/* High precision, low overhead timing functions. powerpc64 version. Copyright (C) 2005, 2008 Free Software Foundation, Inc. This file is part of the GNU C Library. Contributed by Ulrich Drepper , 1998. The GNU C Library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 2.1 of the License, or (at your option) any later version. The GNU C Library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details. You should have received a copy of the GNU Lesser General Public License along with the GNU C Library; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA. */ #ifndef _HP_TIMING_H #define _HP_TIMING_H 1 #include #include #include #include /* The macros defined here use the powerpc 64-bit time base register. The time base is nominally clocked at 1/8th the CPU clock, but this can vary. The list of macros we need includes the following: - HP_TIMING_AVAIL: test for availability. - HP_TIMING_INLINE: this macro is non-zero if the functionality is not implemented using function calls but instead uses some inlined code which might simply consist of a few assembler instructions. We have to know this since we might want to use the macros here in places where we cannot make function calls. - hp_timing_t: This is the type for variables used to store the time values. - HP_TIMING_ZERO: clear `hp_timing_t' object. - HP_TIMING_NOW: place timestamp for current time in variable given as parameter. - HP_TIMING_DIFF_INIT: do whatever is necessary to be able to use the HP_TIMING_DIFF macro. - HP_TIMING_DIFF: compute difference between two times and store it in a third. Source and destination might overlap. - HP_TIMING_ACCUM: add time difference to another variable. This might be a bit more complicated to implement for some platforms as the operation should be thread-safe and 64bit arithmetic on 32bit platforms is not. - HP_TIMING_ACCUM_NT: this is the variant for situations where we know there are no threads involved. - HP_TIMING_PRINT: write decimal representation of the timing value into the given string. This operation need not be inline even though HP_TIMING_INLINE is specified. */ /* We always assume having the timestamp register. */ #define HP_TIMING_AVAIL (1) /* We indeed have inlined functions. */ #define HP_TIMING_INLINE (1) /* We use 64bit values for the times. */ typedef unsigned long long int hp_timing_t; /* Set timestamp value to zero. */ #define HP_TIMING_ZERO(Var) (Var) = (0) /* That's quite simple. Use the `mftb' instruction. Note that the value might not be 100% accurate since there might be some more instructions running in this moment. This could be changed by using a barrier like 'lwsync' right before the `mftb' instruciton. But we are not interested in accurate clock cycles here so we don't do this. */ #define HP_TIMING_NOW(Var) \ do { \ union { long long ll; long ii[2]; } _var; \ long tmp; \ __asm__ __volatile__ ( \ "1: mfspr %0,269;" \ " mfspr %1,268;" \ " mfspr %2,269;" \ " cmpw %0,%2;" \ " bne 1b;" \ : "=r" (_var.ii[0]), "=r" (_var.ii[1]) , "=r" (tmp) \ : : "cr0" \ ); \ Var = _var.ll; \ } while (0) /* Use two 'mftb' instructions in a row to find out how long it takes. On current POWER4, POWER5, and 970 processors mftb take ~10 cycles. */ #define HP_TIMING_DIFF_INIT() \ do { \ if (GLRO(dl_hp_timing_overhead) == 0) \ { \ int __cnt = 5; \ GLRO(dl_hp_timing_overhead) = ~0ull; \ do \ { \ hp_timing_t __t1, __t2; \ HP_TIMING_NOW (__t1); \ HP_TIMING_NOW (__t2); \ if (__t2 - __t1 < GLRO(dl_hp_timing_overhead)) \ GLRO(dl_hp_timing_overhead) = __t2 - __t1; \ } \ while (--__cnt > 0); \ } \ } while (0) /* It's simple arithmetic in 64-bit. */ #define HP_TIMING_DIFF(Diff, Start, End) (Diff) = ((End) - (Start)) /* We need to insure that this add is atomic in threaded environments. We use __arch_atomic_exchange_and_add_64 from atomic.h to get thread safety. */ #define HP_TIMING_ACCUM(Sum, Diff) \ do { \ hp_timing_t __diff = (Diff) - GLRO(dl_hp_timing_overhead); \ __arch_atomic_exchange_and_add_64 (&(Sum), __diff); \ } while (0) /* No threads, no extra work. */ #define HP_TIMING_ACCUM_NT(Sum, Diff) (Sum) += (Diff) /* Print the time value. */ #define HP_TIMING_PRINT(Buf, Len, Val) \ do { \ char __buf[20]; \ char *__cp = _itoa (Val, __buf + sizeof (__buf), 10, 0); \ size_t __len = (Len); \ char *__dest = (Buf); \ while (__len-- > 0 && __cp < __buf + sizeof (__buf)) \ *__dest++ = *__cp++; \ memcpy (__dest, " ticks", MIN (__len, sizeof (" ticks"))); \ } while (0) #endif /* hp-timing.h */