----- Original Message -----
From: "Eric Wilhelm" <[EMAIL PROTECTED]>
To: <inline@perl.org>
Sent: Friday, January 19, 2007 7:51 PM
Subject: Re: Percentiles
# from Sisyphus
# on Thursday 18 January 2007 09:51 pm:
In terms of performance, I don't think it matters much which way you
go about it.
If the caller is routinely passing extremely large arrays, passing as an
array ref should be quite a bit faster than a list of values (each of
which gets copied.)
Well ... I can't detect a significant difference for a 10,000,000 element
array of integers. Perhaps I've been sucked in by a benchmarking sophism.
(Apologies if that's the case.) I also suspect (though I haven't checked
recently), that if you look at the generated .c file, you'll see that
there's buggerall difference.
Here's my test script:
-----------------------------
use warnings;
use Benchmark;
use Inline C => <<'EOC';
void by_list(SV * x, ...) {
Inline_Stack_Vars;
int z, i, len = Inline_Stack_Items;
for(i = 0; i < len; i++) {
z = SvIV(Inline_Stack_Item(i));
//printf("%d\n", z); // to verify that values are correct
}
}
void by_ref(AV * x) {
int len;
int i,z;
SV ** elem;
len = av_len(x) + 1;
for(i = 0; i < len; i++) {
z = SvIV(*(av_fetch(x, i, 0)));
//printf("%d\n", z); // to verify that values are correct
}
}
EOC
@array = ();
$elements = 10000000;
for (1 .. $elements) {push @array, $_}
print "DONE\n";
timethese(1, {
'by_list' => 'by_list(@array);',
'by_ref' => 'by_ref([EMAIL PROTECTED]);',
});
--------------------------------------
For me, that produces:
--------------------------------------
DONE
Benchmark: timing 1 iterations of by_list, by_ref...
by_list: 0 wallclock secs ( 0.47 usr + 0.00 sys = 0.47 CPU) @ 2.13/s
(n=1
)
(warning: too few iterations for a reliable count)
by_ref: 1 wallclock secs ( 0.45 usr + 0.00 sys = 0.45 CPU) @ 2.21/s
(n=1
)
(warning: too few iterations for a reliable count)
------------------------------------
Cheers,
Rob
