/* poly1305.c

 *

 * Copyright (C) 2006-2026 wolfSSL Inc.

 *

 * This file is part of wolfSSL.

 *

 * wolfSSL is free software; you can redistribute it and/or modify

 * it under the terms of the GNU General Public License as published by

 * the Free Software Foundation; either version 3 of the License, or

 * (at your option) any later version.

 *

 * wolfSSL is distributed in the hope that it will be useful,

 * but WITHOUT ANY WARRANTY; without even the implied warranty of

 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

 * GNU General Public License for more details.

 *

 * You should have received a copy of the GNU General Public License

 * along with this program; if not, write to the Free Software

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1335, USA

 */

/*

DESCRIPTION

This library contains implementation for the Poly1305 authenticator.

Based off the public domain implementations by Andrew Moon

and Daniel J. Bernstein

*/

/*

 * WOLFSSL_W64_WRAPPER Uses wrappers around word64 types for a system that does

 *                     not have word64 available. As expected it reduces

 *                     performance. Benchmarks collected July 2024 show

 *                     303.004 MiB/s with and 1874.194 MiB/s without.

 */

#include <wolfssl/wolfcrypt/libwolfssl_sources.h>

/*

 * Poly1305 Build Options:

 *

 * HAVE_POLY1305:            Enable Poly1305 authenticator          default: off

 * POLY130564:               Use 64-bit Poly1305 implementation    default: auto

 * USE_INTEL_POLY1305_SPEEDUP: Intel AVX/AVX2 Poly1305 accel      default: off

 */

#ifdef HAVE_POLY1305

#include <wolfssl/wolfcrypt/poly1305.h>

#include <wolfssl/wolfcrypt/cpuid.h>

#ifdef NO_INLINE

    #include <wolfssl/wolfcrypt/misc.h>

#else

    #define WOLFSSL_MISC_INCLUDED

    #include <wolfcrypt/src/misc.c>

#endif

#ifdef CHACHA_AEAD_TEST

    #include <stdio.h>

#endif

#ifdef _MSC_VER

    /* 4127 warning constant while(1)  */

    #pragma warning(disable: 4127)

#endif

#ifdef USE_INTEL_POLY1305_SPEEDUP

    #include <emmintrin.h>

    #include <immintrin.h>

    #if defined(__GNUC__) && ((__GNUC__ < 4) || \

                              (__GNUC__ == 4 && __GNUC_MINOR__ <= 8))

        #undef  NO_AVX2_SUPPORT

        #define NO_AVX2_SUPPORT

    #endif

    #if defined(__clang__) && ((__clang_major__ < 3) || \

                               (__clang_major__ == 3 && __clang_minor__ <= 5))

        #define NO_AVX2_SUPPORT

    #elif defined(__clang__) && defined(NO_AVX2_SUPPORT)

        #undef NO_AVX2_SUPPORT

    #endif

    #if defined(_MSC_VER) && (_MSC_VER <= 1900)

        #undef  NO_AVX2_SUPPORT

        #define NO_AVX2_SUPPORT

    #endif

    #define HAVE_INTEL_AVX1

    #ifndef NO_AVX2_SUPPORT

        #define HAVE_INTEL_AVX2

    #endif

    /* 8-way path; the struct carries r^5..r^8 only when this is on. */

    #if defined(WOLFSSL_POLY1305_AVX512) && !defined(NO_AVX512_SUPPORT)

        #define HAVE_INTEL_AVX512

    #endif

#endif

#ifdef USE_INTEL_POLY1305_SPEEDUP

static cpuid_flags_t intel_flags = WC_CPUID_INITIALIZER;

/* The fused ChaCha20-Poly1305 kernel drives the 4-way path; when it is active

 * this flag pins Update/Final to the 4-way path so the state layout matches. */

#ifdef WOLFSSL_CHACHA20_POLY1305_FUSED

    #define POLY1305_FORCE_AVX2(ctx) ((ctx)->forceAvx2 != 0)

#else

    #define POLY1305_FORCE_AVX2(ctx) 0

#endif

/* The fused IFMA stitch drives the scalar path; when it is active this flag

 * pins Update/Final to the scalar (poly1305_blocks_avx / _final_avx) path so

 * the state layout matches. */

#ifdef WOLFSSL_CHACHA20_POLY1305_FUSED_IFMA

    #define POLY1305_FORCE_SCALAR(ctx) ((ctx)->forceScalar != 0)

#else

    #define POLY1305_FORCE_SCALAR(ctx) 0

#endif

#ifdef HAVE_INTEL_AVX512

/* Whether to use the 8-way (512-bit) Poly1305 over the 4-way AVX2 path.

 *

 * The zmm path does eight 16-byte blocks per iteration and is the fastest

 * option where 512-bit code runs at full clock: AMD Zen 4/5 (no AVX-512

 * license) and Intel Ice Lake and later.  On Intel Skylake-SP / Cascade

 * Lake-class parts, sustained 512-bit instructions trip the AVX-512 frequency

 * license and downclock the core; the penalty is milder here than for the

 * FP-heavy ChaCha, but the 256-bit AVX2 path can still be the safer default on

 * those parts.  As with the ChaCha gate, VAES presence is used as a

 * generational proxy: the throttling parts predate VAES, whereas every

 * microarchitecture that runs 512-bit without penalty implements it.  A missing

 * VAES only costs a little throughput (fall back to AVX2), never correctness.

 *

 * Override the heuristic with:

 *   WOLFSSL_POLY1305_AVX512_ALWAYS - use zmm whenever AVX-512 is present

 *   WOLFSSL_POLY1305_AVX512_NEVER  - never use zmm (always AVX2 or below)

 */

static WC_INLINE int poly1305_use_avx512(cpuid_flags_t flags)

{

#if defined(WOLFSSL_POLY1305_AVX512_NEVER)

    (void)flags;

    return 0;

#elif defined(WOLFSSL_POLY1305_AVX512_ALWAYS)

    return IS_INTEL_AVX512(flags) != 0;

#else

    return (IS_INTEL_AVX512(flags) != 0) && (IS_INTEL_VAES(flags) != 0);

#endif

}

/* The radix-2^44 IFMA path is preferred on Intel cores with AVX-512 IFMA

 * (Ice Lake and later), where it beats the 26-bit vpmuludq 8-way.  On AMD

 * Zen 4/5 - which also implement IFMA - the vpmuludq 8-way is measurably faster

 * (stronger vector-integer throughput), so IFMA is gated to an Intel vendor.

