#include <stdint.h>
#include <stddef.h>

#define BATCH_SIZE    150000
#define MAX_MSG_LEN   256
#define NONCE_MAX_LEN 128

/* Global state */
static char solution_str[32];
static uint8_t hash_output[32];
static uint8_t message_buffer[MAX_MSG_LEN];

static uint32_t nonce_length     = 0;
static uint32_t difficulty       = 0;
static uint64_t current_solution = 0;
static uint64_t tries            = 0;
static uint32_t zero_bytes       = 0;
static uint32_t half_nibble      = 0;

__attribute__((__import_module__("global"), __import_name__("pow_update"))) void
pow_update(uint64_t tries);
__attribute__((__import_module__("global"), __import_name__("pow_finish"))) void
pow_finish(uint64_t tries, uint64_t solution);

static const uint32_t K[64] = {
    0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1,
    0x923f82a4, 0xab1c5ed5, 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3,
    0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174, 0xe49b69c1, 0xefbe4786,
    0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
    0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147,
    0x06ca6351, 0x14292967, 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13,
    0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85, 0xa2bfe8a1, 0xa81a664b,
    0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
    0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a,
    0x5b9cca4f, 0x682e6ff3, 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208,
    0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2,
};

#define ROTR(x, n)   (((x) >> (n)) | ((x) << (32 - (n))))
#define CH(x, y, z)  (((x) & (y)) ^ (~(x) & (z)))
#define MAJ(x, y, z) (((x) & (y)) ^ ((x) & (z)) ^ ((y) & (z)))
#define EP0(x)       (ROTR(x, 2) ^ ROTR(x, 13) ^ ROTR(x, 22))
#define EP1(x)       (ROTR(x, 6) ^ ROTR(x, 11) ^ ROTR(x, 25))
#define SIG0(x)      (ROTR(x, 7) ^ ROTR(x, 18) ^ ((x) >> 3))
#define SIG1(x)      (ROTR(x, 17) ^ ROTR(x, 19) ^ ((x) >> 10))

static inline void g_sha256_hash(uint32_t message_length) {
    uint32_t h[8] = {0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a,
                     0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19};

    const uint64_t bit_length    = (uint64_t)message_length * 8;
    const uint32_t padded_length = ((message_length + 9 + 63) / 64) * 64;

    static uint8_t padded_msg[128];

    for (uint32_t idx = 0; idx < message_length; ++idx) {
        padded_msg[idx] = message_buffer[idx];
    }
    padded_msg[message_length] = 0x80;
    for (uint32_t idx = message_length + 1; idx < padded_length - 8; ++idx) {
        padded_msg[idx] = 0;
    }
    for (int idx = 0; idx < 8; ++idx) {
        padded_msg[padded_length - 1 - idx] =
            (uint8_t)(bit_length >> (8 * idx));
    }

    for (uint32_t chunk = 0; chunk < padded_length; chunk += 64) {
        uint32_t w[64];
        for (int idx = 0; idx < 16; ++idx) {
            w[idx] = ((uint32_t)padded_msg[chunk + idx * 4] << 24) |
                     ((uint32_t)padded_msg[chunk + idx * 4 + 1] << 16) |
                     ((uint32_t)padded_msg[chunk + idx * 4 + 2] << 8) |
                     ((uint32_t)padded_msg[chunk + idx * 4 + 3]);
        }

        for (int idx = 16; idx < 64; ++idx) {
            w[idx] =
                SIG1(w[idx - 2]) + w[idx - 7] + SIG0(w[idx - 15]) + w[idx - 16];
        }

        uint32_t a = h[0], b = h[1], c = h[2], d = h[3];
        uint32_t e = h[4], f = h[5], g = h[6], h_val = h[7];

        for (int idx = 0; idx < 64; ++idx) {
            const uint32_t t1 = h_val + EP1(e) + CH(e, f, g) + K[idx] + w[idx];
            const uint32_t t2 = EP0(a) + MAJ(a, b, c);
            h_val             = g;
            g                 = f;
            f                 = e;
            e                 = d + t1;
            d                 = c;
            c                 = b;
            b                 = a;
            a                 = t1 + t2;
        }

        h[0] += a;
        h[1] += b;
        h[2] += c;
        h[3] += d;
        h[4] += e;
        h[5] += f;
        h[6] += g;
        h[7] += h_val;
    }

    for (int idx = 0; idx < 8; ++idx) {
        hash_output[idx * 4]     = (h[idx] >> 24) & 0xFF;
        hash_output[idx * 4 + 1] = (h[idx] >> 16) & 0xFF;
        hash_output[idx * 4 + 2] = (h[idx] >> 8) & 0xFF;
        hash_output[idx * 4 + 3] = h[idx] & 0xFF;
    }
}

static inline int uint64_to_str(uint64_t number, char *str) {
    if (number == 0) {
        str[0] = '0';
        str[1] = '\0';
        return 1;
    }

    int length    = 0;
    uint64_t temp = number;

    while (temp > 0) {
        ++length;
        temp /= 10;
    }

    str[length] = '\0';

    for (int idx = length - 1; idx >= 0; --idx) {
        str[idx] = '0' + (number % 10);
        number /= 10;
    }

    return length;
}

static inline int g_meets_difficulty(void) {
    for (uint32_t idx = 0; idx < zero_bytes; ++idx) {
        if (hash_output[idx] != 0) {
            return 0;
        }
    }

    if (half_nibble && (hash_output[zero_bytes] & 0xF0) != 0) {
        return 0;
    }

    return 1;
}

__attribute__((export_name("pow_init"))) void
pow_init(uintptr_t nonce_ptr, uint32_t _nonce_length, uint32_t _difficulty) {
    const uint8_t *nonce = (const uint8_t *)nonce_ptr;

    nonce_length =
        (_nonce_length > NONCE_MAX_LEN) ? NONCE_MAX_LEN : _nonce_length;
    difficulty       = _difficulty;
    zero_bytes       = _difficulty / 2;
    half_nibble      = _difficulty % 2;
    current_solution = 0;
    tries            = 0;

    for (uint32_t idx = 0; idx < nonce_length; ++idx) {
        message_buffer[idx] = nonce[idx];
    }
}

__attribute__((export_name("pow_solve_batch"))) int pow_solve_batch(void) {
    if (nonce_length == 0) {
        return 0;
    }

    for (uint32_t batch_idx = 0; batch_idx < BATCH_SIZE; batch_idx++) {
        const int sol_len = uint64_to_str(current_solution, solution_str);
        const int message_length = nonce_length + sol_len;

        if (message_length > MAX_MSG_LEN) {
            return -1;
        }

        for (int idx = 0; idx < sol_len; ++idx) {
            message_buffer[nonce_length + idx] = (uint8_t)solution_str[idx];
        }

        g_sha256_hash(message_length);

        if (g_meets_difficulty()) {
            pow_finish(tries + batch_idx, current_solution);
            return 1;
        }

        ++current_solution;
    }

    tries += BATCH_SIZE;
    pow_update(tries);

    return 0;
}