sysdeps/powerpc/powerpc32/power4/hp-timing.h

   1 /* High precision, low overhead timing functions.  powerpc64 version.
   2    Copyright (C) 2005, 2008 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 #include <atomic.h>
  28
  29 /* The macros defined here use the powerpc 64-bit time base register.
  30    The time base is nominally clocked at 1/8th the CPU clock, but this
  31    can vary.
  32
  33    The list of macros we need includes the following:
  34
  35    - HP_TIMING_AVAIL: test for availability.
  36
  37    - HP_TIMING_INLINE: this macro is non-zero if the functionality is not
  38      implemented using function calls but instead uses some inlined code
  39      which might simply consist of a few assembler instructions.  We have to
  40      know this since we might want to use the macros here in places where we
  41      cannot make function calls.
  42
  43    - hp_timing_t: This is the type for variables used to store the time
  44      values.
  45
  46    - HP_TIMING_ZERO: clear `hp_timing_t' object.
  47
  48    - HP_TIMING_NOW: place timestamp for current time in variable given as
  49      parameter.
  50
  51    - HP_TIMING_DIFF_INIT: do whatever is necessary to be able to use the
  52      HP_TIMING_DIFF macro.
  53
  54    - HP_TIMING_DIFF: compute difference between two times and store it
  55      in a third.  Source and destination might overlap.
  56
  57    - HP_TIMING_ACCUM: add time difference to another variable.  This might
  58      be a bit more complicated to implement for some platforms as the
  59      operation should be thread-safe and 64bit arithmetic on 32bit platforms
  60      is not.
  61
  62    - HP_TIMING_ACCUM_NT: this is the variant for situations where we know
  63      there are no threads involved.
  64
  65    - HP_TIMING_PRINT: write decimal representation of the timing value into
  66      the given string.  This operation need not be inline even though
  67      HP_TIMING_INLINE is specified.
  68
  69 */
  70
  71 /* We always assume having the timestamp register.  */
  72 #define HP_TIMING_AVAIL         (1)
  73
  74 /* We indeed have inlined functions.  */
  75 #define HP_TIMING_INLINE        (1)
  76
  77 /* We use 64bit values for the times.  */
  78 typedef unsigned long long int hp_timing_t;
  79
  80 /* Set timestamp value to zero.  */
  81 #define HP_TIMING_ZERO(Var)     (Var) = (0)
  82
  83 /* That's quite simple.  Use the `mftb' instruction.  Note that the value
  84    might not be 100% accurate since there might be some more instructions
  85    running in this moment.  This could be changed by using a barrier like
  86    'lwsync' right before the `mftb' instruciton.  But we are not interested
  87    in accurate clock cycles here so we don't do this.  */
  88
  89 #define HP_TIMING_NOW(Var)                                              \
  90   do {                                                                  \
  91         union { long long ll; long ii[2]; } _var;                       \
  92         long tmp;                                                       \
  93         __asm__ __volatile__ (                                          \
  94                 "1:     mfspr   %0,269;"                                \
  95                 "       mfspr   %1,268;"                                \
  96                 "       mfspr   %2,269;"                                \
  97                 "       cmpw    %0,%2;"                                 \
  98                 "       bne     1b;"                                    \
  99                 : "=r" (_var.ii[0]), "=r" (_var.ii[1]) , "=r" (tmp)     \
 100                 : : "cr0"                                               \
 101                 );                                                      \
 102         Var = _var.ll;                                                  \
 103   } while (0)
 104
 105
 106 /* Use two 'mftb' instructions in a row to find out how long it takes.
 107    On current POWER4, POWER5, and 970 processors mftb take ~10 cycles.  */
 108 #define HP_TIMING_DIFF_INIT() \
 109   do {                                                                        \
 110     if (GLRO(dl_hp_timing_overhead) == 0)                                     \
 111       {                                                                       \
 112         int __cnt = 5;                                                        \
 113         GLRO(dl_hp_timing_overhead) = ~0ull;                                  \
 114         do                                                                    \
 115           {                                                                   \
 116             hp_timing_t __t1, __t2;                                           \
 117             HP_TIMING_NOW (__t1);                                             \
 118             HP_TIMING_NOW (__t2);                                             \
 119             if (__t2 - __t1 < GLRO(dl_hp_timing_overhead))                    \
 120               GLRO(dl_hp_timing_overhead) = __t2 - __t1;                      \
 121           }                                                                   \
 122         while (--__cnt > 0);                                                  \
 123       }                                                                       \
 124   } while (0)
 125
 126 /* It's simple arithmetic in 64-bit.  */
 127 #define HP_TIMING_DIFF(Diff, Start, End)        (Diff) = ((End) - (Start))
 128
 129 /* We need to insure that this add is atomic in threaded environments.  We use
 130    __arch_atomic_exchange_and_add_64 from atomic.h to get thread safety.  */
 131 #define HP_TIMING_ACCUM(Sum, Diff) \
 132   do {                                                                        \
 133     hp_timing_t __diff = (Diff) - GLRO(dl_hp_timing_overhead);                \
 134     __arch_atomic_exchange_and_add_64 (&(Sum), __diff);                       \
 135   } while (0)
 136
 137 /* No threads, no extra work.  */
 138 #define HP_TIMING_ACCUM_NT(Sum, Diff)   (Sum) += (Diff)
 139
 140 /* Print the time value.  */
 141 #define HP_TIMING_PRINT(Buf, Len, Val) \
 142   do {                                                                        \
 143     char __buf[20];                                                           \
 144     char *__cp = _itoa (Val, __buf + sizeof (__buf), 10, 0);                  \
 145     size_t __len = (Len);                                                     \
 146     char *__dest = (Buf);                                                     \
 147     while (__len-- > 0 && __cp < __buf + sizeof (__buf))                      \
 148       *__dest++ = *__cp++;                                                    \
 149     memcpy (__dest, " ticks", MIN (__len, sizeof (" ticks")));  \
 150   } while (0)
 151
 152 #endif  /* hp-timing.h */