1 package MinimalPerfectHash;
10 my $FNV_CONST= 16777619;
16 foreach my $char (split //, $key) {
17 $hash = $hash ^ ord($char);
18 $hash = ($hash * $FNV_CONST) & 0xFFFFFFFF;
23 sub build_perfect_hash {
24 my ($hash, $split_pos)= @_;
27 my $max_h= 0xFFFFFFFF;
28 $max_h -= $max_h % $n; # this just avoids a tiny bit bias
29 my $seed1= unpack("N", "Perl") - 1;
35 for ($seed1++;1;$seed1++) {
41 foreach my $key (sort keys %$hash) {
42 $length_all_keys += length $key;
43 my $h= _fnv($key,$seed1);
44 next SEED1 if $h >= $max_h; # check if this hash would bias, and if so find a new seed
45 next SEED1 if exists $hash_to_key{$h};
46 next SEED1 if $high{$h >> $RSHIFT}++;
47 $hash_to_key{$h}= $key;
48 $key_to_hash{$key}= $h;
49 push @{$key_buckets{$h % $n}}, $key;
51 $hash_to_key= \%hash_to_key;
52 $key_to_hash= \%key_to_hash;
53 $key_buckets= \%key_buckets;
60 foreach my $first_idx (sort { @{$key_buckets->{$b}} <=> @{$key_buckets->{$a}} || $a <=> $b } keys %$key_buckets) {
61 my $keys= $key_buckets->{$first_idx};
62 #printf "got %d keys in bucket %d\n", 0+@$keys, $first_idx;
65 for ($seed2=1;1;$seed2++) {
66 goto FIND_SEED if $seed2 > 0xFFFF;
68 ( ( ( $key_to_hash->{$_} >> $RSHIFT ) ^ $seed2 ) & 0xFFFFFFFF ) % $n
71 next SEED2 if grep { $second_level[$_] || $seen{$_}++ } @idx;
72 $first_level[$first_idx]= $seed2;
73 @second_level[@idx]= map {
74 my $sp= $split_pos->{$_} // die "no split pos for '$_':$!";
75 my ($prefix,$suffix)= unpack "A${sp}A*", $_;
81 hash => $key_to_hash->{$_},
90 $second_level[$_]{seed2}= $first_level[$_]||0, $second_level[$_]{idx}= $_ for 0 .. $#second_level;
92 return $seed1, \@second_level, $length_all_keys;
95 sub build_split_words {
96 my ($hash, $preprocess, $blob, $old_res)= @_;
101 foreach my $key (sort {length($b) <=> length($a) || $a cmp $b } keys %$hash) {
102 my ($prefix,$suffix);
103 if ($key=~/^([^=]+=)([^=]+)\z/) {
104 ($prefix,$suffix)= ($1, $2);
113 foreach my $key (sort {length($b) <=> length($a) || $a cmp $b } keys %parts) {
114 $blob .= $key . "\0";
116 printf "Using preprocessing, initial blob size %d\n", length($blob);
118 printf "No preprocessing, initial blob size %d\n", length($blob);
127 (length($b) <=> length($a)) ||
132 next if exists $res{$key};
133 if (index($blob,$key) >= 0 ) {
134 my $idx= length($key);
135 if ($DEBUG and $old_res and $old_res->{$key} != $idx) {
136 print "changing: $key => $old_res->{$key} : $idx\n";
141 my $best= length($key);
143 my $min= 0; #length $key >= 4 ? 4 : 0;
146 foreach my $idx (reverse $min .. length($key)) {
147 my $prefix= substr($key,0,$idx);
148 my $suffix= substr($key,$idx);
149 my $i1= index($blob,$prefix)>=0;
150 my $i2= index($blob,$suffix)>=0;
152 if ($DEBUG and $old_res and $old_res->{$key} != $idx) {
153 print "changing: $key => $old_res->{$key} : $idx\n";
156 $appended{$prefix}++;
157 $appended{$suffix}++;
160 if (length $suffix <= length $append) {
163 $best_prefix= $prefix;
164 $best_suffix= $suffix;
167 if (length $prefix <= length $append) {
170 $best_prefix= $prefix;
171 $best_suffix= $suffix;
175 if ($DEBUG and $old_res and $old_res->{$key} != $best) {
176 print "changing: $key => $old_res->{$key} : $best\n";
178 #print "$best_prefix|$best_suffix => $best => $append\n";
181 $appended{$best_prefix}++;
182 $appended{$best_suffix}++;
185 foreach my $key (sort { length($b) <=> length($a) || $a cmp $b } keys %appended) {
186 $b2 .= $key unless index($b2,$key)>=0;
188 if (length($b2)<length($blob)) {
189 printf "Length old blob: %d length new blob: %d, recomputing using new blob\n", length($blob),length($b2);
194 printf "Length old blob: %d length new blob: %d, keeping old blob\n", length($blob),length($b2);
196 die sprintf "not same size? %d != %d", 0+keys %res, 0+keys %$hash unless keys %res == keys %$hash;
197 return ($blob,\%res);
202 my ($blob,$blob_name)= @_;
204 $blob_name ||= "mph_blob";
206 # output the blob as C code.
