/*****************************************************************************
* cpu.c: cpu detection
*****************************************************************************
- * Copyright (C) 2003-2011 x264 project
+ * Copyright (C) 2003-2012 x264 project
*
* Authors: Loren Merritt <lorenm@u.washington.edu>
* Laurent Aimar <fenrir@via.ecp.fr>
#include "common.h"
#include "cpu.h"
-#if HAVE_PTHREAD && SYS_LINUX
+#if HAVE_POSIXTHREAD && SYS_LINUX
#include <sched.h>
#endif
#if SYS_BEOS
#include <machine/cpu.h>
#endif
-const x264_cpu_name_t x264_cpu_names[] = {
- {"Altivec", X264_CPU_ALTIVEC},
-// {"MMX", X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
- {"MMX2", X264_CPU_MMX|X264_CPU_MMXEXT},
- {"MMXEXT", X264_CPU_MMX|X264_CPU_MMXEXT},
-// {"SSE", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE}, // there are no sse1 functions in x264
- {"SSE2Slow",X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE2_IS_SLOW},
- {"SSE2", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2},
- {"SSE2Fast",X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE2_IS_FAST},
- {"SSE3", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3},
- {"SSSE3", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3},
- {"FastShuffle", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SHUFFLE_IS_FAST},
- {"SSE4.1", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
- {"SSE4.2", X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4|X264_CPU_SSE42},
- {"AVX", X264_CPU_AVX},
- {"Cache32", X264_CPU_CACHELINE_32},
- {"Cache64", X264_CPU_CACHELINE_64},
- {"SSEMisalign", X264_CPU_SSE_MISALIGN},
- {"LZCNT", X264_CPU_LZCNT},
+const x264_cpu_name_t x264_cpu_names[] =
+{
+ {"Altivec", X264_CPU_ALTIVEC},
+// {"MMX", X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
+ {"MMX2", X264_CPU_MMX|X264_CPU_MMX2},
+ {"MMXEXT", X264_CPU_MMX|X264_CPU_MMX2},
+// {"SSE", X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE}, // there are no sse1 functions in x264
+#define SSE2 X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE|X264_CPU_SSE2
+ {"SSE2Slow", SSE2|X264_CPU_SSE2_IS_SLOW},
+ {"SSE2", SSE2},
+ {"SSE2Fast", SSE2|X264_CPU_SSE2_IS_FAST},
+ {"SSE3", SSE2|X264_CPU_SSE3},
+ {"SSSE3", SSE2|X264_CPU_SSE3|X264_CPU_SSSE3},
+ {"FastShuffle", SSE2|X264_CPU_SHUFFLE_IS_FAST},
+ {"SSE4.1", SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
+ {"SSE4", SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
+ {"SSE4.2", SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4|X264_CPU_SSE42},
+#define AVX SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4|X264_CPU_SSE42|X264_CPU_AVX
+ {"AVX", AVX},
+ {"XOP", AVX|X264_CPU_XOP},
+ {"FMA4", AVX|X264_CPU_FMA4},
+ {"AVX2", AVX|X264_CPU_AVX2},
+ {"FMA3", AVX|X264_CPU_FMA3},
+#undef AVX
+#undef SSE2
+ {"Cache32", X264_CPU_CACHELINE_32},
+ {"Cache64", X264_CPU_CACHELINE_64},
+ {"SSEMisalign", X264_CPU_SSE_MISALIGN},
+ {"LZCNT", X264_CPU_LZCNT},
+ {"BMI1", X264_CPU_BMI1},
+ {"BMI2", X264_CPU_BMI1|X264_CPU_BMI2},
+ {"TBM", X264_CPU_TBM},
{"Slow_mod4_stack", X264_CPU_STACK_MOD4},
- {"ARMv6", X264_CPU_ARMV6},
- {"NEON", X264_CPU_NEON},
- {"Fast_NEON_MRC", X264_CPU_FAST_NEON_MRC},
- {"SlowCTZ", X264_CPU_SLOW_CTZ},
- {"SlowAtom", X264_CPU_SLOW_ATOM},
+ {"ARMv6", X264_CPU_ARMV6},
+ {"NEON", X264_CPU_NEON},
+ {"Fast_NEON_MRC", X264_CPU_FAST_NEON_MRC},
+ {"SlowCTZ", X264_CPU_SLOW_CTZ},
+ {"SlowAtom", X264_CPU_SLOW_ATOM},
{"", 0},
};
#if HAVE_MMX
int x264_cpu_cpuid_test( void );
-uint32_t x264_cpu_cpuid( uint32_t op, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx );
+void x264_cpu_cpuid( uint32_t op, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx );
