FIX #2164: check for ThreadRelocated in isAlive()
[ghc.git] / rts / Stats.c
1 /* -----------------------------------------------------------------------------
2 *
3 * (c) The GHC Team, 1998-2005
4 *
5 * Statistics and timing-related functions.
6 *
7 * ---------------------------------------------------------------------------*/
8
9 #include "Rts.h"
10 #include "RtsFlags.h"
11 #include "RtsUtils.h"
12 #include "MBlock.h"
13 #include "Storage.h"
14 #include "Schedule.h"
15 #include "Stats.h"
16 #include "ParTicky.h" /* ToDo: move into Rts.h */
17 #include "Profiling.h"
18 #include "GetTime.h"
19 #include "GC.h"
20
21 #if USE_PAPI
22 #include "Papi.h"
23 #endif
24
25 /* huh? */
26 #define BIG_STRING_LEN 512
27
28 #define TICK_TO_DBL(t) ((double)(t) / TICKS_PER_SECOND)
29
30 static Ticks ElapsedTimeStart = 0;
31
32 static Ticks InitUserTime = 0;
33 static Ticks InitElapsedTime = 0;
34 static Ticks InitElapsedStamp = 0;
35
36 static Ticks MutUserTime = 0;
37 static Ticks MutElapsedTime = 0;
38 static Ticks MutElapsedStamp = 0;
39
40 static Ticks ExitUserTime = 0;
41 static Ticks ExitElapsedTime = 0;
42
43 static ullong GC_tot_alloc = 0;
44 static ullong GC_tot_copied = 0;
45
46 static ullong GC_par_max_copied = 0;
47 static ullong GC_par_avg_copied = 0;
48
49 static Ticks GC_start_time = 0, GC_tot_time = 0; /* User GC Time */
50 static Ticks GCe_start_time = 0, GCe_tot_time = 0; /* Elapsed GC time */
51
52 #ifdef PROFILING
53 static Ticks RP_start_time = 0, RP_tot_time = 0; /* retainer prof user time */
54 static Ticks RPe_start_time = 0, RPe_tot_time = 0; /* retainer prof elap time */
55
56 static Ticks HC_start_time, HC_tot_time = 0; // heap census prof user time
57 static Ticks HCe_start_time, HCe_tot_time = 0; // heap census prof elap time
58 #endif
59
60 #ifdef PROFILING
61 #define PROF_VAL(x) (x)
62 #else
63 #define PROF_VAL(x) 0
64 #endif
65
66 static lnat MaxResidency = 0; // in words; for stats only
67 static lnat AvgResidency = 0;
68 static lnat ResidencySamples = 0; // for stats only
69 static lnat MaxSlop = 0;
70
71 static lnat GC_start_faults = 0, GC_end_faults = 0;
72
73 static Ticks *GC_coll_times = NULL;
74 static Ticks *GC_coll_etimes = NULL;
75
76 static void statsFlush( void );
77 static void statsClose( void );
78
79 Ticks stat_getElapsedGCTime(void)
80 {
81 return GCe_tot_time;
82 }
83
84 Ticks stat_getElapsedTime(void)
85 {
86 return getProcessElapsedTime() - ElapsedTimeStart;
87 }
88
89 /* mut_user_time_during_GC() and mut_user_time()
90 *
91 * The former function can be used to get the current mutator time
92 * *during* a GC, i.e. between stat_startGC and stat_endGC. This is
93 * used in the heap profiler for accurately time stamping the heap
94 * sample.
95 *
96 * ATTENTION: mut_user_time_during_GC() relies on GC_start_time being
97 * defined in stat_startGC() - to minimise system calls,
98 * GC_start_time is, however, only defined when really needed (check
99 * stat_startGC() for details)
100 */
101 double
102 mut_user_time_during_GC( void )
103 {
104 return TICK_TO_DBL(GC_start_time - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time));
105 }
106
107 double
108 mut_user_time( void )
109 {
110 Ticks user;
111 user = getProcessCPUTime();
112 return TICK_TO_DBL(user - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time));
113 }
114
115 #ifdef PROFILING
116 /*
117 mut_user_time_during_RP() is similar to mut_user_time_during_GC();
118 it returns the MUT time during retainer profiling.
