Require the generation of parse-gram.output.

[bison.git] / etc / bench.pl.in

diff --git a/etc/bench.pl.in b/etc/bench.pl.in
index 481f3dda91fc9f7a6f31f170c94246662d48b123..f71611f09a8b3954c0b0a9aebf8c3368e9059013 100755 (executable)
--- a/etc/bench.pl.in
+++ b/etc/bench.pl.in
@@ -19,34 +19,108 @@
 
 =head1 NAME
 
 
 =head1 NAME
 

-bench.pl - perform benches on Bison parsers.
+bench.pl - bench marks for Bison parsers.

 
 =head1 SYNOPSIS
 
 
 =head1 SYNOPSIS
 

-  ./bench.pl
+  ./bench.pl [OPTIONS]... I<directives>
+
+=head1 DIRECTIVES
+
+Specify the set of benches to run.  The following grammar defines the
+I<directives>:
+
+   I<directives> ::= I<directives> | I<directives>  -- Alternation
+                  |  I<directives> & I<directives>  -- Concatenation
+                  |  [ I<directives> ]              -- Optional
+                  |  ( I<directives> )              -- Parentheses
+                  | I<directive>
+
+Parentheses only group to override precedence.  For instance:
+
+  [ %debug ] & [ %error-verbose ] & [ %define variant ]
+
+will generate eight different cases.

 
 =head1 OPTIONS
 
 =over 4
 
 
 =head1 OPTIONS
 
 =over 4
 

+=item B<-b>, B<--bench>
+
+Predefined benches, that is, combimation between a grammar and a I<directives>
+request.
+
+=over 4
+
+=item I<fusion>
+
+Test F<lalr1.cc> with three stacks against F<lalr1-fusion.cc> which
+uses a single one.
+
+=item I<push>
+
+Test the push parser vs. the pull interface.  Use the C parser.
+
+=item I<variant>
+
+Test the use of variants instead of union in the C++ parser.
+
+=back
+

 =item B<-c>, B<--cflags>=I<flags>
 
 =item B<-c>, B<--cflags>=I<flags>
 

-Flags to pass to the C or C++ compiler.
+Flags to pass to the C or C++ compiler.  Defaults to -O2.
+
+=item B<-d>, B<--directive>=I<directives>
+
+Add a set of Bison directives to bench against each other.
+
+=item B<-g>, B<--grammar>=I<grammar>
+
+Select the base I<grammar> to use.  Defaults to I<calc>.
+
+=over 4
+
+=item I<calc>
+
+Traditional calculator.
+
+=item I<list>
+
+C++ grammar that uses std::string and std::list.  Can be used with
+or without %define variant.
+
+=item I<triangular>
+
+Artificial grammar with very long rules.
+
+=back
+
+=item B<-h>, B<--help>
+
+Display this message and exit succesfully.  The more verbose, the more
+details.

 
 =item B<-i>, B<--iterations>=I<integer>
 
 Say how many times a single test of the bench must be run.  If
 negative, specify the minimum number of CPU seconds to run.  Defaults
 
 =item B<-i>, B<--iterations>=I<integer>
 
 Say how many times a single test of the bench must be run.  If
 negative, specify the minimum number of CPU seconds to run.  Defaults

-to -3.
+to -1.
+
+=item B<-q>, B<--quiet>
+
+Decrease the verbosity level (defaults to 1).

 
 =item B<-v>, B<--verbose>
 
 
 =item B<-v>, B<--verbose>
 

-Raise the verbosity level.  Currently only affects B<--help>.
+Raise the verbosity level (defaults to 1).

 
 =back
 
 =cut
 
 
 =back
 
 =cut
 

+use strict;

 use IO::File;
 
 ##################################################################
 use IO::File;
 
 ##################################################################


@@ -55,6 +129,10 @@ use IO::File;
 
 =over 4
 
 
 =over 4
 

+=item C<@bench>
+
+The list of benches to run.
+

 =item C<$bison>
 
 The Bison program to use to compile the grammar.
 =item C<$bison>
 
 The Bison program to use to compile the grammar.


