🚀 go-pugleaf

Does anyone have an algorithm for IN-SITU transposition of a 2-d NON-SQUARE
matrix stored in a linear address space:

source memory location = source width * i + j,

move data to:

destination memory location = source height * j + i

(note: source height = destination width)

without requiring allocation of extra memory - the matrix is large
and I don't have much memory left.

Speed is not too much of an issue. I have found simple coordinate
transformation sequences which work for certain values of width and height,
but in many cases, these transformations produce closed sequences which are
only a subset of the total coordinate space, and a search (analogous to a
sieve) is required to complete the process. The sieve requires extra memory.

Is it even possible in the general case?

Thanks,

Keith Slavin