+void x264_cpu_xgetbv( uint32_t op, uint32_t *eax, uint32_t *edx );
uint32_t x264_cpu_detect( void )
{
uint32_t cpu = 0;
uint32_t eax, ebx, ecx, edx;
uint32_t vendor[4] = {0};
- int max_extended_cap;
+ uint32_t max_extended_cap;
int cache;
#if !ARCH_X86_64
else
return 0;
if( edx&0x02000000 )
- cpu |= X264_CPU_MMXEXT|X264_CPU_SSE;
+ cpu |= X264_CPU_MMX2|X264_CPU_SSE;
if( edx&0x04000000 )
cpu |= X264_CPU_SSE2;
if( ecx&0x00000001 )
cpu |= X264_CPU_SSE4;
if( ecx&0x00100000 )
cpu |= X264_CPU_SSE42;
- if( ecx&0x10000000 )
- cpu |= X264_CPU_AVX;
+ /* Check OXSAVE and AVX bits */
+ if( (ecx&0x18000000) == 0x18000000 )
+ {
+ /* Check for OS support */
+ x264_cpu_xgetbv( 0, &eax, &edx );
+ if( (eax&0x6) == 0x6 )
+ {
+ cpu |= X264_CPU_AVX;
+ if( ecx&0x00001000 )
+ cpu |= X264_CPU_FMA3;
+ }
+ }
+
+ x264_cpu_cpuid( 7, &eax, &ebx, &ecx, &edx );
+ /* AVX2 requires OS support, but BMI1/2 don't. */
+ if( (cpu&X264_CPU_AVX) && (ebx&0x00000020) )
+ cpu |= X264_CPU_AVX2;
+ if( ebx&0x00000008 )
+ {
+ cpu |= X264_CPU_BMI1;
+ if( ebx&0x00000100 )
+ cpu |= X264_CPU_BMI2;
+ }
if( cpu & X264_CPU_SSSE3 )
cpu |= X264_CPU_SSE2_IS_FAST;
cpu |= X264_CPU_SLOW_CTZ;
x264_cpu_cpuid( 0x80000001, &eax, &ebx, &ecx, &edx );
if( edx&0x00400000 )
- cpu |= X264_CPU_MMXEXT;
+ cpu |= X264_CPU_MMX2;
if( cpu & X264_CPU_SSE2 )
{
if( ecx&0x00000040 ) /* SSE4a */
cpu |= X264_CPU_SSE_MISALIGN;
x264_cpu_mask_misalign_sse();
}
+
+ if( cpu & X264_CPU_AVX )
+ {
+ if( ecx&0x00000800 ) /* XOP */
+ cpu |= X264_CPU_XOP;
+ if( ecx&0x00010000 ) /* FMA4 */
+ cpu |= X264_CPU_FMA4;
+ }
+
+ if( ecx&0x00200000 )
+ cpu |= X264_CPU_TBM;
}
}
x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
int family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
int model = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
- /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and 6/14 (core1 "yonah")
- * theoretically support sse2, but it's significantly slower than mmx for
- * almost all of x264's functions, so let's just pretend they don't. */
- if( family == 6 && (model == 9 || model == 13 || model == 14) )
- {
- cpu &= ~(X264_CPU_SSE2|X264_CPU_SSE3);
- assert(!(cpu&(X264_CPU_SSSE3|X264_CPU_SSE4)));
- }
- /* Detect Atom CPU */
- if( family == 6 && model == 28 )
+ if( family == 6 )
{
- cpu |= X264_CPU_SLOW_ATOM;
- cpu |= X264_CPU_SLOW_CTZ;
+ /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and 6/14 (core1 "yonah")
+ * theoretically support sse2, but it's significantly slower than mmx for
+ * almost all of x264's functions, so let's just pretend they don't. */
+ if( model == 9 || model == 13 || model == 14 )
+ {
+ cpu &= ~(X264_CPU_SSE2|X264_CPU_SSE3);
+ assert(!(cpu&(X264_CPU_SSSE3|X264_CPU_SSE4)));
+ }
+ /* Detect Atom CPU */
+ else if( model == 28 )
+ {
+ cpu |= X264_CPU_SLOW_ATOM;
+ cpu |= X264_CPU_SLOW_CTZ;
+ }
+ /* Some Penryns and Nehalems are pointlessly crippled (SSE4 disabled), so
+ * detect them here. */
+ else if( model >= 23 )
+ cpu |= X264_CPU_SHUFFLE_IS_FAST;
}
}
#if !HAVE_THREAD
return 1;
-#elif defined(_WIN32)
+#elif SYS_WINDOWS
return x264_pthread_num_processors_np();
+#elif SYS_CYGWIN
+ return sysconf( _SC_NPROCESSORS_ONLN );
+
#elif SYS_LINUX
- unsigned int bit;
- int np;
cpu_set_t p_aff;
memset( &p_aff, 0, sizeof(p_aff) );
- sched_getaffinity( 0, sizeof(p_aff), &p_aff );
- for( np = 0, bit = 0; bit < sizeof(p_aff); bit++ )
+ if( sched_getaffinity( 0, sizeof(p_aff), &p_aff ) )
+ return 1;
+#if HAVE_CPU_COUNT
+ return CPU_COUNT(&p_aff);
+#else
+ int np = 0;
+ for( unsigned int bit = 0; bit < 8 * sizeof(p_aff); bit++ )
np += (((uint8_t *)&p_aff)[bit / 8] >> (bit % 8)) & 1;
return np;
+#endif
#elif SYS_BEOS
system_info info;
projects / x264 / blobdiff