#endif
#include <assert.h>
-#include "libvlc.h"
-
-#if defined(__APPLE__)
+#ifdef __APPLE__
#include <sys/sysctl.h>
#endif
-#if defined(__OpenBSD__)
-#include <sys/param.h>
-#include <sys/sysctl.h>
-#include <machine/cpu.h>
-#endif
+#include "libvlc.h"
-#if defined(__SunOS)
-#include <unistd.h>
-#include <sys/types.h>
-#include <sys/processor.h>
-#include <sys/pset.h>
-#endif
+static uint32_t cpu_flags;
#if defined( __i386__ ) || defined( __x86_64__ ) || defined( __powerpc__ ) \
|| defined( __ppc__ ) || defined( __ppc64__ ) || defined( __powerpc64__ )
-# ifndef WIN32
+# if !defined( WIN32 ) && !defined( __OS2__ )
static bool check_OS_capability( const char *psz_capability, pid_t pid )
{
int status;
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); \
+# 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) \
+# else /* WIN32 || __OS2__ */
+# 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;
bool b_amd;
/* Needed for x86 CPU capabilities detection */
-# if defined( __x86_64__ )
-# define cpuid( reg ) \
- asm volatile ( "cpuid\n\t" \
- "movl %%ebx,%1\n\t" \
- : "=a" ( i_eax ), \
- "=b" ( i_ebx ), \
- "=c" ( i_ecx ), \
- "=d" ( i_edx ) \
- : "a" ( reg ) \
- : "cc" );
-# else
-# define cpuid( reg ) \
- asm volatile ( "push %%ebx\n\t" \
- "cpuid\n\t" \
- "movl %%ebx,%1\n\t" \
- "pop %%ebx\n\t" \
- : "=a" ( i_eax ), \
- "=r" ( i_ebx ), \
- "=c" ( i_ecx ), \
- "=d" ( i_edx ) \
- : "a" ( reg ) \
- : "cc" );
-# endif
+# if defined (__i386__) && defined (__PIC__)
+# define cpuid(reg) \
+ asm volatile ("xchgl %%ebx,%1\n\t" \
+ "cpuid\n\t" \
+ "xchgl %%ebx,%1\n\t" \
+ : "=a" (i_eax), "=r" (i_ebx), "=c" (i_ecx), "=d" (i_edx) \
+ : "a" (reg) \
+ : "cc");
+# else
+# define cpuid(reg) \
+ asm volatile ("cpuid\n\t" \
+ : "=a" (i_eax), "=b" (i_ebx), "=c" (i_ecx), "=d" (i_edx) \
+ : "a" (reg) \
+ : "cc");
+# endif
/* Check if the OS really supports the requested instructions */
# if defined (__i386__) && !defined (__i486__) && !defined (__i586__) \
&& !defined (__i686__) && !defined (__pentium4__) \
# ifdef CAN_COMPILE_SSE
check_capability( "SSE", CPU_CAPABILITY_SSE,
- "xorps %%xmm0,%%xmm0\n" );
+ "xorps %%xmm0,%%xmm0\n", );
# endif
}
# endif
# elif defined (CAN_COMPILE_SSE2)
if( i_edx & 0x04000000 )
check_capability( "SSE2", CPU_CAPABILITY_SSE2,
- "movupd %%xmm0, %%xmm0\n" );
+ "movupd %%xmm0, %%xmm0\n", );
# endif
# if defined (__SSE3__)
# elif defined (CAN_COMPILE_SSE3)
if( i_ecx & 0x00000001 )
check_capability( "SSE3", CPU_CAPABILITY_SSE3,
- "movsldup %%xmm1, %%xmm0\n" );
+ "movsldup %%xmm1, %%xmm0\n", );
# endif
# if defined (__SSSE3__)
# elif defined (CAN_COMPILE_SSSE3)
if( i_ecx & 0x00000200 )
check_capability( "SSSE3", CPU_CAPABILITY_SSSE3,
- "pabsw %%xmm1, %%xmm0\n" );
+ "pabsw %%xmm1, %%xmm0\n", );
# endif
# if defined (__SSE4_1__)
# elif defined (CAN_COMPILE_SSE4_1)
if( i_ecx & 0x00080000 )
check_capability( "SSE4.1", CPU_CAPABILITY_SSE4_1,
- "pmaxsb %%xmm1, %%xmm0\n" );
+ "pmaxsb %%xmm1, %%xmm0\n", );
# endif
# if defined (__SSE4_2__)
# elif defined (CAN_COMPILE_SSE4_2)
if( i_ecx & 0x00100000 )
check_capability( "SSE4.2", CPU_CAPABILITY_SSE4_2,
- "pcmpgtq %%xmm1, %%xmm0\n" );
+ "pcmpgtq %%xmm1, %%xmm0\n", );
# endif
/* test for additional capabilities */
# elif defined (CAN_COMPILE_3DNOW)
if( i_edx & 0x80000000 )
check_capability( "3D Now!", CPU_CAPABILITY_3DNOW,
- "pfadd %%mm0,%%mm0\n" "femms\n" );
+ "pfadd %%mm0,%%mm0\n" "femms\n", );
# endif
if( b_amd && ( i_edx & 0x00400000 ) )
}
out:
-#elif defined( __arm__ )
-# 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_ALTIVEC;
# 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( 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
#endif
- return i_capabilities;
-}
-
-uint32_t cpu_flags = 0;
+ cpu_flags = i_capabilities;
+}
