FD.io VPP  v19.04.4-rc0-5-ge88582fac
Vector Packet Processing
cpu.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14  */
15 
16 #ifndef included_clib_cpu_h
17 #define included_clib_cpu_h
18 
19 #include <sys/syscall.h>
20 #include <vppinfra/format.h>
21 
22 /*
23  * multiarchitecture support. Adding new entry will produce
24  * new graph node function variant optimized for specific cpu
25  * microarchitecture.
26  * Order is important for runtime selection, as 1st match wins...
27  */
28 
29 #if __x86_64__ && CLIB_DEBUG == 0
30 #define foreach_march_variant(macro, x) \
31  macro(avx2, x, "arch=core-avx2")
32 #else
33 #define foreach_march_variant(macro, x)
34 #endif
35 
36 
37 #if __GNUC__ > 4 && !__clang__ && CLIB_DEBUG == 0
38 #define CLIB_CPU_OPTIMIZED __attribute__ ((optimize ("O3")))
39 #else
40 #define CLIB_CPU_OPTIMIZED
41 #endif
42 
43 
44 #define CLIB_MULTIARCH_ARCH_CHECK(arch, fn, tgt) \
45  if (clib_cpu_supports_ ## arch()) \
46  return & fn ## _ ##arch;
47 
48 /* FIXME to be removed */
49 #define CLIB_MULTIARCH_SELECT_FN(fn,...)
50 
51 #ifdef CLIB_MARCH_VARIANT
52 #define __CLIB_MULTIARCH_FN(a,b) a##_##b
53 #define _CLIB_MULTIARCH_FN(a,b) __CLIB_MULTIARCH_FN(a,b)
54 #define CLIB_MULTIARCH_FN(fn) _CLIB_MULTIARCH_FN(fn,CLIB_MARCH_VARIANT)
55 #else
56 #define CLIB_MULTIARCH_FN(fn) fn
57 #endif
58 
59 #define CLIB_MARCH_SFX CLIB_MULTIARCH_FN
60 
61 typedef struct _clib_march_fn_registration
62 {
63  void *function;
64  int priority;
65  struct _clib_march_fn_registration *next;
66  char *name;
68 
71 {
72  void *rv = 0;
73  int last_prio = -1;
74 
75  while (r)
76  {
77  if (last_prio < r->priority)
78  {
79  last_prio = r->priority;
80  rv = r->function;
81  }
82  r = r->next;
83  }
84  return rv;
85 }
86 
87 #define CLIB_MARCH_FN_POINTER(fn) \
88  clib_march_select_fn_ptr (fn##_march_fn_registrations);
89 
90 #define _CLIB_MARCH_FN_REGISTRATION(fn) \
91 static clib_march_fn_registration \
92 CLIB_MARCH_SFX(fn##_march_fn_registration) = \
93 { \
94  .name = CLIB_MARCH_VARIANT_STR \
95 }; \
96 \
97 static void __clib_constructor \
98 fn##_march_register () \
99 { \
100  clib_march_fn_registration *r; \
101  r = & CLIB_MARCH_SFX (fn##_march_fn_registration); \
102  r->priority = CLIB_MARCH_FN_PRIORITY(); \
103  r->next = fn##_march_fn_registrations; \
104  r->function = CLIB_MARCH_SFX (fn); \
105  fn##_march_fn_registrations = r; \
106 }
107 
108 #ifdef CLIB_MARCH_VARIANT
109 #define CLIB_MARCH_FN_REGISTRATION(fn) \
110 extern clib_march_fn_registration *fn##_march_fn_registrations; \
111 _CLIB_MARCH_FN_REGISTRATION(fn)
112 #else
113 #define CLIB_MARCH_FN_REGISTRATION(fn) \
114 clib_march_fn_registration *fn##_march_fn_registrations = 0; \
115 _CLIB_MARCH_FN_REGISTRATION(fn)
116 #endif
117 #define foreach_x86_64_flags \
118 _ (sse3, 1, ecx, 0) \
119 _ (ssse3, 1, ecx, 9) \
120 _ (sse41, 1, ecx, 19) \
121 _ (sse42, 1, ecx, 20) \
122 _ (avx, 1, ecx, 28) \
123 _ (rdrand, 1, ecx, 30) \
124 _ (avx2, 7, ebx, 5) \
125 _ (rtm, 7, ebx, 11) \
126 _ (pqm, 7, ebx, 12) \
127 _ (pqe, 7, ebx, 15) \
128 _ (avx512f, 7, ebx, 16) \
129 _ (rdseed, 7, ebx, 18) \
130 _ (x86_aes, 1, ecx, 25) \
131 _ (sha, 7, ebx, 29) \
132 _ (vaes, 7, ecx, 9) \
133 _ (vpclmulqdq, 7, ecx, 10) \
134 _ (invariant_tsc, 0x80000007, edx, 8)
135 
136 
137 #define foreach_aarch64_flags \
138 _ (fp, 0) \
139 _ (asimd, 1) \
140 _ (evtstrm, 2) \
141 _ (aarch64_aes, 3) \
142 _ (pmull, 4) \
143 _ (sha1, 5) \
144 _ (sha2, 6) \
145 _ (crc32, 7) \
146 _ (atomics, 8) \
147 _ (fphp, 9) \
148 _ (asimdhp, 10) \
149 _ (cpuid, 11) \
150 _ (asimdrdm, 12) \
151 _ (jscvt, 13) \
152 _ (fcma, 14) \
