1/*
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining
6 * a copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sublicense, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be
14 * included in all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
17 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
18 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
19 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
20 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
21 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
22 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 *
24 */
25
26
27/* machine-dependent cycle counters code. Needs to be inlined. */
28
29/***************************************************************************/
30/* To use the cycle counters in your code, simply #include "cycle.h" (this
31 file), and then use the functions/macros:
32
33 ticks getticks(void);
34
35 ticks is an opaque typedef defined below, representing the current time.
36 You extract the elapsed time between two calls to gettick() via:
37
38 double elapsed(ticks t1, ticks t0);
39
40 which returns a double-precision variable in arbitrary units. You
41 are not expected to convert this into human units like seconds; it
42 is intended only for *comparisons* of time intervals.
43
44 (In order to use some of the OS-dependent timer routines like
45 Solaris' gethrtime, you need to paste the autoconf snippet below
46 into your configure.ac file and #include "config.h" before cycle.h,
47 or define the relevant macros manually if you are not using autoconf.)
48*/
49
50/***************************************************************************/
51/* This file uses macros like HAVE_GETHRTIME that are assumed to be
52 defined according to whether the corresponding function/type/header
53 is available on your system. The necessary macros are most
54 conveniently defined if you are using GNU autoconf, via the tests:
55
56 dnl ---------------------------------------------------------------------
57
58 AC_C_INLINE
59 AC_HEADER_TIME
60 AC_CHECK_HEADERS([sys/time.h c_asm.h intrinsics.h mach/mach_time.h])
61
62 AC_CHECK_TYPE([hrtime_t],[AC_DEFINE(HAVE_HRTIME_T, 1, [Define to 1 if hrtime_t is defined in <sys/time.h>])],,[#if HAVE_SYS_TIME_H
63#include <sys/time.h>
64#endif])
65
66 AC_CHECK_FUNCS([gethrtime read_real_time time_base_to_time clock_gettime mach_absolute_time])
67
68 dnl Cray UNICOS _rtc() (real-time clock) intrinsic
69 AC_MSG_CHECKING([for _rtc intrinsic])
70 rtc_ok=yes
71 AC_TRY_LINK([#ifdef HAVE_INTRINSICS_H
72#include <intrinsics.h>
73#endif], [_rtc()], [AC_DEFINE(HAVE__RTC,1,[Define if you have the UNICOS _rtc() intrinsic.])], [rtc_ok=no])
74 AC_MSG_RESULT($rtc_ok)
75
76 dnl ---------------------------------------------------------------------
77*/
78
79#ifndef QBENCHLIB_CYCLE_H
80#define QBENCHLIB_CYCLE_H
81#define ticks CycleCounterTicks
82
83/***************************************************************************/
84
85#if TIME_WITH_SYS_TIME
86# include <sys/time.h>
87# include <time.h>
88#else
89# if HAVE_SYS_TIME_H
90# include <sys/time.h>
91# else
92# include <time.h>
93# endif
94#endif
95
96#define INLINE_ELAPSED(INL) static INL double elapsed(ticks t1, ticks t0) \
97{ \
98 return (double)t1 - (double)t0; \
99}
100
101/*----------------------------------------------------------------*/
102/* Solaris */
103#if defined(HAVE_GETHRTIME) && defined(HAVE_HRTIME_T) && !defined(HAVE_TICK_COUNTER)
104typedef hrtime_t ticks;
105
106#define getticks gethrtime
107
108INLINE_ELAPSED(inline)
109
110#define HAVE_TICK_COUNTER
111#endif
112
113/*----------------------------------------------------------------*/
114/* AIX v. 4+ routines to read the real-time clock or time-base register */
115#if defined(HAVE_READ_REAL_TIME) && defined(HAVE_TIME_BASE_TO_TIME) && !defined(HAVE_TICK_COUNTER)
116typedef timebasestruct_t ticks;
117
118static __inline ticks getticks(void)
119{
120 ticks t;
121 read_real_time(&t, TIMEBASE_SZ);
122 return t;
123}
124
125static __inline double elapsed(ticks t1, ticks t0) /* time in nanoseconds */
126{
127 time_base_to_time(&t1, TIMEBASE_SZ);
128 time_base_to_time(&t0, TIMEBASE_SZ);
129 return (((double)t1.tb_high - (double)t0.tb_high) * 1.0e9 +
130 ((double)t1.tb_low - (double)t0.tb_low));
131}
132
133#define HAVE_TICK_COUNTER
134#endif
135
136/*----------------------------------------------------------------*/
137/*
138 * PowerPC ``cycle'' counter using the time base register.
