| 1 |
#!/usr/bin/env perl |
| 2 |
|
| 3 |
use Getopt::Std; |
| 4 |
$DEBUG = 0; |
| 5 |
|
| 6 |
sub parse_objdump_file { |
| 7 |
my ($filename) = @_; |
| 8 |
my @result; |
| 9 |
open (INPUT, $filename) or die "$filename: $!\n"; |
| 10 |
print "opened objdump output file $filename\n" if $DEBUG; |
| 11 |
while (<INPUT>) { |
| 12 |
if (/\s*([0-9a-f]*):\t(([0-9a-f]{2} )+) *\t(.*)$/) { |
| 13 |
my ($addr, $bytes, $instr) = ($1, $2, $4); |
| 14 |
$addr = "0x" . $addr; |
| 15 |
$bytes =~ s/\s*(.*\S)\s*/$1/; # trim any remaining whitespace |
| 16 |
$instr =~ s/\s*(.*\S)\s*/$1/; |
| 17 |
push (@result, {'addr' => $addr, 'bytes' => $bytes, 'instr' => $instr}); |
| 18 |
print "addr=$addr bytes='$bytes' instr='$instr'\n" if $DEBUG; |
| 19 |
} |
| 20 |
} |
| 21 |
close INPUT; |
| 22 |
return @result; |
| 23 |
} |
| 24 |
|
| 25 |
sub parse_gdb_file { |
| 26 |
my ($filename) = @_; |
| 27 |
my @result; |
| 28 |
my $got_addr; |
| 29 |
open (INPUT, $filename) or die "$filename: $!\n"; |
| 30 |
print "opened gdb output file $filename\n" if $DEBUG; |
| 31 |
while (<INPUT>) { |
| 32 |
if (/^(0x[0-9a-f]*):\t([^\t]*)\t[^:]*:\t((0x[0-9a-f]{2}\s*)+)\s*$/) { |
| 33 |
my ($addr, $bytes, $instr) = ($1, $3, $2); |
| 34 |
$bytes =~ s/0x//g; |
| 35 |
$bytes =~ s/\s+/ /g; # regularize whitespace |
| 36 |
$bytes =~ s/\s*(.*\S)\s*/$1/; # trim any remaining whitespace |
| 37 |
$instr =~ s/\s*(.*\S)\s*/$1/; |
| 38 |
push (@result, {'addr' => $addr, 'bytes' => $bytes, 'instr' => $instr}); |
| 39 |
print "addr=$addr bytes='$bytes' instr='$instr'\n" if $DEBUG; |
| 40 |
} elsif (/^(0x[0-9a-f]*):\t$/) { # deal with gdb's line breaker |
| 41 |
$got_addr = $1; |
| 42 |
} elsif ($got_addr && /^ ([^\t]*)\t[^:]*:\t((0x[0-9a-f]{2}\s*)+)\s*$/) { |
| 43 |
my ($addr, $bytes, $instr) = ($got_addr, $2, $1); |
| 44 |
$bytes =~ s/0x//g; |
| 45 |
$bytes =~ s/\s+/ /g; # regularize whitespace |
| 46 |
$bytes =~ s/\s*(.*\S)\s*/$1/; # trim any remaining whitespace |
| 47 |
$instr =~ s/\s*(.*\S)\s*/$1/; |
| 48 |
push (@result, {'addr' => $addr, 'bytes' => $bytes, 'instr' => $instr}); |
| 49 |
print "addr=$addr bytes='$bytes' instr='$instr'\n" if $DEBUG; |
| 50 |
undef $got_addr; |
| 51 |
} |
| 52 |
} |
| 53 |
close INPUT; |
| 54 |
return @result; |
| 55 |
} |
| 56 |
|
| 57 |
sub binary_diffs { |
| 58 |
my ($objdump_file, $gdb_file) = @_; |
| 59 |
my @file1 = parse_objdump_file ($objdump_file); |
| 60 |
my @file2 = parse_gdb_file ($gdb_file); |
| 61 |
my $lastrecord = ($#file1 >= $#file2) ? ($#file1) : ($#file2); |
| 62 |
for (my $i = 0; $i <= $lastrecord; ++$i) { |
| 63 |
my $d1 = $file1[$i]; |
| 64 |
my $d2 = $file2[$i]; |
| 65 |
if ($d1->{'bytes'} ne $d2->{'bytes'}) { |
| 66 |
