[ofa-general] Re: [PATCHv3] opensm/osm_mesh.c: Reorder switches for lash

[Sasha Khapyorsky]

On 14:48 Wed 05 Aug     , Hal Rosenstock wrote:
> 
> The goal of this patch is to change the order of the switches in the array kept
> in the lash context from the original order to one in which the switches are
> presented in 'odometer order'.
> 
> When the main routine in lash is called the switches are in an order that is
> likely based on the order that the switches were originally visited by SM 
> topology discovery which is some sort of tree walk. All of the analysis up to
> this point is independent of the actual order of the switches, but lash will
> use that order to enumerate the paths in the fabric and add them to the VL bins.
> 
> Odometer order means that the switches are labelled s[X0, ..., Xn-1] and
> ordered s[0, ..., 0], s[0, ..., 1], s[0, ..., Ln-1], s[0, .. 1, 0] etc.
> The dimensions are also reordered so that the dimension changing the fastest
> has the largest length, i.e. Ln >= Ln-1 >= ... >= L1. [All this is modulo
> possible end to end reversal but the basic idea is that the longest axis
> changes fastest.]
> 
> TO INVESTIGATE: Rather than using an additional switches array in
> sort_switches whether it can be done in place using p_lash->switches.
> 
> Signed-off-by: Robert Pearson <rpearson at systemfabricworks.com>
> Signed-off-by: Hal Rosenstock <hal.rosenstock at gmail.com>

Applied. Thanks.

Sasha