#include <sha512.cuh>
__device__ __constant__ sha512_state sha512_initial_state = { {
    0x6a09e667f3bcc908ULL, 0xbb67ae8584caa73bULL,
    0x3c6ef372fe94f82bULL, 0xa54ff53a5f1d36f1ULL,
    0x510e527fade682d1ULL, 0x9b05688c2b3e6c1fULL,
    0x1f83d9abfb41bd6bULL, 0x5be0cd19137e2179ULL,
} };
__device__ __constant__ unsigned long round_k[80] = {
    0x428a2f98d728ae22ULL, 0x7137449123ef65cdULL, 0xb5c0fbcfec4d3b2fULL, 0xe9b5dba58189dbbcULL,
    0x3956c25bf348b538ULL, 0x59f111f1b605d019ULL, 0x923f82a4af194f9bULL, 0xab1c5ed5da6d8118ULL,
    0xd807aa98a3030242ULL, 0x12835b0145706fbeULL, 0x243185be4ee4b28cULL, 0x550c7dc3d5ffb4e2ULL,
    0x72be5d74f27b896fULL, 0x80deb1fe3b1696b1ULL, 0x9bdc06a725c71235ULL, 0xc19bf174cf692694ULL,
    0xe49b69c19ef14ad2ULL, 0xefbe4786384f25e3ULL, 0x0fc19dc68b8cd5b5ULL, 0x240ca1cc77ac9c65ULL,
    0x2de92c6f592b0275ULL, 0x4a7484aa6ea6e483ULL, 0x5cb0a9dcbd41fbd4ULL, 0x76f988da831153b5ULL,
    0x983e5152ee66dfabULL, 0xa831c66d2db43210ULL, 0xb00327c898fb213fULL, 0xbf597fc7beef0ee4ULL,
    0xc6e00bf33da88fc2ULL, 0xd5a79147930aa725ULL, 0x06ca6351e003826fULL, 0x142929670a0e6e70ULL,
    0x27b70a8546d22ffcULL, 0x2e1b21385c26c926ULL, 0x4d2c6dfc5ac42aedULL, 0x53380d139d95b3dfULL,
    0x650a73548baf63deULL, 0x766a0abb3c77b2a8ULL, 0x81c2c92e47edaee6ULL, 0x92722c851482353bULL,
    0xa2bfe8a14cf10364ULL, 0xa81a664bbc423001ULL, 0xc24b8b70d0f89791ULL, 0xc76c51a30654be30ULL,
    0xd192e819d6ef5218ULL, 0xd69906245565a910ULL, 0xf40e35855771202aULL, 0x106aa07032bbd1b8ULL,
    0x19a4c116b8d2d0c8ULL, 0x1e376c085141ab53ULL, 0x2748774cdf8eeb99ULL, 0x34b0bcb5e19b48a8ULL,
    0x391c0cb3c5c95a63ULL, 0x4ed8aa4ae3418acbULL, 0x5b9cca4f7763e373ULL, 0x682e6ff3d6b2b8a3ULL,
    0x748f82ee5defb2fcULL, 0x78a5636f43172f60ULL, 0x84c87814a1f0ab72ULL, 0x8cc702081a6439ecULL,
    0x90befffa23631e28ULL, 0xa4506cebde82bde9ULL, 0xbef9a3f7b2c67915ULL, 0xc67178f2e372532bULL,
    0xca273eceea26619cULL, 0xd186b8c721c0c207ULL, 0xeada7dd6cde0eb1eULL, 0xf57d4f7fee6ed178ULL,
    0x06f067aa72176fbaULL, 0x0a637dc5a2c898a6ULL, 0x113f9804bef90daeULL, 0x1b710b35131c471bULL,
    0x28db77f523047d84ULL, 0x32caab7b40c72493ULL, 0x3c9ebe0a15c9bebcULL, 0x431d67c49c100d4cULL,
    0x4cc5d4becb3e42b6ULL, 0x597f299cfc657e2aULL, 0x5fcb6fab3ad6faecULL, 0x6c44198c4a475817ULL,
};
__device__ __forceinline__ unsigned long load64(const unsigned char* __restrict__ x) {
    return (static_cast<unsigned long>(x[0]) << 56) | (static_cast<unsigned long>(x[1]) << 48) |
        (static_cast<unsigned long>(x[2]) << 40) | (static_cast<unsigned long>(x[3]) << 32) |
        (static_cast<unsigned long>(x[4]) << 24) | (static_cast<unsigned long>(x[5]) << 16) |
        (static_cast<unsigned long>(x[6]) << 8) | (static_cast<unsigned long>(x[7]));
