import gdb-1999-12-06 snapshot

1 files changed, 258 insertions, 165 deletions

diff --git a/gdb/bcache.c b/gdb/bcache.c
index 97485dd..8ac3305 100644
--- a/gdb/bcache.c
+++ b/gdb/bcache.c
@@ -1,6 +1,7 @@
 /* Implement a cached obstack.
-   Written by Fred Fish (fnf@cygnus.com)
-   Copyright 1995, 1998 Free Software Foundation, Inc.
+   Written by Fred Fish <fnf@cygnus.com>
+   Rewritten by Jim Blandy <jimb@cygnus.com>
+   Copyright 1999 Free Software Foundation, Inc.
 
    This file is part of GDB.
 
@@ -19,198 +20,290 @@
    Foundation, Inc., 59 Temple Place - Suite 330,
    Boston, MA 02111-1307, USA.  */
 
+#include <stddef.h>
+#include <stdlib.h>
+
 #include "defs.h"
 #include "obstack.h"
 #include "bcache.h"
 #include "gdb_string.h"		/* For memcpy declaration */
 
-/* Prototypes for local functions. */
 
-static unsigned int hash PARAMS ((void *, int));
+
+/* The hash function.  */
 
-static void *lookup_cache PARAMS ((void *, int, int, struct bcache *));
+unsigned long
+hash (void *addr, int length)
+{
+  /* If it's a short string, hash on every character.  Otherwise, sample
+     characters from throughout the string.  */
+  if (length <= 64)
+    {
+      char *byte = addr;
+      unsigned long h = 0;
+      int i;
 
-/* FIXME:  Incredibly simplistic hash generator.  Probably way too expensive
-   (consider long strings) and unlikely to have good distribution across hash
-   values for typical input. */
+      for (i = 0; i < length; i++)
+	h = h * 65793 ^ (h >> (sizeof (h) * 8 - 6)) ^ byte[i];
 
-static unsigned int
-hash (bytes, count)
-     void *bytes;
-     int count;
-{
-  unsigned int len;
-  unsigned long hashval;
-  unsigned int c;
-  const unsigned char *data = bytes;
-
-  hashval = 0;
-  len = 0;
-  while (count-- > 0)
+      return h;
+    }
+  else
     {
-      c = *data++;
-      hashval += c + (c << 17);
-      hashval ^= hashval >> 2;
-      ++len;
+      char *byte = addr;
+      int n, i;
+      unsigned long h = 0;
+
+      for (n = i = 0; n < 64; n++)
+	{
+	  h = h * 65793 + (h >> (sizeof (h) * 8 - 6)) + byte[i];
+	  i = h % length;
+	}
+
+      return h;
     }
-  hashval += len + (len << 17);
-  hashval ^= hashval >> 2;
-  return (hashval % BCACHE_HASHSIZE);
 }
 
-static void *
-lookup_cache (bytes, count, hashval, bcachep)
-     void *bytes;
-     int count;
-     int hashval;
-     struct bcache *bcachep;
+
+/* Growing the bcache's hash table.  */
+
+/* If the average chain length grows beyond this, then we want to
+   resize our hash table.  */
+#define CHAIN_LENGTH_THRESHOLD (5)
+
+static void
+expand_hash_table (struct bcache *bcache)
 {
-  void *location = NULL;
-  struct hashlink **hashtablep;
-  struct hashlink *linkp;
+  /* A table of good hash table sizes.  Whenever we grow, we pick the
+     next larger size from this table.  sizes[i] is close to 1 << (i+10),
+     so we roughly double the table size each time.  After we fall off 
+     the end of this table, we just double.  Don't laugh --- there have
+     been executables sighted with a gigabyte of debug info.  */
+  static unsigned long sizes[] = { 
+    1021, 2053, 4099, 8191, 16381, 32771,
+    65537, 131071, 262144, 524287, 1048573, 2097143,
+    4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
+    268435459, 536870923, 1073741827, 2147483659UL
+  };
+  int new_num_buckets;
+  struct bstring **new_buckets;
+  int i;
+
+  /* Find the next size.  */
+  for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
+    if (sizes[i] > bcache->num_buckets)
+      {
+	new_num_buckets = sizes[i];
+	break;
+      }
+  if (i >= (sizeof (sizes) / sizeof (sizes[0])))
+    new_num_buckets = bcache->num_buckets * 2;
+
+  /* Allocate the new table.  */
+  {
+    size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
+    new_buckets = (struct bstring **) xmalloc (new_size);
+    memset (new_buckets, 0, new_size);
+
+    bcache->structure_size -= (bcache->num_buckets
+			       * sizeof (bcache->bucket[0]));
+    bcache->structure_size += new_size;
+  }
 
