Check for FMA4 support and generate appropriate fma functions
[glibc.git] / sysdeps / x86_64 / multiarch / init-arch.h
blob2fb6f75b66527a4abe3ecab5844e3b7bf8517e75
1 /* This file is part of the GNU C Library.
2 Copyright (C) 2008, 2009, 2010, 2011 Free Software Foundation, Inc.
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.
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.
14 You should have received a copy of the GNU Lesser General Public
15 License along with the GNU C Library; if not, write to the Free
16 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17 02111-1307 USA. */
19 #define bit_Fast_Rep_String (1 << 0)
20 #define bit_Fast_Copy_Backward (1 << 1)
21 #define bit_Slow_BSF (1 << 2)
22 #define bit_Prefer_SSE_for_memop (1 << 3)
23 #define bit_Fast_Unaligned_Load (1 << 4)
24 #define bit_Prefer_PMINUB_for_stringop (1 << 5)
26 #ifdef __ASSEMBLER__
28 # include <ifunc-defines.h>
30 # define bit_SSE2 (1 << 26)
31 # define bit_SSSE3 (1 << 9)
32 # define bit_SSE4_1 (1 << 19)
33 # define bit_SSE4_2 (1 << 20)
34 # define bit_AVX (1 << 28)
36 # define index_SSE2 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
37 # define index_SSSE3 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
38 # define index_SSE4_1 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
39 # define index_SSE4_2 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
40 # define index_AVX COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
42 # define index_Fast_Rep_String FEATURE_INDEX_1*FEATURE_SIZE
43 # define index_Fast_Copy_Backward FEATURE_INDEX_1*FEATURE_SIZE
44 # define index_Slow_BSF FEATURE_INDEX_1*FEATURE_SIZE
45 # define index_Prefer_SSE_for_memop FEATURE_INDEX_1*FEATURE_SIZE
46 # define index_Fast_Unaligned_Load FEATURE_INDEX_1*FEATURE_SIZE
47 # define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1*FEATURE_SIZE
49 #else /* __ASSEMBLER__ */
51 # include <sys/param.h>
53 enum
55 COMMON_CPUID_INDEX_1 = 0,
56 COMMON_CPUID_INDEX_80000001, /* for AMD */
57 /* Keep the following line at the end. */
58 COMMON_CPUID_INDEX_MAX
61 enum
63 FEATURE_INDEX_1 = 0,
64 /* Keep the following line at the end. */
65 FEATURE_INDEX_MAX
68 extern struct cpu_features
70 enum cpu_features_kind
72 arch_kind_unknown = 0,
73 arch_kind_intel,
74 arch_kind_amd,
75 arch_kind_other
76 } kind;
77 int max_cpuid;
78 struct cpuid_registers
80 unsigned int eax;
81 unsigned int ebx;
82 unsigned int ecx;
83 unsigned int edx;
84 } cpuid[COMMON_CPUID_INDEX_MAX];
85 unsigned int family;
86 unsigned int model;
87 unsigned int feature[FEATURE_INDEX_MAX];
88 } __cpu_features attribute_hidden;
91 extern void __init_cpu_features (void) attribute_hidden;
92 #define INIT_ARCH()\
93 do \
94 if (__cpu_features.kind == arch_kind_unknown) \
95 __init_cpu_features (); \
96 while (0)
98 /* Used from outside libc.so to get access to the CPU features structure. */
99 extern const struct cpu_features *__get_cpu_features (void)
100 __attribute__ ((const));
102 # ifndef NOT_IN_libc
103 # define __get_cpu_features() (&__cpu_features)
104 # endif
106 # define HAS_CPU_FEATURE(idx, reg, bit) \
107 ((__get_cpu_features ()->cpuid[idx].reg & (1 << (bit))) != 0)
109 /* Following are the feature tests used throughout libc. */
111 # define HAS_SSE2 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, edx, 26)
112 # define HAS_POPCOUNT HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 23)
113 # define HAS_SSSE3 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 9)
114 # define HAS_SSE4_1 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 19)
115 # define HAS_SSE4_2 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 20)
116 # define HAS_FMA HAS_CPU_FEATURE (COMMON_CPUID_INDEX_1, ecx, 12)
117 # define HAS_FMA4 HAS_CPU_FEATURE (COMMON_CPUID_INDEX_80000001, ecx, 16)
119 # define index_Fast_Rep_String FEATURE_INDEX_1
120 # define index_Fast_Copy_Backward FEATURE_INDEX_1
121 # define index_Slow_BSF FEATURE_INDEX_1
122 # define index_Prefer_SSE_for_memop FEATURE_INDEX_1
123 # define index_Fast_Unaligned_Load FEATURE_INDEX_1
125 #define HAS_ARCH_FEATURE(idx, bit) \
126 ((__get_cpu_features ()->feature[idx] & (bit)) != 0)
128 #define HAS_FAST_REP_STRING \
129 HAS_ARCH_FEATURE (index_Fast_Rep_String, bit_Fast_Rep_String)
131 #define HAS_FAST_COPY_BACKWARD \
132 HAS_ARCH_FEATURE (index_Fast_Copy_Backward, bit_Fast_Copy_Backward)
134 #define HAS_SLOW_BSF \
135 HAS_ARCH_FEATURE (index_Slow_BSF, bit_Slow_BSF)
137 #define HAS_PREFER_SSE_FOR_MEMOP \
138 HAS_ARCH_FEATURE (index_Prefer_SSE_for_memop, bit_Prefer_SSE_for_memop)
140 #define HAS_FAST_UNALIGNED_LOAD \
141 HAS_ARCH_FEATURE (index_Fast_Unaligned_Load, bit_Fast_Unaligned_Load)
143 #endif /* __ASSEMBLER__ */