[racket-dev] Demodularizer

Just for some numbers,

http://drdr.racket-lang.org/21386/collects/tests/compiler/demodularizer/

used to take 5.76 minutes and now it takes 2.90

Jay

On Sat, Oct 30, 2010 at 9:17 AM, Jay McCarthy <jay.mccarthy at gmail.com> wrote:
> I just pushed a few commits that should speed it up a bit.
>
> Jay
>
> On Fri, Oct 29, 2010 at 10:50 PM, Jon Rafkind <rafkind at cs.utah.edu> wrote:
>> Ok nevermind about the bug, I had some leftover .zo files. When I erased
>> my compiled directory and reran the demodularizer it worked.
>>
>> Another quick stat:
>> $ time raco demod p.rkt
>> real 3m55.563s
>>
>> 1.8ghz amd
>>
>> On 10/29/2010 10:40 PM, Jon Rafkind wrote:
>>> How do you run the resulting _merged file? I ran 'raco demod p.rkt' and
>>> got compiled/p_rkt_zo_merged.zo. Then I tried running racket on it but
>>> got an error
>>>
>>> $ ~/bin/plt/bin/racket p_rkt_zo_merged
>>> compiled/p_rkt_zo_merged.zo: read (compiled): ill-formed code (bad
>>> count: 436459 != 801349, started at 11951)
>>>
>>> Also just some quick stats:
>>> $ cat p.rkt
>>> #lang racket
>>>
>>> (define (hello)
>>>   (printf "Hello world!"))
>>>
>>> (hello)
>>>
>>> $ ls -l compiled/p_rkt.zo ;; just using raco make
>>> 352
>>> $ ls -l compiled/p_rkt_zo_merged.zo
>>> 448410
>>>
>>> Would you mind renaming the command to 'demodularize' ?
>>>
>>> On 10/29/2010 07:53 PM, Jay McCarthy wrote:
>>>> Here is a message from Blake Johnson about something I just pushed for him:
>>>>
>>>> ====
>>>>
>>>> Jay recently committed my changes implementing raco demod, which will
>>>> flatten a modular program into a single compiled module.
>>>>
>>>> How to use it:
>>>>
>>>> Run "raco demod <filename>". This will produce a demodularized zo file
>>>> called <filename>_merged.zo which can then be run by passing it to
>>>> racket. This process can take quite a while, so if you want to see
>>>> what is going on, you can set PLTSTDERR=debug or info. Once the zo
>>>> file is created, it can be moved and run from anywhere because it has
>>>> no dependencies.
>>>>
>>>> What it doesn't do (yet):
>>>>
>>>> - Dead code elimination
>>>>
>>>> This is partially implemented but doesn't handle every case properly,
>>>> so it is turned off for now.
>>>>
>>>> - Optimize the program
>>>>
>>>> This is the next big thing I'm working on, which involves decompiling
>>>> the zo slightly and converting it into c structs so that the existing
>>>> optimizer can run on it.
>>>>
>>>> - Work with programs that dynamically manipulate modules
>>>>
>>>> This means things like DrRacket won't be able to be demodularized. We
>>>> have some ideas on how to handle this, but it probably won't happen
>>>> any time soon.
>>>>
>>>> Other improvements:
>>>>
>>>> zo-parse and zo-marshal should be able to handle any .zo you throw at them.
>>>>
>>>> Help needed:
>>>>
>>>> Any programs that don't work after demodularization would be helpful.
>>>>
>>>> Thanks,
>>>>
>>>> Blake Johnson
>>>>
>>>> ====
>>>>
>>>> At the moment, we have not measured any performance improvements, etc
>>>> after using the demodularizer. At the moment, we do not anticipate
>>>> that there will be any because we are not doing DCE or the
>>>> re-optimization. We'll let everyone know when we measure that. For the
>>>> moment, this may be of particular interest to any out there that are
>>>> doing program analysis and would like "whole programs".
>>>>
>>>> Jay
>>>>
>>> _________________________________________________
>>>   For list-related administrative tasks:
>>>   http://lists.racket-lang.org/listinfo/dev
>>
>>
>
>
>
> --
> Jay McCarthy <jay at cs.byu.edu>
> Assistant Professor / Brigham Young University
> http://teammccarthy.org/jay
>
> "The glory of God is Intelligence" - D&C 93
>



-- 
Jay McCarthy <jay at cs.byu.edu>
Assistant Professor / Brigham Young University
http://teammccarthy.org/jay

"The glory of God is Intelligence" - D&C 93