sort unique 16 non-zero values using avx512f+avx512bw

ssesort.cpp

#include <algorithm>
#include <iterator>
#include <cstdint>
#include <numeric>
#include <type_traits>
#include <bitset>
#include <cassert>
#include <random>

#include <iostream>
#include <iomanip>

#include <x86intrin.h>

// ls ssesort.cpp | entr -ncs 'clang++ -march=native -O3 -Wall -Wextra -stdlib=libc++ -std=c++23 -o ssesort main.cpp -o ssesort && ./ssesort'

int main()
{
    using I = uint16_t;
    constexpr size_t count = sizeof(__m512i) / sizeof(I);

    alignas(64) I inputs[count];
    std::iota(std::begin(inputs), std::end(inputs), I{1});
    std::random_device rd;
    std::mt19937 g(rd());
    std::shuffle(std::begin(inputs), std::end(inputs), g);
    auto mid = std::next(std::begin(inputs), std::ssize(inputs) / 2);
    std::sample(std::begin(inputs), mid, mid, std::distance(mid, std::end(inputs)), g);
    __m512i values = _mm512_load_si512(&inputs);

    alignas(64) I result[count];
    std::fill(std::begin(result), std::end(result), I{0});
    for (I input : inputs) {
	std::cout << input << " ";
        __m512i value = _mm512_set1_epi16(input);
        __mmask32 is_less_mask = _mm512_cmplt_epu16_mask(values, value);
        uint32_t index = _mm_popcnt_u32(_cvtmask32_u32(is_less_mask));
	result[index] = input;
    }
    std::cout << "\n";

    //__m512i r = _mm512_load_si512(&result);
    //__mmask32 z = _mm512_cmpneq_epi16_mask(_mm512_setzero_si512(), r);
    //__m512i c = _mm512_maskz_compress_epi16(z, r);
    //_mm512_store_si512(&result, c);
    
    //values = _mm512_permutexvar_epi16(_mm512_load_si512(&indices), values);

    auto r = std::remove(std::begin(result), std::end(result), I{0});
    std::copy(std::begin(result), r, std::ostream_iterator<I>(std::cout, " "));
    std::cout << "\n";

    std::sort(std::begin(inputs), std::end(inputs));
    auto i = std::unique(std::begin(inputs), std::end(inputs));
    std::copy(std::begin(inputs), i, std::ostream_iterator<I>(std::cout, " "));
    std::cout << "\n";
}