[ADMB Users] 4 Gb limits
dave fournier
davef at otter-rsch.com
Mon Nov 21 08:20:35 PST 2011
On 11-11-20 03:21 PM, John Sibert wrote:
I don;t think it is needed now. I think it was just during development
looking for errors.
> Hi Alan,
> The problem seems to be in this initialization loop in df_file.cpp
> Adding the additional #ifdef seems to allow the program to execute.
>
> Dave, is this initialization really necessary? It takes a long time
> for sz>4Gb (at least on the test machine).
>> #ifdef __BORLANDC__
>> for (unsigned int i=0;i<sz;i++)
>> #else
>> for (long long int i=0;i<sz;i++)
>> #endif
>> {
>> buff[i]='\0';
>> }
> Also I noticed (rather the compiler noticed) a discrepancy in the
> following two declarations:
>> df_file.cpp: extern char ad_random_part[6];
>> gradstrc.cpp: extern char ad_random_part[7];
> Which is correct?
Looking at the code
void fill_ad_random_part(void)
{
/*
time_t t,tt;
time(&t);
tt=t;
int div=1;
for (int i=0;i<6;i++)
{
ad_random_part[i]=(tt/div)%10+48;
div*=10;
}
*/
}
It seems that it is not used any more but it did use [6]
>
> I will commit the changes eventually.
>
> Cheers,
> John
>
>
> On 11/20/2011 08:39 AM, Allan.Hicks at noaa.gov wrote:
>> Hi John,
>>
>> I have a couple of thoughts, probably of which you have already taken
>> care of. But, here goes:
>>
>> 1) Make sure that you are using ADMB 10.1. I noticed that the inputs
>> were still long int in version 10.0.
>>
>> 2) Make sure that AMB10.1 is compiled as 64-bit. I noticed that there
>> are no 64-bit downloads on the website for ADMB 10.1 in Linux. I also
>> could not find 64-bit executables for Linux gcc on Buildbot (I see a
>> 64-bit version for gcc 4.5 has been successfully built, but I do not see
>> executables). But, given that you are using gcc 4.6, you probably
>> compiled it yourself, but I just wanted to make sure.
>>
>> 3) You may want to try setting the cbs value in the tpl file, just to
>> make sure that there is no issue with the reading in from the command
>> line.
>>
>> If these aren't the answer, I'm not exactly sure where the problem is.
>> Because it is at the exact breaking point of an unsigned long int
>> (2^32-1 is the maximum) then I imagine that there is something in the
>> code that is declared as an unsigned long int and there is a conversion
>> issue. This is a bit over my head, and you warned me a while ago that
>> converting to 64-bit is not as straightforward as simply changing some
>> declarations.
>>
>> In the meantime, I will try to compile ADMB 10.1 in Linux 64-bit and see
>> if I can reproduce your issue and investigate it further.
>>
>> Sorry that I can't be of more help,
>> Allan
>>
>>
>>
>>
>>
>> ----- Original Message -----
>> From: John Sibert<sibert at hawaii.edu>
>> Date: Saturday, November 19, 2011 6:31 pm
>> Subject: Re: [ADMB Users] 4 Gb limits
>>
>>> Nope. The value of the option behind -cbs is declared as a long
>>> long
>>> int. See nh99/xmodelm3.cpp.
>>>
>>> Ian Taylor and Alan Hicks resolved this issue for -gbs.
>>>
>>> John
>>>
>>>
>>>
>>> On 11/19/2011 03:57 PM, Weihai Liu wrote:
>>>> I would guess the size for cbs was defined as an unsigned long,
>>> which
>>>> is 32 bit, its range 0 to 4294967295, so 4294967296 will be just
>>> out
>>>> of range, although it = 2^32.
>>>>
>>>>
>>>> weihai
>>>>
>>>> On Sat, Nov 19, 2011 at 8:23 PM, John Sibert<sibert at hawaii.edu
>>>> <mailto:sibert at hawaii.edu>> wrote:
>>>>
>>>> Using simple.tpl
>>>>
>>>> simple -cbs 4294967295 works fine
>>>> simple -cbs 4294967296 hangs
>>>>
>>>> ubuntu 11.10, 64 bit, 32Gb RAM, gcc 4.6.1
>>>>
>>>> Any thoughts?
>>>> John
>>>>
>>>> --
>>>> John Sibert
>>>> Emeritus Researcher, SOEST
>>>> University of Hawaii at Manoa
>>>>
>>>> Visit the ADMB project http://admb-project.org/
>>>>
>>>> _______________________________________________
>>>> Users mailing list
>>>> Users at admb-project.org<mailto:Users at admb-project.org>
>>>> http://lists.admb-project.org/mailman/listinfo/users
>>>>
>>>>
>>> --
>>> John Sibert
>>> Emeritus Researcher, SOEST
>>> University of Hawaii at Manoa
>>>
>>> Visit the ADMB project http://admb-project.org/
>>>
>>> _______________________________________________
>>> Users mailing list
>>> Users at admb-project.org
>>> http://lists.admb-project.org/mailman/listinfo/users
>>>
>
More information about the Users
mailing list