 * Define WOLFSSL_POLY1305_IFMA_ALWAYS to use it on any IFMA CPU, or

 * WOLFSSL_POLY1305_NO_IFMA to disable it. */

static WC_INLINE int poly1305_use_ifma(cpuid_flags_t flags)

{

#if defined(WOLFSSL_POLY1305_NO_IFMA)

    (void)flags;

    return 0;

#elif defined(WOLFSSL_POLY1305_IFMA_ALWAYS)

    return IS_INTEL_AVX512_IFMA(flags) != 0;

#else

    return (IS_INTEL_AVX512_IFMA(flags) != 0) && (IS_CPU_INTEL(flags) != 0);

#endif

}

#endif /* HAVE_INTEL_AVX512 */

#endif

#if defined(USE_INTEL_POLY1305_SPEEDUP) || defined(POLY130564)

    #if defined(__WATCOMC__)

        #error "POLY130564 || USE_INTEL_POLY1305_SPEEDUP Watcom not supported"

    #elif defined(_MSC_VER)

        #define POLY1305_NOINLINE __declspec(noinline)

    #elif defined(__GNUC__)

        #define POLY1305_NOINLINE __attribute__((noinline))

    #else

        #define POLY1305_NOINLINE

    #endif

    #if defined(_MSC_VER) && !(__WATCOMC__)

        #include <intrin.h>

        typedef struct poly1305_word128 {

            word64 lo;

            word64 hi;

        } poly1305_word128;

        #define MUL(out, x, y) out.lo = _umul128((x), (y), &out.hi)

        #define ADD(out, in) { word64 t = out.lo; out.lo += in.lo; \

                               out.hi += (out.lo < t) + in.hi; }

        #define ADDLO(out, in) { word64 t = out.lo; out.lo += in; \

                                 out.hi += (out.lo < t); }

        #define SHR(in, shift) (__shiftright128(in.lo, in.hi, (shift)))

        #define LO(in) (in.lo)

    #elif defined(__GNUC__)

        #if defined(__SIZEOF_INT128__)

            PEDANTIC_EXTENSION typedef unsigned __int128 poly1305_word128;

        #else

            typedef unsigned poly1305_word128 __attribute__((mode(TI)));

        #endif

        #define MUL(out, x, y) out = ((poly1305_word128)(x) * (y))

        #define ADD(out, in) (out) += (in)

        #define ADDLO(out, in) (out) += (in)

        #define SHR(in, shift) (word64)((in) >> (shift))

        #define LO(in) (word64)(in)

    #endif

#endif

#ifdef USE_INTEL_POLY1305_SPEEDUP

#ifdef __cplusplus

    extern "C" {

#endif

#ifdef HAVE_INTEL_AVX1

/* Process one block (16 bytes) of data.

 *

 * ctx  Poly1305 context.

 * m    One block of message data.

 */

WOLFSSL_LOCAL void poly1305_block_avx(Poly1305* ctx, const unsigned char *m);

/* Process multiple blocks (n * 16 bytes) of data.

 *

 * ctx    Poly1305 context.

 * m      Blocks of message data.

 * bytes  The number of bytes to process.

 */

WOLFSSL_LOCAL void poly1305_blocks_avx(Poly1305* ctx, const unsigned char* m,

                                size_t bytes);

/* Set the key to use when processing data.

 * Initialize the context.

 *

 * ctx  Poly1305 context.

 * key  The key data (16 bytes).

 */

WOLFSSL_LOCAL void poly1305_setkey_avx(Poly1305* ctx, const byte* key);

/* Calculate the final result - authentication data.

 * Zeros out the private data in the context.

 *

 * ctx  Poly1305 context.

 * mac  Buffer to hold 16 bytes.

 */

WOLFSSL_LOCAL void poly1305_final_avx(Poly1305* ctx, byte* mac);

#endif

#ifdef HAVE_INTEL_AVX2

/* Process multiple blocks (n * 16 bytes) of data.

 *

 * ctx    Poly1305 context.

 * m      Blocks of message data.

 * bytes  The number of bytes to process.

 */

WOLFSSL_LOCAL void poly1305_blocks_avx2(Poly1305* ctx, const unsigned char* m,

                                 size_t bytes);

/* Calculate R^1, R^2, R^3 and R^4 and store them in the context.

 *

 * ctx    Poly1305 context.

 */

WOLFSSL_LOCAL void poly1305_calc_powers_avx2(Poly1305* ctx);

/* Set the key to use when processing data.

 * Initialize the context.

 * Calls AVX set key function as final function calls AVX code.

 *

 * ctx  Poly1305 context.

 * key  The key data (16 bytes).

 */

WOLFSSL_LOCAL void poly1305_setkey_avx2(Poly1305* ctx, const byte* key);

/* Calculate the final result - authentication data.

 * Zeros out the private data in the context.

 * Calls AVX final function to quickly process last blocks.

 *

 * ctx  Poly1305 context.

 * mac  Buffer to hold 16 bytes - authentication data.

 */

WOLFSSL_LOCAL void poly1305_final_avx2(Poly1305* ctx, byte* mac);

#endif

#ifdef HAVE_INTEL_AVX512

/* Process multiple 128-byte (8 block) groups of data eight lanes wide. */

WOLFSSL_LOCAL void poly1305_blocks_avx512(Poly1305* ctx,

                                 const unsigned char* m, size_t bytes);

/* Calculate R^1 .. R^8 and store them in the context. */

WOLFSSL_LOCAL void poly1305_calc_powers_avx512(Poly1305* ctx);

/* Calculate the final result - authentication data.  Collapses the eight

 * lanes and finishes any leftover with the AVX final function. */

WOLFSSL_LOCAL void poly1305_final_avx512(Poly1305* ctx, byte* mac);

/* AVX-512 IFMA (radix 2^44) 8-way variants - used when the CPU has IFMA. */

WOLFSSL_LOCAL void poly1305_blocks_avx512ifma(Poly1305* ctx,

                                 const unsigned char* m, size_t bytes);

