Front page | perl.perl5.porters |
Postings from December 2000
On Sun, Dec 17, 2000 at 11:27:32AM -0600, Jarkko Hietaniemi wrote:
> On Sun, Dec 17, 2000 at 04:29:24PM +0000, Nicholas Clark wrote:
> > Why do all the XS constant() routines (that I've looked at) have a second
> > unused integer argument after the name argument?
> > I assume that they return double because perl did all arithmetic (except
> > logical operations such as |) as doubles
>
> I think pretty much all the Perl XS extensions have started their life
> as h2xs outputs, even if they have been mucked around by hand quite
> extensively since then -- and h2xs still has these lines:
>
> print $fh <<"END";
> static double
> constant(char *name, int len, int arg)
> {
> errno = 0;
>
> Not even NV, "double". Sigh.
But the "int len" for the string's length *is* a nifty idea.
So the guilding the lilly department presents a revised Fcntl.xs that
is faster (when you measure it over a 1,000,000 loop) *and* smaller.
Not that this really makes any noticeable difference to the regression
tests. :-) But it's the thought of it being faster that counts.
Before
text data bss dec hex filename
5944 0 0 5944 1738 ext/Fcntl/Fcntl.o
All tests successful.
u=1.77 s=0.47 cu=71.86 cs=10.67 scripts=270 tests=20850
real 2m20.023s
user 1m13.630s
sys 0m11.140s
After
text data bss dec hex filename
5512 0 0 5512 1588 ext/Fcntl/Fcntl.o
All tests successful.
u=1.65 s=0.47 cu=71.12 cs=11.93 scripts=270 tests=20850
real 2m15.466s
user 1m12.770s
sys 0m12.400s
(and I get more variation in those timings than the difference between
them. Benchmarking the constant code shows it to be slightly faster)
not-a-patch because the patch would be longer than the file.
You don't have to use this if you don't feel the need.
Nicholas Clark
Thread Previous
|
Thread Next