#endif
#include <vlc_common.h>
+#include <vlc_cpu.h>
-#include <signal.h> /* SIGHUP, SIGINT, SIGKILL */
-#include <setjmp.h> /* longjmp, setjmp */
-
-#include "libvlc.h"
+#include <sys/types.h>
+#ifndef WIN32
+#include <unistd.h>
+#include <sys/wait.h>
+#include <signal.h>
+#else
+#include <errno.h>
+#endif
+#include <assert.h>
-#if defined(__APPLE__) && (defined(__ppc__) || defined(__ppc64__))
+#ifdef __APPLE__
#include <sys/sysctl.h>
#endif
-/*****************************************************************************
- * Local prototypes
- *****************************************************************************/
-static void SigHandler ( int );
+#include "libvlc.h"
-/*****************************************************************************
- * Global variables - they're needed for signal handling
- *****************************************************************************/
-static jmp_buf env;
-static int i_illegal;
-#if defined( __i386__ ) || defined( __x86_64__ )
-static const char *psz_capability;
+#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 )
+{
+ int status;
+
+ if( pid == -1 )
+ return false; /* fail safe :-/ */
+
+ while( waitpid( pid, &status, 0 ) == -1 );
+
+ if( WIFEXITED( status ) && WEXITSTATUS( status ) == 0 )
+ return true;
+
+ fprintf( stderr, "warning: your CPU has %s instructions, but not your "
+ "operating system.\n", psz_capability );
+ fprintf( stderr, " some optimizations will be disabled unless "
+ "you upgrade your OS\n" );
+ return false;
+}
+
+# define check_capability(name, flag, code) \
+ do { \
+ pid_t pid = fork(); \
+ if( pid == 0 ) \
+ { \
+ signal(SIGILL, SIG_DFL); \
+ __asm__ __volatile__ ( code : : ); \
+ _exit(0); \
+ } \
+ if( check_OS_capability((name), pid )) \
+ i_capabilities |= (flag); \
+ } while(0)
+
+# else /* WIN32 */
+# define check_capability(name, flag, code) \
+ i_capabilities |= (flag);
+# endif
#endif
/*****************************************************************************
*****************************************************************************/
uint32_t CPUCapabilities( void )
{
- volatile uint32_t i_capabilities = CPU_CAPABILITY_NONE;
-
-#if defined(__APPLE__) && (defined(__ppc__) || defined(__ppc64__))
- int selectors[2] = { CTL_HW, HW_VECTORUNIT };
- 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_FPU;
-
- if( i_error == 0 && i_has_altivec != 0 )
- i_capabilities |= CPU_CAPABILITY_ALTIVEC;
-
- return i_capabilities;
+ uint32_t i_capabilities = 0;
-#elif defined( __i386__ ) || defined( __x86_64__ )
- volatile unsigned int i_eax, i_ebx, i_ecx, i_edx;
- volatile bool b_amd;
+#if defined( __i386__ ) || defined( __x86_64__ )
+ unsigned int i_eax, i_ebx, i_ecx, i_edx;
+ bool b_amd;
/* Needed for x86 CPU capabilities detection */
# if defined( __x86_64__ )
: "a" ( reg ) \
: "cc" );
# endif
-
-# if defined( CAN_COMPILE_SSE ) || defined ( CAN_COMPILE_3DNOW )
- void (*pf_sigill) (int) = signal( SIGILL, SigHandler );
-# endif
-
- i_capabilities |= CPU_CAPABILITY_FPU;
-
-# if defined( __i386__ )
+ /* 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__)
/* check if cpuid instruction is supported */
asm volatile ( "push %%ebx\n\t"
"pushf\n\t"
: "cc" );
if( i_eax == i_ebx )
- {
-# if defined( CAN_COMPILE_SSE ) || defined ( CAN_COMPILE_3DNOW )
- signal( SIGILL, pf_sigill );
-# endif
- return i_capabilities;
- }
-# else
