#ifndef WIN32
#include <unistd.h>
#include <sys/wait.h>
+#include <signal.h>
+#else
+#include <errno.h>
#endif
+#include <assert.h>
-#include "libvlc.h"
-
-#if defined(__APPLE__) && (defined(__ppc__) || defined(__ppc64__))
+#ifdef __APPLE__
#include <sys/sysctl.h>
#endif
+#include "libvlc.h"
+
+static uint32_t cpu_flags;
+
#if defined( __i386__ ) || defined( __x86_64__ ) || defined( __powerpc__ ) \
|| defined( __ppc__ ) || defined( __ppc64__ ) || defined( __powerpc64__ )
+# ifndef WIN32
static bool check_OS_capability( const char *psz_capability, pid_t pid )
{
-#ifndef WIN32
int status;
if( pid == -1 )
fprintf( stderr, " some optimizations will be disabled unless "
"you upgrade your OS\n" );
return false;
-#else
-# warning FIXME!
-# define fork() (errno = ENOSYS, -1)
- (void)pid;
- (void)psz_capability;
- return true;
-#endif
}
+
+# define check_capability(name, flag, code, input) \
+ do { \
+ pid_t pid = fork(); \
+ if( pid == 0 ) \
+ { \
+ signal(SIGILL, SIG_DFL); \
+ __asm__ __volatile__ ( code : : input ); \
+ _exit(0); \
+ } \
+ if( check_OS_capability((name), pid )) \
+ i_capabilities |= (flag); \
+ } while(0)
+
+# else /* WIN32 */
+# define check_capability(name, flag, code, input) \
+ i_capabilities |= (flag);
+# endif
#endif
-/*****************************************************************************
- * CPUCapabilities: get the CPU capabilities
- *****************************************************************************
- * This function is called to list extensions the CPU may have.
- *****************************************************************************/
-uint32_t CPUCapabilities( void )
+/**
+ * Determines the CPU capabilities and stores them in cpu_flags.
+ * The result can be retrieved with vlc_CPU().
+ */
+void vlc_CPU_init (void)
{
uint32_t i_capabilities = 0;
: "a" ( reg ) \
: "cc" );
# endif
-
- i_capabilities |= CPU_CAPABILITY_FPU;
-
+ /* Check if the OS really supports the requested instructions */
# if defined (__i386__) && !defined (__i486__) && !defined (__i586__) \
&& !defined (__i686__) && !defined (__pentium4__) \
&& !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
i_capabilities |= CPU_CAPABILITY_MMXEXT;
# ifdef CAN_COMPILE_SSE
- /* We test if OS supports the SSE instructions */
- pid_t pid = fork();
- if( pid == 0 )
- {
- /* Test a SSE instruction */
- __asm__ __volatile__ ( "xorps %%xmm0,%%xmm0\n" : : );
- exit(0);
- }
- if( check_OS_capability( "SSE", pid ) )
- i_capabilities |= CPU_CAPABILITY_SSE;
+ check_capability( "SSE", CPU_CAPABILITY_SSE,
+ "xorps %%xmm0,%%xmm0\n", );
# endif
}
# endif
# if defined (__SSE2__)
i_capabilities |= CPU_CAPABILITY_SSE2;
-# elif defined (CAN_COMPILE_SSE)
+# elif defined (CAN_COMPILE_SSE2)
if( i_edx & 0x04000000 )
- {
- /* We test if OS supports the SSE2 instructions */
- pid_t pid = fork();
- if( pid == 0 )
- {
- /* Test a SSE2 instruction */
- __asm__ __volatile__ ( "movupd %%xmm0, %%xmm0\n" : : );