WOLFSSL_LOCAL void poly1305_calc_powers_avx512ifma(Poly1305* ctx);

WOLFSSL_LOCAL void poly1305_setkey_avx512ifma(Poly1305* ctx, const byte* key);

WOLFSSL_LOCAL void poly1305_final_avx512ifma(Poly1305* ctx, byte* mac);

#endif

#ifdef __cplusplus

    }  /* extern "C" */

#endif

#elif defined(POLY130564)

#if (!defined(WOLFSSL_ARMASM) || defined(WOLFSSL_ARM_POLY1305_NEED_C)) && \

    !defined(WOLFSSL_RISCV_ASM)

    static word64 U8TO64(const byte* p)

    {

        return

            (((word64)(p[0] & 0xff)      ) |

             ((word64)(p[1] & 0xff) <<  8) |

             ((word64)(p[2] & 0xff) << 16) |

             ((word64)(p[3] & 0xff) << 24) |

             ((word64)(p[4] & 0xff) << 32) |

             ((word64)(p[5] & 0xff) << 40) |

             ((word64)(p[6] & 0xff) << 48) |

             ((word64)(p[7] & 0xff) << 56));

    }

    static void U64TO8(byte* p, word64 v) {

        p[0] = (byte)v;

        p[1] = (byte)(v >>  8);

        p[2] = (byte)(v >> 16);

        p[3] = (byte)(v >> 24);

        p[4] = (byte)(v >> 32);

        p[5] = (byte)(v >> 40);

        p[6] = (byte)(v >> 48);

        p[7] = (byte)(v >> 56);

    }

#endif/* !WOLFSSL_ARMASM && !WOLFSSL_RISCV_ASM */

/* if not 64 bit then use 32 bit (also the aarch64 assembly's software

 * fallback when there is no 128-bit integer for POLY130564) */

#elif !defined(WOLFSSL_ARMASM) || defined(WOLFSSL_ARM_POLY1305_NEED_C)

    static word32 U8TO32(const byte *p)

    {

        return

            (((word32)(p[0] & 0xff)      ) |

             ((word32)(p[1] & 0xff) <<  8) |

             ((word32)(p[2] & 0xff) << 16) |

             ((word32)(p[3] & 0xff) << 24));

    }

    static void U32TO8(byte *p, word32 v) {

        p[0] = (byte)((v      ) & 0xff);

        p[1] = (byte)((v >>  8) & 0xff);

        p[2] = (byte)((v >> 16) & 0xff);

        p[3] = (byte)((v >> 24) & 0xff);

    }

#endif

/* convert 32-bit unsigned to little endian 64 bit type as byte array */

static WC_INLINE void u32tole64(const word32 inLe32, byte outLe64[8])

{

#ifndef WOLFSSL_X86_64_BUILD

    outLe64[0] = (byte)(inLe32  & 0x000000FF);

    outLe64[1] = (byte)((inLe32 & 0x0000FF00) >> 8);

    outLe64[2] = (byte)((inLe32 & 0x00FF0000) >> 16);

    outLe64[3] = (byte)((inLe32 & 0xFF000000) >> 24);

    outLe64[4] = 0;

    outLe64[5] = 0;

    outLe64[6] = 0;

    outLe64[7] = 0;

#else

    *(word64*)outLe64 = inLe32;

#endif

}

/* The portable C Poly1305 implementation is needed for builds without an

 * assembly back end, and as the runtime fallback on aarch64 when NEON is

 * unavailable. */

#if (!defined(USE_INTEL_POLY1305_SPEEDUP) && !defined(WOLFSSL_ARMASM) && \

     !defined(WOLFSSL_RISCV_ASM)) || defined(WOLFSSL_ARM_POLY1305_NEED_C)

    #define WOLFSSL_POLY1305_C

#endif

#ifdef WOLFSSL_POLY1305_C

/* On aarch64 the assembly owns the r/h/pad context fields, so the C state lives

 * in the appended c_* fields; every other build uses the standard fields. */

#ifdef WOLFSSL_ARM_POLY1305_NEED_C

    #define POLY1305_CTX_R    c_r

    #define POLY1305_CTX_H    c_h

    #define POLY1305_CTX_PAD  c_pad

#else

    #define POLY1305_CTX_R    r

    #define POLY1305_CTX_H    h

    #define POLY1305_CTX_PAD  pad

#endif

static void poly1305_c_setkey(Poly1305* ctx, const byte* key)

{

#if defined(POLY130564)

    word64 t0,t1;

    /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */

    t0 = U8TO64(key + 0);

    t1 = U8TO64(key + 8);

    ctx->POLY1305_CTX_R[0] = ( t0                    ) & 0xffc0fffffff;

    ctx->POLY1305_CTX_R[1] = ((t0 >> 44) | (t1 << 20)) & 0xfffffc0ffff;

    ctx->POLY1305_CTX_R[2] = ((t1 >> 24)             ) & 0x00ffffffc0f;

    ctx->POLY1305_CTX_H[0] = 0;

    ctx->POLY1305_CTX_H[1] = 0;

    ctx->POLY1305_CTX_H[2] = 0;

    ctx->POLY1305_CTX_PAD[0] = U8TO64(key + 16);

    ctx->POLY1305_CTX_PAD[1] = U8TO64(key + 24);

#else

    /* r &= 0xffffffc0ffffffc0ffffffc0fffffff */

    ctx->POLY1305_CTX_R[0] = (U8TO32(key +  0)     ) & 0x3ffffff;

    ctx->POLY1305_CTX_R[1] = (U8TO32(key +  3) >> 2) & 0x3ffff03;

    ctx->POLY1305_CTX_R[2] = (U8TO32(key +  6) >> 4) & 0x3ffc0ff;

    ctx->POLY1305_CTX_R[3] = (U8TO32(key +  9) >> 6) & 0x3f03fff;

    ctx->POLY1305_CTX_R[4] = (U8TO32(key + 12) >> 8) & 0x00fffff;

    ctx->POLY1305_CTX_H[0] = 0;

    ctx->POLY1305_CTX_H[1] = 0;

    ctx->POLY1305_CTX_H[2] = 0;

    ctx->POLY1305_CTX_H[3] = 0;

    ctx->POLY1305_CTX_H[4] = 0;

    ctx->POLY1305_CTX_PAD[0] = U8TO32(key + 16);

    ctx->POLY1305_CTX_PAD[1] = U8TO32(key + 20);

    ctx->POLY1305_CTX_PAD[2] = U8TO32(key + 24);

    ctx->POLY1305_CTX_PAD[3] = U8TO32(key + 28);

