sysdeps/x86_64/multiarch/init-arch.c

   1 /* Initialize CPU feature data.
   2    This file is part of the GNU C Library.
   3    Copyright (C) 2008, 2009, 2010, 2011 Free Software Foundation, Inc.
   4    Contributed by Ulrich Drepper <drepper@redhat.com>.
   5
   6    The GNU C Library is free software; you can redistribute it and/or
   7    modify it under the terms of the GNU Lesser General Public
   8    License as published by the Free Software Foundation; either
   9    version 2.1 of the License, or (at your option) any later version.
  10
  11    The GNU C Library is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14    Lesser General Public License for more details.
  15
  16    You should have received a copy of the GNU Lesser General Public
  17    License along with the GNU C Library; if not, write to the Free
  18    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  19    02111-1307 USA.  */
  20
  21 #include <atomic.h>
  22 #include <cpuid.h>
  23 #include "init-arch.h"
  24
  25
  26 struct cpu_features __cpu_features attribute_hidden;
  27
  28
  29 static void
  30 get_common_indeces (unsigned int *family, unsigned int *model)
  31 {
  32   __cpuid (1, __cpu_features.cpuid[COMMON_CPUID_INDEX_1].eax,
  33            __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ebx,
  34            __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx,
  35            __cpu_features.cpuid[COMMON_CPUID_INDEX_1].edx);
  36
  37   unsigned int eax = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].eax;
  38   *family = (eax >> 8) & 0x0f;
  39   *model = (eax >> 4) & 0x0f;
  40 }
  41
  42
  43 void
  44 __init_cpu_features (void)
  45 {
  46   unsigned int ebx;
  47   unsigned int ecx;
  48   unsigned int edx;
  49   unsigned int family = 0;
  50   unsigned int model = 0;
  51   enum cpu_features_kind kind;
  52
  53   __cpuid (0, __cpu_features.max_cpuid, ebx, ecx, edx);
  54
  55   /* This spells out "GenuineIntel".  */
  56   if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
  57     {
  58       kind = arch_kind_intel;
  59
  60       get_common_indeces (&family, &model);
  61
  62       /* Intel processors prefer SSE instruction for memory/string
  63          routines if they are available.  */
  64       __cpu_features.feature[index_Prefer_SSE_for_memop]
  65         |= bit_Prefer_SSE_for_memop;
  66
  67       unsigned int eax = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].eax;
  68       unsigned int extended_family = (eax >> 20) & 0xff;
  69       unsigned int extended_model = (eax >> 12) & 0xf0;
  70       if (family == 0x0f)
  71         {
  72           family += extended_family;
  73           model += extended_model;
  74         }
  75       else if (family == 0x06)
  76         {
  77           ecx = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx;
  78           model += extended_model;
  79           switch (model)
  80             {
  81             case 0x1c:
  82             case 0x26:
  83               /* BSF is slow on Atom.  */
  84               __cpu_features.feature[index_Slow_BSF] |= bit_Slow_BSF;
  85               break;
  86
  87             default:
  88               /* Unknown family 0x06 processors.  Assuming this is one
  89                  of Core i3/i5/i7 processors if AVX is available.  */
  90               if ((ecx & bit_AVX) == 0)
  91                 break;
  92
  93             case 0x1a:
  94             case 0x1e:
  95             case 0x1f:
  96             case 0x25:
  97             case 0x2c:
  98             case 0x2e:
  99             case 0x2f:
 100               /* Rep string instructions, copy backward and unaligned loads
 101                  are fast on Intel Core i3, i5 and i7.  */
 102 #if index_Fast_Rep_String != index_Fast_Copy_Backward
 103 # error index_Fast_Rep_String != index_Fast_Copy_Backward
 104 #endif
 105 #if index_Fast_Rep_String != index_Fast_Unaligned_Load
 106 # error index_Fast_Rep_String != index_Fast_Unaligned_Load
 107 #endif
 108               __cpu_features.feature[index_Fast_Rep_String]
 109                 |= (bit_Fast_Rep_String
 110                     | bit_Fast_Copy_Backward
 111                     | bit_Fast_Unaligned_Load);
 112               break;
 113             }
 114         }
 115     }
 116   /* This spells out "AuthenticAMD".  */
 117   else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
 118     {
 119       kind = arch_kind_amd;
 120
 121       get_common_indeces (&family, &model);
 122
 123       unsigned int ecx = __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx;
 124
 125       /* AMD processors prefer SSE instructions for memory/string routines
 126          if they are available, otherwise they prefer integer instructions.  */
 127       if ((ecx & 0x200))
 128         __cpu_features.feature[index_Prefer_SSE_for_memop]
 129           |= bit_Prefer_SSE_for_memop;
 130     }
 131   else
 132     kind = arch_kind_other;
 133
 134   __cpu_features.family = family;
 135   __cpu_features.model = model;
 136   atomic_write_barrier ();
 137   __cpu_features.kind = kind;
 138 }
 139
 140 #undef __get_cpu_features
 141
 142 const struct cpu_features *
 143 __get_cpu_features (void)
 144 {
 145   if (__cpu_features.kind == arch_kind_unknown)
 146     __init_cpu_features ();
 147
 148   return &__cpu_features;
 149 }