Re: [Caml-list] Question about Optimization

[Gabriel Scherer <gabriel.scherer@gmail.com>]

Another approach that might be worth trying (sorry for not thinking
about it earlier) is MetaOCaml. I tend of think of it as a tool to
explicitly specify and control partial evaluation strategies.

Jeremy Yallop teaches MetaOCaml and some optimization tricks using it.
See this excellent IOCaml notebook showing how to optimize a sequence
of operations on a simple stack machine using MetaOCaml, which is not
far from the discussed application.

  http://ocamllabs.github.io/iocamljs/staging2.html

(I don't have an opinion on effectiveness of user-defined rewriting
for OCaml code, I guess one should try to see how it goes. My
take-away from the usage in the Haskell community is that it's very
useful in some situations but also very fragile.)

On Thu, Apr 21, 2016 at 11:45 AM, Gregory Malecha <gmalecha@gmail.com> wrote:
> Thanks for the feedback. It sounds like there are reasonable ways to do this
> if you write your programs in such a way that everything can be done via
> inlining. While I agree that this works in many cases, it doesn't seem
> powerful enough to solve everything. For example, equations that are only
> provable by induction are not optimizable by this strategy (without some
> pretty fancy tricks).
>
> Do you believe that having a rewriting facility is unnecessary?
> Undersireable (it makes the compiler too complex, too slow, too
> unpredictable)? Or do you think that it would be useful, but no one has done
> anything on it?
>
> Thanks.
>
>
> On Thu, Apr 21, 2016, 4:45 AM Yaron Minsky <yminsky@janestreet.com> wrote:
>>
>> Also, Core_kernel's Sequence type fills a similar purpose.  And Flambda
>> does a good job of optimizing the iteration in Sequence, from what I've
>> overheard about our experiments.
>>
>> On Thu, Apr 21, 2016 at 5:32 AM, Jonas Jensen <jj@issuu.com> wrote:
>>>
>>> On 21 April 2016 at 09:13, Gregory Malecha <gmalecha@gmail.com> wrote:
>>> >
>>> > I'm wondering if there is any work (and interest) on supporting
>>> > user-defined optimizations similar to GHC's rewrite rules in the Ocaml
>>> > compiler. For example, a standard example would be specifying map fusion:
>>> >
>>> > map f (map g ls) = map (fun x -> f (g x)) ls
>>>
>>> A "boring" and practical answer is that you get this optimization by
>>> writing your long chain of map, filter, bind, etc. using Batteries'
>>> Enum
>>> (http://ocaml-batteries-team.github.io/batteries-included/hdoc2/BatEnum.html)
>>> or the stand-alone Gen package
>>> (http://cedeela.fr/~simon/software/gen/Gen.S.html). It looks
>>> superficially like list map, but the order of execution will be like
>>> after a fusion, which should improve cache locality and avoid
>>> allocations of intermediate lists.
>>>
>>> Cheers,
>>> Jonas
>>>
>>>
>>> --
>>> Caml-list mailing list.  Subscription management and archives:
>>> https://sympa.inria.fr/sympa/arc/caml-list
>>> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>>> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
> --
>
> - gregory malecha
>   gmalecha.github.io