/*****************************************************************************
- * cpu.c: h264 encoder library
+ * cpu.c: cpu detection
*****************************************************************************
- * Copyright (C) 2003-2008 x264 project
+ * Copyright (C) 2003-2016 x264 project
*
* Authors: Loren Merritt <lorenm@u.washington.edu>
* Laurent Aimar <fenrir@via.ecp.fr>
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
+ *
+ * This program is also available under a commercial proprietary license.
+ * For more information, contact us at licensing@x264.com.
*****************************************************************************/
#define _GNU_SOURCE // for sched_getaffinity
#include "common.h"
#include "cpu.h"
-#if defined(HAVE_PTHREAD) && defined(SYS_LINUX)
+#if HAVE_POSIXTHREAD && SYS_LINUX
#include <sched.h>
#endif
-#ifdef SYS_BEOS
+#if SYS_BEOS
#include <kernel/OS.h>
#endif
-#if defined(SYS_MACOSX) || defined(SYS_FREEBSD)
+#if SYS_MACOSX || SYS_FREEBSD
#include <sys/types.h>
#include <sys/sysctl.h>
#endif
-#ifdef SYS_OPENBSD
+#if SYS_OPENBSD
#include <sys/param.h>
#include <sys/sysctl.h>
#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},
- {"Cache32", X264_CPU_CACHELINE_32},
- {"Cache64", X264_CPU_CACHELINE_64},
- {"SSEMisalign", X264_CPU_SSE_MISALIGN},
- {"LZCNT", X264_CPU_LZCNT},
- {"Slow_mod4_stack", X264_CPU_STACK_MOD4},
- {"ARMv6", X264_CPU_ARMV6},
- {"NEON", X264_CPU_NEON},
- {"Fast_NEON_MRC", X264_CPU_FAST_NEON_MRC},
+const x264_cpu_name_t x264_cpu_names[] =
+{
+#if HAVE_MMX
+// {"MMX", X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
+// {"CMOV", X264_CPU_CMOV}, // we require this unconditionally, so don't print it
+#define MMX2 X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_CMOV
+ {"MMX2", MMX2},
+ {"MMXEXT", MMX2},
+ {"SSE", MMX2|X264_CPU_SSE},
+#define SSE2 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},
+ {"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},
+ {"FMA3", AVX|X264_CPU_FMA3},
+ {"AVX2", AVX|X264_CPU_FMA3|X264_CPU_AVX2},
+#undef AVX
+#undef SSE2
+#undef MMX2
+ {"Cache32", X264_CPU_CACHELINE_32},
+ {"Cache64", X264_CPU_CACHELINE_64},
+ {"LZCNT", X264_CPU_LZCNT},
+ {"BMI1", X264_CPU_BMI1},
+ {"BMI2", X264_CPU_BMI1|X264_CPU_BMI2},
+ {"SlowCTZ", X264_CPU_SLOW_CTZ},
+ {"SlowAtom", X264_CPU_SLOW_ATOM},
+ {"SlowPshufb", X264_CPU_SLOW_PSHUFB},
+ {"SlowPalignr", X264_CPU_SLOW_PALIGNR},
+ {"SlowShuffle", X264_CPU_SLOW_SHUFFLE},
+ {"UnalignedStack", X264_CPU_STACK_MOD4},
+#elif ARCH_PPC
+ {"Altivec", X264_CPU_ALTIVEC},
+#elif ARCH_ARM
+ {"ARMv6", X264_CPU_ARMV6},
+ {"NEON", X264_CPU_NEON},
+ {"FastNeonMRC", X264_CPU_FAST_NEON_MRC},
+#elif ARCH_AARCH64
+ {"ARMv8", X264_CPU_ARMV8},
+ {"NEON", X264_CPU_NEON},
+#elif ARCH_MIPS
+ {"MSA", X264_CPU_MSA},
+#endif
{"", 0},
};
-#if (defined(ARCH_PPC) && defined(SYS_LINUX)) || (defined(ARCH_ARM) && !defined(HAVE_NEON))
+#if (ARCH_PPC && SYS_LINUX) || (ARCH_ARM && !HAVE_NEON)
#include <signal.h>
#include <setjmp.h>
static sigjmp_buf jmpbuf;
}
#endif
-#ifdef HAVE_MMX