@@ -71,6 +149,10 @@ The C++ compiler.
 
 Compiler flags (C or C++).
 
 
 Compiler flags (C or C++).
 

+=item C<@directive>
+
+A list of directive sets to measure against each other.
+

 =item C<$iterations>
 
 The number of times the parser is run for a bench.
 =item C<$iterations>
 
 The number of times the parser is run for a bench.


@@ -83,12 +165,15 @@ Verbosity level.
 
 =cut
 
 
 =cut
 

+my $bench;

 my $bison = $ENV{'BISON'} || '@abs_top_builddir@/tests/bison';
 my $cc = $ENV{'CC'} || 'gcc';
 my $cxx = $ENV{'CXX'} || 'g++';
 my $bison = $ENV{'BISON'} || '@abs_top_builddir@/tests/bison';
 my $cc = $ENV{'CC'} || 'gcc';
 my $cxx = $ENV{'CXX'} || 'g++';

-my $cflags = '';
-my $iterations = -3;
-my $verbose = 0;
+my $cflags = '-O2';
+my @directive = ();
+my $grammar = 'calc';
+my $iterations = -1;
+my $verbose = 1;

 
 =head1 FUNCTIONS
 
 
 =head1 FUNCTIONS
 


@@ -107,30 +192,27 @@ sub verbose($$)
     if $level <= $verbose;
 }
 
     if $level <= $verbose;
 }
 

+
+######################################################################
+

 =item C<directives($bench, @directive)>
 
 Format the list of directives for Bison for bench named C<$bench>.
 
 =item C<directives($bench, @directive)>
 
 Format the list of directives for Bison for bench named C<$bench>.
 

-The special fake C<%variant> directive requests the use of
-Boost.Variants instead of a regular union.  So don't pass it, it is
-not a valid directive.
-

 =cut
 
 sub directives($@)
 {
   my ($bench, @directive) = @_;
   my $res = "/* Directives for bench `$bench'. */\n";
 =cut
 
 sub directives($@)
 {
   my ($bench, @directive) = @_;
   my $res = "/* Directives for bench `$bench'. */\n";

-  for my $d (@directive)
-    {
-      $res .= $d . "\n"
-        unless $d eq '%variant';
-    }
+  $res .= join ("\n", @directive) . "\n";

   $res .= "/* End of directives for bench `$bench'. */\n";
   return $res;
 }
 
   $res .= "/* End of directives for bench `$bench'. */\n";
   return $res;
 }
 

-=item C<triangular_grammar ($base, $max, @directive)>
+######################################################################
+
+=item C<generate_grammar_triangular ($base, $max, @directive)>

 
 Create a large triangular grammar which looks like :
 
 
 Create a large triangular grammar which looks like :
 


@@ -156,7 +238,7 @@ The created parser is self contained: it includes its scanner, and
 source of input.
 =cut
 
 source of input.
 =cut
 

-sub triangular_grammar ($$$)
+sub generate_grammar_triangular ($$@)

 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);


@@ -267,7 +349,8 @@ sub calc_input ($$)
 }
 
 ##################################################################
 }
 
 ##################################################################

-=item C<calc_grammar ($base, $max, @directive)>
+
+=item C<generate_grammar_calc ($base, $max, @directive)>

 
 Generate a Bison file F<$base.y> for a calculator parser in C.  Pass
 the additional Bison C<@directive>.  C<$max> is ignored, but left to
 
 Generate a Bison file F<$base.y> for a calculator parser in C.  Pass
 the additional Bison C<@directive>.  C<$max> is ignored, but left to


@@ -275,11 +358,15 @@ have the same interface as C<triangular_grammar>.
 
