#include <signal.h>
#else
#include <errno.h>
-#include <windows.h>
-#include <winbase.h>
-#define PF_SSE3_INSTRUCTIONS_AVAILABLE 13
#endif
+#include <assert.h>
-#include "libvlc.h"
-
-#if defined(__APPLE__) && (defined(__ppc__) || defined(__ppc64__))
+#ifdef __APPLE__
#include <sys/sysctl.h>
#endif
+#include "libvlc.h"
+
#if defined( __i386__ ) || defined( __x86_64__ ) || defined( __powerpc__ ) \
|| defined( __ppc__ ) || defined( __ppc64__ ) || defined( __powerpc64__ )
# ifndef WIN32
# else /* WIN32 */
# define check_capability(name, flag, code) \
- do { \
- i_capabilities |= (flag); \
- } while(0)
+ i_capabilities |= (flag);
# endif
#endif
/* test for the MMX flag */
cpuid( 0x00000001 );
-
# if !defined (__MMX__)
if( ! (i_edx & 0x00800000) )
goto out;
# endif
i_capabilities |= CPU_CAPABILITY_MMX;
+
# if defined (__SSE__)
i_capabilities |= CPU_CAPABILITY_MMXEXT | CPU_CAPABILITY_SSE;
# else
i_capabilities |= CPU_CAPABILITY_MMXEXT;
# ifdef CAN_COMPILE_SSE
-# ifdef WIN32
- if( IsProcessorFeaturePresent( PF_XMMI_INSTRUCTIONS_AVAILABLE ) )
- i_capabilities |= CPU_CAPABILITY_SSE;
-# else
check_capability( "SSE", CPU_CAPABILITY_SSE,
"xorps %%xmm0,%%xmm0\n" );
-# endif
# endif
}
# endif
i_capabilities |= CPU_CAPABILITY_SSE2;
# elif defined (CAN_COMPILE_SSE2)
if( i_edx & 0x04000000 )
- {
-# ifdef WIN32
- if( IsProcessorFeaturePresent( PF_XMMI64_INSTRUCTIONS_AVAILABLE ) )
- i_capabilities |= CPU_CAPABILITY_SSE2;
-# else
check_capability( "SSE2", CPU_CAPABILITY_SSE2,
"movupd %%xmm0, %%xmm0\n" );
-# endif
- }
# endif
# if defined (__SSE3__)
i_capabilities |= CPU_CAPABILITY_SSE3;
# elif defined (CAN_COMPILE_SSE3)
if( i_ecx & 0x00000001 )
- {
-# ifdef WIN32
- if( IsProcessorFeaturePresent( PF_SSE3_INSTRUCTIONS_AVAILABLE ) )
- i_capabilities |= CPU_CAPABILITY_SSE3;
-# else
check_capability( "SSE3", CPU_CAPABILITY_SSE3,
"movsldup %%xmm1, %%xmm0\n" );
-# endif
- }
# endif
# if defined (__SSSE3__)
i_capabilities |= CPU_CAPABILITY_SSSE3;
# elif defined (CAN_COMPILE_SSSE3)
-# ifdef WIN32
- /* FIXME: IsProcessorFeaturePresent can't check for SSSE3 */
-# else
if( i_ecx & 0x00000200 )
check_capability( "SSSE3", CPU_CAPABILITY_SSSE3,
"pabsw %%xmm1, %%xmm0\n" );
-# endif
# endif
# if defined (__SSE4_1__)
i_capabilities |= CPU_CAPABILITY_SSE4_1;
# elif defined (CAN_COMPILE_SSE4_1)
-# ifdef WIN32
- /* FIXME: IsProcessorFeaturePresent can't check for SSE4.1 */
-# else
if( i_ecx & 0x00080000 )
check_capability( "SSE4.1", CPU_CAPABILITY_SSE4_1,
"pmaxsb %%xmm1, %%xmm0\n" );
-# endif
# endif
# if defined (__SSE4_2__)
i_capabilities |= CPU_CAPABILITY_SSE4_2;
# elif defined (CAN_COMPILE_SSE4_2)
-# ifdef WIN32
- /* FIXME: IsProcessorFeaturePresent can't check for SSE4.2 */
-# else
if( i_ecx & 0x00100000 )
check_capability( "SSE4.2", CPU_CAPABILITY_SSE4_2,
"pcmpgtq %%xmm1, %%xmm0\n" );
-# endif
# endif
/* test for additional capabilities */
# if defined (__3dNOW__)
i_capabilities |= CPU_CAPABILITY_3DNOW;
# elif defined (CAN_COMPILE_3DNOW)
-
if( i_edx & 0x80000000 )
- {
-# ifdef WIN32
- if( IsProcessorFeaturePresent( PF_3DNOW_INSTRUCTIONS_AVAILABLE ) )
- i_capabilities |= CPU_CAPABILITY_3DNOW;
-# else
check_capability( "3D Now!", CPU_CAPABILITY_3DNOW,
"pfadd %%mm0,%%mm0\n" "femms\n" );
-# endif
- }
# 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;
+# 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__)
+# 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);
return cpu_flags;
}
-const struct
-{
- 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" },
-#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;
-}
-
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
}