1/* This file is part of the GNU C Library.
2 Copyright (C) 2008-2016 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#ifndef cpu_features_h
19#define cpu_features_h
20
21#define bit_arch_Fast_Rep_String (1 << 0)
22#define bit_arch_Fast_Copy_Backward (1 << 1)
23#define bit_arch_Slow_BSF (1 << 2)
24#define bit_arch_Fast_Unaligned_Load (1 << 4)
25#define bit_arch_Prefer_PMINUB_for_stringop (1 << 5)
26#define bit_arch_AVX_Usable (1 << 6)
27#define bit_arch_FMA_Usable (1 << 7)
28#define bit_arch_FMA4_Usable (1 << 8)
29#define bit_arch_Slow_SSE4_2 (1 << 9)
30#define bit_arch_AVX2_Usable (1 << 10)
31#define bit_arch_AVX_Fast_Unaligned_Load (1 << 11)
32#define bit_arch_AVX512F_Usable (1 << 12)
33#define bit_arch_AVX512DQ_Usable (1 << 13)
34#define bit_arch_I586 (1 << 14)
35#define bit_arch_I686 (1 << 15)
36#define bit_arch_Prefer_MAP_32BIT_EXEC (1 << 16)
37#define bit_arch_Prefer_No_VZEROUPPER (1 << 17)
38#define bit_arch_Fast_Unaligned_Copy (1 << 18)
39#define bit_arch_Prefer_ERMS (1 << 19)
40
41/* CPUID Feature flags. */
42
43/* COMMON_CPUID_INDEX_1. */
44#define bit_cpu_CX8 (1 << 8)
45#define bit_cpu_CMOV (1 << 15)
46#define bit_cpu_SSE2 (1 << 26)
47#define bit_cpu_SSSE3 (1 << 9)
48#define bit_cpu_SSE4_1 (1 << 19)
49#define bit_cpu_SSE4_2 (1 << 20)
50#define bit_cpu_OSXSAVE (1 << 27)
51#define bit_cpu_AVX (1 << 28)
52#define bit_cpu_POPCOUNT (1 << 23)
53#define bit_cpu_FMA (1 << 12)
54#define bit_cpu_FMA4 (1 << 16)
55#define bit_cpu_HTT (1 << 28)
56
57/* COMMON_CPUID_INDEX_7. */
58#define bit_cpu_ERMS (1 << 9)
59#define bit_cpu_RTM (1 << 11)
60#define bit_cpu_AVX2 (1 << 5)
61#define bit_cpu_AVX512F (1 << 16)
62#define bit_cpu_AVX512DQ (1 << 17)
63
64/* XCR0 Feature flags. */
65#define bit_XMM_state (1 << 1)
66#define bit_YMM_state (2 << 1)
67#define bit_Opmask_state (1 << 5)
68#define bit_ZMM0_15_state (1 << 6)
69#define bit_ZMM16_31_state (1 << 7)
70
71/* The integer bit array index for the first set of internal feature bits. */
72#define FEATURE_INDEX_1 0
73
74/* The current maximum size of the feature integer bit array. */
75#define FEATURE_INDEX_MAX 1
76
77#ifdef __ASSEMBLER__
78
79# include <cpu-features-offsets.h>
80
81# define index_cpu_CX8 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
82# define index_cpu_CMOV COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
83# define index_cpu_SSE2 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_EDX_OFFSET
84# define index_cpu_SSSE3 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
85# define index_cpu_SSE4_1 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
86# define index_cpu_SSE4_2 COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
87# define index_cpu_AVX COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET
88# define index_cpu_AVX2 COMMON_CPUID_INDEX_7*CPUID_SIZE+CPUID_EBX_OFFSET