139 */
140#if ((((defined(__GNUC__) && (defined(__powerpc__) || defined(__ppc__))) || (defined(__MWERKS__) && defined(macintosh)))) || (defined(__IBM_GCC_ASM) && (defined(__powerpc__) || defined(__ppc__)))) && !defined(HAVE_TICK_COUNTER)
141typedef unsigned long long ticks;
142
143static __inline__ ticks getticks(void)
144{
145 unsigned int tbl, tbu0, tbu1;
146
147 do {
148 __asm__ __volatile__ ("mftbu %0" : "=r"(tbu0));
149 __asm__ __volatile__ ("mftb %0" : "=r"(tbl));
150 __asm__ __volatile__ ("mftbu %0" : "=r"(tbu1));
151 } while (tbu0 != tbu1);
152
153 return (((unsigned long long)tbu0) << 32) | tbl;
154}
155
156INLINE_ELAPSED(__inline__)
157
158#define HAVE_TICK_COUNTER
159#endif
160
161/* MacOS/Mach (Darwin) time-base register interface (unlike UpTime,
162 from Carbon, requires no additional libraries to be linked). */
163#if defined(HAVE_MACH_ABSOLUTE_TIME) && defined(HAVE_MACH_MACH_TIME_H) && !defined(HAVE_TICK_COUNTER)
164#include <mach/mach_time.h>
165typedef uint64_t ticks;
166#define getticks mach_absolute_time
167INLINE_ELAPSED(__inline__)
168#define HAVE_TICK_COUNTER
169#endif
170
171/*----------------------------------------------------------------*/
172/*
173 * Pentium cycle counter
174 */
175#if (defined(__GNUC__) || defined(__ICC)) && defined(__i386__) && !defined(HAVE_TICK_COUNTER)
176typedef unsigned long long ticks;
177
178static __inline__ ticks getticks(void)
179{
180 ticks ret;
181
182 __asm__ __volatile__("rdtsc": "=A" (ret));
183 /* no input, nothing else clobbered */
184 return ret;
185}
186
187INLINE_ELAPSED(__inline__)
188
189#define HAVE_TICK_COUNTER
190#define TIME_MIN 5000.0 /* unreliable pentium IV cycle counter */
191#endif
192
193/* Visual C++ -- thanks to Morten Nissov for his help with this */
194#ifdef _MSC_VER
195#if _MSC_VER >= 1200 && _M_IX86 >= 500 && !defined(HAVE_TICK_COUNTER)
196#include <windows.h>
197typedef LARGE_INTEGER ticks;
198#define RDTSC __asm __emit 0fh __asm __emit 031h /* hack for VC++ 5.0 */
199
200static __inline ticks getticks(void)
201{
202 ticks retval;
203
204 __asm {
205 RDTSC
206 mov retval.HighPart, edx
207 mov retval.LowPart, eax
208 }
209 return retval;
210}
211
212static __inline double elapsed(ticks t1, ticks t0)
213{
214 return (double)t1.QuadPart - (double)t0.QuadPart;
215}
216
217#define HAVE_TICK_COUNTER
218#define TIME_MIN 5000.0 /* unreliable pentium IV cycle counter */
219#endif
220#endif // _MSC_VER
221
222/*----------------------------------------------------------------*/
223/*
224 * X86-64 cycle counter
225 */
226#if (defined(__GNUC__) || defined(__ICC) || defined(__SUNPRO_C)) && defined(__x86_64__) && !defined(HAVE_TICK_COUNTER)
227typedef unsigned long long ticks;
228
229static __inline__ ticks getticks(void)
230{
231 unsigned a, d;
232 __asm__ __volatile__ ("rdtsc" : "=a" (a), "=d" (d));
233 return ((ticks)a) | (((ticks)d) << 32);
234}
235
236INLINE_ELAPSED(__inline__)
237
238#define HAVE_TICK_COUNTER
239#define TIME_MIN 5000.0
240#endif
241
242/* PGI compiler, courtesy Cristiano Calonaci, Andrea Tarsi, & Roberto Gori.