119 The same is for mut_user_time_during_HC();
120 */
121 double
122 mut_user_time_during_RP( void )
123 {
124 return TICK_TO_DBL(RP_start_time - GC_tot_time - RP_tot_time - HC_tot_time);
125 }
126
127 double
128 mut_user_time_during_heap_census( void )
129 {
130 return TICK_TO_DBL(HC_start_time - GC_tot_time - RP_tot_time - HC_tot_time);
131 }
132 #endif /* PROFILING */
133
134 // initStats0() has no dependencies, it can be called right at the beginning
135 void
136 initStats0(void)
137 {
138 ElapsedTimeStart = 0;
139
140 InitUserTime = 0;
141 InitElapsedTime = 0;
142 InitElapsedStamp = 0;
143
144 MutUserTime = 0;
145 MutElapsedTime = 0;
146 MutElapsedStamp = 0;
147
148 ExitUserTime = 0;
149 ExitElapsedTime = 0;
150
151 GC_tot_alloc = 0;
152 GC_tot_copied = 0;
153 GC_par_max_copied = 0;
154 GC_par_avg_copied = 0;
155 GC_start_time = 0;
156 GC_tot_time = 0;
157 GCe_start_time = 0;
158 GCe_tot_time = 0;
159
160 #ifdef PROFILING
161 RP_start_time = 0;
162 RP_tot_time = 0;
163 RPe_start_time = 0;
164 RPe_tot_time = 0;
165
166 HC_start_time = 0;
167 HC_tot_time = 0;
168 HCe_start_time = 0;
169 HCe_tot_time = 0;
170 #endif
171
172 MaxResidency = 0;
173 AvgResidency = 0;
174 ResidencySamples = 0;
175 MaxSlop = 0;
176
177 GC_start_faults = 0;
178 GC_end_faults = 0;
179 }
180
181 // initStats1() can be called after setupRtsFlags()
182 void
183 initStats1 (void)
184 {
185 nat i;
186
187 if (RtsFlags.GcFlags.giveStats >= VERBOSE_GC_STATS) {
188 statsPrintf(" Alloc Copied Live GC GC TOT TOT Page Flts\n");
189 statsPrintf(" bytes bytes bytes user elap user elap\n");
190 }
191 GC_coll_times =
192 (Ticks *)stgMallocBytes(
193 sizeof(Ticks)*RtsFlags.GcFlags.generations,
194 "initStats");
195 GC_coll_etimes =
196 (Ticks *)stgMallocBytes(
197 sizeof(Ticks)*RtsFlags.GcFlags.generations,
198 "initStats");
199 for (i = 0; i < RtsFlags.GcFlags.generations; i++) {
200 GC_coll_times[i] = 0;
201 GC_coll_etimes[i] = 0;
202 }
203 }
204
205 /* -----------------------------------------------------------------------------
206 Initialisation time...
207 -------------------------------------------------------------------------- */
208
209 void
210 stat_startInit(void)
211 {
212 Ticks elapsed;
213
214 elapsed = getProcessElapsedTime();
215 ElapsedTimeStart = elapsed;
216 }
217
218 void
219 stat_endInit(void)
220 {
221 Ticks user, elapsed;
222
223 getProcessTimes(&user, &elapsed);
224
225 InitUserTime = user;
226 InitElapsedStamp = elapsed;
227 if (ElapsedTimeStart > elapsed) {
228 InitElapsedTime = 0;
229 } else {
230 InitElapsedTime = elapsed - ElapsedTimeStart;
231 }
232 #if USE_PAPI
233 /* We start counting events for the mutator
234 * when garbage collection starts
235 * we switch to the GC event set. */
236 papi_start_mutator_count();
237
238 /* This flag is needed to avoid counting the last GC */
239 papi_is_reporting = 1;
240
241 #endif
242 }
243
244 /* -----------------------------------------------------------------------------
