libcilkrts/runtime/stats.h

   1 /* stats.h                  -*-C++-*-
   2  *
   3  *************************************************************************
   4  *
   5  *  Copyright (C) 2009-2016, Intel Corporation
   6  *  All rights reserved.
   7  *
   8  *  Redistribution and use in source and binary forms, with or without
   9  *  modification, are permitted provided that the following conditions
  10  *  are met:
  11  *
  12  *    * Redistributions of source code must retain the above copyright
  13  *      notice, this list of conditions and the following disclaimer.
  14  *    * Redistributions in binary form must reproduce the above copyright
  15  *      notice, this list of conditions and the following disclaimer in
  16  *      the documentation and/or other materials provided with the
  17  *      distribution.
  18  *    * Neither the name of Intel Corporation nor the names of its
  19  *      contributors may be used to endorse or promote products derived
  20  *      from this software without specific prior written permission.
  21  *
  22  *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  25  *  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  *  HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  *  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  28  *  BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
  29  *  OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  30  *  AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  31  *  LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
  32  *  WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  *  POSSIBILITY OF SUCH DAMAGE.
  34  *
  35  *  *********************************************************************
  36  *
  37  *  PLEASE NOTE: This file is a downstream copy of a file mainitained in
  38  *  a repository at cilkplus.org. Changes made to this file that are not
  39  *  submitted through the contribution process detailed at
  40  *  http://www.cilkplus.org/submit-cilk-contribution will be lost the next
  41  *  time that a new version is released. Changes only submitted to the
  42  *  GNU compiler collection or posted to the git repository at
  43  *  https://bitbucket.org/intelcilkruntime/intel-cilk-runtime.git are
  44  *  not tracked.
  45  *
  46  *  We welcome your contributions to this open source project. Thank you
  47  *  for your assistance in helping us improve Cilk Plus.
  48  **************************************************************************/
  49
  50 /**
  51  * @file stats.h
  52  *
  53  * @brief Support for gathering and reporting statistics on Cilk applications.
  54  *
  55  * Note that stats are normally NOT compiled in because it increases the
  56  * overhead of stealing.  To compile in profiling support, define CILK_PROFILE.
  57  */
  58
  59 #ifndef INCLUDED_STATS_DOT_H
  60 #define INCLUDED_STATS_DOT_H
  61
  62 /* #define CILK_PROFILE 1 */
  63 // @note  The CILK_PROFILE flag and intervals is known to be broken
  64 //        in at least programs with Windows exceptions.
  65 //        Enable this flag at your own peril. :)
  66
  67 #include <cilk/common.h>
  68 #include "rts-common.h"
  69 #include "internal/abi.h"
  70
  71 #ifdef CILK_PROFILE
  72 #include <stdio.h>     // Define FILE *
  73 #endif
  74
  75 __CILKRTS_BEGIN_EXTERN_C
  76
  77 // Some notes on the stats we are counting for the Cilk Plus runtime.
  78 //
  79 // INTERVAL_IN_SCHEDULER:
  80 //
  81 //   For a user worker, it starts counting when the worker thread is
  82 //   bound, and stops when the worker thread is unbound.
  83 //
  84 //   For a system worker, it starts counting when the worker thread is
  85 //   created, and stops after it returns from the scheduling loop
  86 //   (which usually won't happen unless __cilkrts_end_cilk() is
  87 //   called.)
  88 //
  89 // INTERVAL_WORKING + INTERVAL_IN_RUNTIME should be equal to
  90 // INTERVAL_IN_SCHEDULER (modulo overheads in instrumentation).
  91 //
  92 // INTERVAL_IN_RUNTIME tries to count all the time spent inside
  93 // runtime code.  (leave_frame and reudctions blur the accounting a
  94 // bit).  We switch from WORKING to IN_RUNTIME from:
  95 //
  96 //    1. __cilkrts_leave_frame, when we start doing an exception check
  97 //       and might jump into the runtime, either when returning from a
  98 //       spawn or have an exception to process.
  99 //
 100 //    2. __cilkrts_c_sync, when we are at a sync and the frame has
 101 //       been stolen from.
 102 //
 103 // We switch back to WORKING when we are returning to user code.