- exit(0);
- }
- if( check_OS_capability( "SSE2", pid ) )
- i_capabilities |= CPU_CAPABILITY_SSE2;
- }
+ check_capability( "SSE2", CPU_CAPABILITY_SSE2,
+ "movupd %%xmm0, %%xmm0\n", );
# endif
# if defined (__SSE3__)
i_capabilities |= CPU_CAPABILITY_SSE3;
# elif defined (CAN_COMPILE_SSE3)
if( i_ecx & 0x00000001 )
- {
- /* We test if OS supports the SSE3 instructions */
- pid_t pid = fork();
- if( pid == 0 )
- {
- /* Test a SSE3 instruction */
- __asm__ __volatile__ ( "movsldup %%xmm1, %%xmm0\n" : : );
- exit(0);
- }
- if( check_OS_capability( "SSE3", pid ) )
- i_capabilities |= CPU_CAPABILITY_SSE3;
- }
+ check_capability( "SSE3", CPU_CAPABILITY_SSE3,
+ "movsldup %%xmm1, %%xmm0\n", );
+# endif
+
+# if defined (__SSSE3__)
+ i_capabilities |= CPU_CAPABILITY_SSSE3;
+# elif defined (CAN_COMPILE_SSSE3)
+ if( i_ecx & 0x00000200 )
+ check_capability( "SSSE3", CPU_CAPABILITY_SSSE3,
+ "pabsw %%xmm1, %%xmm0\n", );
+# endif
+
+# if defined (__SSE4_1__)
+ i_capabilities |= CPU_CAPABILITY_SSE4_1;
+# elif defined (CAN_COMPILE_SSE4_1)
+ if( i_ecx & 0x00080000 )
+ check_capability( "SSE4.1", CPU_CAPABILITY_SSE4_1,
+ "pmaxsb %%xmm1, %%xmm0\n", );
+# endif
+
+# if defined (__SSE4_2__)
+ i_capabilities |= CPU_CAPABILITY_SSE4_2;
+# elif defined (CAN_COMPILE_SSE4_2)
+ if( i_ecx & 0x00100000 )
+ check_capability( "SSE4.2", CPU_CAPABILITY_SSE4_2,
+ "pcmpgtq %%xmm1, %%xmm0\n", );
# endif
/* test for additional capabilities */
i_capabilities |= CPU_CAPABILITY_3DNOW;
# elif defined (CAN_COMPILE_3DNOW)
if( i_edx & 0x80000000 )
- {
- pid_t pid = fork();
- if( pid == 0 )
- {
- /* Test a 3D Now! instruction */
- __asm__ __volatile__ ( "pfadd %%mm0,%%mm0\n" "femms\n" : : );
- exit(0);
- }
- if( check_OS_capability( "3D Now!", pid ) )
- i_capabilities |= CPU_CAPABILITY_3DNOW;
- }
+ check_capability( "3D Now!", CPU_CAPABILITY_3DNOW,
+ "pfadd %%mm0,%%mm0\n" "femms\n", );
# endif
if( b_amd && ( i_edx & 0x00400000 ) )
}
out:
-#elif defined( __arm__ )
-# if defined( __ARM_EABI__ ) && !defined( __SOFTFP__ )
-// i_capabilities |= CPU_CAPABILITY_FPU;
-# endif
-# if defined( __ARM_NEON__ )
+#elif defined (__arm__)
+
+# if defined (__ARM_NEON__)
i_capabilities |= CPU_CAPABILITY_NEON;
+# elif defined (CAN_COMPILE_NEON)
+# define NEED_RUNTIME_CPU_CHECK 1
+# endif
+
+# ifdef NEED_RUNTIME_CPU_CHECK
+# if defined (__linux__)
+ FILE *info = fopen ("/proc/cpuinfo", "rt");
+ if (info != NULL)
+ {
+ char *line = NULL;
+ size_t linelen = 0;
+
+ while (getline (&line, &linelen, info) != -1)
+ {
+ const char *cap;
+
+ if (strncmp (line, "Features\t:", 10))
+ continue;
+
+ /* TODO: detect other CPU features when we use them */
+# if defined (CAN_COMPILE_NEON) && !defined (__ARM_NEON__)
+ cap = strstr (line + 10, " neon");
+ if (cap != NULL && (cap[5] == '\0' || cap[5] == ' '))
+ i_capabilities |= CPU_CAPABILITY_NEON;
# endif
+ break;
+ }
+ fclose (info);
+ free (line);
+ }
+# else
+# warning Run-time CPU detection missing: optimizations disabled!
