Estimate number of users who can't use Translations because of the SIMD requirements
Categories
(Firefox :: Translations, task)
Tracking
()
People
(Reporter: marco, Unassigned)
References
Details
Using Telemetry data, we should figure out how many users would be left out if we don't get rid of the requirement.
Yury, in which situations exactly do we disable WASM SIMD? On Intel it's SSE < 4.1, what about ARM?
Comment 1•2 years ago
•
|
||
(In reply to Marco Castelluccio [:marco] from comment #0)
Yury, in which situations exactly do we disable WASM SIMD? On Intel it's SSE < 4.1, what about ARM?
For Aarch64, WASM SIMD is always enabled. For ARM, it is not available, based on assumption (maybe some inside information?) that we have low amount of users with ARM, and support of SIMD/Neon only available for small amount of modern chips.
Comment 2•2 years ago
|
||
I think we could write a telemetry query to see via env_build_arch
in clients_daily_v6
Reporter | ||
Comment 3•2 years ago
|
||
Here are my queries for SSE 4.1: https://sql.telemetry.mozilla.org/queries/89605 and https://sql.telemetry.mozilla.org/queries/89917.
Comment 4•2 years ago
|
||
There are users with ARM Neon extensions -- would be nice to collect this information too.
Reporter | ||
Comment 5•2 years ago
|
||
I have created a new query for Desktop, including Aarch64 too (even if that's a basically insignificant number of users): https://sql.telemetry.mozilla.org/queries/91228.
Currently, 2% of users on Windows >= 10, 20% of users on Windows < 10, 0,5% on Mac and 7% on Linux don't have access to WASM SIMD.
Regarding mobile, I've asked around to figure out what data source I can use to perform the query.
Comment 6•2 years ago
|
||
(In reply to Yury Delendik (:yury) from comment #1)
(In reply to Marco Castelluccio [:marco] from comment #0)
Yury, in which situations exactly do we disable WASM SIMD? On Intel it's SSE < 4.1, what about ARM?
For Aarch64, WASM SIMD is always enabled. For ARM, it is not available, based on assumption (maybe some inside information?) that we have low amount of users with ARM, and support of SIMD/Neon only available for small amount of modern chips.
I'm slightly confused about the ARM support. I believe all of the new MacBook chips are ARM, and transaltions works fine on my MacBook.
Am I misunderstanding something?
Comment 7•2 years ago
|
||
I believe all of the new MacBook chips are ARM, and transaltions works fine on my MacBook.
M1s are Aarch64 (and Wasm SIMD is supported for it)
Comment 8•2 years ago
|
||
I'm slightly confused about the ARM support.
The "ARM" is used here as legacy processors that 32-bit (ARMv7 or below). Though some ARMv8 can be forced to run in 32-bit -- we count them as "ARM".
For sake of this discussion, it is better to refer ARM processors that run in 64-bit mode as Aarch64 (or ARM64), which is they are.
Reporter | ||
Comment 9•2 years ago
|
||
And here's a query for Fenix: https://sql.telemetry.mozilla.org/queries/91248#225920.
We have around 10% of users on ARM 32 bit with NEON.
Reporter | ||
Comment 10•2 years ago
|
||
We now have numbers, we can make a choice whether we want to exclude these users and keep things as they are, or support all users and thus expand WASM SIMD support or stop using WASM SIMD.
Description
•