 104 //
 105 // We also switch back to WORKING when we need to do a slow merge of
 106 // reducer maps.  Otherwise, parallel reductions can mess up the stat
 107 // collection.
 108
 109 /** @brief Events that we measure. */
 110 enum interval
 111 {
 112     INTERVAL_IN_SCHEDULER,                  ///< Time threads spend "bound" to Cilk
 113     INTERVAL_WORKING,                       ///< Time spent working
 114     INTERVAL_IN_RUNTIME,                    ///< Time spent executing runtime code
 115     INTERVAL_SCHED_LOOP,                    ///< Time spent in scheduling loop
 116     INTERVAL_STEALING,                      ///< Time spent stealing work
 117     INTERVAL_STEAL_SUCCESS,                 ///< Time to do a successful steal
 118     INTERVAL_STEAL_FAIL_EMPTYQ,             ///< Count of steal failures due to lack of stealable work
 119     INTERVAL_STEAL_FAIL_LOCK,               ///< Count of steal failures due to failure to lock worker
 120     INTERVAL_STEAL_FAIL_USER_WORKER,        ///< Count of steal failures by user workers which attempt to steal from another team
 121     INTERVAL_STEAL_FAIL_DEKKER,             ///< Count of steal failures due to Dekker protocol failure
 122     INTERVAL_SYNC_CHECK,                    ///< Time spent processing syncs
 123     INTERVAL_THE_EXCEPTION_CHECK,           ///< Time spent performing THE exception checks
 124     INTERVAL_THE_EXCEPTION_CHECK_USELESS,   ///< Count of useless THE exception checks
 125     INTERVAL_RETURNING,                     ///< Time spent returning from calls
 126     INTERVAL_FINALIZE_CHILD,                ///< Time spent in finalize_child
 127     INTERVAL_PROVABLY_GOOD_STEAL,           ///< Time spent in provably_good_steal
 128     INTERVAL_UNCONDITIONAL_STEAL,           ///< Time spent in unconditional_steal
 129     INTERVAL_ALLOC_FULL_FRAME,              ///< Time spent in __cilkrts_make_full_frame
 130     INTERVAL_FRAME_ALLOC_LARGE,             ///< Count of calls to __cilkrts_frame_malloc for buffers bigger than FRAME_MALLOC_MAX_SIZE or with a NULL worker
 131     INTERVAL_FRAME_ALLOC,                   ///< Time spent allocating memory from worker buckets
 132     INTERVAL_FRAME_ALLOC_GLOBAL,            ///< Time spent calling memory allocator when buckets are empty
 133     INTERVAL_FRAME_FREE_LARGE,              ///< Count of calls to __cilkrts_frame_malloc for buffers bigger than FRAME_MALLOC_MAX_SIZE or with a NULL worker
 134     INTERVAL_FRAME_FREE,                    ///< Time spent freeing memory to worker buckets
 135     INTERVAL_FRAME_FREE_GLOBAL,             ///< Time spent calling memory deallocator when buckets are full
 136     INTERVAL_MUTEX_LOCK,                    ///< Count of calls to __cilkrts_mutex_lock for a worker
 137     INTERVAL_MUTEX_LOCK_SPINNING,           ///< Time spent spinning in __cilkrts_mutex_lock for a worker
 138     INTERVAL_MUTEX_LOCK_YIELDING,           ///< Time spent yielding in __cilkrts_mutex_lock for a worker
 139     INTERVAL_MUTEX_TRYLOCK,                 ///< Count of calls to __cilkrts_mutex_trylock
 140     INTERVAL_FIBER_ALLOCATE,                ///< Time spent calling cilk_fiber_allocate
 141     INTERVAL_FIBER_DEALLOCATE,              ///< Time spent calling cilk_fiber_deallocate (not from thread)
 142     INTERVAL_FIBER_ALLOCATE_FROM_THREAD,    ///< Time spent calling cilk_fiber_allocate_from_thread
 143     INTERVAL_FIBER_DEALLOCATE_FROM_THREAD,  ///< Time spent calling cilk_fiber_deallocate (from thread)
 144     INTERVAL_SUSPEND_RESUME_OTHER,          ///< Count of fiber suspend_self_and_resume_other
 145     INTERVAL_DEALLOCATE_RESUME_OTHER,       ///< Count of fiber deallocate_self_and_resume_other
 146     INTERVAL_INIT_WORKER,                   ///< Time spent on initialization of worker thread
 147     INTERVAL_SCHEDULE_WAIT,                 ///< Time worker spends in SCHEDULE_WAIT state
 148     INTERVAL_N                              ///< Number of intervals, must be last
 149 };
 150
 151 /**
 152  * @brief Struct that collects of all runtime statistics.
