Updated CPU detection code for SSE3 and SSE4 and removed obsolete 3DNow! and Altivec support.

configure.in

@@ -501,33 +501,6 @@ 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 $amd3dnow_CFLAGS"
-        fi
-    fi
-
     AC_ARG_ENABLE(sse,
 AC_HELP_STRING([--enable-sse], [use SSE assembly routines [[default=yes]]]),
                   , enable_sse=yes)
@@ -599,82 +572,6 @@ AC_HELP_STRING([--enable-sse2], [use SSE2 assembly routines [[default=no]]]),
             EXTRA_CFLAGS="$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

@@ -82,7 +82,6 @@
 #undef HAVE_MATH_H
 #undef HAVE_ICONV_H
 #undef HAVE_SIGNAL_H
-#undef HAVE_ALTIVEC_H
 
 /* C library functions */
 #undef HAVE_MALLOC
@@ -303,6 +302,5 @@
 
 /* Enable assembly routines */
 #undef SDL_ASSEMBLY_ROUTINES
-#undef SDL_ALTIVEC_BLITTERS
 
 #endif /* _SDL_config_h */

include/SDL_config_macosx.h

@@ -168,8 +168,5 @@
 
 /* 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

@@ -69,21 +69,6 @@ extern DECLSPEC SDL_bool SDLCALL SDL_HasRDTSC(void);
  */
 extern DECLSPEC SDL_bool SDLCALL SDL_HasMMX(void);
 
-/**
- *  This function returns true if the CPU has MMX Ext.\ features.
- */
-extern DECLSPEC SDL_bool SDLCALL SDL_HasMMXExt(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 3DNow!\ Ext.\ features.
- */
-extern DECLSPEC SDL_bool SDLCALL SDL_Has3DNowExt(void);
-
 /**
  *  This function returns true if the CPU has SSE features.
  */
@@ -95,9 +80,15 @@ extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE(void);
 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE2(void);
 
 /**
- *  This function returns true if the CPU has AltiVec features.
+ *  This function returns true if the CPU has SSE3 features.
+ */
+extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE3(void);
+
+/**
+ *  This function returns true if the CPU has SSE4 features.
  */
-extern DECLSPEC SDL_bool SDLCALL SDL_HasAltiVec(void);
+extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE4(void);
+
 
 /* Ends C function definitions when using C++ */
 #ifdef __cplusplus

src/cpuinfo/SDL_cpuinfo.c

@@ -32,36 +32,17 @@
 #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_MMXEXT  0x00000004
-#define CPU_HAS_3DNOW   0x00000010
-#define CPU_HAS_3DNOWEXT 0x00000020
-#define CPU_HAS_SSE     0x00000040
-#define CPU_HAS_SSE2    0x00000080
-#define CPU_HAS_ALTIVEC 0x00000100
-
-#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 */
+#define CPU_HAS_SSE     0x00000010
+#define CPU_HAS_SSE2    0x00000020
+#define CPU_HAS_SSE3    0x00000040
+#define CPU_HAS_SSE4    0x00000080
+
 
 static __inline__ int
 CPU_haveCPUID(void)
@@ -201,20 +182,6 @@ CPU_getCPUIDFeatures(void)
     return features;
 }
 
-static __inline__ int
-CPU_getCPUIDFeaturesExt(void)
-{
-    int features = 0;
-    int a, b, c, d;
-
-    cpuid(0x80000000, a, b, c, d);
-    if (a >= 0x80000001) {
-        cpuid(0x80000001, a, b, c, d);
-        features = d;
-    }
-    return features;
-}
-
 static __inline__ int
 CPU_haveRDTSC(void)
 {
@@ -234,73 +201,53 @@ CPU_haveMMX(void)
 }
 
 static __inline__ int
-CPU_haveMMXExt(void)
+CPU_haveSSE(void)
 {
     if (CPU_haveCPUID()) {
-        return (CPU_getCPUIDFeaturesExt() & 0x00400000);
+        return (CPU_getCPUIDFeatures() & 0x02000000);
     }
     return 0;
 }
 
 static __inline__ int
-CPU_have3DNow(void)
+CPU_haveSSE2(void)
 {
     if (CPU_haveCPUID()) {
-        return (CPU_getCPUIDFeaturesExt() & 0x80000000);
+        return (CPU_getCPUIDFeatures() & 0x04000000);
     }
     return 0;
 }
 
 static __inline__ int
-CPU_have3DNowExt(void)
+CPU_haveSSE3(void)
 {
     if (CPU_haveCPUID()) {
-        return (CPU_getCPUIDFeaturesExt() & 0x40000000);
-    }
-    return 0;
-}
+        int a, b, c, d;
 
-static __inline__ int
-CPU_haveSSE(void)
-{
-    if (CPU_haveCPUID()) {
-        return (CPU_getCPUIDFeatures() & 0x02000000);
+        cpuid(0, a, b, c, d);
+        if (a >= 1) {
+            cpuid(1, a, b, c, d);
+            return (c & 0x00000001);
+        }
     }
     return 0;
 }
 
 static __inline__ int
-CPU_haveSSE2(void)
+CPU_haveSSE4(void)
 {
     if (CPU_haveCPUID()) {
-        return (CPU_getCPUIDFeatures() & 0x04000000);
+        int a, b, c, d;
+
+        cpuid(0, a, b, c, d);
+        if (a >= 1) {
+            cpuid(1, a, b, c, d);
+            return (c & 0x00000100);
+        }
     }
     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 int SDL_CPUCount = 0;
 