#endif

    ctx->leftover = 0;

    ctx->finished = 0;

}

static int poly1305_c_blocks(Poly1305* ctx, const byte* m, size_t bytes)

{

#if defined(POLY130564)

    const word64 hibit = (ctx->finished) ? 0 : ((word64)1 << 40); /* 1 << 128 */

    word64 r0,r1,r2;

    word64 s1,s2;

    word64 h0,h1,h2;

    word64 c;

    poly1305_word128 d0,d1,d2,d;

    r0 = ctx->POLY1305_CTX_R[0];

    r1 = ctx->POLY1305_CTX_R[1];

    r2 = ctx->POLY1305_CTX_R[2];

    h0 = ctx->POLY1305_CTX_H[0];

    h1 = ctx->POLY1305_CTX_H[1];

    h2 = ctx->POLY1305_CTX_H[2];

    s1 = r1 * (5 << 2);

    s2 = r2 * (5 << 2);

    while (bytes >= POLY1305_BLOCK_SIZE) {

        word64 t0,t1;

        /* h += m[i] */

        t0 = U8TO64(&m[0]);

        t1 = U8TO64(&m[8]);

        h0 += (( t0                    ) & 0xfffffffffff);

        h1 += (((t0 >> 44) | (t1 << 20)) & 0xfffffffffff);

        h2 += (((t1 >> 24)             ) & 0x3ffffffffff) | hibit;

        /* h *= r */

        MUL(d0, h0, r0); MUL(d, h1, s2); ADD(d0, d); MUL(d, h2, s1); ADD(d0, d);

        MUL(d1, h0, r1); MUL(d, h1, r0); ADD(d1, d); MUL(d, h2, s2); ADD(d1, d);

        MUL(d2, h0, r2); MUL(d, h1, r1); ADD(d2, d); MUL(d, h2, r0); ADD(d2, d);

        /* (partial) h %= p */

                      c = SHR(d0, 44); h0 = LO(d0) & 0xfffffffffff;

        ADDLO(d1, c); c = SHR(d1, 44); h1 = LO(d1) & 0xfffffffffff;

        ADDLO(d2, c); c = SHR(d2, 42); h2 = LO(d2) & 0x3ffffffffff;

        h0  += c * 5; c = (h0 >> 44);  h0 =    h0  & 0xfffffffffff;

        h1  += c;

        m += POLY1305_BLOCK_SIZE;

        bytes -= POLY1305_BLOCK_SIZE;

    }

    ctx->POLY1305_CTX_H[0] = h0;

    ctx->POLY1305_CTX_H[1] = h1;

    ctx->POLY1305_CTX_H[2] = h2;

    return 0;

#else /* if not 64 bit then use 32 bit */

    const word32 hibit = (ctx->finished) ? 0 : ((word32)1 << 24); /* 1 << 128 */

    word32 r0,r1,r2,r3,r4;

    word32 s1,s2,s3,s4;

    word32 h0,h1,h2,h3,h4;

    word32 c;

#ifdef WOLFSSL_W64_WRAPPER

    #ifdef WOLFSSL_SMALL_STACK

    w64wrapper* d;

    d = (w64wrapper*)XMALLOC(5 * sizeof(w64wrapper), NULL,

        DYNAMIC_TYPE_TMP_BUFFER);

    if (d == NULL) {

        return MEMORY_E;

    }

    #else

    w64wrapper d[5];

    #endif

#else

    word64 d0,d1,d2,d3,d4;

#endif

    r0 = ctx->POLY1305_CTX_R[0];

    r1 = ctx->POLY1305_CTX_R[1];

    r2 = ctx->POLY1305_CTX_R[2];

    r3 = ctx->POLY1305_CTX_R[3];

    r4 = ctx->POLY1305_CTX_R[4];

    s1 = r1 * 5;

    s2 = r2 * 5;

    s3 = r3 * 5;

    s4 = r4 * 5;

    h0 = ctx->POLY1305_CTX_H[0];

    h1 = ctx->POLY1305_CTX_H[1];

    h2 = ctx->POLY1305_CTX_H[2];

    h3 = ctx->POLY1305_CTX_H[3];

    h4 = ctx->POLY1305_CTX_H[4];

    while (bytes >= POLY1305_BLOCK_SIZE) {

        /* h += m[i] */

        h0 += (U8TO32(m+ 0)     ) & 0x3ffffff;

        h1 += (U8TO32(m+ 3) >> 2) & 0x3ffffff;

        h2 += (U8TO32(m+ 6) >> 4) & 0x3ffffff;

        h3 += (U8TO32(m+ 9) >> 6) & 0x3ffffff;

        h4 += (U8TO32(m+12) >> 8) | hibit;

        /* h *= r */

#ifdef WOLFSSL_W64_WRAPPER

        {

            w64wrapper tmp;

            d[0] = w64Mul(h0, r0); tmp = w64Mul(h1, s4);

            d[0] = w64Add(d[0], tmp, NULL); tmp = w64Mul(h2, s3);

            d[0] = w64Add(d[0], tmp, NULL); tmp = w64Mul(h3, s2);

            d[0] = w64Add(d[0], tmp, NULL); tmp = w64Mul(h4, s1);

            d[0] = w64Add(d[0], tmp, NULL);

            d[1] = w64Mul(h0, r1); tmp = w64Mul(h1, r0);

            d[1] = w64Add(d[1], tmp, NULL); tmp = w64Mul(h2, s4);

            d[1] = w64Add(d[1], tmp, NULL); tmp = w64Mul(h3, s3);

            d[1] = w64Add(d[1], tmp, NULL); tmp = w64Mul(h4, s2);

            d[1] = w64Add(d[1], tmp, NULL);

