#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];
};

struct tint_module_vars_struct {
  device tint_array<int, 10>* output;
};

[[max_total_threads_per_threadgroup(1)]]
kernel void foo(device tint_array<int, 10>* output [[buffer(0)]]) {
  tint_module_vars_struct const tint_module_vars = tint_module_vars_struct{.output=output};
  int i = 0;
  {
    uint2 tint_loop_idx = uint2(4294967295u);
    while(true) {
      if (all((tint_loop_idx == uint2(0u)))) {
        break;
      }
      int x = (*tint_module_vars.output)[min(uint(i), 9u)];
      {
        uint const tint_low_inc = (tint_loop_idx.x - 1u);
        tint_loop_idx.x = tint_low_inc;
        uint const tint_carry = uint((tint_low_inc == 4294967295u));
        tint_loop_idx.y = (tint_loop_idx.y - tint_carry);
        int x_1 = (*tint_module_vars.output)[min(uint(x), 9u)];
        i = as_type<int>((as_type<uint>(i) + as_type<uint>(x_1)));
        if ((i > 10)) { break; }
      }
    }
  }
  (*tint_module_vars.output)[0u] = i;
}