#pragma once
#include <cstdint>
#include <cstddef>
#include <span>

namespace halfsiphash {

namespace detail {

constexpr uint32_t rotl(uint32_t x, int b) {
  return (x << b) | (x >> (32 - b));
}

constexpr uint32_t load_le32(const uint8_t *p) {
  return static_cast<uint32_t>(p[0])
       | (static_cast<uint32_t>(p[1]) << 8)
       | (static_cast<uint32_t>(p[2]) << 16)
       | (static_cast<uint32_t>(p[3]) << 24);
}

inline void store_le32(uint8_t *p, uint32_t v) {
  p[0] = static_cast<uint8_t>(v);
  p[1] = static_cast<uint8_t>(v >> 8);
  p[2] = static_cast<uint8_t>(v >> 16);
  p[3] = static_cast<uint8_t>(v >> 24);
}

inline void sipround(uint32_t &v0, uint32_t &v1, uint32_t &v2, uint32_t &v3) {
  v0 += v1; v1 = rotl(v1, 5);  v1 ^= v0; v0 = rotl(v0, 16);
  v2 += v3; v3 = rotl(v3, 8);  v3 ^= v2;
  v0 += v3; v3 = rotl(v3, 7);  v3 ^= v0;
  v2 += v1; v1 = rotl(v1, 13); v1 ^= v2; v2 = rotl(v2, 16);
}

} // namespace detail

// Compute HalfSipHash-2-4 with an 8-byte key, returning a 32-bit hash.
inline uint32_t hash32(std::span<const uint8_t> data, const uint8_t key[8]) {
  using namespace detail;

  uint32_t k0 = load_le32(key);
  uint32_t k1 = load_le32(key + 4);

  uint32_t v0 = 0 ^ k0;
  uint32_t v1 = 0 ^ k1;
  uint32_t v2 = UINT32_C(0x6c796765) ^ k0;
  uint32_t v3 = UINT32_C(0x74656462) ^ k1;

  const uint8_t *end = data.data() + data.size() - (data.size() % 4);
  for (const uint8_t *p = data.data(); p != end; p += 4) {
    uint32_t m = load_le32(p);
    v3 ^= m;
    sipround(v0, v1, v2, v3);
    sipround(v0, v1, v2, v3);
    v0 ^= m;
  }

  uint32_t b = static_cast<uint32_t>(data.size()) << 24;
  switch (data.size() & 3) {
  case 3: b |= static_cast<uint32_t>(end[2]) << 16; [[fallthrough]];
  case 2: b |= static_cast<uint32_t>(end[1]) << 8;  [[fallthrough]];
  case 1: b |= static_cast<uint32_t>(end[0]);        break;
  case 0: break;
  }

  v3 ^= b;
  sipround(v0, v1, v2, v3);
  sipround(v0, v1, v2, v3);
  v0 ^= b;

  v2 ^= 0xff;
  sipround(v0, v1, v2, v3);
  sipround(v0, v1, v2, v3);
  sipround(v0, v1, v2, v3);
  sipround(v0, v1, v2, v3);

  return v1 ^ v3;
}

} // namespace halfsiphash