maximum array size

[Neil_S_]

Is there a maximum array size that is allocatable using the intel compiler ?

[Steven_L_Intel1]

No - you're limited by what the operating system allows. The compiler does not impose its own limit.

[edouardoudet]

Hi,

How is it possible to evaluate this maximum size : I work on an 64b opteron and installed successfully Intel's fortran compiler 8.* for 64b processors. I have no problem for compiling or linking but I was surprised to observe that the maximum array size didn't change from my 32b installation ... !? Is it normal ??

Thanks for your answer(s), Edouard.

[Steven_L_Intel1]

There is no easy way to determine it. If you are declaring a static array, there are limits of the linker and the image activator. Have you tried an allocatable array? What was the maximum size you managed to see?

[TimP]

Does the -mcmodel= switch influence the maximum size of static arrays? I don't find documentation readily accessible.

[Steven_L_Intel1]

Hmm - yes, it might. Try -mcmodel=medium and see if it helps.

[Neil_S_]

/opt/fc-8.1.023/bin/ifort -mcmodel=medium -c frxx5.f

ifort: Command line warning: ignoring unknown option '-mcmodel=medium'

I have seen this option on other compilers, not yours.

[Neil_S_]

Try -fpic on your 64b machine, i think thismight help.

[edouardoudet]

Actually the limit of size on my computer (the 64b Opteron AND on my 32b PC) is between 1E8 and 1E9 for the maximum size of an allocatable array for a DOUBLE PRECISION. I checked also that the stack size and the data seg size are unlimited.

The /proc/meminfo file gives me :

total: used: free: shared: buffers: cached:
Mem: 4088872960 97042432 3991830528 0 11104256 59793408
Swap: 4293910528 270786560 4023123968
MemTotal: 3993040 kB
MemFree: 3898272 kB

which seems to be sufficient !?

On the other hand, the greatest integer of kind=8 (obtained with the huge command) is the same as on a 32b : 2147483647. Again, I do not know if it is normal.

Thanks for your comments, Edouard.

[Steven_L_Intel1]

If you are seeing the same result for HUGE of an INTEGER(8) as an INTEGER(4), there's an error in your program. Perhaps you are also misinterpreting the size numbers.

[edouardoudet]

As you suggest, there was a problem in my program : I used a "Integer :: variable=kind(8)" statement instead of the declaration "Integer(8):: variable". I thought it was equivalent !

Yet I still do not understand why it is not possible to allocate a bigger array on my 64b processor than on my 32b PC. Do you have any idea considering the datas in a previous message ?

Thanks a lot for your answers.

[Steven_L_Intel1]

Let's see an example program, the commands you used to build it, and the results of running it. I know this works - I have seen it myself.

[edouardoudet]

I used the followig test program :

**********************************************************
PROGRAM test64

IMPLICIT NONE

INTEGER(8) :: dimarray
DOUBLE PRECISION, DIMENSION(:), ALLOCATABLE :: testalloc

dimarray = 260000000
PRINT*,HUGE(dimarray)

PRINT*,"BEGIN ALLOCATION"
ALLOCATE(testalloc(dimarray))
PRINT*,"ALLOCATION SUCCESFULL",dimarray
DEALLOCATE(testalloc)

END PROGRAM test64
**********************************************************

+ I compiled and executed it on my PC (32b) with the standard

ifort test64.f90 -o test64
./test64

instructions. It is successfull. If ichange the value dimarray to 270000000. Iget the message :

Allocate error 494: Allocation of Array with extent of 270000000 failed



+ On the 64b processor compiled with the instruction :

/opt/intel_fce_80/bin/ifort test64.f90

the limit (computed manually changing the value of dimarray) is close from 1000000000. That is less than 5 times larger than the previous one.

With a greater integer I got the following message :

forrtl: severe (41): insufficient virtual memory
Image PC Routine Line Source
a.out 000000000043A2BE Unknown Unknown Unknown
a.out 0000000000439CE2 Unknown Unknown Unknown
...

Thanks.

[Steven_L_Intel1]

You're hitting an OS limit determined by your configuration. I don't know enough about Linux to advise you on this, except to say that it's not straightforward to configure a system that will allow more than 2GB allocations.

[hanswb]

Hi,

I am able to compile and run the test program up to the size of the physical memory limit of 32 GB:

c5i5n13 .../development/x86_64 $ ifort test64.f90
c5i5n13 .../development/x86_64 $ a.out
9223372036854775807
BEGIN ALLOCATION
ALLOCATIONSUCCESFULL 4000000000

It's a SuSE 9.1 x86_64 with Kernel 2.6.5-7.97-smpand Intel l_fce_pc_8.1.025

Hans