Performance impact of how a large array is accessed

It’s still the case as of today.

If for instance you have a 2D array and you need to alternatively work on columns and on rows, whatever the reason, you have two options:

  • each time copy/transpose the array, with the associated cost and occupying even more memory
  • use strides when working on the rows

There is no “best option”, all depends on each specific case, and you cannot say “don’t use strides”.