- /* x86_64 supports cpuid instruction, so we dont need to check it */
-# endif
-
- i_capabilities |= CPU_CAPABILITY_486;
+ goto out;
+# endif
/* the CPU supports the CPUID instruction - get its level */
cpuid( 0x00000000 );
+# if defined (__i386__) && !defined (__i586__) \
+ && !defined (__i686__) && !defined (__pentium4__) \
+ && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
if( !i_eax )
- {
-# if defined( CAN_COMPILE_SSE ) || defined ( CAN_COMPILE_3DNOW )
- signal( SIGILL, pf_sigill );
-# endif
- return i_capabilities;
- }
-
- /* FIXME: this isn't correct, since some 486s have cpuid */
- i_capabilities |= CPU_CAPABILITY_586;
+ goto out;
+#endif
/* borrowed from mpeg2dec */
b_amd = ( i_ebx == 0x68747541 ) && ( i_ecx == 0x444d4163 )
/* test for the MMX flag */
cpuid( 0x00000001 );
-
+# if !defined (__MMX__)
if( ! (i_edx & 0x00800000) )
- {
-# if defined( CAN_COMPILE_SSE ) || defined ( CAN_COMPILE_3DNOW )
- signal( SIGILL, pf_sigill );
-# endif
- return i_capabilities;
- }
-
+ goto out;
+# endif
i_capabilities |= CPU_CAPABILITY_MMX;
+# if defined (__SSE__)
+ i_capabilities |= CPU_CAPABILITY_MMXEXT | CPU_CAPABILITY_SSE;
+# else
if( i_edx & 0x02000000 )
{
i_capabilities |= CPU_CAPABILITY_MMXEXT;
# ifdef CAN_COMPILE_SSE
- /* We test if OS supports the SSE instructions */
- psz_capability = "SSE";
- i_illegal = 0;
-
- if( setjmp( env ) == 0 )
- {
- /* Test a SSE instruction */
- __asm__ __volatile__ ( "xorps %%xmm0,%%xmm0\n" : : );
- }
-
- if( i_illegal == 0 )
- {
- 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_SSE2)
if( i_edx & 0x04000000 )
- {
-# if defined(CAN_COMPILE_SSE)
- /* We test if OS supports the SSE instructions */
- psz_capability = "SSE2";
- i_illegal = 0;
-
- if( setjmp( env ) == 0 )
- {
- /* Test a SSE2 instruction */
- __asm__ __volatile__ ( "movupd %%xmm0, %%xmm0\n" : : );
- }
-
- if( i_illegal == 0 )
- {
- i_capabilities |= CPU_CAPABILITY_SSE2;
- }
-# endif
- }
+ 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 )
+ 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 */
cpuid( 0x80000000 );
if( i_eax < 0x80000001 )
- {
-# if defined( CAN_COMPILE_SSE ) || defined ( CAN_COMPILE_3DNOW )
- signal( SIGILL, pf_sigill );
-# endif
- return i_capabilities;
- }
+ goto out;
/* list these additional capabilities */
cpuid( 0x80000001 );
-# ifdef CAN_COMPILE_3DNOW
+# if defined (__3dNOW__)
+ i_capabilities |= CPU_CAPABILITY_3DNOW;
+# elif defined (CAN_COMPILE_3DNOW)
if( i_edx & 0x80000000 )
- {
- psz_capability = "3D Now!";
- i_illegal = 0;
-
- if( setjmp( env ) == 0 )
- {
- /* Test a 3D Now! instruction */
- __asm__ __volatile__ ( "pfadd %%mm0,%%mm0\n" "femms\n" : : );
- }
-
- if( i_illegal == 0 )
- {
- i_capabilities |= CPU_CAPABILITY_3DNOW;
- }
- }
-# endif
+ check_capability( "3D Now!", CPU_CAPABILITY_3DNOW,
+ "pfadd %%mm0,%%mm0\n" "femms\n" );
+# endif
if( b_amd && ( i_edx & 0x00400000 ) )
{
i_capabilities |= CPU_CAPABILITY_MMXEXT;
}
+out:
-# if defined( CAN_COMPILE_SSE ) || defined ( CAN_COMPILE_3DNOW )
- signal( SIGILL, pf_sigill );
-# endif
- return i_capabilities;
+#elif defined (__arm__)
-#elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __ppc64__ )
+# if defined (__ARM_NEON__)
+ i_capabilities |= CPU_CAPABILITY_NEON;
+# elif defined (CAN_COMPILE_NEON)
+# define NEED_RUNTIME_CPU_CHECK 1
+# endif
-# ifdef CAN_COMPILE_ALTIVEC
- void (*pf_sigill) (int) = signal( SIGILL, SigHandler );
