#include <metal_stdlib>

using namespace metal;
struct Params {
  /* 0x0000 */ uint filterDim;
  /* 0x0004 */ uint blockDim;
};
struct Flip {
  /* 0x0000 */ uint value;
};
struct tint_array_wrapper_1 {
  float3 arr[256];
};
struct tint_array_wrapper {
  tint_array_wrapper_1 arr[4];
};

kernel void tint_symbol(texture2d<float, access::sample> tint_symbol_3 [[texture(1)]], sampler tint_symbol_4 [[sampler(0)]], texture2d<float, access::write> tint_symbol_5 [[texture(2)]], uint3 WorkGroupID [[threadgroup_position_in_grid]], uint3 LocalInvocationID [[thread_position_in_threadgroup]], uint local_invocation_index [[thread_index_in_threadgroup]], constant Params& params [[buffer(1)]], constant Flip& flip [[buffer(3)]]) {
  threadgroup tint_array_wrapper tint_symbol_2;
  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_2.arr[i_1].arr[i_2] = float3();
  }
  threadgroup_barrier(mem_flags::mem_threadgroup);
  uint const filterOffset = ((params.filterDim - 1u) / 2u);
  int2 const dims = int2(tint_symbol_3.get_width(0), tint_symbol_3.get_height(0));
  int2 const baseIndex = as_type<int2>((as_type<uint2>(int2(((WorkGroupID.xy * uint2(params.blockDim, 4u)) + (LocalInvocationID.xy * uint2(4u, 1u))))) - as_type<uint2>(int2(int(filterOffset), 0))));
  for(uint r = 0u; (r < 4u); r = (r + 1u)) {
    for(uint c = 0u; (c < 4u); c = (c + 1u)) {
      int2 loadIndex = as_type<int2>((as_type<uint2>(baseIndex) + as_type<uint2>(int2(int(c), int(r)))));
      if ((flip.value != 0u)) {
        loadIndex = loadIndex.yx;
      }
      tint_symbol_2.arr[r].arr[((4u * LocalInvocationID.x) + c)] = tint_symbol_3.sample(tint_symbol_4, ((float2(loadIndex) + float2(0.25f, 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)) {
      int2 writeIndex = as_type<int2>((as_type<uint2>(baseIndex) + as_type<uint2>(int2(int(c), int(r)))));
      if ((flip.value != 0u)) {
        writeIndex = writeIndex.yx;
      }
      uint const center = ((4u * LocalInvocationID.x) + c);
      if ((((center >= filterOffset) && (center < (256u - filterOffset))) && all((writeIndex < dims)))) {
        float3 acc = float3(0.0f, 0.0f, 0.0f);
        for(uint f = 0u; (f < params.filterDim); f = (f + 1u)) {
          uint i = ((center + f) - filterOffset);
          acc = (acc + ((1.0f / float(params.filterDim)) * tint_symbol_2.arr[r].arr[i]));
        }
        tint_symbol_5.write(float4(acc, 1.0f), uint2(writeIndex));
      }
    }
  }
  return;
}