round_up Round to the next multiple

template <typename T, typename = U>
[[nodiscard]] __host__ __device__ inline
constexpr cuda::std::common_type_t<T, U> round_up(T value, U base_multiple) noexcept;

value: The value to be rounded up. base_multiple: The base multiple to which the value rounds up.

  • Requires: T and U are integral types (including 128-bit integers) or enumerators.

  • Preconditions: a >= 0 is true and b > 0 is true.

  • Returns: a rounded up to the smallest multiple of b greater than or equal to a. If a is already a multiple of b, return a.

  • Note: the result can overflow if ceil(a / b) * b exceeds the maximum value of the common type of

    a and b. The condition is checked in debug mode.

Note

The function requires C++17 onwards

Performance considerations:

  • The function performs a ceiling division (cuda::ceil_div()) followed by a multiplication

Example:

#include <cuda/cmath>

__global__ void example_kernel(int a, unsigned b, unsigned* result) {
  // a = 7, b = 3 -> result = 9
  *result = cuda::round_up(a, b);
}

See it on Godbolt TODO