When is CONTIGUOUS useful?

[CRquantum]

Dear all,

Quick question, when I am using Intel OneAPI’s Intel Advisor, I remember there was a time I saw a performance hints, saying that using F2008’s CONTIGUOUS property. So I checked it,

CONTIGUOUS

Intel’s example,

REAL, CONTIGUOUS, DIMENSION(:,:) :: A
REAL, POINTER, CONTIGUOUS :: MY_POINTER(:)

It is perhaps to emphasis the assumed-shape array is CONTIGUOUS so it make the compiler easier to do vectorization or some optimization.

Just curious, does anyone use CONTIGUOUS and get some performance gains?
In what scenario CONTIGUOUS can be useful?

Thanks much in advance!

[Pap]

It has its uses, although not very often, at least not for me. One case that contiguous might be useful is when you have to deal with C interoperability. Say, you call a C function from within Fortran code that allocates memory and returns a pointer to that memory block. The corresponding Fortran pointer should be something like
integer(kind=c_signed_char), dimension(:), pointer, contiguous :: data
I can recall a few cases that this didn’t work without the contiguous attribute.
Another example, actually taken from a program I wrote some time ago: Assume you have a C subroutine that renders an image using an array of RGB triplets for each pixel. Now, another subroutine, written in Fortran, has to read the data from an image file. We do not know in advance how big the image is, so we have to use an assumed-shape array to store the data; that array must be contiguous, otherwise the rendering subroutine will render garbage.

[alozada]

This thread could be useful

Fortran pointers, and aliasing

In C, the restrict keyword allows a programmer to make a promise to the compiler, saying approximately “I promise that this pointer doesn’t overlap with memory referred to by other pointers that are passed to this function”. That’s the default state in Fortran, for assumed-shape arrays, for example. But what bout when actually using pointers in Fortran? If the programmer knows that there is no aliasing, is there a way to communicate that to the compiler? Or is it unnecessary, since only targets …

[Arjen]

The programmer is her/himself responsible to make sure that is the case. Unlike in C where the compiler must take into account that things might overlap, the Fortran compiler can simply assume that to be the case. I guess the volatile attribute might help if you know that is not going to be the case?