Optimize: Use a hash table in compute_partition.

This reduces the execution time of gperf on large inputs by ca. 30%.

* autogen.sh (GNULIB_MODULES): Add map-c++, hash-map.
* src/keyword.h: Include <stddef.h>.
(struct KeywordExt): Add fields _undetermined_chars,
_undetermined_chars_length, _undetermined_chars_hashcode.
* src/search.cc: Include gl_map.hh, gl_hash_map.h.
(Search::prepare_asso_values): Initialize the _undetermined_chars field.
(struct EquivalenceClass): Remove the fields _undetermined_chars,
_undetermined_chars_length.
(undetermined_equals, undetermined_hashcode): New functions.
(Search::compute_partition): Initialize the _undetermined_chars* fields
of all keywords. Use a hash map instead of a loop over the equivalence
classes.
(Search::find_good_asso_values): Deallocate the _undetermined_chars field.

ChangeLog

@@ -1,3 +1,21 @@
+2025-04-19  Bruno Haible  <bruno@clisp.org>
+
+	Optimize: Use a hash table in compute_partition.
+	This reduces the execution time of gperf on large inputs by ca. 30%.
+	* autogen.sh (GNULIB_MODULES): Add map-c++, hash-map.
+	* src/keyword.h: Include <stddef.h>.
+	(struct KeywordExt): Add fields _undetermined_chars,
+	_undetermined_chars_length, _undetermined_chars_hashcode.
+	* src/search.cc: Include gl_map.hh, gl_hash_map.h.
+	(Search::prepare_asso_values): Initialize the _undetermined_chars field.
+	(struct EquivalenceClass): Remove the fields _undetermined_chars,
+	_undetermined_chars_length.
+	(undetermined_equals, undetermined_hashcode): New functions.
+	(Search::compute_partition): Initialize the _undetermined_chars* fields
+	of all keywords. Use a hash map instead of a loop over the equivalence
+	classes.
+	(Search::find_good_asso_values): Deallocate the _undetermined_chars field.
+
 2025-04-19  Bruno Haible  <bruno@clisp.org>
 
 	Optimize: Make Bool_Array take less memory.

autogen.sh

@@ -66,6 +66,7 @@ if test $skip_gnulib = false; then
   GNULIB_MODULES='
     filename
     getopt-gnu
+    map-c++ hash-map
     read-file
     package-version
   '

lib/.gitignore

@@ -5,12 +5,19 @@
 /alloca.in.h
 /arg-nonnull.h
 /assert.in.h
+/attribute.h
+/basename-lgpl.c
+/basename-lgpl.h
 /c++defs.h
 /cloexec.c
 /cloexec.h
 /close.c
 /dup2.c
 /errno.in.h
+/error.c
+/error.in.h
+/exitfail.c
+/exitfail.h
 /fcntl.c
 /fcntl.in.h
 /fd-hook.c
@@ -31,9 +38,22 @@
 /getopt.in.h
 /getopt1.c
 /getopt_int.h
+/getprogname.c
+/getprogname.h
 /gettext.h
+/gl_anyhash1.h
+/gl_anyhash2.h
+/gl_anyhash_primes.h
+/gl_hash_map.c
+/gl_hash_map.h
+/gl_map.c
+/gl_map.h
+/gl_map.hh
+/gl_xmap.c
+/gl_xmap.h
 /idx.h
 /intprops-internal.h
+/intprops.h
 /inttypes.in.h
 /limits.in.h
 /lseek.c
@@ -51,6 +71,7 @@
 /read-file.c
 /read-file.h
 /realloc.c
+/size_max.h
 /stat-time.c
 /stat-time.h
 /stat-w32.c
@@ -65,6 +86,9 @@
 /stdio.in.h
 /stdlib.c
 /stdlib.in.h
+/strerror-override.c
+/strerror-override.h
+/strerror.c
 /string.in.h
 /sys_stat.in.h
 /sys_types.in.h
@@ -74,7 +98,11 @@
 /verify.h
 /warn-on-use.h
 /wchar.in.h
+/xalloc-die.c
 /xalloc-oversized.h
+/xalloc.h
+/xsize.c
+/xsize.h
 
 # Files generated by the autotools:
 /aclocal.m4

src/keyword.h

@@ -24,6 +24,8 @@
 #ifndef keyword_h
 #define keyword_h 1
 
+#include <stddef.h> /* defines size_t */
+
 /* Class defined in "positions.h".  */
 class Positions;
 
@@ -72,6 +74,13 @@ struct KeywordExt : public Keyword
   /* Deletes selchars.  */
   void                  delete_selchars ();
 
+  /* Data members used by the algorithm, specifically compute_partition.  */
+  /* The undetermined selected characters for this keyword, as a
+     canonically reordered multiset.  */
+  unsigned int *        _undetermined_chars;
+  unsigned int          _undetermined_chars_length;
+  size_t                _undetermined_chars_hashcode;
+
   /* Data members used by the algorithm.  */
   int                   _hash_value; /* Hash value for the keyword.  */
 

src/search.cc

@@ -28,7 +28,9 @@
 #include <string.h> /* declares memset(), memcmp() */
 #include <time.h> /* declares time() */
 #include <math.h> /* declares exp() */
-#include <limits.h> /* defines INT_MIN, INT_MAX, UINT_MAX */
+#include <limits.h> /* defines INT_MIN, INT_MAX, UINT_MAX, CHAR_BIT */
+#include "gl_map.hh"
+#include "gl_hash_map.h"
 #include "options.h"
 #include "hash-table.h"
 
@@ -839,6 +841,13 @@ Search::prepare_asso_values ()
   /* Memory allocation.  */
   _asso_values = new int[_alpha_size];
 
