C++Now 2018: You Can Do Better than std::unordered_map: New Improvements to Hash Table Performance

Many videos on hash table are great. I found this one particularly good. [ ▶ this video ] C++Now 2018: You Can Do Better than std::unordered_map: New Improvements to Hash Table Performance

In my example the table stores the hash codes themselves instead of the keys (because the hash function is invertible) Oh, I see, right. If determining the home bucket is trivial, then the back-shifting method is great. The issue is just that it’s not as much of a general-purpose solution as it may initially seem. In the bad cases, it's really bad. With a different algorithm (Robin Hood or bidirectional linear probing), the load factor can be kept well over 90% with good performance, as the benchmarks in the same repo demonstrate. I’ve seen the 90% claim made several times in literature on Robin Hood hash tables. In my experience, the claim is a bit exaggerated, although I suppose it depends on what our idea of “good performance” is. See these benchmarks, which again go up to a maximum load factor of 0.95 (although boost and Absl forcibly grow/rehash at 0.85-0.9): link Tsl, Martinus, and CC are all Robin Hood tables ( link , link , and link , respectively). Absl (from Google) and Boost are well-known C++ SIMD-based hash tables. Khash ( link ) is, I think, an ordinary open-addressing table using quadratic probing. Fastmap is a new, yet-to-be-published design that is fundamentally similar to bytell ( ▶ this video ) but also incorporates some aspects of the aforementioned SIMD maps (in particular, it stores a 4-bit fragment of the hash code to avoid most false key comparisons). As you can see, all the Robin Hood maps spike upwards dramatically as the load factor gets high,

It has the right asymptotic guarantees for a hash map, but beyond that it doesn't have much going for it. The degree to which it emphasizes pointer/iterator stability and arbitrary load factors is not a good tradeoff in almost any practical case (unless you do not care about performance). See ▶ this video for a more comprehensive discussion, or the sibling threads here.

> some kind of bucket-optimisation (multiple elements per allocation, instead of one element per allocation) should also meet all constraints the standard sets for this type Pretty sure that makes it impossible to implement merge without pointer/iterator invalidation, which is a constraint imposed by the standard ( link ). > it would be good if the person I originally responded to came out and defended his POV See link if you want numbers, or for example this talk by the same author ▶ this video .