next if (($d1->{'instr'} eq $d2->{'instr'}) && $opt_d); |
| 67 |
printf "0x%08x:\t%30s \t%s\n", 0+$d1->{'addr'}, $d1->{'bytes'}, $d1->{'instr'}; |
| 68 |
printf "0x%08x:\t%30s \t%s\n\n", 0+$d2->{'addr'}, $d2->{'bytes'}, $d2->{'instr'}; |
| 69 |
} |
| 70 |
} |
| 71 |
} |
| 72 |
|
| 73 |
&getopts('d'); |
| 74 |
$objdump_file = $ARGV[0]; |
| 75 |
$gdb_file = $ARGV[1]; |
| 76 |
binary_diffs ($objdump_file, $gdb_file); |
| 77 |
exit (0); |
| 78 |
__END__ |
| 79 |
=pod |
| 80 |
|
| 81 |
=head1 NAME |
| 82 |
|
| 83 |
codegen-diff |
| 84 |
|
| 85 |
=head1 SYNOPSIS |
| 86 |
|
| 87 |
codegen-diff [-d] I<OBJDUMP-OUTPUT-FILE> I<GDB-DISASSEMBLY-FILE> |
| 88 |
|
| 89 |
=head1 DESCRIPTION |
| 90 |
|
| 91 |
B<codegen-diff> is a program that tries to show you the differences |
| 92 |
between the code that B<llc> generated and the code that B<lli> generated. |
| 93 |
|
| 94 |
The way you use it is as follows: first, you create I<OBJDUMP-OUTPUT-FILE> |
| 95 |
by running B<objdump> on the B<llc> compiled and linked binary. You need to |
| 96 |
trim down the result so it contains only the function of interest. |
| 97 |
|
| 98 |
Second, you create I<GDB-DISASSEMBLY-FILE> by running B<gdb>, with my patch |
| 99 |
to print out hex bytes in the B<disassemble> command output, on |
| 100 |
B<lli>. Set a breakpoint in C<Emitter::finishFunction()> and wait until |
| 101 |
the function you want is compiled. Then use the B<disassemble> command |
| 102 |
to print out the assembly dump of the function B<lli> just compiled. |
| 103 |
(Use C<lli -debug> to find out where the function starts and ends in memory.) |
| 104 |
It's easiest to save this output by using B<script>. |
| 105 |
|
| 106 |
Finally, you run B<codegen-diff>, as indicated in the Synopsis section of |
| 107 |
this manpage. It will print out a two-line stanza for each mismatched |
| 108 |
instruction, with the B<llc> version first, and the B<lli> version second. |
| 109 |
|
| 110 |
=head1 OPTIONS |
| 111 |
|
| 112 |
=over 4 |
| 113 |
|
| 114 |
=item -d |
| 115 |
|
| 116 |
Don't show instructions where the bytes are different but they |
| 117 |
disassemble to the same thing. This puts a lot of trust in the |
| 118 |
disassembler, but it might help you highlight the more egregious cases |
| 119 |
of misassembly. |
| 120 |
|
| 121 |
=back |
| 122 |
|
| 123 |
=head1 AUTHOR |
| 124 |
|
| 125 |
B<codegen-diff> was written by Brian Gaeke. |
| 126 |
|
| 127 |
=head1 SEE ALSO |
| 128 |
|
| 129 |
L<gdb(1)>, L<objdump(1)>, L<script(1)>. |
| 130 |
|
| 131 |
You will need my B<gdb> patch: |
| 132 |
|
| 133 |
http://llvm.cs.uiuc.edu/~gaeke/gdb-disassembly-print-bytes.patch |
| 134 |
|
| 135 |
=cut |