nasm/perllib/phash.ph
H. Peter Anvin (Intel) 671f204ed1 phash: bloat the hashes somewhat, reducing the likelihood of false positives
Set the hash size scaling constant to 1.6, signifying 3.2 times the
hash load. This both reduces the convergence time and makes it less
likely (< 25%) that a non-entry will require a secondary comparison,
and after all, in most of our use cases non-entries are by far the
more common.

Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
2020-07-27 13:24:59 -07:00

203 lines
4.4 KiB
Perl

# -*- perl -*-
#
# Perfect Minimal Hash Generator written in Perl, which produces
# C output.
#
require 'random_sv_vectors.ph';
require 'crc64.ph';
#
# Compute the prehash for a key
#
# prehash(key, sv, N)
#
sub prehash($$$) {
my($key, $n, $sv) = @_;
my @c = crc64($sv, $key);
# Create a bipartite graph...
$k1 = (($c[1] & ($n-1)) << 1) + 0; # low word
$k2 = (($c[0] & ($n-1)) << 1) + 1; # high word
return ($k1, $k2);
}
#
# Walk the assignment graph, return true on success
#
sub walk_graph($$$$) {
my($nodeval,$nodeneigh,$n,$v) = @_;
my $nx;
# print STDERR "Vertex $n value $v\n";
$$nodeval[$n] = $v;
foreach $nx (@{$$nodeneigh[$n]}) {
# $nx -> [neigh, hash]
my ($o, $e) = @$nx;
# print STDERR "Edge $n,$o value $e: ";
my $ov;
if (defined($ov = $$nodeval[$o])) {
if ($v+$ov != $e) {
# Cyclic graph with collision
# print STDERR "error, should be ", $v+$ov, "\n";
return 0;
} else {
# print STDERR "ok\n";
}
} else {
return 0 unless (walk_graph($nodeval, $nodeneigh, $o, $e-$v));
}
}
return 1;
}
#
# Generate the function assuming a given N.
#
# gen_hash_n(N, sv, \%data, run)
#
sub gen_hash_n($$$$) {
my($n, $sv, $href, $run) = @_;
my @keys = keys(%{$href});
my $i;
my $gr;
my ($k, $v);
my $gsize = 2*$n;
my @nodeval;
my @nodeneigh;
my %edges;
for ($i = 0; $i < $gsize; $i++) {
$nodeneigh[$i] = [];
}
%edges = ();
foreach $k (@keys) {
my ($pf1, $pf2) = prehash($k, $n, $sv);
($pf1,$pf2) = ($pf2,$pf1) if ($pf1 > $pf2); # Canonicalize order
my $pf = "$pf1,$pf2";
my $e = ${$href}{$k};
my $xkey;
if (defined($xkey = $edges{$pf})) {
next if ($e == ${$href}{$xkey}); # Duplicate hash, safe to ignore
if (defined($run)) {
print STDERR "$run: Collision: $pf: $k with $xkey\n";
}
return;
}
# print STDERR "Edge $pf value $e from $k\n";
$edges{$pf} = $k;
push(@{$nodeneigh[$pf1]}, [$pf2, $e]);
push(@{$nodeneigh[$pf2]}, [$pf1, $e]);
}
# Now we need to assign values to each vertex, so that for each
# edge, the sum of the values for the two vertices give the value
# for the edge (which is our hash index.) If we find an impossible
# sitation, the graph was cyclic.
@nodeval = (undef) x $gsize;
for ($i = 0; $i < $gsize; $i++) {
if (scalar(@{$nodeneigh[$i]})) {
# This vertex has neighbors (is used)
if (!defined($nodeval[$i])) {
# First vertex in a cluster
unless (walk_graph(\@nodeval, \@nodeneigh, $i, 0)) {
if (defined($run)) {
print STDERR "$run: Graph is cyclic\n";
}
return;
}
}
}
}
# for ($i = 0; $i < $n; $i++) {
# print STDERR "Vertex ", $i, ": ", $g[$i], "\n";
# }
if (defined($run)) {
printf STDERR "$run: Done: n = $n, sv = [0x%08x, 0x%08x]\n",
$$sv[0], $$sv[1];
}
return ($n, $sv, \@nodeval);
}
#
# Driver for generating the function
#
# gen_perfect_hash(\%data)
#
sub gen_perfect_hash($) {
my($href) = @_;
my @keys = keys(%{$href});
my @hashinfo;
my ($n, $i, $j, $sv, $maxj);
my $run = 1;
# Minimal power of 2 value for N with enough wiggle room.
# The scaling constant must be larger than 0.5 in order for the
# algorithm to ever terminate. The higher the scaling constant,
# the more space does the hash take up, but the less likely is it
# that an invalid token will require a string comparison.
my $room = int(scalar(@keys)*1.6);
$n = 1;
while ($n < $room) {
$n <<= 1;
}
# Number of times to try...
$maxj = scalar @random_sv_vectors;
for ($i = 0; $i < 4; $i++) {
printf STDERR "%d vectors, trying n = %d...\n",
scalar @keys, $n;
for ($j = 0; $j < $maxj; $j++) {
$sv = $random_sv_vectors[$j];
@hashinfo = gen_hash_n($n, $sv, $href, $run++);
return @hashinfo if (@hashinfo);
}
$n <<= 1;
}
return;
}
#
# Verify that the hash table is actually correct...
#
sub verify_hash_table($$)
{
my ($href, $hashinfo) = @_;
my ($n, $sv, $g) = @{$hashinfo};
my $k;
my $err = 0;
foreach $k (keys(%$href)) {
my ($pf1, $pf2) = prehash($k, $n, $sv);
my $g1 = ${$g}[$pf1];
my $g2 = ${$g}[$pf2];
if ($g1+$g2 != ${$href}{$k}) {
printf STDERR "%s(%d,%d): %d+%d = %d != %d\n",
$k, $pf1, $pf2, $g1, $g2, $g1+$g2, ${$href}{$k};
$err = 1;
} else {
# printf STDERR "%s: %d+%d = %d ok\n",
# $k, $g1, $g2, $g1+$g2;
}
}
die "$0: hash validation error\n" if ($err);
}
1;