cudaFlow Algorithms » Parallel Merge

Merge two Sorted Ranges of Items

tf::cudaFlow::merge performs a parallel merge over two ranges of elements into a sorted range of items.

The following code merges two sorted arrays input_1 and input_2, each of 1000 items, into a sorted array output of 2000 items.

const size_t N = 1000;
int* input_1 = tf::cuda_malloc_shared<int>(N);    // input vector 1
int* input_2 = tf::cuda_malloc_shared<int>(N);    // input vector 2
int* output  = tf::cuda_malloc_shared<int>(2*N);  // output vector

// initializes the data
for(size_t i=0; i<N; i++) {
  input_1[i] = rand()%100;
  input_2[i] = rand()%100;
}
std::sort(input_1, input1 + N);
std::sort(input_2, input2 + N);

// merge input_1 and input_2 to output
tf::cudaFlow cf;
tf::cudaTask task = cf.merge(
  input_1, input_1 + N, input_2, input_2 + N, output, 
  []__device__ (int a, int b) { return a < b; }  // comparator
);
cf.offload();

Miscellaneous Items

Parallel merge algorithms are also available in tf::cudaFlowCapturer with the same API.