How to use Thrust (or cub) reduction maximum for key/object?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I can’t work out what function thrust::reduce needs for scoring my objects.
I think I might need to change just one or two lines, but I don’t see how.
I’d appreciate your insights.

// Needs fixing - but how?

Thanks,

Paul

class scored_orientation_t {
public:
   __device__ scored_orientation_t(float score_in, int idx_in) : score(score_in), idx(idx_in) {}
   int idx;
   float score;
   __device__ bool operator<(const scored_orientation_t &o) const {
     return (score < o.score);
   }
   __device__ bool operator>(const scored_orientation_t &o) const {
     return (score > o.score);
   }
};

an array of which holds the scores of various orientations.

I want to find the orientation parameters (currently merely idx) with the maximum score

unsigned int N_trials = 10000;
scored_orientation_t *scored_orientations_p;
cudaMalloc(&scored_orientations_p, N_trials * sizeof(scored_orientation_t));

         // .. code to set the scores here ..

// wrap raw pointer with a device_ptr 
thrust::device_ptr<scored_orientation_t> dev_ptr(scored_orientations_p);

// Needs fixing - but how?
thrust::device_ptr<scored_orientation_t> best_via_thrust = thrust::reduce(dev_ptr, dev_ptr + N_trials, 0, thrust::max_element<scored_orientation_t>());
2

Found it (don’t use thrust::reduce, use thrust::max_element):

scored_orientation_t *best_result_thrust_dev_p = thrust::max_element(thrust::device, scored_orientations_p, scored_orientations_p + N_trials);