I am using the intel Fortran compiler on a quad core Mac OS 10.4.11 with 8GB of memory and 64 bit cores. When I increase my array size above a certain point I get the following error:
ld: /tmp/ifort9IeJKr.o r_value (0x9b640d2f) field of relocation entry 318 in section (__TEXT,__text) out of range
Is there a way to allow array dimensions to be increased to take advantage of having so much memory? I can run the code on a Linux machine that has 2GB of physical memory and the same thing happens with the intel compiler at the same point. So it isn't a question of a physical memory limitation. That is, despite having 4 times the memory on the Mac I can still only use the same array sizes as on the Linux machine with 2GB.
In order to use 8GB effectively, you need a 64-bit OS and the 64-bit ifort. Even then, the easiest way to make arrays >2GB would be with ALLOCATABLE. The software doesn't do any magic mode switching simply because you have large physical memory.
Thanks - I have the 64 bit set up. I don't follow what you mean by 'the easiest way to make arrays >2GB would be with ALLOCATABLE.' Please could you explain what this involves doing? I can't find a compiler option to do this. Or does this only apply to itanium based systems?
ALLOCATABLE is a Fortran standard way to declare dynamic arrays, since 15 years ago. There are examples in the documentation of every f90 or newer compiler, including ifort. I don't know of a compiler with an option to disable it. Are you thinking of -mc-model, which changes x86-64 linkage to allow larger static objects? Itanium compilers have larger default limits for static arrays, 4GB or more, depending on the compiler.
This has to do with the way MacOS lays out memory. Try building with -static-intel (I think that's the one). This will allow some increase in static allocation. Or, as Tim suggests, use dynamic allocation.
MacOS does not have the -mc-model option that Linux does.
