TreePredictor Class Reference

Collaboration diagram for TreePredictor:

List of all members.

---

Detailed Description

The following collectives prediction is based on our implementation with the recursive formula T(r) = predict_p2p(r, getChild(r)) + max(T(getChild(r)),T(r \ <subtree with root getChild(r)>)) if r is not a leaf where getChild(r) = rightmost child of r if R2L & TOP-DOWN = BCAST getChild(r) = leftmost child of r if L2R & DOWN-TOP = REDUCE T(r) = 0 if r is a leaf (probably too idealistic) assumptions:

• a P2P communication (a,b) takes places exactly when both a and b are not involved in any other P2P communication
• any two pairs (a,b) and (c,d) can communicate in parallel
• symmetry - predict_p2p (a,b) = predict_p2p(b,a) - otherwise bcast and reduce would have different implementations (reverse directions)

---

The documentation for this class was generated from the following file:

• models/cpm_predict_trees.hpp