undefined | Better HN

story

0 pointsamelius6mo ago0 comments

But if you explicitly handle exceptions using IF statements then that's overhead too, right?

0 comments

yes but i think branch prediction essentialy makes them zero overhead

That's a different type of overhead than having unwind tables. With exceptions you wouldn't need a branch after each function call at all.

ameliusOP6mo ago

But a branch that is (almost) never taken has an overhead close to the overhead of a NOP instruction, which may be negligible on modern architectures.

neonz806mo ago

The CPU can not remember an infinite number of branches. Also, many branches will increase code size. With exceptions the unwind tables and unwind code can be placed elsewhere and not take up valuable L1 cache.

1 more reply

j / k navigate · click thread line to collapse

0 comments

arbitrandomuser6mo ago

yes but i think branch prediction essentialy makes them zero overhead

neonz806mo ago

That's a different type of overhead than having unwind tables. With exceptions you wouldn't need a branch after each function call at all.

ameliusOP6mo ago

But a branch that is (almost) never taken has an overhead close to the overhead of a NOP instruction, which may be negligible on modern architectures.

neonz806mo ago

1 more reply

j / k navigate · click thread line to collapse