Stream: t-compiler/wg-nll

Topic: those-who-would-sacrifice-completeness-for-performance


nikomatsakis (May 10 2018 at 13:49, on Zulip):

@pnkfelix you saw https://github.com/rust-lang/rust/pull/50593 I suppose

lqd (May 10 2018 at 14:03, on Zulip):

(just in case niko, today is a bank holiday in France)

nikomatsakis (May 10 2018 at 14:04, on Zulip):

oh, right, forgot

pnkfelix (May 11 2018 at 09:35, on Zulip):

this is an amazing topic title. Though perhaps it works better the other way?

pnkfelix (May 11 2018 at 09:39, on Zulip):

(after all, non-performant systems don't get used? There's some pithy quote, not sure if its Butler Lampson or someone else, about how the problem with fault-tolerance as applied to bugs is that tolerated bugs don't get fixed...)

nikomatsakis (May 11 2018 at 10:33, on Zulip):

ha, a fair point.

nikomatsakis (May 15 2018 at 12:17, on Zulip):

@pnkfelix you saw the perf results?

nikomatsakis (May 15 2018 at 12:17, on Zulip):

lookin' decent, though still room for improvement

pnkfelix (May 15 2018 at 12:19, on Zulip):

no i've been distracted

pnkfelix (May 15 2018 at 12:59, on Zulip):

hmm so we saw at best a 60% improvement

nikomatsakis (May 15 2018 at 12:59, on Zulip):

yeah, though I think we can get more

pnkfelix (May 15 2018 at 12:59, on Zulip):

this was from your PR to remove the per-location computationm?

pnkfelix (May 15 2018 at 12:59, on Zulip):

I would have expected much more improvement

pnkfelix (May 15 2018 at 12:59, on Zulip):

hmm

nikomatsakis (May 15 2018 at 12:59, on Zulip):

well, I didn't change the data structures or anything

nikomatsakis (May 15 2018 at 12:59, on Zulip):

it's still e.g. a bitset

nikomatsakis (May 15 2018 at 13:00, on Zulip):

I'm not sure but maybe SEME gives a win

nikomatsakis (May 15 2018 at 13:00, on Zulip):

not sure

pnkfelix (May 15 2018 at 13:00, on Zulip):

right, but the per-location stuff is the cause of quadratic blowup, right?

nikomatsakis (May 15 2018 at 13:00, on Zulip):

not really

pnkfelix (May 15 2018 at 13:00, on Zulip):

I guess I'm over-estimating the effect from asymptotic inefficiency

nikomatsakis (May 15 2018 at 13:00, on Zulip):

or at least there are many potential causes

nikomatsakis (May 15 2018 at 13:00, on Zulip):

I mean we still have O(n) regions and O(n) points per region

nikomatsakis (May 15 2018 at 13:00, on Zulip):

so in some sense ther eis still a O(n^2)

pnkfelix (May 15 2018 at 13:00, on Zulip):

or I guess I'm misunderstanding how many causes of blow up there are. :)

pnkfelix (May 15 2018 at 13:01, on Zulip):

I guess that is true

pnkfelix (May 15 2018 at 13:01, on Zulip):

okay

Last update: Nov 22 2019 at 00:00UTC