1 /*****************************************************************************
2 * cpu.c: CPU detection code
3 *****************************************************************************
4 * Copyright (C) 1998-2004 the VideoLAN team
7 * Authors: Samuel Hocevar <sam@zoy.org>
8 * Christophe Massiot <massiot@via.ecp.fr>
9 * Eugenio Jarosiewicz <ej0@cise.ufl.eduEujenio>
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
24 *****************************************************************************/
26 /*****************************************************************************
28 *****************************************************************************/
33 #include <vlc_common.h>
36 #include <sys/types.h>
47 #include <sys/sysctl.h>
52 static uint32_t cpu_flags;
54 #if defined( __i386__ ) || defined( __x86_64__ ) || defined( __powerpc__ ) \
55 || defined( __ppc__ ) || defined( __ppc64__ ) || defined( __powerpc64__ )
57 static bool check_OS_capability( const char *psz_capability, pid_t pid )
62 return false; /* fail safe :-/ */
64 while( waitpid( pid, &status, 0 ) == -1 );
66 if( WIFEXITED( status ) && WEXITSTATUS( status ) == 0 )
69 fprintf( stderr, "warning: your CPU has %s instructions, but not your "
70 "operating system.\n", psz_capability );
71 fprintf( stderr, " some optimizations will be disabled unless "
72 "you upgrade your OS\n" );
76 # define check_capability(name, flag, code, input) \
81 signal(SIGILL, SIG_DFL); \
82 __asm__ __volatile__ ( code : : input ); \
85 if( check_OS_capability((name), pid )) \
86 i_capabilities |= (flag); \
90 # define check_capability(name, flag, code, input) \
91 i_capabilities |= (flag);
96 * Determines the CPU capabilities and stores them in cpu_flags.
97 * The result can be retrieved with vlc_CPU().
99 void vlc_CPU_init (void)
101 uint32_t i_capabilities = 0;
103 #if defined( __i386__ ) || defined( __x86_64__ )
104 unsigned int i_eax, i_ebx, i_ecx, i_edx;
107 /* Needed for x86 CPU capabilities detection */
108 # if defined( __x86_64__ )
109 # define cpuid( reg ) \
110 asm volatile ( "cpuid\n\t" \
111 "movl %%ebx,%1\n\t" \
119 # define cpuid( reg ) \
120 asm volatile ( "push %%ebx\n\t" \
122 "movl %%ebx,%1\n\t" \
131 /* Check if the OS really supports the requested instructions */
132 # if defined (__i386__) && !defined (__i486__) && !defined (__i586__) \
133 && !defined (__i686__) && !defined (__pentium4__) \
134 && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
135 /* check if cpuid instruction is supported */
136 asm volatile ( "push %%ebx\n\t"
139 "movl %%eax, %%ebx\n\t"
140 "xorl $0x200000, %%eax\n\t"
156 /* the CPU supports the CPUID instruction - get its level */
159 # if defined (__i386__) && !defined (__i586__) \
160 && !defined (__i686__) && !defined (__pentium4__) \
161 && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