245 stat_startExit and stat_endExit
246
247 These two measure the time taken in shutdownHaskell().
248 -------------------------------------------------------------------------- */
249
250 void
251 stat_startExit(void)
252 {
253 Ticks user, elapsed;
254
255 getProcessTimes(&user, &elapsed);
256
257 MutElapsedStamp = elapsed;
258 MutElapsedTime = elapsed - GCe_tot_time -
259 PROF_VAL(RPe_tot_time + HCe_tot_time) - InitElapsedStamp;
260 if (MutElapsedTime < 0) { MutElapsedTime = 0; } /* sometimes -0.00 */
261
262 MutUserTime = user - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time) - InitUserTime;
263 if (MutUserTime < 0) { MutUserTime = 0; }
264
265 #if USE_PAPI
266 /* We stop counting mutator events
267 * GC events are not being counted at this point */
268 papi_stop_mutator_count();
269
270 /* This flag is needed, because GC is run once more after this function */
271 papi_is_reporting = 0;
272
273 #endif
274 }
275
276 void
277 stat_endExit(void)
278 {
279 Ticks user, elapsed;
280
281 getProcessTimes(&user, &elapsed);
282
283 ExitUserTime = user - MutUserTime - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time) - InitUserTime;
284 ExitElapsedTime = elapsed - MutElapsedStamp;
285 if (ExitUserTime < 0) {
286 ExitUserTime = 0;
287 }
288 if (ExitElapsedTime < 0) {
289 ExitElapsedTime = 0;
290 }
291 }
292
293 /* -----------------------------------------------------------------------------
294 Called at the beginning of each GC
295 -------------------------------------------------------------------------- */
296
297 static nat rub_bell = 0;
298
299 /* initialise global variables needed during GC
300 *
301 * * GC_start_time is read in mut_user_time_during_GC(), which in turn is
302 * needed if either PROFILING or DEBUGing is enabled
303 */
304 void
305 stat_startGC(void)
306 {
307 nat bell = RtsFlags.GcFlags.ringBell;
308
309 if (bell) {
310 if (bell > 1) {
311 debugBelch(" GC ");
312 rub_bell = 1;
313 } else {
314 debugBelch("\007");
315 }
316 }
317
318 #if defined(PROFILING) || defined(DEBUG)
319 GC_start_time = getProcessCPUTime(); // needed in mut_user_time_during_GC()
320 #endif
321
322 if (RtsFlags.GcFlags.giveStats != NO_GC_STATS) {
323 #if !defined(PROFILING) && !defined(DEBUG)
324 GC_start_time = getProcessCPUTime();
325 #endif
326 GCe_start_time = getProcessElapsedTime();
327 if (RtsFlags.GcFlags.giveStats) {
328 GC_start_faults = getPageFaults();
329 }
330 }
331
332 #if USE_PAPI
333 if(papi_is_reporting) {
334 /* Switch to counting GC events */
335 papi_stop_mutator_count();
336 papi_start_gc_count();
337 }
338 #endif
339
340 }
341
342 /* -----------------------------------------------------------------------------
343 Called at the end of each GC
344 -------------------------------------------------------------------------- */
345
346 void
347 stat_endGC (lnat alloc, lnat live, lnat copied, lnat gen,
348 lnat max_copied, lnat avg_copied, lnat slop)
349 {
350 if (RtsFlags.GcFlags.giveStats != NO_GC_STATS) {
351 Ticks time, etime, gc_time, gc_etime;
352
353 getProcessTimes(&time, &etime);
354 gc_time = time - GC_start_time;
355 gc_etime = etime - GCe_start_time;
356
357 if (RtsFlags.GcFlags.giveStats == VERBOSE_GC_STATS) {
358 nat faults = getPageFaults();
359
360 statsPrintf("%9ld %9ld %9ld",
361 alloc*sizeof(W_), copied*sizeof(W_),
362 live*sizeof(W_));
363 statsPrintf(" %5.2f %5.2f %7.2f %7.2f %4ld %4ld (Gen: %2ld)\n",
364 TICK_TO_DBL(gc_time),
365 TICK_TO_DBL(gc_etime),
366 TICK_TO_DBL(time),
367 TICK_TO_DBL(etime - ElapsedTimeStart),
368 faults - GC_start_faults,
369 GC_start_faults - GC_end_faults,
370 gen);
371
372 GC_end_faults = faults;
373 statsFlush();
374 }
375
376 GC_coll_times[gen] += gc_time;