 153  *
 154  * There is an instance of this structure in each worker's
 155  * local_state, as well as one in the @c global_state_t which will be
 156  * used to accumulate the per-worker stats.
 157  */
 158 typedef struct statistics
 159 {
 160     /** Number of times each interval is entered */
 161     unsigned long long count[INTERVAL_N];
 162
 163     /**
 164      * Time when the system entered each interval, in system-dependent
 165      * "ticks"
 166      */
 167     unsigned long long start[INTERVAL_N];
 168
 169     /** Total time spent in each interval, in system-dependent "ticks" */
 170     unsigned long long accum[INTERVAL_N];
 171
 172     /**
 173      * Largest global number of stacks seen by this worker.
 174      * The true maximum at end of execution is the max of the
 175      * worker maxima.
 176      */
 177     long stack_hwm;
 178 } statistics;
 179
 180 /**
 181  * Initializes a statistics structure
 182  *
 183  * @param s The statistics structure to be initialized.
 184  */
 185 COMMON_PORTABLE void __cilkrts_init_stats(statistics *s);
 186
 187 /**
 188  * @brief Sums statistics from worker to the global struct
 189  *
 190  * @param to   The statistics structure that will accumulate the information.
 191  *             This structure is usually @c g->stats.
 192  * @param from The statistics structure that will be accumulated.
 193  *             This structure is usually statistics kept per worker.
 194  */
 195 COMMON_PORTABLE
 196 void __cilkrts_accum_stats(statistics *to, statistics *from);
 197
 198 /**
 199  * @brief Mark the start of an interval by saving the current tick count.
 200  *
 201  * @pre Start time == INVALID_START
 202  *
 203  * @param w The worker we're accumulating stats for.
 204  * @param i The interval we're accumulating stats for.
 205  */
 206 COMMON_PORTABLE
 207 void __cilkrts_start_interval(__cilkrts_worker *w, enum interval i);
 208
 209 /**
 210  * @brief Mark the end of an interval by adding the ticks since the
 211  * start to the accumulated time.
 212  *
 213  * @pre Start time != INVALID_START
 214  *
 215  * @param w The worker we're accumulating stats for.
 216  * @param i The interval we're accumulating stats for.
 217  */
 218 COMMON_PORTABLE
 219 void __cilkrts_stop_interval(__cilkrts_worker *w, enum interval i);
 220
 221 /**
 222  * @brief Start and stop interval I, charging zero time against it
 223  *
 224  * Precondition:
 225  * - Start time == INVALID_START
 226  *
 227  * @param w The worker we're accumulating stats for.
 228  * @param i The interval we're accumulating stats for.
 229  */
 230 COMMON_PORTABLE
 231 void __cilkrts_note_interval(__cilkrts_worker *w, enum interval i);
 232
 233 #ifdef CILK_PROFILE
 234 COMMON_PORTABLE
 235 void dump_stats_to_file(FILE *stat_file, statistics *s);
 236 #endif
 237
 238
 239 #ifdef CILK_PROFILE
 240 # define START_INTERVAL(w, i) __cilkrts_start_interval(w, i);
 241 # define STOP_INTERVAL(w, i) __cilkrts_stop_interval(w, i);
 242 # define NOTE_INTERVAL(w, i) __cilkrts_note_interval(w, i);
 243 #else
 244 /** Start an interval.  No effect unless CILK_PROFILE is defined. */
 245 # define START_INTERVAL(w, i)
 246 /** End an interval.  No effect unless CILK_PROFILE is defined. */
 247 # define STOP_INTERVAL(w, i)
 248 /** Increment a counter.  No effect unless CILK_PROFILE is defined. */
 249 # define NOTE_INTERVAL(w, i)
 250 #endif
 251
 252 __CILKRTS_END_EXTERN_C
 253
 254 #endif // ! defined(INCLUDED_STATS_DOT_H)