207 my @code= (sprintf "STATIC const unsigned char %s[] =\n",$blob_name);
208 my $blob_len= length $blob;
209 while (length($blob)) {
210 push @code, sprintf qq( "%s"), substr($blob,0,65,"");
211 push @code, length $blob ? "\n" : ";\n";
213 push @code, "/* $blob_name length: $blob_len */\n";
214 return join "",@code;
219 print $ofh "#include <stdio.h>\n";
220 print $ofh "#include <string.h>\n";
221 print $ofh "#include <stdint.h>\n";
226 my ($ofh,$defines)= @_;
229 foreach my $def (keys %$defines) {
230 $key_len //= length $def;
231 $key_len= length $def if $key_len < length $def;
233 foreach my $def (sort keys %$defines) {
234 printf $ofh "#define %*s %5d\n", -$key_len, $def, $defines->{$def};
240 sub build_array_of_struct {
241 my ($second_level,$blob)= @_;
246 foreach my $row (@$second_level) {
247 $defines{$row->{value}}= $row->{idx}+1;
248 $tests{$row->{key}}= $defines{$row->{value}};
251 index($blob,$row->{prefix}//0),
252 index($blob,$row->{suffix}//0),
254 $_ > 0xFFFF and die "panic: value exceeds range of U16"
257 length($row->{prefix}),
258 length($row->{suffix}),
260 $_ > 0xFF and die "panic: value exceeds range of U8"
262 push @rows, sprintf(" { %5d, %5d, %5d, %3d, %3d, %s } /* %s%s */",
263 @u16, @u8, $row->{value}, $row->{prefix}, $row->{suffix});
265 return \@rows,\%defines,\%tests;
269 my ($second_level, $seed1, $length_all_keys, $smart_blob, $rows,
270 $blob_name, $struct_name, $table_name, $match_name, $prefix) = @_;
272 $blob_name ||= "mph_blob";
273 $struct_name ||= "mph_struct";
274 $table_name ||= "mph_table";
277 my $n= 0+@$second_level;
278 my $data_size= 0+@$second_level * 8 + length $smart_blob;
280 my @code = "#define ${prefix}_VALt I16\n\n";
282 push @code, sprintf "rows: %s\n", $n;
283 push @code, sprintf "seed: %s\n", $seed1;
284 push @code, sprintf "full length of keys: %d\n", $length_all_keys;
285 push @code, sprintf "blob length: %d\n", length $smart_blob;
286 push @code, sprintf "ref length: %d\n", 0+@$second_level * 8;
287 push @code, sprintf "data size: %d (%%%.2f)\n", $data_size, ($data_size / $length_all_keys) * 100;
288 push @code, "*/\n\n";
290 push @code, blob_as_code($smart_blob, $blob_name);
291 push @code, <<"EOF_CODE";
293 struct $struct_name {
299 ${prefix}_VALt value;
304 push @code, "#define ${prefix}_RSHIFT $RSHIFT\n";
305 push @code, "#define ${prefix}_BUCKETS $n\n\n";
306 push @code, sprintf "STATIC const U32 ${prefix}_SEED1 = 0x%08x;\n", $seed1;
307 push @code, sprintf "STATIC const U32 ${prefix}_FNV_CONST = 0x%08x;\n\n", $FNV_CONST;
309 push @code, "/* The comments give the input key for the row it is in */\n";
310 push @code, "STATIC const struct $struct_name $table_name\[${prefix}_BUCKETS] = {\n", join(",\n", @$rows)."\n};\n\n";
311 push @code, <<"EOF_CODE";
312 ${prefix}_VALt $match_name( const unsigned char * const key, const U16 key_len ) {
313 const unsigned char * ptr= key;
314 const unsigned char * ptr_end= key + key_len;
315 U32 h= ${prefix}_SEED1;
320 h *= ${prefix}_FNV_CONST;
321 } while ( ++ptr < ptr_end );
322 n= h % ${prefix}_BUCKETS;
323 s = $table_name\[n].seed2;
325 h= (h >> ${prefix}_RSHIFT) ^ s;
326 n = h % ${prefix}_BUCKETS;
328 ( $table_name\[n].pfx_len + $table_name\[n].sfx_len == key_len ) &&
329 ( memcmp($blob_name + $table_name\[n].pfx, key, $table_name\[n].pfx_len) == 0 ) &&
330 ( !$table_name\[n].sfx_len || memcmp($blob_name + $table_name\[n].sfx,
331 key + $table_name\[n].pfx_len, $table_name\[n].sfx_len) == 0 )
333 return $table_name\[n].value;
340 return join "", @code;
344 my ($ofh, $second_level, $seed1, $long_blob, $smart_blob, $rows,
345 $blob_name, $struct_name, $table_name, $match_name ) = @_;
350 open $ofh, ">", $file
351 or die "Failed to open '$file': $!";
354 my $code = make_algo(
355 $second_level, $seed1, $long_blob, $smart_blob, $rows,
356 $blob_name, $struct_name, $table_name, $match_name );
361 my ($ofh,$h_file,$match_name,$prefix)=@_;
362 print $ofh <<"EOF_CODE";
365 int main(int argc, char *argv[]){
367 for (i=1; i<argc; i++) {
368 unsigned char *key = (unsigned char *)argv[i];
369 int key_len = strlen(argv[i]);
370 printf("key: %s got: %d\\n", key, $match_name((unsigned char *)key,key_len));
377 # output the test Perl code.