            d[2] = w64Mul(h0, r2); tmp = w64Mul(h1, r1);

            d[2] = w64Add(d[2], tmp, NULL); tmp = w64Mul(h2, r0);

            d[2] = w64Add(d[2], tmp, NULL); tmp = w64Mul(h3, s4);

            d[2] = w64Add(d[2], tmp, NULL); tmp = w64Mul(h4, s3);

            d[2] = w64Add(d[2], tmp, NULL);

            d[3] = w64Mul(h0, r3); tmp = w64Mul(h1, r2);

            d[3] = w64Add(d[3], tmp, NULL); tmp = w64Mul(h2, r1);

            d[3] = w64Add(d[3], tmp, NULL); tmp = w64Mul(h3, r0);

            d[3] = w64Add(d[3], tmp, NULL); tmp = w64Mul(h4, s4);

            d[3] = w64Add(d[3], tmp, NULL);

            d[4] = w64Mul(h0, r4); tmp = w64Mul(h1, r3);

            d[4] = w64Add(d[4], tmp, NULL); tmp = w64Mul(h2, r2);

            d[4] = w64Add(d[4], tmp, NULL); tmp = w64Mul(h3, r1);

            d[4] = w64Add(d[4], tmp, NULL); tmp = w64Mul(h4, r0);

            d[4] = w64Add(d[4], tmp, NULL);

        }

#else

        d0 = ((word64)h0 * r0) + ((word64)h1 * s4) + ((word64)h2 * s3) +

             ((word64)h3 * s2) + ((word64)h4 * s1);

        d1 = ((word64)h0 * r1) + ((word64)h1 * r0) + ((word64)h2 * s4) +

             ((word64)h3 * s3) + ((word64)h4 * s2);

        d2 = ((word64)h0 * r2) + ((word64)h1 * r1) + ((word64)h2 * r0) +

             ((word64)h3 * s4) + ((word64)h4 * s3);

        d3 = ((word64)h0 * r3) + ((word64)h1 * r2) + ((word64)h2 * r1) +

             ((word64)h3 * r0) + ((word64)h4 * s4);

        d4 = ((word64)h0 * r4) + ((word64)h1 * r3) + ((word64)h2 * r2) +

             ((word64)h3 * r1) + ((word64)h4 * r0);

#endif

        /* (partial) h %= p */

#ifdef WOLFSSL_W64_WRAPPER

        c = w64GetLow32(w64ShiftRight(d[0], 26));

        h0 = w64GetLow32(d[0]) & 0x3ffffff;

        d[1] = w64Add32(d[1], c, NULL);

        c = w64GetLow32(w64ShiftRight(d[1], 26));

        h1 = w64GetLow32(d[1]) & 0x3ffffff;

        d[2] = w64Add32(d[2], c, NULL);

        c = w64GetLow32(w64ShiftRight(d[2], 26));

        h2 = w64GetLow32(d[2]) & 0x3ffffff;

        d[3] = w64Add32(d[3], c, NULL);

        c = w64GetLow32(w64ShiftRight(d[3], 26));

        h3 = w64GetLow32(d[3]) & 0x3ffffff;

        d[4] = w64Add32(d[4], c, NULL);

        c = w64GetLow32(w64ShiftRight(d[4], 26));

        h4 = w64GetLow32(d[4]) & 0x3ffffff;

#else

                      c = (word32)(d0 >> 26); h0 = (word32)d0 & 0x3ffffff;

        d1 += c;      c = (word32)(d1 >> 26); h1 = (word32)d1 & 0x3ffffff;

        d2 += c;      c = (word32)(d2 >> 26); h2 = (word32)d2 & 0x3ffffff;

        d3 += c;      c = (word32)(d3 >> 26); h3 = (word32)d3 & 0x3ffffff;

        d4 += c;      c = (word32)(d4 >> 26); h4 = (word32)d4 & 0x3ffffff;

#endif

        h0 += c * 5;  c =  (h0 >> 26); h0 =                h0 & 0x3ffffff;

        h1 += c;

        m += POLY1305_BLOCK_SIZE;

        bytes -= POLY1305_BLOCK_SIZE;

    }

    ctx->POLY1305_CTX_H[0] = h0;

    ctx->POLY1305_CTX_H[1] = h1;

    ctx->POLY1305_CTX_H[2] = h2;

    ctx->POLY1305_CTX_H[3] = h3;

    ctx->POLY1305_CTX_H[4] = h4;

#if defined(WOLFSSL_W64_WRAPPER) && defined(WOLFSSL_SMALL_STACK)

    XFREE(d, NULL, DYNAMIC_TYPE_TMP_BUFFER);

#endif

    return 0;

#endif /* end of 64 bit cpu blocks or 32 bit cpu */

}

static void poly1305_c_final(Poly1305* ctx, byte* mac)

{

#if defined(POLY130564)

    word64 h0,h1,h2,c;

    word64 g0,g1,g2;

    word64 t0,t1;

    /* fully carry h */

    h0 = ctx->POLY1305_CTX_H[0];

    h1 = ctx->POLY1305_CTX_H[1];

    h2 = ctx->POLY1305_CTX_H[2];

                 c = (h1 >> 44); h1 &= 0xfffffffffff;

    h2 += c;     c = (h2 >> 42); h2 &= 0x3ffffffffff;

    h0 += c * 5; c = (h0 >> 44); h0 &= 0xfffffffffff;

    h1 += c;     c = (h1 >> 44); h1 &= 0xfffffffffff;

    h2 += c;     c = (h2 >> 42); h2 &= 0x3ffffffffff;

    h0 += c * 5; c = (h0 >> 44); h0 &= 0xfffffffffff;

    h1 += c;

    /* compute h + -p */

    g0 = h0 + 5; c = (g0 >> 44); g0 &= 0xfffffffffff;

    g1 = h1 + c; c = (g1 >> 44); g1 &= 0xfffffffffff;

    g2 = h2 + c - ((word64)1 << 42);

    /* select h if h < p, or h + -p if h >= p */

    c = (g2 >> ((sizeof(word64) * CHAR_BIT) - 1)) - 1;

    g0 &= c;

    g1 &= c;

    g2 &= c;

    c = ~c;

    h0 = (h0 & c) | g0;

    h1 = (h1 & c) | g1;

    h2 = (h2 & c) | g2;

