1 --- !ditz.rubyforge.org,2008-03-06/issue
2 title: Determine memory architecture for median calculation.
4 Determine the number of buffers ('planes') to allocate for median calculation.
5 See {issue 2a9ae9d6db1fda0097a62dd69e52e33b33754e26} (Investigate median calculation ...) for more.
9 reporter: David Hilvert <dhilvert@auricle.dyndns.org>
12 creation_time: 2009-02-07 11:33:45.567276 Z
15 id: 876c9d5bb7e4b7b461c023043a74bd62085be7c2
17 - - 2009-02-07 11:33:46.863110 Z
18 - David Hilvert <dhilvert@auricle.dyndns.org>
21 - - 2009-02-07 13:11:08.840409 Z
22 - David Hilvert <dhilvert@auricle.dyndns.org>
25 It appears that this is not a new problem, as it's discussed in a comment in
26 ale:./d2/image_weighted_median.h, regarding cache performance in particular. A
27 multiple-plane approach seems reasonable, as it is probably the most
28 straightforward approach; whether such planes should be allocated incrementally
29 or at the outset is less clear, however.
31 Given that not all interfaces (or uses) of the code will necessarily know
32 capacity at the outset, however, perhaps incremental allocation would be the
33 best approach (where allocating a zero-valued, zero-weighted prefix plane
34 should always be safe).