mostly fftw using christophe troessler binding and a kind of plain convolution, actully very comparable to plain convolution 100% in caml (not using any c binding).

plus minor things like binary search, vector multiplications etc. (audio DSP).

Le 3 déc. 2010 à 22:22, Jon Harrop a écrit :

> Philippe wrote:
>> I'm totally noob on compilers internals, but if the processing of float
>> arrays can be improved a lot by a LLVM ocamlopt, I would use it
>> exclusively.
> 
> LLVM's x86 code gen will generate more efficient floating point code than
> ocamlopt if and only if the type information is available to it. With
> OCaml's current design, that is unlikely and you'll still have things like
> the 16Mb limit.
> 
> What algorithms are you running?
> 
> Cheers,
> Jon.