First FOR loop is never entered. Compiler can optimize this out.
Second FOR Loop is entered but immediately terminated by "break" (possibly compiler will optimize this out). So this should take the same time as first case.
Third FOR Loop will take "t" to execute
If you're looking for loop overheads:
On GCN chips (HD77xx+) the overhead is small: 1..4 cycles. If the loop is the same for all the workitems in a wavefront, it can be realized in 1 cycle with the Scalar ALU.
On older VLIW chips it costs a clause switch which can take longer time. 10-40 cycles I guess.