}
__device__ __forceinline__ void store64(unsigned char* __restrict__ x, unsigned long v) {
#pragma unroll 8
    for (int i = 0; i < 8; i++) {
        x[i] = (unsigned char)(v >> (56 - i * 8));
    }
}
__device__ __forceinline__ unsigned long rot64(unsigned long x, int bits) {
    return (x >> bits) | (x << (64 - bits));
}
__device__ __forceinline__ void sha512_block(sha512_state* __restrict__ s, const unsigned char* __restrict__ blk) {
    unsigned long w[16];
#pragma unroll 16
    for (int i = 0; i < 16; i++) {
        w[i] = load64(blk + i * 8);
    }
    unsigned long a = s->h[0];
    unsigned long b = s->h[1];
    unsigned long c = s->h[2];
    unsigned long d = s->h[3];
    unsigned long e = s->h[4];
    unsigned long f = s->h[5];
    unsigned long g = s->h[6];
    unsigned long h = s->h[7];
#pragma unroll 80
    for (int i = 0; i < 80; i++) {
        const int idx = i & 15;
        const int idx1 = (i + 1) & 15;
        const int idx7 = (i + 9) & 15;
        const int idx14 = (i + 14) & 15;
        unsigned long s0 = rot64(w[idx1], 1) ^ rot64(w[idx1], 8) ^ (w[idx1] >> 7);
        unsigned long s1 = rot64(w[idx14], 19) ^ rot64(w[idx14], 61) ^ (w[idx14] >> 6);
        unsigned long S0 = rot64(a, 28) ^ rot64(a, 34) ^ rot64(a, 39);
        unsigned long S1 = rot64(e, 14) ^ rot64(e, 18) ^ (rot64(e, 41));
        unsigned long temp1 = h + S1 + ((e & f) ^ (~e & g)) + round_k[i] + w[idx];
        unsigned long temp2 = S0 + ((a & b) ^ (a & c) ^ (b & c));
        h = g;
        g = f;
        f = e;
        e = d + temp1;
        d = c;
        c = b;
        b = a;
        a = temp1 + temp2;
        w[idx] += s0 + w[idx7] + s1;
    }
    s->h[0] += a;
    s->h[1] += b;
    s->h[2] += c;
    s->h[3] += d;
    s->h[4] += e;
    s->h[5] += f;
    s->h[6] += g;
    s->h[7] += h;
}
__device__ void sha512_final(sha512_state* s, const unsigned char* blk) {
    unsigned char temp[128];
    unsigned long last_size = 32 & (128 - 1);
    memcpy(temp, blk, last_size);
    temp[last_size] = 0x80;
    if (last_size > (128 - 9)) {
        sha512_block(s, temp);
        memset(temp, 0, sizeof(temp));
    }
    store64(temp + 128 - 8, 256);
    sha512_block(s, temp);
}
__device__ void sha512_get(const sha512_state* s, unsigned char* hash) {
    unsigned len = 64;
    if (len > 128) len = 128;
    unsigned i = 0, c = (len < 8) ? len : 8;
    store64(hash, s->h[i]);
    hash += c;
    len -= c;
    i++;
    while (len >= 8) {
        store64(hash, s->h[i]);
        hash += 8;
        len -= 8;
        i++;
    }
    if (len > 0) {
        unsigned char tmp[8];
        store64(tmp, s->h[i]);
        memcpy(hash, tmp, len);
    }
}