#version 430 core

layout (local_size_x = 128, local_size_y = 1, local_size_z = 1) in;

layout (r16i, binding = 0) readonly uniform iimageBuffer ImageIn;
layout (rg16i, binding = 1) writeonly uniform iimageBuffer ImageOut;

layout (location = 0) uniform int GlobalOffset;
layout (location = 1) uniform int WindowLength;

void main()
{
  int x = int(gl_GlobalInvocationID.x);

  int offset = GlobalOffset + x * WindowLength;

  vec2 sum = vec2(0, 0);

  float scale = 1.0 / float(WindowLength);
  for (int i = 0; i < WindowLength; i++) {
    ivec4 v = imageLoad(ImageIn, offset + i);
    float value = float(v.x);
    sum += vec2(clamp(value, 0.0, 32767.0), clamp(value, -32768.0, 0.0)) * scale;
  }

  imageStore(ImageOut, x, ivec4(int(sum.x), int(sum.y), 0, 0));
}