Tim Sweeney: Eliminating C++ exceptions gave 15% speedup

twitter.com

18 points·JabavuAdams··4 comments

Quite an improvement. However, i feel like that the tweet also displays the (sometimes emotional) feelings toward exceptions that a large part of the C++ community has.

Herb Sutter made an excellent presentation about this "fragmentation" of the community and steps towards fixing this split:

https://www.youtube.com/watch?v=ARYP83yNAWk

Other people (e.g. Bjarne) report a 13% speedup by switching to exceptions, away from checking and propagating status codes. It seems hard to get a speedup by not using exceptions unless your code was littered with catch blocks, and throws from normal system events.

The extra code to check and handle status codes creates places for latent bugs. It's usually very poorly exercised, and it obscures the logic for the normal, non-failure case.

My favored way to fix bugs is to delete the code they are in. I don't even need to know where they were.

The code was already written without exceptions in this case, it was more mysterious compiler behaviour apparently. As for which is faster, people writing games don't tend to spend a lot of time validating data in real time, and thus have a constant factor to lose and not much to gain

Thank you, that puts the report in context. I can't get the tweet to come up... does it identify the compiler?

Reply on news.ycombinator.com