keccak.c

// ethash: C/C++ implementation of Ethash, the Ethereum Proof of Work algorithm.
// Copyright 2018 Pawel Bylica.
// SPDX-License-Identifier: Apache-2.0

#include "../support/attributes.h"
#include <ethash/keccak.h>

#if !__has_builtin(__builtin_memcpy) && !defined(__GNUC__)
#include <string.h>
#define __builtin_memcpy memcpy
#define __builtin_memset memset
#endif

#if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
#define to_le64(X) __builtin_bswap64(X)
#else
#define to_le64(X) X
#endif

/// Loads 64-bit integer from given memory location as little-endian number.
static inline ALWAYS_INLINE uint64_t load_le(const uint8_t* data)
{
    /* memcpy is the best way of expressing the intention. Every compiler will
       optimize is to single load instruction if the target architecture
       supports unaligned memory access (GCC and clang even in O0).
       This is great trick because we are violating C/C++ memory alignment
       restrictions with no performance penalty. */
    uint64_t word;
    __builtin_memcpy(&word, data, sizeof(word));
    return to_le64(word);
}

/// Rotates the bits of x left by the count value specified by s.
/// The s must be in range <0, 64> exclusively, otherwise the result is undefined.
static inline uint64_t rol(uint64_t x, unsigned s)
{
    return (x << s) | (x >> (64 - s));
}

static const uint64_t round_constants[24] = {  //
    0x0000000000000001, 0x0000000000008082, 0x800000000000808a, 0x8000000080008000,
    0x000000000000808b, 0x0000000080000001, 0x8000000080008081, 0x8000000000008009,
    0x000000000000008a, 0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
    0x000000008000808b, 0x800000000000008b, 0x8000000000008089, 0x8000000000008003,
    0x8000000000008002, 0x8000000000000080, 0x000000000000800a, 0x800000008000000a,
    0x8000000080008081, 0x8000000000008080, 0x0000000080000001, 0x8000000080008008};

/// The Keccak-f[1600] function.
///
/// The implementation of the Keccak-f function with 1600-bit width of the permutation (b).
/// The size of the state is also 1600 bit what gives 25 64-bit words.
///
/// @param state  The state of 25 64-bit words on which the permutation is to be performed.
///
/// The implementation based on:
/// - "simple" implementation by Ronny Van Keer, included in "Reference and optimized code in C",
///   https://keccak.team/archives.html, CC0-1.0 / Public Domain.
static inline ALWAYS_INLINE void keccakf1600_implementation(uint64_t state[25])
{
    uint64_t Aba, Abe, Abi, Abo, Abu;
    uint64_t Aga, Age, Agi, Ago, Agu;
    uint64_t Aka, Ake, Aki, Ako, Aku;
    uint64_t Ama, Ame, Ami, Amo, Amu;
    uint64_t Asa, Ase, Asi, Aso, Asu;

    uint64_t Eba, Ebe, Ebi, Ebo, Ebu;
    uint64_t Ega, Ege, Egi, Ego, Egu;
    uint64_t Eka, Eke, Eki, Eko, Eku;
    uint64_t Ema, Eme, Emi, Emo, Emu;
    uint64_t Esa, Ese, Esi, Eso, Esu;

    uint64_t Ba, Be, Bi, Bo, Bu;

    uint64_t Da, De, Di, Do, Du;

    Aba = state[0];
    Abe = state[1];
    Abi = state[2];
    Abo = state[3];
    Abu = state[4];
    Aga = state[5];
    Age = state[6];
    Agi = state[7];
    Ago = state[8];
    Agu = state[9];
    Aka = state[10];
    Ake = state[11];
    Aki = state[12];
    Ako = state[13];
    Aku = state[14];
    Ama = state[15];
    Ame = state[16];
    Ami = state[17];
    Amo = state[18];
    Amu = state[19];
    Asa = state[20];
    Ase = state[21];
    Asi = state[22];
    Aso = state[23];
    Asu = state[24];

    for (size_t n = 0; n < 24; n += 2)
    {
        // Round (n + 0): Axx -> Exx

        Ba = Aba ^ Aga ^ Aka ^ Ama ^ Asa;
        Be = Abe ^ Age ^ Ake ^ Ame ^ Ase;
        Bi = Abi ^ Agi ^ Aki ^ Ami ^ Asi;
        Bo = Abo ^ Ago ^ Ako ^ Amo ^ Aso;
        Bu = Abu ^ Agu ^ Aku ^ Amu ^ Asu;