377 GC_coll_etimes[gen] += gc_etime;
378
379 GC_tot_copied += (ullong) copied;
380 GC_tot_alloc += (ullong) alloc;
381 GC_par_max_copied += (ullong) max_copied;
382 GC_par_avg_copied += (ullong) avg_copied;
383 GC_tot_time += gc_time;
384 GCe_tot_time += gc_etime;
385
386 #if defined(THREADED_RTS)
387 {
388 Task *task;
389 if ((task = myTask()) != NULL) {
390 task->gc_time += gc_time;
391 task->gc_etime += gc_etime;
392 }
393 }
394 #endif
395
396 if (gen == RtsFlags.GcFlags.generations-1) { /* major GC? */
397 if (live > MaxResidency) {
398 MaxResidency = live;
399 }
400 ResidencySamples++;
401 AvgResidency += live;
402 }
403
404 if (slop > MaxSlop) MaxSlop = slop;
405 }
406
407 if (rub_bell) {
408 debugBelch("\b\b\b \b\b\b");
409 rub_bell = 0;
410 }
411
412 #if USE_PAPI
413 if(papi_is_reporting) {
414 /* Switch to counting mutator events */
415 if (gen == 0) {
416 papi_stop_gc0_count();
417 } else {
418 papi_stop_gc1_count();
419 }
420 papi_start_mutator_count();
421 }
422 #endif
423 }
424
425 /* -----------------------------------------------------------------------------
426 Called at the beginning of each Retainer Profiliing
427 -------------------------------------------------------------------------- */
428 #ifdef PROFILING
429 void
430 stat_startRP(void)
431 {
432 Ticks user, elapsed;
433 getProcessTimes( &user, &elapsed );
434
435 RP_start_time = user;
436 RPe_start_time = elapsed;
437 }
438 #endif /* PROFILING */
439
440 /* -----------------------------------------------------------------------------
441 Called at the end of each Retainer Profiliing
442 -------------------------------------------------------------------------- */
443
444 #ifdef PROFILING
445 void
446 stat_endRP(
447 nat retainerGeneration,
448 #ifdef DEBUG_RETAINER
449 nat maxCStackSize,
450 int maxStackSize,
451 #endif
452 double averageNumVisit)
453 {
454 Ticks user, elapsed;
455 getProcessTimes( &user, &elapsed );
456
457 RP_tot_time += user - RP_start_time;
458 RPe_tot_time += elapsed - RPe_start_time;
459
460 fprintf(prof_file, "Retainer Profiling: %d, at %f seconds\n",
461 retainerGeneration, mut_user_time_during_RP());
462 #ifdef DEBUG_RETAINER
463 fprintf(prof_file, "\tMax C stack size = %u\n", maxCStackSize);
464 fprintf(prof_file, "\tMax auxiliary stack size = %u\n", maxStackSize);
465 #endif
466 fprintf(prof_file, "\tAverage number of visits per object = %f\n", averageNumVisit);
467 }
468 #endif /* PROFILING */
469
470 /* -----------------------------------------------------------------------------
471 Called at the beginning of each heap census
472 -------------------------------------------------------------------------- */
473 #ifdef PROFILING
474 void
475 stat_startHeapCensus(void)
476 {
477 Ticks user, elapsed;
478 getProcessTimes( &user, &elapsed );
479
480 HC_start_time = user;
481 HCe_start_time = elapsed;
482 }
483 #endif /* PROFILING */
484
485 /* -----------------------------------------------------------------------------
486 Called at the end of each heap census
487 -------------------------------------------------------------------------- */
488 #ifdef PROFILING
489 void
490 stat_endHeapCensus(void)
491 {
492 Ticks user, elapsed;
493 getProcessTimes( &user, &elapsed );
494
495 HC_tot_time += user - HC_start_time;
496 HCe_tot_time += elapsed - HCe_start_time;
497 }
498 #endif /* PROFILING */
499
500 /* -----------------------------------------------------------------------------
501 Called at the end of execution
502
503 NOTE: number of allocations is not entirely accurate: it doesn't
504 take into account the few bytes at the end of the heap that
505 were left unused when the heap-check failed.
