# Position of arguments in function definition and performance

Hal Daume III hdaume@ISI.EDU
Wed, 6 Feb 2002 23:45:55 -0800 (PST)

Well, I assume you meant:

reverse1 [] ys =3D ys
reverse1 (x:xs) ys =3D reverse1 xs (x:ys)

reverse2 ys [] =3D ys
reverse2 ys (x:xs) =3D reverse1 (x:ys) xs

If so, and you make two programs:

main =3D print (length \$! reverse1 [1..2000000] [])

and

main =3D print (length \$! reverse2 [] [1..2000000])

compile them with ghc -O2 -fvia-c, and time them we get:

FOR REVERSE1:

11:42pm enescu:~/ time a.out
2000000
4.84u 0.28s 0:06.01 85.1%
11:42pm enescu:~/ time a.out
2000000
4.71u 0.24s 0:05.25 94.2%

FOR REVERSE2:

11:43pm enescu:~/ time a.out
2000000
1.00u 0.03s 0:01.09 94.4%
11:43pm enescu:~/ time a.out
2000000
0.99u 0.01s 0:00.99 101.0%

curiously, REVERSE2 did significantly better; I have no idea why.  Perhaps
one of the Simons could comment on this.  Moreover, if this is a general
phenomenon, why doesn't GHC simply permute the order of parameters to
allow it to optimize best?

Regards,

Hal

--
Hal Daume III

"Computer science is no more about computers    | hdaume@isi.edu
than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

On Wed, 6 Feb 2002, [iso-8859-1] Jos=E9 Romildo Malaquias wrote:

> Hello.
>=20
> Please, tell me which set of definitions below should I expected
> to be more efficient: the reverse1 or the reverse2 functions.
>=20
> reverse1 []     ys =3D ys
> reverse1 (x:xs) ys =3D reverse2 (x:ys) xs
>=20
> reverse2 ys []     =3D ys
> reverse2 ys (x:xs) =3D reverse2 (x:ys) xs
>=20
> The difference rely on the position of the argument in which the
> pattern matching is done in the function definition.
>=20
> Regards.
>=20
> Romildo
> --=20
> Prof. Jos=E9 Romildo Malaquias               Departamento de Computa=E7=
=E3o
> http://iceb.ufop.br/~romildo       Universidade Federal de Ouro Preto
> romildo@iceb.ufop.br                                           Brasil
> romildo@uber.com.br
> _______________________________________________