    /* h = (h + pad) */

    t0 = ctx->POLY1305_CTX_PAD[0];

    t1 = ctx->POLY1305_CTX_PAD[1];

    h0 += (( t0                    ) & 0xfffffffffff)    ;

    c = (h0 >> 44); h0 &= 0xfffffffffff;

    h1 += (((t0 >> 44) | (t1 << 20)) & 0xfffffffffff) + c;

    c = (h1 >> 44); h1 &= 0xfffffffffff;

    h2 += (((t1 >> 24)             ) & 0x3ffffffffff) + c;

    h2 &= 0x3ffffffffff;

    /* mac = h % (2^128) */

    h0 = ((h0      ) | (h1 << 44));

    h1 = ((h1 >> 20) | (h2 << 24));

    U64TO8(mac + 0, h0);

    U64TO8(mac + 8, h1);

    /* zero out the state */

    ctx->POLY1305_CTX_H[0] = 0;

    ctx->POLY1305_CTX_H[1] = 0;

    ctx->POLY1305_CTX_H[2] = 0;

    ctx->POLY1305_CTX_R[0] = 0;

    ctx->POLY1305_CTX_R[1] = 0;

    ctx->POLY1305_CTX_R[2] = 0;

    ctx->POLY1305_CTX_PAD[0] = 0;

    ctx->POLY1305_CTX_PAD[1] = 0;

#else /* if not 64 bit then use 32 bit */

    word32 h0,h1,h2,h3,h4,c;

    word32 g0,g1,g2,g3,g4;

#ifdef WOLFSSL_W64_WRAPPER

    w64wrapper f;

#else

    word64 f;

#endif

    word32 mask;

    /* fully carry h */

    h0 = ctx->POLY1305_CTX_H[0];

    h1 = ctx->POLY1305_CTX_H[1];

    h2 = ctx->POLY1305_CTX_H[2];

    h3 = ctx->POLY1305_CTX_H[3];

    h4 = ctx->POLY1305_CTX_H[4];

                 c = h1 >> 26; h1 = h1 & 0x3ffffff;

    h2 +=     c; c = h2 >> 26; h2 = h2 & 0x3ffffff;

    h3 +=     c; c = h3 >> 26; h3 = h3 & 0x3ffffff;

    h4 +=     c; c = h4 >> 26; h4 = h4 & 0x3ffffff;

    h0 += c * 5; c = h0 >> 26; h0 = h0 & 0x3ffffff;

    h1 +=     c;

    /* compute h + -p */

    g0 = h0 + 5; c = g0 >> 26; g0 &= 0x3ffffff;

    g1 = h1 + c; c = g1 >> 26; g1 &= 0x3ffffff;

    g2 = h2 + c; c = g2 >> 26; g2 &= 0x3ffffff;

    g3 = h3 + c; c = g3 >> 26; g3 &= 0x3ffffff;

    g4 = h4 + c - ((word32)1 << 26);

    /* select h if h < p, or h + -p if h >= p */

    mask = ((word32)g4 >> ((sizeof(word32) * CHAR_BIT) - 1)) - 1;

    g0 &= mask;

    g1 &= mask;

    g2 &= mask;

    g3 &= mask;

    g4 &= mask;

    mask = ~mask;

    h0 = (h0 & mask) | g0;

    h1 = (h1 & mask) | g1;

    h2 = (h2 & mask) | g2;

    h3 = (h3 & mask) | g3;

    h4 = (h4 & mask) | g4;

    /* h = h % (2^128) */

    h0 = ((h0      ) | (h1 << 26)) & 0xffffffff;

    h1 = ((h1 >>  6) | (h2 << 20)) & 0xffffffff;

    h2 = ((h2 >> 12) | (h3 << 14)) & 0xffffffff;

    h3 = ((h3 >> 18) | (h4 <<  8)) & 0xffffffff;

    /* mac = (h + pad) % (2^128) */

#ifdef WOLFSSL_W64_WRAPPER

    f = w64From32(0, h0);

    f = w64Add32(f, ctx->POLY1305_CTX_PAD[0], NULL);

    h0 = w64GetLow32(f);

    f = w64ShiftRight(f, 32);

    f = w64Add32(f, h1, NULL);

    f = w64Add32(f, ctx->POLY1305_CTX_PAD[1], NULL);

    h1 = w64GetLow32(f);

    f = w64ShiftRight(f, 32);

    f = w64Add32(f, h2, NULL);

    f = w64Add32(f, ctx->POLY1305_CTX_PAD[2], NULL);

    h2 = w64GetLow32(f);

    f = w64ShiftRight(f, 32);

    f = w64Add32(f, h3, NULL);

    f = w64Add32(f, ctx->POLY1305_CTX_PAD[3], NULL);

    h3 = w64GetLow32(f);

#else

    f = (word64)h0 + ctx->POLY1305_CTX_PAD[0]            ; h0 = (word32)f;

    f = (word64)h1 + ctx->POLY1305_CTX_PAD[1] + (f >> 32); h1 = (word32)f;

    f = (word64)h2 + ctx->POLY1305_CTX_PAD[2] + (f >> 32); h2 = (word32)f;

    f = (word64)h3 + ctx->POLY1305_CTX_PAD[3] + (f >> 32); h3 = (word32)f;

#endif

    U32TO8(mac + 0, h0);

    U32TO8(mac + 4, h1);

    U32TO8(mac + 8, h2);

    U32TO8(mac + 12, h3);

    /* zero out the state */

    ctx->POLY1305_CTX_H[0] = 0;

    ctx->POLY1305_CTX_H[1] = 0;

    ctx->POLY1305_CTX_H[2] = 0;

    ctx->POLY1305_CTX_H[3] = 0;

    ctx->POLY1305_CTX_H[4] = 0;

    ctx->POLY1305_CTX_R[0] = 0;

    ctx->POLY1305_CTX_R[1] = 0;

    ctx->POLY1305_CTX_R[2] = 0;

    ctx->POLY1305_CTX_R[3] = 0;

    ctx->POLY1305_CTX_R[4] = 0;

    ctx->POLY1305_CTX_PAD[0] = 0;

    ctx->POLY1305_CTX_PAD[1] = 0;

    ctx->POLY1305_CTX_PAD[2] = 0;

    ctx->POLY1305_CTX_PAD[3] = 0;

