Open Bug 1984610 Opened 1 year ago Updated 1 day ago

Evaluate potential performance gains of moving to ssse3, sse4.2 or avx2 as baseline

Categories

(Firefox Build System :: General, task)

task

Tracking

(Not tracked)

People

(Reporter: sergesanspaille, Assigned: sergesanspaille)

References

Details

Changing the base hardware requirements have a lot of impact on users, so this is just a study: what are the potential gain of raising the bar?

Assignee: nobody → sguelton
Summary: Evanuate potential performance gains of moving to ssse3, sse4.2 or avx2 as baseline → Evaluate potential performance gains of moving to ssse3, sse4.2 or avx2 as baseline

Runs

All run made 20x , speedometer3, linux.

Results

Slight loss with ssse3 (-0.27%, probably in the noise), slight gain with sse4.2 (+0.42%), and a decent gain with avx2 (plus some extra flags like fma just to prevent compilation failure) (+1%)

Please keep in mind that speedometer3 does not execute the network stack at all. I’m still wondering if we would see gains in higher-bandwidth scenarios where we are currently performing poorly especially over QUIC.

Thanks for looking into this. If the Linux runners are still on datacenter-oriented CPUs, it would be great to see results also for laptop/desktop CPUs. (My previous attempt at getting gains from SSSE3 showed results that suggested a difference between datacenter CPUs vs. non-datacenter CPUs.)

See Also: → CVE-2026-2760
You need to log in before you can comment on or make changes to this bug.