 int
@@ -471,23 +418,17 @@ SDL_GetCPUFeatures(void)
         if (CPU_haveMMX()) {
             SDL_CPUFeatures |= CPU_HAS_MMX;
         }
-        if (CPU_haveMMXExt()) {
-            SDL_CPUFeatures |= CPU_HAS_MMXEXT;
-        }
-        if (CPU_have3DNow()) {
-            SDL_CPUFeatures |= CPU_HAS_3DNOW;
-        }
-        if (CPU_have3DNowExt()) {
-            SDL_CPUFeatures |= CPU_HAS_3DNOWEXT;
-        }
         if (CPU_haveSSE()) {
             SDL_CPUFeatures |= CPU_HAS_SSE;
         }
         if (CPU_haveSSE2()) {
             SDL_CPUFeatures |= CPU_HAS_SSE2;
         }
-        if (CPU_haveAltiVec()) {
-            SDL_CPUFeatures |= CPU_HAS_ALTIVEC;
+        if (CPU_haveSSE3()) {
+            SDL_CPUFeatures |= CPU_HAS_SSE3;
+        }
+        if (CPU_haveSSE4()) {
+            SDL_CPUFeatures |= CPU_HAS_SSE4;
         }
     }
     return SDL_CPUFeatures;
@@ -512,54 +453,36 @@ SDL_HasMMX(void)
 }
 
 SDL_bool
-SDL_HasMMXExt(void)
-{
-    if (SDL_GetCPUFeatures() & CPU_HAS_MMXEXT) {
-        return SDL_TRUE;
-    }
-    return SDL_FALSE;
-}
-
-SDL_bool
-SDL_Has3DNow(void)
-{
-    if (SDL_GetCPUFeatures() & CPU_HAS_3DNOW) {
-        return SDL_TRUE;
-    }
-    return SDL_FALSE;
-}
-
-SDL_bool
-SDL_Has3DNowExt(void)
+SDL_HasSSE(void)
 {
-    if (SDL_GetCPUFeatures() & CPU_HAS_3DNOWEXT) {
+    if (SDL_GetCPUFeatures() & CPU_HAS_SSE) {
         return SDL_TRUE;
     }
     return SDL_FALSE;
 }
 
 SDL_bool
-SDL_HasSSE(void)
+SDL_HasSSE2(void)
 {
-    if (SDL_GetCPUFeatures() & CPU_HAS_SSE) {
+    if (SDL_GetCPUFeatures() & CPU_HAS_SSE2) {
         return SDL_TRUE;
     }
     return SDL_FALSE;
 }
 
 SDL_bool
-SDL_HasSSE2(void)
+SDL_HasSSE3(void)
 {
-    if (SDL_GetCPUFeatures() & CPU_HAS_SSE2) {
+    if (SDL_GetCPUFeatures() & CPU_HAS_SSE3) {
         return SDL_TRUE;
     }
     return SDL_FALSE;
 }
 
 SDL_bool
-SDL_HasAltiVec(void)
+SDL_HasSSE4(void)
 {
-    if (SDL_GetCPUFeatures() & CPU_HAS_ALTIVEC) {
+    if (SDL_GetCPUFeatures() & CPU_HAS_SSE4) {
         return SDL_TRUE;
     }
     return SDL_FALSE;
@@ -578,12 +501,10 @@ main()
     printf("CacheLine size: %d\n", SDL_GetCPUCacheLineSize());
     printf("RDTSC: %d\n", SDL_HasRDTSC());
     printf("MMX: %d\n", SDL_HasMMX());
-    printf("MMXExt: %d\n", SDL_HasMMXExt());
-    printf("3DNow: %d\n", SDL_Has3DNow());
-    printf("3DNowExt: %d\n", SDL_Has3DNowExt());
     printf("SSE: %d\n", SDL_HasSSE());
     printf("SSE2: %d\n", SDL_HasSSE2());
-    printf("AltiVec: %d\n", SDL_HasAltiVec());
+    printf("SSE3: %d\n", SDL_HasSSE3());
+    printf("SSE4: %d\n", SDL_HasSSE4());
     return 0;
 }
 

src/video/SDL_blit.c

@@ -100,30 +100,6 @@ 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,
@@ -145,22 +121,12 @@ 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

@@ -34,9 +34,6 @@
 #ifdef __MMX__
 #include <mmintrin.h>
 #endif
-#ifdef __3dNOW__
-#include <mm3dnow.h>
-#endif
 #ifdef __SSE__
 #include <xmmintrin.h>
 #endif
@@ -65,11 +62,8 @@
 /* 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/testplatform.c

@@ -143,13 +143,10 @@ TestCPUInfo(SDL_bool verbose)
 	printf("CPU cache line size: %d\n", SDL_GetCPUCacheLineSize());
         printf("RDTSC %s\n", SDL_HasRDTSC()? "detected" : "not detected");
         printf("MMX %s\n", SDL_HasMMX()? "detected" : "not detected");
-        printf("MMX Ext %s\n", SDL_HasMMXExt()? "detected" : "not detected");
-        printf("3DNow %s\n", SDL_Has3DNow()? "detected" : "not detected");
-        printf("3DNow Ext %s\n",
-               SDL_Has3DNowExt()? "detected" : "not detected");
         printf("SSE %s\n", SDL_HasSSE()? "detected" : "not detected");
         printf("SSE2 %s\n", SDL_HasSSE2()? "detected" : "not detected");
-        printf("AltiVec %s\n", SDL_HasAltiVec()? "detected" : "not detected");
+        printf("SSE3 %s\n", SDL_HasSSE3()? "detected" : "not detected");
+        printf("SSE4 %s\n", SDL_HasSSE4()? "detected" : "not detected");
     }
     return (0);
 }