memeq_alligned

static inline constexpr uintptr_t roundup_n(uintptr_t a, uintptr_t n) {
    return ((a + (n - 1)) & ~(n - 1));
}

static inline constexpr uintptr_t rounddown_n(uintptr_t a, uintptr_t n) {
    return (a & ~(n - 1));
}

static inline bool isaligned_n(uintptr_t ptr, uintptr_t n) {
    return ((ptr & ((n)-1)) == 0);
}

static __m128i _mm_alignr_epi8_nonconst(__m128i a, __m128i b, int n) {
#define ALIGNR_CASE(n) case n: return _mm_alignr_epi8(a, b, n); break;
    switch (n) {
        ALIGNR_CASE(0x0);
        ALIGNR_CASE(0x1);
        ALIGNR_CASE(0x2);
        ALIGNR_CASE(0x3);
        ALIGNR_CASE(0x4);
        ALIGNR_CASE(0x5);
        ALIGNR_CASE(0x6);
        ALIGNR_CASE(0x7);
        ALIGNR_CASE(0x8);
        ALIGNR_CASE(0x9);
        ALIGNR_CASE(0xA);
        ALIGNR_CASE(0xB);
        ALIGNR_CASE(0xC);
        ALIGNR_CASE(0xD);
        ALIGNR_CASE(0xE);
        ALIGNR_CASE(0xF);
    }
#ifdef _MSC_VER
    _assume(0);
#else
    __builtin_unreachable();
#endif
}

bool memeq(const uint8_t* pu1, const uint8_t* pu2, size_t n) {
    auto* pa1 = (const uint8_t*)rounddown_n((uintptr_t)pu1, 16);
    auto* pa2 = (const uint8_t*)rounddown_n((uintptr_t)pu2, 16);
    auto diff1 = pu1 - pa1;
    auto diff2 = pu2 - pa2;

    auto xmm1_1 = _mm_load_si128((const __m128i*)pa1);
    auto xmm2_1 = _mm_load_si128((const __m128i*)pa2);

    size_t off;
    for (off = sizeof(__m128i); off <= n; off += sizeof(__m128i)) {
        auto xmm1_2 = _mm_load_si128((const __m128i*)(pa1 + off));
        auto xmm2_2 = _mm_load_si128((const __m128i*)(pa2 + off));
        auto xmm1_a = _mm_alignr_epi8_nonconst(xmm1_2, xmm1_1, diff1);
        auto xmm2_a = _mm_alignr_epi8_nonconst(xmm2_2, xmm2_1, diff2);
        if (!_mm_test_all_ones(_mm_cmpeq_epi8(xmm1_a, xmm2_a))) return false;
        xmm1_1 = xmm1_2;
        xmm2_1 = xmm2_2;
    }

    if (uint16_t left = n + sizeof(__m128i) - off; left > 0) {
        auto xmm1_2 = _mm_load_si128((const __m128i*)(pa1 + off));
        auto xmm2_2 = _mm_load_si128((const __m128i*)(pa2 + off));
        auto xmm1_a = _mm_alignr_epi8_nonconst(xmm1_2, xmm1_1, diff1);
        auto xmm2_a = _mm_alignr_epi8_nonconst(xmm2_2, xmm2_1, diff2);
        auto res = _mm_cmpeq_epi8(xmm1_a, xmm2_a);
        auto mask = (uint16_t)_mm_movemask_epi8(res);
        uint16_t bits = (1 << left) - 1;
        return (mask & bits) == bits;
    }

    return true;
}