inline __attribute__()

in common/checksum/crc.cpp [673:713]

• 41 lines of code
• 4 McCabe index (conditional complexity)

inline __attribute__((always_inline))
__m128i crc64ecma_hw_big_avx512(const uint8_t*& data, size_t& nbytes, uint64_t crc) {
    assert(nbytes >= 256);
    __attribute__((aligned(16)))
    v512 crc0 = {(long)~crc};
    auto& ptr = (const v512*&)data;
    auto zmm0 = _mm512_loadu_si512(ptr++); zmm0 ^= crc0;
    auto zmm4 = _mm512_loadu_si512(ptr++);
    nbytes -= 128;
    if (nbytes < 384) {
        auto rk3 = _mm512_broadcast_i32x4(*(__m128i*)_RK(3));
        do { // fold 128 bytes each iteration
            zmm0 = OP(zmm0, rk3, ptr++);
            zmm4 = OP(zmm4, rk3, ptr++);
            nbytes -= 128;
        } while (nbytes >= 128);
    } else { // nbytes >= 384
        auto rk_1_2 = _mm512_broadcast_i32x4(*(__m128i*)&rk512[0]);
        auto zmm7   = _mm512_loadu_si512(ptr++);
        auto zmm8   = _mm512_loadu_si512(ptr++);
        nbytes -= 128;
        do { // fold 256 bytes each iteration
            zmm0 = OP(zmm0, rk_1_2, ptr++);
            zmm4 = OP(zmm4, rk_1_2, ptr++);
            zmm7 = OP(zmm7, rk_1_2, ptr++);
            zmm8 = OP(zmm8, rk_1_2, ptr++);
            nbytes -= 256;
        } while (nbytes >= 256);
        auto rk3 = _mm512_broadcast_i32x4(*(__m128i*)_RK(3));
        zmm0 = OP(zmm0, rk3, zmm7);
        zmm4 = OP(zmm4, rk3, zmm8);
    }
    auto t = _mm512_extracti64x2_epi64(zmm4, 0x03);
    auto zmm7 = v512{t[0], t[1]};
    auto zmm1 = OP(zmm0, *(v512*)_RK(9), zmm7);
         zmm1 = OP(zmm4, *(v512*)_RK(17), zmm1);
    auto zmm8 = _mm512_shuffle_i64x2(zmm1, zmm1, 0x4e);
    auto ymm8 = ((__m256i&)zmm8) ^ ((__m256i&)zmm1);
    return _mm256_extracti64x2_epi64(ymm8, 0) ^
           _mm256_extracti64x2_epi64(ymm8, 1) ;
}