#include <metal_stdlib>
using namespace metal;

struct Params {
	uint count;
	float scale;
};

struct DataBuffer {
	array<float, 64> values;
};

kernel void main(uint __luma_work_local_index [[thread_index_in_threadgroup]],
	uint3 __luma_work_global_id [[thread_position_in_grid]],
	uint3 __luma_work_local_id [[thread_position_in_threadgroup]],
	uint3 __luma_work_group_id [[threadgroup_position_in_grid]],
	device Params* params [[buffer(0)]],
	device DataBuffer* data [[buffer(1)]]) {
	threadgroup float scratch[64];
	
	uint idx = __luma_work_local_index;
	uint gid = __luma_work_global_id.x;
	uint lid = __luma_work_local_id.x;
	uint group_x = __luma_work_group_id.x;
	if ((gid < params.count)) {
		scratch[idx] = (data.values[idx] * params.scale);
		threadgroup_barrier(mem_flags::mem_threadgroup);
		data.values[idx] = (scratch[idx] + static_cast<float>((lid + group_x)));
	}
}