1449496 - Percentage of x86/x86_64 installed base with fast unaligned SSE2

Status: RESOLVED FIXED

(Data & BI Services Team :: DI: New request, task)

Description

[Henri Sivonen (:hsivonen)]

Goal: Understanding how our x86/x86_64 installed base is split across CPUs that have fast unaligned SSE2 loads/stores vs. CPUs that have slower unaligned SSE2 loads/stores. If slow unaligned is insignificant, we could simplify code.

With a dataset that best represents the CPU family and model data for our installed base (do we collect this as part of the hardware report? if not, crash reports have these), could you, please, run this query:

For what percentage of installs is the following true?

(cpu_family >= 0x15) OR (cpu_family == 0x6 AND cpu_model >= 0x1A AND cpu_model != 0x1C AND cpu_model != 0x36 AND cpu_model != 0x35 AND cpu_model != 0x27 AND cpu_model != 0x26)

Comment 1

[Henri Sivonen (:hsivonen)]

(In reply to Henri Sivonen (:hsivonen) from comment #0)
> For what percentage of installs is the following true?

That should have said "percentage of x86/x86_64 installs". I.e. ARM installs should not be part of the percentage divisor.

Comment 2

[Gabriele Svelto [:gsvelto]]

Telemetry pings should have this information as part of the telemetry environment field (specifically the environment.system.cpu.model and environment.system.cpu.stepping fields). application.architecture or build.architecture can be used to distinguish between ARM and x86 builds. That should be the most reliable source we have to gather this kind of statistics.

Comment 3

[Chris H-C :chutten]

I wrote https://sql.telemetry.mozilla.org/queries/55156/source#145215 just now. Double-check my conditionals, but it looks to be about an 80/20 split in release.

Comment 4

[Henri Sivonen (:hsivonen)]

Thank you! I checked the conditionals and consider this FIXED.