1 / 11

Fast Ring reconstruction in the CBM RICH detector

Fast Ring reconstruction in the CBM RICH detector. Semen Lebedev GSI, Darmstadt, Germany and LIT JINR, Dubna, Russia Claudia Höhne GSI, Darmstadt Gennady Ososkov LIT JINR, Dubna, Russia. Two main innovations were developed in ring reconstruction:

jetta
Download Presentation

Fast Ring reconstruction in the CBM RICH detector

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Fast Ring reconstruction in the CBM RICH detector Semen Lebedev GSI, Darmstadt, Germany and LIT JINR, Dubna, Russia Claudia Höhne GSI, Darmstadt Gennady Ososkov LIT JINR, Dubna, Russia

  2. Two main innovations were developed in ring reconstruction: SIMDized (vectors) version of Hough Transform Multithreading version of Hough Transform using ITBB (Intel threading building blocks) Outlook

  3. Many cores CPUs -> algorithms can be parallelized Implementation of parallel algorithm is a challenge!!! But we have no choice!!! Parallelization with threads S. Borkar et al. (Intel), "Platform 2015: Intel Platform Evolution for the Next Decade", 2005. • Fast algorithm -> less computers requirements • Possibility to use on-line reconstruction

  4. Ring recognition algorithm (basic idea) Preliminary selection of hits Histogram of ring centers Hough Transform Remove hits of found ring(only best matched hits) Ring quality calculation Ring array Ellipse fitter

  5. Divide hits into a several groups Make Hough Transform of each group independently Optimization of combinatory First group of hits Second group of hits Hough Transform Hough Transform Can be done in parallel!!! Sum up histogram

  6. Parallelization with threads Parallelization of Hough Transform algorithm can be done on different levels. 1 ) Two independent photodetectors 2 ) Many independent local Ring reconstructions 3 ) Divide hits into a several groups. Independent HT for each group

  7. SSE is available now!!! Today: 128 bit registers (4 float operations) Tomorrow: AVX (256 bits or 8 float operations) LRB (512 bits or 16 float operations) Introduction to SIMD float float float float 32 bits 32 bits 32 bits 32 bits 128 bit register 128 bit register 4 float operations concurrently 128 bit register

  8. Requirements: The best candidate for SIMDitaion is a “heavy” procedure without branches (if, break, continue), which can work in parallel for multiple data. Goal: rethink (reimplement) your algorithm in a way of eliminating all branches in the code. Hard task!!! How to implement SIMD

  9. Implementation of SIMD version of Hough Transform is a challenge (no “heavy” procedure and many branches). Idea: calculate ring parameters (x,y,r) from triplets in parallel. 4 parameters instead of 1. To implement this idea one needed to eliminate all branches in the code and reimplement HT. Hough Transform and SIMD Detector space HT Space r Hough Transform yc xc

  10. 82 msec - fast serial version 55 msec – SIMDized version (1.5) 62 msec – Multithreading version (1.35) Results of parallelization • Simulation: 500 UrQMD events at 25 AGeV + 10 e+ and e- embedded in each event. • Compact RICH

  11. Ring reconstruction works good with 93% efficiency for Compact RICH Parallel (SIMD and Multithreading) version of the Hough Transform was developed and resulted in 62 and55 msec/event time. → continue speed up optimization of the algorithm Summary

More Related