+# endif
+# endif
#elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __powerpc64__ ) \
|| defined( __ppc64__ )
- i_capabilities |= CPU_CAPABILITY_FPU;
-
-# if defined(__APPLE__)
+# if defined(__APPLE__) || defined(__OpenBSD__)
+# if defined(__OpenBSD__)
+ int selectors[2] = { CTL_MACHDEP, CPU_ALTIVEC };
+# else
int selectors[2] = { CTL_HW, HW_VECTORUNIT };
+# endif
int i_has_altivec = 0;
size_t i_length = sizeof( i_has_altivec );
int i_error = sysctl( selectors, 2, &i_has_altivec, &i_length, NULL, 0);
i_capabilities |= CPU_CAPABILITY_ALTIVEC;
# elif defined( CAN_COMPILE_ALTIVEC )
- pid_t pid = fork();
- if( pid == 0 )
- {
- asm volatile ("mtspr 256, %0\n\t"
- "vand %%v0, %%v0, %%v0"
- :
- : "r" (-1));
- exit(0);
- }
-
- if( check_OS_capability( "Altivec", pid ) )
- i_capabilities |= CPU_CAPABILITY_ALTIVEC;
+ check_capability( "Altivec", CPU_CAPABILITY_ALTIVEC,
+ "mtspr 256, %0\n\t"
+ "vand %%v0, %%v0, %%v0",
+ "r" (-1));
# endif
-#elif defined( __sparc__ )
- i_capabilities |= CPU_CAPABILITY_FPU;
+#endif
-#elif defined( _MSC_VER ) && !defined( UNDER_CE )
- i_capabilities |= CPU_CAPABILITY_FPU;
+ cpu_flags = i_capabilities;
+}
+/**
+ * Retrieves pre-computed CPU capability flags
+ */
+unsigned vlc_CPU (void)
+{
+#ifndef WIN32 /* On Windows, initialized from DllMain() instead */
+ static pthread_once_t once = PTHREAD_ONCE_INIT;
+ pthread_once (&once, vlc_CPU_init);
#endif
- return i_capabilities;
+ return cpu_flags;
}
-uint32_t cpu_flags = 0;
+void vlc_CPU_dump (vlc_object_t *obj)
+{
+ const unsigned flags = vlc_CPU();
+ char buf[200], *p = buf;
+
+#define PRINT_CAPABILITY( capability, string ) \
+ if (flags & (capability)) \
+ p += sprintf (p, "%s ", (string) )
+
+#if defined (__i386__) || defined (__x86_64__)
+ PRINT_CAPABILITY(CPU_CAPABILITY_MMX, "MMX");
+ PRINT_CAPABILITY(CPU_CAPABILITY_3DNOW, "3DNow!");
+ PRINT_CAPABILITY(CPU_CAPABILITY_MMXEXT, "MMXEXT");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSE, "SSE");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSE2, "SSE2");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSE3, "SSE3");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSSE3, "SSSE3");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSE4_1, "SSE4.1");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSE4_2, "SSE4.2");
+ PRINT_CAPABILITY(CPU_CAPABILITY_SSE4A, "SSE4A");
+
+#elif defined (__powerpc__) || defined (__ppc__) || defined (__ppc64__)
+ PRINT_CAPABILITY(CPU_CAPABILITY_ALTIVEC, "AltiVec");
+
+#elif defined (__arm__)
+ PRINT_CAPABILITY(CPU_CAPABILITY_NEON, "NEONv1");
+#endif
-/*****************************************************************************
- * vlc_CPU: get pre-computed CPU capability flags
- ****************************************************************************/
-unsigned vlc_CPU (void)
-{
- return cpu_flags;
+#if HAVE_FPU
+ p += sprintf (p, "FPU ");
+#endif
+
+ if (p > buf)
+ msg_Dbg (obj, "CPU has capabilities %s", buf);
}
+
static vlc_memcpy_t pf_vlc_memcpy = memcpy;
-static vlc_memset_t pf_vlc_memset = memset;
-void vlc_fastmem_register (vlc_memcpy_t cpy, vlc_memset_t set)
+void vlc_fastmem_register (vlc_memcpy_t cpy)
{
- if (cpy)
- pf_vlc_memcpy = cpy;
- if (set)
- pf_vlc_memset = set;
+ assert (cpy != NULL);
+ pf_vlc_memcpy = cpy;
}
/**
{
return pf_vlc_memcpy (tgt, src, n);
}
-
-/**
- * vlc_memset: fast CPU-dependent memset
- */
-void *vlc_memset (void *tgt, int c, size_t n)
-{
- return pf_vlc_memset (tgt, c, n);
-}