89# define index_cpu_ERMS COMMON_CPUID_INDEX_7*CPUID_SIZE+CPUID_EBX_OFFSET
90
91# define index_arch_Fast_Rep_String FEATURE_INDEX_1*FEATURE_SIZE
92# define index_arch_Fast_Copy_Backward FEATURE_INDEX_1*FEATURE_SIZE
93# define index_arch_Slow_BSF FEATURE_INDEX_1*FEATURE_SIZE
94# define index_arch_Fast_Unaligned_Load FEATURE_INDEX_1*FEATURE_SIZE
95# define index_arch_Prefer_PMINUB_for_stringop FEATURE_INDEX_1*FEATURE_SIZE
96# define index_arch_AVX_Usable FEATURE_INDEX_1*FEATURE_SIZE
97# define index_arch_FMA_Usable FEATURE_INDEX_1*FEATURE_SIZE
98# define index_arch_FMA4_Usable FEATURE_INDEX_1*FEATURE_SIZE
99# define index_arch_Slow_SSE4_2 FEATURE_INDEX_1*FEATURE_SIZE
100# define index_arch_AVX2_Usable FEATURE_INDEX_1*FEATURE_SIZE
101# define index_arch_AVX_Fast_Unaligned_Load FEATURE_INDEX_1*FEATURE_SIZE
102# define index_arch_AVX512F_Usable FEATURE_INDEX_1*FEATURE_SIZE
103# define index_arch_AVX512DQ_Usable FEATURE_INDEX_1*FEATURE_SIZE
104# define index_arch_I586 FEATURE_INDEX_1*FEATURE_SIZE
105# define index_arch_I686 FEATURE_INDEX_1*FEATURE_SIZE
106# define index_arch_Prefer_MAP_32BIT_EXEC FEATURE_INDEX_1*FEATURE_SIZE
107# define index_arch_Prefer_No_VZEROUPPER FEATURE_INDEX_1*FEATURE_SIZE
108# define index_arch_Fast_Unaligned_Copy FEATURE_INDEX_1*FEATURE_SIZE
109# define index_arch_Prefer_ERMS FEATURE_INDEX_1*FEATURE_SIZE
110
111
112# if defined (_LIBC) && !IS_IN (nonlib)
113# ifdef __x86_64__
114# ifdef SHARED
115# if IS_IN (rtld)
116# define LOAD_RTLD_GLOBAL_RO_RDX
117# define HAS_FEATURE(offset, field, name) \
118 testl $(bit_##field##_##name), \
119 _rtld_local_ro+offset+(index_##field##_##name)(%rip)
120# else
121# define LOAD_RTLD_GLOBAL_RO_RDX \
122 mov _rtld_global_ro@GOTPCREL(%rip), %RDX_LP
123# define HAS_FEATURE(offset, field, name) \
124 testl $(bit_##field##_##name), \
125 RTLD_GLOBAL_RO_DL_X86_CPU_FEATURES_OFFSET+offset+(index_##field##_##name)(%rdx)
126# endif
127# else /* SHARED */
128# define LOAD_RTLD_GLOBAL_RO_RDX
129# define HAS_FEATURE(offset, field, name) \
130 testl $(bit_##field##_##name), \
131 _dl_x86_cpu_features+offset+(index_##field##_##name)(%rip)
132# endif /* !SHARED */
133# else /* __x86_64__ */
134# ifdef SHARED
135# define LOAD_FUNC_GOT_EAX(func) \
136 leal func@GOTOFF(%edx), %eax
137# if IS_IN (rtld)
138# define LOAD_GOT_AND_RTLD_GLOBAL_RO \
139 LOAD_PIC_REG(dx)
140# define HAS_FEATURE(offset, field, name) \
141 testl $(bit_##field##_##name), \
142 offset+(index_##field##_##name)+_rtld_local_ro@GOTOFF(%edx)
143# else
144# define LOAD_GOT_AND_RTLD_GLOBAL_RO \
145 LOAD_PIC_REG(dx); \
146 mov _rtld_global_ro@GOT(%edx), %ecx
147# define HAS_FEATURE(offset, field, name) \
148 testl $(bit_##field##_##name), \
149 RTLD_GLOBAL_RO_DL_X86_CPU_FEATURES_OFFSET+offset+(index_##field##_##name)(%ecx)
150# endif
151# else /* SHARED */