        Da = Bu ^ rol(Be, 1);
        De = Ba ^ rol(Bi, 1);
        Di = Be ^ rol(Bo, 1);
        Do = Bi ^ rol(Bu, 1);
        Du = Bo ^ rol(Ba, 1);

        Ba = Aba ^ Da;
        Be = rol(Age ^ De, 44);
        Bi = rol(Aki ^ Di, 43);
        Bo = rol(Amo ^ Do, 21);
        Bu = rol(Asu ^ Du, 14);
        Eba = Ba ^ (~Be & Bi) ^ round_constants[n];
        Ebe = Be ^ (~Bi & Bo);
        Ebi = Bi ^ (~Bo & Bu);
        Ebo = Bo ^ (~Bu & Ba);
        Ebu = Bu ^ (~Ba & Be);

        Ba = rol(Abo ^ Do, 28);
        Be = rol(Agu ^ Du, 20);
        Bi = rol(Aka ^ Da, 3);
        Bo = rol(Ame ^ De, 45);
        Bu = rol(Asi ^ Di, 61);
        Ega = Ba ^ (~Be & Bi);
        Ege = Be ^ (~Bi & Bo);
        Egi = Bi ^ (~Bo & Bu);
        Ego = Bo ^ (~Bu & Ba);
        Egu = Bu ^ (~Ba & Be);

        Ba = rol(Abe ^ De, 1);
        Be = rol(Agi ^ Di, 6);
        Bi = rol(Ako ^ Do, 25);
        Bo = rol(Amu ^ Du, 8);
        Bu = rol(Asa ^ Da, 18);
        Eka = Ba ^ (~Be & Bi);
        Eke = Be ^ (~Bi & Bo);
        Eki = Bi ^ (~Bo & Bu);
        Eko = Bo ^ (~Bu & Ba);
        Eku = Bu ^ (~Ba & Be);

        Ba = rol(Abu ^ Du, 27);
        Be = rol(Aga ^ Da, 36);
        Bi = rol(Ake ^ De, 10);
        Bo = rol(Ami ^ Di, 15);
        Bu = rol(Aso ^ Do, 56);
        Ema = Ba ^ (~Be & Bi);
        Eme = Be ^ (~Bi & Bo);
        Emi = Bi ^ (~Bo & Bu);
        Emo = Bo ^ (~Bu & Ba);
        Emu = Bu ^ (~Ba & Be);

        Ba = rol(Abi ^ Di, 62);
        Be = rol(Ago ^ Do, 55);
        Bi = rol(Aku ^ Du, 39);
        Bo = rol(Ama ^ Da, 41);
        Bu = rol(Ase ^ De, 2);
        Esa = Ba ^ (~Be & Bi);
        Ese = Be ^ (~Bi & Bo);
        Esi = Bi ^ (~Bo & Bu);
        Eso = Bo ^ (~Bu & Ba);
        Esu = Bu ^ (~Ba & Be);


        // Round (n + 1): Exx -> Axx

        Ba = Eba ^ Ega ^ Eka ^ Ema ^ Esa;
        Be = Ebe ^ Ege ^ Eke ^ Eme ^ Ese;
        Bi = Ebi ^ Egi ^ Eki ^ Emi ^ Esi;
        Bo = Ebo ^ Ego ^ Eko ^ Emo ^ Eso;
        Bu = Ebu ^ Egu ^ Eku ^ Emu ^ Esu;

        Da = Bu ^ rol(Be, 1);
        De = Ba ^ rol(Bi, 1);
        Di = Be ^ rol(Bo, 1);
        Do = Bi ^ rol(Bu, 1);
        Du = Bo ^ rol(Ba, 1);

        Ba = Eba ^ Da;
        Be = rol(Ege ^ De, 44);
        Bi = rol(Eki ^ Di, 43);
        Bo = rol(Emo ^ Do, 21);
        Bu = rol(Esu ^ Du, 14);
        Aba = Ba ^ (~Be & Bi) ^ round_constants[n + 1];
        Abe = Be ^ (~Bi & Bo);
        Abi = Bi ^ (~Bo & Bu);
        Abo = Bo ^ (~Bu & Ba);
        Abu = Bu ^ (~Ba & Be);

        Ba = rol(Ebo ^ Do, 28);
        Be = rol(Egu ^ Du, 20);
        Bi = rol(Eka ^ Da, 3);
        Bo = rol(Eme ^ De, 45);
        Bu = rol(Esi ^ Di, 61);
        Aga = Ba ^ (~Be & Bi);
        Age = Be ^ (~Bi & Bo);
        Agi = Bi ^ (~Bo & Bu);
        Ago = Bo ^ (~Bu & Ba);
        Agu = Bu ^ (~Ba & Be);

