+2009-07-31 H.J. Lu <hongjiu.lu@intel.com>
+
+ * sysdeps/i386/i686/Makefile (sysdep_routines): Add cacheinfo.
+ * sysdeps/i386/i686/cacheinfo.c: New file.
+ * sysdeps/i386/i686/multiarch/Makefile: New file.
+ * sysdeps/i386/i686/multiarch/ifunc-defines.sym: New file.
+ * sysdeps/i386/i686/multiarch/init-arch.c: New file.
+ * sysdeps/i386/i686/multiarch/init-arch.h: New file.
+ * sysdeps/i386/i686/multiarch/sched_cpucount.c: New file.
+ * sysdeps/unix/sysv/linux/i386/i686/sysconf.c: New file.
+ * sysdeps/x86_64/cacheinfo.c: Include <cpuid.h>.
+ (__cpuid_count): New. Provide the default.
+ (__x86_64_prefetchw): Define only if DISABLE_PREFETCHW is not defined.
+ (__x86_64_preferred_memory_instruction): Define only if
+ DISABLE_PREFERRED_MEMORY_INSTRUCTION is not defined.
+ (intel_check_word): Use __cpuid.
+ (handle_intel): Likewise.
+ (handle_amd): Likewise.
+ (__cache_sysconf): Likewise.
+ (init_cacheinfo): Updated. Use __cpuid and __cpuid_count.
+ (__cache_sysconf): Likewise.
+ (init_cacheinfo): Updated. Use __cpuid and __cpuid_count.
+ * sysdeps/x86_64/multiarch/init-arch.c: Include <cpuid.h>.
+ (get_common_indeces): Use __cpuid.
+ (__init_cpu_features): Likewise. Disable SSSE3 on Atom only
+ if ENABLE_SSSE3_ON_ATOM is not defined.
+ * sysdeps/x86_64/multiarch/init-arch.h (HAS_SSE2): Define.
+ * sysdeps/x86_64/multiarch/sched_cpucount.c (POPCNT): Use
+ popcnt instead of popcntq.
+
2009-07-31 Jakub Jelinek <jakub@redhat.com>
* malloc/Makefile (CFLAGS-obstack.c): Add $(uses-callbacks).
#include <stdbool.h>
#include <stdlib.h>
#include <unistd.h>
+#include <cpuid.h>
+
+#ifndef __cpuid_count
+/* FIXME: Provide __cpuid_count if it isn't defined. Copied from gcc
+ 4.4.0. Remove this if gcc 4.4 is the minimum requirement. */
+# if defined(__i386__) && defined(__PIC__)
+/* %ebx may be the PIC register. */
+# define __cpuid_count(level, count, a, b, c, d) \
+ __asm__ ("xchg{l}\t{%%}ebx, %1\n\t" \
+ "cpuid\n\t" \
+ "xchg{l}\t{%%}ebx, %1\n\t" \
+ : "=a" (a), "=r" (b), "=c" (c), "=d" (d) \
+ : "0" (level), "2" (count))
+# else
+# define __cpuid_count(level, count, a, b, c, d) \
+ __asm__ ("cpuid\n\t" \
+ : "=a" (a), "=b" (b), "=c" (c), "=d" (d) \
+ : "0" (level), "2" (count))
+# endif
+#endif
#ifdef USE_MULTIARCH
# include "multiarch/init-arch.h"
unsigned int ebx;
unsigned int ecx;
unsigned int edx;
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (1));
+ __cpuid (1, eax, ebx, ecx, edx);
family = ((eax >> 20) & 0xff) + ((eax >> 8) & 0xf);
model = (((eax >>16) & 0xf) << 4) + ((eax >> 4) & 0xf);
unsigned int ebx;
unsigned int ecx;
unsigned int edx;
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (2));
+ __cpuid (2, eax, ebx, ecx, edx);
/* The low byte of EAX in the first round contain the number of
rounds we have to make. At least one, the one we are already
unsigned int ebx;
unsigned int ecx;
unsigned int edx;
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (0x80000000));
+ __cpuid (0x80000000, eax, ebx, ecx, edx);
/* No level 4 cache (yet). */
if (name > _SC_LEVEL3_CACHE_LINESIZE)
if (eax < fn)
return 0;
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (fn));
+ __cpuid (fn, eax, ebx, ecx, edx);
if (name < _SC_LEVEL1_DCACHE_SIZE)
{
unsigned int ebx;
unsigned int ecx;
unsigned int edx;
- asm volatile ("cpuid"
- : "=a" (max_cpuid), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (0));
+ __cpuid (0, max_cpuid, ebx, ecx, edx);
#endif
if (is_intel)
L2 or L3 size. */
long int __x86_64_shared_cache_size_half attribute_hidden = 1024 * 1024 / 2;
long int __x86_64_shared_cache_size attribute_hidden = 1024 * 1024;
+
+#ifndef DISABLE_PREFETCHW
/* PREFETCHW support flag for use in memory and string routines. */
int __x86_64_prefetchw attribute_hidden;
+#endif
+#ifndef DISABLE_PREFERRED_MEMORY_INSTRUCTION
/* Instructions preferred for memory and string routines.
