Tue, 22 Jan 2013

parMap to the Rescue.

I had a long running, CPU intensive Haskell program that I wanted to speed up. The program was basically a loop consisting of a a small sequential part followed by a map of a CPU intensive pure function over a list of 1500 elements.

Obviously I needed some sort of parallel map function and I had a faint recollection of a function called parMap. The wonderful Hoogle search engine pointed me to the parMap documentation.

Changing the existing sequential map operation into a parallel map required a 3 line change (one of them to import the required module). I then added "-threaded" to the compile command line to enable the threaded runtime system in the generated executable and "+RTS -N6" to the executable's command line. The resulting program went from using 100% of 1 CPU to using 560% of 1 CPU on an 8 CPU box. Win!

I wish all code was this easy to parallelize.

Posted at: 22:08 | Category: CodeHacking/Haskell | Permalink