Question Branching in GPU architecture

[lordoftherings]

How does branching in GPU architecture work? For example if I have a simple if else statement and the execution gets serialized. If I need to run both execution parts(if and else), why don't I simply compare first? Why do threads need to run at different time during execution, why not at the same time?

 

[kerberos_20]

How does branching in GPU architecture work? For example if I have a simple if else statement and the execution gets serialized. If I need to run both execution parts(if and else), why don't I simply compare first? Why do threads need to run at different time during execution, why not at the same time?

for coding usualy stackowerflow has some answers already
https://stackoverflow.com/questions/5891082/simt-warp-question