 =cut
 
 
 =cut
 

-sub calc_grammar ($$$)
+sub generate_grammar_calc ($$@)

 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 
 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 

+  # Putting this request here is stupid, since the input will be
+  # generated each time we generate a grammar.
+  calc_input ('calc', 200);
+

   my $out = new IO::File ">$base.y"
     or die;
   print $out <<EOF;
   my $out = new IO::File ">$base.y"
     or die;
   print $out <<EOF;


@@ -477,18 +564,18 @@ EOF
 
 ##################################################################
 
 
 ##################################################################
 

-=item C<variant_grammar ($base, $max, @directive)>
+=item C<generate_grammar_list ($base, $max, @directive)>

 
 

-Generate a Bison file F<$base.y> that uses, or not, the Boost.Variants
-depending on the C<@directive>.
+Generate a Bison file F<$base.y> for a C++ parser that uses C++
+objects (std::string, std::list).  Tailored for using %define variant.

 
 =cut
 
 
 =cut
 

-sub variant_grammar ($$$)
+sub generate_grammar_list ($$@)

 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);
 {
   my ($base, $max, @directive) = @_;
   my $directives = directives ($base, @directive);

-  my $variant = grep { $_ eq '%variant' } @directive;
+  my $variant = grep { $_ eq '%define variant' } @directive;

 
   my $out = new IO::File ">$base.y"
     or die;
 
   my $out = new IO::File ">$base.y"
     or die;


@@ -497,12 +584,12 @@ sub variant_grammar ($$$)
 %defines
 $directives
 
 %defines
 $directives
 

-%code requires // variant.h
+%code requires // *.h

 {
 #include <string>
 }
 
 {
 #include <string>
 }
 

-%code // variant.c
+%code // *.c

 {
 #include <algorithm>
 #include <iostream>
 {
 #include <algorithm>
 #include <iostream>


@@ -511,7 +598,7 @@ $directives
 // Prototype of the yylex function providing subsequent tokens.
 static yy::parser::token_type yylex(yy::parser::semantic_type* yylval);
 
 // Prototype of the yylex function providing subsequent tokens.
 static yy::parser::token_type yylex(yy::parser::semantic_type* yylval);
 

-#define STAGE_MAX    ($max * 10)
+#define STAGE_MAX    ($max * 10) // max = $max

 #define USE_VARIANTS $variant
 #if USE_VARIANTS
 # define IF_VARIANTS(True, False) True
 #define USE_VARIANTS $variant
 #if USE_VARIANTS
 # define IF_VARIANTS(True, False) True


@@ -524,7 +611,6 @@ EOF
   if ($variant)
     {
       print $out <<'EOF';
   if ($variant)
     {
       print $out <<'EOF';

-%code variant {int,std::string}

 %token <std::string> TEXT
 %token <int> NUMBER
 %printer { std::cerr << "Number: " << $$; } <int>
 %token <std::string> TEXT
 %token <int> NUMBER
 %printer { std::cerr << "Number: " << $$; } <int>


@@ -551,7 +637,7 @@ EOF
     }
   else
     {
     }
   else
     {

-      # Not using Boost variants.
+      # Not using Bison variants.

       print $out <<'EOF';
 %union {int ival; std::string* sval;}
 %token <sval> TEXT
       print $out <<'EOF';
 %union {int ival; std::string* sval;}
 %token <sval> TEXT


@@ -590,12 +676,12 @@ yylex(yy::parser::semantic_type* yylval)
     return yy::parser::token::END_OF_FILE;
   else if (stage % 2)
     {
     return yy::parser::token::END_OF_FILE;
   else if (stage % 2)
     {

-      IF_VARIANTS(*yylval, yylval->ival) = stage;
+      IF_VARIANTS(yylval->build<int>(), yylval->ival) = stage;

       return yy::parser::token::NUMBER;
     }
   else
     {
       return yy::parser::token::NUMBER;
     }
   else
     {

-      IF_VARIANTS(*yylval =, yylval->sval = new) std::string("A string.");
+      IF_VARIANTS(yylval->build<std::string>() =, yylval->sval = new) std::string("A string.");

       return yy::parser::token::TEXT;
     }
   abort();
       return yy::parser::token::TEXT;
     }
   abort();


