From fb5a599c8ac2c67139e2d37c0b7031ea75e1da84 Mon Sep 17 00:00:00 2001
From: "H. Peter Anvin" <hpa@zytor.com>
Date: Thu, 30 Aug 2007 23:42:39 +0000
Subject: [PATCH] phash.ph: use a bipartite graph to reduce the storage
 requirements

Since we fold the f- and g-functions together, if we guarantee that g is
bipartite, we can make g twice the size of f without cost.  This greatly
improves the odds of generating a smaller hash.
---
 Makefile.in      |  2 +-
 perllib/phash.ph | 32 +++++++++++++++++++-------------
 2 files changed, 20 insertions(+), 14 deletions(-)

diff --git a/Makefile.in b/Makefile.in
index 7cec801c..0651b8e7 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -113,7 +113,7 @@ regs.h: regs.dat regs.pl
 	$(PERL) $(srcdir)/regs.pl h $(srcdir)/regs.dat > regs.h
 
 # Token hash
-tokhash.c: insns.dat regs.dat tokens.dat tokhash.pl
+tokhash.c: insns.dat regs.dat tokens.dat tokhash.pl perllib/phash.ph
 	$(PERL) $(srcdir)/tokhash.pl $(srcdir)/insns.dat $(srcdir)/regs.dat \
 		$(srcdir)/tokens.dat > tokhash.c
 
diff --git a/perllib/phash.ph b/perllib/phash.ph
index 438e4660..017155d7 100644
--- a/perllib/phash.ph
+++ b/perllib/phash.ph
@@ -3,7 +3,7 @@
 # Perfect Minimal Hash Generator written in Perl, which produces
 # C output.
 #
-# Requires the CPAN Graph module (tested against 0.83)
+# Requires the CPAN Graph module (tested against 0.81, 0.83, 0.84)
 
 use Graph::Undirected;
 
@@ -65,10 +65,15 @@ sub shuffle(@) {
 #
 # ffunc(N)
 #
-sub ffunc($) {
-    my($n) = @_;
+sub ffunc($$$) {
+    my($n,$s,$i) = @_;
+    my(@l) = ();
 
-    return shuffle(0..$n-1);
+    while ($n--) {
+	push(@l, $i);
+	$i += $s;
+    }
+    return shuffle(@l);
 }
 
 #
@@ -107,12 +112,13 @@ sub gen_hash_n($$$) {
     my $i, $sv, @f1, @f2, @g;
     my $gr;
     my $k, $v;
+    my $gsize = 2*$n;
 
-    @f1 = ffunc($n);
-    @f2 = ffunc($n);
+    @f1 = ffunc($n, 2, 0);
+    @f2 = ffunc($n, 2, 1);
 
     $gr = Graph::Undirected->new;
-    for ($i = 0; $i < $n; $i++) {
+    for ($i = 0; $i < $gsize; $i++) {
 	$gr->add_vertex($i);
     }
 
@@ -149,7 +155,7 @@ sub gen_hash_n($$$) {
     # edge, the sum of the values for the two vertices give the value
     # for the edge (which is our hash index.)  Since the graph is
     # acyclic, this is always doable.
-    for ($i = 0; $i < $n; $i++) {
+    for ($i = 0; $i < $gsize; $i++) {
 	if (!$gr->has_vertex_attribute($i, "val")) {
 	    walk_graph($gr,$i,0); # First vertex in a cluster
 	}
@@ -160,7 +166,7 @@ sub gen_hash_n($$$) {
     #	print STDERR "Vertex ", $i, ": ", $g[$i], "\n";
     # }
 
-    print STDERR "Done: n = $n, sv = $sv\n";
+    print STDERR "Done: n = $n, sv = [", join(',', @$sv), "]\n";
 
     return ($n, $sv, \@f1, \@f2, \@g);
 }
@@ -184,8 +190,10 @@ sub gen_perfect_hash($) {
     my @hashinfo;
     my $n, $i, $j, $sv, $maxj;
 
-    # Minimal power of 2 value for N with enough wiggle room
-    my $room = scalar(@keys)*1.3;
+    # Minimal power of 2 value for N with enough wiggle room.
+    # The scaling constant must be larger than 0.5 in order for the
+    # algorithm to ever terminate.
+    my $room = scalar(@keys)*0.8;
     $n = 1;
     while ($n < $room) {
 	$n <<= 1;
@@ -243,8 +251,6 @@ sub verify_hash_table($$)
     my $k;
     my $err = 0;
 
-    print STDERR "Verify: n = $n, sv = $sv\n";
-
     foreach $k (keys(%$href)) {
 	my ($p1, $p2) = prehash($k, $n, $sv);
 	my $pf1 = ${$f1}[$p1];