+# ifdef NEED_RUNTIME_CPU_CHECK
+# if defined (__linux__)
+ FILE *info = fopen ("/proc/cpuinfo", "rt");
+ if (info != NULL)
+ {
+ char *line = NULL;
+ size_t linelen = 0;
- i_capabilities |= CPU_CAPABILITY_FPU;
+ while (getline (&line, &linelen, info) != -1)
+ {
+ const char *cap;
+
+ if (strncmp (line, "Features\t:", 10))
+ continue;
+# 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__ )
+
+# 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_illegal = 0;
+ if( i_error == 0 && i_has_altivec != 0 )
+ i_capabilities |= CPU_CAPABILITY_ALTIVEC;
- if( setjmp( env ) == 0 )
+# elif defined( CAN_COMPILE_ALTIVEC )
+ pid_t pid = fork();
+ if( pid == 0 )
{
+ signal(SIGILL, SIG_DFL);
asm volatile ("mtspr 256, %0\n\t"
"vand %%v0, %%v0, %%v0"
:
: "r" (-1));
+ _exit(0);
}
- if( i_illegal == 0 )
- {
+ if( check_OS_capability( "Altivec", pid ) )
i_capabilities |= CPU_CAPABILITY_ALTIVEC;
- }
- signal( SIGILL, pf_sigill );
-# else
- (void)SigHandler; /* Don't complain about dead code here */
# endif
- return i_capabilities;
-
-#elif defined( __sparc__ )
-
- i_capabilities |= CPU_CAPABILITY_FPU;
- return i_capabilities;
-
-#elif defined( _MSC_VER ) && !defined( UNDER_CE )
- i_capabilities |= CPU_CAPABILITY_FPU;
- return i_capabilities;
-
-#else
- /* default behaviour */
- return i_capabilities;
-
#endif
+ return i_capabilities;
}
-/*****************************************************************************
- * SigHandler: system signal handler
- *****************************************************************************
- * This function is called when an illegal instruction signal is received by
- * the program. We use this function to test OS and CPU capabilities
- *****************************************************************************/
-static void SigHandler( int i_signal )
-{
- /* Acknowledge the signal received */
- i_illegal = 1;
-
-#ifdef HAVE_SIGRELSE
- sigrelse( i_signal );
-#else
- VLC_UNUSED( i_signal );
-#endif
-
-#if defined( __i386__ )
- fprintf( stderr, "warning: your CPU has %s instructions, but not your "
- "operating system.\n", psz_capability );
- fprintf( stderr, " some optimizations will be disabled unless "
- "you upgrade your OS\n" );
-# if defined( __linux__ )
- fprintf( stderr, " (for instance Linux kernel 2.4.x or later)\n" );
-# endif
-#endif
-
- longjmp( env, 1 );
-}
-
-
uint32_t cpu_flags = 0;
}
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;
}
/**
}
/**
- * vlc_memset: fast CPU-dependent memset
+ * Returned an aligned pointer on newly allocated memory.
+ * \param alignment must be a power of 2 and a multiple of sizeof(void*)
+ * \param size is the size of the usable memory returned.
+ *
+ * It must not be freed directly, *base must.
*/
-void *vlc_memset (void *tgt, int c, size_t n)
+void *vlc_memalign(void **base, size_t alignment, size_t size)
{
- return pf_vlc_memset (tgt, c, n);
+ assert(alignment >= sizeof(void*));
+ for (size_t t = alignment; t > 1; t >>= 1)
+ assert((t&1) == 0);
+#if defined(HAVE_POSIX_MEMALIGN)
+ if (posix_memalign(base, alignment, size)) {
+ *base = NULL;
+ return NULL;
+ }
+ return *base;
+#elif defined(HAVE_MEMALIGN)
+ return *base = memalign(alignment, size);
+#else
+ unsigned char *p = *base = malloc(size + alignment - 1);
+ if (!p)
+ return NULL;
+ return (void*)((uintptr_t)(p + alignment - 1) & ~(alignment - 1));
+#endif
}