@@ -623,6 +709,43 @@ EOF
 
 ##################################################################
 
 
 ##################################################################
 

+=item C<generate_grammar ($name, $base, @directive)>
+
+Generate F<$base.y> by calling C<&generate_grammar_$name>.
+
+=cut
+
+sub generate_grammar ($$@)
+{
+  my ($name, $base, @directive) = @_;
+  verbose 2, "Generating $base.y\n";
+  my %generator =
+    (
+      "calc"       => \&generate_grammar_calc,
+      "list"       => \&generate_grammar_list,
+      "triangular" => \&generate_grammar_triangular,
+    );
+  &{$generator{$name}}($base, 200, @directive);
+}
+
+##################################################################
+
+=item C<run ($command)>
+
+Run, possibly verbosely, the shell C<$command>.
+
+=cut
+
+sub run ($)
+{
+  my ($command) = @_;
+  verbose 2, "$command\n";
+  system ("$command") == 0
+    or die "$command failed";
+}
+
+##################################################################
+

 =item C<compile ($base)>
 
 Compile C<$base.y> to an executable C, Using the C or C++ compiler
 =item C<compile ($base)>
 
 Compile C<$base.y> to an executable C, Using the C or C++ compiler


@@ -638,41 +761,43 @@ sub compile ($)
 
   my $compiler = $language eq 'C++' ? $cxx : $cc;
 
 
   my $compiler = $language eq 'C++' ? $cxx : $cc;
 

-  system ("$bison $base.y -o $base.c") == 0
-    or die;
-  system ("$compiler -o $base $cflags $base.c") == 0
-    or die;
+  run "$bison $base.y -o $base.c";
+  run "$compiler -o $base $cflags $base.c";

 }
 
 }
 

-=item C<bench_grammar ($gram, %bench)>
-
-Generate benches for C<$gram>.  C<$gram> should be C<calc> or
-C<triangle>.  C<%bench> is a hash of the form:
+######################################################################

 
 

-  $name => @directive
+=item C<bench ($grammar, @token)>

 
 

-where C<$name> is the name of the bench, and C<@directive> are the
-Bison directive to use for this bench.  All the benches are compared
-against each other, repeated 50 times.
+Generate benches for the C<$grammar> and the directive specification
+given in the list of C<@token>.

 
 =cut
 
 
 =cut
 

-sub bench_grammar ($%)
+sub bench ($@)

 {
 {

-  my ($gram, %test) = @_;
-
+  my ($grammar, @token) = @_;

   use Benchmark qw (:all :hireswallclock);
 
   use Benchmark qw (:all :hireswallclock);
 

+  my @directive = parse (@token);
+

   # Set up the benches as expected by timethese.
   my %bench;
   # Set up the benches as expected by timethese.
   my %bench;

+  # A counter of directive sets.
+  my $count = 1;
+  for my $d (@directive)
+    {
+      $bench{$count} = $d;
+      printf " %2d. %s\n", $count, join (' ', split ("\n", $d));
+      $count++;
+    };
+

   # For each bench, capture the size.
   my %size;
   # For each bench, capture the size.
   my %size;

-  while (my ($name, $directives) = each %test)
+
+  while (my ($name, $directives) = each %bench)

     {
     {

-      verbose 1, "Generating $name\n";
-      # Call the Bison input file generator.
-      my $generator = "$gram" . "_grammar";
-      &$generator ($name, 200, @$directives);
+      generate_grammar ($grammar, $name, $directives);

       # Compile the executable.
       compile ($name);
       $bench{$name} = "system ('./$name');";
       # Compile the executable.
       compile ($name);
       $bench{$name} = "system ('./$name');";


@@ -688,20 +813,29 @@ sub bench_grammar ($%)
   # shows only wallclock and the two children times.  'auto' (the
   # default) will act as 'all' unless the children times are both
   # zero, in which case it acts as 'noc'.  'none' prevents output.
   # shows only wallclock and the two children times.  'auto' (the
   # default) will act as 'all' unless the children times are both
   # zero, in which case it acts as 'noc'.  'none' prevents output.

