sysdeps/i386/i686/hp-timing.h

   1 /* High precision, low overhead timing functions.  i686 version.
   2    Copyright (C) 1998-2014 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998.
   5
   6    The GNU C Library is free software; you can redistribute it and/or
   7    modify it under the terms of the GNU Lesser General Public
   8    License as published by the Free Software Foundation; either
   9    version 2.1 of the License, or (at your option) any later version.
  10
  11    The GNU C Library is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14    Lesser General Public License for more details.
  15
  16    You should have received a copy of the GNU Lesser General Public
  17    License along with the GNU C Library; if not, see
  18    <http://www.gnu.org/licenses/>.  */
  19
  20 #ifndef _HP_TIMING_H
  21 #define _HP_TIMING_H    1
  22
  23 #include <string.h>
  24 #include <sys/param.h>
  25 #include <_itoa.h>
  26
  27 /* The macros defined here use the timestamp counter in i586 and up versions
  28    of the x86 processors.  They provide a very accurate way to measure the
  29    time with very little overhead.  The time values themself have no real
  30    meaning, only differences are interesting.
  31
  32    This version is for the i686 processors.  The difference to the i586
  33    version is that the timerstamp register is unconditionally used.  This is
  34    not the case for the i586 version where we have to perform runtime test
  35    whether the processor really has this capability.  We have to make this
  36    distinction since the sysdeps/i386/i586 code is supposed to work on all
  37    platforms while the i686 already contains i686-specific code.
  38
  39    The list of macros we need includes the following:
  40
  41    - HP_TIMING_AVAIL: test for availability.
  42
  43    - HP_TIMING_INLINE: this macro is non-zero if the functionality is not
  44      implemented using function calls but instead uses some inlined code
  45      which might simply consist of a few assembler instructions.  We have to
  46      know this since we might want to use the macros here in places where we
  47      cannot make function calls.
  48
  49    - hp_timing_t: This is the type for variables used to store the time
  50      values.
  51
  52    - HP_TIMING_ZERO: clear `hp_timing_t' object.
  53
  54    - HP_TIMING_NOW: place timestamp for current time in variable given as
  55      parameter.
  56
  57    - HP_TIMING_DIFF_INIT: do whatever is necessary to be able to use the
  58      HP_TIMING_DIFF macro.
  59
  60    - HP_TIMING_DIFF: compute difference between two times and store it
  61      in a third.  Source and destination might overlap.
  62
  63    - HP_TIMING_ACCUM: add time difference to another variable.  This might
  64      be a bit more complicated to implement for some platforms as the
  65      operation should be thread-safe and 64bit arithmetic on 32bit platforms
  66      is not.
  67
  68    - HP_TIMING_ACCUM_NT: this is the variant for situations where we know
  69      there are no threads involved.
  70
  71    - HP_TIMING_PRINT: write decimal representation of the timing value into
  72      the given string.  This operation need not be inline even though
  73      HP_TIMING_INLINE is specified.
  74
  75 */
  76
  77 /* We always assume having the timestamp register.  */
  78 #define HP_TIMING_AVAIL         (1)
  79
  80 /* We indeed have inlined functions.  */
  81 #define HP_TIMING_INLINE        (1)
  82
  83 /* We use 64bit values for the times.  */
  84 typedef unsigned long long int hp_timing_t;
  85
  86 /* Set timestamp value to zero.  */
  87 #define HP_TIMING_ZERO(Var)     (Var) = (0)
  88
  89 /* That's quite simple.  Use the `rdtsc' instruction.  Note that the value
  90    might not be 100% accurate since there might be some more instructions
  91    running in this moment.  This could be changed by using a barrier like
  92    'cpuid' right before the `rdtsc' instruciton.  But we are not interested
  93    in accurate clock cycles here so we don't do this.  */
  94 #define HP_TIMING_NOW(Var)      __asm__ __volatile__ ("rdtsc" : "=A" (Var))
  95
  96 /* Use two 'rdtsc' instructions in a row to find out how long it takes.  */
  97 #define HP_TIMING_DIFF_INIT() \
  98   do {                                                                        \
  99     if (GLRO(dl_hp_timing_overhead) == 0)                                     \
 100       {                                                                       \
 101         int __cnt = 5;                                                        \
 102         GLRO(dl_hp_timing_overhead) = ~0ull;                                  \
 103         do                                                                    \
 104           {                                                                   \
 105             hp_timing_t __t1, __t2;                                           \
 106             HP_TIMING_NOW (__t1);                                             \
 107             HP_TIMING_NOW (__t2);                                             \
 108             if (__t2 - __t1 < GLRO(dl_hp_timing_overhead))                    \
 109               GLRO(dl_hp_timing_overhead) = __t2 - __t1;                      \
 110           }                                                                   \
 111         while (--__cnt > 0);                                                  \
 112       }                                                                       \
 113   } while (0)
 114
 115 /* It's simple arithmetic for us.  */
 116 #define HP_TIMING_DIFF(Diff, Start, End)        (Diff) = ((End) - (Start))
 117
 118 /* We have to jump through hoops to get this correctly implemented.  */
 119 #define HP_TIMING_ACCUM(Sum, Diff) \
 120   do {                                                                        \
 121     int __not_done;                                                           \
 122     hp_timing_t __oldval = (Sum);                                             \
 123     hp_timing_t __diff = (Diff) - GLRO(dl_hp_timing_overhead);                \
 124     do                                                                        \
 125       {                                                                       \
 126         hp_timing_t __newval = __oldval + __diff;                             \
 127         int __temp0, __temp1;                                                 \
 128         __asm__ __volatile__ ("xchgl %0, %%ebx\n\t"                           \
 129                               "lock; cmpxchg8b %1\n\t"                        \
 130                               "sete %%bl\n\t"                                 \
 131                               "xchgl %0, %%ebx"                               \
 132                               : "=SD" (__not_done), "=m" (Sum),               \
 133                                 "=A" (__oldval), "=c" (__temp0)               \
 134                               : "m" (Sum), "2" (__oldval),                    \
 135                                 "3" ((unsigned int) (__newval >> 32)),        \
 136                                 "0" ((unsigned int) __newval));               \
 137       }                                                                       \
 138     while ((unsigned char) __not_done);                                       \
 139   } while (0)
 140
 141 /* No threads, no extra work.  */
 142 #define HP_TIMING_ACCUM_NT(Sum, Diff)   (Sum) += (Diff)
 143
 144 /* Print the time value.  */
 145 #define HP_TIMING_PRINT(Buf, Len, Val) \
 146   do {                                                                        \
 147     char __buf[20];                                                           \
 148     char *__cp = _itoa (Val, __buf + sizeof (__buf), 10, 0);                  \
 149     size_t __len = (Len);                                                     \
 150     char *__dest = (Buf);                                                     \
 151     while (__len-- > 0 && __cp < __buf + sizeof (__buf))                      \
 152       *__dest++ = *__cp++;                                                    \
 153     memcpy (__dest, " clock cycles", MIN (__len, sizeof (" clock cycles")));  \
 154   } while (0)
 155
 156 #endif  /* hp-timing.h */