Project 1: Henry Zhu

[Maknee]

• Repo Link

Features

• Has naive, uniform, and coherent implementations of boid simulation
• Has an awesome repo cover that includes answers to questions mentioned in INSTRUCTIONS, FPS performance graph and Nsight performance summaries for these different implementations (with different number of boids and blocks)
• checkCUDAErrorWithLine("error") gets reduced to checkCUDAError(), which tells the file, line and tracks the error.
• Uses c++11 features (type traits, variadic templates, lamdas...):

/*type_traits*/
//swap pointers (helps with ping pong)
template<typename T, typename std::enable_if<std::is_pointer<T>::value>::type* = nullptr>
void swap_pointers(T& p1, T& p2)
{
  T temp = p1;
  p1 = p2;
  p2 = temp;
}

/*variadic templates*/
template<typename Type = int, typename... Ts>
__device__ auto truncate_to(Ts*... ts)
{
  (void)std::initializer_list<int>{(*ts = static_cast<Type>(*ts), 0)...};
}

• Uses plenty templating and lambdas to abstract details.
• For example (abstracts checking rules):

template<typename CheckSuccessCallback>
__device__ void check_rule(float rule_distance, int this_boid, int other_boid, const glm::vec3* pos, CheckSuccessCallback check_success_callback)
{
  const auto& this_boid_pos = pos[this_boid];
  const auto& other_boid_pos = pos[other_boid];

  if (this_boid != other_boid && glm::distance(this_boid_pos, other_boid_pos) < rule_distance)
  {
    check_success_callback();
  }
}

• Rule 1 uses above by specifying the arguments more

__device__ void check_rule1(int this_boid, int other_boid, const glm::vec3* pos, glm::vec3& perceived_center, int& num_neighbors)
{
  check_rule(rule1Distance, this_boid, other_boid, pos,
             [&]()
             {
               const auto& other_boid_pos = pos[other_boid];
               perceived_center += other_boid_pos;
               num_neighbors++;
             });
}