a good design

[ktprime]

hi,
I have added your qc-hash into my benchmark (https://github.com/ktprime/emhash/blob/master/bench/ebench.cpp),
some cases qc-hash is very fast but others quite slow. and it can not run(hanged) and martin bench(martin_bench.cpp).
can you fix the issuse?
I add my owern emhash7/emhash5 into your bench file benchmark.cpp(https://github.com/ktprime/emhash/blob/master/bench/qbench.cpp), it's a good benchmark.

[Daskie]

Interesting, thanks for taking the time to add this to your benchmark. I have not encountered any hanging in my own tests, but if you can provide the operation and kind of data that makes it hang, I would love to fix any bugs I may have missed.

As a side note, I noticed you're using std::hash with the map in your benchmark. I recommend using qc::hash::IdentityHash (the default) for keys with reasonable low-order entropy (which I think your random gamma distributed keys are) or qc::hash::FastHash otherwise. I would expect either to yield a notable step-up in speed.

[ktprime]

qc-hash is slow in insert-erase case. (ex https://martin.ankerl.com/2019/04/01/hashmap-benchmarks-03-03-result-RandomInsertErase/), you can test and reproduce https://github.com/ktprime/emhash/blob/master/bench/martin_bench.cpp (case 6 bench_insert, which is copyed from martin bench)
mybe the main reason is simple linear probing(I use 2-3 complex probling).

how to build my bench on bench dir
hyb@PC210500096:/mnt/f/emhash/bench$ clang++ -O3 -march=native -mtune=native -I.. -I../thirdparty -DQC_HASH=2 -std=c++20 martin_bench.cpp -o mbenc

or
clang++ -O3 -march=native -I.. -I../thirdparty -DFIB_HASH=2 -DQC_HASH=2 -std=c++20 ebench.cpp -o ebench

my bench can choose 6 hash functions by set FIB_HASH=(1-6) in build cmd.

template<typename T>
struct Int64Hasher
{
    static constexpr uint64_t KC = UINT64_C(11400714819323198485);
    inline std::size_t operator()(T key) const
    {
#if FIB_HASH == 1
        return key;
#elif FIB_HASH == 2
        return hashfib(key);
#elif FIB_HASH == 3
        return hash_mur3(key);
#elif FIB_HASH == 4
        return hashmix(key);
#elif FIB_HASH == 5
        return rrxmrrxmsx_0(key);
#elif FIB_HASH > 10000
        return key % FIB_HASH; //bad hash
#elif FIB_HASH > 100
        return key / FIB_HASH; //bad hash
#elif FIB_HASH == 6
        return wyhash64(key, KC);
#else
        auto x = key;
        x = (x ^ (x >> 30)) * UINT64_C(0xbf58476d1ce4e5b9);
        x = (x ^ (x >> 27)) * UINT64_C(0x94d049bb133111eb);
        x = x ^ (x >> 31);
        return x;
#endif
    }
};

[Daskie]

Thanks for the feedback! That's not a performance case I had considered. When I get some time I'll expand my benchmarking and decide whether that case is worth the increased complexity of a more sophisticated probing strategy, like the one you've come up with

[ktprime]

https://www.reddit.com/r/cpp/comments/ve3qbx/updating_map_benchmarks_send_your_hashmaps/