153 _ (lrcpc, 15) \
154 _ (dcpop, 16) \
155 _ (sha3, 17) \
156 _ (sm3, 18) \
157 _ (sm4, 19) \
158 _ (asimddp, 20) \
159 _ (sha512, 21) \
160 _ (sve, 22)
161 
162 static inline u32
164 {
165  unsigned cpu, node;
166  syscall (__NR_getcpu, &cpu, &node, 0);
167  return cpu;
168 }
169 
170 static inline u32
172 {
173  unsigned cpu, node;
174  syscall (__NR_getcpu, &cpu, &node, 0);
175  return node;
176 }
177 
178 #if defined(__x86_64__)
179 #include "cpuid.h"
180 
181 static inline int
182 clib_get_cpuid (const u32 lev, u32 * eax, u32 * ebx, u32 * ecx, u32 * edx)
183 {
184  if ((u32) __get_cpuid_max (0x80000000 & lev, 0) < lev)
185  return 0;
186  if (lev == 7)
187  __cpuid_count (lev, 0, *eax, *ebx, *ecx, *edx);
188  else
189  __cpuid (lev, *eax, *ebx, *ecx, *edx);
190  return 1;
191 }
192 
193 
194 #define _(flag, func, reg, bit) \
195 static inline int \
196 clib_cpu_supports_ ## flag() \
197 { \
198  u32 __attribute__((unused)) eax, ebx = 0, ecx = 0, edx = 0; \
199  clib_get_cpuid (func, &eax, &ebx, &ecx, &edx); \
200  \
201  return ((reg & (1 << bit)) != 0); \
202 }
204 #undef _
205 #else /* __x86_64__ */
206 
207 #define _(flag, func, reg, bit) \
208 static inline int clib_cpu_supports_ ## flag() { return 0; }
210 #undef _
211 #endif /* __x86_64__ */
212 #if defined(__aarch64__)
213 #include <sys/auxv.h>
214 #define _(flag, bit) \
215 static inline int \
216 clib_cpu_supports_ ## flag() \
217 { \
218  unsigned long hwcap = getauxval(AT_HWCAP); \
219  return (hwcap & (1 << bit)); \
220 }
222 #undef _
223 #else /* ! __x86_64__ && !__aarch64__ */
224 #define _(flag, bit) \
225 static inline int clib_cpu_supports_ ## flag() { return 0; }
227 #undef _
228 #endif /* __x86_64__, __aarch64__ */
229 /*
230  * aes is the only feature with the same name in both flag lists
231  * handle this by prefixing it with the arch name, and handling it
232  * with the custom function below
233  */
234  static inline int
236 {
237 #if defined (__aarch64__)
238  return clib_cpu_supports_x86_aes ();
239 #elif defined (__aarch64__)
240  return clib_cpu_supports_aarch64_aes ();
241 #else
242  return 0;
243 #endif
244 }
245 
246 static inline int
248 {
249  if (clib_cpu_supports_avx512f ())
250  return 20;
251  return -1;
252 }
253 
254 static inline int
256 {
257  if (clib_cpu_supports_avx2 ())
258  return 50;
259  return -1;
260 }
261 
262 static inline u32
264 {
265  char buf[128];
266  static u32 implementer = -1;
267 
268  if (-1 != implementer)
269  return implementer;
270 
271  FILE *fp = fopen ("/proc/cpuinfo", "r");
272  if (!fp)
273  return implementer;
274 
275  while (!feof (fp))
276  {
277  if (!fgets (buf, sizeof (buf), fp))
278  break;
279  buf[127] = '\0';
280  if (strstr (buf, "CPU implementer"))
281  implementer = (u32) strtol (memchr (buf, ':', 128) + 2, NULL, 0);
282  if (-1 != implementer)
283  break;
284  }
285  fclose (fp);
286 
287  return implementer;
288 }
289 
290 static inline u32
292 {
293  char buf[128];
294  static u32 part = -1;
295 
296  if (-1 != part)
297  return part;
298 
299  FILE *fp = fopen ("/proc/cpuinfo", "r");
300  if (!fp)
301  return part;
302 
303  while (!feof (fp))
304  {
305  if (!fgets (buf, sizeof (buf), fp))
306  break;
307  buf[127] = '\0';
308  if (strstr (buf, "CPU part"))
309  part = (u32) strtol (memchr (buf, ':', 128) + 2, NULL, 0);
310  if (-1 != part)
311  break;
312  }
313  fclose (fp);
314 
315  return part;
316 }
317 
318 #define AARCH64_CPU_IMPLEMENTER_THUNERDERX2 0x43
319 #define AARCH64_CPU_PART_THUNERDERX2 0x0af
320 #define AARCH64_CPU_IMPLEMENTER_QDF24XX 0x51
321 #define AARCH64_CPU_PART_QDF24XX 0xc00
