sysdeps/x86_64/multiarch/init-arch.h

   1 /* This file is part of the GNU C Library.
   2    Copyright (C) 2008-2015 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_Fast_Unaligned_Load         (1 << 4)
  22 #define bit_Prefer_PMINUB_for_stringop  (1 << 5)
  23 #define bit_AVX_Usable                  (1 << 6)
  24 #define bit_FMA_Usable                  (1 << 7)
  25 #define bit_FMA4_Usable                 (1 << 8)
  26 #define bit_Slow_SSE4_2                 (1 << 9)
  27 #define bit_AVX2_Usable                 (1 << 10)
  28
  29 /* CPUID Feature flags.  */
  30
  31 /* COMMON_CPUID_INDEX_1.  */
  32 #define bit_SSE2        (1 << 26)
  33 #define bit_SSSE3       (1 << 9)
  34 #define bit_SSE4_1      (1 << 19)
  35 #define bit_SSE4_2      (1 << 20)
  36 #define bit_OSXSAVE     (1 << 27)
  37 #define bit_AVX         (1 << 28)
  38 #define bit_POPCOUNT    (1 << 23)
  39 #define bit_FMA         (1 << 12)
  40 #define bit_FMA4        (1 << 16)
  41
  42 /* COMMON_CPUID_INDEX_7.  */
  43 #define bit_RTM         (1 << 11)
  44 #define bit_AVX2        (1 << 5)
  45
  46 /* XCR0 Feature flags.  */
  47 #define bit_XMM_state  (1 << 1)
  48 #define bit_YMM_state  (2 << 1)
  49
  50 /* The integer bit array index for the first set of internal feature bits.  */
  51 # define FEATURE_INDEX_1 0
  52
  53 /* The current maximum size of the feature integer bit array.  */
  54 # define FEATURE_INDEX_MAX 1
  55
  56 #ifdef  __ASSEMBLER__
  57
  58 # include <ifunc-defines.h>
  59
  60 # define index_SSE2     COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
  61 # define index_SSSE3    COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  62 # define index_SSE4_1   COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  63 # define index_SSE4_2   COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  64 # define index_AVX      COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
  65 # define index_AVX2     COMMON_CPUID_INDEX_7*CPUID_SIZE+CPUID_EBX_OFFSET
  66
  67 # define index_Fast_Rep_String          FEATURE_INDEX_1*FEATURE_SIZE
  68 # define index_Fast_Copy_Backward       FEATURE_INDEX_1*FEATURE_SIZE
  69 # define index_Slow_BSF                 FEATURE_INDEX_1*FEATURE_SIZE
  70 # define index_Fast_Unaligned_Load      FEATURE_INDEX_1*FEATURE_SIZE
  71 # define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1*FEATURE_SIZE
  72 # define index_AVX_Usable               FEATURE_INDEX_1*FEATURE_SIZE
  73 # define index_FMA_Usable               FEATURE_INDEX_1*FEATURE_SIZE
  74 # define index_FMA4_Usable              FEATURE_INDEX_1*FEATURE_SIZE
  75 # define index_Slow_SSE4_2              FEATURE_INDEX_1*FEATURE_SIZE
  76 # define index_AVX2_Usable              FEATURE_INDEX_1*FEATURE_SIZE
  77
  78 #else   /* __ASSEMBLER__ */
  79
  80 # include <sys/param.h>
  81
  82 enum
  83   {
  84     COMMON_CPUID_INDEX_1 = 0,
  85     COMMON_CPUID_INDEX_7,
  86     COMMON_CPUID_INDEX_80000001,        /* for AMD */
  87     /* Keep the following line at the end.  */
  88     COMMON_CPUID_INDEX_MAX
  89   };
  90
  91 extern struct cpu_features
  92 {
  93   enum cpu_features_kind
  94     {
  95       arch_kind_unknown = 0,
  96       arch_kind_intel,
  97       arch_kind_amd,
  98       arch_kind_other
  99     } kind;
 100   int max_cpuid;
 101   struct cpuid_registers
 102   {
 103     unsigned int eax;
 104     unsigned int ebx;
 105     unsigned int ecx;
 106     unsigned int edx;
 107   } cpuid[COMMON_CPUID_INDEX_MAX];
 108   unsigned int family;
 109   unsigned int model;
 110   unsigned int feature[FEATURE_INDEX_MAX];
 111 } __cpu_features attribute_hidden;
 112
 113
 114 extern void __init_cpu_features (void) attribute_hidden;
 115 # define INIT_ARCH() \
 116   do                                                    \
 117     if (__cpu_features.kind == arch_kind_unknown)       \
 118       __init_cpu_features ();                           \
 119   while (0)
 120
 121 /* Used from outside libc.so to get access to the CPU features structure.  */
 122 extern const struct cpu_features *__get_cpu_features (void)
 123      __attribute__ ((const));
 124
 125 # if IS_IN (libc)
 126 #  define __get_cpu_features()  (&__cpu_features)
 127 # endif
 128
 129 # define HAS_CPU_FEATURE(idx, reg, bit) \
 130   ((__get_cpu_features ()->cpuid[idx].reg & (bit)) != 0)
 131
 132 /* Following are the feature tests used throughout libc.  */
 133
 134 /* CPUID_* evaluates to true if the feature flag is enabled.
