[Haskell-cafe] Fwd: shootout

[Ben Gamari]

On Sun, 31 Jul 2011 02:27:07 +0200, Thorsten Hater <th at tp1.rub.de> wrote:
Non-text part: multipart/mixed
> Good Evening,
> 
> can anybody confirm that this implementation is somewhat faster
> than the current benchmark (at expense of memory consumption)?
> 
> Cheers, Thorsten
> 
Somewhat faster is an understatement I would say:

$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 7.0.3
$ ghc -O2 -threaded -rtsopts fasta-old.hs -XBangPatterns
[1 of 1] Compiling Main             ( fasta-old.hs, fasta-old.o )
Linking fasta-old ...
$ ghc -O2 -threaded -rtsopts fasta-new.hs
[1 of 1] Compiling Main             ( fasta-new.hs, fasta-new.o )
Linking fasta-new ...
$ time ./fasta-old +RTS -N2 -RTS 25000000 >| old

real	0m21.173s
user	0m18.380s
sys	0m0.910s
$ time ./fasta-new +RTS -N2 -RTS 25000000 >| new

real	0m4.809s
user	0m2.190s
sys	0m0.730s
$ diff -q old new
$ 
$ time ./fasta-old +RTS -N1 -RTS 25000000 >| old

real	0m19.069s
user	0m16.670s
sys	0m0.630s
$ time ./fasta-new +RTS -N1 -RTS 25000000 >| new

real	0m3.797s
user	0m1.500s
sys	0m0.600s
$ diff -q old new
$ 

This is on a dual-core Core 2 running at 2.1GHz. I'm honestly not sure
why performance doesn't improve with two threads, but I think I've made
the point.

- Ben