Perf considerations when sanitizing bounds of C-style arrays

[Kalvin Lee]

Hi folks,

We'd like to enable the subject matter Chromium-wide. We were cautioned that some projects may carry perf considerations not visible from top-level benchmarks (Speedometer and friends). Is there anything we should investigate, test, or expose (e.g. build configuration) before trying to land this?

Cheers,

Kalvin

[Shahbaz Youssefi]

Hi,

For the Chromium side, I'll reach out to you internally. For ANGLE as a standalone project, we'd definitely need an off-by-default flag that Chromium can override to enable this.

[Kalvin Lee]

WRT standalone ANGLE, I think we should be safe. This would be implemented Chromium-side as a default compiler config, so unless you're directly importing Chromium's //build/config for use in standalone ANGLE, I don't think anything changes for you. (Please correct me if I'm wrong.)

[Shahbaz Youssefi]

Actually we _do_ import Chromium's //build/config, this sort of perf-degrading flag has affected us in the past a number of times.

[Kalvin Lee]

Ouch. Do you need a special GN-level escape-hatch? Or is there a specific benchmark we can run to quantify your perf hit?

[Shahbaz Youssefi]

GN-level escape-hatch! `enable_safe_libcxx` is an example of a similar flag added to ANGLE in the past: https://chromium-review.googlesource.com/c/angle/angle/+/4264463

[Kalvin Lee]

This feature is now enabled. The opt-out CLs for ANGLE will come along soon.

--
You received this message because you are subscribed to a topic in the Google Groups "angleproject" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/angleproject/b3waFJdP4Wc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to angleproject...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/angleproject/19bb8ec8-478a-4eee-8752-4473ae1b3fdbn%40googlegroups.com.