506 -------------------------------------------------------------------------- */
507
508 #ifdef DEBUG
509 #define TICK_VAR(arity) \
510 extern StgInt SLOW_CALLS_##arity; \
511 extern StgInt RIGHT_ARITY_##arity; \
512 extern StgInt TAGGED_PTR_##arity;
513
514 #define TICK_VAR_INI(arity) \
515 StgInt SLOW_CALLS_##arity = 1; \
516 StgInt RIGHT_ARITY_##arity = 1; \
517 StgInt TAGGED_PTR_##arity = 0;
518
519 extern StgInt TOTAL_CALLS;
520
521 TICK_VAR(1)
522 TICK_VAR(2)
523
524 TICK_VAR_INI(1)
525 TICK_VAR_INI(2)
526
527 StgInt TOTAL_CALLS=1;
528 #endif
529
530 /* Report the value of a counter */
531 #define REPORT(counter) \
532 { \
533 ullong_format_string(counter,temp,rtsTrue/*commas*/); \
534 statsPrintf(" (" #counter ") : %s\n",temp); \
535 }
536
537 /* Report the value of a counter as a percentage of another counter */
538 #define REPORT_PCT(counter,countertot) \
539 statsPrintf(" (" #counter ") %% of (" #countertot ") : %.1f%%\n", \
540 counter*100.0/countertot)
541
542 #define TICK_PRINT(arity) \
543 REPORT(SLOW_CALLS_##arity); \
544 REPORT_PCT(RIGHT_ARITY_##arity,SLOW_CALLS_##arity); \
545 REPORT_PCT(TAGGED_PTR_##arity,RIGHT_ARITY_##arity); \
546 REPORT(RIGHT_ARITY_##arity); \
547 REPORT(TAGGED_PTR_##arity)
548
549 #define TICK_PRINT_TOT(arity) \
550 statsPrintf(" (SLOW_CALLS_" #arity ") %% of (TOTAL_CALLS) : %.1f%%\n", \
551 SLOW_CALLS_##arity * 100.0/TOTAL_CALLS)
552
553 extern lnat hw_alloc_blocks;
554
555 void
556 stat_exit(int alloc)
557 {
558 if (RtsFlags.GcFlags.giveStats != NO_GC_STATS) {
559
560 char temp[BIG_STRING_LEN];
561 Ticks time;
562 Ticks etime;
563 nat g, total_collections = 0;
564
565 getProcessTimes( &time, &etime );
566 etime -= ElapsedTimeStart;
567
568 GC_tot_alloc += alloc;
569
570 /* Count total garbage collections */
571 for (g = 0; g < RtsFlags.GcFlags.generations; g++)
572 total_collections += generations[g].collections;
573
574 /* avoid divide by zero if time is measured as 0.00 seconds -- SDM */
575 if (time == 0.0) time = 1;
576 if (etime == 0.0) etime = 1;
577
578 if (RtsFlags.GcFlags.giveStats >= VERBOSE_GC_STATS) {
579 statsPrintf("%9ld %9.9s %9.9s", (lnat)alloc*sizeof(W_), "", "");
580 statsPrintf(" %5.2f %5.2f\n\n", 0.0, 0.0);
581 }
582
583 if (RtsFlags.GcFlags.giveStats >= SUMMARY_GC_STATS) {
584 ullong_format_string(GC_tot_alloc*sizeof(W_),
585 temp, rtsTrue/*commas*/);
586 statsPrintf("%16s bytes allocated in the heap\n", temp);
587
588 ullong_format_string(GC_tot_copied*sizeof(W_),
589 temp, rtsTrue/*commas*/);
590 statsPrintf("%16s bytes copied during GC\n", temp);
591
592 if ( ResidencySamples > 0 ) {
593 ullong_format_string(MaxResidency*sizeof(W_),
594 temp, rtsTrue/*commas*/);
