OpenCL

No news.

GPU Servers

CI Server: Sergio finished CI integration.

Looking into setting up workflow to run standalone GPU tool.

Need new job variant ci-jobs? Repo is private.

Highly Ionizing Particles

Biggest changes since last meeting:

• Fixed issue where filter missed saturated bins (-> All bins with ADC == 1023 are now cleared)
• Added rising edge detection
• Added exponential filter for tail cutoff
• Cluster creation is now working

Filter Performance

Rough parameter sweep over threshold and exp filter alpha:

• Performance increased to 60% - 70% depending on tolerated false positive rate.
• Filter performance looks better when checking #ADC instead of #digit cleared. -> Reach +80% of ADCs cleared 
• Caveats: cluster pad width appears to be a fat tail distribution. Filter currently goes +- 5 pads, but ~10% digits and ADC outside of that window and never seen by filter (=> SOL is 90% for filter performance)
• Low alpha (slow update rate): Higher recall, but false positive rate explodes

Cluster Creation

Short example for cluster creation. Also shows failure modes of HIP filter.

MC:

Cluster + MC:

Runtime

Cluster creation currently parallel across rows. Still need some idea how to parallelize across tails, or at least make tail -> cluster matching faster single threaded.