152# define LOAD_FUNC_GOT_EAX(func) \
153 leal func, %eax
154# define LOAD_GOT_AND_RTLD_GLOBAL_RO
155# define HAS_FEATURE(offset, field, name) \
156 testl $(bit_##field##_##name), \
157 _dl_x86_cpu_features+offset+(index_##field##_##name)
158# endif /* !SHARED */
159# endif /* !__x86_64__ */
160# else /* _LIBC && !nonlib */
161# error "Sorry, <cpu-features.h> is unimplemented for assembler"
162# endif /* !_LIBC || nonlib */
163
164/* HAS_* evaluates to true if we may use the feature at runtime. */
165# define HAS_CPU_FEATURE(name) HAS_FEATURE (CPUID_OFFSET, cpu, name)
166# define HAS_ARCH_FEATURE(name) HAS_FEATURE (FEATURE_OFFSET, arch, name)
167
168#else /* __ASSEMBLER__ */
169
170enum
171 {
172 COMMON_CPUID_INDEX_1 = 0,
173 COMMON_CPUID_INDEX_7,
174 COMMON_CPUID_INDEX_80000001, /* for AMD */
175 /* Keep the following line at the end. */
176 COMMON_CPUID_INDEX_MAX
177 };
178
179struct cpu_features
180{
181 enum cpu_features_kind
182 {
183 arch_kind_unknown = 0,
184 arch_kind_intel,
185 arch_kind_amd,
186 arch_kind_other
187 } kind;
188 int max_cpuid;
189 struct cpuid_registers
190 {
191 unsigned int eax;
192 unsigned int ebx;
193 unsigned int ecx;
194 unsigned int edx;
195 } cpuid[COMMON_CPUID_INDEX_MAX];
196 unsigned int family;
197 unsigned int model;
198 unsigned int feature[FEATURE_INDEX_MAX];
199};
200
201/* Used from outside of glibc to get access to the CPU features
202 structure. */
203extern const struct cpu_features *__get_cpu_features (void)
204 __attribute__ ((const));
205
206# if defined (_LIBC) && !IS_IN (nonlib)
207/* Unused for x86. */
208# define INIT_ARCH()
209# define __get_cpu_features() (&GLRO(dl_x86_cpu_features))
210# endif
211
212
213/* Only used directly in cpu-features.c. */
214# define CPU_FEATURES_CPU_P(ptr, name) \
215 ((ptr->cpuid[index_cpu_##name].reg_##name & (bit_cpu_##name)) != 0)
216# define CPU_FEATURES_ARCH_P(ptr, name) \
217 ((ptr->feature[index_arch_##name] & (bit_arch_##name)) != 0)
218
219/* HAS_* evaluates to true if we may use the feature at runtime. */
220# define HAS_CPU_FEATURE(name) \
221 CPU_FEATURES_CPU_P (__get_cpu_features (), name)
222# define HAS_ARCH_FEATURE(name) \
223 CPU_FEATURES_ARCH_P (__get_cpu_features (), name)
224
225# define index_cpu_CX8 COMMON_CPUID_INDEX_1
226# define index_cpu_CMOV COMMON_CPUID_INDEX_1
227# define index_cpu_SSE2 COMMON_CPUID_INDEX_1
228# define index_cpu_SSSE3 COMMON_CPUID_INDEX_1
229# define index_cpu_SSE4_1 COMMON_CPUID_INDEX_1
230# define index_cpu_SSE4_2 COMMON_CPUID_INDEX_1
231# define index_cpu_AVX COMMON_CPUID_INDEX_1
232# define index_cpu_AVX2 COMMON_CPUID_INDEX_7
233# define index_cpu_AVX512F COMMON_CPUID_INDEX_7
234# define index_cpu_AVX512DQ COMMON_CPUID_INDEX_7
235# define index_cpu_ERMS COMMON_CPUID_INDEX_7
236# define index_cpu_RTM COMMON_CPUID_INDEX_7
