[FEA]: Add round up/down to multiple

[fbusato]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

libcu++

Is your feature request related to a problem? Please describe.

The following patterns are very common:

• floor(a / b) * b
• ceil(a / b) * b

Introduce two functions to compute round up/down to a given multiple

Describe the solution you'd like

template<typename T>
//HOST_DEVICE_NODISCARD
constexpr T round_up_to_multiple(T value, T mul) {
    static_assert(::cuda::is_integral<T>::value);
    //ASSERT_OR_ASSUME(value >= 0)
    //ASSERT_OR_ASSUME(mul   > 0)
    auto  cdiv = ceil_div(value, mul);
    //ASSERT(!is_mul_overflow(cdiv, mul))
    using U    = ::cuda::std::__make_unsigned_t<T>;
    auto  mul1 = static_cast<U>(mul);  // for optimization purposes
    auto  ret  = static_cast<T>(cdiv * mul1);
    //ASSERT_OR_ASSUME(ret >= value)
    return ret;
}

template<typename T>
//HOST_DEVICE_NODISCARD
constexpr T round_down_to_multiple(T value, T mul) {
    static_assert(::cuda::is_integral<T>::value);
    //ASSERT_OR_ASSUME(value >= 0)
    //ASSERT_OR_ASSUME(mul   > 0)
    //ASSERT_OR_ASSUME(mul <= value)
    using U    = ::cuda::std::__make_unsigned_t<T>;
    auto  value1 = static_cast<U>(value); // for optimization purposes
    auto  mul1 = static_cast<U>(mul);  // for optimization purposes
    auto  ret    = static_cast<T>((value1 / mul1) * mul1);
    //ASSERT_OR_ASSUME(ret >= value - mul)
    return ret;
}

Describe alternatives you've considered

round_up_to_multiple(T, U);
round_down_to_multiple(T, U);

Additional context

#1517
#2375

[fbusato]

@bernhardmgruber @pauleonix for visibility

[bernhardmgruber]

Thx a lot for these suggestion! I would love to see those in libcu++!