Boost.Foreach performance
So, I'm loving Foreach and as I watch it spread throughout our codebase, of course it runs across the desk of our optimization obsessed team member. So, I was curious if you could address runtime performance of Foreach. I would expect it to be essentially equivalent to a hand written for loop with caching of end(). But what's the overhead of the auto_any_t, etc.? Thanks, Tom
Thomas Matelich wrote:
So, I'm loving Foreach and as I watch it spread throughout our codebase, of course it runs across the desk of our optimization obsessed team member. So, I was curious if you could address runtime performance of Foreach. I would expect it to be essentially equivalent to a hand written for loop with caching of end(). But what's the overhead of the auto_any_t, etc.?
FOREACH is ~5% slower than the equivalent hand-coded loop, but for most loops (e.g., that have non-empty loop bodies) the small overhead is lost in the noise. What overhead there is is due to an extra test/set of a boolean at each loop iteration. You can find some performance comparisons here: http://tinyurl.com/2loc29 -- Eric Niebler Boost Consulting www.boost-consulting.com The Astoria Seminar ==> http://www.astoriaseminar.com
On 7/31/07, Eric Niebler <eric@boost-consulting.com> wrote:
Thomas Matelich wrote:
So, I'm loving Foreach and as I watch it spread throughout our codebase, of course it runs across the desk of our optimization obsessed team member. So, I was curious if you could address runtime performance of Foreach. I would expect it to be essentially equivalent to a hand written for loop with caching of end(). But what's the overhead of the auto_any_t, etc.?
FOREACH is ~5% slower than the equivalent hand-coded loop, but for most loops (e.g., that have non-empty loop bodies) the small overhead is lost in the noise. What overhead there is is due to an extra test/set of a boolean at each loop iteration.
You can find some performance comparisons here: http://tinyurl.com/2loc29
Thanks for the info. /me happily evangelizes Foreach throughout the group.
participants (6)
-
Eric Niebler -
Thomas Matelich