595 statsPrintf("%16s bytes maximum residency (%ld sample(s))\n",
596 temp, ResidencySamples);
597 }
598
599 ullong_format_string(MaxSlop*sizeof(W_), temp, rtsTrue/*commas*/);
600 statsPrintf("%16s bytes maximum slop\n", temp);
601
602 statsPrintf("%16ld MB total memory in use (%ld MB lost due to fragmentation)\n\n",
603 mblocks_allocated * MBLOCK_SIZE_W / (1024 * 1024 / sizeof(W_)),
604 (mblocks_allocated * MBLOCK_SIZE_W - hw_alloc_blocks * BLOCK_SIZE_W) / (1024 * 1024 / sizeof(W_)));
605
606 /* Print garbage collections in each gen */
607 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
608 statsPrintf(" Generation %d: %5d collections, %5d parallel, %5.2fs, %5.2fs elapsed\n",
609 g, generations[g].collections,
610 generations[g].par_collections,
611 TICK_TO_DBL(GC_coll_times[g]),
612 TICK_TO_DBL(GC_coll_etimes[g]));
613 }
614
615 #if defined(THREADED_RTS)
616 if (RtsFlags.ParFlags.gcThreads > 1) {
617 statsPrintf("\n Parallel GC work balance: %.2f (%ld / %ld, ideal %d)\n",
618 (double)GC_par_avg_copied / (double)GC_par_max_copied,
619 (lnat)GC_par_avg_copied, (lnat)GC_par_max_copied,
620 RtsFlags.ParFlags.gcThreads
621 );
622 }
623 #endif
624
625 statsPrintf("\n");
626
627 #if defined(THREADED_RTS)
628 {
629 nat i;
630 Task *task;
631 for (i = 0, task = all_tasks;
632 task != NULL;
633 i++, task = task->all_link) {
634 statsPrintf(" Task %2d %-8s : MUT time: %6.2fs (%6.2fs elapsed)\n"
635 " GC time: %6.2fs (%6.2fs elapsed)\n\n",
636 i,
637 (task->tso == NULL) ? "(worker)" : "(bound)",
638 TICK_TO_DBL(task->mut_time),
639 TICK_TO_DBL(task->mut_etime),
640 TICK_TO_DBL(task->gc_time),
641 TICK_TO_DBL(task->gc_etime));
642 }
643 }
644 #endif
645
646 statsPrintf(" INIT time %6.2fs (%6.2fs elapsed)\n",
647 TICK_TO_DBL(InitUserTime), TICK_TO_DBL(InitElapsedTime));
648 statsPrintf(" MUT time %6.2fs (%6.2fs elapsed)\n",
649 TICK_TO_DBL(MutUserTime), TICK_TO_DBL(MutElapsedTime));
650 statsPrintf(" GC time %6.2fs (%6.2fs elapsed)\n",
651 TICK_TO_DBL(GC_tot_time), TICK_TO_DBL(GCe_tot_time));
652 #ifdef PROFILING
653 statsPrintf(" RP time %6.2fs (%6.2fs elapsed)\n",
654 TICK_TO_DBL(RP_tot_time), TICK_TO_DBL(RPe_tot_time));
655 statsPrintf(" PROF time %6.2fs (%6.2fs elapsed)\n",
656 TICK_TO_DBL(HC_tot_time), TICK_TO_DBL(HCe_tot_time));
657 #endif
658 statsPrintf(" EXIT time %6.2fs (%6.2fs elapsed)\n",
659 TICK_TO_DBL(ExitUserTime), TICK_TO_DBL(ExitElapsedTime));
660 statsPrintf(" Total time %6.2fs (%6.2fs elapsed)\n\n",
661 TICK_TO_DBL(time), TICK_TO_DBL(etime));
662 statsPrintf(" %%GC time %5.1f%% (%.1f%% elapsed)\n\n",
663 TICK_TO_DBL(GC_tot_time)*100/TICK_TO_DBL(time),
664 TICK_TO_DBL(GCe_tot_time)*100/TICK_TO_DBL(etime));
665
