Re: relocation errors - ReadList.com

1 msg	gcc-4.1.2 build help
2 msg	Gcc build error
2 msg	regarding optimization flags

relocation errors
\ Huliyar Nataraj (25 Feb 2007)
. \ Tim Prince (25 Feb 2007)

4 msg	help for math fuctions
2 msg	Error in GCC build
10 msg	Is fixinclude needed on modern systems?
2 msg	Hi can you please help me
1 msg	to: Unable to compile large projects
1 msg	Building gcc 4.1.2 - gengtype can't find files
2 msg	need help with DGRAM socket
4 msg	What happens when -O1 flag is used?
3 msg	Best mcpu option for amcc440 with fpu ?
1 msg	makedepend.c error
9 msg	list of optimizations done by -O1 level
1 msg	WinPorts GCC 4.1.2 is out [win32 build]
4 msg	strange floating point problem
1 msg	Re: 'Error: suffix or operands invalid for `pus...
2 msg	problem with 'nanosleep()'
5 msg	g++ compile time calculations

Subject:	Re: relocation errors
Group:	Gcc-help
From:	Tim Prince
Date:	25 Feb 2007

Huliyar Nataraj wrote:
> Dear Sir,
>
> I am using a 64-bit AMD apteron cluster with 8 GB RAM (RHEL platform).
> I face "relocation overflow" and "relocation truncated to fit errors"
> when I compile big fortran 77 codes using f77/g77 (even with ifort and
> pathf95) on it. Though it compiles successfully it cannot assemble or
> link the executables. I understand this may be because of large arrays
> or so. But the executable size is not more than 2GB. so I could not
> run my codes with large dimensions.
>
> Can you please tell me what options I can use while compiling these
> kind of codes. Sometimes, ulimit command or -m32 option worked but
> when I run the executable it simply dumps the core. Please suggest me
> what can I do to get rid of these problems?.
>
> Note:
> The gcc version I am using is gcc-3.2.3-42
>

Since your gcc is so old, and you are using RH, I might suspect you also
have an old unreliable version of binutils (ld --version). Current gnu
binutils release (2.17 or newer) tends to be more reliable on x86-64.
If you have any static data objects (e.g. COMMON) larger than 2GB, you
must use 64-bit compilation and set an appropriate mc-model (probably
not supported by g77).
g77 is clearly not supported well for x86-64; a current gcc (gfortran)
or commercial compiler would be a better choice.
If you are using cluster computing, in the usual sense (MPI or the
like), you can't use -m32 unless everything is consistently 32-bit,
including your MPI library build. Then your large RAM would not be
visible, as your total address space limit would be 2GB.