n", SDL_HasRDTSC()? "detected" : "not detected" );
+   SDL_ATprintVerbose( 1, "   AltiVec %s\n", SDL_HasAltiVec()? "detected" : "not detected" );
    SDL_ATprintVerbose( 1, "   MMX %s\n", SDL_HasMMX()? "detected" : "not detected" );
+   SDL_ATprintVerbose( 1, "   3DNow! %s\n", SDL_Has3DNow()? "detected" : "not detected" );
    SDL_ATprintVerbose( 1, "   SSE %s\n", SDL_HasSSE()? "detected" : "not detected" );
    SDL_ATprintVerbose( 1, "   SSE2 %s\n", SDL_HasSSE2()? "detected" : "not detected" );
    SDL_ATprintVerbose( 1, "   SSE3 %s\n", SDL_HasSSE3()? "detected" : "not detected" );

test/testplatform.c

@@ -140,9 +140,11 @@ TestCPUInfo(SDL_bool verbose)
 {
     if (verbose) {
         printf("CPU count: %d\n", SDL_GetCPUCount());
-	printf("CPU cache line size: %d\n", SDL_GetCPUCacheLineSize());
+        printf("CPU cache line size: %d\n", SDL_GetCPUCacheLineSize());
         printf("RDTSC %s\n", SDL_HasRDTSC()? "detected" : "not detected");
+        printf("AltiVec %s\n", SDL_HasAltiVec()? "detected" : "not detected");
         printf("MMX %s\n", SDL_HasMMX()? "detected" : "not detected");
+        printf("3DNow! %s\n", SDL_Has3DNow()? "detected" : "not detected");
         printf("SSE %s\n", SDL_HasSSE()? "detected" : "not detected");
         printf("SSE2 %s\n", SDL_HasSSE2()? "detected" : "not detected");
         printf("SSE3 %s\n", SDL_HasSSE3()? "detected" : "not detected");