- DCA Fitter on GPU:
- As tests worked, a first showcase application on an analysis task (plot attached for reference): link.
- Currently, we need to copy the track parameters from the arrow table to construct the tracks. Ideally, this could be done on the GPU as well.
- Will try to adapt a task (this or similar in HF, where DCA fits are a bottleneck) with the
processBulk
version (see below).
- Added
processBulk()
CPU handler to perform simultaneous fitting (1 Fit/thread) of a vector of TrackParCov
s: PR.
- Added kernel timing. On 10000 fits, 20 blocks, and 512 threads, we are comparable with the CPU, a scaling study is planned.
- ITS Tracking on GPU:
- No progress this week; I will start benchmarking on EPN this afternoon.
- Re-create consistency checks