Skip to content

Use less shared memory in looking forward triplet seeding

Daniel Hugo Campora Perez requested to merge dcampora_less_shared_memory_looking_forward into master

The logic of triplet seeding has been divided into two kernels, lf_triplet_seeding and lf_triplet_keep_best. That way, less shared memory is used. Additionally:

  • Calling number of threads optimized
  • Shared memory usage in Tensor core code changed to require less resources

Merge request reports