-/*****************************************************************************
- * vlc_CPU: get pre-computed CPU capability flags
- ****************************************************************************/
+/**
+ * Retrieves pre-computed CPU capability flags
+ */
unsigned vlc_CPU (void)
{
+/* On Windows and OS/2,
+ * initialized from DllMain() and _DLL_InitTerm() respectively, instead */
+#if !defined(WIN32) && !defined(__OS2__)
+ static pthread_once_t once = PTHREAD_ONCE_INIT;
+ pthread_once (&once, vlc_CPU_init);
+#endif
return cpu_flags;
}
-const struct
+void vlc_CPU_dump (vlc_object_t *obj)
{
- uint32_t value;
- char name[12];
-} cap_dirs[] = {
-#if defined ( __i386__ ) || defined ( __x86_64__ )
- { CPU_CAPABILITY_MMX, "mmx" },
- { CPU_CAPABILITY_MMXEXT, "mmxext" },
- { CPU_CAPABILITY_3DNOW, "3dnow" },
- { CPU_CAPABILITY_SSE, "sse" },
-#endif
-#if defined (__ppc__) || defined (__ppc64__) || defined (__powerpc__)
- { CPU_CAPABILITY_ALTIVEC, "altivec" },
-#endif
-#if defined (__arm__)
- { CPU_CAPABILITY_NEON, "arm_neon" },
+ 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
-};
-/**
- * Return the number of available logical CPU.
- */
-unsigned vlc_GetCPUCount(void)
-{
-#if defined(WIN32) && !defined(UNDER_CE)
- DWORD process_mask;
- DWORD system_mask;
- if (!GetProcessAffinityMask(GetCurrentProcess(), &process_mask, &system_mask))
- return 1;
-
- unsigned count = 0;
- while (system_mask) {
- count++;
- system_mask >>= 1;
- }
- return count;
-#elif defined(HAVE_SCHED_GETAFFINITY)
- cpu_set_t cpu;
- CPU_ZERO(&cpu);
- if (sched_getaffinity(0, sizeof(cpu), &cpu) < 0)
- return 1;
- unsigned count = 0;
- for (unsigned i = 0; i < CPU_SETSIZE; i++)
- count += CPU_ISSET(i, &cpu) != 0;
- return count;
-#elif defined(__APPLE__)
- int count;
- size_t size = sizeof(count) ;
- if (sysctlbyname("hw.ncpu", &count, &size, NULL, 0))
- return 1; /* Failure */
- return count;
-#elif defined(__OpenBSD__)
- int selectors[2] = { CTL_HW, HW_NCPU };
- int count;
- size_t size = sizeof(count) ;
- if (sysctl(selectors, 2, &count, &size, NULL, 0))
- return 1; /* Failure */
- return count;
-#elif defined(__SunOS)
- unsigned count = 0;
- int type;
- u_int numcpus;
- processorid_t *cpulist;
- processor_info_t cpuinfo;
- cpulist = malloc(sizeof(processorid_t) * sysconf(_SC_NPROCESSORS_MAX));
- if (!cpulist) return 1;
- if (pset_info(PS_MYID, &type, &numcpus, cpulist)==0)
- {
- for (u_int i = 0; i < numcpus; i++)
- {
- if (!processor_info(cpulist[i], &cpuinfo))
- count += (cpuinfo.pi_state == P_ONLINE)?1:0;
- }
- } else {
- count = sysconf(_SC_NPROCESSORS_ONLN);
- }
- free(cpulist);
- return (count>0)?count:1;
-#else
-# warning "vlc_GetCPUCount is not implemented for your platform"
- return 1;
+#if HAVE_FPU
+ p += sprintf (p, "FPU ");
#endif
-}
-/**
- * Check if a directory name contains usable plugins w.r.t. the hardware
- * capabilities. Loading a plugin when the hardware has insufficient
- * capabilities may lead to illegal instructions (SIGILL) and must be avoided.
- *
- * @param name the name of the directory (<b>not</b> the path)
- *
- * @return true if the hardware has sufficient capabilities or the directory
- * does not require any special capability; false if the running hardware has
- * insufficient capabilities.
- */
-bool vlc_CPU_CheckPluginDir (const char *name)
-{
- const unsigned flags = vlc_CPU ();
- for (size_t i = 0; i < sizeof (cap_dirs) / sizeof (cap_dirs[0]); i++)
- {
- if (strcmp (name, cap_dirs[i].name))
- continue;
- return (flags & cap_dirs[i].value) != 0;
- }
- return true;
+ 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);
-}
-
-/**
- * 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_memalign(void **base, size_t alignment, size_t size)
-{
- 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
-}
-
projects / vlc / blobdiff