sysdeps/x86_64/multiarch/init-arch.h

   1 /* This file is part of the GNU C Library.
   2    Copyright (C) 2008-2013 Free Software Foundation, Inc.
   3
   4    The GNU C Library is free software; you can redistribute it and/or
   5    modify it under the terms of the GNU Lesser General Public
   6    License as published by the Free Software Foundation; either
   7    version 2.1 of the License, or (at your option) any later version.
   8
   9    The GNU C Library is distributed in the hope that it will be useful,
  10    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12    Lesser General Public License for more details.
  13
  14    You should have received a copy of the GNU Lesser General Public
  15    License along with the GNU C Library; if not, see
  16    <http://www.gnu.org/licenses/>.  */
  17
  18 #define bit_Fast_Rep_String             (1 << 0)
  19 #define bit_Fast_Copy_Backward          (1 << 1)
  20 #define bit_Slow_BSF                    (1 << 2)
  21 #define bit_Prefer_SSE_for_memop        (1 << 3)
  22 #define bit_Fast_Unaligned_Load         (1 << 4)
  23 #define bit_Prefer_PMINUB_for_stringop  (1 << 5)
  24 #define bit_AVX_Usable                  (1 << 6)
  25 #define bit_FMA_Usable                  (1 << 7)
  26 #define bit_FMA4_Usable                 (1 << 8)
  27
  28 /* CPUID Feature flags.  */
  29
  30 /* COMMON_CPUID_INDEX_1.  */
  31 #define bit_SSE2        (1 << 26)
  32 #define bit_SSSE3       (1 << 9)
  33 #define bit_SSE4_1      (1 << 19)
  34 #define bit_SSE4_2      (1 << 20)
  35 #define bit_OSXSAVE     (1 << 27)
  36 #define bit_AVX         (1 << 28)
  37 #define bit_POPCOUNT    (1 << 23)
  38 #define bit_FMA         (1 << 12)
  39 #define bit_FMA4        (1 << 16)
  40
  41 /* COMMON_CPUID_INDEX_7.  */
  42 #define bit_RTM         (1 << 11)
  43
  44 /* XCR0 Feature flags.  */
  45 #define bit_XMM_state  (1 << 1)
  46 #define bit_YMM_state  (2 << 1)
  47
  48 #ifdef  __ASSEMBLER__
  49
  50 # include <ifunc-defines.h>
  51
  52 # define index_SSE2     COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
  53 # define index_SSSE3    COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  54 # define index_SSE4_1   COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  55 # define index_SSE4_2   COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  56 # define index_AVX      COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  57
  58 # define index_Fast_Rep_String          FEATURE_INDEX_1*FEATURE_SIZE
  59 # define index_Fast_Copy_Backward       FEATURE_INDEX_1*FEATURE_SIZE
  60 # define index_Slow_BSF                 FEATURE_INDEX_1*FEATURE_SIZE
  61 # define index_Prefer_SSE_for_memop     FEATURE_INDEX_1*FEATURE_SIZE
  62 # define index_Fast_Unaligned_Load      FEATURE_INDEX_1*FEATURE_SIZE
  63 # define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1*FEATURE_SIZE
  64 # define index_AVX_Usable               FEATURE_INDEX_1*FEATURE_SIZE
  65 # define index_FMA_Usable               FEATURE_INDEX_1*FEATURE_SIZE
  66 # define index_FMA4_Usable              FEATURE_INDEX_1*FEATURE_SIZE
  67
  68 #else   /* __ASSEMBLER__ */
  69
  70 # include <sys/param.h>
  71
  72 enum
  73   {
  74     COMMON_CPUID_INDEX_1 = 0,
  75     COMMON_CPUID_INDEX_7,
  76     COMMON_CPUID_INDEX_80000001,        /* for AMD */
  77     /* Keep the following line at the end.  */
  78     COMMON_CPUID_INDEX_MAX
  79   };
  80
  81 enum
  82   {
  83     FEATURE_INDEX_1 = 0,
  84     /* Keep the following line at the end.  */
  85     FEATURE_INDEX_MAX
  86   };
  87
  88 extern struct cpu_features
  89 {
  90   enum cpu_features_kind
  91     {
  92       arch_kind_unknown = 0,
  93       arch_kind_intel,
  94       arch_kind_amd,
  95       arch_kind_other
  96     } kind;
  97   int max_cpuid;
  98   struct cpuid_registers
  99   {
 100     unsigned int eax;
 101     unsigned int ebx;
 102     unsigned int ecx;
 103     unsigned int edx;
 104   } cpuid[COMMON_CPUID_INDEX_MAX];
 105   unsigned int family;
 106   unsigned int model;
 107   unsigned int feature[FEATURE_INDEX_MAX];
 108 } __cpu_features attribute_hidden;
 109
 110
 111 extern void __init_cpu_features (void) attribute_hidden;
 112
 113 /* Implicitly call __init_cpu_features before accessing the CPU features
 114    structure.  */
 115
 116 extern __always_inline attribute_hidden
 117 const struct cpu_features *
 118 __get_cpu_features (void)
 119 {
 120   if (__cpu_features.kind == arch_kind_unknown)
 121     __init_cpu_features ();
 122   return &__cpu_features;
 123 }
 124
 125 # define HAS_CPU_FEATURE(idx, reg, bit) \
 126   ((__get_cpu_features ()->cpuid[idx].reg & (bit)) != 0)
 127
 128 /* Following are the feature tests used throughout libc.  */
 129
 130 /* CPUID_* evaluates to true if the feature flag is enabled.