666 if (time - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time) == 0)
667 ullong_format_string(0, temp, rtsTrue/*commas*/);
668 else
669 ullong_format_string(
670 (ullong)((GC_tot_alloc*sizeof(W_))/
671 TICK_TO_DBL(time - GC_tot_time -
672 PROF_VAL(RP_tot_time + HC_tot_time))),
673 temp, rtsTrue/*commas*/);
674
675 statsPrintf(" Alloc rate %s bytes per MUT second\n\n", temp);
676
677 statsPrintf(" Productivity %5.1f%% of total user, %.1f%% of total elapsed\n\n",
678 TICK_TO_DBL(time - GC_tot_time -
679 PROF_VAL(RP_tot_time + HC_tot_time) - InitUserTime) * 100
680 / TICK_TO_DBL(time),
681 TICK_TO_DBL(time - GC_tot_time -
682 PROF_VAL(RP_tot_time + HC_tot_time) - InitUserTime) * 100
683 / TICK_TO_DBL(etime));
684
685 /*
686 TICK_PRINT(1);
687 TICK_PRINT(2);
688 REPORT(TOTAL_CALLS);
689 TICK_PRINT_TOT(1);
690 TICK_PRINT_TOT(2);
691 */
692
693 #if USE_PAPI
694 papi_stats_report();
695 #endif
696 }
697
698 if (RtsFlags.GcFlags.giveStats == ONELINE_GC_STATS) {
699 /* print the long long separately to avoid bugginess on mingwin (2001-07-02, mingw-0.5) */
700 statsPrintf("<<ghc: %llu bytes, ", GC_tot_alloc*(ullong)sizeof(W_));
701 statsPrintf("%d GCs, %ld/%ld avg/max bytes residency (%ld samples), %luM in use, %.2f INIT (%.2f elapsed), %.2f MUT (%.2f elapsed), %.2f GC (%.2f elapsed) :ghc>>\n",
702 total_collections,
703 ResidencySamples == 0 ? 0 :
704 AvgResidency*sizeof(W_)/ResidencySamples,
705 MaxResidency*sizeof(W_),
706 ResidencySamples,
707 (unsigned long)(mblocks_allocated * MBLOCK_SIZE / (1024L * 1024L)),
708 TICK_TO_DBL(InitUserTime), TICK_TO_DBL(InitElapsedTime),
709 TICK_TO_DBL(MutUserTime), TICK_TO_DBL(MutElapsedTime),
710 TICK_TO_DBL(GC_tot_time), TICK_TO_DBL(GCe_tot_time));
711 }
712
713 #if defined(THREADED_RTS) && defined(PROF_SPIN)
714 {
715 nat g, s;
716
717 statsPrintf("recordMutableGen_sync: %"FMT_Word64"\n", recordMutableGen_sync.spin);
718 statsPrintf("gc_alloc_block_sync: %"FMT_Word64"\n", gc_alloc_block_sync.spin);
719 statsPrintf("whitehole_spin: %"FMT_Word64"\n", whitehole_spin);
720 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
721 for (s = 0; s < generations[g].n_steps; s++) {
722 statsPrintf("gen[%d].steps[%d].sync_todo: %"FMT_Word64"\n", g, s, generations[g].steps[s].sync_todo.spin);
723 statsPrintf("gen[%d].steps[%d].sync_large_objects: %"FMT_Word64"\n", g, s, generations[g].steps[s].sync_large_objects.spin);
724 }
725 }
726 }
727 #endif
728
729 statsFlush();
730 statsClose();
731 }
732
733 if (GC_coll_times)
734 stgFree(GC_coll_times);
735 GC_coll_times = NULL;
736 if (GC_coll_etimes)
737 stgFree(GC_coll_etimes);
738 GC_coll_etimes = NULL;
739 }
740
741 /* -----------------------------------------------------------------------------
