Skip to content

algo: Optimize host code in GPU Hitfinder.

Felix Weiglhoferrequested to merge
fweig/cbmroot:gpu-optimization into master

Introduces three optimizations to the host code of the STS Hitfinder:

  • Allocate all buffers beforehand on GPU, do dynamic allocation per TS on CPU (allocating the pinned memory for GPU is very slow)
  • Speedup sorting Digis by module by using a lookup table to map the sensor address to an index and parallelizing the copy
  • Copy hits to pinned memory to speed up copy from the GPU
Edited by Felix Weiglhofer

Merge request reports