I have tried swiss tables but I have been unable to make these match the performance of robin hood tables with a single-stage lookup. Fundamentally the issue se

I have tried swiss tables but I have been unable to make these match the performance of robin hood tables with a single-stage lookup. Fundamentally the issue seems to be that swiss tables require one to read 2 or 3 cache lines while the alternative requires one to read only 1 cache line per lookup (most of the time). The included benchmarks are not that useful for inserting many items or looking up many items because one can achieve much higher throughput using prefetching (like <<10ns/op), but this sort of consideration never makes it into any graphs.