20 Mar
2007
20 Mar
'07
7:55 a.m.
When you tried using Arrays, I presume you used an array indexed by a pair (i,j), and just reversed the order of the index pair to switch from row-wise to column-wise access? It's hard to see how that would slow you down. Perhaps the slowdown was caused by excessive array copying?
the difference can be in locality wrt the memory hierarchy: is the next element nearby most of the time (apart from array borders), or a row-/column-width away most of the time? i understand that, eg, fortran and c differ in their default interpretations of array layout, so naively translated benchmarks might suffer from running against the grain in one of the two (row-major loops over a column-major layout, or the other way round). claus