The Ember part of Speedometer has some small-ish functions that we Ion-compile more eagerly but these functions then inline larger functions so we're stuck in IonBuilder for a while. Increasing our small-function limit seems to be a nice win and doesn't regress anything so we should try that first.
This patch uses the normal warmup threshold for small functions too. I think it's worth trying to see how it affects AWFY. If it doesn't regress anything we could later simplify the code a bit, if it does regress something we can back out or try something else.
Assignee: nobody → jdemooij
Status: NEW → ASSIGNED
Attachment #8845893 - Flags: review?(hv1989)
Attachment #8845893 - Flags: review?(hv1989) → review+
Pushed by firstname.lastname@example.org: https://hg.mozilla.org/integration/mozilla-inbound/rev/230e0c324669 Change Ion warmup threshold for small functions to be the same as the one for other functions. r=h4writer
Pushed by email@example.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/4e048aaa479f followup - Add array-isArray-proxy-recursion.js to cgc-jstests-slow.txt. r=red
This regressed SunSpider by a few ms, but it improved Speedometer by ~4 points (7% or so) and the react-shell test also got 5-9% faster.
Seems the link between perfherder and awfy is cut. Fixing now to get those results. But I think this is an acceptable trade-off. I did try optimizations levels to reinstate this, but I keep getting regressions on sunspider, since we compile more. (IonBuilder).
You need to log in before you can comment on or make changes to this bug.