-  verbose 1, "Running the benches for $gram\n";
+  verbose 2, "Running the benches for $grammar\n";

   my $res = timethese ($iterations, \%bench, 'nop');
 
   # Output the speed result.
   cmpthese ($res, 'nop');
 
   # Display the sizes.
   my $res = timethese ($iterations, \%bench, 'nop');
 
   # Output the speed result.
   cmpthese ($res, 'nop');
 
   # Display the sizes.

-  print "Sizes:\n";
+  print "Sizes (decreasing):\n";
+  my $width = 10;

   for my $bench (keys %size)
     {
   for my $bench (keys %size)
     {

-      printf "%10s: %10dkB\n", $bench, int ($size{$bench} / 1024);
+      $width = length $bench
+        if $width < length $bench;
+    }
+  # Benches sorted by decreasing size.
+  my @benches_per_size = sort {$size{$b} <=> $size{$a}} keys %size;
+  for my $bench (@benches_per_size)
+    {
+      printf "%${width}s: %5.2fkB\n", $bench, $size{$bench} / 1024;

     }
 }
 
     }
 }
 

+######################################################################

 
 =item C<bench_push_parser ()>
 
 
 =item C<bench_push_parser ()>
 


@@ -712,18 +846,16 @@ interfaces.
 
 sub bench_push_parser ()
 {
 
 sub bench_push_parser ()
 {

-  calc_input ('calc', 200);
-  bench_grammar
-    ('calc',
-     (
-      "pull-impure" => [],
-      "pull-pure"   => ['%define api.pure'],
-      "push-impure" => ['%define api.push_pull "both"'],
-      "push-pure"   => ['%define api.push_pull "both"', '%define api.pure'],
-     )
-    );
+  bench ('calc',
+         (
+          '[', '%define api.pure', ']',
+          '&',
+          '[', '%define api.push_pull "both"', ']'
+         ));

 }
 
 }
 

+######################################################################
+

 =item C<bench_variant_parser ()>
 
 Bench the C++ lalr1.cc parser using Boost.Variants or %union.
 =item C<bench_variant_parser ()>
 
 Bench the C++ lalr1.cc parser using Boost.Variants or %union.


@@ -732,15 +864,31 @@ Bench the C++ lalr1.cc parser using Boost.Variants or %union.
 
 sub bench_variant_parser ()
 {
 
 sub bench_variant_parser ()
 {

-  bench_grammar
-    ('variant',
-     (
-      "union"         => [],
-      "variant"       => ['%variant'],
-      "union-debug"   => ['%debug'],
-      "variant-debug" => ['%debug', '%variant'],
-     )
-    );
+  bench ('variant',
+         ('%skeleton "lalr1.cc"',
+          '&',
+          '[', '%debug', ']',
+          '&',
+          '[', '%define variant', ']',
+          '&',
+          '[', "%code {\n#define VARIANT_DESTROY\n}", ']'
+         ));
+}
+
+######################################################################
+
+=item C<bench_fusion_parser ()>
+
+Bench the C++ lalr1.cc parser using Boost.Variants or %union.
+
+=cut
+
+sub bench_fusion_parser ()
+{
+  bench ('list',
+         ('%skeleton "lalr1-split.cc"',
+          '|',
+          '%skeleton "lalr1.cc"'));

