Currently: developing Road finding (traverse the tree of neighbour cells and create all candidated topologies)
- Traversing the tree is recursive function, no easy prediction on number of combinations
- Operations with indexes (Cells), reduced amount of candidates at this point
- Dryrun, fill index table, rerun with known starting points
PRs pending:
- https://github.com/AliceO2Group/AliceO2/pull/10955
CHEP plans:
- Road finding can be a realistic and reasonable point to arrive at.
- Understanding reasonably approvable plots for that time
- Overall I never see timing performance regression wrt CPU (with no discussion on optimisation), nor exaggerated speedups, there are features which can still be valuable (automatic code generation for HIP, intra TF flexibility, memory size configurability)
TL;DR: ITS GPU Vertexer (TimeFrame "splitting")

Step	on GPU	CPU consistency	Speedup	Optimised	Comments/Improvements	CUDA	HIP
Tracklet Finder	✅	✅	~	🚧		✅	✅
Tracklet Selection	✅	✅	~	🚧		✅	✅
Cell finder	❌	/	/	/	Investigate after tracking	❌	❌

Step	on GPU	CPU consistency	Speedup	Optimised	Comments/Improvements	CUDA	HIP
Tracklet Finder	✅	✅ (δ~1‰)	~	🚧	multi-ROF sliding window	✅	✅
Trkl duplicate finder	✅	✅	~	✅	thrust & CUB mainly	✅	✅
Cell finder	✅	🚧	~	🚧		✅	✅
Cell neghbour finder	✅	🚧	~	🚧		✅	✅
Road finder	🚧	🚧	~	🚧		🚧	🚧
Track fitting	❌	❌		❌		❌	❌