Feature wall distance

[economon]

This contains a new implementation of tree-based searches with a first application to computing wall distances for RANS calculations (thanks due to Edwin!). Initial scaling tests look good, but still completing some... checking for correctness now.

[economon]

Everything looks good here with the ADT search. Merging this into the develop branch. Two notes:

1. I switched MPI_Allgatherv() to SU2_MPI:Allgather() temporarily until we can properly treat the wrapper. This will be easy to change. The code is slightly less pretty, but same results.
2. Scaling results look very good out to 96 cores and can be ~100x faster than the brute force search. I'll run on O(1000) cores as soon as I have access again to make sure there aren't any hidden problems at scale.

[hlkline]

Looks good, thanks @vdweide, @economon.
I took a quick peak, it looks like there wasn't any change to the interpolation routines - can this same tree search be applied to finding nearest neighbors in the interpolation routines, and if so did you look into that yet or would that be a project for the future?

[vdweide]

I expect that the tree search can be used for the interpolation routines without a problem. However, we have not looked into that yet.

[LaSerpe]

Heather, all,

As you know I'm working on sliding meshes and that's something we were really looking forward.

In my opinion including the ADT search inside the interpolator class to find the nearest neighbour will be mandatory when we will get to simulate multi-stage turbines.

I did't have a proper look at those routines yet but I also do not expect any particular problem in using them inside the interpolator class.

(Actually one item in my "todo" list is indeed to include those routines in the interpolator class.)

Giulio

---

Da: Edwin van der Weide [email protected]
Inviato: giovedì 7 luglio 2016 23.47.21
A: su2code/SU2
Oggetto: Re: [su2code/SU2] Feature wall distance (#288)

I expect that the tree search can be used for the interpolation routines without a problem. However, we have not looked into that yet.

You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHubhttps://github.com//pull/288#issuecomment-231217312, or mute the threadhttps://github.com/notifications/unsubscribe/AHg4OBXfHwCsJhsAn3mhMiYi_44mr1A0ks5qTXPpgaJpZM4JF5wW.