 }
 
 ############################################################################
 }
 
 ############################################################################


@@ -756,16 +904,118 @@ sub help ($)
                -output  => \*STDOUT });
 }
 
                -output  => \*STDOUT });
 }
 

+######################################################################
+
+# The list of tokens parsed by the following functions.
+my @token;
+
+# Parse directive specifications:
+#   expr: term (| term)*
+#   term: fact (& fact)*
+#   fact: ( expr ) | [ expr ] | dirs
+sub parse (@)
+{
+  @token = @_;
+  verbose 2, "Parsing: @token\n";
+  return parse_expr ();
+}
+
+sub parse_expr ()
+{
+  my @res = parse_term ();
+  while (defined $token[0] && $token[0] eq '|')
+    {
+      shift @token;
+      # Alternation.
+      push @res, parse_term ();
+    }
+  return @res;
+}
+
+sub parse_term ()
+{
+  my @res = parse_fact ();
+  while (defined $token[0] && $token[0] eq '&')
+    {
+      shift @token;
+      # Cartesian product.
+      my @lhs = @res;
+      @res = ();
+      for my $rhs (parse_fact ())
+        {
+          for my $lhs (@lhs)
+            {
+              push @res, "$lhs\n$rhs";
+            }
+        }
+    }
+  return @res;
+}
+
+sub parse_fact ()
+{
+  my @res;
+  die "unexpected end of expression"
+    unless defined $token[0];
+
+  if ($token[0] eq '(')
+    {
+      shift @token;
+      @res = parse_expr ();
+      die "unexpected $token[0], expected )"
+        unless $token[0] eq ')';
+      shift @token;
+    }
+  elsif ($token[0] eq '[')
+    {
+      shift @token;
+      @res = (parse_expr (), '');
+      die "unexpected $token[0], expected ]"
+        unless $token[0] eq ']';
+      shift @token;
+    }
+  else
+    {
+      @res = $token[0];
+      shift @token;
+    }
+  return @res;
+}
+
+######################################################################
+

 sub getopt ()
 {
   use Getopt::Long;
 sub getopt ()
 {
   use Getopt::Long;

-  %option = ("h|help"     => sub { help ($verbose) },
-            "v|verbose"  => sub { ++$verbose },
-             "c|cflags=s" => \$cflags,
-             "i|iterations=i" => \$iterations);
+  my %option = (
+    "b|bench=s"      => \$bench,
+    "c|cflags=s"     => \$cflags,
+    "d|directive=s"  => \@directive,
+    "g|grammar=s"    => \$grammar,
+    "h|help"         => sub { help ($verbose) },
+    "i|iterations=i" => \$iterations,
+    "q|quiet"        => sub { --$verbose },
+    "v|verbose"      => sub { ++$verbose },
+    );

   Getopt::Long::Configure ("bundling", "pass_through");
   GetOptions (%option)
     or exit 1;
   Getopt::Long::Configure ("bundling", "pass_through");
   GetOptions (%option)
     or exit 1;

+
+  # Support -b: predefined benches.
+  my %bench =
+    (
+     "fusion"   => \&bench_fusion_parser,
+     "push"     => \&bench_push_parser,
+     "variant"  => \&bench_variant_parser,
+    );
+
+  if (defined $bench)
+    {
+      die "invalid argument for --bench: $bench"
+        unless defined $bench{$bench};
+      &{$bench{$bench}}();
+      exit 0;
+    }

 }
 
 ######################################################################
 }
 
 ######################################################################


@@ -775,8 +1025,11 @@ verbose 1, "Using bison=$bison.\n";
 verbose 1, "Using cc=$cc.\n";
 verbose 1, "Using cxx=$cxx.\n";
 verbose 1, "Using cflags=$cflags.\n";
 verbose 1, "Using cc=$cc.\n";
 verbose 1, "Using cxx=$cxx.\n";
 verbose 1, "Using cflags=$cflags.\n";

-# bench_push_parser();
-bench_variant_parser();
+verbose 2, "Grammar: $grammar\n";
+
+# Launch the bench marking.
+bench ($grammar, @ARGV);
+

 
 ### Setup "GNU" style for perl-mode and cperl-mode.
 ## Local Variables:
 
 ### Setup "GNU" style for perl-mode and cperl-mode.
 ## Local Variables: