X-Git-Url: https://git.saurik.com/bison.git/blobdiff_plain/922730fe36cb5fd4db11f42af6b67e9e320501ab..5679f31101d0bc5aab6adbce37212b2fdedd32cd:/etc/bench.pl.in

diff --git a/etc/bench.pl.in b/etc/bench.pl.in
index 804d4c8f..f8fca8ac 100755
--- a/etc/bench.pl.in
+++ b/etc/bench.pl.in
@@ -19,26 +19,186 @@
 
 =head1 NAME
 
-bench.pl - perform benches on Bison parsers.
+bench.pl - bench marks for Bison parsers.
 
 =head1 SYNOPSIS
 
-  ./bench.pl
+  ./bench.pl [OPTIONS]... DIRECTIVES
+
+=head1 DIRECTIVES
+
+Specify the set of benches to run.  The following grammar defines the
+I<directives>:
+
+   directives ::=
+       directives | directives  -- Alternation
+     | directives & directives  -- Concatenation
+     | [ directives> ]          -- Optional
+     | ( directives> )          -- Parentheses
+     | #d NAME[=VALUE]          -- %code { #define NAME [VALUE] }
+     | %d NAME[=VALUE]          -- %define NAME ["VALUE"]
+     | %s skeleton              -- %skeleton "skeleton"
+     | directive
+
+Parentheses only group to override precedence.  For instance:
+
+  [ %debug ] & [ %error-verbose ] & [ %define variant ]
+
+will generate eight different cases.
+
+=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>
+
+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
+to -1.
+
+=item B<-q>, B<--quiet>
+
+Decrease the verbosity level (defaults to 1).
+
+=item B<-v>, B<--verbose>
+
+Raise the verbosity level (defaults to 1).
+
+=back
 
 =cut
 
+use strict;
 use IO::File;
-use Benchmark qw (:all);
 
+##################################################################
+
+=head1 VARIABLES
+
+=over 4
+
+=item C<@bench>
+
+The list of benches to run.
+
+=item C<$bison>
+
+The Bison program to use to compile the grammar.
+
+=item C<$cc>
+
+The C compiler.
+
+=item C<$cxx>
+
+The C++ compiler.
+
+=item C<$cflags>
+
+Compiler flags (C or C++).
+
+=item C<@directive>
+
+A list of directive sets to measure against each other.
+
+=item C<$iterations>
+
+The number of times the parser is run for a bench.
+
+=item C<$verbose>
+
+Verbosity level.
+
+=back
+
+=cut
+
+my $bench;
 my $bison = $ENV{'BISON'} || '@abs_top_builddir@/tests/bison';
 my $cc = $ENV{'CC'} || 'gcc';
+my $cxx = $ENV{'CXX'} || 'g++';
+my $cflags = '-O2';
+my @directive = ();
+my $grammar = 'calc';
+my $iterations = -1;
+my $verbose = 1;
 
-##################################################################
-
-=head2 Functions
+=head1 FUNCTIONS
 
 =over 4
 
+=item C<verbose($level, $message)>
+
+Report the C<$message> is C<$level> E<lt>= C<$verbose>.
+
+=cut
+
+sub verbose($$)
+{
+  my ($level, $message) = @_;
+  print STDERR $message
+    if $level <= $verbose;
+}
+
+
+######################################################################
+
 =item C<directives($bench, @directive)>
 
 Format the list of directives for Bison for bench named C<$bench>.
@@ -47,14 +207,16 @@ Format the list of directives for Bison for bench named C<$bench>.
 
 sub directives($@)
 {
-  my ($bench, @directives) = @_;
+  my ($bench, @directive) = @_;
   my $res = "/* Directives for bench `$bench'. */\n";
-  $res .= join ("\n", @directives);
+  $res .= join ("\n", @directive) . "\n";
   $res .= "/* End of directives for bench `$bench'. */\n";
   return $res;
 }
 
-=item C<triangular_grammar ($base, $max, @directives)>
+######################################################################
+
+=item C<generate_grammar_triangular ($base, $max, @directive)>
 
 Create a large triangular grammar which looks like :
 
@@ -72,18 +234,18 @@ Create a large triangular grammar which looks like :
   | "1" "2" "3" "4" "5"  END    { $$ = 5; }
   ;
 
-C<$base> is the base name for the file to create (C<$base.y>).
+C<$base> is the base name for the file to create (F<$base.y>).
 C<$max> is the number of such rules (here, 5).  You may pass
-additional Bison C<@directives>.
+additional Bison C<@directive>.
 
 The created parser is self contained: it includes its scanner, and
 source of input.
 =cut
 
-sub triangular_grammar ($$$)
+sub generate_grammar_triangular ($$@)
 {
-  my ($base, $max, @directives) = @_;
-  my $directives = directives ($base, @directives);
+  my ($base, $max, @directive) = @_;
+  my $directives = directives ($base, @directive);
 
   my $out = new IO::File ">$base.y"
     or die;
@@ -171,7 +333,7 @@ EOF
 
 =item C<calc_input ($base, $max)>
 
-Generate the input file C<$base.input> for the calc parser.  The input
+Generate the input file F<$base.input> for the calc parser.  The input
 is composed of two expressions.  The first one is using left recursion
 only and consumes no stack.  The second one requires a deep stack.
 These two expressions are repeated C<$max> times in the output file.
@@ -191,18 +353,23 @@ sub calc_input ($$)
 }
 
 ##################################################################
-=item C<calc_grammar ($base, $max, @directives)>
 
-Generate a Bison file C<$base.y> that for a calculator parser in C.
-Pass the additional Bison C<@directives>.  C<$max> is ignored, but
-left to have the same interface as C<triangular_grammar>.
+=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
+have the same interface as C<triangular_grammar>.
 
 =cut
 
-sub calc_grammar ($$$)
+sub generate_grammar_calc ($$@)
 {
-  my ($base, $max, @directives) = @_;
-  my $directives = directives ($base, @directives);
+  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;
@@ -401,57 +568,324 @@ EOF
 
 ##################################################################
 
+=item C<generate_grammar_list ($base, $max, @directive)>
+
+Generate a Bison file F<$base.y> for a C++ parser that uses C++
+objects (std::string, std::list).  Tailored for using %define variant.
+
+=cut
+
+sub generate_grammar_list ($$@)
+{
+  my ($base, $max, @directive) = @_;
+  my $directives = directives ($base, @directive);
+  my $variant = grep { /%define "?variant"?/ } @directive;
+  my $lex_symbol = grep { /%define "?lex_symbol"?/ } @directive;
+  my $out = new IO::File ">$base.y"
+    or die;
+  print $out <<EOF;
+%language "C++"
+%defines
+%locations
+$directives
+
+%code requires // *.h
+{
+#include <string>
+}
+
+%code // *.c
+{
+#include <algorithm>
+#include <iostream>
+#include <sstream>
+
+#define STAGE_MAX    ($max * 10) // max = $max
+
+#define USE_LEX_SYMBOL $lex_symbol
+#define USE_VARIANTS $variant
+
+  // Prototype of the yylex function providing subsequent tokens.
+  static
+#if USE_LEX_SYMBOL
+  yy::parser::symbol_type yylex();
+#else
+  yy::parser::token_type yylex(yy::parser::semantic_type* yylval,
+                               yy::parser::location_type* yylloc);
+#endif
+
+  // Conversion to string.
+  template <typename T>
+    inline
+    std::string
+    string_cast (const T& t)
+  {
+    std::ostringstream o;
+    o << t;
+    return o.str ();
+  }
+}
+
+%token END_OF_FILE 0
+EOF
+
+  if ($variant)
+    {
+      print $out <<'EOF';
+%token <std::string> TEXT
+%token <int> NUMBER
+%printer { std::cerr << "Number: " << $$; } <int>
+%printer { std::cerr << "Text: " << $$; } <std::string>
+%type <std::string> text result
+
+%%
+result:
+  text			{ /* Throw away the result. */ }
+;
+
+text:
+  /* nothing */		{ /* This will generate an empty string */ }
+| text TEXT		{ std::swap ($$, $2); }
+| text NUMBER		{ $$ = string_cast($2); }
+;
+EOF
+    }
+  else
+    {
+      # Not using Bison variants.
+      print $out <<'EOF';
+%union {int ival; std::string* sval;}
+%token <sval> TEXT
+%token <ival> NUMBER
+%printer { std::cerr << "Number: " << $$; } <ival>
+%printer { std::cerr << "Text: " << *$$; } <sval>
+%type <sval> text result
+
+%%
+result:
+  text			{ delete $1; }
+;
+
+text:
+  /* nothing */		{ $$ = new std::string; }
+| text TEXT		{ delete $1; $$ = $2; }
+| text NUMBER		{ delete $1; $$ = new std::string (string_cast ($2)); }
+;
+EOF
+    }
+
+  print $out <<'EOF';
+%%
+#
+
+static
+#if USE_LEX_SYMBOL
+yy::parser::symbol_type yylex()
+#else
+yy::parser::token_type yylex(yy::parser::semantic_type* yylval,
+                             yy::parser::location_type* yylloc)
+#endif
+{
+  typedef yy::parser::token token;
+  static int stage = -1;
+  ++stage;
+  if (stage == STAGE_MAX)
+    {
+#if USE_LEX_SYMBOL
+      return yy::parser::make_symbol <token::END_OF_FILE> (yy::location());
+#else
+      *yylloc = yy::location ();
+      return token::END_OF_FILE;
+#endif
+    }
+  else if (stage % 2)
+    {
+#if USE_LEX_SYMBOL
+      return yy::parser::make_symbol <token::NUMBER> (stage, yy::location());
+#elif defined ONE_STAGE_BUILD
+      yylval->build(stage);
+      *yylloc = yy::location ();
+      return token::NUMBER;
+#elif USE_VARIANTS
+      yylval->build<int>() = stage;
+      *yylloc = yy::location ();
+      return token::NUMBER;
+#else
+      yylval->ival = stage;
+      *yylloc = yy::location ();
+      return token::NUMBER;
+#endif
+    }
+  else
+    {
+#if USE_LEX_SYMBOL
+      return yy::parser::make_symbol <token::TEXT> ("A string.", yy::location());
+#elif defined ONE_STAGE_BUILD
+      yylval->build(std::string("A string."));
+      *yylloc = yy::location ();
+      return token::TEXT;
+#elif USE_VARIANTS
+      yylval->build<std::string>() = std::string("A string.");
+      *yylloc = yy::location ();
+      return token::TEXT;
+#else
+      yylval->sval = new std::string("A string.");
+      *yylloc = yy::location ();
+      return token::TEXT;
+#endif
+    }
+  abort();
+}
+
+// Mandatory error function
+void
+yy::parser::error(const yy::parser::location_type& yylloc,
+                  const std::string& message)
+{
+  std::cerr << yylloc << ": " << message << std::endl;
+}
+
+int main(int argc, char *argv[])
+{
+  yy::parser p;
+#if YYDEBUG
+  p.set_debug_level(!!getenv("YYDEBUG"));
+#endif
+  p.parse();
+  return 0;
+}
+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 3, "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 3, "$command\n";
+  system ("$command") == 0
+    or die "$command failed";
+}
+
+##################################################################
+
 =item C<compile ($base)>
 
-Compile C<$base.y> to an executable C<$base> using the C compiler.
+Compile C<$base.y> to an executable C, Using the C or C++ compiler
+depending on the %language specification in C<$base.y>.
 
 =cut
 
 sub compile ($)
 {
   my ($base) = @_;
-  system ("$bison $base.y -o $base.c") == 0
-    or die;
-  system ("$cc -o $base $base.c") == 0
-    or die;
-}
+  my $language = `sed -ne '/%language "\\(.*\\)"/{s//\\1/;p;q;}' $base.y`;
+  chomp $language;
+
+  my $compiler = $language eq 'C++' ? $cxx : $cc;
 
-=item C<bench_grammar ($gram, %bench)>
+  run "$bison $base.y -o $base.c";
+  run "$compiler -o $base $cflags $base.c";
+}
 
-Generate benches for C<$gram>.  C<$gram> should be C<calc> or
-C<triangle>.  C<%bench> is a hash of the form:
+######################################################################
 
-  C<$name> => C<@directives>
+=item C<bench ($grammar, @token)>
 
-where C<$name> is the name of the bench, and C<@directives> are the
-Bison directive to use for this bench.  All the benches are compared
-against each other, repeated 50 times.
+Generate benches for the C<$grammar> and the directive specification
+given in the list of C<@token>.
 
