Implementing and optimizing a simple graph metric

[Joseph Rushton Wakeling]

On Wednesday, 18 September 2013 at 13:39:29 UTC, bearophile wrote:
>     auto centrality = 
> minimallyInitializedArray!(typeof(return))(g.vertexCount);
>     centrality[] = T0;
>     auto stack = new size_t[g.vertexCount];
>     auto sigma = minimallyInitializedArray!T(g.vertexCount);
>     sigma[] = T0;
>     auto delta = minimallyInitializedArray!T(g.vertexCount);
>     delta[] = T0;
>     auto d = minimallyInitializedArray!long(g.vertexCount);
>     d[] = -1;
>     auto q = VertexQueue(g.vertexCount);
>     auto p = new Appender!(size_t[])[g.vertexCount];

As an experiment I tried something along these lines -- using 
uninitializedArray for most of the arrays here and 
minimallyInitializedArray for p.

It seems to result in a small speed improvement, although I'm not 
certain about that -- certainly not a speed loss.

On the other hand, if inside the VertexQueue implementation, I 
replace the "new" declaration with an uninitializedArray call, 
the code gets slower by a noticeable amount.

Very odd -- any ideas why that might be?