sysdeps/powerpc/powerpc64/hp-timing.h

   1 /* High precision, low overhead timing functions.  powerpc64 version.
   2    Copyright (C) 2005-2014 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 #include <atomic.h>
  27
  28 /* The macros defined here use the powerpc 64-bit time base register.
  29    The time base is nominally clocked at 1/8th the CPU clock, but this
  30    can vary.
  31
  32    The list of macros we need includes the following:
  33
  34    - HP_TIMING_AVAIL: test for availability.
  35
  36    - HP_TIMING_INLINE: this macro is non-zero if the functionality is not
  37      implemented using function calls but instead uses some inlined code
  38      which might simply consist of a few assembler instructions.  We have to
  39      know this since we might want to use the macros here in places where we
  40      cannot make function calls.
  41
  42    - hp_timing_t: This is the type for variables used to store the time
  43      values.
  44
  45    - HP_TIMING_ZERO: clear `hp_timing_t' object.
  46
  47    - HP_TIMING_NOW: place timestamp for current time in variable given as
  48      parameter.
  49
  50    - HP_TIMING_DIFF_INIT: do whatever is necessary to be able to use the
  51      HP_TIMING_DIFF macro.
  52
  53    - HP_TIMING_DIFF: compute difference between two times and store it
  54      in a third.  Source and destination might overlap.
  55
  56    - HP_TIMING_ACCUM: add time difference to another variable.  This might
  57      be a bit more complicated to implement for some platforms as the
  58      operation should be thread-safe and 64bit arithmetic on 32bit platforms
  59      is not.
  60
  61    - HP_TIMING_ACCUM_NT: this is the variant for situations where we know
  62      there are no threads involved.
  63
  64    - HP_TIMING_PRINT: write decimal representation of the timing value into
  65      the given string.  This operation need not be inline even though
  66      HP_TIMING_INLINE is specified.
  67
  68 */
  69
  70 /* We always assume having the timestamp register.  */
  71 #define HP_TIMING_AVAIL         (1)
  72
  73 /* We indeed have inlined functions.  */
  74 #define HP_TIMING_INLINE        (1)
  75
  76 /* We use 64bit values for the times.  */
  77 typedef unsigned long long int hp_timing_t;
  78
  79 /* Set timestamp value to zero.  */
  80 #define HP_TIMING_ZERO(Var)     (Var) = (0)
  81
  82 /* That's quite simple.  Use the `mftb' instruction.  Note that the value
  83    might not be 100% accurate since there might be some more instructions
  84    running in this moment.  This could be changed by using a barrier like
  85    'lwsync' right before the `mftb' instruction.  But we are not interested
  86    in accurate clock cycles here so we don't do this.  */
  87 #ifdef _ARCH_PWR4
  88 #define HP_TIMING_NOW(Var)      __asm__ __volatile__ ("mfspr %0,268" : "=r" (Var))
  89 #else
  90 #define HP_TIMING_NOW(Var)      __asm__ __volatile__ ("mftb %0" : "=r" (Var))
  91 #endif
  92
  93 /* Use two 'mftb' instructions in a row to find out how long it takes.
  94    On current POWER4, POWER5, and 970 processors mftb take ~10 cycles.  */
  95 #define HP_TIMING_DIFF_INIT() \
  96   do {                                                                        \
  97     if (GLRO(dl_hp_timing_overhead) == 0)                                     \
  98       {                                                                       \
  99         int __cnt = 5;                                                        \
 100         GLRO(dl_hp_timing_overhead) = ~0ull;                                  \
 101         do                                                                    \
 102           {                                                                   \
 103             hp_timing_t __t1, __t2;                                           \
 104             HP_TIMING_NOW (__t1);                                             \
 105             HP_TIMING_NOW (__t2);                                             \
 106             if (__t2 - __t1 < GLRO(dl_hp_timing_overhead))                    \
 107               GLRO(dl_hp_timing_overhead) = __t2 - __t1;                      \
 108           }                                                                   \
 109         while (--__cnt > 0);                                                  \
 110       }                                                                       \
 111   } while (0)
 112
 113 /* It's simple arithmetic in 64-bit.  */
 114 #define HP_TIMING_DIFF(Diff, Start, End)        (Diff) = ((End) - (Start))
 115
 116 /* We need to insure that this add is atomic in threaded environments.  We use
 117    __arch_atomic_exchange_and_add_64 from atomic.h to get thread safety.  */
 118 #define HP_TIMING_ACCUM(Sum, Diff) \
 119   do {                                                                        \
 120     hp_timing_t __diff = (Diff) - GLRO(dl_hp_timing_overhead);                \
 121     __arch_atomic_exchange_and_add_64 (&(Sum), __diff);                       \
 122   } while (0)
 123
 124 /* No threads, no extra work.  */
 125 #define HP_TIMING_ACCUM_NT(Sum, Diff)   (Sum) += (Diff)
 126
 127 /* Print the time value.  */
 128 #define HP_TIMING_PRINT(Buf, Len, Val) \
 129   do {                                                                        \
 130     char __buf[20];                                                           \
 131     char *__cp = _itoa (Val, __buf + sizeof (__buf), 10, 0);                  \
 132     size_t __len = (Len);                                                     \
 133     char *__dest = (Buf);                                                     \
 134     while (__len-- > 0 && __cp < __buf + sizeof (__buf))                      \
 135       *__dest++ = *__cp++;                                                    \
 136     memcpy (__dest, " ticks", MIN (__len, sizeof (" ticks")));  \
 137   } while (0)
 138
 139 #endif  /* hp-timing.h */