 135    We always use &__cpu_features because the HAS_CPUID_* macros
 136    are called only within __init_cpu_features, where we can't
 137    call __get_cpu_features without infinite recursion.  */
 138 # define HAS_CPUID_FLAG(idx, reg, bit) \
 139   (((&__cpu_features)->cpuid[idx].reg & (bit)) != 0)
 140
 141 # define CPUID_OSXSAVE \
 142   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_1, ecx, bit_OSXSAVE)
 143 # define CPUID_AVX \
 144   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_1, ecx, bit_AVX)
 145 # define CPUID_FMA \
 146   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_1, ecx, bit_FMA)
 147 # define CPUID_FMA4 \
 148   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_80000001, ecx, bit_FMA4)
 149 # define CPUID_RTM \
 150   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_7, ebx, bit_RTM)
 151 # define CPUID_AVX2 \
 152   HAS_CPUID_FLAG (COMMON_CPUID_INDEX_7, ebx, bit_AVX2)
 153
 154 /* HAS_* evaluates to true if we may use the feature at runtime.  */
 155 # define HAS_SSE2       HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, edx, bit_SSE2)
 156 # define HAS_POPCOUNT   HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_POPCOUNT)
 157 # define HAS_SSSE3      HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSSE3)
 158 # define HAS_SSE4_1     HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSE4_1)
 159 # define HAS_SSE4_2     HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, bit_SSE4_2)
 160 # define HAS_RTM        HAS_CPU_FEATURE (COMMON_CPUID_INDEX_7, ebx, bit_RTM)
 161
 162 # define index_Fast_Rep_String          FEATURE_INDEX_1
 163 # define index_Fast_Copy_Backward       FEATURE_INDEX_1
 164 # define index_Slow_BSF                 FEATURE_INDEX_1
 165 # define index_Fast_Unaligned_Load      FEATURE_INDEX_1
 166 # define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1
 167 # define index_AVX_Usable               FEATURE_INDEX_1
 168 # define index_FMA_Usable               FEATURE_INDEX_1
 169 # define index_FMA4_Usable              FEATURE_INDEX_1
 170 # define index_Slow_SSE4_2              FEATURE_INDEX_1
 171 # define index_AVX2_Usable              FEATURE_INDEX_1
 172
 173 # define HAS_ARCH_FEATURE(name) \
 174   ((__get_cpu_features ()->feature[index_##name] & (bit_##name)) != 0)
 175
 176 # define HAS_FAST_REP_STRING            HAS_ARCH_FEATURE (Fast_Rep_String)
 177 # define HAS_FAST_COPY_BACKWARD         HAS_ARCH_FEATURE (Fast_Copy_Backward)
 178 # define HAS_SLOW_BSF                   HAS_ARCH_FEATURE (Slow_BSF)
 179 # define HAS_FAST_UNALIGNED_LOAD        HAS_ARCH_FEATURE (Fast_Unaligned_Load)
 180 # define HAS_AVX                        HAS_ARCH_FEATURE (AVX_Usable)
 181 # define HAS_AVX2                       HAS_ARCH_FEATURE (AVX2_Usable)
 182 # define HAS_FMA                        HAS_ARCH_FEATURE (FMA_Usable)
 183 # define HAS_FMA4                       HAS_ARCH_FEATURE (FMA4_Usable)
 184
 185 #endif  /* __ASSEMBLER__ */