Fortran Structures: What Are They?

[cammo12]

Hi - I expect this is quite a simple question.

What are fortran structures? I'm trying to convert fortran code into MATLAB and I don't know what a structure / endstructure statement is.

e.g.


structure /pixel/
union
map
INTEGER*4 cl
INTEGER*4 rw
endmap
map
INTEGER*4 x
INTEGER*4 y
endmap
endunion
endstructure

 

[ChrisLeslie]

A structure is a defined block of related variables, a feature common to many languages. I am not sure about Fortran structure specifically, but it appears that you have structure called "pixel" that allows you to define specific instances of a pixel with the characteristics of cl, rw, x and y. Somewhere in the code there will be a statement that instantiates a pixel instance, eg something like "myPixel pixel...". "myPixel" is then the name of your pixel instance, and the pixel members for "myPixel" will likely be identified as "myPixel.cl", "myPixel.rw", myPixel.x" and "myPixel.y". For example, "myPixel.x = 100" and "myPixel.y = 200" will establish the x-y coordinates for the "myPixel" pixel.

Chris

 

[jim mcnamara]

Here is the same thing in C. I have not done MATLAB in so long I'd be afraid to post anything that purports any relationship.

typedef union
{
      struct
      {
        long cl;
        long rw;
      } s1;
      struct
      {
        long x;
        long y;
      } s2;
} pixel_t;

pixel_t pix;

It's a union of two structs.

 

[cammo12]

So what is the point in this?

Do I have to have structures? Can't I just use an array somehow?

 

[cammo12]

Thanks for your help by the way :)

 

[AlephZero]

No you don't have to use structures (and that particular union of two structures seems rather complicated, for what it actually does).

The point of structures is to logically group a set of data items together so you can refer to the group by one name - for instance

pixel p1,p2, pfun() // pfun is a function that returns a pixel structure
...
p1 = p2
call sub(p1)
p2 = pfun(1,2,3)
...
etc

Yes you could do all this without using structures - except for the function return. But if you had a structure with 20 elements not 2, the code would be about 20 times longer.

 

[ChrisLeslie]

Yes you can probably manage with an array, depending on the complexity of the problem. eg, something like:

INTEGER pixel(4)
* then address the members as element of the array
cl = pixel(1)
rw= pixel(2)
x = pixel(3)
y = pixel(4)

 

[jim mcnamara]

..thought C would help. :)

In English, not computer, the FORTRAN code means the following:
cl occupies the exact same memory location as does x
rw does the same with y
cl, rw, x, and y are all signed integers - 32 bit int -- what INTEGER*4 means.

Your array does not seem to do all that - ie., overlap memory. I'm assuming INTEGER pixel(4) creates four elements in vector of 32 bit signed ints.

 

[ChrisLeslie]

Hi jim

Yes, the array method does not physically overlap in memory the elements cl with x, and rw with y. cammo12, that is what the union does in your original code.
Without structures or unions this can be achieved in FORTRAN by using "equivalence". However, equivalence will not work with members from the same array. So if this type of union is required then two arrays will have to be created:

integer pixelx(2),pixely(2)
equivalence (pixelx(1),pixely(1)), (pixelx(2),pixely(2))
etc, etc

This may not help with the Matlab translation, but hopefully jim's C example and the explanations may help cammo12 to interpret the Fortran snippet.

Chris