- Uses Blelloch's Algorithm (exclusive scan)
- Not limited by 2048 items (a former restriction on the initial implementation of the algorithm due to the maximum threads that can run in a thread block on current GPUs)
- Not limited by input sizes that are powers of 2 (a former restriction due to inherent binary tree-approach of the algorithm)
- Free of shared memory bank conflicts using the index padding method in this paper.
amallia / gpu-prefix-sum Goto Github PK
View Code? Open in Web Editor NEWThis project forked from mark-poscablo/gpu-prefix-sum
CUDA implementation of exclusive prefix sum via Blelloch's algorithm