243 NOTE: this code will fail to link unless you use the -Masmkeyword compiler
244 option (grrr). */
245#if defined(__PGI) && defined(__x86_64__) && !defined(HAVE_TICK_COUNTER)
246typedef unsigned long long ticks;
247static ticks getticks(void)
248{
249 asm(" rdtsc; shl $0x20,%rdx; mov %eax,%eax; or %rdx,%rax; ");
250}
251INLINE_ELAPSED(__inline__)
252#define HAVE_TICK_COUNTER
253#define TIME_MIN 5000.0
254#endif
255
256/* Visual C++, courtesy of Dirk Michaelis */
257#ifdef _MSC_VER
258#if _MSC_VER >= 1400 && (defined(_M_AMD64) || defined(_M_X64)) && !defined(HAVE_TICK_COUNTER)
259
260#include <intrin.h>
261#pragma intrinsic(__rdtsc)
262typedef unsigned __int64 ticks;
263#define getticks __rdtsc
264INLINE_ELAPSED(__inline)
265
266#define HAVE_TICK_COUNTER
267#define TIME_MIN 5000.0
268#endif
269#endif // _MSC_VER
270
271/*----------------------------------------------------------------*/
272/*
273 * IA64 cycle counter
274 */
275
276/* intel's icc/ecc compiler */
277#if (defined(__EDG_VERSION) || defined(__ECC)) && defined(__ia64__) && !defined(HAVE_TICK_COUNTER)
278typedef unsigned long ticks;
279#include <ia64intrin.h>
280
281static __inline__ ticks getticks(void)
282{
283 return __getReg(_IA64_REG_AR_ITC);
284}
285
286INLINE_ELAPSED(__inline__)
287
288#define HAVE_TICK_COUNTER
289#endif
290
291/* gcc */
292#if defined(__GNUC__) && defined(__ia64__) && !defined(HAVE_TICK_COUNTER)
293typedef unsigned long ticks;
294
295static __inline__ ticks getticks(void)
296{
297 ticks ret;
298
299 __asm__ __volatile__ ("mov %0=ar.itc" : "=r"(ret));
300 return ret;
301}
302
303INLINE_ELAPSED(__inline__)
304
305#define HAVE_TICK_COUNTER
306#endif
307
308/* HP/UX IA64 compiler, courtesy Teresa L. Johnson: */
309#if defined(__hpux) && defined(__ia64) && !defined(HAVE_TICK_COUNTER)
310#include <machine/sys/inline.h>
311typedef unsigned long ticks;
312
313static inline ticks getticks(void)
314{
315 ticks ret;
316
317 ret = _Asm_mov_from_ar (_AREG_ITC);
318 return ret;
319}
320
321INLINE_ELAPSED(inline)
322
323#define HAVE_TICK_COUNTER
324#endif
325
326/* Microsoft Visual C++ */
327#if defined(_MSC_VER) && defined(_M_IA64) && !defined(HAVE_TICK_COUNTER)
328typedef unsigned __int64 ticks;
329
330# ifdef __cplusplus
331extern "C"
332# endif
333ticks __getReg(int whichReg);
334#pragma intrinsic(__getReg)
335
336static __inline ticks getticks(void)
337{
338 volatile ticks temp;
339 temp = __getReg(3116);
340 return temp;
341}
342
343INLINE_ELAPSED(inline)
344
345#define HAVE_TICK_COUNTER
346#endif
347
348/*----------------------------------------------------------------*/
349/*
350 * PA-RISC cycle counter
351 */
352#if (defined(__hppa__) || defined(__hppa)) && !defined(HAVE_TICK_COUNTER)
353typedef unsigned long ticks;
354
355# ifdef __GNUC__
