How to reuse temporary buffer across multiple calls of thrust::sort_by_key?

Accelerated Computing CUDA CUDA Programming and Performance
1

I need to call thrust::sort_by_key multiple times with same input(same device array with different data).
How to tell thrust to reuse temporary buffers?
I don’t want cudaMalloc/cudaFree be called each time I call thrust::sort_by_key because it lower down performance.

2

You also have direct control over this using cub.

3

Nice, thanks a lot :)