simon:cuda blyth$ grep atomic *.*
daq.cu: atomicMin(earliest_time_int + channel_index, time_int);
daq.cu: atomicAdd(channel_q_int + channel_index, charge_int);
daq.cu: atomicOr(channel_histories + channel_index, history);
daq.cu: atomicMin(earliest_time_int + channel_offset, time_int);
daq.cu: atomicAdd(channel_q_int + channel_offset, charge_int);
daq.cu: atomicOr(channel_histories + channel_offset, history);
hybrid_render.cu: data = atomicExch(addr, data+atomicExch(addr, 0.0f));
pdf.cu: atomicMin(&distance_table_len, i);
propagate.cu: atomicAdd(&counter, 1);
propagate.cu: atomicAdd(index_counter, counter);
propagate.cu: int offset = atomicAdd(index_counter, 1);
propagate.cu: int out_idx = atomicAdd(output_queue, 1);
Capability <2.0 (but >=1.3)
// For capability <2.0 (but >=1.3) GPUs
#define ATOMIC_ADD(x,y) MyAtomicAdd (x, (float)(y))
#else
#define ATOMIC_ADD(x,y) atomicAdd (x, (float)(y))
#endif