379 my ($file, $tests_hash)= @_;
380 open my $ofh, ">", $file
381 or die "Failed to open '$file' for writing: $!";
382 my $num_tests= 2 + keys %$tests_hash;
383 print $ofh "use strict;\nuse warnings;\nuse Test::More tests => $num_tests;\nmy \@res;";
385 my @tests= sort keys %$tests_hash;
386 print $ofh "\@res=`./mph_test '$tests[0]/should-not-match' 'should-not-match/$tests[0]'`;\n";
387 print $ofh "ok( \$res[0] =~ /got: 0/,'proper prefix does not match');\n";
388 print $ofh "ok( \$res[1] =~ /got: 0/,'proper suffix does not match');\n";
390 my @batch= splice @tests,0,10;
391 my $batch_args= join " ", map { "'$_'" } @batch;
392 print $ofh "\@res=`./mph_test $batch_args`;\n";
393 foreach my $i (0..$#batch) {
395 my $want= $tests_hash->{$key};
396 print $ofh "ok(\$res[$i]=~/got: (\\d+)/ && \$1 == $want, '$key');\n";
402 sub print_test_binary {
403 my ($file,$h_file, $second_level, $seed1, $length_all_keys,
404 $smart_blob, $rows, $defines, $match_name, $prefix)= @_;
405 open my $ofh, ">", $file
406 or die "Failed to open '$file': $!";
407 print_includes($ofh);
408 print_defines($ofh, $defines);
409 print_main($ofh,$h_file,$match_name,$prefix);
413 sub make_mph_from_hash {
416 # we do this twice because often we can find longer prefixes on the second pass.
417 my @keys= sort {length($b) <=> length($a) || $a cmp $b } keys %$hash;
419 my ($smart_blob, $res_to_split)= build_split_words($hash,0);
421 my ($smart_blob2, $res_to_split2)= build_split_words($hash,1);
422 if (length($smart_blob) > length($smart_blob2)) {
423 printf "Using preprocess-smart blob, length: %d (vs %d)\n", length $smart_blob2, length $smart_blob;
424 $smart_blob= $smart_blob2;
425 $res_to_split= $res_to_split2;
427 printf "Using greedy-smart blob, length: %d (vs %d)\n", length $smart_blob, length $smart_blob2;
430 my ($seed1, $second_level, $length_all_keys)= build_perfect_hash($hash, $res_to_split);
431 my ($rows, $defines, $tests)= build_array_of_struct($second_level, $smart_blob);
432 return ($second_level, $seed1, $length_all_keys, $smart_blob, $rows, $defines, $tests);
436 my ($hash,$base_name)= @_;
438 my $h_name= $base_name . "_algo.h";
439 my $c_name= $base_name . "_test.c";
440 my $p_name= $base_name . "_test.pl";
441 my $blob_name= $base_name . "_blob";
442 my $struct_name= $base_name . "_bucket_info";
443 my $table_name= $base_name . "_table";
444 my $match_name= $base_name . "_match";
445 my $prefix= uc($base_name);
447 my ($second_level, $seed1, $length_all_keys,
448 $smart_blob, $rows, $defines, $tests)= make_mph_from_hash( $hash );
450 $second_level, $seed1, $length_all_keys, $smart_blob, $rows,
451 $blob_name, $struct_name, $table_name, $match_name, $prefix );
452 print_test_binary( $c_name, $h_name, $second_level, $seed1, $length_all_keys,
453 $smart_blob, $rows, $defines,
454 $match_name, $prefix );
455 print_tests( $p_name, $tests );
462 do "../perl/lib/unicore/Heavy.pl";
463 %hash= %utf8::loose_to_file_of;
465 if ($ENV{MERGE_KEYS}) {
466 my @keys= keys %hash;
467 foreach my $loose (keys %utf8::loose_property_name_of) {
468 my $to= $utf8::loose_property_name_of{$loose};
469 next if $to eq $loose;
470 foreach my $key (@keys) {
472 if ($copy=~s/^\Q$to\E(=|\z)/$loose$1/) {
473 #print "$key => $copy\n";
479 foreach my $key (keys %hash) {
480 my $munged= uc($key);
482 $hash{$key} = $munged;
485 my $name= shift @ARGV;
487 make_files(\%hash,$name);