[racket] The performance of fannkuch-redux

From: James Swaine (james.swaine at eecs.northwestern.edu)
Date: Mon Feb 25 11:15:09 EST 2013

It can be difficult if you aren't familiar with the JIT implementation
code, but yes.  It seems like Sam's package is a much more tractable
option, though.

On Mon, Feb 25, 2013 at 9:24 AM, Robby Findler
<robby at eecs.northwestern.edu>wrote:

> Can't you set a break point somewhere and then use gdb's 'disassemble'
> command?
>
> Robby
>
>
> On Monday, February 25, 2013, James Swaine wrote:
>
>> Hi Haiwei,
>>
>> AFAIK there isn't any way to see the actual instructions that the JIT is
>> generating.  Why do you think it would be helpful in this case?
>>
>> It may be helpful to use raco decompile, however -- that allows you to
>> see what the bytecode version of your program looks like (which is what is
>> actually being fed to the JIT).
>>
>> -James
>>
>>
>> Message: 6
>>> Date: Mon, 25 Feb 2013 17:53:34 +0800
>>> From: Haiwei Zhou <highfly22 at gmail.com>
>>> To: Danny Yoo <dyoo at hashcollision.org>
>>> Cc: users <users at racket-lang.org>
>>> Subject: Re: [racket] The performance of fannkuch-redux
>>> Message-ID:
>>>         <CABLXuO9nJAUa=Xn7779377NPnoQX6Ze3Tdkf6bvF=
>>> sta+bG6wQ at mail.gmail.com>
>>> Content-Type: text/plain; charset="iso-8859-1"
>>>
>>> Updated a parallel version to
>>> https://github.com/highfly22/fannkuch-redux
>>> It takes about 1m50s in the four cores PC, while the script in the
>>> shoutout
>>> takes about 4m30s. It looks like the task division is not perfect.
>>>
>>> I am just curious on the native implement of JIT. How to dump assemble
>>> code
>>> after JIT?
>>>
>>> When I try to define count-flips as a macro, racket complains "to many
>>> forms"? What does that mean?
>>>
>>> Thanks,
>>> Haiwei
>>>
>>>
>>> On 24 February 2013 11:03, Danny Yoo <dyoo at hashcollision.org> wrote:
>>>
>>> > >> > It's amazing fast. If it is only 6 times slower than C version. If
>>> > >> > parallelizing the time consuming operations, it would be 2x slower
>>> > than
>>> > >> > C
>>> > >> > version in the four cores CPU.
>>> > >>
>>> > >>
>>> > >> Do you mean the version you've written, or the one in the Racket
>>> > >> benchmark suite?
>>> > >
>>> > > It's the one in the benchmark suite.
>>> >
>>> >
>>> > What does the code look like if you add the parallelizing?  Can you
>>> > show us?   Thanks!
>>> >
>>> -------------- next part --------------
>>> An HTML attachment was scrubbed...
>>> URL: <
>>> http://lists.racket-lang.org/users/archive/attachments/20130225/8e591afb/attachment-0001.html
>>> >
>>>
>>> ------------------------------
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.racket-lang.org/users/archive/attachments/20130225/1a7bb0e3/attachment.html>

Posted on the users mailing list.