#pragma clang diagnostic ignored "-Wmissing-prototypes"

#include <metal_stdlib>
#include <simd/simd.h>

using namespace metal;

struct SSBO0
{
    float4 as[1];
};

struct SSBO1
{
    float4 bs[1];
};

constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(1u);

template<typename T, uint A>
inline void spvArrayCopyFromConstantToStack1(thread T (&dst)[A], constant T (&src)[A])
{
    for (uint i = 0; i < A; i++)
    {
        dst[i] = src[i];
    }
}

template<typename T, uint A>
inline void spvArrayCopyFromConstantToThreadGroup1(threadgroup T (&dst)[A], constant T (&src)[A])
{
    for (uint i = 0; i < A; i++)
    {
        dst[i] = src[i];
    }
}

template<typename T, uint A>
inline void spvArrayCopyFromStackToStack1(thread T (&dst)[A], thread const T (&src)[A])
{
    for (uint i = 0; i < A; i++)
    {
        dst[i] = src[i];
    }
}

template<typename T, uint A>
inline void spvArrayCopyFromStackToThreadGroup1(threadgroup T (&dst)[A], thread const T (&src)[A])
{
    for (uint i = 0; i < A; i++)
    {
        dst[i] = src[i];
    }
}

template<typename T, uint A>
inline void spvArrayCopyFromThreadGroupToStack1(thread T (&dst)[A], threadgroup const T (&src)[A])
{
    for (uint i = 0; i < A; i++)
    {
        dst[i] = src[i];
    }
}

template<typename T, uint A>
inline void spvArrayCopyFromThreadGroupToThreadGroup1(threadgroup T (&dst)[A], threadgroup const T (&src)[A])
{
    for (uint i = 0; i < A; i++)
    {
        dst[i] = src[i];
    }
}

kernel void main0(device SSBO0& _16 [[buffer(0)]], device SSBO1& _32 [[buffer(1)]], uint3 gl_GlobalInvocationID [[thread_position_in_grid]], uint gl_LocalInvocationIndex [[thread_index_in_threadgroup]])
{
    float4 _37[2] = { _16.as[gl_GlobalInvocationID.x], _32.bs[gl_GlobalInvocationID.x] };
    float4 values[2];
    spvArrayCopyFromStackToStack1(values, _37);
    _16.as[0] = values[gl_LocalInvocationIndex];
    _32.bs[1] = float4(40.0);
}