make the smp way RTS-only, normal libraries now work with -smp
[ghc-hetmet.git] / ghc / rts / ProfHeap.c
index 19beb9c..a50f2f0 100644 (file)
@@ -1,5 +1,4 @@
 /* -----------------------------------------------------------------------------
- * $Id: ProfHeap.c,v 1.47 2003/08/22 22:24:12 sof Exp $
  *
  * (c) The GHC Team, 1998-2003
  *
@@ -24,7 +23,6 @@
 #include "ProfHeap.h"
 #include "Stats.h"
 #include "Hash.h"
-#include "StrHash.h"
 #include "RetainerProfile.h"
 #include "LdvProfile.h"
 #include "Arena.h"
@@ -32,6 +30,7 @@
 
 #include <string.h>
 #include <stdlib.h>
+#include <math.h>
 
 /* -----------------------------------------------------------------------------
  * era stores the current time period.  It is the same as the
@@ -46,7 +45,7 @@
  * When era reaches max_era, the profiling stops because a closure can
  * store only up to (max_era - 1) as its creation or last use time.
  * -------------------------------------------------------------------------- */
-nat era;
+unsigned int era;
 static nat max_era;
 
 /* -----------------------------------------------------------------------------
@@ -72,7 +71,7 @@ typedef struct _counter {
     struct _counter *next;
 } counter;
 
-static inline void
+STATIC_INLINE void
 initLDVCtr( counter *ctr )
 {
     ctr->c.ldv.prim = 0;
@@ -150,17 +149,17 @@ static char *type_names[] = {
     , "STOP_FRAME"
 
     , "BLACKHOLE"
-    , "BLACKHOLE_BQ"
     , "MVAR"
 
     , "ARR_WORDS"
 
-    , "MUT_ARR_PTRS"
+    , "MUT_ARR_PTRS_CLEAN"
+    , "MUT_ARR_PTRS_DIRTY"
     , "MUT_ARR_PTRS_FROZEN"
-    , "MUT_VAR"
+    , "MUT_VAR_CLEAN"
+    , "MUT_VAR_DIRTY"
 
     , "WEAK"
-    , "FOREIGN"
   
     , "TSO"
 
@@ -177,7 +176,7 @@ static char *type_names[] = {
  * the band to which this closure's heap space is attributed in the
  * heap profile.
  * ------------------------------------------------------------------------- */
-static inline void *
+STATIC_INLINE void *
 closureIdentity( StgClosure *p )
 {
     switch (RtsFlags.ProfFlags.doHeapProfile) {
@@ -215,20 +214,20 @@ closureIdentity( StgClosure *p )
  * Profiling type predicates
  * ----------------------------------------------------------------------- */
 #ifdef PROFILING
-static inline rtsBool
+STATIC_INLINE rtsBool
 doingLDVProfiling( void )
 {
     return (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV 
            || RtsFlags.ProfFlags.bioSelector != NULL);
 }
 
-static inline rtsBool
+STATIC_INLINE rtsBool
 doingRetainerProfiling( void )
 {
     return (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER
            || RtsFlags.ProfFlags.retainerSelector != NULL);
 }
-#endif // PROFILING
+#endif /* PROFILING */
 
 // Precesses a closure 'c' being destroyed whose size is 'size'.
 // Make sure that LDV_recordDead() is not invoked on 'inherently used' closures
@@ -249,6 +248,7 @@ LDV_recordDead( StgClosure *c, nat size )
 
     if (era > 0 && closureSatisfiesConstraints(c)) {
        size -= sizeofW(StgProfHeader);
+       ASSERT(LDVW(c) != 0);
        if ((LDVW((c)) & LDV_STATE_MASK) == LDV_STATE_CREATE) {
            t = (LDVW((c)) & LDV_CREATE_MASK) >> LDV_SHIFT;
            if (t < era) {
@@ -304,7 +304,7 @@ LDV_recordDead( StgClosure *c, nat size )
 /* --------------------------------------------------------------------------
  * Initialize censuses[era];
  * ----------------------------------------------------------------------- */
-static inline void
+STATIC_INLINE void
 initEra(Census *census)
 {
     census->hash  = allocHashTable();
@@ -330,7 +330,7 @@ nextEra( void )
        era++;
 
        if (era == max_era) {
-           prog_belch("maximum number of censuses reached; use +RTS -i to reduce");
+           errorBelch("maximum number of censuses reached; use +RTS -i to reduce");
            stg_exit(EXIT_FAILURE);
        }
        
@@ -340,7 +340,7 @@ nextEra( void )
                                       "nextEra");
        }
     }
-#endif // PROFILING
+#endif /* PROFILING */
 
     initEra( &censuses[era] );
 }
@@ -351,6 +351,7 @@ nextEra( void )
 
