Re: State of the Art

kamal <>
Mon, 28 Jul 2008 04:05:16 -0700 (PDT)

          From comp.compilers

Related articles
[5 earlier articles]
Re: State of the Art (Aleksey Demakov) (2008-07-23)
Re: State of the Art (Chris F Clark) (2008-07-22)
Re: State of the Art (2008-07-23)
Re: State of the Art (Aaron Gray) (2008-07-24)
Re: State of the Art (Tony Finch) (2008-07-25)
Re: State of the Art (johnhull2008) (2008-07-28)
Re: State of the Art (kamal) (2008-07-28)
Re: State of the Art (Matt Luckman) (2008-07-29)
Re: State of the Art (2008-08-03)
| List of all articles for this month |

From: kamal <>
Newsgroups: comp.compilers
Date: Mon, 28 Jul 2008 04:05:16 -0700 (PDT)
Organization: Compilers Central
References: 08-07-033
Keywords: practice
Posted-Date: 28 Jul 2008 09:49:24 EDT

On Jul 18, 7:40 pm, Peter <> wrote:
> Hi all,
> I havn't worked in compiler construction and programming languages for
> some years, but now I have a chance to return to this area. I would
> like to find out what happened while I was absent.

can you put a lid on how long you were absent?

> So, let me ask the following questions:
> - In your opinion, what are the greatest advances in compiler
> construction in the last ten years?

some posters [incl. their authors] referred to parsing techniques (PEG/

IMO, Java came in, -then its lack of performance brought up JIT/
hotspot editors/dynamic optimization etc. I haven't seen a technique
come up that improves generated code speed significantly. Most changes
I see in the code base are a reaction to customer input/language
standards etc..

    Major changes in compiler optimization (at the back-end) are
invariably driven by changes in processor micro-architecture. So,
tracking architectural changes is probably a good way to locate
changes in compiler back-end technique.

  In that regard, POWER archiecture from IBM shifted many optimixations
to h/w via on-chip scheduler/branch-miss prediction. Itanium OTOH
shifted optimizations to s/w. Itanium has a concept of instruction
bundles, which are taken advantage of by the backend to schedule
instructions in parallel on different functional units. In general,
new/emerging changes to hw include :-

-energy efficiency
-multi core (coz we are hitting moore;s law)

> - What are the most important current trends?
the above 3 issues should be the major drivers [if you depend on
processor roadmaps as a hint]


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.