For those interested in improving the performance of GHC's generated code, especially in the area of tight loops, I have collected a bunch of ideas on this wiki page: http://hackage.haskell.org/trac/ghc/wiki/BackEndNotes This is partly the outcome of a discussion between myself and Roman Leshchinskiy, who is working on nested data parallelism. Roman has already implemented some special-case optimisations to improve performance of tight loops, and he plans to do more. If you have any thoughts, ideas, or comments on the proposals, please let us know! Also, it's not clear if/when any of this will get done. Talented compiler hackers are most welcome to help out. Cheers, Simon