-  hashtablep = bcachep->indextable[count];
-  if (hashtablep != NULL)
+  /* Rehash all existing strings.  */
+  for (i = 0; i < bcache->num_buckets; i++)
     {
-      linkp = hashtablep[hashval];
-      while (linkp != NULL)
+      struct bstring *s, *next;
+
+      for (s = bcache->bucket[i]; s; s = next)
 	{
-	  if (memcmp (BCACHE_DATA (linkp), bytes, count) == 0)
-	    {
-	      location = BCACHE_DATA (linkp);
-	      break;
-	    }
-	  linkp = linkp->next;
+	  struct bstring **new_bucket;
+	  next = s->next;
+
+	  new_bucket = &new_buckets[(hash (&s->d.data, s->length)
+				     % new_num_buckets)];
+	  s->next = *new_bucket;
+	  *new_bucket = s;
 	}
     }
-  return (location);
+
+  /* Plug in the new table.  */
+  if (bcache->bucket)
+    free (bcache->bucket);
+  bcache->bucket = new_buckets;
+  bcache->num_buckets = new_num_buckets;
 }
 
+
+/* Looking up things in the bcache.  */
+
+/* The number of bytes needed to allocate a struct bstring whose data
+   is N bytes long.  */
+#define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
+
+/* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
+   never seen those bytes before, add a copy of them to BCACHE.  In
+   either case, return a pointer to BCACHE's copy of that string.  */
 void *
-bcache (bytes, count, bcachep)
-     void *bytes;
-     int count;
-     struct bcache *bcachep;
+bcache (void *addr, int length, struct bcache *bcache)
 {
-  int hashval;
-  void *location;
-  struct hashlink *newlink;
-  struct hashlink **linkpp;
-  struct hashlink ***hashtablepp;
+  int hash_index;
+  struct bstring *s;
 
-  if (count >= BCACHE_MAXLENGTH)
-    {
-      /* Rare enough to just stash unique copies */
-      location = (void *) obstack_alloc (&bcachep->cache, count);
-      bcachep->cache_bytes += count;
-      memcpy (location, bytes, count);
-      bcachep->bcache_overflows++;
-    }
-  else
-    {
-      hashval = hash (bytes, count);
-      location = lookup_cache (bytes, count, hashval, bcachep);
-      if (location != NULL)
-	{
-	  bcachep->cache_savings += count;
-	  bcachep->cache_hits++;
-	}
-      else
-	{
-	  bcachep->cache_misses++;
-	  hashtablepp = &bcachep->indextable[count];
-	  if (*hashtablepp == NULL)
-	    {
-	      *hashtablepp = (struct hashlink **)
-		obstack_alloc (&bcachep->cache, BCACHE_HASHSIZE * sizeof (struct hashlink *));
-	      bcachep->cache_bytes += BCACHE_HASHSIZE * sizeof (struct hashlink *);
-	      memset (*hashtablepp, 0, BCACHE_HASHSIZE * sizeof (struct hashlink *));
-	    }
-	  linkpp = &(*hashtablepp)[hashval];
-	  newlink = (struct hashlink *)
-	    obstack_alloc (&bcachep->cache, BCACHE_DATA_ALIGNMENT + count);
-	  bcachep->cache_bytes += BCACHE_DATA_ALIGNMENT + count;
-	  memcpy (BCACHE_DATA (newlink), bytes, count);
-	  newlink->next = *linkpp;
-	  *linkpp = newlink;
-	  location = BCACHE_DATA (newlink);
-	}
-    }
-  return (location);
+  /* If our average chain length is too high, expand the hash table.  */
+  if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
+    expand_hash_table (bcache);
+
+  bcache->total_count++;
+  bcache->total_size += length;
+
+  hash_index = hash (addr, length) % bcache->num_buckets;
+
+  /* Search the hash bucket for a string identical to the caller's.  */
+  for (s = bcache->bucket[hash_index]; s; s = s->next)
+    if (s->length == length
+	&& ! memcmp (&s->d.data, addr, length))
+      return &s->d.data;
+
+  /* The user's string isn't in the list.  Insert it after *ps.  */
+  {
+    struct bstring *new
+      = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
+    memcpy (&new->d.data, addr, length);
+    new->length = length;
+    new->next = bcache->bucket[hash_index];
+    bcache->bucket[hash_index] = new;
+
+    bcache->unique_count++;
+    bcache->unique_size += length;
+    bcache->structure_size += BSTRING_SIZE (length);
+
+    return &new->d.data;
+  }
 }
 
