#version 450
layout(local_size_x = 16, local_size_y = 8, local_size_z = 1) in;

layout(set = 0, binding = 0) uniform sampler2D in_depth; 
layout(set = 0, binding = 1, rg16f) writeonly uniform image2D reduced;

shared lowp float local_tile_min[gl_WorkGroupSize.x*gl_WorkGroupSize.y*2];
shared lowp float local_tile_max[gl_WorkGroupSize.x*gl_WorkGroupSize.y*2];

void main()
{
    //load into group memory
    local_tile_min[gl_LocalInvocationIndex] = min(texelFetch(in_depth, ivec2(ivec2(1, 2)*gl_GlobalInvocationID.xy), 0).x,
                                                  texelFetch(in_depth, ivec2(ivec2(1, 2)*gl_GlobalInvocationID.xy+ivec2(8, 0)), 0).x);

    local_tile_max[gl_LocalInvocationIndex] = max(texelFetch(in_depth, ivec2(ivec2(1, 2)*gl_GlobalInvocationID.xy), 0).x,
                                                  texelFetch(in_depth, ivec2(ivec2(1, 2)*gl_GlobalInvocationID.xy+ivec2(8, 0)), 0).x);
    groupMemoryBarrier();
    barrier();
    for(int i = 64; i >= 1; i /= 2) {
        if(gl_LocalInvocationIndex < i) {
            local_tile_min[gl_LocalInvocationIndex] = min(local_tile_min[gl_LocalInvocationIndex],
                                                          local_tile_min[gl_LocalInvocationIndex+i]);

            local_tile_max[gl_LocalInvocationIndex] = max(local_tile_max[gl_LocalInvocationIndex],
                                                          local_tile_max[gl_LocalInvocationIndex+i]);
            groupMemoryBarrier();
            barrier();
        }   
    }

    if(gl_LocalInvocationIndex == 0)
    {
        imageStore(reduced, ivec2(gl_WorkGroupID.xy), vec2(local_tile_min[0], local_tile_max[0]).xyxx);
    }
}