166 /* borrowed from mpeg2dec */
167 b_amd = ( i_ebx == 0x68747541 ) && ( i_ecx == 0x444d4163 )
168 && ( i_edx == 0x69746e65 );
170 /* test for the MMX flag */
172 # if !defined (__MMX__)
173 if( ! (i_edx & 0x00800000) )
176 i_capabilities |= CPU_CAPABILITY_MMX;
178 # if defined (__SSE__)
179 i_capabilities |= CPU_CAPABILITY_MMXEXT | CPU_CAPABILITY_SSE;
181 if( i_edx & 0x02000000 )
183 i_capabilities |= CPU_CAPABILITY_MMXEXT;
185 # ifdef CAN_COMPILE_SSE
186 check_capability( "SSE", CPU_CAPABILITY_SSE,
187 "xorps %%xmm0,%%xmm0\n", );
192 # if defined (__SSE2__)
193 i_capabilities |= CPU_CAPABILITY_SSE2;
194 # elif defined (CAN_COMPILE_SSE2)
195 if( i_edx & 0x04000000 )
196 check_capability( "SSE2", CPU_CAPABILITY_SSE2,
197 "movupd %%xmm0, %%xmm0\n", );
200 # if defined (__SSE3__)
201 i_capabilities |= CPU_CAPABILITY_SSE3;
202 # elif defined (CAN_COMPILE_SSE3)
203 if( i_ecx & 0x00000001 )
204 check_capability( "SSE3", CPU_CAPABILITY_SSE3,
205 "movsldup %%xmm1, %%xmm0\n", );
208 # if defined (__SSSE3__)
209 i_capabilities |= CPU_CAPABILITY_SSSE3;
210 # elif defined (CAN_COMPILE_SSSE3)
211 if( i_ecx & 0x00000200 )
212 check_capability( "SSSE3", CPU_CAPABILITY_SSSE3,
213 "pabsw %%xmm1, %%xmm0\n", );
216 # if defined (__SSE4_1__)
217 i_capabilities |= CPU_CAPABILITY_SSE4_1;
218 # elif defined (CAN_COMPILE_SSE4_1)
219 if( i_ecx & 0x00080000 )
220 check_capability( "SSE4.1", CPU_CAPABILITY_SSE4_1,
221 "pmaxsb %%xmm1, %%xmm0\n", );
224 # if defined (__SSE4_2__)
225 i_capabilities |= CPU_CAPABILITY_SSE4_2;
226 # elif defined (CAN_COMPILE_SSE4_2)
227 if( i_ecx & 0x00100000 )
228 check_capability( "SSE4.2", CPU_CAPABILITY_SSE4_2,
229 "pcmpgtq %%xmm1, %%xmm0\n", );
232 /* test for additional capabilities */
235 if( i_eax < 0x80000001 )
238 /* list these additional capabilities */
241 # if defined (__3dNOW__)
242 i_capabilities |= CPU_CAPABILITY_3DNOW;
243 # elif defined (CAN_COMPILE_3DNOW)
244 if( i_edx & 0x80000000 )
245 check_capability( "3D Now!", CPU_CAPABILITY_3DNOW,
246 "pfadd %%mm0,%%mm0\n" "femms\n", );
249 if( b_amd && ( i_edx & 0x00400000 ) )
251 i_capabilities |= CPU_CAPABILITY_MMXEXT;
255 #elif defined (__arm__)
257 # if defined (__ARM_NEON__)
258 i_capabilities |= CPU_CAPABILITY_NEON;
259 # elif defined (CAN_COMPILE_NEON)
260 # define NEED_RUNTIME_CPU_CHECK 1
263 # ifdef NEED_RUNTIME_CPU_CHECK
264 # if defined (__linux__)
265 FILE *info = fopen ("/proc/cpuinfo", "rt");
271 while (getline (&line, &linelen, info) != -1)
275 if (strncmp (line, "Features\t:", 10))
278 /* TODO: detect other CPU features when we use them */
279 # if defined (CAN_COMPILE_NEON) && !defined (__ARM_NEON__)
280 cap = strstr (line + 10, " neon");
281 if (cap != NULL && (cap[5] == '\0' || cap[5] == ' '))
282 i_capabilities |= CPU_CAPABILITY_NEON;
290 # warning Run-time CPU detection missing: optimizations disabled!