-extern int x264_cpu_cpuid_test( void );
-extern uint32_t x264_cpu_cpuid( uint32_t op, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx );
+#if HAVE_MMX
+int x264_cpu_cpuid_test( void );
+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, max_basic_cap;
int cache;
-#ifndef ARCH_X86_64
+#if !ARCH_X86_64
if( !x264_cpu_cpuid_test() )
return 0;
#endif
x264_cpu_cpuid( 0, &eax, vendor+0, vendor+2, vendor+1 );
- if( eax == 0 )
+ max_basic_cap = eax;
+ if( max_basic_cap == 0 )
return 0;
x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
if( edx&0x00800000 )
cpu |= X264_CPU_MMX;
else
- return 0;
+ return cpu;
if( edx&0x02000000 )
- cpu |= X264_CPU_MMXEXT|X264_CPU_SSE;
+ cpu |= X264_CPU_MMX2|X264_CPU_SSE;
+ if( edx&0x00008000 )
+ cpu |= X264_CPU_CMOV;
+ else
+ return cpu;
if( edx&0x04000000 )
cpu |= X264_CPU_SSE2;
if( ecx&0x00000001 )
cpu |= X264_CPU_SSE4;
if( ecx&0x00100000 )
cpu |= X264_CPU_SSE42;
+ /* 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;
+ }
+ }
+
+ if( max_basic_cap >= 7 )
+ {
+ 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;
- if( cpu & X264_CPU_SSE4 )
- cpu |= X264_CPU_SHUFFLE_IS_FAST;
x264_cpu_cpuid( 0x80000000, &eax, &ebx, &ecx, &edx );
max_extended_cap = eax;
- if( !strcmp((char*)vendor, "AuthenticAMD") && max_extended_cap >= 0x80000001 )
+ if( max_extended_cap >= 0x80000001 )
{
x264_cpu_cpuid( 0x80000001, &eax, &ebx, &ecx, &edx );
- if( edx&0x00400000 )
- cpu |= X264_CPU_MMXEXT;
- if( cpu & X264_CPU_SSE2 )
+
+ if( ecx&0x00000020 )
+ cpu |= X264_CPU_LZCNT; /* Supported by Intel chips starting with Haswell */
+ if( ecx&0x00000040 ) /* SSE4a, AMD only */
{
- if( ecx&0x00000040 ) /* SSE4a */
+ int family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
+ cpu |= X264_CPU_SSE2_IS_FAST; /* Phenom and later CPUs have fast SSE units */
+ if( family == 0x14 )
{
- cpu |= X264_CPU_SSE2_IS_FAST;
- cpu |= X264_CPU_LZCNT;
- cpu |= X264_CPU_SHUFFLE_IS_FAST;
+ cpu &= ~X264_CPU_SSE2_IS_FAST; /* SSSE3 doesn't imply fast SSE anymore... */
+ cpu |= X264_CPU_SSE2_IS_SLOW; /* Bobcat has 64-bit SIMD units */
+ cpu |= X264_CPU_SLOW_PALIGNR; /* palignr is insanely slow on Bobcat */
}
- else
- cpu |= X264_CPU_SSE2_IS_SLOW;
-
- if( ecx&0x00000080 ) /* Misalign SSE */
+ if( family == 0x16 )
{
- cpu |= X264_CPU_SSE_MISALIGN;
- x264_cpu_mask_misalign_sse();
+ cpu |= X264_CPU_SLOW_PSHUFB; /* Jaguar's pshufb isn't that slow, but it's slow enough
+ * compared to alternate instruction sequences that this
+ * is equal or faster on almost all such functions. */
}
}
+
+ if( cpu & X264_CPU_AVX )
+ {
+ if( ecx&0x00000800 ) /* XOP */
+ cpu |= X264_CPU_XOP;
+ if( ecx&0x00010000 ) /* FMA4 */
+ cpu |= X264_CPU_FMA4;
+ }
+
+ if( !strcmp((char*)vendor, "AuthenticAMD") )
+ {
+ if( edx&0x00400000 )
+ cpu |= X264_CPU_MMX2;
+ if( !(cpu&X264_CPU_LZCNT) )
+ cpu |= X264_CPU_SLOW_CTZ;
+ if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_FAST) )
+ cpu |= X264_CPU_SSE2_IS_SLOW; /* AMD CPUs come in two types: terrible at SSE and great at it */
+ }
}
if( !strcmp((char*)vendor, "GenuineIntel") )
{
