gem5/tests/diff-out

410 lines
10 KiB
Text
Raw Normal View History

#!/usr/bin/perl
# Copyright (c) 2001-2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Steve Reinhardt
#
# This script diffs two SimpleScalar statistics output files.
#
use Getopt::Std;
#
# -t thresh sets threshold for ignoring differences (in %)
# -p sorts differences by % chg (default is alphabetic)
# -f ignores fetch-loss statistics
# -d ignores all distributions
#
getopts('dfn:pt:h');
if ($#ARGV < 1)
{
print "\nError: need two file arguments (<reference> <new>).\n";
print " Options: -d = Ignore distributions\n";
print " -f = Ignore fetch-loss stats\n";
print " -p = Sort errors by percentage\n";
print " -h = Diff header info separately from stats\n";
print " -n <num> = Print top <num> errors (default 20)\n";
print " -t <num> = Error threshold in percent (default 1)\n\n";
die -1;
}
open(REF, "<$ARGV[0]") or die "Error: can't open $ARGV[0].\n";
open(NEW, "<$ARGV[1]") or die "Error: can't open $ARGV[1].\n";
#
# Things that really should be adjustable via the command line
#
# Ignorable error (in percent)
$err_thresh = ($opt_t) ? $opt_t : 0;
# Number of stats to print before omitting
$omit_count = ($opt_n) ? $opt_n : 20;
#
# First copy everything up to the simulation statistics to a pair of
# temporary files, stripping out date-related items, and do a plain
# diff. Any differences in the arguments are not necessarily an issue;
# any differences in the program output should be caught by the EIO
# mechanism if an EIO file is used.
#
# copy_header takes input filehandle and output filename
sub copy_header
{
my ($inhandle, $outname) = @_;
open(OUTPUT, ">$outname") or die "Error: can't open $outname.\n";
while (<$inhandle>)
{
# strip out lines that can vary
next if /^(command line:|M5 compiled on |M5 simulation started |M5 executing on )/;
last if /Begin Simulation Statistics/;
print OUTPUT;
}
close OUTPUT;
}
if ($opt_h) {
# Diff header separately from stats
$refheader = "/tmp/smt-test.refheader.$$";
$newheader = "/tmp/smt-test.newheader.$$";
copy_header(\*REF, $refheader);
copy_header(\*NEW, $newheader);
print "\n===== Header and program output differences =====\n\n";
print `diff $refheader $newheader`;
print "\n===== Statistics differences =====\n\n";
}
#
# Now parse statistics
#
#
# This function takes an open filehandle and returns a reference to
# a hash containing all the statistics variables and their values.
#
sub parse_file
{
$stathandle = shift;
$in_dist = undef;
$hashref = { }; # initialize hash for values
while (<$stathandle>)
{
next if /^\s*$/; # skip blank lines
next if /^\*\*Ignore/; # temporary, to make totaling scripts easy for ISCA 03
last if /End Simulation Statistics/;
s/ *#.*//; # strip comments
if (/^Memory usage: (\d+) KBytes/) {
$stat = 'memory usage';
$value = $1;
}
elsif ($in_dist) {
if ($in_dist =~ /^fetch_loss_counters/) {
if (/^fetch_loss_counters_\d+\.end/) {
# end line of distribution: clear $in_dist flag
$in_dist = undef;
next;
}
else {
next if $opt_f;
($stat, $value) = /^(\S+)\s+(.*)/;
}
}
else {
if (/(.*)\.end_dist/) {
# end line of distribution: clear $in_dist flag
$in_dist = undef;
next;
}
if ($opt_d) {
next; # bail out if we are ignoring dists...
}
elsif (/(.*)\.(min|max)_value/) {
# treat these like normal stats
($stat, $value) = /^(\S+)\s+(.*)/;
}
else {
# this is ugly because labels in the distribution
# buckets don't start in column 0 and may include
# embedded spaces
($stat, $value) =
/^\s*(\S+(?:.*\S)?)\s+(\d+)\s+\d+\.\d+%/;
$stat = $in_dist . '::' . $stat;
}
}
}
else {
if (/(.*)\.start_dist/) {
# start line of distribution: set $in_dist flag
# and save distribution name for future reference
$in_dist = $1;
$stat = $1;
$value = 0;
}
elsif (/^(fetch_loss_counters_\d+)\.start/) {
# treat fetch loss counters like distribution, sort of
$in_dist = $1;
$stat = $1;
$value = 0;
}
else {
($stat, $value) = /^(\S+)\s+(.*)/;
}
}
$$hashref{$stat} = $value;
}
close($stathandle);
return $hashref;
}
#
# pct_diff($old, $new) returns percent difference from $old to $new.
#
sub pct_diff
{
my ($old, $new) = @_;
return ($old == 0) ? (($new == 0) ? 0 : 9999) : 100 * ($new - $old) / $old;
