remove total slice buffer and reuse buffer from loader

[frickt]

for performane reason
this commit obsoletes #548 and #558

[coveralls]

Coverage decreased (-0.1%) to 62.315% when pulling 8b2e4c4 on frickt:remove_total_slices_buffer_and_reuse_from_loader into c0f2f05 on ivmartel:develop.

[ivmartel]

Thanks for your work! My problem is that you seem to have removed the frame extra dimension, If I understand your code, you consider the first dimension to be either frame or slice. The buffer is supposed to be an array of arrays, the first dimension being the frame and then the slice. The idea is to support data as the one provided in #493.

[frickt]

hm... I see, didn't know that.
So should we use 3 dimensions [frameIndex][sliceIndex][buffer]?

[ivmartel]

At the moment, there are just 2: [frameIndex][buffer]. Adding a third one could help to have an easier append slice like you do in this PR and avoid the copy.

[frickt]

ok, i rewrote it to use 3d buffer.

[ivmartel]

Looking good! So how does it compare to the current version of the code? Storing the 3D data in a huge 1d array is kind of old fashion, I'm not sure there is a real benefit compared to this array of array solution.

[frickt]

we don't need to copy the buffer anymore, therefore gain some performance. also garbage collection is reduced. Depending on the 3d data size, the performance (for buffer handling) is improved by ((numberOfSlices - 1)/2) times.

[frickt]

is there somethink i need to change get this merged?

[ivmartel]

No, looking good, I was just finishing the 0.24.1patch release.

[ivmartel]

And now, I'm reviving the dcmbench project to evaluate your changes, I'm nearly there...

[ivmartel]

Looking good on the default demo data of dcmbench, will integrate soon.

[ivmartel]

I know its an old one, but I need more time to investigate...

[ivmartel]

Closing for now, follow #905 for new strategy.