#endif

}

#undef POLY1305_CTX_R

#undef POLY1305_CTX_H

#undef POLY1305_CTX_PAD

#endif /* WOLFSSL_POLY1305_C */

#ifdef WOLFSSL_ARM_POLY1305_NEED_C

#ifdef WOLFSSL_ARM_POLY1305_NEON_FALLBACK

static cpuid_flags_t poly1305_cpuid_flags = WC_CPUID_INITIALIZER;

/* Return non-zero when NEON/ASIMD is present and the asm path should run. */

static WC_INLINE int poly1305_use_neon(void)

{

    cpuid_get_flags_ex(&poly1305_cpuid_flags);

    return IS_AARCH64_ASIMD(poly1305_cpuid_flags);

}

#endif

/* Dispatch each Poly1305 operation to the NEON assembly or the C

 * implementation, choosing at runtime when both are available. */

static void poly1305_blocks_aarch64(Poly1305* ctx, const unsigned char* m,

    size_t bytes)

{

#ifdef WOLFSSL_ARM_POLY1305_C_ONLY

    poly1305_c_blocks(ctx, m, bytes);

#else

    if (poly1305_use_neon())

        poly1305_arm64_blocks(ctx, m, bytes);

    else

        poly1305_c_blocks(ctx, m, bytes);

#endif

}

static void poly1305_block_aarch64(Poly1305* ctx, const unsigned char* m)

{

#ifdef WOLFSSL_ARM_POLY1305_C_ONLY

    poly1305_c_blocks(ctx, m, POLY1305_BLOCK_SIZE);

#else

    if (poly1305_use_neon())

        poly1305_arm64_block_16(ctx, m);

    else

        poly1305_c_blocks(ctx, m, POLY1305_BLOCK_SIZE);

#endif

}

static void poly1305_setkey_aarch64(Poly1305* ctx, const byte* key)

{

#ifdef WOLFSSL_ARM_POLY1305_C_ONLY

    poly1305_c_setkey(ctx, key);

#else

    if (poly1305_use_neon())

        poly1305_set_key(ctx, key);

    else

        poly1305_c_setkey(ctx, key);

#endif

}

static void poly1305_final_aarch64(Poly1305* ctx, byte* mac)

{

#ifdef WOLFSSL_ARM_POLY1305_C_ONLY

    poly1305_c_final(ctx, mac);

#else

    if (poly1305_use_neon())

        poly1305_final(ctx, mac);

    else

        poly1305_c_final(ctx, mac);

#endif

}

#endif /* WOLFSSL_ARM_POLY1305_NEED_C */

#if !defined(WOLFSSL_RISCV_ASM)

/*

This local function operates on a message with a given number of bytes

with a given ctx pointer to a Poly1305 structure.

*/

static int poly1305_blocks(Poly1305* ctx, const unsigned char *m,

                     size_t bytes)

{

#ifdef USE_INTEL_POLY1305_SPEEDUP

    /* AVX2 is handled in wc_Poly1305Update. */

    SAVE_VECTOR_REGISTERS(return _svr_ret;);

    poly1305_blocks_avx(ctx, m, bytes);

    RESTORE_VECTOR_REGISTERS();

    return 0;

#elif defined(WOLFSSL_ARMASM) && defined(__aarch64__)

    poly1305_blocks_aarch64(ctx, m, bytes);

    return 0;

#elif defined(WOLFSSL_ARMASM) && defined(WOLFSSL_ARMASM_THUMB2)

    poly1305_blocks_thumb2_16(ctx, m, bytes, 1);

    return 0;

#elif defined(WOLFSSL_ARMASM)

#ifndef WOLFSSL_ARMASM_NO_NEON

    poly1305_arm32_blocks(ctx, m, bytes);

    return 0;

#else

    poly1305_arm32_blocks_16(ctx, m, bytes, 1);

    return 0;

#endif

#else

    return poly1305_c_blocks(ctx, m, bytes);

#endif

}

#endif /* !WOLFSSL_RISCV_ASM */

/*

This local function is used for the last call when a message with a given

number of bytes is less than the block size.

*/

static WC_INLINE int poly1305_block(Poly1305* ctx, const unsigned char *m)

{

#ifdef USE_INTEL_POLY1305_SPEEDUP

    /* No call to poly1305_block when AVX2, AVX2 does 4 blocks at a time. */

    SAVE_VECTOR_REGISTERS(return _svr_ret;);

    poly1305_block_avx(ctx, m);

    RESTORE_VECTOR_REGISTERS();

    return 0;

#elif defined(WOLFSSL_ARMASM) && defined(WOLFSSL_ARMASM_THUMB2)

    poly1305_blocks_thumb2_16(ctx, m, POLY1305_BLOCK_SIZE, !ctx->finished);

    return 0;

#elif defined(WOLFSSL_ARMASM) && !defined(__aarch64__)

    poly1305_arm32_blocks_16(ctx, m, POLY1305_BLOCK_SIZE, !ctx->finished);

    return 0;

#elif defined(WOLFSSL_ARMASM)

    poly1305_block_aarch64(ctx, m);

    return 0;

#elif defined(WOLFSSL_RISCV_ASM)

    poly1305_block_16_riscv64(ctx, m);

    return 0;

#else

    return poly1305_blocks(ctx, m, POLY1305_BLOCK_SIZE);

#endif

}

int wc_Poly1305SetKey(Poly1305* ctx, const byte* key, word32 keySz)

{

    if (key == NULL)

        return BAD_FUNC_ARG;

#ifdef CHACHA_AEAD_TEST

    word32 k;

    printf("Poly key used:\n");

    for (k = 0; k < keySz; k++) {

        printf("%02x", key[k]);

        if ((k+1) % 8 == 0)

            printf("\n");

    }

    printf("\n");

#endif

    if ((ctx == NULL) || (key == NULL) || (keySz != 32)) {

        return BAD_FUNC_ARG;

    }

#ifdef USE_INTEL_POLY1305_SPEEDUP

    cpuid_get_flags_ex(&intel_flags);

