Repository navigation
chore(benchmark): Samsung SM-E146B results with the GPU cache floor - #32
Merged
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
riteshshukla04
force-pushed
the
bench/m14-gpu-cache-floor-results
branch
from
November 11, 2026 08:21
f01a496 to
ace3bcc
Compare
Before and after, interleaved by view. tgfx at 10,000 paths goes from 8.8 to 21.8 FPS on SurfaceView and from 4.8 to 20.4 on TextureView; no other row moved beyond noise. The performance page gets the SM-E146B table under the S25 Edge one, which is kept, with a note that it predates the fix.
Drops the note under the S25 Edge table and the prose around the new table.
The S25 Edge table and the notes around it move out of the docs; its result files stay in apps/benchmark/bench.
riteshshukla04
force-pushed
the
bench/m14-gpu-cache-floor-results
branch
from
November 11, 2026 08:28
ace3bcc to
22c79de
Compare
Creating the 5,000 handles and attaching the scene takes 12-20 s on the SM-E146B, so the combo's 28 s clock and the runner's windows landed on setup. The clock and the parity pose now wait for the attach; every other combo starts as before.
13.6 fps on SurfaceView and 5.0 on TextureView, against skia's 53.2 and 23.2. UI-thread bound at 98 %; the run and its raw output are in the cache-floor results.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #31.
The SM-E146B matrix again with the GPU cache floor from #31. The two builds ran interleaved by view (before surface, after surface, after texture, before texture), so neither got the cooler phone.
apps/benchmark/bench/results-2026-11-11-samsung-90hz-skia310-cache-floor.md: before/after, plus the full after table.docs/content/docs/performance.mdx: the benchmark section is now the SM-E146B table alone, with a pointer toapps/benchmarkfor the method. The S25 Edge result files stay inapps/benchmark/bench.apps/benchmark/App.tsx,AnimScenes.tsx: tgfxanim-imp's 28 s clock and parity pose now start once its scene is attached. Creating the 5,000 handles and attaching the scene takes 12-20 s, so before this the runner's windows landed on setup. Every other combo starts as before.tgfx now draws 2x skia at 10,000 paths (skia 10.3 and 10.1). No other row moved beyond noise; skia's static rows are within 1.2 %.
tgfx
anim-imp, now measured (two runs, skia alternating):UI-thread bound (98 %): one worklet writing 5,000 handles per frame. skia's figures are the matrix's; its own run alongside tgfx read 59.4 and 22.3.