search

uzh-rpg / vilib

CUDA Visual Library by RPG
Other
428 stars 89 forks source link

Fixing synchronized behaviour on Volta and Turing microarchitectures #5

Closed baliika closed 4 years ago

baliika commented 4 years ago

Within our Grid NMS, __shfl_down_sync() was not called by all the threads, which resulted in a deadlock. I also cleaned up some unused functionality within the DetectorBaseGPU to make this fix more seamless.