356static __inline__ ticks getticks(void)
357{
358 ticks ret;
359
360 __asm__ __volatile__("mfctl 16, %0": "=r" (ret));
361 /* no input, nothing else clobbered */
362 return ret;
363}
364
365INLINE_ELAPSED(inline)
366
367#define HAVE_TICK_COUNTER
368# elif 0 // Doesn't compile
369# include <machine/inline.h>
370static inline unsigned long getticks(void)
371{
372 register ticks ret;
373 _MFCTL(16, ret);
374 return ret;
375}
376# endif
377#endif
378
379/*----------------------------------------------------------------*/
380/* S390, courtesy of James Treacy */
381#if defined(__GNUC__) && defined(__s390__) && !defined(HAVE_TICK_COUNTER)
382typedef unsigned long long ticks;
383
384static __inline__ ticks getticks(void)
385{
386 ticks cycles;
387 __asm__("stck 0(%0)" : : "a" (&(cycles)) : "memory", "cc");
388 return cycles;
389}
390
391INLINE_ELAPSED(__inline__)
392
393#define HAVE_TICK_COUNTER
394#endif
395/*----------------------------------------------------------------*/
396#if defined(__GNUC__) && defined(__alpha__) && !defined(HAVE_TICK_COUNTER)
397/*
398 * The 32-bit cycle counter on alpha overflows pretty quickly,
399 * unfortunately. A 1GHz machine overflows in 4 seconds.
400 */
401typedef unsigned int ticks;
402
403static __inline__ ticks getticks(void)
404{
405 unsigned long cc;
406 __asm__ __volatile__ ("rpcc %0" : "=r"(cc));
407 return (cc & 0xFFFFFFFF);
408}
409
410INLINE_ELAPSED(__inline__)
411
412#define HAVE_TICK_COUNTER
413#endif
414
415/*----------------------------------------------------------------*/
416#if defined(__GNUC__) && defined(__sparc_v9__) && !defined(HAVE_TICK_COUNTER)
417typedef unsigned long ticks;
418
419static __inline__ ticks getticks(void)
420{
421 ticks ret;
422 __asm__ __volatile__("rd %%tick, %0" : "=r" (ret));
423 return ret;
424}
425
426INLINE_ELAPSED(__inline__)
427
428#define HAVE_TICK_COUNTER
429#endif
430
431/*----------------------------------------------------------------*/
432#if (defined(__DECC) || defined(__DECCXX)) && defined(__alpha) && defined(HAVE_C_ASM_H) && !defined(HAVE_TICK_COUNTER)
433# include <c_asm.h>
434typedef unsigned int ticks;
435
436static __inline ticks getticks(void)
437{
438 unsigned long cc;
439 cc = asm("rpcc %v0");
440 return (cc & 0xFFFFFFFF);
441}
442
443INLINE_ELAPSED(__inline)
444
445#define HAVE_TICK_COUNTER
446#endif
447/*----------------------------------------------------------------*/
448/* SGI/Irix */
449#if defined(HAVE_CLOCK_GETTIME) && defined(CLOCK_SGI_CYCLE) && !defined(HAVE_TICK_COUNTER) && !defined(__ANDROID__)
450typedef struct timespec ticks;
451
452static inline ticks getticks(void)
453{
454 struct timespec t;
455 clock_gettime(CLOCK_SGI_CYCLE, &t);
456 return t;
457}
458
459static inline double elapsed(ticks t1, ticks t0)
