https://hgpu.org/?p=11072
Evaluating tradeoff between recall and performance of GPU permutation index