Max common block size, global array size on ia32

[Craig Tierney]

I was pretty sure that mallocs work.  We have users that
do use more than 1GB of ram in C and Fortran 90.  It is just the ones 
that have 1GB of common blocks in fortran 77 and they aren't going
to rewrite them.

Thanks,
Craig


On Wed, Jul 24, 2002 at 10:33:33AM -0700, Martin Siegert wrote:
> For what it is worth: the following works on my system
> (RH7.2, Linux-2.4.18, pgf90)
> 
> --------------------------------------------------
> module block
> integer, parameter :: k9=selected_int_kind(9)
> integer(kind=k9), parameter :: size=896*1024*1024
> character, allocatable :: a(:)
> end module
> 
> program memtest
> use block
>   allocate (a(size))
>   write(*,*) "hello"
> end
> --------------------------------------------------
> 
> whereas
> 
> --------------------------------------------------
> module block
> integer, parameter :: k9=selected_int_kind(9)
> integer(kind=k9), parameter :: size=896*1024*1024
> character :: a(size)
> end module
> 
> program memtest
> use block
>   write(*,*) "hello"
> end
> --------------------------------------------------
> 
> segfaults. This is straight out of the box without any manipulation of
> the stack limit or anything else.
> 
> Martin
> 
> ========================================================================
> Martin Siegert
> Academic Computing Services                        phone: (604) 291-4691
> Simon Fraser University                            fax:   (604) 291-4242
> Burnaby, British Columbia                          email: siegert at sfu.ca
> Canada  V5A 1S6
> ========================================================================
> 
> On Tue, Jul 23, 2002 at 07:51:38PM -0700, Chris Smith wrote:
> > You're actually hitting the data segment limit, not the stack limit. 
> > 
> > I brought the array declaration inside of main() in the C program to
> > force it to be allocated on the stack, bumped up the stack limit as you
> > did, and the program ran. Alternatively, you can use malloc() to
> > allocate this memory on the heap.
> > 
> > I tried bumping up the data segment limit (even though it was unlimited
> > already), but still had the seg fault. Maybe this is hard limit on the
> > way the process address space is carved up.
> > 
> > Sorry ... I don't know how to do the same kind of manipulation in
> > Fortran.
> > 
> > -- Chris
> > 
> > On Tue, 2002-07-23 at 16:43, Craig Tierney wrote:
> > > Sorry if this is a bit off topic.  I am not sure
> > > where to ask this question.  The following
> > > two codes fail on my system (Dual Xeon, 2 GB Ram,
> > > Linux-2.4.18, redhat 7.2).
> > > 
> > > program memtest
> > > 
> > > integer*8 size
> > > parameter(size=896*1024*1024)
> > > haracter a(size)   
> > > common /block/ a
> > > 
> > > write(*,*) "hello"
> > > 
> > > stop
> > > end
> > > 
> > > OR:
> > > 
> > > #include<stdio.h>
> > > #include<memory.h>
> > > char ar[896*1024*1024];
> > > 
> > > int main() { printf("Hello\n"); }
> > > 
> > > I get a segmentation fault before the codes
> > > start.  I can use ifc, icc, pgf77 and gcc and
> > > get the same results.   If I change the array size to 895 MB,
> > > the codes run.  If I change the C code to
> > > define the array as 'static char ar[blah]' I can
> > > allocate more than 895MB.
> > > 
> > > I have bumped up the max stack size with:
> > > 
> > > ulimit -Hs 2048000
> > > ulimit -s 2048000
> > > 
> > > But this does not help.
> > > 
> > > I cannot find anywhere in the linux source where
> > > the max stacksize might be set.  It seems that
> > > it might be tied to 1 GB, but I cannot find it.
> > > 
> > > Does anyone know how I can get around this
> > > issue?
> > > 
> > > Thanks,
> > > Craig
> > > 
> > > 
> > > -- 
> > > Craig Tierney (ctierney at hpti.com)
> > > _______________________________________________
> > > Beowulf mailing list, Beowulf at beowulf.org
> > > To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> > 
> > 
> > 
> > _______________________________________________
> > Beowulf mailing list, Beowulf at beowulf.org
> > To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
Craig Tierney (ctierney at hpti.com)