Fortran Need some help with dynamic memory allocation in Fortran

[sue132]

Hi,

I need some help with allocating arrays dynamically.

I have an array whose size keeps changing at each step - I start with a row vector and add another row vector at each step of a cycle. I need to use this new array for further calculations.

I've been trying to declare the arrays as allocatable, and then allocate them within a do-loop, but I haven't been able to get it working properly.

I am still not very clear with what happens when you deallocate the memory for that particular array. If the data is lost, then to store it before deallocation, do I use another array? How do I specify the size of that array?

Any help would be greatly appreciated.

Thank you

 

[chiro]

Hey sue132 and welcome to the forums.

Do you mind posting your code so that the readers can make specific suggestions? Without this kind of thing neither you nor the other readers benefit from the discourse.

To post code surround your code with *CODE* */CODE* tags but replace the *'s with square brackets (left [ bracket before C and ] bracket after E for each tag).

 

[sue132]

Thanks a lot.

I've tried to include enough comments to make the code clear, hope it makes some sense.

The current program is giving me a segmentation fault error, which I'm trying to rectify now.

! Program to generate a matrix whose rows are linearly independent vectors. 
! The elements of the matrix are either 1 or -1

   program trial
   
   implicit none 
   integer, parameter ::n=10,p=5   !where 'p' gives the no. of rows, & 'n' gives the no. of columns
   real(kind=8),dimension(p,n) ::z1 
! The array to be generated - this needs to be made allocatable
! I'll need the values of this array for further calculations, so should I store
! them in another array?

   integer::i,j,k,s1  !counters used
   real(kind=8),dimension(n)::x

   do i=1,p
15	call random_number(x) !generate random row vectors 
	z1(i,:)=x

! To make the array elements 1's and -1's
	do j=1,n
		if (z1(i,j).ge.0.5) then
			z1(i,j)=1.d0
!			write(*,*),z1(i,j)
		else
			z1(i,j)=-1.d0
!			write(*,*),z1(i,j)
		endif
	enddo

! Checking the output so far
	print*,'Z1'
	do k=1,i
 		write(*,*)(z1(k,j),j=1,n)
	enddo

! To ckeck for linear independence, get the reduced row echelon form. Check the rows
! of the rref array, if any of the rows contains only zeros, the vectors are not
! linearly independent. Hence, regenerate that particular row of the array. 

! Check for linear independence as each row is added

	call to_rref(z1)
		do k=1,i
			s1=0
			do j=1,n
			if(z1(k,j)==0)then
				s1=s1+1
			endif
			if (s1==n) then
				print *,'THE VECTORS ARE NOT LINEARLY INDEPENDENT'
			goto 15
			endif
			enddo
		enddo
	
	do k=1,i
 		write(*,*)(z1(k,j),j=1,n)
	enddo
   enddo
	
   end

! Subroutine for getting the reduced row echelon form 
subroutine to_rref(matrix)
    implicit none
    real, dimension(:,:), intent(inout) :: matrix
 
    integer :: pivot, norow, nocolumn
    integer :: r, i
    real, dimension(:), allocatable :: trow
 
    pivot = 1
    norow = size(matrix, 1)
    nocolumn = size(matrix, 2)
 
    allocate(trow(nocolumn))
 
    do r = 1, norow
       if ( nocolumn <= pivot ) exit
       i = r
       do while ( matrix(i, pivot) == 0 )
          i = i + 1
          if ( norow == i ) then
             i = r
             pivot = pivot + 1
             if ( nocolumn == pivot ) return
          end if
       end do
       trow = matrix(i, :)
       matrix(i, :) = matrix(r, :)
       matrix(r, :) = trow
       matrix(r, :) = matrix(r, :) / matrix(r, pivot)
       do i = 1, norow
          if ( i /= r ) matrix(i, :) = matrix(i, :) - matrix(r, :) * matrix(i, pivot) 
       end do
       pivot = pivot + 1
    end do
    deallocate(trow)
  end subroutine to_rref

P.S.: I've used the rref function from numerical recipes, as I've heard that they are highly optimized.

 

[chiro]

Do you have any debug output in terms of the line of code and the variable dump that gave you the error?

Running it through a debugger should give you an error message and the line in which the program crashes.

 

[gsal]

sue:

Quick reply here

First of all, I don't see why you need your z1 matrix allocatable when you are defining n and p with initial values and you even have your do-loop as "i=1,p"...

Now, if the rref subroutine is destructive, then, yes, I would declared two matrices and call rref with both of them, one with intent(in) and one with intent(out) so that z1 does not gets destroyed.

Also, if I understand your program correctly, I think I would turn the hard set "do i=1,p" for some index-less "do - enddo" loop with an 'exit' somewhere in between...after all, if you are checking linear independence every iteration, you do not want to throw away the entire matrix, you just want to re-do the last row and, so, you need to control the row number during your iteration...you may need to set it back 1...to that end, you cannot let it be a loop variable...maybe I am getting ahead of myself and you already knew that.

my 2 cents.

 

[sue132]

Thanks for the reply, gsal.

I need the array to be allocatable, as my array size is ultimately pXn, but not throughout the program. My array starts with a 1Xn row vector and at each step, another row is added. Hence, it needs to be allocatable.

And, I'm not discarding the entire array - I just go back to the step within the loop where a new row is generated, and as I'm still within the loop, I'm only generating the row afresh. At least, this is what I'm trying to do.

Hope I'm clear...