Searched refs:logical_lane_id (Results 1 – 2 of 2) sorted by relevance
64 uint32_t logical_lane_id = __kmpc_impl_popc(Liveness & lanemask_lt) * 2; in gpu_irregular_simd_reduce() local70 logical_lane_id /= 2; in gpu_irregular_simd_reduce()71 shflFct(reduce_data, /*LaneId =*/logical_lane_id, in gpu_irregular_simd_reduce()73 } while (logical_lane_id % 2 == 0 && size > 1); in gpu_irregular_simd_reduce()74 return (logical_lane_id == 0); in gpu_irregular_simd_reduce()
183 int logical_lane_id = find_number_of_dispersed_active_lanes_before_me() * 2;189 logical_lane_id /= 2;190 ShuffleReduceFn(reduce_data, logical_lane_id, remote_id-1-threadIdx.x, 2);191 } while (logical_lane_id % 2 == 0 && size > 1);212 concept of logical_lane_id, and it is important to distinguish it214 1. //logical_lane_id//: as the name suggests, it refers to the calculated216 our algorithm logically concise. A thread with logical_lane_id k means233 f(x): logical_lane_id -> physical_lane_id;235 f^-1(x): physical_lane_id -> logical_lane_id243 logical_lane_id is recalculated; so is the remote_id.