X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=rts%2FStats.c;h=73479181f906ef5e70f7a74a8aa8e3f9d050fe38;hb=5244f537d12bdb5e016ce7b04e9181b364e5cb27;hp=aabe259014a756b70096a5a95e7111928e89cbee;hpb=a7d236fbcf664fde8c81ba840efc1c471051814d;p=ghc-hetmet.git diff --git a/rts/Stats.c b/rts/Stats.c index aabe2590..7347918 100644 --- a/rts/Stats.c +++ b/rts/Stats.c @@ -16,6 +16,13 @@ #include "ParTicky.h" /* ToDo: move into Rts.h */ #include "Profiling.h" #include "GetTime.h" +#include "GC.h" +#include "GCUtils.h" +#include "Evac.h" + +#if USE_PAPI +#include "Papi.h" +#endif /* huh? */ #define BIG_STRING_LEN 512 @@ -37,7 +44,6 @@ static Ticks ExitElapsedTime = 0; static ullong GC_tot_alloc = 0; static ullong GC_tot_copied = 0; -static ullong GC_tot_scavd_copied = 0; static Ticks GC_start_time = 0, GC_tot_time = 0; /* User GC Time */ static Ticks GCe_start_time = 0, GCe_tot_time = 0; /* Elapsed GC time */ @@ -62,10 +68,8 @@ static lnat ResidencySamples = 0; // for stats only static lnat GC_start_faults = 0, GC_end_faults = 0; -static Ticks *GC_coll_times; - -static void statsPrintf( char *s, ... ) - GNUC3_ATTRIBUTE(format (printf, 1, 2)); +static Ticks *GC_coll_times = NULL; +static Ticks *GC_coll_etimes = NULL; static void statsFlush( void ); static void statsClose( void ); @@ -138,8 +142,13 @@ initStats(void) (Ticks *)stgMallocBytes( sizeof(Ticks)*RtsFlags.GcFlags.generations, "initStats"); + GC_coll_etimes = + (Ticks *)stgMallocBytes( + sizeof(Ticks)*RtsFlags.GcFlags.generations, + "initStats"); for (i = 0; i < RtsFlags.GcFlags.generations; i++) { GC_coll_times[i] = 0; + GC_coll_etimes[i] = 0; } } @@ -170,6 +179,16 @@ stat_endInit(void) } else { InitElapsedTime = elapsed - ElapsedTimeStart; } +#if USE_PAPI + /* We start counting events for the mutator + * when garbage collection starts + * we switch to the GC event set. */ + papi_start_mutator_count(); + + /* This flag is needed to avoid counting the last GC */ + papi_is_reporting = 1; + +#endif } /* ----------------------------------------------------------------------------- @@ -192,6 +211,16 @@ stat_startExit(void) MutUserTime = user - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time) - InitUserTime; if (MutUserTime < 0) { MutUserTime = 0; } + +#if USE_PAPI + /* We stop counting mutator events + * GC events are not being counted at this point */ + papi_stop_mutator_count(); + + /* This flag is needed, because GC is run once more after this function */ + papi_is_reporting = 0; + +#endif } void @@ -249,6 +278,15 @@ stat_startGC(void) GC_start_faults = getPageFaults(); } } + +#if USE_PAPI + if(papi_is_reporting) { + /* Switch to counting GC events */ + papi_stop_mutator_count(); + papi_start_gc_count(); + } +#endif + } /* ----------------------------------------------------------------------------- @@ -256,8 +294,7 @@ stat_startGC(void) -------------------------------------------------------------------------- */ void -stat_endGC (lnat alloc, lnat live, lnat copied, - lnat scavd_copied, lnat gen) +stat_endGC (lnat alloc, lnat live, lnat copied, lnat gen) { if (RtsFlags.GcFlags.giveStats != NO_GC_STATS) { Ticks