The fastest recommended hash function by SMHasher.

The fastest passing hash in SMHasher3.

rapidhash is wyhash' official successor, with improved speed, quality and compatibility.

Fast
Extremely fast for both short and large inputs.
Surpasses 70GB/s on Apple's M4 cpus.
The fastest hash function passing all tests in SMHasher.
The fastest hash function passing all tests in SMHasher3.

Universal
Optimized for both AMD64 and AArch64 systems.
Compatible with gcc, clang, icx and MSVC.
It does not use machine-specific vectorized or cryptographic instruction sets.
Prepared for both C and C++ compilation.

Excellent
Passes all tests in both SMHasher and SMHasher3.
Collision-based study showed a collision probability lower than wyhash and close to ideal.
Outstanding collision ratio when tested with datasets of 16B and 66B keys:

More results can be found in the collisions folder

Average latency when hashing keys of 4, 8 and 16 bytes

Peak throughput when hashing files of 16Kb-2Mb

The benchmarking program can be found in the bench folder

A perfect hash function distributes its domain uniformly onto the image.
When the domain's cardinality is a multiple of the image's cardinality, each potential output has the same probability of being produced.
A function producing 64-bit hashes should have a $p=1/2^{64}$ of generating each output.

If we compute $n$ hashes, the expected amount of collisions should be the number of unique input pairs times the probability of producing a given hash.
This should be $(n*(n-1))/2 * 1/2^{64}$, or simplified: $(n*(n-1))/2^{65}$.
In the case of hashing $15*2^{30}$ (~16.1B) different keys, we should expect to see $7.03$ collisions.

We present an experiment in which we use rapidhash to hash $77$ datasets of $15*2^{30}$ (15Gi) keys each.
For each dataset, the amount of collisions produced is recorded as measurement.
Ideally, the average among measurements should be $7.031$ and the results collection should be a binomial distribution.
We obtained a mean value of $8.026$, just $14$% over $7.031$.
Each dataset individual result and the collisions test program can be found in the collisions folder.
The default seed $0$ was used in all experiments.