 #ifdef DEBUG_HEAP_PROF
 FILE *hp_file;
+static char *hp_filename;
 
 void initProfiling1( void )
 {
@@ -358,6 +359,20 @@ void initProfiling1( void )
 
 void initProfiling2( void )
 {
+  if (RtsFlags.ProfFlags.doHeapProfile) {
+    /* Initialise the log file name */
+    hp_filename = stgMallocBytes(strlen(prog_name) + 6, "hpFileName");
+    sprintf(hp_filename, "%s.hp", prog_name);
+    
+    /* open the log file */
+    if ((hp_file = fopen(hp_filename, "w")) == NULL) {
+      debugBelch("Can't open profiling report file %s\n", 
+             hp_filename);
+      RtsFlags.ProfFlags.doHeapProfile = 0;
+      return;
+    }
+  }
+  
   initHeapProfiling();
 }
 
@@ -367,6 +382,16 @@ void endProfiling( void )
 }
 #endif /* DEBUG_HEAP_PROF */
 
+static void
+printSample(rtsBool beginSample, StgDouble sampleValue)
+{
+    StgDouble fractionalPart, integralPart;
+    fractionalPart = modf(sampleValue, &integralPart);
+    fprintf(hp_file, "%s %d.%02d\n",
+            (beginSample ? "BEGIN_SAMPLE" : "END_SAMPLE"),
+            (int)integralPart, (int)(fractionalPart * 100));
+}
+
 /* --------------------------------------------------------------------------
  * Initialize the heap profilier
  * ----------------------------------------------------------------------- */
@@ -379,8 +404,8 @@ initHeapProfiling(void)
 
 #ifdef PROFILING
     if (doingLDVProfiling() && doingRetainerProfiling()) {
-       prog_belch("cannot mix -hb and -hr");
-       stg_exit(1);
+       errorBelch("cannot mix -hb and -hr");
+       stg_exit(EXIT_FAILURE);
     }
 #endif
 
@@ -407,6 +432,7 @@ initHeapProfiling(void)
 
     initEra( &censuses[era] );
 
+    /* initProfilingLogFile(); */
     fprintf(hp_file, "JOB \"%s", prog_name);
 
 #ifdef PROFILING
@@ -414,10 +440,9 @@ initHeapProfiling(void)
        int count;
        for(count = 1; count < prog_argc; count++)
            fprintf(hp_file, " %s", prog_argv[count]);
-       fprintf(hp_file, " +RTS ");
+       fprintf(hp_file, " +RTS");
        for(count = 0; count < rts_argc; count++)
-           fprintf(hp_file, "%s ", rts_argv[count]);
-       fprintf(hp_file, "\n");
+           fprintf(hp_file, " %s", rts_argv[count]);
     }
 #endif /* PROFILING */
 
@@ -428,8 +453,8 @@ initHeapProfiling(void)
     fprintf(hp_file, "SAMPLE_UNIT \"seconds\"\n");
     fprintf(hp_file, "VALUE_UNIT \"bytes\"\n");
 
-    fprintf(hp_file, "BEGIN_SAMPLE 0.00\n");
-    fprintf(hp_file, "END_SAMPLE 0.00\n");
+    printSample(rtsTrue, 0);
+    printSample(rtsFalse, 0);
 
 #ifdef DEBUG_HEAP_PROF
     DEBUG_LoadSymbols(prog_name);
@@ -471,8 +496,8 @@ endHeapProfiling(void)
 #endif
 
     seconds = mut_user_time();
-    fprintf(hp_file, "BEGIN_SAMPLE %0.2f\n", seconds);
-    fprintf(hp_file, "END_SAMPLE %0.2f\n", seconds);
+    printSample(rtsTrue, seconds);
+    printSample(rtsFalse, seconds);
     fclose(hp_file);
 }
 
@@ -495,8 +520,6 @@ static void
 fprint_ccs(FILE *fp, CostCentreStack *ccs, nat max_length)
 {
     char buf[max_length+1], *p, *buf_end;
-    nat next_offset = 0;
-    nat written;
 
     // MAIN on its own gets printed as "MAIN", otherwise we ignore MAIN.
     if (ccs == CCS_MAIN) {
@@ -504,7 +527,7 @@ fprint_ccs(FILE *fp, CostCentreStack *ccs, nat max_length)
        return;
     }
 