+
+/* Freeing bcaches.  */
+
+/* Free all the storage associated with BCACHE.  */
 void
-print_bcache_statistics (bcachep, id)
-     struct bcache *bcachep;
-     char *id;
+free_bcache (struct bcache *bcache)
 {
-  struct hashlink **hashtablep;
-  struct hashlink *linkp;
-  int tidx, tcount, hidx, hcount, lcount, lmax, temp, lmaxt, lmaxh;
+  obstack_free (&bcache->cache, 0);
+  free (bcache->bucket);
 
-  for (lmax = lcount = tcount = hcount = tidx = 0; tidx < BCACHE_MAXLENGTH; tidx++)
-    {
-      hashtablep = bcachep->indextable[tidx];
-      if (hashtablep != NULL)
-	{
-	  tcount++;
-	  for (hidx = 0; hidx < BCACHE_HASHSIZE; hidx++)
-	    {
-	      linkp = hashtablep[hidx];
-	      if (linkp != NULL)
-		{
-		  hcount++;
-		  for (temp = 0; linkp != NULL; linkp = linkp->next)
-		    {
-		      lcount++;
-		      temp++;
-		    }
-		  if (temp > lmax)
-		    {
-		      lmax = temp;
-		      lmaxt = tidx;
-		      lmaxh = hidx;
-		    }
-		}
-	    }
-	}
-    }
-  printf_filtered ("  Cached '%s' statistics:\n", id);
-  printf_filtered ("    Cache hits: %d\n", bcachep->cache_hits);
-  printf_filtered ("    Cache misses: %d\n", bcachep->cache_misses);
-  printf_filtered ("    Cache hit ratio: ");
-  if (bcachep->cache_hits + bcachep->cache_misses > 0)
-    {
-      printf_filtered ("%d%%\n", ((bcachep->cache_hits) * 100) /
-		       (bcachep->cache_hits + bcachep->cache_misses));
-    }
-  else
-    {
-      printf_filtered ("(not applicable)\n");
-    }
-  printf_filtered ("    Space used for caching: %d\n", bcachep->cache_bytes);
-  printf_filtered ("    Space saved by cache hits: %d\n", bcachep->cache_savings);
-  printf_filtered ("    Number of bcache overflows: %d\n", bcachep->bcache_overflows);
-  printf_filtered ("    Number of index buckets used: %d\n", tcount);
-  printf_filtered ("    Number of hash table buckets used: %d\n", hcount);
-  printf_filtered ("    Number of chained items: %d\n", lcount);
-  printf_filtered ("    Average hash table population: ");
-  if (tcount > 0)
-    {
-      printf_filtered ("%d%%\n", (hcount * 100) / (tcount * BCACHE_HASHSIZE));
-    }
+  /* This isn't necessary, but at least the bcache is always in a
+     consistent state.  */
+  memset (bcache, 0, sizeof (*bcache));
+}
+
+
+
+/* Printing statistics.  */
+
+static int
+compare_ints (const void *ap, const void *bp)
+{
+  /* Because we know we're comparing two ints which are positive,
+     there's no danger of overflow here.  */
+  return * (int *) ap - * (int *) bp;
+}
+
+
+static void
+print_percentage (int portion, int total)
+{
+  if (total == 0)
+    printf_filtered ("(not applicable)\n");
   else
-    {
-      printf_filtered ("(not applicable)\n");
-    }
-  printf_filtered ("    Average chain length ");
-  if (hcount > 0)
-    {
-      printf_filtered ("%d\n", lcount / hcount);
-    }
+    printf_filtered ("%3d%%\n", portion * 100 / total);
+}
+
+
+/* Print statistics on BCACHE's memory usage and efficacity at
