Re: [Haskell-cafe] Small question

On Fri, Aug 10, 2007 at 02:28:09PM +0800, Hugh Perkins wrote:

On 8/10/07, Stefan O'Rear wrote:

...

Haskell's purpose: To be a generally cool language Haskell's competition: C++, SML, ... hundreds of thousands more and I make no assertion of a representative sample ...

Wrt imperative languages, it probably makes more sense to compare Haskell with imperative languages that do have a GC and for which it's near impossible to accidentally corrupt the stack/heap. You'll find by the way that the imperative GC'd, stack/heap protected languages run *significantly* faster for many (not all I guess?) algorithms and applications.

I don't have any numbers, but I've got a strong suspicion that this is almost entirely an issue of "programmer culture"; IOW, if you wrote a O'Caml to Haskell compiler, and fed idiomatic O'Caml through that and then GHC, the resulting binaries would be about as fast as if you used ocamlopt (INRIA's native code O'Caml compiler); and conversely, Haskell code translated naïvely into O'Caml would be no faster than before. (I do have the case of my Unlambda compiler, which was somewhat faster with ghc -O2 than with ocamlopt, but between CPS-conversion and the complete lack of data types in Unlambda, the resulting code was sufficiently unidiomatic in either language as to render my numbers mostly useless).

This will change with threading of course, but still if you've got a 1024-core Niagara 2012 machine, and the Haskell algorithm runs 65536 times as slowly as a single-core imperative GC'd language program, you're not going to see a significant speed-up ;-)

Just wait 12 years, and if the price of processors follows Moore's extrapolation and the Haskell keeps its parallelism, Haskell will win :) Stefan