460{
461 return ((double)t1.tv_sec - (double)t0.tv_sec) * 1.0E9 +
462 ((double)t1.tv_nsec - (double)t0.tv_nsec);
463}
464#define HAVE_TICK_COUNTER
465#endif
466
467/*----------------------------------------------------------------*/
468/* Cray UNICOS _rtc() intrinsic function */
469#if defined(HAVE__RTC) && !defined(HAVE_TICK_COUNTER)
470#ifdef HAVE_INTRINSICS_H
471# include <intrinsics.h>
472#endif
473
474typedef long long ticks;
475
476#define getticks _rtc
477
478INLINE_ELAPSED(inline)
479
480#define HAVE_TICK_COUNTER
481#endif
482
483/*----------------------------------------------------------------*/
484/* MIPS ZBus */
485#if HAVE_MIPS_ZBUS_TIMER
486#if defined(__mips__) && !defined(HAVE_TICK_COUNTER)
487#include <sys/mman.h>
488#include <unistd.h>
489#include <fcntl.h>
490
491typedef uint64_t ticks;
492
493static inline ticks getticks(void)
494{
495 static uint64_t* addr = 0;
496
497 if (addr == 0)
498 {
499 uint32_t rq_addr = 0x10030000;
500 int fd;
501 int pgsize;
502
503 pgsize = getpagesize();
504 fd = open ("/dev/mem", O_RDONLY | O_SYNC, 0);
505 if (fd < 0) {
506 perror("open");
507 return NULL;
508 }
509 addr = mmap(0, pgsize, PROT_READ, MAP_SHARED, fd, rq_addr);
510 close(fd);
511 if (addr == (uint64_t *)-1) {
512 perror("mmap");
513 return NULL;
514 }
515 }
516
517 return *addr;
518}
519
520INLINE_ELAPSED(inline)
521
522#define HAVE_TICK_COUNTER
523#endif
524#endif /* HAVE_MIPS_ZBUS_TIMER */
525
526#if defined(HAVE_ARMV7A_CNTVCT)
527typedef uint64_t ticks;
528static inline ticks getticks(void)
529{
530 uint32_t Rt, Rt2 = 0;
531 asm volatile("mrrc p15, 1, %0, %1, c14" : "=r"(Rt), "=r"(Rt2));
532 return ((uint64_t)Rt) | (((uint64_t)Rt2) << 32);
533}
534INLINE_ELAPSED(inline)
535#define HAVE_TICK_COUNTER
536#endif
537
538#if defined(HAVE_ARMV7A_PMCCNTR)
539typedef uint64_t ticks;
540static inline ticks getticks(void)
541{
542 uint32_t r;
543 asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r"(r) );
544 return r;
545}
546INLINE_ELAPSED(inline)
547#define HAVE_TICK_COUNTER
548#endif
549
550#if defined(__aarch64__) && defined(HAVE_ARMV8_CNTVCT_EL0) && !defined(HAVE_ARMV8_PMCCNTR_EL0)
551typedef uint64_t ticks;
552static inline ticks getticks(void)
553{
554 uint64_t Rt;
555 asm volatile("mrs %0, CNTVCT_EL0" : "=r" (Rt));
556 return Rt;
557}
558INLINE_ELAPSED(inline)
559#define HAVE_TICK_COUNTER
560#endif
561
562#if defined(__aarch64__) && defined(HAVE_ARMV8_PMCCNTR_EL0)
563typedef uint64_t ticks;
564static inline ticks getticks(void)
565{
566 uint64_t cc = 0;
567 asm volatile("mrs %0, PMCCNTR_EL0" : "=r"(cc));
568 return cc;
569}
570INLINE_ELAPSED(inline)
571#define HAVE_TICK_COUNTER
572#endif
573
574#undef ticks
575#endif // QBENCHLIB_CYCLE_H
576

source code of qtbase/src/testlib/3rdparty/cycle/cycle_p.h