+   eliminating duplication.  NAME should describe the kind of data
+   BCACHE holds.  Statistics are printed using `printf_filtered' and
+   its ilk.  */
+void
+print_bcache_statistics (struct bcache *c, char *type)
+{
+  int occupied_buckets;
+  int max_chain_length;
+  int median_chain_length;
+
+  /* Count the number of occupied buckets, and measure chain lengths.  */
+  {
+    int b;
+    int *chain_length
+      = (int *) alloca (c->num_buckets * sizeof (*chain_length));
+
+    occupied_buckets = 0;
+
+    for (b = 0; b < c->num_buckets; b++)
+      {
+	struct bstring *s = c->bucket[b];
+
+	chain_length[b] = 0;
+
+	if (s)
+	  {
+	    occupied_buckets++;
+	    
+	    while (s)
+	      {
+		chain_length[b]++;
+		s = s->next;
+	      }
+	  }
+      }
+
+    /* To compute the median, we need the set of chain lengths sorted.  */
+    qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
+	   compare_ints);
+
+    if (c->num_buckets > 0)
+      {
+	max_chain_length = chain_length[c->num_buckets - 1];
+	median_chain_length = chain_length[c->num_buckets / 2];
+      }
+    else
+      {
+	max_chain_length = 0;
+	median_chain_length = 0;
+      }
+  }
+
+  printf_filtered ("  Cached '%s' statistics:\n", type);
+  printf_filtered ("    Total object count:  %ld\n", c->total_count);
+  printf_filtered ("    Unique object count: %ld\n", c->unique_count);
+  printf_filtered ("    Percentage of duplicates, by count: ");
+  print_percentage (c->total_count - c->unique_count, c->total_count);
+  printf_filtered ("\n");
+
+  printf_filtered ("    Total object size:   %ld\n", c->total_size);
+  printf_filtered ("    Unique object size:  %ld\n", c->unique_size);
+  printf_filtered ("    Percentage of duplicates, by size:  ");
+  print_percentage (c->total_size - c->unique_size, c->total_size);
+  printf_filtered ("\n");
+
+  printf_filtered ("    Total memory used by bcache, including overhead: %ld\n",
+		   c->structure_size);
+  printf_filtered ("    Percentage memory overhead: ");
+  print_percentage (c->structure_size - c->unique_size, c->unique_size);
+  printf_filtered ("    Net memory savings:         ");
+  print_percentage (c->total_size - c->structure_size, c->total_size);
+  printf_filtered ("\n");
+
+  printf_filtered ("    Hash table size:           %3d\n", c->num_buckets);
+  printf_filtered ("    Hash table population:     ");
+  print_percentage (occupied_buckets, c->num_buckets);
+  printf_filtered ("    Median hash chain length:  %3d\n",
+		   median_chain_length);
+  printf_filtered ("    Average hash chain length: ");
+  if (c->num_buckets > 0)
+    printf_filtered ("%3ld\n", c->unique_count / c->num_buckets);
   else
-    {
-      printf_filtered ("(not applicable)\n");
-    }
-  printf_filtered ("    Maximum chain length %d at %d:%d\n", lmax, lmaxt, lmaxh);
+    printf_filtered ("(not applicable)\n");
+  printf_filtered ("    Maximum hash chain length: %3d\n", max_chain_length);
+  printf_filtered ("\n");
 }