_umul128 is not a valid intrinsic for x86 Windows.

[Convery]

Code: https://github.com/greg7mdp/parallel-hashmap/blob/master/parallel_hashmap/phmap_bits.h#L470-L483
MSDN: https://docs.microsoft.com/en-us/cpp/intrinsics/umul128?view=vs-2017
SO examples: https://stackoverflow.com/questions/46870373/umul128-on-windows-32-bits

[Convery]

Example fix for line 470-483 as stolen from SO:

#ifdef PHMAP_HAVE_INTRINSIC_INT128
    inline uint64_t umul128(uint64_t a, uint64_t b, uint64_t* high)
    {
        auto result = static_cast<unsigned __int128>(a) * static_cast<unsigned __int128>(b);
        *high = static_cast<uint64_t>(result >> 64);
        return static_cast<uint64_t>(result);
    }
#elif (defined(_MSC_VER))
    #if defined(_WIN64)
    #pragma intrinsic(_umul128)
    inline uint64_t umul128(uint64_t a, uint64_t b, uint64_t* high)
    {
        return _umul128(a, b, high);
    }
    #else
    #pragma intrinsic(__emulu)
    inline uint64_t umul128(uint64_t multiplier, uint64_t multiplicand, uint64_t *product_hi)
    {
        uint64_t a = multiplier >> 32;
        uint64_t b = (uint32_t)multiplier; // & 0xFFFFFFFF;
        uint64_t c = multiplicand >> 32;
        uint64_t d = (uint32_t)multiplicand; // & 0xFFFFFFFF;

        uint64_t ad = __emulu(a, d);
        uint64_t bd = __emulu(b, d);

        uint64_t adbc = ad + __emulu(b, c);
        uint64_t adbc_carry = (adbc < ad); // ? 1 : 0;
        // MSVC gets confused by the ternary and makes worse code than using a boolean in an integer context for 1 : 0

        // multiplier * multiplicand = product_hi * 2^64 + product_lo
        uint64_t product_lo = bd + (adbc << 32);
        uint64_t product_lo_carry = (product_lo < bd); // ? 1 : 0;
        *product_hi = __emulu(a, c) + (adbc >> 32) + (adbc_carry << 32) + product_lo_carry;

        return product_lo;
    }
    #endif
#endif

[greg7mdp]

Thanks for the report and suggesting a solution. Not closing the issue yet because there are a couple compilation warnings I need to take care of.

[greg7mdp]

I updated the code and checked on x86... works fine now. Thanks again for the bug report.