+  /* Memory allocation in each Keyword.  */
+  for (temp = _head; temp; temp = temp->rest())
+    {
+      KeywordExt *keyword = temp->first();
+      keyword->_undetermined_chars = new unsigned int[keyword->_selchars_length];
+    }
+
   int non_linked_length = _list_len;
   unsigned int asso_value_max;
 
@@ -943,10 +952,6 @@ struct EquivalenceClass
   KeywordExt_List *     _keywords_last;
   /* The number of keywords in this equivalence class.  */
   unsigned int          _cardinality;
-  /* The undetermined selected characters for the keywords in this
-     equivalence class, as a canonically reordered multiset.  */
-  unsigned int *        _undetermined_chars;
-  unsigned int          _undetermined_chars_length;
 
   EquivalenceClass *    _next;
 };
@@ -984,48 +989,78 @@ equals (const unsigned int *ptr1, const unsigned int *ptr2, unsigned int len)
   return true;
 }
 
+static bool
+undetermined_equals (KeywordExt *key1, KeywordExt *key2)
+{
+  return (key1->_undetermined_chars_length == key2 ->_undetermined_chars_length)
+         && equals (key1->_undetermined_chars, key2->_undetermined_chars,
+                    key1->_undetermined_chars_length);
+}
+
+static size_t
+undetermined_hashcode (KeywordExt *key)
+{
+  return key->_undetermined_chars_hashcode;
+}
+
 EquivalenceClass *
 Search::compute_partition (bool *undetermined) const
 {
-  EquivalenceClass *partition = NULL;
-  EquivalenceClass *partition_last = NULL;
+  /* Prepare the use of the hash-map: For each keyword,
+     compute the undetermined characters and their hash code.  */
   for (KeywordExt_List *temp = _head; temp; temp = temp->rest())
     {
       KeywordExt *keyword = temp->first();
 
-      /* Compute the undetermined characters for this keyword.  */
-      unsigned int *undetermined_chars =
-        new unsigned int[keyword->_selchars_length];
+      /* This scratch memory, an array of length keyword->_selchars_length,
+         was allocated earlier.  */
+      unsigned int *undetermined_chars = keyword->_undetermined_chars;
       unsigned int undetermined_chars_length = 0;
 
       for (int i = 0; i < keyword->_selchars_length; i++)
         if (undetermined[keyword->_selchars[i]])
           undetermined_chars[undetermined_chars_length++] = keyword->_selchars[i];
+      keyword->_undetermined_chars_length = undetermined_chars_length;
+
+      {
+        const int SIZE_BITS = sizeof (size_t) * CHAR_BIT;
+        size_t h = undetermined_chars_length;
+        for (unsigned int i = 0; i < undetermined_chars_length; i++)
+          h = undetermined_chars[i] * 641 + ((h << 9) | (h >> (SIZE_BITS - 9)));
+        keyword->_undetermined_chars_hashcode = h;
+      }
+    }
+
+  EquivalenceClass *partition = NULL;
+  EquivalenceClass *partition_last = NULL;
+  /* A hash-map that maps each keyword to the EquivalenceClass that contains
+     it.  */
+  gl_Map<KeywordExt *, EquivalenceClass const *>
+    map (GL_HASH_MAP, undetermined_equals, undetermined_hashcode, NULL, NULL);
+  for (KeywordExt_List *temp = _head; temp; temp = temp->rest())
+    {
+      KeywordExt *keyword = temp->first();
 
       /* Look up the equivalence class to which this keyword belongs.  */
-      EquivalenceClass *equclass;
-      for (equclass = partition; equclass; equclass = equclass->_next)
-        if (equclass->_undetermined_chars_length == undetermined_chars_length
-            && equals (equclass->_undetermined_chars, undetermined_chars,
-                       undetermined_chars_length))
-          break;
+      EquivalenceClass *equclass = const_cast<EquivalenceClass *>(map.get(keyword));
       if (equclass == NULL)
         {
           equclass = new EquivalenceClass();
           equclass->_keywords = NULL;
           equclass->_keywords_last = NULL;
           equclass->_cardinality = 0;
-          equclass->_undetermined_chars = undetermined_chars;
-          equclass->_undetermined_chars_length = undetermined_chars_length;
           equclass->_next = NULL;
+
+          /* Map this keyword (and all equivalent ones that will be seen later)
+             to equclass.  */
+          map.put(keyword, equclass);
+
           if (partition)
             partition_last->_next = equclass;
           else
             partition = equclass;
           partition_last = equclass;
         }
-      else
-        delete[] undetermined_chars;
 
       /* Add the keyword to the equivalence class.  */
       KeywordExt_List *cons = new KeywordExt_List(keyword);
@@ -1037,10 +1072,6 @@ Search::compute_partition (bool *undetermined) const
       equclass->_cardinality++;
     }
 
-  /* Free some of the allocated memory.  The caller doesn't need it.  */
-  for (EquivalenceClass *cls = partition; cls; cls = cls->_next)
-    delete[] cls->_undetermined_chars;
-
   return partition;
 }
 
@@ -1052,7 +1083,6 @@ delete_partition (EquivalenceClass *partition)
       EquivalenceClass *equclass = partition;
       partition = equclass->_next;
       delete_list (equclass->_keywords);
-      //delete[] equclass->_undetermined_chars; // already freed above
       delete equclass;
     }
 }
@@ -1572,6 +1602,13 @@ Search::find_good_asso_values ()
       delete[] best_asso_values;
       /* The keywords' _hash_value fields are recomputed below.  */
     }
+
+  /* Memory deallocation in each Keyword.  */
+  for (KeywordExt_List *temp = _head; temp; temp = temp->rest())
+    {
+      KeywordExt *keyword = temp->first();
+      delete[] keyword->_undetermined_chars;
+    }
 }
 
 /* ========================================================================= */