Subroutine to set diagonal elements of a matrix

[milancurcic]

In support of: j3-fortran/fortran_proposals#14

Prior art:

• Python numpy.fill_diagonal
• Julia LinearAlgebra.diagind (returns indices of diagonal elements which can then be used to assign to a diagonal)
• MATLAB diag (enabled by MATLAB's special assignment syntax)

Target module: stdlib_linalg

Proposed signature:

! Scalar diagonal variant
pure subroutine set_diag(d, A, k)
  {integer(*), real(*)}, complex(*)}, intent(in) :: d !! Value to set the diagonal to 
  {integer(*), real(*)}, complex(*)}, intent(inout) :: A(:,:) !! Matrix to set the diagonal in
  integer, intent(in), optional :: k !! Number of the diagonal

! Array diagonal variant
pure subroutine set_diag(d, A, k)
  {integer(*), real(*)}, complex(*)}, intent(in) :: d(:) !! Values to set the diagonal to 
  {integer(*), real(*)}, complex(*)}, intent(inout) :: A(:,:) !! Matrix to set the diagonal in
  integer, intent(in), optional :: k !! Number of the diagonal

[ivan-pi]

I presume this is supposed to be the subroutine version of diag that modifies the array in place.

[milancurcic]

@ivan-pi correct, though it would need a different name, thus the proposed set_diag.

[vickysharma0812]

Nice idea @milancurcic
Thanks, @ivan-pi, for the link.

I have the following implementation in my library.
I have defined the Generic function called set(). From the viewpoint of the current issue, the interface of set is as follows.

PURE SUBROUTINE set(mat, value, index, filter)
REAL, INTENT(INOUT) :: mat(:,:)
REAL, INTENT(IN) :: value(:)
REAL, INTENT(IN) :: index
CHARACTER(LEN=1), INTENT(IN) :: filter 
END SUBROUTINE set

• filter can be R for a row, C for the column, and D for diagonal
• index denotes row number, column number, diagonal number, depending upon the filter
• Val is the row/ col/ diag value
• if size(Val) == 1, then all the row/col/diag values are set to the singleton value.

I would be happy to know your opinion on this.

Regards
Vikas