0: Regular instructions
*/
int __x86_64_preferred_memory_instruction attribute_hidden;
+#endif
static void
__init_cpu_features ();
#else
int max_cpuid;
- asm volatile ("cpuid"
- : "=a" (max_cpuid), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (0));
+ __cpuid (0, max_cpuid, ebx, ecx, edx);
#endif
if (is_intel)
ecx = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx;
edx = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].edx;
#else
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (1));
+ __cpuid (1, eax, ebx, ecx, edx);
#endif
+#ifndef DISABLE_PREFERRED_MEMORY_INSTRUCTION
/* Intel prefers SSSE3 instructions for memory/string routines
if they are avaiable. */
if ((ecx & 0x200))
__x86_64_preferred_memory_instruction = 3;
else
__x86_64_preferred_memory_instruction = 2;
+#endif
/* Figure out the number of logical threads that share the
highest cache level. */
/* Query until desired cache level is enumerated. */
do
{
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (4), "2" (i++));
+ __cpuid_count (4, i++, eax, ebx, ecx, edx);
/* There seems to be a bug in at least some Pentium Ds
which sometimes fail to iterate all cache parameters.
shared = handle_amd (_SC_LEVEL3_CACHE_SIZE);
/* Get maximum extended function. */
- asm volatile ("cpuid"
- : "=a" (max_cpuid_ex), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (0x80000000));
+ __cpuid (0x80000000, max_cpuid_ex, ebx, ecx, edx);
if (shared <= 0)
/* No shared L3 cache. All we have is the L2 cache. */
if (max_cpuid_ex >= 0x80000008)
{
/* Get width of APIC ID. */
- asm volatile ("cpuid"
- : "=a" (max_cpuid_ex), "=b" (ebx), "=c" (ecx),
- "=d" (edx)
- : "0" (0x80000008));
+ __cpuid (0x80000008, max_cpuid_ex, ebx, ecx, edx);
threads = 1 << ((ecx >> 12) & 0x0f);
}
{
/* If APIC ID width is not available, use logical
processor count. */
- asm volatile ("cpuid"
- : "=a" (max_cpuid_ex), "=b" (ebx), "=c" (ecx),
- "=d" (edx)
- : "0" (0x00000001));
+ __cpuid (0x00000001, max_cpuid_ex, ebx, ecx, edx);
if ((edx & (1 << 28)) != 0)
threads = (ebx >> 16) & 0xff;
shared += core;
}
+#ifndef DISABLE_PREFETCHW
if (max_cpuid_ex >= 0x80000001)
{
- asm volatile ("cpuid"
- : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
- : "0" (0x80000001));
+ __cpuid (0x80000001, eax, ebx, ecx, edx);
/* PREFETCHW || 3DNow! */
if ((ecx & 0x100) || (edx & 0x80000000))
__x86_64_prefetchw = -1;
}
+#endif
}
if (data > 0)
Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
02111-1307 USA. */
+#include <cpuid.h>
#include "init-arch.h"
static void
get_common_indeces (void)
{
- asm volatile ("cpuid"
- : "=a" (__cpu_features.cpuid[COMMON_CPUID_INDEX_1].eax),
- "=b" (__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ebx),
- "=c" (__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx),
- "=d" (__cpu_features.cpuid[COMMON_CPUID_INDEX_1].edx)
- : "0" (1));
+ __cpuid (1, __cpu_features.cpuid[COMMON_CPUID_INDEX_1].eax,
+ __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ebx,
+ __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx,
+ __cpu_features.cpuid[COMMON_CPUID_INDEX_1].edx);
unsigned int eax = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].eax;
__cpu_features.family = (eax >> 8) & 0x0f;
unsigned int ecx;
unsigned int edx;
- asm volatile ("cpuid"
- : "=a" (__cpu_features.max_cpuid), "=b" (ebx), "=c" (ecx),
- "=d" (edx)
- : "0" (0));
+ __cpuid (0, __cpu_features.max_cpuid, ebx, ecx, edx);
/* This spells out "GenuineIntel". */
if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
{
__cpu_features.model += extended_model;
+#ifndef ENABLE_SSSE3_ON_ATOM
if (__cpu_features.model == 0x1c)
/* Avoid SSSE3 on Atom since it is slow. */
__cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx &= ~(1 << 9);
+#endif
}
}
/* This spells out "AuthenticAMD". */
projects / platform / upstream / glibc.git / commitdiff