In Fastest way to lookup a point in a set we concluded, in isolation, that hash lookups were faster using split/join rather than pack/unpack.

With nothing to lose though, I tried changing split /:/, $z to unpack 'ii', $z and

join ':', $x - 1, $y - 1
join ':', @_
[download]

to:

pack 'ii', $x - 1, $y - 1
pack 'ii', @_
[download]

... and almost fell off my chair when the run time for three million cells dropped from 287 seconds way down to 204 seconds!! What gives?

As shown below, there is no difference in the number of op codes:

> perl -MO=Terse -e "split /:/, $z"
LISTOP (0x2bf88f8) leave [1]
    OP (0x25b59a0) enter
    COP (0x2bf8940) nextstate
    LISTOP (0x25b59d8) split [2]
        PMOP (0x25b5a60) pushre
        UNOP (0x25b5a20) null [15]
            PADOP (0x25b5ad0) gvsv  GV (0x25b0660) *z
        SVOP (0x25b5938) const [3] IV (0x25b0d80) 0
> perl -MO=Terse -e "unpack 'ii', $z"
LISTOP (0x2c387f8) leave [1]
    OP (0x645908) enter
    COP (0x645940) nextstate
    LISTOP (0x6459d8) unpack
        OP (0x6459a0) null [3]
        SVOP (0x645aa0) const [2] PV (0x63fd40) "ii"
        UNOP (0x645a20) null [15]
            PADOP (0x645a60) gvsv  GV (0x63fe00) *z
> perl -MO=Terse -e "join ':', @_"
LISTOP (0x2a9d798) leave [1]
    OP (0x24f5938) enter
    COP (0x24f5970) nextstate
    LISTOP (0x24f5a08) join [3]
        OP (0x24f59d0) pushmark
        SVOP (0x24f5ad0) const [4] PV (0x24f0960) ":"
        UNOP (0x24f5a50) rv2av [2]
            PADOP (0x24f5a90) gv  GV (0xe9b2c8) *_
> perl -MO=Terse -e "pack 'ii', @_"
LISTOP (0x2c4e318) leave [1]
    OP (0x645938) enter
    COP (0x645970) nextstate
    LISTOP (0x645a08) pack [3]
        OP (0x6459d0) pushmark
        SVOP (0x645ad0) const [4] PV (0x640a80) "ii"
        UNOP (0x645a50) rv2av [2]
            PADOP (0x645a90) gv  GV (0xffb2c8) *_
[download]

There didn't appear to be any significant difference in memory consumption either.

Anyone got any ideas? The slowdown may be caused by split using a /:/ regex - and regexes are slow. Note that in Fastest way to lookup a point in a set, we were measuring lookups in isolation and lookups use join, not split. How to investigate further? Devel::NYTProf?

Update:: From running:

for my $r ( [-123456789, 987654321], [1,2] ) {
   my $pp = pack 'ii', @{$r};
   my $jj = join ':', @{$r};
   my $pplen = length $pp;
   my $jjlen = length $jj;
   print "$r->[0]:$r->[1] packlen=$pplen joinlen=$jjlen\n";
   my ($xpp, $ypp) = unpack 'ii', $pp;
   my ($xjj, $yjj) = split /:/, $jj;
   $xpp == $r->[0] or die;
   $ypp == $r->[1] or die;
   $xjj == $r->[0] or die;
   $yjj == $r->[1] or die;
}
[download]

we see:

-123456789:987654321 packlen=8 joinlen=20
1:2 packlen=8 joinlen=3
[download]

That is, pack/unpack always has a hash key length of 8 bytes, while with split/join the key length varies, depending on the size of the x and y coordinates.

Update: As discovered by marioroy, 'i2' is faster than 'ii' in pack and unpack.

---

---

• Are you posting in the right place? Check out Where do I post X? to know for sure.
• Posts may use any of the Perl Monks Approved HTML tags. Currently these include the following:

  <code> <a> <b> <big> <blockquote> <br /> <dd> <dl> <dt> <em> <font> <h1> <h2> <h3> <h4> <h5> <h6> <hr /> <i> <li> <nbsp> <ol> <p> <small> <strike> <strong> <sub> <sup> <table> <td> <th> <tr> <tt> <u> <ul>

• Snippets of code should be wrapped in <code> tags not <pre> tags. In fact, <pre> tags should generally be avoided. If they must be used, extreme care should be taken to ensure that their contents do not have long lines (<70 chars), in order to prevent horizontal scrolling (and possible janitor intervention).
• Want more info? How to link or How to display code and escape characters are good places to start.