 131    We always use &__cpu_features because the HAS_CPUID_* macros
 132    are called only within __init_cpu_features, where we can't
 133    call __get_cpu_features without infinite recursion.  */
 134 # define HAS_CPUID_FLAG(idx, reg, bit) \
 135   (((&__cpu_features)->cpuid[idx].reg & (bit)) != 0)
 136
 137 # define CPUID_OSXSAVE \
 138   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_1, ecx, bit_OSXSAVE)
 139 # define CPUID_AVX \
 140   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_1, ecx, bit_AVX)
 141 # define CPUID_FMA \
 142   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_1, ecx, bit_FMA)
 143 # define CPUID_FMA4 \
 144   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_80000001, ecx, bit_FMA4)
 145 # define CPUID_RTM \
 146   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_7, ebx, bit_RTM)
 147
 148 /* HAS_* evaluates to true if we may use the feature at runtime.  */
 149 # define HAS_SSE2       HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, edx, bit_SSE2)
 150 # define HAS_POPCOUNT   HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_POPCOUNT)
 151 # define HAS_SSSE3      HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSSE3)
 152 # define HAS_SSE4_1     HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSE4_1)
 153 # define HAS_SSE4_2     HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSE4_2)
 154 # define HAS_RTM        HAS_CPU_FEATURE (COMMON_CPUID_INDEX_7, ebx, bit_RTM)
 155
 156 # define index_Fast_Rep_String          FEATURE_INDEX_1
 157 # define index_Fast_Copy_Backward       FEATURE_INDEX_1
 158 # define index_Slow_BSF                 FEATURE_INDEX_1
 159 # define index_Prefer_SSE_for_memop     FEATURE_INDEX_1
 160 # define index_Fast_Unaligned_Load      FEATURE_INDEX_1
 161 # define index_AVX_Usable               FEATURE_INDEX_1
 162 # define index_FMA_Usable               FEATURE_INDEX_1
 163 # define index_FMA4_Usable              FEATURE_INDEX_1
 164
 165 # define HAS_ARCH_FEATURE(name) \
 166   ((__get_cpu_features ()->feature[index_##name] & (bit_##name)) != 0)
 167
 168 # define HAS_FAST_REP_STRING            HAS_ARCH_FEATURE (Fast_Rep_String)
 169 # define HAS_FAST_COPY_BACKWARD         HAS_ARCH_FEATURE (Fast_Copy_Backward)
 170 # define HAS_SLOW_BSF                   HAS_ARCH_FEATURE (Slow_BSF)
 171 # define HAS_PREFER_SSE_FOR_MEMOP       HAS_ARCH_FEATURE (Prefer_SSE_for_memop)
 172 # define HAS_FAST_UNALIGNED_LOAD        HAS_ARCH_FEATURE (Fast_Unaligned_Load)
 173 # define HAS_AVX                        HAS_ARCH_FEATURE (AVX_Usable)
 174 # define HAS_FMA                        HAS_ARCH_FEATURE (FMA_Usable)
 175 # define HAS_FMA4                       HAS_ARCH_FEATURE (FMA4_Usable)
 176
 177 #endif  /* __ASSEMBLER__ */