237# define index_cpu_FMA COMMON_CPUID_INDEX_1
238# define index_cpu_FMA4 COMMON_CPUID_INDEX_80000001
239# define index_cpu_POPCOUNT COMMON_CPUID_INDEX_1
240# define index_cpu_OSXSAVE COMMON_CPUID_INDEX_1
241# define index_cpu_HTT COMMON_CPUID_INDEX_1
242
243# define reg_CX8 edx
244# define reg_CMOV edx
245# define reg_SSE2 edx
246# define reg_SSSE3 ecx
247# define reg_SSE4_1 ecx
248# define reg_SSE4_2 ecx
249# define reg_AVX ecx
250# define reg_AVX2 ebx
251# define reg_AVX512F ebx
252# define reg_AVX512DQ ebx
253# define reg_ERMS ebx
254# define reg_RTM ebx
255# define reg_FMA ecx
256# define reg_FMA4 ecx
257# define reg_POPCOUNT ecx
258# define reg_OSXSAVE ecx
259# define reg_HTT edx
260
261# define index_arch_Fast_Rep_String FEATURE_INDEX_1
262# define index_arch_Fast_Copy_Backward FEATURE_INDEX_1
263# define index_arch_Slow_BSF FEATURE_INDEX_1
264# define index_arch_Fast_Unaligned_Load FEATURE_INDEX_1
265# define index_arch_Prefer_PMINUB_for_stringop FEATURE_INDEX_1
266# define index_arch_AVX_Usable FEATURE_INDEX_1
267# define index_arch_FMA_Usable FEATURE_INDEX_1
268# define index_arch_FMA4_Usable FEATURE_INDEX_1
269# define index_arch_Slow_SSE4_2 FEATURE_INDEX_1
270# define index_arch_AVX2_Usable FEATURE_INDEX_1
271# define index_arch_AVX_Fast_Unaligned_Load FEATURE_INDEX_1
272# define index_arch_AVX512F_Usable FEATURE_INDEX_1
273# define index_arch_AVX512DQ_Usable FEATURE_INDEX_1
274# define index_arch_I586 FEATURE_INDEX_1
275# define index_arch_I686 FEATURE_INDEX_1
276# define index_arch_Prefer_MAP_32BIT_EXEC FEATURE_INDEX_1
277# define index_arch_Prefer_No_VZEROUPPER FEATURE_INDEX_1
278# define index_arch_Fast_Unaligned_Copy FEATURE_INDEX_1
279# define index_arch_Prefer_ERMS FEATURE_INDEX_1
280
281#endif /* !__ASSEMBLER__ */
282
283#ifdef __x86_64__
284# define HAS_CPUID 1
285#elif defined __i586__ || defined __pentium__
286# define HAS_CPUID 1
287# define HAS_I586 1
288# define HAS_I686 HAS_ARCH_FEATURE (I686)
289#elif (defined __i686__ || defined __pentiumpro__ \
290 || defined __pentium4__ || defined __nocona__ \
291 || defined __atom__ || defined __core2__ \
292 || defined __corei7__ || defined __corei7_avx__ \
293 || defined __core_avx2__ || defined __nehalem__ \
294 || defined __sandybridge__ || defined __haswell__ \
295 || defined __knl__ || defined __bonnell__ \
296 || defined __silvermont__ \
297 || defined __k6__ || defined __k8__ \
298 || defined __athlon__ || defined __amdfam10__ \
299 || defined __bdver1__ || defined __bdver2__ \
300 || defined __bdver3__ || defined __bdver4__ \
301 || defined __btver1__ || defined __btver2__)
302# define HAS_CPUID 1
303# define HAS_I586 1
304# define HAS_I686 1
305#else
306# define HAS_CPUID 0
307# define HAS_I586 HAS_ARCH_FEATURE (I586)
308# define HAS_I686 HAS_ARCH_FEATURE (I686)
309#endif
310
311#endif /* cpu_features_h */
312