[Haskell-cafe] Re: GHC predictability

[Achim Schneider]

Andrew Coppin <andrewcoppin at btinternet.com> wrote:

> You're probably right about all that. I would humbly suggest that
> what is somewhat lacking is a good, introductory, high-level text on
> what makes Haskell go fast and what makes it go slow. As with many
> things in the Haskell world, there are bits and pieces of information
> out there, but it's difficult to track them all down and present a
> coherant picture. We've got a section on the wiki containing scraps
> and snippets of information. There are various GHC-related papers [if
> you can find them online]. GHC has various profiling possibilities,
> but thus far I've found it difficult to digest the results. We need a
> good, thorough body of text on this subject, I feel. [Of course, that
> still means somebody has to write one...]
> 
Something like the history paper but concentrating on algorithms,
techniques & tricks would be great, yes. And, most importantly, less
buzzwords where you're lucky if you find a definition of it by googling.

> 1. What is "ghc-core"?
> 
An intermediate language, I'm quoting from memory:

First comes a Syntax tree, then the type checker, then the translation
to core, then optimisations on core (, then the printout) and finally
c/assembly.  

> 2. Does anybody know how to actually read GHC's Core output anyway?
> To me, it looks almost exactly like very, very complicated Haskell
> source with a suspicious concentration of case expressions - but I
> understand that in the Core language, many constructs actually mean
> something quite different.
> 
I found it rather easy to parse... as long as you succeed in finding
what you're looking for behind all that inlining. Types ending in #
mean they're unboxed.

It's particularly useful to find out how much ghc specialises your
code.

> 3. Any idea where the STG paper is? Is it still an accurate
> reflection of GHC's current technology?
>
http://research.microsoft.com/~simonpj/Papers/papers.html
Implementing lazy functional languages on stock hardware: the Spineless
Tagless G-machine, SL Peyton Jones, Journal of Functional Programming
2(2), Apr 1992, pp127-202.

while googling for it, I stumbled across 
http://citeseer.ist.psu.edu/reid98putting.html

which might be more actual, but I neither read it yet or have any
idea whatsoever.

-- 
(c) this sig last receiving data processing entity. Inspect headers for
past copyright information. All rights reserved. Unauthorised copying,
hiring, renting, public performance and/or broadcasting of this
signature prohibited.