Front page | perl.perl5.porters |
Postings from April 2008
Graham Barr wrote:
> On Apr 1, 2008, at 8:49 AM, Nicholas Clark wrote:
> > > @xs = map { f($_) } @ys;
> > >
> > > In that code, `map` will construct an array in memory, but when it
> > > returns it, the array will be copied cell by cell into a new array
> > > constructed as @xs. We can optimize this to remove that intermediate
> > > array by using the following code instead:
> > >
> > > @xs = ();
> > > push @xs, f($_) foreach @ys;
>
> I am not convinced that is always "better" performance. It depends what
> you metric is. While it may use less memory, it is more OPs to run and
> may result in slower runtime performance in some cases.
In all my benchmarking I was only looking at the runtime. By that metric
performance is much better (+60~+100%) for large lists, and approaching
equivalent as the size of lists decreases; which is consistent with
asymptotic differences due to building intermediate values, garbage
collecting them, and generally wrecking the cache.
Even at the low end of the spectrum I didn't notice the `map` version
being even marginally faster. I'll grant there may be some corner cases
where that happens, but I didn't find them.
Live well,
~wren
Thread Previous