TiledCUDA is a highly efficient kernel template library designed to elevate CUDA C’s level of abstraction for processing tiles.
157
stars
10
forks
source link
fix(cell): store tiles to shared memory with no bank-conflicts. #149
Closed
haruhi55 closed 1 month ago
This PR implements bank-conflict-free storage of tiles from registers to shared memory and from shared memory to global memory.