Challenge taken.
I've made it super fast now (at least 6x, means it runs in 5x5 resolution with same speed as before in 2x2), and it still looks good ole lottes.
The bloom effect was really slowing things down, not to mention power functions in gamma management.
I think it should be the default version here since it's really more usable with higher output resolutions.
IGP friendly...
Edit: New version (17.5.2017). Almost 20% faster, still accurate.
Edit2: New fast version (18.5.2017). Almost 2x faster and nice looking.
Edit3: Renamed and added original version so you can have all np.