- int family, model, stepping;
x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
- family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
- model = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
- stepping = eax&0xf;
- /* 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) )
+ int family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
+ int model = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
+ if( family == 6 )
{
- cpu &= ~(X264_CPU_SSE2|X264_CPU_SSE3);
- assert(!(cpu&(X264_CPU_SSSE3|X264_CPU_SSE4)));
+ /* 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;
+ cpu |= X264_CPU_SLOW_PSHUFB;
+ }
+ /* Conroe has a slow shuffle unit. Check the model number to make sure not
+ * to include crippled low-end Penryns and Nehalems that don't have SSE4. */
+ else if( (cpu&X264_CPU_SSSE3) && !(cpu&X264_CPU_SSE4) && model < 23 )
+ cpu |= X264_CPU_SLOW_SHUFFLE;
}
}
x264_cpu_cpuid( 0x80000006, &eax, &ebx, &ecx, &edx );
cache = ecx&0xff; // cacheline size
}
- if( !cache )
+ if( !cache && max_basic_cap >= 2 )
{
// Cache and TLB Information
static const char cache32_ids[] = { 0x0a, 0x0c, 0x41, 0x42, 0x43, 0x44, 0x45, 0x82, 0x83, 0x84, 0x85, 0 };
x264_log( NULL, X264_LOG_WARNING, "unable to determine cacheline size\n" );
}
-#ifdef BROKEN_STACK_ALIGNMENT
+#if STACK_ALIGNMENT < 16
cpu |= X264_CPU_STACK_MOD4;
#endif
return cpu;
}
-#elif defined( ARCH_PPC )
+#elif ARCH_PPC
-#if defined(SYS_MACOSX) || defined(SYS_OPENBSD)
+#if SYS_MACOSX || SYS_OPENBSD || SYS_FREEBSD
#include <sys/sysctl.h>
uint32_t x264_cpu_detect( void )
{
/* Thank you VLC */
uint32_t cpu = 0;
-#ifdef SYS_OPENBSD
+#if SYS_OPENBSD
int selectors[2] = { CTL_MACHDEP, CPU_ALTIVEC };
-#else
+#elif SYS_MACOSX
int selectors[2] = { CTL_HW, HW_VECTORUNIT };
#endif
int has_altivec = 0;
size_t length = sizeof( has_altivec );
+#if SYS_MACOSX || SYS_OPENBSD
int error = sysctl( selectors, 2, &has_altivec, &length, NULL, 0 );
+#else
+ int error = sysctlbyname( "hw.altivec", &has_altivec, &length, NULL, 0 );
+#endif
if( error == 0 && has_altivec != 0 )
cpu |= X264_CPU_ALTIVEC;
return cpu;
}
-#elif defined( SYS_LINUX )
+#elif SYS_LINUX
uint32_t x264_cpu_detect( void )
{
+#ifdef __NO_FPRS__
+ return 0;
+#else
static void (*oldsig)( int );
oldsig = signal( SIGILL, sigill_handler );
signal( SIGILL, oldsig );
return X264_CPU_ALTIVEC;
+#endif
}
#endif
-#elif defined( ARCH_ARM )
+#elif ARCH_ARM
-void x264_cpu_neon_test();
-int x264_cpu_fast_neon_mrc_test();
+void x264_cpu_neon_test( void );
+int x264_cpu_fast_neon_mrc_test( void );
uint32_t x264_cpu_detect( void )
{
int flags = 0;
-#ifdef HAVE_ARMV6
+#if HAVE_ARMV6
flags |= X264_CPU_ARMV6;
// don't do this hack if compiled with -mfpu=neon
-#ifndef HAVE_NEON
+#if !HAVE_NEON
static void (* oldsig)( int );
oldsig = signal( SIGILL, sigill_handler );
if( sigsetjmp( jmpbuf, 1 ) )
// Note that there is potential for a race condition if another program or
// x264 instance disables or reinits the counters while x264 is using them,
// which may result in incorrect detection and the counters stuck enabled.
