Main Content

Thrust Example

With Thrust library support in GPU Coder™, you can take advantage of GPU-accelerated primitives such as sort to implement complex high-performance parallel applications. When your MATLAB® code uses gpucoder.sort function instead of sort, GPU Coder can generate calls to the Thrust sort primitives.

This example generates CUDA® code to sort the columns of a matrix in descending order. In one file, write an entry-point function mySort that accepts a matrix inputs A. Use the gpucoder.sort function to sort the columns of A in descending order.

function B = mySort(A)
     B = gpucoder.sort(A, 1, 'descend');
end

Use the codegen function to generate CUDA MEX function.

codegen -config coder.gpuConfig('mex') -args {ones(1024,1024,'double')} -report mySort

Generated CUDA Code

The following is a snippet of the generated code. The Thrust library call is denoted by thrustSortImpl

...
cudaMalloc(&gpu_inDims, 8ULL);
cudaMalloc(&gpu_B, 8388608ULL);
cudaMalloc(&gpu_A, 8388608ULL);
mySort_kernel1<<<dim3(1U, 1U, 1U), dim3(32U, 1U, 1U)>>>(*gpu_inDims);
cudaMemcpy(gpu_A, (void *)&A[0], 8388608ULL, cudaMemcpyHostToDevice);
mySort_kernel2<<<dim3(2048U, 1U, 1U), dim3(512U, 1U, 1U)>>>(*gpu_A, *gpu_B);
cudaMemcpy(&inDims[0], gpu_inDims, 8ULL, cudaMemcpyDeviceToHost);
thrustSortImpl(&(*gpu_B)[0], 2, &inDims[0], 1, 'd', false);
cudaMemcpy(&B[0], gpu_B, 8388608ULL, cudaMemcpyDeviceToHost);
...