#include <metal_stdlib>

using namespace metal;

template<typename T, size_t N>
struct tint_array {
    const constant T& operator[](size_t i) const constant { return elements[i]; }
    device T& operator[](size_t i) device { return elements[i]; }
    const device T& operator[](size_t i) const device { return elements[i]; }
    thread T& operator[](size_t i) thread { return elements[i]; }
    const thread T& operator[](size_t i) const thread { return elements[i]; }
    threadgroup T& operator[](size_t i) threadgroup { return elements[i]; }
    const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; }
    T elements[N];
};

void compute_main_inner(uint local_invocation_index, threadgroup tint_array<atomic_uint, 4>* const tint_symbol) {
  for(uint idx = local_invocation_index; (idx < 4u); idx = (idx + 1u)) {
    __asm__("");
    uint const i = idx;
    atomic_store_explicit(&((*(tint_symbol))[i]), 0u, memory_order_relaxed);
  }
  threadgroup_barrier(mem_flags::mem_threadgroup);
  atomic_store_explicit(&((*(tint_symbol))[1]), 1u, memory_order_relaxed);
}

kernel void compute_main(uint local_invocation_index [[thread_index_in_threadgroup]]) {
  threadgroup tint_array<atomic_uint, 4> tint_symbol_1;
  compute_main_inner(local_invocation_index, &(tint_symbol_1));
  return;
}