Isn't the efficiency here < 1/2 despite what was said in lecture?
ellenluo
Yes the efficiency is technically less than 1/2 since the branches have unequal work. I think the professor meant that each ALU only operates for half the branches since it must wait for the other branch (correct me if I'm wrong).
mishywangiepie
Would organizing fragments based on their likelihood to pass branch condition be realistic/efficient? ex: if one area being shaded is more likely to have specular shading, group them together
AnastasiaMegabit
I was thinking that too. If we are doing 128 calculations at a time, for example, is there some way to send like calculations down the same ALU based on some probability of the image. It could have a major impact when we are talking about that many calculations at a time.
Isn't the efficiency here < 1/2 despite what was said in lecture?
Yes the efficiency is technically less than 1/2 since the branches have unequal work. I think the professor meant that each ALU only operates for half the branches since it must wait for the other branch (correct me if I'm wrong).
Would organizing fragments based on their likelihood to pass branch condition be realistic/efficient? ex: if one area being shaded is more likely to have specular shading, group them together
I was thinking that too. If we are doing 128 calculations at a time, for example, is there some way to send like calculations down the same ALU based on some probability of the image. It could have a major impact when we are talking about that many calculations at a time.