Iteration performance

[9il]

aSlice[i] requires 1 addition and 1 multiplication
aSlice[i, j] requires 2 additions and 2 multiplications
90% of algorithm can be iterated using ndslice.algorithm (ndMap, ndReduce). Other 10% can use front!d, popFront!d methods

[9il]

For example, this code can be optimizaed using assumeSameStructure (zip analog) and ndEach

        foreach(r; iota(f1.height).parallel)
        {
            foreach(c; 0 .. f1.width)
            {
                auto id = r * f1.width + c;
                f1s[r, c] = cast(float)f1d[id];
                f2s[r, c] = cast(float)f2d[id];
            }
        }

As Walter said, idiomatic D code should not have loops. This is really true for ndslice.

[ljubobratovicrelja]

I agree with everything you've stated. I assume we could include these changes in same basket as migration to Mir and ndslice.algorithm?

[9il]

Yes

[9il]

The code like [i, j] still exist, for example in https://github.com/ljubobratovicrelja/dcv/blob/master/source/dcv/tracking/opticalflow/lucaskanade.d

[ljubobratovicrelja]

Aw, man... :| Yea, I haven't cleaned optical flow modules, wanted to cut #58 short. Reopening...