It's generally more straight-forward to optimize CPU-bound Haskell because there are less external moving parts to wrestle with (i.e. OS's graphics / networking stack).
Any code that doesn't allocate ought to be as fast as C with clang as it's pretty much up to LLVM to optimize those inner-loops. GHC's deforestation is the big killer feature to make number-crunching code more elegant.
Once one goes the way of fancy tree structures and all that, well one is writing very different code than they would in C, so expect different performance tradeoffs. GHC should still do a very good job, however, and Haskells various synchronization primitives are quite good too, if your domain requires that.
Thanks... but the alioth benchmarks do not really measure production readiness, although they are very educational.
To me it seems that for writing high performance dense numerical code the tradeoff still is: either write idiomatic code in a horrible language, or write horrible kludges in an beautiful language. Sigh.
Would there be any "Scientific Haskell" resources available that would straightforwardly explain how to write performant code that wrangles data from one data sequence to another?
It's generally more straight-forward to optimize CPU-bound Haskell because there are less external moving parts to wrestle with (i.e. OS's graphics / networking stack).
Any code that doesn't allocate ought to be as fast as C with clang as it's pretty much up to LLVM to optimize those inner-loops. GHC's deforestation is the big killer feature to make number-crunching code more elegant.
Once one goes the way of fancy tree structures and all that, well one is writing very different code than they would in C, so expect different performance tradeoffs. GHC should still do a very good job, however, and Haskells various synchronization primitives are quite good too, if your domain requires that.