+ // right now Apple does not seem to support performance counters for this test
+#ifndef __MACH__
flags |= x264_cpu_fast_neon_mrc_test() ? X264_CPU_FAST_NEON_MRC : 0;
+#endif
// TODO: write dual issue test? currently it's A8 (dual issue) vs. A9 (fast mrc)
#endif
return flags;
}
-#else
+#elif ARCH_AARCH64
uint32_t x264_cpu_detect( void )
{
- return 0;
+ return X264_CPU_ARMV8 | X264_CPU_NEON;
}
+#elif ARCH_MIPS
+
+uint32_t x264_cpu_detect( void )
+{
+ uint32_t flags = 0;
+#if HAVE_MSA
+ flags |= X264_CPU_MSA;
#endif
+ return flags;
+}
+
+#else
-#ifndef HAVE_MMX
-void x264_emms( void )
+uint32_t x264_cpu_detect( void )
{
+ return 0;
}
-#endif
+#endif
int x264_cpu_num_processors( void )
{
-#if !defined(HAVE_PTHREAD)
+#if !HAVE_THREAD
return 1;
-#elif defined(_WIN32)
- return pthread_num_processors_np();
+#elif SYS_WINDOWS
+ return x264_pthread_num_processors_np();
+
+#elif SYS_CYGWIN || SYS_SunOS
+ return sysconf( _SC_NPROCESSORS_ONLN );
-#elif defined(SYS_LINUX)
- unsigned int bit;
- int np;
+#elif SYS_LINUX
+#ifdef __ANDROID__
+ // Android NDK does not expose sched_getaffinity
+ return sysconf( _SC_NPROCESSORS_CONF );
+#else
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
+#endif
-#elif defined(SYS_BEOS)
+#elif SYS_BEOS
system_info info;
get_system_info( &info );
return info.cpu_count;
-#elif defined(SYS_MACOSX) || defined(SYS_FREEBSD) || defined(SYS_OPENBSD)
- int numberOfCPUs;
- size_t length = sizeof( numberOfCPUs );
-#ifdef SYS_OPENBSD
+#elif SYS_MACOSX || SYS_FREEBSD || SYS_OPENBSD
+ int ncpu;
+ size_t length = sizeof( ncpu );
+#if SYS_OPENBSD
int mib[2] = { CTL_HW, HW_NCPU };
- if( sysctl(mib, 2, &numberOfCPUs, &length, NULL, 0) )
+ if( sysctl(mib, 2, &ncpu, &length, NULL, 0) )
#else
- if( sysctlbyname("hw.ncpu", &numberOfCPUs, &length, NULL, 0) )
+ if( sysctlbyname("hw.ncpu", &ncpu, &length, NULL, 0) )
#endif
{
- numberOfCPUs = 1;
+ ncpu = 1;
}
- return numberOfCPUs;
+ return ncpu;
#else
return 1;
projects / x264 / blobdiff