#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 Params {
  /* 0x0000 */ uint filterDim;
  /* 0x0004 */ uint blockDim;
};

struct Flip {
  /* 0x0000 */ uint value;
};

void tint_symbol_inner(uint3 WorkGroupID, uint3 LocalInvocationID, uint local_invocation_index, threadgroup tint_array<tint_array<float3, 256>, 4>* const tint_symbol_1, const constant Params* const tint_symbol_2, texture2d<float, access::sample> tint_symbol_3, const constant Flip* const tint_symbol_4, sampler tint_symbol_5, texture2d<float, access::write> tint_symbol_6) {
  for(uint idx = local_invocation_index; (idx < 1024u); idx = (idx + 64u)) {
    uint const i_1 = (idx / 256u);
    uint const i_2 = (idx % 256u);
    (*(tint_symbol_1))[i_1][i_2] = float3(0.0f);
  }
  threadgroup_barrier(mem_flags::mem_threadgroup);
  uint const filterOffset = (((*(tint_symbol_2)).filterDim - 1u) / 2u);
  uint2 const dims = uint2(tint_symbol_3.get_width(0), tint_symbol_3.get_height(0));
  uint2 const baseIndex = (((uint3(WorkGroupID).xy * uint2((*(tint_symbol_2)).blockDim, 4u)) + (uint3(LocalInvocationID).xy * uint2(4u, 1u))) - uint2(filterOffset, 0u));
  for(uint r = 0u; (r < 4u); r = (r + 1u)) {
    for(uint c = 0u; (c < 4u); c = (c + 1u)) {
      uint2 loadIndex = (baseIndex + uint2(c, r));
      if (((*(tint_symbol_4)).value != 0u)) {
        loadIndex = uint2(loadIndex).yx;
      }
      (*(tint_symbol_1))[r][((4u * LocalInvocationID[0]) + c)] = float4(tint_symbol_3.sample(tint_symbol_5, ((float2(loadIndex) + float2(0.25f)) / float2(dims)), level(0.0f))).rgb;
    }
  }
  threadgroup_barrier(mem_flags::mem_threadgroup);
  for(uint r = 0u; (r < 4u); r = (r + 1u)) {
    for(uint c = 0u; (c < 4u); c = (c + 1u)) {
      uint2 writeIndex = (baseIndex + uint2(c, r));
      if (((*(tint_symbol_4)).value != 0u)) {
        writeIndex = uint2(writeIndex).yx;
      }
      uint const center = ((4u * LocalInvocationID[0]) + c);
      if ((((center >= filterOffset) && (center < (256u - filterOffset))) && all((writeIndex < dims)))) {
        float3 acc = float3(0.0f);
        for(uint f = 0u; (f < (*(tint_symbol_2)).filterDim); f = (f + 1u)) {
          uint i = ((center + f) - filterOffset);
          acc = (acc + ((1.0f / float((*(tint_symbol_2)).filterDim)) * (*(tint_symbol_1))[r][i]));
        }
        tint_symbol_6.write(float4(acc, 1.0f), uint2(writeIndex));
      }
    }
  }
}

kernel void tint_symbol(const constant Params* tint_symbol_8 [[buffer(0)]], texture2d<float, access::sample> tint_symbol_9 [[texture(0)]], const constant Flip* tint_symbol_10 [[buffer(1)]], sampler tint_symbol_11 [[sampler(0)]], texture2d<float, access::write> tint_symbol_12 [[texture(1)]], uint3 WorkGroupID [[threadgroup_position_in_grid]], uint3 LocalInvocationID [[thread_position_in_threadgroup]], uint local_invocation_index [[thread_index_in_threadgroup]]) {
  threadgroup tint_array<tint_array<float3, 256>, 4> tint_symbol_7;
  tint_symbol_inner(WorkGroupID, LocalInvocationID, local_invocation_index, &(tint_symbol_7), tint_symbol_8, tint_symbol_9, tint_symbol_10, tint_symbol_11, tint_symbol_12);
  return;
}