uniform shader image;
uniform float2 imageSize;
uniform float amount;
uniform float passthrough;
half4 main(float2 coord) {

  float3x3 G[2];
  G[0] = mat3(1.0, 2.0, 1.0, 0.0, 0.0, 0.0, -1.0, -2.0, -1.0);
  G[1] = mat3(1.0, 0.0, -1.0, 2.0, 0.0, -2.0, 1.0, 0.0, -1.0);

  // There used to be a bug in this shader which caused us to always use 1/512 as the 
  // relative texel size. Using correct pixel offsets instead results in far better
  // outputs for lower resolution images but also makes the outline a lot less pronounced
  // in higher resolution images. In order to mitigate this, we now always offset at
  // least 1/512 of the image size so that the output is the same as before for higher resolution
  // images.
  float2 texelScale = max(float2(1.0), imageSize / float2(512.0));

  /* fetch the 3x3 neighbourhood and use the RGB vectors length as intensity value */
  float3x3 I;
  for (float i = 0.0; i < 3.0; i++) {
    for (float j = 0.0; j < 3.0; j++) {
      float3 probe = unpremul(image.eval(coord + texelScale * float2(i - 1, j - 1))).rgb;
      I[int(i)][int(j)] = length(probe);
    }
  }

  /* calculate the convolution values for all the masks */
  float cnv[2];
  for (int i = 0; i < 2; i++) {
    float dp3 = dot(G[i][0], I[0]) + dot(G[i][1], I[1]) + dot(G[i][2], I[2]);
    cnv[i] = dp3 * dp3;
  }                                                                                                                        

  float4 orig = unpremul(image.eval(coord));

  float4 color = orig * passthrough + float4(0.5 * sqrt(cnv[0] * cnv[0] + cnv[1] * cnv[1])) * amount;
  color.rgb *= color.a;
  return color;
}