294 #elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __powerpc64__ ) \
295 || defined( __ppc64__ )
297 # if defined(__APPLE__) || defined(__OpenBSD__)
298 # if defined(__OpenBSD__)
299 int selectors[2] = { CTL_MACHDEP, CPU_ALTIVEC };
301 int selectors[2] = { CTL_HW, HW_VECTORUNIT };
303 int i_has_altivec = 0;
304 size_t i_length = sizeof( i_has_altivec );
305 int i_error = sysctl( selectors, 2, &i_has_altivec, &i_length, NULL, 0);
307 if( i_error == 0 && i_has_altivec != 0 )
308 i_capabilities |= CPU_CAPABILITY_ALTIVEC;
310 # elif defined( CAN_COMPILE_ALTIVEC )
311 check_capability( "Altivec", CPU_CAPABILITY_ALTIVEC,
313 "vand %%v0, %%v0, %%v0",
320 cpu_flags = i_capabilities;
324 * Retrieves pre-computed CPU capability flags
326 unsigned vlc_CPU (void)
328 #ifndef WIN32 /* On Windows, initialized from DllMain() instead */
329 static pthread_once_t once = PTHREAD_ONCE_INIT;
330 pthread_once (&once, vlc_CPU_init);
335 void vlc_CPU_dump (vlc_object_t *obj)
337 const unsigned flags = vlc_CPU();
338 char buf[200], *p = buf;
340 #define PRINT_CAPABILITY( capability, string ) \
341 if (flags & (capability)) \
342 p += sprintf (p, "%s ", (string) )
344 #if defined (__i386__) || defined (__x86_64__)
345 PRINT_CAPABILITY(CPU_CAPABILITY_MMX, "MMX");
346 PRINT_CAPABILITY(CPU_CAPABILITY_3DNOW, "3DNow!");
347 PRINT_CAPABILITY(CPU_CAPABILITY_MMXEXT, "MMXEXT");
348 PRINT_CAPABILITY(CPU_CAPABILITY_SSE, "SSE");
349 PRINT_CAPABILITY(CPU_CAPABILITY_SSE2, "SSE2");
350 PRINT_CAPABILITY(CPU_CAPABILITY_SSE3, "SSE3");
351 PRINT_CAPABILITY(CPU_CAPABILITY_SSSE3, "SSSE3");
352 PRINT_CAPABILITY(CPU_CAPABILITY_SSE4_1, "SSE4.1");
353 PRINT_CAPABILITY(CPU_CAPABILITY_SSE4_2, "SSE4.2");
354 PRINT_CAPABILITY(CPU_CAPABILITY_SSE4A, "SSE4A");
356 #elif defined (__powerpc__) || defined (__ppc__) || defined (__ppc64__)
357 PRINT_CAPABILITY(CPU_CAPABILITY_ALTIVEC, "AltiVec");
359 #elif defined (__arm__)
360 PRINT_CAPABILITY(CPU_CAPABILITY_NEON, "NEONv1");
365 p += sprintf (p, "FPU ");
369 msg_Dbg (obj, "CPU has capabilities %s", buf);
373 static vlc_memcpy_t pf_vlc_memcpy = memcpy;
375 void vlc_fastmem_register (vlc_memcpy_t cpy)
377 assert (cpy != NULL);
382 * vlc_memcpy: fast CPU-dependent memcpy
384 void *vlc_memcpy (void *tgt, const void *src, size_t n)
386 return pf_vlc_memcpy (tgt, src, n);
390 * Returned an aligned pointer on newly allocated memory.
391 * \param alignment must be a power of 2 and a multiple of sizeof(void*)
392 * \param size is the size of the usable memory returned.
394 * It must not be freed directly, *base must.
396 void *vlc_memalign(void **base, size_t alignment, size_t size)
398 assert(alignment >= sizeof(void*));
399 for (size_t t = alignment; t > 1; t >>= 1)
401 #if defined(HAVE_POSIX_MEMALIGN)
402 if (posix_memalign(base, alignment, size)) {
407 #elif defined(HAVE_MEMALIGN)
408 return *base = memalign(alignment, size);
410 unsigned char *p = *base = malloc(size + alignment - 1);
413 return (void*)((uintptr_t)(p + alignment - 1) & ~(alignment - 1));