 =cut
 
-sub bench_grammar ($%)
+sub bench ($@)
 {
-  my ($gram, %test) = @_;
+  my ($grammar, @token) = @_;
+  use Benchmark qw (:all :hireswallclock);
+
+  my @directive = parse (@token);
 
   # Set up the benches as expected by timethese.
   my %bench;
-  while (my ($name, $directives) = each %test)
+  # A counter of directive sets.
+  my $count = 1;
+  for my $d (@directive)
     {
-      print STDERR "$name\n";
-      # Call the Bison input file generator.
-      my $generator = "$gram" . "_grammar";
-      &$generator ($name, 200, @$directives);
+      $bench{$count} = $d;
+      printf " %2d. %s\n", $count, join (' ', split ("\n", $d));
+      $count++;
+    };
+
+  # For each bench, capture the size.
+  my %size;
+
+  while (my ($name, $directives) = each %bench)
+    {
+      generate_grammar ($grammar, $name, $directives);
+      # Compile the executable.
       compile ($name);
       $bench{$name} = "system ('./$name');";
+      chop($size{$name} = `wc -c <$name`);
     }
 
-  print "$gram:\n";
   # Run the benches.
-  my $res = timethese (50, \%bench, 'nop');
-  # Output the result.
+  #
+  # STYLE can be any of 'all', 'none', 'noc', 'nop' or 'auto'.  'all'
+  # shows each of the 5 times available ('wallclock' time, user time,
+  # system time, user time of children, and system time of
+  # children). 'noc' shows all except the two children times. 'nop'
+  # 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 3, "Running the benches for $grammar\n";
+  my $res = timethese ($iterations, \%bench, 'nop');
+
+  # Output the speed result.
   cmpthese ($res, 'nop');
+
+  # Display the sizes.
+  print "Sizes (decreasing):\n";
+  my $width = 10;
+  for my $bench (keys %size)
+    {
+      $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 ()>
 
@@ -462,20 +896,256 @@ interfaces.
 
 sub bench_push_parser ()
 {
-  print STDERR "Using $bison, $cc.\n";
-  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',
+         qw(
+            [ %d api.pure ]
+            &
+            [ %d api.push_pull=both ]
+         ));
+}
+
+######################################################################
+
+=item C<bench_variant_parser ()>
+
+Bench the C++ lalr1.cc parser using variants or %union.
+
+=cut
+
+sub bench_variant_parser ()
+{
+  bench ('list',
+         qw(
+            [
+              %d variant
+              &
+              [ #d ONE_STAGE_BUILD | %d lex_symbol ]
+            ]
+         )
     );
 }
 