-    fprintf(fp, "(%d)", ccs->ccsID);
+    fprintf(fp, "(%ld)", ccs->ccsID);
 
     p = buf;
     buf_end = buf + max_length + 1;
@@ -528,19 +551,17 @@ fprint_ccs(FILE *fp, CostCentreStack *ccs, nat max_length)
        if (p >= buf_end) {
            sprintf(buf+max_length-4, "...");
            break;
-       } else {
-           next_offset += written;
        }
     }
     fprintf(fp, "%s", buf);
 }
-#endif // PROFILING
+#endif /* PROFILING */
 
 rtsBool
 strMatchesSelector( char* str, char* sel )
 {
    char* p;
-   // fprintf(stderr, "str_matches_selector %s %s\n", str, sel);
+   // debugBelch("str_matches_selector %s %s\n", str, sel);
    while (1) {
        // Compare str against wherever we've got to in sel.
        p = str;
@@ -568,6 +589,7 @@ rtsBool
 closureSatisfiesConstraints( StgClosure* p )
 {
 #ifdef DEBUG_HEAP_PROF
+    (void)p;   /* keep gcc -Wall happy */
     return rtsTrue;
 #else
    rtsBool b;
@@ -668,8 +690,8 @@ aggregateCensusInfo( void )
                // totals *must* be zero.
                ASSERT(c->c.ldv.void_total == 0 && c->c.ldv.drag_total == 0);
 
-               // fprintCCS(stderr,c->identity);
-               // fprintf(stderr," census=%d void_total=%d drag_total=%d\n",
+               // debugCCS(c->identity);
+               // debugBelch(" census=%d void_total=%d drag_total=%d\n",
                //         t, c->c.ldv.void_total, c->c.ldv.drag_total);
            } else {
                d->c.ldv.void_total += c->c.ldv.void_total;
@@ -714,20 +736,20 @@ dumpCensus( Census *census )
     counter *ctr;
     int count;
 
-    fprintf(hp_file, "BEGIN_SAMPLE %0.2f\n", census->time);
+    printSample(rtsTrue, census->time);
 
 #ifdef PROFILING
     if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV) {
-       fprintf(hp_file, "VOID\t%u\n", census->void_total * sizeof(W_));
-       fprintf(hp_file, "LAG\t%u\n", 
-               (census->not_used - census->void_total) * sizeof(W_));
-       fprintf(hp_file, "USE\t%u\n", 
-               (census->used - census->drag_total) * sizeof(W_));
-       fprintf(hp_file, "INHERENT_USE\t%u\n", 
-               census->prim * sizeof(W_));
-       fprintf(hp_file, "DRAG\t%u\n", census->drag_total *
-               sizeof(W_));
-       fprintf(hp_file, "END_SAMPLE %0.2f\n", census->time);
+      fprintf(hp_file, "VOID\t%lu\n", (unsigned long)(census->void_total) * sizeof(W_));
+       fprintf(hp_file, "LAG\t%lu\n", 
+               (unsigned long)(census->not_used - census->void_total) * sizeof(W_));
+       fprintf(hp_file, "USE\t%lu\n", 
+               (unsigned long)(census->used - census->drag_total) * sizeof(W_));
+       fprintf(hp_file, "INHERENT_USE\t%lu\n", 
+               (unsigned long)(census->prim) * sizeof(W_));
+       fprintf(hp_file, "DRAG\t%lu\n",
+               (unsigned long)(census->drag_total) * sizeof(W_));
+       printSample(rtsFalse, census->time);
        return;
     }
 #endif
@@ -804,10 +826,10 @@ dumpCensus( Census *census )
        }
 #endif
 
-       fprintf(hp_file, "\t%d\n", count * sizeof(W_));
+       fprintf(hp_file, "\t%lu\n", (unsigned long)count * sizeof(W_));
     }
 
-    fprintf(hp_file, "END_SAMPLE %0.2f\n", census->time);
+    printSample(rtsFalse, census->time);
 }
 
 /* -----------------------------------------------------------------------------
@@ -825,6 +847,15 @@ heapCensusChain( Census *census, bdescr *bd )
     rtsBool prim;
 
     for (; bd != NULL; bd = bd->link) {
+
+       // HACK: ignore pinned blocks, because they contain gaps.
+       // It's not clear exactly what we'd like to do here, since we
+       // can't tell which objects in the block are actually alive.
+       // Perhaps the whole block should be counted as SYSTEM memory.
+       if (bd->flags & BF_PINNED) {
+           continue;
+       }
+
        p = bd->start;
        while (p < bd->free) {
            info = get_itbl((StgClosure *)p);
@@ -832,9 +863,24 @@ heapCensusChain( Census *census, bdescr *bd )
            
            switch (info->type) {
 
+           case THUNK:
+               size = thunk_sizeW_fromITBL(info);
+               break;
+
+           case THUNK_1_1:
+           case THUNK_0_2:
+           case THUNK_2_0:
+               size = sizeofW(StgThunkHeader) + 2;
+               break;
+
+           case THUNK_1_0:
+           case THUNK_0_1:
+           case THUNK_SELECTOR:
+               size = sizeofW(StgThunkHeader) + 1;
+               break;
+
            case CONSTR:
            case FUN:
-           case THUNK:
            case IND_PERM:
            case IND_OLDGEN:
            case IND_OLDGEN_PERM:
@@ -842,7 +888,6 @@ heapCensusChain( Census *census, bdescr *bd )
            case SE_CAF_BLACKHOLE:
            case SE_BLACKHOLE:
            case BLACKHOLE:
-           case BLACKHOLE_BQ:
            case CONSTR_INTLIKE:
            case CONSTR_CHARLIKE:
            case FUN_1_0:
@@ -850,9 +895,6 @@ heapCensusChain( Census *census, bdescr *bd )
            case FUN_1_1:
            case FUN_0_2:
            case FUN_2_0:
-           case THUNK_1_1:
-           case THUNK_0_2:
-           case THUNK_2_0:
            case CONSTR_1_0:
            case CONSTR_0_1:
            case CONSTR_1_1:
@@ -860,7 +902,18 @@ heapCensusChain( Census *census, bdescr *bd )
            case CONSTR_2_0:
                size = sizeW_fromITBL(info);
                break;
-               
+
+           case IND:
+               // Special case/Delicate Hack: INDs don't normally
+               // appear, since we're doing this heap census right
+               // after GC.  However, GarbageCollect() also does
+               // resurrectThreads(), which can update some
+               // blackholes when it calls raiseAsync() on the
+               // resurrected threads.  So we know that any IND will
+               // be the size of a BLACKHOLE.
+               size = BLACKHOLE_sizeW();
+               break;
+
            case BCO:
                prim = rtsTrue;
                size = bco_sizeW((StgBCO *)p);
@@ -868,21 +921,17 @@ heapCensusChain( Census *census, bdescr *bd )
 
            case MVAR:
            case WEAK:
-           case FOREIGN:
            case STABLE_NAME:
-           case MUT_VAR:
-           case MUT_CONS:
+           case MUT_VAR_CLEAN:
+           case MUT_VAR_DIRTY:
                prim = rtsTrue;
                size = sizeW_fromITBL(info);
                break;
 
-           case THUNK_1_0:             /* ToDo - shouldn't be here */
-           case THUNK_0_1:             /* "  ditto  " */
-           case THUNK_SELECTOR:
-               size = sizeofW(StgHeader) + MIN_UPD_SIZE;
+           case AP:
+               size = ap_sizeW((StgAP *)p);
                break;
 
-           case AP:
            case PAP:
                size = pap_sizeW((StgPAP *)p);
                break;
@@ -896,8 +945,10 @@ heapCensusChain( Census *census, bdescr *bd )
                size = arr_words_sizeW(stgCast(StgArrWords*,p));
                break;
                
-           case MUT_ARR_PTRS:
+           case MUT_ARR_PTRS_CLEAN:
+           case MUT_ARR_PTRS_DIRTY:
            case MUT_ARR_PTRS_FROZEN:
+           case MUT_ARR_PTRS_FROZEN0:
                prim = rtsTrue;
                size = mut_arr_ptrs_sizeW((StgMutArrPtrs *)p);
                break;
@@ -918,8 +969,28 @@ heapCensusChain( Census *census, bdescr *bd )
                }
 #endif
 
+           case TREC_HEADER: 
+               prim = rtsTrue;
+               size = sizeofW(StgTRecHeader);
+               break;
+
+           case TVAR_WAIT_QUEUE:
+               prim = rtsTrue;
+               size = sizeofW(StgTVarWaitQueue);
+               break;
+               
+           case TVAR:
+               prim = rtsTrue;
+               size = sizeofW(StgTVar);
+               break;
+               
+           case TREC_CHUNK:
+               prim = rtsTrue;
+               size = sizeofW(StgTRecChunk);
+               break;
+
            default:
-               barf("heapCensus");
+               barf("heapCensus, unknown object: %d", info->type);
            }
            
            identity = NULL;
@@ -1021,7 +1092,7 @@ heapCensus( void )
 
   // Now traverse the heap in each generation/step.
   if (RtsFlags.GcFlags.generations == 1) {
-      heapCensusChain( census, g0s0->to_blocks );
+      heapCensusChain( census, g0s0->blocks );
   } else {
       for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
          for (s = 0; s < generations[g].n_steps; s++) {