#version 310 es
layout(local_size_x = 64) in;

layout(set = 0, binding = 0, std430) readonly buffer SSBO0
{
	float inputs[];
};

layout(set = 0, binding = 1, std430) writeonly buffer SSBO1
{
	float outputs[];
};

shared float tmp[gl_WorkGroupSize.x];

void main()
{
	uint local = gl_LocalInvocationIndex;
	uint work_group = gl_WorkGroupID.x;

	// Does a trivial parallel reduction through shared memory.
	tmp[local] = inputs[work_group * gl_WorkGroupSize.x * 2u + local] + inputs[work_group * gl_WorkGroupSize.x * 2u + local + gl_WorkGroupSize.x];
	memoryBarrierShared();
	barrier();

	for (uint limit = 32u; limit > 1u; limit >>= 1u)
	{
		if (local < limit)
			tmp[local] = tmp[local] + tmp[local + limit];
		memoryBarrierShared();
		barrier();
	}

	if (local == 0u)
		outputs[work_group] = tmp[0] + tmp[1];
}