}
#
# Statistics to ignore: these relate to simulator performance, not
# correctness, so don't fail on changes here.
#
%ignore = (
'host_seconds' => 1,
'host_tick_rate' => 1,
'host_inst_rate' => 1,
'host_mem_usage' => 1
);
#
# List of key statistics (always displayed)
# ==> list stats here WITHOUT trailing thread ID
#
@key_stat_list = (
'COM:IPC',
'ISSUE:MSIPC',
'COM:count',
'host_inst_rate',
'sim_insts',
'sim_ticks',
'host_mem_usage'
);
$key_stat_pattern = join('|', @key_stat_list);
# initialize first statistics from each file
$max_err_mag = 0;
$refhash = parse_file(\*REF);
$newhash = parse_file(\*NEW);
# The string sim-smt prints on a divide by zero
$divbyzero = '<err: divide by zero>';
foreach $stat (sort keys %$refhash)
{
$refvalue = $$refhash{$stat};
$newvalue = $$newhash{$stat};
if (!defined($newvalue)) {
# stat missing from new file
push @missing_stats, $stat;
next;
}
if ($stat =~ /($key_stat_pattern)/o) {
# key statistics: always record & display changes in these
push @key_stats, [$stat, $refvalue, $newvalue];
}
if ($ignore{$stat} or $refvalue eq $newvalue) {
# stat is in "ignore" list, or hasn't changed
}
else {
if ($refvalue eq $divbyzero || $newvalue eq $divbyzero) {
# one or the other was a divide by zero:
# no point in trying to quantify error
print "$stat: $refvalue --> $newvalue\n";
}
else {
$reldiff = pct_diff($refvalue, $newvalue);
$diffmag = abs($reldiff);
if ($diffmag > $err_thresh) {
push @errs,
[$stat, $refvalue, $newvalue, $reldiff];
}
if ($diffmag > $max_err_mag) {
$max_err_mag = $diffmag;
}
}
}
# remove from new hash so we can detect added stats
delete $$newhash{$stat};
}
#
# All done. Print comparison summary.
#
printf("Maximum error magnitude: %+f%%\n\n", $max_err_mag);
printf(" %-30s %10s %10s %10s %7s\n", ' ', 'Reference', 'New Value', 'Abs Diff', 'Pct Chg');
printf("Key statistics:\n\n");
foreach $key_stat (@key_stats)
{
($statname, $refvalue, $newvalue, $reldiff) = @$key_stat;
# deduce format from reference value
$pointpos = rindex($refvalue, '.');
$digits = ($pointpos < 0) ? 0 :(length($refvalue) - $pointpos - 1);
$fmt = "%10.${digits}f";
# print differing values with absolute and relative error
printf(" %-30s $fmt $fmt $fmt %+7.2f%%\n",
$statname, $refvalue, $newvalue,
$newvalue - $refvalue, pct_diff($refvalue, $newvalue));
}
printf("\nLargest $omit_count relative errors (> %d%%):\n\n", $err_thresh);
$num_errs = 0;
if ($opt_p)
{
# sort differences by percent change
@errs = sort { abs($$b[3]) <=> abs($$a[3]) } @errs;
}
foreach $err (@errs)
{
($statname, $refvalue, $newvalue, $reldiff) = @$err;
# deduce format from reference value
$pointpos1 = rindex($refvalue, '.');
$digits1 = ($pointpos1 < 0) ? 0 :(length($refvalue) - $pointpos1 - 1);
$pointpos2 = rindex($newvalue, '.');
$digits2 = ($pointpos2 < 0) ? 0 :(length($newvalue) - $pointpos2 - 1);
$digits = ($digits1 > $digits2) ? $digits1 : $digits2;
$fmt = "%10.${digits}f";
# print differing values with absolute and relative error
printf(" %-30s $fmt $fmt $fmt %+7.2f%%\n",
$statname, $refvalue, $newvalue, $newvalue - $refvalue, $reldiff);
# only print top N errors
if (++$num_errs >= $omit_count)
{
print "[... additional errors omitted ...]\n";
last;
}
}
#
# Report missing stats, but first filter out distribution buckets:
# these are mostly noise
@missing_stats = grep { !/::(\d+|overflows)?$/ } @missing_stats;
# get count
$missing_stats = scalar(@missing_stats);
if ($missing_stats)
{
print "\nMissing $missing_stats reference statistics:\n\n";
foreach $stat (@missing_stats)
{
# print "\t$stat\n";
printf " %-50s ", $stat;
print "$$refhash{$stat}\n";
}
}
#
# Any stats left in newhash are added since the reference file
#
@added_stats = keys %$newhash;
# first filter out distribution buckets: mostly noise
@added_stats = grep { !/::(\d+|overflows)?$/ } @added_stats;
# get count
$added_stats = scalar(@added_stats);
if ($added_stats)
{
print "\nFound $added_stats new statistics:\n\n";
foreach $stat (sort @added_stats)
{
# print "\t$stat\n";
printf " %-50s ", $stat;
print "$$newhash{$stat}\n";
}
}
cleanup();
# Exit code is 0 if no stats error, 1 otherwise
$status = ($max_err_mag == 0.0) ? 0 : 1;
exit $status;
sub cleanup
{
unlink($refheader) if ($refheader);
unlink($newheader) if ($newheader);
}