time, etime, gc_time, gc_etime; @@ -270,7 +307,7 @@ stat_endGC (lnat alloc, lnat live, lnat copied, nat faults = getPageFaults(); statsPrintf("%9ld %9ld %9ld", - alloc*sizeof(W_), (copied+scavd_copied)*sizeof(W_), + alloc*sizeof(W_), copied*sizeof(W_), live*sizeof(W_)); statsPrintf(" %5.2f %5.2f %7.2f %7.2f %4ld %4ld (Gen: %2ld)\n", TICK_TO_DBL(gc_time), @@ -286,9 +323,9 @@ stat_endGC (lnat alloc, lnat live, lnat copied, } GC_coll_times[gen] += gc_time; + GC_coll_etimes[gen] += gc_etime; GC_tot_copied += (ullong) copied; - GC_tot_scavd_copied += (ullong) scavd_copied; GC_tot_alloc += (ullong) alloc; GC_tot_time += gc_time; GCe_tot_time += gc_etime; @@ -316,6 +353,18 @@ stat_endGC (lnat alloc, lnat live, lnat copied, debugBelch("\b\b\b \b\b\b"); rub_bell = 0; } + +#if USE_PAPI + if(papi_is_reporting) { + /* Switch to counting mutator events */ + if (gen == 0) { + papi_stop_gc0_count(); + } else { + papi_stop_gc1_count(); + } + papi_start_mutator_count(); + } +#endif } /* ----------------------------------------------------------------------------- @@ -401,6 +450,52 @@ stat_endHeapCensus(void) were left unused when the heap-check failed. -------------------------------------------------------------------------- */ +#ifdef DEBUG +#define TICK_VAR(arity) \ + extern StgInt SLOW_CALLS_##arity; \ + extern StgInt RIGHT_ARITY_##arity; \ + extern StgInt TAGGED_PTR_##arity; + +#define TICK_VAR_INI(arity) \ + StgInt SLOW_CALLS_##arity = 1; \ + StgInt RIGHT_ARITY_##arity = 1; \ + StgInt TAGGED_PTR_##arity = 0; + +extern StgInt TOTAL_CALLS; + +TICK_VAR(1) +TICK_VAR(2) + +TICK_VAR_INI(1) +TICK_VAR_INI(2) + +StgInt TOTAL_CALLS=1; +#endif + +/* Report the value of a counter */ +#define REPORT(counter) \ + { \ + ullong_format_string(counter,temp,rtsTrue/*commas*/); \ + statsPrintf(" (" #counter ") : %s\n",temp); \ + } + +/* Report the value of a counter as a percentage of another counter */ +#define REPORT_PCT(counter,countertot) \ + statsPrintf(" (" #counter ") %% of (" #countertot ") : %.1f%%\n", \ + counter*100.0/countertot) + +#define TICK_PRINT(arity) \ + REPORT(SLOW_CALLS_##arity); \ + REPORT_PCT(RIGHT_ARITY_##arity,SLOW_CALLS_##arity); \ + REPORT_PCT(TAGGED_PTR_##arity,RIGHT_ARITY_##arity); \ + REPORT(RIGHT_ARITY_##arity); \ + REPORT(TAGGED_PTR_##arity) + +#define TICK_PRINT_TOT(arity) \ + statsPrintf(" (SLOW_CALLS_" #arity ") %% of (TOTAL_CALLS) : %.1f%%\n", \ + SLOW_CALLS_##arity * 100.0/TOTAL_CALLS) + + void stat_exit(int alloc) { @@ -436,12 +531,8 @@ stat_exit(int alloc) ullong_format_string(GC_tot_copied*sizeof(W_), temp, rtsTrue/*commas*/); - statsPrintf("%11s bytes copied during GC (scavenged)\n", temp); + statsPrintf("%11s bytes copied during GC\n", temp); - ullong_format_string(GC_tot_scavd_copied*sizeof(W_), - temp, rtsTrue/*commas*/); - statsPrintf("%11s bytes copied during GC (not scavenged)\n", temp); - if ( ResidencySamples > 0 ) { ullong_format_string(MaxResidency*sizeof(W_), temp, rtsTrue/*commas*/); @@ -452,12 +543,13 @@ stat_exit(int alloc) /* Print garbage collections in each gen */ for (g = 0; g < RtsFlags.GcFlags.generations; g++) { - statsPrintf("%11d collections in generation %d (%6.2fs)\n", + statsPrintf("%11d collections in generation %d, %6.2fs, %6.2fs elapsed\n", generations[g].collections, g, - TICK_TO_DBL(GC_coll_times[g])); + TICK_TO_DBL(GC_coll_times[g]), + TICK_TO_DBL(GC_coll_etimes[g])); } - statsPrintf("\n%11ld Mb total memory in use\n\n", + statsPrintf("\n%11ld MB total memory in use\n\n", mblocks_allocated * MBLOCK_SIZE / (1024 * 1024)); #if defined(THREADED_RTS) @@ -517,6 +609,18 @@ stat_exit(int alloc) TICK_TO_DBL(time - GC_tot_time - PROF_VAL(RP_tot_time + HC_tot_time) - InitUserTime) * 100 / TICK_TO_DBL(etime)); + + /* + TICK_PRINT(1); + TICK_PRINT(2); + REPORT(TOTAL_CALLS); + TICK_PRINT_TOT(1); + TICK_PRINT_TOT(2); + */ + +#if USE_PAPI + papi_stats_report(); +#endif } if (RtsFlags.GcFlags.giveStats == ONELINE_GC_STATS) { @@ -534,12 +638,33 @@ stat_exit(int alloc) TICK_TO_DBL(GC_tot_time), TICK_TO_DBL(GCe_tot_time)); } +#if defined(THREADED_RTS) && defined(PROF_SPIN) + { + nat g, s; + + statsPrintf("recordMutableGen_sync: %"FMT_Word64"\n", recordMutableGen_sync.spin); + statsPrintf("gc_alloc_block_sync: %"FMT_Word64"\n", gc_alloc_block_sync.spin); + statsPrintf("static_objects_sync: %"FMT_Word64"\n", static_objects_sync.spin); + statsPrintf("whitehole_spin: %"FMT_Word64"\n", whitehole_spin); + for (g = 0; g < RtsFlags.GcFlags.generations; g++) { + for (s = 0; s < generations[g].n_steps; s++) { + statsPrintf("gen[%d].steps[%d].sync_todo: %"FMT_Word64"\n", g, s, generations[g].steps[s].sync_todo.spin); + statsPrintf("gen[%d].steps[%d].sync_large_objects: %"FMT_Word64"\n", g, s, generations[g].steps[s].sync_large_objects.spin); + } + } + } +#endif + statsFlush(); statsClose(); } + if (GC_coll_times) stgFree(GC_coll_times); GC_coll_times = NULL; + if (GC_coll_etimes) + stgFree(GC_coll_etimes); + GC_coll_etimes = NULL; } /* ----------------------------------------------------------------------------- @@ -547,7 +672,6 @@ stat_exit(int alloc) Produce some detailed info on the state of the generational GC. -------------------------------------------------------------------------- */ -#ifdef DEBUG void statDescribeGens(void) { @@ -575,14 +699,10 @@ statDescribeGens(void) for (bd = step->large_objects, lge = 0; bd; bd = bd->link) { lge++; } - live = step->n_large_blocks * BLOCK_SIZE; - bd = step->blocks; // This live figure will be slightly less that the "live" figure // given by +RTS -Sstderr, because we take don't count the // slop at the end of each block. - for (; bd; bd = bd->link) { - live += (bd->free - bd->start) * sizeof(W_); - } + live += countOccupied(step->blocks) + countOccupied(step->large_objects); if (s != 0) { debugBelch("%36s",""); } @@ -592,7 +712,6 @@ statDescribeGens(void) } debugBelch("\n"); } -#endif /* ----------------------------------------------------------------------------- Stats available via a programmatic interface, so eg. GHCi can time @@ -606,7 +725,7 @@ extern HsInt64 getAllocations( void ) Dumping stuff in the stats file, or via the debug message interface -------------------------------------------------------------------------- */ -static void +void statsPrintf( char *s, ... ) { FILE *sf = RtsFlags.GcFlags.statsFile;