CODEWITHSUNDEEP
arraysfortranvariable-assignmentin-placeaugmented-assignment
rayhem
rayhem

Reputation: 771

How does Fortran handle augmented assignment of arrays?

I've been working on some code lately that requires me to shift elements in an array left, right, up, and down (depending on an index i). My first thought was to try something like this:

subroutine shift(f)
    real, intent(inout) :: f(0:8, rDim, cDim)
    real :: periodicHor(rDim)

    periodicHor(:) = f(1,:,cDim)
    f(1,:,2:cDim)  = f(1,:,1:cDim - 1)
    f(1,:,1)       = periodicHor(:)

    !and so on for directions 2:8
end subroutine

However, when I lay it out this way, column 1 gets copied into column 2, which gets copied into column 3 which gets copied into... It's not a shift of the data so much as the first column overwriting everything to the left.

If I flip the indices, though, 

subroutine betterShift(f)
    real, intent(inout) :: f(rDim, cDim, 0:8)
    real :: periodicHor(rDim)

    periodicHor(:) = f(:,cDim,1)
    f(:,2:cDim,1)  = f(:,1:cDim - 1,1)
    f(:,1,1)       = periodicHor(:)
end subroutine

things work fine, shifting left or right. I suspect the compiler detects an in-place update of continuous memory in the 2nd case, so it makes a temporary copy "under the hood" to avoid the overwrite issue, but that's just speculation on my part. Can anyone give a more detailed explanation as to why the shift works one way and not the other?

Upvotes: 0

Views: 176

Answers (0)

Related Questions