def atomicAggInc(counter):
active = cuda.cg.coalesced_threads()
mask = active.ballot(True)
# Select the leader
leader = cuda.ffs(mask) - 1
# Leader does the update
if active.thread_rank == leader:
res = cuda.atomic.add(counter, 0, cuda.popc(mask))
# Broadcast result
res = active.shfl(res, leader)
# Each thread computes its own value
return res + cuda.popc(mask & ((1 << active.thread_rank) - 1))
def simple_ffs(ary, c):
ary[0] = cuda.ffs(c)
def simple_ffs(ary, c):
ary[0] = cuda.ffs(c)
