package halfsiphash

import "encoding/binary"

func rotl(x uint32, b uint) uint32 {
	return (x << b) | (x >> (32 - b))
}

func sipround(v0, v1, v2, v3 *uint32) {
	*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)
}

// Sum32 computes HalfSipHash-2-4 with an 8-byte key and returns a 4-byte hash.
func Sum32(data []byte, key [8]byte) uint32 {
	k0 := binary.LittleEndian.Uint32(key[0:4])
	k1 := binary.LittleEndian.Uint32(key[4:8])

	v0 := uint32(0) ^ k0
	v1 := uint32(0) ^ k1
	v2 := uint32(0x6c796765) ^ k0
	v3 := uint32(0x74656462) ^ k1

	// Process full 4-byte blocks.
	nblocks := len(data) / 4
	for i := 0; i < nblocks; i++ {
		m := binary.LittleEndian.Uint32(data[i*4:])
		v3 ^= m
		for j := 0; j < 2; j++ {
			sipround(&v0, &v1, &v2, &v3)
		}
		v0 ^= m
	}

	// Process remaining bytes.
	b := uint32(len(data)) << 24
	tail := data[nblocks*4:]
	switch len(tail) {
	case 3:
		b |= uint32(tail[2]) << 16
		fallthrough
	case 2:
		b |= uint32(tail[1]) << 8
		fallthrough
	case 1:
		b |= uint32(tail[0])
	}

	v3 ^= b
	for i := 0; i < 2; i++ {
		sipround(&v0, &v1, &v2, &v3)
	}
	v0 ^= b

	v2 ^= 0xff

	for i := 0; i < 4; i++ {
		sipround(&v0, &v1, &v2, &v3)
	}

	return v1 ^ v3
}