sysdeps/i386/i686/hp-timing.h

   1 /* High precision, low overhead timing functions.  i686 version.
   2    Copyright (C) 1998, 2002, 2003, 2004, 2005 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, write to the Free
  18    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  19    02111-1307 USA.  */
  20
  21 #ifndef _HP_TIMING_H
  22 #define _HP_TIMING_H    1
  23
  24 #include <string.h>
  25 #include <sys/param.h>
  26 #include <stdio-common/_itoa.h>
  27
  28 /* The macros defined here use the timestamp counter in i586 and up versions
  29    of the x86 processors.  They provide a very accurate way to measure the
  30    time with very little overhead.  The time values themself have no real
  31    meaning, only differences are interesting.
  32
  33    This version is for the i686 processors.  The difference to the i586
  34    version is that the timerstamp register is unconditionally used.  This is
  35    not the case for the i586 version where we have to perform runtime test
  36    whether the processor really has this capability.  We have to make this
  37    distinction since the sysdeps/i386/i586 code is supposed to work on all
  38    platforms while the i686 already contains i686-specific code.
  39
  40    The list of macros we need includes the following:
  41
  42    - HP_TIMING_AVAIL: test for availability.
  43
  44    - HP_TIMING_INLINE: this macro is non-zero if the functionality is not
  45      implemented using function calls but instead uses some inlined code
  46      which might simply consist of a few assembler instructions.  We have to
  47      know this since we might want to use the macros here in places where we
  48      cannot make function calls.
  49
  50    - hp_timing_t: This is the type for variables used to store the time
  51      values.
  52
  53    - HP_TIMING_ZERO: clear `hp_timing_t' object.
  54
  55    - HP_TIMING_NOW: place timestamp for current time in variable given as
  56      parameter.
  57
  58    - HP_TIMING_DIFF_INIT: do whatever is necessary to be able to use the
  59      HP_TIMING_DIFF macro.
  60
  61    - HP_TIMING_DIFF: compute difference between two times and store it
  62      in a third.  Source and destination might overlap.
  63
  64    - HP_TIMING_ACCUM: add time difference to another variable.  This might
  65      be a bit more complicated to implement for some platforms as the
  66      operation should be thread-safe and 64bit arithmetic on 32bit platforms
  67      is not.
  68
  69    - HP_TIMING_ACCUM_NT: this is the variant for situations where we know
  70      there are no threads involved.
  71
  72    - HP_TIMING_PRINT: write decimal representation of the timing value into
  73      the given string.  This operation need not be inline even though
  74      HP_TIMING_INLINE is specified.
  75
  76 */
  77
  78 /* We always assume having the timestamp register.  */
  79 #define HP_TIMING_AVAIL         (1)
  80
  81 /* We indeed have inlined functions.  */
  82 #define HP_TIMING_INLINE        (1)
  83
  84 /* We use 64bit values for the times.  */
  85 typedef unsigned long long int hp_timing_t;
  86
  87 /* Set timestamp value to zero.  */
  88 #define HP_TIMING_ZERO(Var)     (Var) = (0)
  89
  90 /* That's quite simple.  Use the `rdtsc' instruction.  Note that the value
  91    might not be 100% accurate since there might be some more instructions
  92    running in this moment.  This could be changed by using a barrier like
  93    'cpuid' right before the `rdtsc' instruciton.  But we are not interested
  94    in accurate clock cycles here so we don't do this.  */
  95 #define HP_TIMING_NOW(Var)      __asm__ __volatile__ ("rdtsc" : "=A" (Var))
  96
  97 /* Use two 'rdtsc' instructions in a row to find out how long it takes.  */
  98 #define HP_TIMING_DIFF_INIT() \
  99   do {                                                                        \
 100     if (GLRO(dl_hp_timing_overhead) == 0)                                     \
 101       {                                                                       \
 102         int __cnt = 5;                                                        \
 103         GLRO(dl_hp_timing_overhead) = ~0ull;                                  \
 104         do                                                                    \
 105           {                                                                   \
 106             hp_timing_t __t1, __t2;                                           \
 107             HP_TIMING_NOW (__t1);                                             \
 108             HP_TIMING_NOW (__t2);                                             \
 109             if (__t2 - __t1 < GLRO(dl_hp_timing_overhead))                    \
 110               GLRO(dl_hp_timing_overhead) = __t2 - __t1;                      \
 111           }                                                                   \
 112         while (--__cnt > 0);                                                  \
 113       }                                                                       \
 114   } while (0)
 115
 116 /* It's simple arithmetic for us.  */
 117 #define HP_TIMING_DIFF(Diff, Start, End)        (Diff) = ((End) - (Start))
 118
 119 /* We have to jump through hoops to get this correctly implemented.  */
 120 #define HP_TIMING_ACCUM(Sum, Diff) \
 121   do {                                                                        \
 122     int __not_done;                                                           \
 123     hp_timing_t __oldval = (Sum);                                             \
 124     hp_timing_t __diff = (Diff) - GLRO(dl_hp_timing_overhead);                \
 125     do                                                                        \
 126       {                                                                       \
 127         hp_timing_t __newval = __oldval + __diff;                             \
 128         int __temp0, __temp1;                                                 \
 129         __asm__ __volatile__ ("xchgl %0, %%ebx\n\t"                           \
 130                               "lock; cmpxchg8b %1\n\t"                        \
 131                               "sete %%bl\n\t"                                 \
 132                               "xchgl %0, %%ebx"                               \
 133                               : "=SD" (__not_done), "=m" (Sum),               \
 134                                 "=A" (__oldval), "=c" (__temp0)               \
 135                               : "m" (Sum), "2" (__oldval),                    \
 136                                 "3" ((unsigned int) (__newval >> 32)),        \
 137                                 "0" ((unsigned int) __newval));               \
 138       }                                                                       \
 139     while ((unsigned char) __not_done);                                       \
 140   } while (0)
 141
 142 /* No threads, no extra work.  */
 143 #define HP_TIMING_ACCUM_NT(Sum, Diff)   (Sum) += (Diff)
 144
 145 /* Print the time value.  */
 146 #define HP_TIMING_PRINT(Buf, Len, Val) \
 147   do {                                                                        \
 148     char __buf[20];                                                           \
 149     char *__cp = _itoa (Val, __buf + sizeof (__buf), 10, 0);                  \
 150     size_t __len = (Len);                                                     \
 151     char *__dest = (Buf);                                                     \
 152     while (__len-- > 0 && __cp < __buf + sizeof (__buf))                      \
 153       *__dest++ = *__cp++;                                                    \
 154     memcpy (__dest, " clock cycles", MIN (__len, sizeof (" clock cycles")));  \
 155   } while (0)
 156
 157 #endif  /* hp-timing.h */