topic Re: Passing a contiguous allocatable array section to a explici in Intel® Fortran Compiler

Passing a contiguous allocatable array section to a explicit shape array

OP1 — Tue, 26 Aug 2008 14:39:26 GMT

Is passing a contiguous array section of an allocatable array to a subroutine where the corresponding dummy argument is an explicit shape array efficient? In other words, is a temporary array created?

Consider the following example:

PROGRAM PROG
INTEGER(4) N,P
REAL(8),ALLOCATABLE :: M(:,:)
! Allocate and initialize M(N,P)
CALL SUB(M(:,1),N)
END PROGRAM PROG

SUBROUTINE SUB(A,SIZE)
INTEGER(4) SIZE
REAL(8) A(SIZE)
! Do some work here...
END SUBROUTINE

Is this efficient? Is the array passed by address or by value in SUB (in the later case a copy of the array section would need to be done I suppose).

How does the compiler recognize that the data (the array section) is contiguous?

Thanks,

Olivier

Re: Passing a contiguous allocatable array section to a explici

Steven_L_Intel1 — Tue, 26 Aug 2008 15:54:16 GMT

The compiler does a run-time test and creates the copy only if needed. To test this, add the /check:arg_temp_created option (Run-Time - Check for actual arguments using temporary storage) and see if you get a run-time warning when the call is made.

Re: Passing a contiguous allocatable array section to a explici

jimdempseyatthecove — Tue, 26 Aug 2008 18:13:49 GMT

Olivier,

Consider using

CALL SUB(M(:,1), size(M(:,1)))

This reduces potential for error in the event the allocation is changed without changing N.

The is nothing wrong with using the N as you had before...
as long as the N isn't reused with different values for different arrays.

The above syntax should be within a few memory cycles of using N.

Jim Dempsey

Re: Passing a contiguous allocatable array section to a explici

OP1 — Tue, 26 Aug 2008 18:49:13 GMT

Thank you Jim and Steve for your help. I found the run-time check solution proposed by Steve a very valuable tool.

Olivier