// smooths the ambient occlusion of gtao_main.comp: the average of the 4x4 pixels around each one, which takes out the
// repeating 4x4 pattern of the slices, with the pixels at a clearly different distance left out so the occlusion does
// not leak across the edges of objects. one invocation per pixel of the half size target

layout(local_size_x = 8, local_size_y = 8) in;

layout(binding = 0) uniform sampler2D depth_tex;
layout(binding = 1) uniform sampler2D ao_tex;
layout(binding = 0, r8) writeonly uniform image2D ao_target;

layout(location = 0) uniform mat4 inv_projection;

// relative difference of the distance at which a neighbour stops counting
const float DEPTH_TOLERANCE = 0.05;

float view_distance(ivec2 texel)
{
	ivec2 full_size = textureSize(depth_tex, 0);
	float depth = texelFetch(depth_tex, min(texel * 2, full_size - 1), 0).r;
	if (depth <= 0.0)
		return 1e30;
	vec2 uv = (vec2(texel * 2) + 0.5) / vec2(full_size);
	vec4 position = inv_projection * vec4(uv * 2.0 - 1.0, depth, 1.0);
	return length(position.xyz / position.w);
}

void main()
{
	ivec2 texel = ivec2(gl_GlobalInvocationID.xy);
	ivec2 size = imageSize(ao_target);
	if (texel.x >= size.x || texel.y >= size.y)
		return;

	float distance = view_distance(texel);
	float sum = 0.0;
	float weights = 0.0;
	for (int y = -2; y < 2; ++y)
	{
		for (int x = -2; x < 2; ++x)
		{
			ivec2 sampletexel = clamp(texel + ivec2(x, y), ivec2(0), size - 1);
			float sampledistance = view_distance(sampletexel);
			float weight = clamp(1.0 - abs(sampledistance - distance) / (DEPTH_TOLERANCE * distance), 0.0, 1.0);
			sum += texelFetch(ao_tex, sampletexel, 0).r * weight;
			weights += weight;
		}
	}
	// the pixel itself is in the window with the full weight, so weights >= 1
	imageStore(ao_target, texel, vec4(sum / weights));
}