Newtrack

[JosephPeterDevlin]

No description provided.

[mattsignorelli]

@JosephPeterDevlin could you take a look at my modifcations?

[JosephPeterDevlin]

@mattsignorelli There is a problem with them, this line with the Toy AGS as ring causes a seg fault:

M = track(ring, [0.0 0.0 2.5e-4 0.0 0.0 0.0], [0.0 1.0 0.0]; n_turns=1000);

[mattsignorelli]

Ok I kinda fixed it but it is super slow. Was it this slow in your original implementation?

[mattsignorelli]

1 particle for 1000 turns should not take several seconds

[mattsignorelli]

I guarantee most of the tracking time is spent writing to the tensor and poor performance practices in the track loop (which i too am guilty of).

I wonder if we can try to quickly eek-out a memcopy kernel and just skip this 3-vector every single turn thing instead just tracking the quaternion always

[JosephPeterDevlin]

No, it wasn't that slow before. 1 particle for 10,000 turns was 6 seconds before.

[JosephPeterDevlin]

I just tried it with storing 3-vectors and 1 particle for 1,000 turns only took 0.76 seconds. This is about the same speed as before. I don't understand the advantage of tracking the quaternion always if I'm going to want to turn every single turn into a spin vector in the end anyway.