#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_packed_vec3_f32_array_element {
  packed_float3 elements;
};

float3x3 tint_unpack_vec3_in_composite(tint_array<tint_packed_vec3_f32_array_element, 3> in) {
  float3x3 result = float3x3(float3(in[0].elements), float3(in[1].elements), float3(in[2].elements));
  return result;
}

float3x3 tint_workgroupUniformLoad(threadgroup tint_array<tint_packed_vec3_f32_array_element, 3>* const p) {
  threadgroup_barrier(mem_flags::mem_threadgroup);
  float3x3 const result = tint_unpack_vec3_in_composite(*(p));
  threadgroup_barrier(mem_flags::mem_threadgroup);
  return result;
}

float3x3 foo(threadgroup tint_array<tint_packed_vec3_f32_array_element, 3>* const tint_symbol) {
  return tint_workgroupUniformLoad(tint_symbol);
}