[EMAIL PROTECTED] wrote:
I've written some files to go into examples/benchmark which time the raw throughput of the md5 part, which excluding the load/compile phases.
That's not really needed. Startup time is almost nothing with file sizes bigger then some MB. And below that, you don't get reasonable timings.
We can compare Digest::MD5 and python's md5sum.py against parrot -j. Both competitors are written in C --that's fine.
We're about a quarter of the throughput of pure C, which I think is darn good.
Yep, but let's compare with perl XS - we are around 1/2 perl XS speed now, see below.
(And we outperformed the pure Perl version by quite a bit, although I'm guessing the slow core has similar timings)
Well, we are comparing a pure Parrot version with a pure Perl version ;-)
Now its your turn to process files blockwise ;-)
I'll see what I can do over the weekend. I'm not totally convinced that the reduced memory consumption merits the extra code, but we'll see.
No extra code neeeded, just a bit different - have a look at Python's md5*.c - written very clearly + OO interface.
Ok lets compare some real time numbers:
parrot -j perl XS Python md5c.c md5sum of core (~400KB) [1] 0.10 0.07 0.06 ops/core_ops.c (~1MB) [2] 0.16 0.10 0.09 parrot (~4.5MB) [3] 0.50 0.25 0.22 perl-5.8.0.tar.gz(11MB)[4] 1.07 0.51 0.44 22mb [5] 2.10 0.98 0.86 100mb [6] 14.0 4.26 3.73
leo
[1] 401408 bytes [2] 1014056 [3] 4725838 [4] 11023084 [5] 22528000 # dd if=/dev/zero of=22mb count=4400 [6]102400000 # of=100mb count=20000
programs used: perl 5.8.0, python 2.4, parrot svn head
$ time parrot -j md5sum.pbc the_file # changed to load MD5.pbc and of course JIT/x86 with these recent hacks ;-)
$ time python ~/src/Python-2.4/Tools/scripts/md5sum.py the_file
$ time perl md5.pl the_file
$ l ~/src/perl-5.8.0.tar.gz [..] 11023084 Nov 30 2002 /home/lt/src/perl-5.8.0.tar.gz
# cat md5.pl
use lib 'lib'; #use Digest::Perl::MD5; use Digest::MD5;
my $file = shift; open(FILE, $file) or die "Can't open '$file': $!"; binmode(FILE);
# print Digest::Perl::MD5->new->addfile(*FILE)->hexdigest, " $file\n"; print Digest::MD5->new->addfile(*FILE)->hexdigest, " $file\n";
