X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=src%2Fmisc%2Fcpu.c;h=71aed9dc16187a0e401e7bef1c73bdb039975ffe;hb=b4c9cb160e24388a8c6092717f5257cdf46f146a;hp=ac42227006cc1b98d86854eba0a9856d35f0e3c1;hpb=600d29043eed8f9c3e848dddb79bb0e2e9a0a391;p=vlc diff --git a/src/misc/cpu.c b/src/misc/cpu.c index ac42227006..71aed9dc16 100644 --- a/src/misc/cpu.c +++ b/src/misc/cpu.c @@ -31,24 +31,36 @@ #endif #include +#include #include #ifndef WIN32 #include #include +#include +#else +#include #endif +#include #include "libvlc.h" -#if defined(__APPLE__) && (defined(__ppc__) || defined(__ppc64__)) +#if defined(__APPLE__) #include #endif -#if defined( __i386__ ) || defined( __x86_64__ ) \ - || defined( __ppc__ ) || defined( __ppc64__ ) +#if defined(__SunOS) +#include +#include +#include +#include +#endif + +#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 ) @@ -64,14 +76,25 @@ static bool check_OS_capability( const char *psz_capability, pid_t pid ) 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) \ + 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 /***************************************************************************** @@ -81,7 +104,7 @@ static bool check_OS_capability( const char *psz_capability, pid_t pid ) *****************************************************************************/ uint32_t CPUCapabilities( void ) { - uint32_t i_capabilities = CPU_CAPABILITY_NONE; + uint32_t i_capabilities = 0; #if defined( __i386__ ) || defined( __x86_64__ ) unsigned int i_eax, i_ebx, i_ecx, i_edx; @@ -111,10 +134,10 @@ uint32_t CPUCapabilities( void ) : "a" ( reg ) \ : "cc" ); # 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" @@ -134,20 +157,17 @@ uint32_t CPUCapabilities( void ) if( i_eax == i_ebx ) goto out; -# else - /* x86_64 supports cpuid instruction, so we dont need to check it */ -# endif - - i_capabilities |= CPU_CAPABILITY_486; +# 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 ) goto out; - - /* FIXME: this isn't correct, since some 486s have cpuid */ - i_capabilities |= CPU_CAPABILITY_586; +#endif /* borrowed from mpeg2dec */ b_amd = ( i_ebx == 0x68747541 ) && ( i_ecx == 0x444d4163 ) @@ -155,45 +175,65 @@ uint32_t CPUCapabilities( void ) /* 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 if( i_edx & 0x02000000 ) { 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_SSE2) if( i_edx & 0x04000000 ) - { -# if defined(CAN_COMPILE_SSE) - /* 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; -# 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 ); @@ -204,20 +244,13 @@ uint32_t CPUCapabilities( void ) /* 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 ) - { - 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; - } -# endif + check_capability( "3D Now!", CPU_CAPABILITY_3DNOW, + "pfadd %%mm0,%%mm0\n" "femms\n" ); +# endif if( b_amd && ( i_edx & 0x00400000 ) ) { @@ -226,16 +259,12 @@ uint32_t CPUCapabilities( void ) out: #elif defined( __arm__ ) -# if defined( __ARM_EABI__ ) && !defined( __SOFTFP__ ) -// i_capabilities |= CPU_CAPABILITY_FPU; -# endif # if defined( __ARM_NEON__ ) i_capabilities |= CPU_CAPABILITY_NEON; # endif -#elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __ppc64__ ) - - i_capabilities |= CPU_CAPABILITY_FPU; +#elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __powerpc64__ ) \ + || defined( __ppc64__ ) # if defined(__APPLE__) int selectors[2] = { CTL_HW, HW_VECTORUNIT }; @@ -250,11 +279,12 @@ out: 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); + _exit(0); } if( check_OS_capability( "Altivec", pid ) ) @@ -262,12 +292,6 @@ out: # endif -#elif defined( __sparc__ ) - i_capabilities |= CPU_CAPABILITY_FPU; - -#elif defined( _MSC_VER ) && !defined( UNDER_CE ) - i_capabilities |= CPU_CAPABILITY_FPU; - #endif return i_capabilities; } @@ -283,6 +307,106 @@ unsigned vlc_CPU (void) 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 +}; + +/** + * 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(__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; +#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 (not 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; @@ -309,3 +433,32 @@ 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 +} +