search

fastmachinelearning / SonicCMS

Services for Optimized Network Inference on Coprocessors (for CMS)
8 stars 8 forks source link

Example of converting GPU algorithm to Triton custom backend #19

Open kpedro88 opened 4 years ago

kpedro88 commented 4 years ago
  1. pick a classical algorithm (e.g. HCAL reconstruction) that has already been ported to CUDA
  2. get GPU code for that algorithm
  3. set up custom backend using library for that algorithm
  4. set up associated Sonic producer to handle inputs/outputs
  5. document procedures for steps 3 and 4

assigned to: @violatingcp, ?

kpedro88 commented 4 years ago

@pierinim are you interested in this?