[Haskell-cafe] GHC predictability

[Darrin Thompson]

On Tue, May 13, 2008 at 4:30 PM, Andrew Coppin
<andrewcoppin at btinternet.com> wrote:
>  Well, I'm the sort of contrary person who reads random papers like that
> just for the fun of it. But when somebody says something like this, I don't
> think "ooo, that's scary", I think "ooo, somebody really ought to sit down
> and write a more gentle introduction". You really shouldn't *need* to know
> the exact implementation details to get some idea of what will perform well
> and what won't. But obviously you do need some kind of high-level
> understanding of what's going on. The STG paper isn't a good way to get that
> high-level overview.
>

I don't think anyone would disagree with that. Reflecting on what I
already know, I can optimize python pretty well and the principles are
pretty similar for C. The reason I can't just port that knowledge is
that with GHC I'm in the land of optimizing for cache hits and at the
same time I'm at a really high level of abstraction so I have to have
some mental picture of how the plumbing connects.

I'm hoping that the optimization chapter of RWH covers a lot of
individual techniques. I think the sum of the techniques will shed
light on the compiler internals in a practical way. But then I'm not
the one doing the work.

--
Darrin