        Ba = rol(Ebe ^ De, 1);
        Be = rol(Egi ^ Di, 6);
        Bi = rol(Eko ^ Do, 25);
        Bo = rol(Emu ^ Du, 8);
        Bu = rol(Esa ^ Da, 18);
        Aka = Ba ^ (~Be & Bi);
        Ake = Be ^ (~Bi & Bo);
        Aki = Bi ^ (~Bo & Bu);
        Ako = Bo ^ (~Bu & Ba);
        Aku = Bu ^ (~Ba & Be);

        Ba = rol(Ebu ^ Du, 27);
        Be = rol(Ega ^ Da, 36);
        Bi = rol(Eke ^ De, 10);
        Bo = rol(Emi ^ Di, 15);
        Bu = rol(Eso ^ Do, 56);
        Ama = Ba ^ (~Be & Bi);
        Ame = Be ^ (~Bi & Bo);
        Ami = Bi ^ (~Bo & Bu);
        Amo = Bo ^ (~Bu & Ba);
        Amu = Bu ^ (~Ba & Be);

        Ba = rol(Ebi ^ Di, 62);
        Be = rol(Ego ^ Do, 55);
        Bi = rol(Eku ^ Du, 39);
        Bo = rol(Ema ^ Da, 41);
        Bu = rol(Ese ^ De, 2);
        Asa = Ba ^ (~Be & Bi);
        Ase = Be ^ (~Bi & Bo);
        Asi = Bi ^ (~Bo & Bu);
        Aso = Bo ^ (~Bu & Ba);
        Asu = Bu ^ (~Ba & Be);
    }

    state[0] = Aba;
    state[1] = Abe;
    state[2] = Abi;
    state[3] = Abo;
    state[4] = Abu;
    state[5] = Aga;
    state[6] = Age;
    state[7] = Agi;
    state[8] = Ago;
    state[9] = Agu;
    state[10] = Aka;
    state[11] = Ake;
    state[12] = Aki;
    state[13] = Ako;
    state[14] = Aku;
    state[15] = Ama;
    state[16] = Ame;
    state[17] = Ami;
    state[18] = Amo;
    state[19] = Amu;
    state[20] = Asa;
    state[21] = Ase;
    state[22] = Asi;
    state[23] = Aso;
    state[24] = Asu;
}

static void keccakf1600_generic(uint64_t state[25])
{
    keccakf1600_implementation(state);
}

/// The pointer to the best Keccak-f[1600] function implementation,
/// selected during runtime initialization.
static void (*keccakf1600_best)(uint64_t[25]) = keccakf1600_generic;


#if !defined(_MSC_VER) && defined(__x86_64__) && __has_attribute(target)
__attribute__((target("bmi,bmi2"))) static void keccakf1600_bmi(uint64_t state[25])
{
    keccakf1600_implementation(state);
}

__attribute__((constructor)) static void select_keccakf1600_implementation(void)
{
    // Init CPU information.
    // This is needed on macOS because of the bug: https://bugs.llvm.org/show_bug.cgi?id=48459.
    __builtin_cpu_init();

    // Check if both BMI and BMI2 are supported. Some CPUs like Intel E5-2697 v2 incorrectly
    // report BMI2 but not BMI being available.
    if (__builtin_cpu_supports("bmi") && __builtin_cpu_supports("bmi2"))
        keccakf1600_best = keccakf1600_bmi;
}
#endif


static inline ALWAYS_INLINE void keccak(
    uint64_t* out, size_t bits, const uint8_t* data, size_t size)
{
    static const size_t word_size = sizeof(uint64_t);
    const size_t hash_size = bits / 8;
    const size_t block_size = (1600 - bits * 2) / 8;

    size_t i;
    uint64_t* state_iter;
    uint64_t last_word = 0;
    uint8_t* last_word_iter = (uint8_t*)&last_word;

    uint64_t state[25] = {0};

    while (size >= block_size)
    {
        for (i = 0; i < (block_size / word_size); ++i)
        {
            state[i] ^= load_le(data);
            data += word_size;
        }

        keccakf1600_best(state);

        size -= block_size;
    }

    state_iter = state;

    while (size >= word_size)
    {
        *state_iter ^= load_le(data);
        ++state_iter;
        data += word_size;
        size -= word_size;
    }

    while (size > 0)
    {
        *last_word_iter = *data;
        ++last_word_iter;
        ++data;
        --size;
    }

    *last_word_iter = 0x01;
    *state_iter ^= to_le64(last_word);

    state[(block_size / word_size) - 1] ^= 0x8000000000000000;
    keccakf1600_best(state);

    for (i = 0; i < (hash_size / word_size); ++i)
        out[i] = to_le64(state[i]);
}

union ethash_hash256 ethash_keccak256(const uint8_t* data, size_t size)
{
    union ethash_hash256 hash;
    keccak(hash.word64s, 256, data, size);
    return hash;
}

union ethash_hash256 ethash_keccak256_32(const uint8_t data[32])
{
    union ethash_hash256 hash;
    keccak(hash.word64s, 256, data, 32);
    return hash;
}

union ethash_hash512 ethash_keccak512(const uint8_t* data, size_t size)
{
    union ethash_hash512 hash;
    keccak(hash.word64s, 512, data, size);
    return hash;
}

union ethash_hash512 ethash_keccak512_64(const uint8_t data[64])
{
    union ethash_hash512 hash;
    keccak(hash.word64s, 512, data, 64);
    return hash;
}

static inline ALWAYS_INLINE void keccak_init(struct ethash_keccak256_context* ctx, size_t bits)
{
    __builtin_memset((uint8_t*)ctx->state, 0, sizeof ctx->state);
    ctx->hash_size = bits / 8;
    ctx->block_size = (1600 - bits * 2) / 8;

    ctx->last_word = 0;
    ctx->last_word_iter = (uint8_t*)&(ctx->last_word);

    ctx->state_iter = ctx->state;
}

static inline ALWAYS_INLINE void keccak_update(
    struct ethash_keccak256_context* ctx, const uint8_t* data, size_t size)
{
    static const size_t word_size = sizeof(uint64_t);
    size_t i;

    size_t block_size_b = ctx->block_size / word_size;  // block size in bytes
    size_t last_word_unfilled_size =  // calculate unfilled space in last word
        (word_size - (size_t)(ctx->last_word_iter - (uint8_t*)&(ctx->last_word)));
    size_t state_unfilled_size =  // calculate unfilled space in state
        (block_size_b - (size_t)(ctx->state_iter - ctx->state));

    // fill the last word unfilled space with bytes until it's full
    while(last_word_unfilled_size > 0 && size > 0)
    {
        *ctx->last_word_iter = *data;
        ++ctx->last_word_iter;
        ++data;
        --size;
        --last_word_unfilled_size;
    }

    // if the last word is full, move it to state.
    if(ctx->last_word_iter == (uint8_t*)&(ctx->last_word) + word_size)
    {
        *ctx->state_iter ^= to_le64(ctx->last_word);
        ++ctx->state_iter;
        ctx->last_word = 0;
        ctx->last_word_iter = (uint8_t*)&(ctx->last_word);
        --state_unfilled_size;
    }

    // fill the state unfilled space with words until it's full
    while(state_unfilled_size > 0 && size >= word_size)
    {
        *ctx->state_iter ^= load_le(data);
        ++ctx->state_iter;
        data += word_size;
        size -= word_size;
        --state_unfilled_size;
    }

    // if the state is full, calculate keccak and reset the state iterator
    if(ctx->state_iter == ctx->state + (ctx->block_size / word_size))
    {
        keccakf1600_best(ctx->state);
        ctx->state_iter = ctx->state;
    }

    // if there is more data then block size, fill the whole blocks
    if(size >= ctx->block_size) 
    {
        while(size >= ctx->block_size)
        {
            for (i = 0; i < (ctx->block_size / word_size); ++i)
            {
                ctx->state[i] ^= load_le(data);
                data += word_size;
            }

            keccakf1600_best(ctx->state);

            size -= ctx->block_size;
        }

        ctx->state_iter = ctx->state;
    }

    // if there is more data then word size, fill the whole words
    while (size >= word_size)
    {
        *ctx->state_iter ^= load_le(data);
        ++ctx->state_iter;
        data += word_size;
        size -= word_size;
    }

    // if there is still some data put it into last word.
    while (size > 0)
    {
        *ctx->last_word_iter = *data;
        ++ctx->last_word_iter;
        ++data;
        --size;
    }

    // if the last word is full, move it to state.
    if(ctx->last_word_iter == (uint8_t*)&(ctx->last_word) + word_size)
    {
        *ctx->state_iter ^= to_le64(ctx->last_word);
        ++ctx->state_iter;
        ctx->last_word = 0;
        ctx->last_word_iter = (uint8_t*)&(ctx->last_word);
    }

    // if the state is full, calculate keccak and reset the state iterator
    if(ctx->state_iter == ctx->state + (ctx->block_size / word_size))
    {
        keccakf1600_best(ctx->state);
        ctx->state_iter = ctx->state;
    }
}

static inline ALWAYS_INLINE void keccak_final(struct ethash_keccak256_context* ctx, uint64_t* out)
{
    static const size_t word_size = sizeof(uint64_t);
    size_t i;

    *ctx->last_word_iter = 0x01;
    *ctx->state_iter ^= to_le64(ctx->last_word);

    ctx->state[(ctx->block_size / word_size) - 1] ^= 0x8000000000000000;

    keccakf1600_best(ctx->state);

    for (i = 0; i < (ctx->hash_size / word_size); ++i)
        out[i] = to_le64(ctx->state[i]);
}

void ethash_keccak256_init(struct ethash_keccak256_context* ctx)
{
    keccak_init(ctx, 256);
}

void ethash_keccak256_update(struct ethash_keccak256_context* ctx, const uint8_t* data, size_t size)
{
    keccak_update(ctx, data, size);
}

union ethash_hash256 ethash_keccak256_final(struct ethash_keccak256_context* ctx)
{
    union ethash_hash256 hash;
    keccak_final(ctx, hash.word64s);
    return hash;
}

static inline ALWAYS_INLINE void keccak_init_2(struct ethash_keccak256_context* ctx, size_t bits)
{
    __builtin_memset((uint8_t*)ctx->state, 0, sizeof ctx->state);
    ctx->state_iter = ctx->state;

    ctx->hash_size = bits / 8;
    ctx->block_size = (1600 - bits * 2) / 8;
    ctx->last_word = 0;
    ctx->last_word_iter = (uint8_t*)&ctx->last_word;

    __builtin_memset(ctx->buffer, 0, sizeof ctx->buffer);
    ctx->buffer_index = 0;
}

static inline ALWAYS_INLINE void keccak_update_2(
    struct ethash_keccak256_context* ctx, const uint8_t* data, size_t size)
{
    static const size_t word_size = sizeof(uint64_t);

    while(size > 0) 
    {
        size_t empty_space_size = ctx->block_size - ctx->buffer_index;
        size_t data_to_load_size = size >= empty_space_size ? empty_space_size : size;

        __builtin_memcpy(&ctx->buffer[ctx->buffer_index], data, data_to_load_size);
        ctx->buffer_index += data_to_load_size;
        size -= data_to_load_size;
        data += data_to_load_size;

        if(ctx->buffer_index == ctx->block_size)
        {
            size_t i;
            uint8_t* d = &ctx->buffer[0];

            for (i = 0; i < (ctx->block_size / word_size); ++i)
            {
                *ctx->state_iter ^= load_le(d);
                ++ctx->state_iter;
                d += word_size;
            }

            keccakf1600_best(ctx->state);
            ctx->state_iter = ctx->state;
            ctx->buffer_index = 0;
        }
    }
}

static inline ALWAYS_INLINE void keccak_final_2(struct ethash_keccak256_context* ctx, uint64_t* out)
{
    static const size_t word_size = sizeof(uint64_t);
    size_t i;

    if(ctx->buffer_index != 0) 
    {
        uint8_t* d = ctx->buffer;
        for (i = 0; i < (ctx->buffer_index / word_size); ++i)
        {
            *ctx->state_iter ^= load_le(d);
            ++ctx->state_iter;
            d += word_size;
        }

        size_t last_word_size = ctx->buffer_index % word_size;
        d = &ctx->buffer[ctx->buffer_index - last_word_size];
        ctx->last_word_iter = (uint8_t*)&ctx->last_word;

        while (last_word_size > 0)    
        {
            *ctx->last_word_iter = *d;
            ++ctx->last_word_iter;
            ++d;
            --last_word_size;
        }
    }

    *ctx->last_word_iter = 0x01;
    *ctx->state_iter ^= to_le64(ctx->last_word);

    ctx->state[(ctx->block_size / word_size) - 1] ^= 0x8000000000000000;

    keccakf1600_best(ctx->state);

    for (i = 0; i < (ctx->hash_size / word_size); ++i)
        out[i] = to_le64(ctx->state[i]);
}

void ethash_keccak256_init_2(struct ethash_keccak256_context* ctx)
{
    keccak_init_2(ctx, 256);
}

void ethash_keccak256_update_2(struct ethash_keccak256_context* ctx, const uint8_t* data, size_t size)
{
    keccak_update_2(ctx, data, size);
}

union ethash_hash256 ethash_keccak256_final_2(struct ethash_keccak256_context* ctx)
{
    union ethash_hash256 hash;
    keccak_final_2(ctx, hash.word64s);
    return hash;
}