322 #define AARCH64_CPU_IMPLEMENTER_CORTEXA72 0x41
323 #define AARCH64_CPU_PART_CORTEXA72 0xd08
324 
325 static inline int
327 {
330  return 20;
331  return -1;
332 }
333 
334 static inline int
336 {
339  return 20;
340  return -1;
341 }
342 
343 static inline int
345 {
348  return 10;
349  return -1;
350 }
351 
352 #ifdef CLIB_MARCH_VARIANT
353 #define CLIB_MARCH_FN_PRIORITY() CLIB_MARCH_SFX(clib_cpu_march_priority)()
354 #else
355 #define CLIB_MARCH_FN_PRIORITY() 0
356 #endif
357 #endif /* included_clib_cpu_h */
358 
359 #define CLIB_MARCH_FN_CONSTRUCTOR(fn) \
360 static void __clib_constructor \
361 CLIB_MARCH_SFX(fn ## _march_constructor) (void) \
362 { \
363  if (CLIB_MARCH_FN_PRIORITY() > fn ## _selected_priority) \
364  { \
365  fn ## _selected = & CLIB_MARCH_SFX (fn ## _ma); \
366  fn ## _selected_priority = CLIB_MARCH_FN_PRIORITY(); \
367  } \
368 } \
369 
370 #ifndef CLIB_MARCH_VARIANT
371 #define CLIB_MARCH_FN(fn, rtype, _args...) \
372  static rtype CLIB_CPU_OPTIMIZED CLIB_MARCH_SFX (fn ## _ma)(_args); \
373  rtype (*fn ## _selected) (_args) = & CLIB_MARCH_SFX (fn ## _ma); \
374  int fn ## _selected_priority = 0; \
375  static inline rtype CLIB_CPU_OPTIMIZED \
376  CLIB_MARCH_SFX (fn ## _ma)(_args)
377 #else
378 #define CLIB_MARCH_FN(fn, rtype, _args...) \
379  static rtype CLIB_CPU_OPTIMIZED CLIB_MARCH_SFX (fn ## _ma)(_args); \
380  extern rtype (*fn ## _selected) (_args); \
381  extern int fn ## _selected_priority; \
382  CLIB_MARCH_FN_CONSTRUCTOR (fn) \
383  static rtype CLIB_CPU_OPTIMIZED CLIB_MARCH_SFX (fn ## _ma)(_args)
384 #endif
385 
386 #define CLIB_MARCH_FN_SELECT(fn) (* fn ## _selected)
387 
391 
392 /*
393  * fd.io coding-style-patch-verification: ON
394  *
395  * Local Variables:
396  * eval: (c-set-style "gnu")
397  * End:
398  */
static int clib_cpu_march_priority_cortexa72()
Definition: cpu.h:344
#define AARCH64_CPU_IMPLEMENTER_THUNERDERX2
Definition: cpu.h:318
static u32 clib_get_current_numa_node()
Definition: cpu.h:171
#define NULL
Definition: clib.h:58
static u32 clib_get_current_cpu_id()
Definition: cpu.h:163
format_function_t format_cpu_flags
Definition: cpu.h:390
#define foreach_aarch64_flags
Definition: cpu.h:137
u8 *() format_function_t(u8 *s, va_list *args)
Definition: format.h:48
#define static_always_inline
Definition: clib.h:99
unsigned int u32
Definition: types.h:88
static u32 clib_cpu_part()
Definition: cpu.h:291
u8 name[64]
Definition: memclnt.api:152
static int clib_cpu_march_priority_avx512()
Definition: cpu.h:247
#define AARCH64_CPU_PART_QDF24XX
Definition: cpu.h:321
i32 priority
Definition: ipsec.api:91
static int clib_cpu_march_priority_thunderx2t99()
Definition: cpu.h:326
static int clib_get_cpuid(const u32 lev, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
Definition: cpu.h:182
static u32 clib_cpu_implementer()
Definition: cpu.h:263
static int clib_cpu_march_priority_avx2()
Definition: cpu.h:255
#define foreach_x86_64_flags
Definition: cpu.h:117
struct _clib_march_fn_registration clib_march_fn_registration
#define AARCH64_CPU_PART_THUNERDERX2
Definition: cpu.h:319
static_always_inline void * clib_march_select_fn_ptr(clib_march_fn_registration *r)
Definition: cpu.h:70
static foreach_aarch64_flags int clib_cpu_supports_aes()
Definition: cpu.h:235
#define AARCH64_CPU_IMPLEMENTER_CORTEXA72
Definition: cpu.h:322
static int clib_cpu_march_priority_qdf24xx()
Definition: cpu.h:335
format_function_t format_cpu_uarch
Definition: cpu.h:388
format_function_t format_cpu_model_name
Definition: cpu.h:389
#define AARCH64_CPU_PART_CORTEXA72
Definition: cpu.h:323
#define AARCH64_CPU_IMPLEMENTER_QDF24XX
Definition: cpu.h:320