The code was applying std::abs to the difference of two unsigned integers, which results in an unsigned integer, which means there is no reason to take the absolute value. So I think you want to take the difference of two signed integers according to:
for (uint i = 0; i < rows; ++i)
ASSERT_LT (std::abs ((int)refMax[i] - (int)results[i]), eps);
The code was applying std::abs to the difference of two unsigned integers, which results in an unsigned integer, which means there is no reason to take the absolute value. So I think you want to take the difference of two signed integers according to:
that made the compiler error go away.