[SciPy-dev] random module not working on opteron

[Charles Harris]

The mt1997 code on the author's site masks all the operations for 64 bit
compatability, although I suspect this is compiler dependent and shouldn't
matter with gcc. But apparently there was a reason... ;) Also, device
urandom delivers bytes, IIRC, like any other unix file read, so we need to
read 624*sizeof(long) bytes to fill the array.

Chuck

-----Original Message-----
From: scipy-dev-bounces at scipy.net on behalf of Robert Kern
Sent: Sun 10/23/2005 6:09 PM
To: SciPy Developers List
Subject: Re: [SciPy-dev] random module not working on opteron
 
Arnd Baecker wrote:
> On Sun, 23 Oct 2005, Robert Kern wrote:
> 
> 
>>Arnd Baecker wrote:
> 
> 
> [...]
> 
> 
>>>Can anyone with a 64 Bit machine confirm this behaviour?
>>>What could one do to hunt this one down?
>>
>>I would first check if the RandomKit functions work correctly from C.
>>See the files randomkit.[ch] ; the functions should be self-explanatory.
>>It's entirely possible that they're not 64-bit safe; I never had the
>>opportunity to check.
> 
> 
> Ok, referring to a famous scene of the Life of Brian:
> after more than 5 years of not speaking C at all
> you are stepping on my feet and I speak C again - a miracle!
> (What a pleasure, no need to pay attention to white space,
> I love all those variable declarations and ";"... Thank you ;-).
> 
> Now to the facts:
> - the code below works fine on my PIV
> - on the opteron it gives problems and hangs:
> 
> Value: 1791095845l max=4294967295l
> Value: 0.997185
> Now with random seed:
> Value: 920722385l max=4294967295l
>    Value: 915381373.352906                    # <----- not ok! (rk_double)
> Now test the distributions:
> rk_uniform: 566004099.864036                  # <----- not ok!
> rk_normal:                                    # (hangs)
> 
> The problem is caused by calling rk_random_seed
> (Commenting this one out, everything seems fine).
> 
> I am not sure what is going on in that routine
> (and have to leave investigating this at the moment ...)
> Maybe you can have a look at that routine
> and spot anything which might be problematic?
> (either 64 Bit, or the clock stuff???)

I think the issue is that the state is an array of 624 unsigned longs.
rk_seed() essentially just does 32-bit arithmetic and so fills those
longs with numbers up to 2**32-1. rk_randomseed(), however, takes raw
bits from /dev/urandom and drops them into the array. I believe that the
algorithm is expecting 32-bit integers.

I've applied a mask to the state after it gets seeded by /dev/urandom.
Please checkout the latest version of randomkit.c from SVN and give it a
try. We may need to make this more robust later, but for now it might work.

-- 
Robert Kern
rkern at ucsd.edu

"In the fields of hell where the grass grows high
 Are the graves of dreams allowed to die."
  -- Richard Harter

_______________________________________________
Scipy-dev mailing list
Scipy-dev at scipy.net
http://www.scipy.net/mailman/listinfo/scipy-dev

-------------- next part --------------
A non-text attachment was scrubbed...
Name: winmail.dat
Type: application/ms-tnef
Size: 4445 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20051023/09624f81/attachment.bin>