ocb.c

/*------------------------------------------------------------------------
/ OCB Version 3 Reference Code (Optimized C)     Last modified 12-JUN-2013
/-------------------------------------------------------------------------
/ Copyright (c) 2013 Ted Krovetz.
/
/ Permission to use, copy, modify, and/or distribute this software for any
/ purpose with or without fee is hereby granted, provided that the above
/ copyright notice and this permission notice appear in all copies.
/
/ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
/ WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
/ MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
/ ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
/ WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
/ ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
/ OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
/
/ Phillip Rogaway holds patents relevant to OCB. See the following for
/ his patent grant: http://www.cs.ucdavis.edu/~rogaway/ocb/grant.htm
/
/ Special thanks to Keegan McAllister for suggesting several good improvements
/
/ Comments are welcome: Ted Krovetz <ted@krovetz.net> - Dedicated to Laurel K
/------------------------------------------------------------------------- */

/* ----------------------------------------------------------------------- */
/* Usage notes                                                             */
/* ----------------------------------------------------------------------- */

/* - When AE_PENDING is passed as the 'final' parameter of any function,
/    the length parameters must be a multiple of (BPI*16).
/  - When available, SSE or AltiVec registers are used to manipulate data.
/    So, when on machines with these facilities, all pointers passed to
/    any function should be 16-byte aligned.
/  - Plaintext and ciphertext pointers may be equal (ie, plaintext gets
/    encrypted in-place), but no other pair of pointers may be equal.
/  - This code assumes all x86 processors have SSE2 and SSSE3 instructions
/    when compiling under MSVC. If untrue, alter the #define.
/  - This code is tested for C99 and recent versions of GCC and MSVC.      */

/* ----------------------------------------------------------------------- */
/* User configuration options                                              */
/* ----------------------------------------------------------------------- */

/* Set the AES key length to use and length of authentication tag to produce.
/  Setting either to 0 requires the value be set at runtime via ae_init().
/  Some optimizations occur for each when set to a fixed value.            */
#define OCB_KEY_LEN 16 /* 0, 16, 24 or 32. 0 means set in ae_init */
#define OCB_TAG_LEN 16 /* 0 to 16. 0 means set in ae_init         */

/* This implementation has built-in support for multiple AES APIs. Set any
/  one of the following to non-zero to specify which to use.               */
#define USE_OPENSSL_AES 1   /* http://openssl.org                      */
#define USE_REFERENCE_AES 0 /* Internet search: rijndael-alg-fst.c     */
#define USE_AES_NI 0        /* Uses compiler's intrinsics              */

/* During encryption and decryption, various "L values" are required.
/  The L values can be precomputed during initialization (requiring extra
/  space in ae_ctx), generated as needed (slightly slowing encryption and
/  decryption), or some combination of the two. L_TABLE_SZ specifies how many
/  L values to precompute. L_TABLE_SZ must be at least 3. L_TABLE_SZ*16 bytes
/  are used for L values in ae_ctx. Plaintext and ciphertexts shorter than
/  2^L_TABLE_SZ blocks need no L values calculated dynamically.            */
#define L_TABLE_SZ 16

/* Set L_TABLE_SZ_IS_ENOUGH non-zero iff you know that all plaintexts
/  will be shorter than 2^(L_TABLE_SZ+4) bytes in length. This results
/  in better performance.                                                  */
#define L_TABLE_SZ_IS_ENOUGH 1

/* ----------------------------------------------------------------------- */
/* Includes and compiler specific definitions                              */
/* ----------------------------------------------------------------------- */

#include "ae.h"
#include <stdlib.h>
#include <string.h>

/* Define standard sized integers                                          */
#if defined(_MSC_VER) && (_MSC_VER < 1600)
typedef unsigned __int8 uint8_t;
typedef unsigned __int32 uint32_t;
typedef unsigned __int64 uint64_t;
typedef __int64 int64_t;
#else
#include <stdint.h>
#endif

/* Compiler-specific intrinsics and fixes: bswap64, ntz                    */
#if _MSC_VER
#define inline __inline                           /* MSVC doesn't recognize "inline" in C */
#define restrict __restrict                       /* MSVC doesn't recognize "restrict" in C */
#define __SSE2__ (_M_IX86 || _M_AMD64 || _M_X64)  /* Assume SSE2  */
#define __SSSE3__ (_M_IX86 || _M_AMD64 || _M_X64) /* Assume SSSE3 */
#include <intrin.h>
#pragma intrinsic(_byteswap_uint64, _BitScanForward, memcpy)
#define bswap64(x) _byteswap_uint64(x)
static inline unsigned ntz(unsigned x) {
    _BitScanForward(&x, x);
    return x;
}
#elif __GNUC__
#define inline __inline__                   /* No "inline" in GCC ansi C mode */
#define restrict __restrict__               /* No "restrict" in GCC ansi C mode */
#define bswap64(x) __builtin_bswap64(x)     /* Assuming GCC 4.3+ */
#define ntz(x) __builtin_ctz((unsigned)(x)) /* Assuming GCC 3.4+ */
#else /* Assume some C99 features: stdint.h, inline, restrict */
#define bswap32(x)                                                                                 \
    ((((x)&0xff000000u) >> 24) | (((x)&0x00ff0000u) >> 8) | (((x)&0x0000ff00u) << 8) |             \
     (((x)&0x000000ffu) << 24))

static inline uint64_t bswap64(uint64_t x) {
    union {
        uint64_t u64;
        uint32_t u32[2];
    } in, out;
    in.u64 = x;
    out.u32[0] = bswap32(in.u32[1]);
    out.u32[1] = bswap32(in.u32[0]);
    return out.u64;
}

#if (L_TABLE_SZ <= 9) && (L_TABLE_SZ_IS_ENOUGH) /* < 2^13 byte texts */
static inline unsigned ntz(unsigned x) {
    static const unsigned char tz_table[] = {
        0, 2, 3, 2, 4, 2, 3, 2, 5, 2, 3, 2, 4, 2, 3, 2, 6, 2, 3, 2, 4, 2, 3, 2, 5, 2,
        3, 2, 4, 2, 3, 2, 7, 2, 3, 2, 4, 2, 3, 2, 5, 2, 3, 2, 4, 2, 3, 2, 6, 2, 3, 2,
        4, 2, 3, 2, 5, 2, 3, 2, 4, 2, 3, 2, 8, 2, 3, 2, 4, 2, 3, 2, 5, 2, 3, 2, 4, 2,
        3, 2, 6, 2, 3, 2, 4, 2, 3, 2, 5, 2, 3, 2, 4, 2, 3, 2, 7, 2, 3, 2, 4, 2, 3, 2,
        5, 2, 3, 2, 4, 2, 3, 2, 6, 2, 3, 2, 4, 2, 3, 2, 5, 2, 3, 2, 4, 2, 3, 2};
    return tz_table[x / 4];
}
#else                                           /* From http://supertech.csail.mit.edu/papers/debruijn.pdf */
static inline unsigned ntz(unsigned x) {
    static const unsigned char tz_table[32] = {0,  1,  28, 2,  29, 14, 24, 3,  30, 22, 20,
                                               15, 25, 17, 4,  8,  31, 27, 13, 23, 21, 19,
                                               16, 7,  26, 12, 18, 6,  11, 5,  10, 9};
    return tz_table[((uint32_t)((x & -x) * 0x077CB531u)) >> 27];
}
#endif
#endif

/* ----------------------------------------------------------------------- */
/* Define blocks and operations -- Patch if incorrect on your compiler.    */
/* ----------------------------------------------------------------------- */

#if __SSE2__ && !KEYMASTER_CLANG_TEST_BUILD
#include <xmmintrin.h> /* SSE instructions and _mm_malloc */
#include <emmintrin.h> /* SSE2 instructions               */
typedef __m128i block;
#define xor_block(x, y) _mm_xor_si128(x, y)
#define zero_block() _mm_setzero_si128()
#define unequal_blocks(x, y) (_mm_movemask_epi8(_mm_cmpeq_epi8(x, y)) != 0xffff)
#if __SSSE3__ || USE_AES_NI
#include <tmmintrin.h> /* SSSE3 instructions              */
#define swap_if_le(b)                                                                              \
    _mm_shuffle_epi8(b, _mm_set_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15))
#else
static inline block swap_if_le(block b) {
    block a = _mm_shuffle_epi32(b, _MM_SHUFFLE(0, 1, 2, 3));
    a = _mm_shufflehi_epi16(a, _MM_SHUFFLE(2, 3, 0, 1));
    a = _mm_shufflelo_epi16(a, _MM_SHUFFLE(2, 3, 0, 1));
    return _mm_xor_si128(_mm_srli_epi16(a, 8), _mm_slli_epi16(a, 8));
}
#endif
static inline block gen_offset(uint64_t KtopStr[3], unsigned bot) {
    block hi = _mm_load_si128((__m128i*)(KtopStr + 0));  /* hi = B A */
    block lo = _mm_loadu_si128((__m128i*)(KtopStr + 1)); /* lo = C B */
    __m128i lshift = _mm_cvtsi32_si128(bot);
    __m128i rshift = _mm_cvtsi32_si128(64 - bot);
    lo = _mm_xor_si128(_mm_sll_epi64(hi, lshift), _mm_srl_epi64(lo, rshift));
#if __SSSE3__ || USE_AES_NI
    return _mm_shuffle_epi8(lo, _mm_set_epi8(8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7));
#else
    return swap_if_le(_mm_shuffle_epi32(lo, _MM_SHUFFLE(1, 0, 3, 2)));
#endif
}
static inline block double_block(block bl) {
    const __m128i mask = _mm_set_epi32(135, 1, 1, 1);
    __m128i tmp = _mm_srai_epi32(bl, 31);
    tmp = _mm_and_si128(tmp, mask);
    tmp = _mm_shuffle_epi32(tmp, _MM_SHUFFLE(2, 1, 0, 3));
    bl = _mm_slli_epi32(bl, 1);
    return _mm_xor_si128(bl, tmp);
}
#elif __ALTIVEC__
#include <altivec.h>
typedef vector unsigned block;
#define xor_block(x, y) vec_xor(x, y)
#define zero_block() vec_splat_u32(0)
#define unequal_blocks(x, y) vec_any_ne(x, y)
#define swap_if_le(b) (b)
#if __PPC64__
block gen_offset(uint64_t KtopStr[3], unsigned bot) {
    union {
        uint64_t u64[2];
        block bl;
    } rval;
    rval.u64[0] = (KtopStr[0] << bot) | (KtopStr[1] >> (64 - bot));
    rval.u64[1] = (KtopStr[1] << bot) | (KtopStr[2] >> (64 - bot));
    return rval.bl;
}
#else
/* Special handling: Shifts are mod 32, and no 64-bit types */
block gen_offset(uint64_t KtopStr[3], unsigned bot) {
    const vector unsigned k32 = {32, 32, 32, 32};
    vector unsigned hi = *(vector unsigned*)(KtopStr + 0);
    vector unsigned lo = *(vector unsigned*)(KtopStr + 2);
    vector unsigned bot_vec;
    if (bot < 32) {
        lo = vec_sld(hi, lo, 4);
    } else {
        vector unsigned t = vec_sld(hi, lo, 4);
        lo = vec_sld(hi, lo, 8);
        hi = t;
        bot = bot - 32;
    }
    if (bot == 0)
        return hi;
    *(unsigned*)&bot_vec = bot;
    vector unsigned lshift = vec_splat(bot_vec, 0);
    vector unsigned rshift = vec_sub(k32, lshift);
    hi = vec_sl(hi, lshift);
    lo = vec_sr(lo, rshift);
    return vec_xor(hi, lo);
}
#endif
static inline block double_block(block b) {
    const vector unsigned char mask = {135, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
    const vector unsigned char perm = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0};
    const vector unsigned char shift7 = vec_splat_u8(7);
    const vector unsigned char shift1 = vec_splat_u8(1);
    vector unsigned char c = (vector unsigned char)b;
    vector unsigned char t = vec_sra(c, shift7);
    t = vec_and(t, mask);
    t = vec_perm(t, t, perm);
    c = vec_sl(c, shift1);
    return (block)vec_xor(c, t);
}
#elif __ARM_NEON__
#include <arm_neon.h>
typedef int8x16_t block; /* Yay! Endian-neutral reads! */
#define xor_block(x, y) veorq_s8(x, y)
#define zero_block() vdupq_n_s8(0)
static inline int unequal_blocks(block a, block b) {
    int64x2_t t = veorq_s64((int64x2_t)a, (int64x2_t)b);
    return (vgetq_lane_s64(t, 0) | vgetq_lane_s64(t, 1)) != 0;
}
#define swap_if_le(b) (b) /* Using endian-neutral int8x16_t */
/* KtopStr is reg correct by 64 bits, return mem correct */
block gen_offset(uint64_t KtopStr[3], unsigned bot) {
    const union {
        unsigned x;
        unsigned char endian;
    } little = {1};
    const int64x2_t k64 = {-64, -64};
    /* Copy hi and lo into local variables to ensure proper alignment */
    uint64x2_t hi = vld1q_u64(KtopStr + 0); /* hi = A B */
    uint64x2_t lo = vld1q_u64(KtopStr + 1); /* lo = B C */
    int64x2_t ls = vdupq_n_s64(bot);
    int64x2_t rs = vqaddq_s64(k64, ls);
    block rval = (block)veorq_u64(vshlq_u64(hi, ls), vshlq_u64(lo, rs));
    if (little.endian)
        rval = vrev64q_s8(rval);
    return rval;
}
static inline block double_block(block b) {
    const block mask = {135, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
    block tmp = vshrq_n_s8(b, 7);
    tmp = vandq_s8(tmp, mask);
    tmp = vextq_s8(tmp, tmp, 1); /* Rotate high byte to end */
    b = vshlq_n_s8(b, 1);
    return veorq_s8(tmp, b);
}
#else
typedef struct { uint64_t l, r; } block;
static inline block xor_block(block x, block y) {
    x.l ^= y.l;
    x.r ^= y.r;
    return x;
}
static inline block zero_block(void) {
    const block t = {0, 0};
    return t;
}
#define unequal_blocks(x, y) ((((x).l ^ (y).l) | ((x).r ^ (y).r)) != 0)
static inline block swap_if_le(block b) {
    const union {
        unsigned x;
        unsigned char endian;
    } little = {1};
    if (little.endian) {
        block r;
        r.l = bswap64(b.l);
        r.r = bswap64(b.r);
        return r;
    } else
        return b;
}

/* KtopStr is reg correct by 64 bits, return mem correct */
block gen_offset(uint64_t KtopStr[3], unsigned bot) {
    block rval;
    if (bot != 0) {
        rval.l = (KtopStr[0] << bot) | (KtopStr[1] >> (64 - bot));
        rval.r = (KtopStr[1] << bot) | (KtopStr[2] >> (64 - bot));
    } else {
        rval.l = KtopStr[0];
        rval.r = KtopStr[1];
    }
    return swap_if_le(rval);
}

#if __GNUC__ && __arm__
static inline block double_block(block b) {
    __asm__("adds %1,%1,%1\n\t"
            "adcs %H1,%H1,%H1\n\t"
            "adcs %0,%0,%0\n\t"
            "adcs %H0,%H0,%H0\n\t"
            "it cs\n\t"
            "eorcs %1,%1,#135"
            : "+r"(b.l), "+r"(b.r)
            :
            : "cc");
    return b;
}
#else
static inline block double_block(block b) {
    uint64_t t = (uint64_t)((int64_t)b.l >> 63);
    b.l = (b.l + b.l) ^ (b.r >> 63);
    b.r = (b.r + b.r) ^ (t & 135);
    return b;
}
#endif

#endif

/* ----------------------------------------------------------------------- */
/* AES - Code uses OpenSSL API. Other implementations get mapped to it.    */
/* ----------------------------------------------------------------------- */

/*---------------*/
#if USE_OPENSSL_AES
/*---------------*/

#include <openssl/aes.h> /* http://openssl.org/ */

/* How to ECB encrypt an array of blocks, in place                         */
static inline void AES_ecb_encrypt_blks(block* blks, unsigned nblks, AES_KEY* key) {
    while (nblks) {
        --nblks;
        AES_encrypt((unsigned char*)(blks + nblks), (unsigned char*)(blks + nblks), key);
    }
}

static inline void AES_ecb_decrypt_blks(block* blks, unsigned nblks, AES_KEY* key) {
    while (nblks) {
        --nblks;
        AES_decrypt((unsigned char*)(blks + nblks), (unsigned char*)(blks + nblks), key);
    }
}

#define BPI 4 /* Number of blocks in buffer per ECB call */

/*-------------------*/
#elif USE_REFERENCE_AES
/*-------------------*/

#include "rijndael-alg-fst.h" /* Barreto's Public-Domain Code */
#if (OCB_KEY_LEN == 0)
typedef struct {
    uint32_t rd_key[60];
    int rounds;
} AES_KEY;
#define ROUNDS(ctx) ((ctx)->rounds)
#define AES_set_encrypt_key(x, y, z)                                                               \
    do {                                                                                           \
        rijndaelKeySetupEnc((z)->rd_key, x, y);                                                    \
        (z)->rounds = y / 32 + 6;                                                                  \
    } while (0)
#define AES_set_decrypt_key(x, y, z)                                                               \
    do {                                                                                           \
        rijndaelKeySetupDec((z)->rd_key, x, y);                                                    \
        (z)->rounds = y / 32 + 6;                                                                  \
    } while (0)
#else
typedef struct { uint32_t rd_key[OCB_KEY_LEN + 28]; } AES_KEY;
#define ROUNDS(ctx) (6 + OCB_KEY_LEN / 4)
#define AES_set_encrypt_key(x, y, z) rijndaelKeySetupEnc((z)->rd_key, x, y)
#define AES_set_decrypt_key(x, y, z) rijndaelKeySetupDec((z)->rd_key, x, y)
#endif
#define AES_encrypt(x, y, z) rijndaelEncrypt((z)->rd_key, ROUNDS(z), x, y)
#define AES_decrypt(x, y, z) rijndaelDecrypt((z)->rd_key, ROUNDS(z), x, y)

static void AES_ecb_encrypt_blks(block* blks, unsigned nblks, AES_KEY* key) {
    while (nblks) {
        --nblks;
        AES_encrypt((unsigned char*)(blks + nblks), (unsigned char*)(blks + nblks), key);
    }
}

void AES_ecb_decrypt_blks(block* blks, unsigned nblks, AES_KEY* key) {
    while (nblks) {
        --nblks;
        AES_decrypt((unsigned char*)(blks + nblks), (unsigned char*)(blks + nblks), key);
    }
}

#define BPI 4 /* Number of blocks in buffer per ECB call */

/*----------*/
#elif USE_AES_NI
/*----------*/

#include <wmmintrin.h>

#if (OCB_KEY_LEN == 0)
typedef struct {
    __m128i rd_key[15];
    int rounds;
} AES_KEY;
#define ROUNDS(ctx) ((ctx)->rounds)
#else
typedef struct { __m128i rd_key[7 + OCB_KEY_LEN / 4]; } AES_KEY;
#define ROUNDS(ctx) (6 + OCB_KEY_LEN / 4)
#endif

#define EXPAND_ASSIST(v1, v2, v3, v4, shuff_const, aes_const)                                      \
    v2 = _mm_aeskeygenassist_si128(v4, aes_const);                                                 \
    v3 = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(v3), _mm_castsi128_ps(v1), 16));         \
    v1 = _mm_xor_si128(v1, v3);                                                                    \
    v3 = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(v3), _mm_castsi128_ps(v1), 140));        \
    v1 = _mm_xor_si128(v1, v3);                                                                    \
    v2 = _mm_shuffle_epi32(v2, shuff_const);                                                       \
    v1 = _mm_xor_si128(v1, v2)

#define EXPAND192_STEP(idx, aes_const)                                                             \
    EXPAND_ASSIST(x0, x1, x2, x3, 85, aes_const);                                                  \
    x3 = _mm_xor_si128(x3, _mm_slli_si128(x3, 4));                                                 \
    x3 = _mm_xor_si128(x3, _mm_shuffle_epi32(x0, 255));                                            \
    kp[idx] = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(tmp), _mm_castsi128_ps(x0), 68));   \
    kp[idx + 1] =                                                                                  \
        _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(x0), _mm_castsi128_ps(x3), 78));          \
    EXPAND_ASSIST(x0, x1, x2, x3, 85, (aes_const * 2));                                            \
    x3 = _mm_xor_si128(x3, _mm_slli_si128(x3, 4));                                                 \
    x3 = _mm_xor_si128(x3, _mm_shuffle_epi32(x0, 255));                                            \
    kp[idx + 2] = x0;                                                                              \
    tmp = x3

static void AES_128_Key_Expansion(const unsigned char* userkey, void* key) {
    __m128i x0, x1, x2;
    __m128i* kp = (__m128i*)key;
    kp[0] = x0 = _mm_loadu_si128((__m128i*)userkey);
    x2 = _mm_setzero_si128();
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 1);
    kp[1] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 2);
    kp[2] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 4);
    kp[3] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 8);
    kp[4] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 16);
    kp[5] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 32);
    kp[6] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 64);
    kp[7] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 128);
    kp[8] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 27);
    kp[9] = x0;
    EXPAND_ASSIST(x0, x1, x2, x0, 255, 54);
    kp[10] = x0;
}

static void AES_192_Key_Expansion(const unsigned char* userkey, void* key) {
    __m128i x0, x1, x2, x3, tmp, *kp = (__m128i*)key;
    kp[0] = x0 = _mm_loadu_si128((__m128i*)userkey);
    tmp = x3 = _mm_loadu_si128((__m128i*)(userkey + 16));
    x2 = _mm_setzero_si128();
    EXPAND192_STEP(1, 1);
    EXPAND192_STEP(4, 4);
    EXPAND192_STEP(7, 16);
    EXPAND192_STEP(10, 64);
}

static void AES_256_Key_Expansion(const unsigned char* userkey, void* key) {
    __m128i x0, x1, x2, x3, *kp = (__m128i*)key;
    kp[0] = x0 = _mm_loadu_si128((__m128i*)userkey);
    kp[1] = x3 = _mm_loadu_si128((__m128i*)(userkey + 16));
    x2 = _mm_setzero_si128();
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 1);
    kp[2] = x0;
    EXPAND_ASSIST(x3, x1, x2, x0, 170, 1);
    kp[3] = x3;
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 2);
    kp[4] = x0;
    EXPAND_ASSIST(x3, x1, x2, x0, 170, 2);
    kp[5] = x3;
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 4);
    kp[6] = x0;
    EXPAND_ASSIST(x3, x1, x2, x0, 170, 4);
    kp[7] = x3;
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 8);
    kp[8] = x0;
    EXPAND_ASSIST(x3, x1, x2, x0, 170, 8);
    kp[9] = x3;
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 16);
    kp[10] = x0;
    EXPAND_ASSIST(x3, x1, x2, x0, 170, 16);
    kp[11] = x3;
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 32);
    kp[12] = x0;
    EXPAND_ASSIST(x3, x1, x2, x0, 170, 32);
    kp[13] = x3;
    EXPAND_ASSIST(x0, x1, x2, x3, 255, 64);
    kp[14] = x0;
}

static int AES_set_encrypt_key(const unsigned char* userKey, const int bits, AES_KEY* key) {
    if (bits == 128) {
        AES_128_Key_Expansion(userKey, key);
    } else if (bits == 192) {
        AES_192_Key_Expansion(userKey, key);
    } else if (bits == 256) {
        AES_256_Key_Expansion(userKey, key);
    }
#if (OCB_KEY_LEN == 0)
    key->rounds = 6 + bits / 32;
#endif
    return 0;
}

static void AES_set_decrypt_key_fast(AES_KEY* dkey, const AES_KEY* ekey) {
    int j = 0;
    int i = ROUNDS(ekey);
#if (OCB_KEY_LEN == 0)
    dkey->rounds = i;
#endif
    dkey->rd_key[i--] = ekey->rd_key[j++];
    while (i)
        dkey->rd_key[i--] = _mm_aesimc_si128(ekey->rd_key[j++]);
    dkey->rd_key[i] = ekey->rd_key[j];
}

static int AES_set_decrypt_key(const unsigned char* userKey, const int bits, AES_KEY* key) {
    AES_KEY temp_key;
    AES_set_encrypt_key(userKey, bits, &temp_key);
    AES_set_decrypt_key_fast(key, &temp_key);
    return 0;
}

static inline void AES_encrypt(const unsigned char* in, unsigned char* out, const AES_KEY* key) {
    int j, rnds = ROUNDS(key);
    const __m128i* sched = ((__m128i*)(key->rd_key));
    __m128i tmp = _mm_load_si128((__m128i*)in);
    tmp = _mm_xor_si128(tmp, sched[0]);
    for (j = 1; j < rnds; j++)
        tmp = _mm_aesenc_si128(tmp, sched[j]);
    tmp = _mm_aesenclast_si128(tmp, sched[j]);
    _mm_store_si128((__m128i*)out, tmp);
}

static inline void AES_decrypt(const unsigned char* in, unsigned char* out, const AES_KEY* key) {
    int j, rnds = ROUNDS(key);
    const __m128i* sched = ((__m128i*)(key->rd_key));
    __m128i tmp = _mm_load_si128((__m128i*)in);
    tmp = _mm_xor_si128(tmp, sched[0]);
    for (j = 1; j < rnds; j++)
        tmp = _mm_aesdec_si128(tmp, sched[j]);
    tmp = _mm_aesdeclast_si128(tmp, sched[j]);
    _mm_store_si128((__m128i*)out, tmp);
}

static inline void AES_ecb_encrypt_blks(block* blks, unsigned nblks, AES_KEY* key) {
    unsigned i, j, rnds = ROUNDS(key);
    const __m128i* sched = ((__m128i*)(key->rd_key));
    for (i = 0; i < nblks; ++i)
        blks[i] = _mm_xor_si128(blks[i], sched[0]);
    for (j = 1; j < rnds; ++j)
        for (i = 0; i < nblks; ++i)
            blks[i] = _mm_aesenc_si128(blks[i], sched[j]);
    for (i = 0; i < nblks; ++i)
        blks[i] = _mm_aesenclast_si128(blks[i], sched[j]);
}

static inline void AES_ecb_decrypt_blks(block* blks, unsigned nblks, AES_KEY* key) {
    unsigned i, j, rnds = ROUNDS(key);
    const __m128i* sched = ((__m128i*)(key->rd_key));
    for (i = 0; i < nblks; ++i)
        blks[i] = _mm_xor_si128(blks[i], sched[0]);
    for (j = 1; j < rnds; ++j)
        for (i = 0; i < nblks; ++i)
            blks[i] = _mm_aesdec_si128(blks[i], sched[j]);
    for (i = 0; i < nblks; ++i)
        blks[i] = _mm_aesdeclast_si128(blks[i], sched[j]);
}

#define BPI 8 /* Number of blocks in buffer per ECB call   */
/* Set to 4 for Westmere, 8 for Sandy Bridge */

#endif

/* ----------------------------------------------------------------------- */
/* Define OCB context structure.                                           */
/* ----------------------------------------------------------------------- */

/*------------------------------------------------------------------------
/ Each item in the OCB context is stored either "memory correct" or
/ "register correct". On big-endian machines, this is identical. On
/ little-endian machines, one must choose whether the byte-string
/ is in the correct order when it resides in memory or in registers.
/ It must be register correct whenever it is to be manipulated
/ arithmetically, but must be memory correct whenever it interacts
/ with the plaintext or ciphertext.
/------------------------------------------------------------------------- */

struct _ae_ctx {
    block offset;        /* Memory correct               */
    block checksum;      /* Memory correct               */
    block Lstar;         /* Memory correct               */
    block Ldollar;       /* Memory correct               */
    block L[L_TABLE_SZ]; /* Memory correct               */
    block ad_checksum;   /* Memory correct               */
    block ad_offset;     /* Memory correct               */
    block cached_Top;    /* Memory correct               */
    uint64_t KtopStr[3]; /* Register correct, each item  */
    uint32_t ad_blocks_processed;
    uint32_t blocks_processed;
    AES_KEY decrypt_key;
    AES_KEY encrypt_key;
#if (OCB_TAG_LEN == 0)
    unsigned tag_len;
#endif
};

/* ----------------------------------------------------------------------- */
/* L table lookup (or on-the-fly generation)                               */
/* ----------------------------------------------------------------------- */

#if L_TABLE_SZ_IS_ENOUGH
#define getL(_ctx, _tz) ((_ctx)->L[_tz])
#else
static block getL(const ae_ctx* ctx, unsigned tz) {
    if (tz < L_TABLE_SZ)
        return ctx->L[tz];
    else {
        unsigned i;
        /* Bring L[MAX] into registers, make it register correct */
        block rval = swap_if_le(ctx->L[L_TABLE_SZ - 1]);
        rval = double_block(rval);
        for (i = L_TABLE_SZ; i < tz; i++)
            rval = double_block(rval);
        return swap_if_le(rval); /* To memory correct */
    }
}
#endif

/* ----------------------------------------------------------------------- */
/* Public functions                                                        */
/* ----------------------------------------------------------------------- */

/* 32-bit SSE2 and Altivec systems need to be forced to allocate memory
   on 16-byte alignments. (I believe all major 64-bit systems do already.) */

ae_ctx* ae_allocate(void* misc) {
    void* p;
    (void)misc; /* misc unused in this implementation */
#if (__SSE2__ && !_M_X64 && !_M_AMD64 && !__amd64__)
    p = _mm_malloc(sizeof(ae_ctx), 16);
#elif(__ALTIVEC__ && !__PPC64__)
    if (posix_memalign(&p, 16, sizeof(ae_ctx)) != 0)
        p = NULL;
#else
    p = malloc(sizeof(ae_ctx));
#endif
    return (ae_ctx*)p;
}

void ae_free(ae_ctx* ctx) {
#if (__SSE2__ && !_M_X64 && !_M_AMD64 && !__amd64__)
    _mm_free(ctx);
#else
    free(ctx);
#endif
}

/* ----------------------------------------------------------------------- */

int ae_clear(ae_ctx* ctx) /* Zero ae_ctx and undo initialization          */
{
    memset(ctx, 0, sizeof(ae_ctx));
    return AE_SUCCESS;
}

int ae_ctx_sizeof(void) {
    return (int)sizeof(ae_ctx);
}

/* ----------------------------------------------------------------------- */

int ae_init(ae_ctx* ctx, const void* key, int key_len, int nonce_len, int tag_len) {
    unsigned i;
    block tmp_blk;

    if (nonce_len != 12)
        return AE_NOT_SUPPORTED;

/* Initialize encryption & decryption keys */
#if (OCB_KEY_LEN > 0)
    key_len = OCB_KEY_LEN;
#endif
    AES_set_encrypt_key((unsigned char*)key, key_len * 8, &ctx->encrypt_key);
#if USE_AES_NI
    AES_set_decrypt_key_fast(&ctx->decrypt_key, &ctx->encrypt_key);
#else
    AES_set_decrypt_key((unsigned char*)key, (int)(key_len * 8), &ctx->decrypt_key);
#endif

    /* Zero things that need zeroing */
    ctx->cached_Top = ctx->ad_checksum = zero_block();
    ctx->ad_blocks_processed = 0;

    /* Compute key-dependent values */
    AES_encrypt((unsigned char*)&ctx->cached_Top, (unsigned char*)&ctx->Lstar, &ctx->encrypt_key);
    tmp_blk = swap_if_le(ctx->Lstar);
    tmp_blk = double_block(tmp_blk);
    ctx->Ldollar = swap_if_le(tmp_blk);
    tmp_blk = double_block(tmp_blk);
    ctx->L[0] = swap_if_le(tmp_blk);
    for (i = 1; i < L_TABLE_SZ; i++) {
        tmp_blk = double_block(tmp_blk);
        ctx->L[i] = swap_if_le(tmp_blk);
    }

#if (OCB_TAG_LEN == 0)
    ctx->tag_len = tag_len;
#else
    (void)tag_len; /* Suppress var not used error */
#endif

    return AE_SUCCESS;
}

/* ----------------------------------------------------------------------- */

static block gen_offset_from_nonce(ae_ctx* ctx, const void* nonce) {
    const union {
        unsigned x;
        unsigned char endian;
    } little = {1};
    union {
        uint32_t u32[4];
        uint8_t u8[16];
        block bl;
    } tmp;
    unsigned idx;

/* Replace cached nonce Top if needed */
#if (OCB_TAG_LEN > 0)
    if (little.endian)
        tmp.u32[0] = 0x01000000 + ((OCB_TAG_LEN * 8 % 128) << 1);
    else
        tmp.u32[0] = 0x00000001 + ((OCB_TAG_LEN * 8 % 128) << 25);
#else
    if (little.endian)
        tmp.u32[0] = 0x01000000 + ((ctx->tag_len * 8 % 128) << 1);
    else
        tmp.u32[0] = 0x00000001 + ((ctx->tag_len * 8 % 128) << 25);
#endif
    tmp.u32[1] = ((uint32_t*)nonce)[0];
    tmp.u32[2] = ((uint32_t*)nonce)[1];
    tmp.u32[3] = ((uint32_t*)nonce)[2];
    idx = (unsigned)(tmp.u8[15] & 0x3f);           /* Get low 6 bits of nonce  */
    tmp.u8[15] = tmp.u8[15] & 0xc0;                /* Zero low 6 bits of nonce */
    if (unequal_blocks(tmp.bl, ctx->cached_Top)) { /* Cached?       */
        ctx->cached_Top = tmp.bl;                  /* Update cache, KtopStr    */
        AES_encrypt(tmp.u8, (unsigned char*)&ctx->KtopStr, &ctx->encrypt_key);
        if (little.endian) { /* Make Register Correct    */
            ctx->KtopStr[0] = bswap64(ctx->KtopStr[0]);
            ctx->KtopStr[1] = bswap64(ctx->KtopStr[1]);
        }
        ctx->KtopStr[2] = ctx->KtopStr[0] ^ (ctx->KtopStr[0] << 8) ^ (ctx->KtopStr[1] >> 56);
    }
    return gen_offset(ctx->KtopStr, idx);
}

static void process_ad(ae_ctx* ctx, const void* ad, int ad_len, int final) {
    union {
        uint32_t u32[4];
        uint8_t u8[16];
        block bl;
    } tmp;
    block ad_offset, ad_checksum;
    const block* adp = (block*)ad;
    unsigned i, k, tz, remaining;

    ad_offset = ctx->ad_offset;
    ad_checksum = ctx->ad_checksum;
    i = ad_len / (BPI * 16);
    if (i) {
        unsigned ad_block_num = ctx->ad_blocks_processed;
        do {
            block ta[BPI], oa[BPI];
            ad_block_num += BPI;
            tz = ntz(ad_block_num);
            oa[0] = xor_block(ad_offset, ctx->L[0]);
            ta[0] = xor_block(oa[0], adp[0]);
            oa[1] = xor_block(oa[0], ctx->L[1]);
            ta[1] = xor_block(oa[1], adp[1]);
            oa[2] = xor_block(ad_offset, ctx->L[1]);
            ta[2] = xor_block(oa[2], adp[2]);
#if BPI == 4
            ad_offset = xor_block(oa[2], getL(ctx, tz));
            ta[3] = xor_block(ad_offset, adp[3]);
#elif BPI == 8
            oa[3] = xor_block(oa[2], ctx->L[2]);
            ta[3] = xor_block(oa[3], adp[3]);
            oa[4] = xor_block(oa[1], ctx->L[2]);
            ta[4] = xor_block(oa[4], adp[4]);
            oa[5] = xor_block(oa[0], ctx->L[2]);
            ta[5] = xor_block(oa[5], adp[5]);
            oa[6] = xor_block(ad_offset, ctx->L[2]);
            ta[6] = xor_block(oa[6], adp[6]);
            ad_offset = xor_block(oa[6], getL(ctx, tz));
            ta[7] = xor_block(ad_offset, adp[7]);
#endif
            AES_ecb_encrypt_blks(ta, BPI, &ctx->encrypt_key);
            ad_checksum = xor_block(ad_checksum, ta[0]);
            ad_checksum = xor_block(ad_checksum, ta[1]);
            ad_checksum = xor_block(ad_checksum, ta[2]);
            ad_checksum = xor_block(ad_checksum, ta[3]);
#if (BPI == 8)
            ad_checksum = xor_block(ad_checksum, ta[4]);
            ad_checksum = xor_block(ad_checksum, ta[5]);
            ad_checksum = xor_block(ad_checksum, ta[6]);
            ad_checksum = xor_block(ad_checksum, ta[7]);
#endif
            adp += BPI;
        } while (--i);
        ctx->ad_blocks_processed = ad_block_num;
        ctx->ad_offset = ad_offset;
        ctx->ad_checksum = ad_checksum;
    }

    if (final) {
        block ta[BPI];

        /* Process remaining associated data, compute its tag contribution */
        remaining = ((unsigned)ad_len) % (BPI * 16);
        if (remaining) {
            k = 0;
#if (BPI == 8)
            if (remaining >= 64) {
                tmp.bl = xor_block(ad_offset, ctx->L[0]);
                ta[0] = xor_block(tmp.bl, adp[0]);
                tmp.bl = xor_block(tmp.bl, ctx->L[1]);
                ta[1] = xor_block(tmp.bl, adp[1]);
                ad_offset = xor_block(ad_offset, ctx->L[1]);
                ta[2] = xor_block(ad_offset, adp[2]);
                ad_offset = xor_block(ad_offset, ctx->L[2]);
                ta[3] = xor_block(ad_offset, adp[3]);
                remaining -= 64;
                k = 4;
            }
#endif
            if (remaining >= 32) {
                ad_offset = xor_block(ad_offset, ctx->L[0]);
                ta[k] = xor_block(ad_offset, adp[k]);
                ad_offset = xor_block(ad_offset, getL(ctx, ntz(k + 2)));
                ta[k + 1] = xor_block(ad_offset, adp[k + 1]);
                remaining -= 32;
                k += 2;
            }
            if (remaining >= 16) {
                ad_offset = xor_block(ad_offset, ctx->L[0]);
                ta[k] = xor_block(ad_offset, adp[k]);
                remaining = remaining - 16;
                ++k;
            }
            if (remaining) {
                ad_offset = xor_block(ad_offset, ctx->Lstar);
                tmp.bl = zero_block();
                memcpy(tmp.u8, adp + k, remaining);
                tmp.u8[remaining] = (unsigned char)0x80u;
                ta[k] = xor_block(ad_offset, tmp.bl);
                ++k;
            }
            AES_ecb_encrypt_blks(ta, k, &ctx->encrypt_key);
            switch (k) {
#if (BPI == 8)
            case 8:
                ad_checksum = xor_block(ad_checksum, ta[7]);
            case 7:
                ad_checksum = xor_block(ad_checksum, ta[6]);
            case 6:
                ad_checksum = xor_block(ad_checksum, ta[5]);
            case 5:
                ad_checksum = xor_block(ad_checksum, ta[4]);
#endif
            case 4:
                ad_checksum = xor_block(ad_checksum, ta[3]);
            case 3:
                ad_checksum = xor_block(ad_checksum, ta[2]);
            case 2:
                ad_checksum = xor_block(ad_checksum, ta[1]);
            case 1:
                ad_checksum = xor_block(ad_checksum, ta[0]);
            }
            ctx->ad_checksum = ad_checksum;
        }
    }
}

/* ----------------------------------------------------------------------- */

int ae_encrypt(ae_ctx* ctx, const void* nonce, const void* pt, int pt_len, const void* ad,
               int ad_len, void* ct, void* tag, int final) {
    union {
        uint32_t u32[4];
        uint8_t u8[16];
        block bl;
    } tmp;
    block offset, checksum;
    unsigned i, k;
    block* ctp = (block*)ct;
    const block* ptp = (block*)pt;

    /* Non-null nonce means start of new message, init per-message values */
    if (nonce) {
        ctx->offset = gen_offset_from_nonce(ctx, nonce);
        ctx->ad_offset = ctx->checksum = zero_block();
        ctx->ad_blocks_processed = ctx->blocks_processed = 0;
        if (ad_len >= 0)
            ctx->ad_checksum = zero_block();
    }

    /* Process associated data */
    if (ad_len > 0)
        process_ad(ctx, ad, ad_len, final);

    /* Encrypt plaintext data BPI blocks at a time */
    offset = ctx->offset;
    checksum = ctx->checksum;
    i = pt_len / (BPI * 16);
    if (i) {
        block oa[BPI];
        unsigned block_num = ctx->blocks_processed;
        oa[BPI - 1] = offset;
        do {
            block ta[BPI];
            block_num += BPI;
            oa[0] = xor_block(oa[BPI - 1], ctx->L[0]);
            ta[0] = xor_block(oa[0], ptp[0]);
            checksum = xor_block(checksum, ptp[0]);
            oa[1] = xor_block(oa[0], ctx->L[1]);
            ta[1] = xor_block(oa[1], ptp[1]);
            checksum = xor_block(checksum, ptp[1]);
            oa[2] = xor_block(oa[1], ctx->L[0]);
            ta[2] = xor_block(oa[2], ptp[2]);
            checksum = xor_block(checksum, ptp[2]);
#if BPI == 4
            oa[3] = xor_block(oa[2], getL(ctx, ntz(block_num)));
            ta[3] = xor_block(oa[3], ptp[3]);
            checksum = xor_block(checksum, ptp[3]);
#elif BPI == 8
            oa[3] = xor_block(oa[2], ctx->L[2]);
            ta[3] = xor_block(oa[3], ptp[3]);
            checksum = xor_block(checksum, ptp[3]);
            oa[4] = xor_block(oa[1], ctx->L[2]);
            ta[4] = xor_block(oa[4], ptp[4]);
            checksum = xor_block(checksum, ptp[4]);
            oa[5] = xor_block(oa[0], ctx->L[2]);
            ta[5] = xor_block(oa[5], ptp[5]);
            checksum = xor_block(checksum, ptp[5]);
            oa[6] = xor_block(oa[7], ctx->L[2]);
            ta[6] = xor_block(oa[6], ptp[6]);
            checksum = xor_block(checksum, ptp[6]);
            oa[7] = xor_block(oa[6], getL(ctx, ntz(block_num)));
            ta[7] = xor_block(oa[7], ptp[7]);
            checksum = xor_block(checksum, ptp[7]);
#endif
            AES_ecb_encrypt_blks(ta, BPI, &ctx->encrypt_key);
            ctp[0] = xor_block(ta[0], oa[0]);
            ctp[1] = xor_block(ta[1], oa[1]);
            ctp[2] = xor_block(ta[2], oa[2]);
            ctp[3] = xor_block(ta[3], oa[3]);
#if (BPI == 8)
            ctp[4] = xor_block(ta[4], oa[4]);
            ctp[5] = xor_block(ta[5], oa[5]);
            ctp[6] = xor_block(ta[6], oa[6]);
            ctp[7] = xor_block(ta[7], oa[7]);
#endif
            ptp += BPI;
            ctp += BPI;
        } while (--i);
        ctx->offset = offset = oa[BPI - 1];
        ctx->blocks_processed = block_num;
        ctx->checksum = checksum;
    }

    if (final) {
        block ta[BPI + 1], oa[BPI];

        /* Process remaining plaintext and compute its tag contribution    */
        unsigned remaining = ((unsigned)pt_len) % (BPI * 16);
        k = 0; /* How many blocks in ta[] need ECBing */
        if (remaining) {
#if (BPI == 8)
            if (remaining >= 64) {
                oa[0] = xor_block(offset, ctx->L[0]);
                ta[0] = xor_block(oa[0], ptp[0]);
                checksum = xor_block(checksum, ptp[0]);
                oa[1] = xor_block(oa[0], ctx->L[1]);
                ta[1] = xor_block(oa[1], ptp[1]);
                checksum = xor_block(checksum, ptp[1]);
                oa[2] = xor_block(oa[1], ctx->L[0]);
                ta[2] = xor_block(oa[2], ptp[2]);
                checksum = xor_block(checksum, ptp[2]);
                offset = oa[3] = xor_block(oa[2], ctx->L[2]);
                ta[3] = xor_block(offset, ptp[3]);
                checksum = xor_block(checksum, ptp[3]);
                remaining -= 64;
                k = 4;
            }
#endif
            if (remaining >= 32) {
                oa[k] = xor_block(offset, ctx->L[0]);
                ta[k] = xor_block(oa[k], ptp[k]);
                checksum = xor_block(checksum, ptp[k]);
                offset = oa[k + 1] = xor_block(oa[k], ctx->L[1]);
                ta[k + 1] = xor_block(offset, ptp[k + 1]);
                checksum = xor_block(checksum, ptp[k + 1]);
                remaining -= 32;
                k += 2;
            }
            if (remaining >= 16) {
                offset = oa[k] = xor_block(offset, ctx->L[0]);
                ta[k] = xor_block(offset, ptp[k]);
                checksum = xor_block(checksum, ptp[k]);
                remaining -= 16;
                ++k;
            }
            if (remaining) {
                tmp.bl = zero_block();
                memcpy(tmp.u8, ptp + k, remaining);
                tmp.u8[remaining] = (unsigned char)0x80u;
                checksum = xor_block(checksum, tmp.bl);
                ta[k] = offset = xor_block(offset, ctx->Lstar);
                ++k;
            }
        }
        offset = xor_block(offset, ctx->Ldollar); /* Part of tag gen */
        ta[k] = xor_block(offset, checksum);      /* Part of tag gen */
        AES_ecb_encrypt_blks(ta, k + 1, &ctx->encrypt_key);
        offset = xor_block(ta[k], ctx->ad_checksum); /* Part of tag gen */
        if (remaining) {
            --k;
            tmp.bl = xor_block(tmp.bl, ta[k]);
            memcpy(ctp + k, tmp.u8, remaining);
        }
        switch (k) {
#if (BPI == 8)
        case 7:
            ctp[6] = xor_block(ta[6], oa[6]);
        case 6:
            ctp[5] = xor_block(ta[5], oa[5]);
        case 5:
            ctp[4] = xor_block(ta[4], oa[4]);
        case 4:
            ctp[3] = xor_block(ta[3], oa[3]);
#endif
        case 3:
            ctp[2] = xor_block(ta[2], oa[2]);
        case 2:
            ctp[1] = xor_block(ta[1], oa[1]);
        case 1:
            ctp[0] = xor_block(ta[0], oa[0]);
        }

        /* Tag is placed at the correct location
         */
        if (tag) {
#if (OCB_TAG_LEN == 16)
            *(block*)tag = offset;
#elif(OCB_TAG_LEN > 0)
            memcpy((char*)tag, &offset, OCB_TAG_LEN);
#else
            memcpy((char*)tag, &offset, ctx->tag_len);
#endif
        } else {
#if (OCB_TAG_LEN > 0)
            memcpy((char*)ct + pt_len, &offset, OCB_TAG_LEN);
            pt_len += OCB_TAG_LEN;
#else
            memcpy((char*)ct + pt_len, &offset, ctx->tag_len);
            pt_len += ctx->tag_len;
#endif
        }
    }
    return (int)pt_len;
}

/* ----------------------------------------------------------------------- */

/* Compare two regions of memory, taking a constant amount of time for a
   given buffer size -- under certain assumptions about the compiler
   and machine, of course.

   Use this to avoid timing side-channel attacks.

   Returns 0 for memory regions with equal contents; non-zero otherwise. */
static int constant_time_memcmp(const void* av, const void* bv, size_t n) {
    const uint8_t* a = (const uint8_t*)av;
    const uint8_t* b = (const uint8_t*)bv;
    uint8_t result = 0;
    size_t i;

    for (i = 0; i < n; i++) {
        result |= *a ^ *b;
        a++;
        b++;
    }

    return (int)result;
}

int ae_decrypt(ae_ctx* ctx, const void* nonce, const void* ct, int ct_len, const void* ad,
               int ad_len, void* pt, const void* tag, int final) {
    union {
        uint32_t u32[4];
        uint8_t u8[16];
        block bl;
    } tmp;
    block offset, checksum;
    unsigned i, k;
    block* ctp = (block*)ct;
    block* ptp = (block*)pt;

    /* Reduce ct_len tag bundled in ct */
    if ((final) && (!tag))
#if (OCB_TAG_LEN > 0)
        ct_len -= OCB_TAG_LEN;
#else
        ct_len -= ctx->tag_len;
#endif

    /* Non-null nonce means start of new message, init per-message values */
    if (nonce) {
        ctx->offset = gen_offset_from_nonce(ctx, nonce);
        ctx->ad_offset = ctx->checksum = zero_block();
        ctx->ad_blocks_processed = ctx->blocks_processed = 0;
        if (ad_len >= 0)
            ctx->ad_checksum = zero_block();
    }

    /* Process associated data */
    if (ad_len > 0)
        process_ad(ctx, ad, ad_len, final);

    /* Encrypt plaintext data BPI blocks at a time */
    offset = ctx->offset;
    checksum = ctx->checksum;
    i = ct_len / (BPI * 16);
    if (i) {
        block oa[BPI];
        unsigned block_num = ctx->blocks_processed;
        oa[BPI - 1] = offset;
        do {
            block ta[BPI];
            block_num += BPI;
            oa[0] = xor_block(oa[BPI - 1], ctx->L[0]);
            ta[0] = xor_block(oa[0], ctp[0]);
            oa[1] = xor_block(oa[0], ctx->L[1]);
            ta[1] = xor_block(oa[1], ctp[1]);
            oa[2] = xor_block(oa[1], ctx->L[0]);
            ta[2] = xor_block(oa[2], ctp[2]);
#if BPI == 4
            oa[3] = xor_block(oa[2], getL(ctx, ntz(block_num)));
            ta[3] = xor_block(oa[3], ctp[3]);
#elif BPI == 8
            oa[3] = xor_block(oa[2], ctx->L[2]);
            ta[3] = xor_block(oa[3], ctp[3]);
            oa[4] = xor_block(oa[1], ctx->L[2]);
            ta[4] = xor_block(oa[4], ctp[4]);
            oa[5] = xor_block(oa[0], ctx->L[2]);
            ta[5] = xor_block(oa[5], ctp[5]);
            oa[6] = xor_block(oa[7], ctx->L[2]);
            ta[6] = xor_block(oa[6], ctp[6]);
            oa[7] = xor_block(oa[6], getL(ctx, ntz(block_num)));
            ta[7] = xor_block(oa[7], ctp[7]);
#endif
            AES_ecb_decrypt_blks(ta, BPI, &ctx->decrypt_key);
            ptp[0] = xor_block(ta[0], oa[0]);
            checksum = xor_block(checksum, ptp[0]);
            ptp[1] = xor_block(ta[1], oa[1]);
            checksum = xor_block(checksum, ptp[1]);
            ptp[2] = xor_block(ta[2], oa[2]);
            checksum = xor_block(checksum, ptp[2]);
            ptp[3] = xor_block(ta[3], oa[3]);
            checksum = xor_block(checksum, ptp[3]);
#if (BPI == 8)
            ptp[4] = xor_block(ta[4], oa[4]);
            checksum = xor_block(checksum, ptp[4]);
            ptp[5] = xor_block(ta[5], oa[5]);
            checksum = xor_block(checksum, ptp[5]);
            ptp[6] = xor_block(ta[6], oa[6]);
            checksum = xor_block(checksum, ptp[6]);
            ptp[7] = xor_block(ta[7], oa[7]);
            checksum = xor_block(checksum, ptp[7]);
#endif
            ptp += BPI;
            ctp += BPI;
        } while (--i);
        ctx->offset = offset = oa[BPI - 1];
        ctx->blocks_processed = block_num;
        ctx->checksum = checksum;
    }

    if (final) {
        block ta[BPI + 1], oa[BPI];

        /* Process remaining plaintext and compute its tag contribution    */
        unsigned remaining = ((unsigned)ct_len) % (BPI * 16);
        k = 0; /* How many blocks in ta[] need ECBing */
        if (remaining) {
#if (BPI == 8)
            if (remaining >= 64) {
                oa[0] = xor_block(offset, ctx->L[0]);
                ta[0] = xor_block(oa[0], ctp[0]);
                oa[1] = xor_block(oa[0], ctx->L[1]);
                ta[1] = xor_block(oa[1], ctp[1]);
                oa[2] = xor_block(oa[1], ctx->L[0]);
                ta[2] = xor_block(oa[2], ctp[2]);
                offset = oa[3] = xor_block(oa[2], ctx->L[2]);
                ta[3] = xor_block(offset, ctp[3]);
                remaining -= 64;
                k = 4;
            }
#endif
            if (remaining >= 32) {
                oa[k] = xor_block(offset, ctx->L[0]);
                ta[k] = xor_block(oa[k], ctp[k]);
                offset = oa[k + 1] = xor_block(oa[k], ctx->L[1]);
                ta[k + 1] = xor_block(offset, ctp[k + 1]);
                remaining -= 32;
                k += 2;
            }
            if (remaining >= 16) {
                offset = oa[k] = xor_block(offset, ctx->L[0]);
                ta[k] = xor_block(offset, ctp[k]);
                remaining -= 16;
                ++k;
            }
            if (remaining) {
                block pad;
                offset = xor_block(offset, ctx->Lstar);
                AES_encrypt((unsigned char*)&offset, tmp.u8, &ctx->encrypt_key);
                pad = tmp.bl;
                memcpy(tmp.u8, ctp + k, remaining);
                tmp.bl = xor_block(tmp.bl, pad);
                tmp.u8[remaining] = (unsigned char)0x80u;
                memcpy(ptp + k, tmp.u8, remaining);
                checksum = xor_block(checksum, tmp.bl);
            }
        }
        AES_ecb_decrypt_blks(ta, k, &ctx->decrypt_key);
        switch (k) {
#if (BPI == 8)
        case 7:
            ptp[6] = xor_block(ta[6], oa[6]);
            checksum = xor_block(checksum, ptp[6]);
        case 6:
            ptp[5] = xor_block(ta[5], oa[5]);
            checksum = xor_block(checksum, ptp[5]);
        case 5:
            ptp[4] = xor_block(ta[4], oa[4]);
            checksum = xor_block(checksum, ptp[4]);
        case 4:
            ptp[3] = xor_block(ta[3], oa[3]);
            checksum = xor_block(checksum, ptp[3]);
#endif
        case 3:
            ptp[2] = xor_block(ta[2], oa[2]);
            checksum = xor_block(checksum, ptp[2]);
        case 2:
            ptp[1] = xor_block(ta[1], oa[1]);
            checksum = xor_block(checksum, ptp[1]);
        case 1:
            ptp[0] = xor_block(ta[0], oa[0]);
            checksum = xor_block(checksum, ptp[0]);
        }

        /* Calculate expected tag */
        offset = xor_block(offset, ctx->Ldollar);
        tmp.bl = xor_block(offset, checksum);
        AES_encrypt(tmp.u8, tmp.u8, &ctx->encrypt_key);
        tmp.bl = xor_block(tmp.bl, ctx->ad_checksum); /* Full tag */

        /* Compare with proposed tag, change ct_len if invalid */
        if ((OCB_TAG_LEN == 16) && tag) {
            if (unequal_blocks(tmp.bl, *(block*)tag))
                ct_len = AE_INVALID;
        } else {
#if (OCB_TAG_LEN > 0)
            int len = OCB_TAG_LEN;
#else
            int len = ctx->tag_len;
#endif
            if (tag) {
                if (constant_time_memcmp(tag, tmp.u8, len) != 0)
                    ct_len = AE_INVALID;
            } else {
                if (constant_time_memcmp((char*)ct + ct_len, tmp.u8, len) != 0)
                    ct_len = AE_INVALID;
            }
        }
    }
    return ct_len;
}

/* ----------------------------------------------------------------------- */
/* Simple test program                                                     */
/* ----------------------------------------------------------------------- */

#if 0

#include <stdio.h>
#include <time.h>

#if __GNUC__
#define ALIGN(n) __attribute__((aligned(n)))
#elif _MSC_VER
#define ALIGN(n) __declspec(align(n))
#else /* Not GNU/Microsoft: delete alignment uses.     */
#define ALIGN(n)
#endif

static void pbuf(void *p, unsigned len, const void *s)
{
    unsigned i;
    if (s)
        printf("%s", (char *)s);
    for (i = 0; i < len; i++)
        printf("%02X", (unsigned)(((unsigned char *)p)[i]));
    printf("\n");
}

static void vectors(ae_ctx *ctx, int len)
{
    ALIGN(16) char pt[128];
    ALIGN(16) char ct[144];
    ALIGN(16) char nonce[] = {0,1,2,3,4,5,6,7,8,9,10,11};
    int i;
    for (i=0; i < 128; i++) pt[i] = i;
    i = ae_encrypt(ctx,nonce,pt,len,pt,len,ct,NULL,AE_FINALIZE);
    printf("P=%d,A=%d: ",len,len); pbuf(ct, i, NULL);
    i = ae_encrypt(ctx,nonce,pt,0,pt,len,ct,NULL,AE_FINALIZE);
    printf("P=%d,A=%d: ",0,len); pbuf(ct, i, NULL);
    i = ae_encrypt(ctx,nonce,pt,len,pt,0,ct,NULL,AE_FINALIZE);
    printf("P=%d,A=%d: ",len,0); pbuf(ct, i, NULL);
}

void validate()
{
    ALIGN(16) char pt[1024];
    ALIGN(16) char ct[1024];
    ALIGN(16) char tag[16];
    ALIGN(16) char nonce[12] = {0,};
    ALIGN(16) char key[32] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};
    ae_ctx ctx;
    char *val_buf, *next;
    int i, len;

    val_buf = (char *)malloc(22400 + 16);
    next = val_buf = (char *)(((size_t)val_buf + 16) & ~((size_t)15));

    if (0) {
		ae_init(&ctx, key, 16, 12, 16);
		/* pbuf(&ctx, sizeof(ctx), "CTX: "); */
		vectors(&ctx,0);
		vectors(&ctx,8);
		vectors(&ctx,16);
		vectors(&ctx,24);
		vectors(&ctx,32);
		vectors(&ctx,40);
    }

    memset(key,0,32);
    memset(pt,0,128);
    ae_init(&ctx, key, OCB_KEY_LEN, 12, OCB_TAG_LEN);

    /* RFC Vector test */
    for (i = 0; i < 128; i++) {
        int first = ((i/3)/(BPI*16))*(BPI*16);
        int second = first;
        int third = i - (first + second);

        nonce[11] = i;

        if (0) {
            ae_encrypt(&ctx,nonce,pt,i,pt,i,ct,NULL,AE_FINALIZE);
            memcpy(next,ct,(size_t)i+OCB_TAG_LEN);
            next = next+i+OCB_TAG_LEN;

            ae_encrypt(&ctx,nonce,pt,i,pt,0,ct,NULL,AE_FINALIZE);
            memcpy(next,ct,(size_t)i+OCB_TAG_LEN);
            next = next+i+OCB_TAG_LEN;

            ae_encrypt(&ctx,nonce,pt,0,pt,i,ct,NULL,AE_FINALIZE);
            memcpy(next,ct,OCB_TAG_LEN);
            next = next+OCB_TAG_LEN;
        } else {
            ae_encrypt(&ctx,nonce,pt,first,pt,first,ct,NULL,AE_PENDING);
            ae_encrypt(&ctx,NULL,pt+first,second,pt+first,second,ct+first,NULL,AE_PENDING);
            ae_encrypt(&ctx,NULL,pt+first+second,third,pt+first+second,third,ct+first+second,NULL,AE_FINALIZE);
            memcpy(next,ct,(size_t)i+OCB_TAG_LEN);
            next = next+i+OCB_TAG_LEN;

            ae_encrypt(&ctx,nonce,pt,first,pt,0,ct,NULL,AE_PENDING);
            ae_encrypt(&ctx,NULL,pt+first,second,pt,0,ct+first,NULL,AE_PENDING);
            ae_encrypt(&ctx,NULL,pt+first+second,third,pt,0,ct+first+second,NULL,AE_FINALIZE);
            memcpy(next,ct,(size_t)i+OCB_TAG_LEN);
            next = next+i+OCB_TAG_LEN;

            ae_encrypt(&ctx,nonce,pt,0,pt,first,ct,NULL,AE_PENDING);
            ae_encrypt(&ctx,NULL,pt,0,pt+first,second,ct,NULL,AE_PENDING);
            ae_encrypt(&ctx,NULL,pt,0,pt+first+second,third,ct,NULL,AE_FINALIZE);
            memcpy(next,ct,OCB_TAG_LEN);
            next = next+OCB_TAG_LEN;
        }

    }
    nonce[11] = 0;
    ae_encrypt(&ctx,nonce,NULL,0,val_buf,next-val_buf,ct,tag,AE_FINALIZE);
    pbuf(tag,OCB_TAG_LEN,0);


    /* Encrypt/Decrypt test */
    for (i = 0; i < 128; i++) {
        int first = ((i/3)/(BPI*16))*(BPI*16);
        int second = first;
        int third = i - (first + second);

        nonce[11] = i%128;

        if (1) {
            len = ae_encrypt(&ctx,nonce,val_buf,i,val_buf,i,ct,tag,AE_FINALIZE);
            len = ae_encrypt(&ctx,nonce,val_buf,i,val_buf,-1,ct,tag,AE_FINALIZE);
            len = ae_decrypt(&ctx,nonce,ct,len,val_buf,-1,pt,tag,AE_FINALIZE);
            if (len == -1) { printf("Authentication error: %d\n", i); return; }
            if (len != i) { printf("Length error: %d\n", i); return; }
            if (memcmp(val_buf,pt,i)) { printf("Decrypt error: %d\n", i); return; }
        } else {
            len = ae_encrypt(&ctx,nonce,val_buf,i,val_buf,i,ct,NULL,AE_FINALIZE);
            ae_decrypt(&ctx,nonce,ct,first,val_buf,first,pt,NULL,AE_PENDING);
            ae_decrypt(&ctx,NULL,ct+first,second,val_buf+first,second,pt+first,NULL,AE_PENDING);
            len = ae_decrypt(&ctx,NULL,ct+first+second,len-(first+second),val_buf+first+second,third,pt+first+second,NULL,AE_FINALIZE);
            if (len == -1) { printf("Authentication error: %d\n", i); return; }
            if (memcmp(val_buf,pt,i)) { printf("Decrypt error: %d\n", i); return; }
        }

    }
    printf("Decrypt: PASS\n");
}

int main()
{
    validate();
    return 0;
}
#endif

#if USE_AES_NI
char infoString[] = "OCB3 (AES-NI)";
#elif USE_REFERENCE_AES
char infoString[] = "OCB3 (Reference)";
#elif USE_OPENSSL_AES
char infoString[] = "OCB3 (OpenSSL)";
#endif