    SAVE_VECTOR_REGISTERS(return _svr_ret;);

    #ifdef HAVE_INTEL_AVX512

    if (poly1305_use_ifma(intel_flags))

        poly1305_setkey_avx512ifma(ctx, key);

    else

    #endif

    #ifdef HAVE_INTEL_AVX2

    if (IS_INTEL_AVX2(intel_flags))

        poly1305_setkey_avx2(ctx, key);

    else

    #endif

        poly1305_setkey_avx(ctx, key);

    RESTORE_VECTOR_REGISTERS();

    ctx->started = 0;

#ifdef WOLFSSL_CHACHA20_POLY1305_FUSED

    ctx->forceAvx2 = 0;

#endif

#ifdef WOLFSSL_CHACHA20_POLY1305_FUSED_IFMA

    ctx->forceScalar = 0;

#endif

#elif defined(WOLFSSL_ARMASM)

#ifdef __aarch64__

    poly1305_setkey_aarch64(ctx, key);

#else

    poly1305_set_key(ctx, key);

#endif

    ctx->finished = 0;

#elif defined(WOLFSSL_RISCV_ASM)

    poly1305_set_key_riscv64(ctx, key);

#else

    poly1305_c_setkey(ctx, key);

#endif

    return 0;

}

int wc_Poly1305Final(Poly1305* ctx, byte* mac)

{

    if (ctx == NULL || mac == NULL)

        return BAD_FUNC_ARG;

#ifdef USE_INTEL_POLY1305_SPEEDUP

    SAVE_VECTOR_REGISTERS(return _svr_ret;);

    #ifdef HAVE_INTEL_AVX512

    if (!POLY1305_FORCE_AVX2(ctx) && !POLY1305_FORCE_SCALAR(ctx) &&

            poly1305_use_ifma(intel_flags))

        poly1305_final_avx512ifma(ctx, mac);

    else

    if (!POLY1305_FORCE_AVX2(ctx) && !POLY1305_FORCE_SCALAR(ctx) &&

            poly1305_use_avx512(intel_flags))

        poly1305_final_avx512(ctx, mac);

    else

    #endif

    #ifdef HAVE_INTEL_AVX2

    if (!POLY1305_FORCE_SCALAR(ctx) && IS_INTEL_AVX2(intel_flags))

        poly1305_final_avx2(ctx, mac);

    else

    #endif

        poly1305_final_avx(ctx, mac);

    RESTORE_VECTOR_REGISTERS();

#elif defined(WOLFSSL_ARMASM)

    #if !defined(WOLFSSL_ARMASM_THUMB2) && !defined(WOLFSSL_ARMASM_NO_NEON) && \

        !defined(__aarch64__)

        if (ctx->leftover >= POLY1305_BLOCK_SIZE) {

             size_t len = ctx->leftover & (~(POLY1305_BLOCK_SIZE - 1));

             poly1305_arm32_blocks(ctx, ctx->buffer, len);

             ctx->leftover -= len;

             if (ctx->leftover) {

                 XMEMCPY(ctx->buffer, ctx->buffer + len, ctx->leftover);

             }

        }

    #endif

        if (ctx->leftover) {

             size_t i = ctx->leftover;

             ctx->buffer[i++] = 1;

             for (; i < POLY1305_BLOCK_SIZE; i++) {

                 ctx->buffer[i] = 0;

             }

            ctx->finished = 1;

            poly1305_block(ctx, ctx->buffer);

        }

#ifdef __aarch64__

        poly1305_final_aarch64(ctx, mac);

#else

        poly1305_final(ctx, mac);

#endif

#elif defined(WOLFSSL_RISCV_ASM)

    /* process the remaining block */

    if (ctx->leftover) {

        size_t i = ctx->leftover;

        ctx->buffer[i] = 1;

        for (i = i + 1; i < POLY1305_BLOCK_SIZE; i++)

            ctx->buffer[i] = 0;

        ctx->finished = 1;

        poly1305_block_16_riscv64(ctx, ctx->buffer);

    }

    poly1305_final_riscv64(ctx, mac);

#else

    /* process the remaining block */

    if (ctx->leftover) {

        size_t i = ctx->leftover;

        ctx->buffer[i++] = 1;

        for (; i < POLY1305_BLOCK_SIZE; i++)

            ctx->buffer[i] = 0;

        ctx->finished = 1;

        poly1305_block(ctx, ctx->buffer);

    }

    poly1305_c_final(ctx, mac);

#endif

    return 0;

}

int wc_Poly1305Update(Poly1305* ctx, const byte* m, word32 bytes)

{

    size_t i;

    if (ctx == NULL || (m == NULL && bytes > 0))

        return BAD_FUNC_ARG;

    if (bytes == 0) {

        /* valid, but do nothing */

        return 0;

    }

#ifdef CHACHA_AEAD_TEST

    word32 k;

    printf("Raw input to poly:\n");

    for (k = 0; k < bytes; k++) {

        printf("%02x", m[k]);

        if ((k+1) % 16 == 0)

            printf("\n");

    }

    printf("\n");

#endif

#if defined(WOLFSSL_ARMASM) && !defined(WOLFSSL_ARMASM_THUMB2) && \

    !defined(WOLFSSL_ARMASM_NO_NEON)

    /* handle leftover */

    if (ctx->leftover) {

        size_t want = sizeof(ctx->buffer) - ctx->leftover;

        if (want > bytes)

            want = bytes;

        for (i = 0; i < want; i++)

            ctx->buffer[ctx->leftover + i] = m[i];

        bytes -= (word32)want;

        m += want;

        ctx->leftover += want;

        if (ctx->leftover < sizeof(ctx->buffer)) {

            return 0;

        }

        poly1305_blocks(ctx, ctx->buffer, sizeof(ctx->buffer));

        ctx->leftover = 0;

    }

    /* process full blocks */

    if (bytes >= sizeof(ctx->buffer)) {

        size_t want = bytes & ~((size_t)POLY1305_BLOCK_SIZE - 1);

        poly1305_blocks(ctx, m, want);

        m += want;

        bytes -= (word32)want;

    }

    /* store leftover */

    if (bytes) {