742 stat_describe_gens
743
744 Produce some detailed info on the state of the generational GC.
745 -------------------------------------------------------------------------- */
746 void
747 statDescribeGens(void)
748 {
749 nat g, s, mut, lge;
750 lnat live, slop;
751 lnat tot_live, tot_slop;
752 bdescr *bd;
753 step *step;
754
755 debugBelch(
756 "-----------------------------------------------------------------\n"
757 " Gen Max Mut-list Step Blocks Large Live Slop\n"
758 " Blocks Bytes Objects \n"
759 "-----------------------------------------------------------------\n");
760
761 tot_live = 0;
762 tot_slop = 0;
763 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
764 mut = 0;
765 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
766 mut += (bd->free - bd->start) * sizeof(W_);
767 }
768
769 debugBelch("%5d %7d %9d", g, generations[g].max_blocks, mut);
770
771 for (s = 0; s < generations[g].n_steps; s++) {
772 step = &generations[g].steps[s];
773 for (bd = step->large_objects, lge = 0; bd; bd = bd->link) {
774 lge++;
775 }
776 live = step->n_words + countOccupied(step->large_objects);
777 if (s != 0) {
778 debugBelch("%23s","");
779 }
780 slop = (step->n_blocks + step->n_large_blocks) * BLOCK_SIZE_W - live;
781 debugBelch("%6d %8d %8d %8ld %8ld\n", s, step->n_blocks, lge,
782 live*sizeof(W_), slop*sizeof(W_));
783 tot_live += live;
784 tot_slop += slop;
785 }
786 }
787 debugBelch("-----------------------------------------------------------------\n");
788 debugBelch("%48s%8ld %8ld\n","",tot_live*sizeof(W_),tot_slop*sizeof(W_));
789 debugBelch("-----------------------------------------------------------------\n");
790 debugBelch("\n");
791 }
792
793 /* -----------------------------------------------------------------------------
794 Stats available via a programmatic interface, so eg. GHCi can time
795 each compilation and expression evaluation.
796 -------------------------------------------------------------------------- */
797
798 extern HsInt64 getAllocations( void )
799 { return (HsInt64)total_allocated * sizeof(W_); }
800
801 /* -----------------------------------------------------------------------------
802 Dumping stuff in the stats file, or via the debug message interface
803 -------------------------------------------------------------------------- */
804
805 void
806 statsPrintf( char *s, ... )
807 {
808 FILE *sf = RtsFlags.GcFlags.statsFile;
809 va_list ap;
810
811 va_start(ap,s);
812 if (sf == NULL) {
813 vdebugBelch(s,ap);
814 } else {
815 vfprintf(sf, s, ap);
816 }
817 va_end(ap);
818 }
819
820 static void
821 statsFlush( void )
822 {
823 FILE *sf = RtsFlags.GcFlags.statsFile;
824 if (sf != NULL) {
825 fflush(sf);
826 }
827 }
828
829 static void
830 statsClose( void )
831 {
832 FILE *sf = RtsFlags.GcFlags.statsFile;
833 if (sf != NULL) {
834 fclose(sf);
835 }
836 }