-bench_push_parser();
+######################################################################
+
+=item C<bench_fusion_parser ()>
+
+Bench the C++ lalr1.cc parser using Boost.Variants or %union.
+
+=cut
+
+sub bench_fusion_parser ()
+{
+  bench ('list',
+         qw(
+             %s lalr1-split.cc
+           |
+             %s lalr1.cc
+         )
+    );
+}
+
+############################################################################
+
+sub help ($)
+{
+  my ($verbose) = @_;
+  use Pod::Usage;
+  # See <URL:http://perldoc.perl.org/pod2man.html#NOTES>.
+  pod2usage( { -message => "Bench Bison parsers",
+               -exitval => 0,
+               -verbose => $verbose,
+               -output  => \*STDOUT });
+}
+
+######################################################################
+
+# The end of the directives to parse.
+my $eod = "end of directives";
+# The list of tokens parsed by the following functions.
+my @token;
+
+# eat ($EXPECTED)
+# ---------------
+# Check that the current token is $EXPECTED, and move to the next.
+sub eat ($)
+{
+  my ($expected) = @_;
+  die "expected $expected, unexpected: $token[0] (@token)\n"
+    unless $token[0] eq $expected;
+  shift @token;
+}
+
+# Parse directive specifications:
+#   expr: term (| term)*
+#   term: fact (& fact)*
+#   fact: ( expr ) | [ expr ] | dirs
+#   dirs: %s SKELETON | #d NAME[=VALUE] | %d NAME[=VALUE] | directive
+sub parse (@)
+{
+  @token = (@_, $eod);
+  verbose 3, "Parsing: @token\n";
+  my @res = parse_expr ();
+  eat ($eod);
+  return @res;
+}
+
+sub parse_expr ()
+{
+  my @res = parse_term ();
+  while ($token[0] eq '|')
+    {
+      eat ('|');
+      # Alternation.
+      push @res, parse_term ();
+    }
+  return @res;
+}
+
+sub parse_term ()
+{
+  my @res = parse_fact ();
+  while ($token[0] eq '&')
+    {
+      eat ('&');
+      # Cartesian product.
+      my @lhs = @res;
+      @res = ();
+      for my $rhs (parse_fact ())
+        {
+          for my $lhs (@lhs)
+            {
+              push @res, $lhs . ($lhs && $rhs ? "\n" : "") . $rhs;
+            }
+        }
+    }
+  return @res;
+}
+
+sub parse_fact ()
+{
+  my @res;
+  die "unexpected end of expression"
+    unless defined $token[0];
+
+  if ($token[0] eq '(')
+    {
+      eat ('(');
+      @res = parse_expr ();
+      eat (')');
+    }
+  elsif ($token[0] eq '[')
+    {
+      eat ('[');
+      @res = (parse_expr (), '');
+      eat (']');
+    }
+  else
+    {
+      @res = parse_dirs ();
+    }
+  return @res;
+}
+
+sub parse_dirs ()
+{
+  my @res;
+  die "unexpected end of expression"
+    unless defined $token[0];
+
+  if ($token[0] eq '#d')
+    {
+      eat ('#d');
+      $token[0] =~ s/(.*?)=(.*)/$1 $2/;
+      @res = ("%code {\n#define $token[0]\n}");
+      shift @token;
+    }
+  elsif ($token[0] eq '%d')
+    {
+      shift @token;
+      $token[0] =~ s/(.*?)=(.*)/$1 "$2"/;
+      @res = ("%define $token[0]");
+      shift @token;
+    }
+  elsif ($token[0] eq '%s')
+    {
+      shift @token;
+      @res = ("%skeleton \"$token[0]\"");
+      shift @token;
+    }
+  else
+    {
+      @res = $token[0];
+      shift @token;
+    }
+
+  return @res;
+}
+
+######################################################################
+
+sub getopt ()
+{
+  use Getopt::Long;
+  my %option = (
+    "b|bench=s"      => \$bench,
+    "c|cflags=s"     => \$cflags,
+    "d|directive=s"  => \@directive,
+    "g|grammar=s"    => \$grammar,
+    "h|help"         => sub { help ($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;
+
+# Create the directory we work in.
+mkdir "benches" or die "cannot create benches"
+  unless -d "benches";
+my $count = 1;
+++$count
+  while -d "benches/$count";
+my $dir = "benches/$count";
+mkdir $dir
+  or die "cannot create $dir";
+chdir $dir
+  or die "cannot chdir $dir";
+
+# The following message is tailored to please Emacs' compilation-mode.
+verbose 1, "Entering directory `$dir'\n";
+verbose 1, "Using bison=$bison.\n";
+verbose 2, "Using cc=$cc.\n";
+verbose 2, "Using cxx=$cxx.\n";
+verbose 2, "Using cflags=$cflags.\n";
+verbose 2, "Grammar: $grammar\n";
+
+
+# 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;
+}
+else
+{
+  # Launch the bench marking.
+  bench ($grammar, @ARGV);
+}
 
 ### Setup "GNU" style for perl-mode and cperl-mode.
 ## Local Variables: