sparse strips: Bump to fearless_simd v0.4.0#1462
Merged
tomcur merged 4 commits intolinebender:mainfrom Feb 20, 2026
Merged
Conversation
This brings a nice little boost to flattening, but badly regresses analytic AA in strip rendering. I've traced the regression back to `max`/`max_precise` semantics introduced here: linebender/fearless_simd#136. I'll open a PR based on top of this one to fix the issue. Benchmarks: Flattening ``` flatten/Ghostscript_Tiger time: [180.70 µs 181.08 µs 181.46 µs] change: [-6.4035% -6.0571% -5.6921%] (p = 0.00 < 0.05) Performance has improved. Found 2 outliers among 100 measurements (2.00%) 1 (1.00%) high mild 1 (1.00%) high severe flatten/paris-30k time: [9.2019 ms 9.2259 ms 9.2511 ms] change: [-2.0521% -1.5976% -1.1616%] (p = 0.00 < 0.05) Performance has improved. Found 2 outliers among 100 measurements (2.00%) 2 (2.00%) high mild ``` Strip generation ``` render_strips/Ghostscript_Tiger_simd time: [244.27 µs 244.83 µs 245.40 µs] change: [+25.354% +25.929% +26.391%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 50 measurements (2.00%) 1 (2.00%) high severe render_strips/paris-30k_simd time: [26.936 ms 27.010 ms 27.093 ms] change: [+14.664% +15.809% +16.826%] (p = 0.00 < 0.05) Performance has regressed. Found 3 outliers among 50 measurements (6.00%) 3 (6.00%) high mild ```
Collaborator
|
Have you already tested whether that fix works? |
Member
Author
Yep! Some more info here: #simd > fearless_simd v0.4 @ 💬. |
tomcur
force-pushed
the
push-vzsvxnpmplwo
branch
from
8214859 to
37fb31e
Compare
LaurenzV
reviewed
Feb 20, 2026
Collaborator
There was a problem hiding this comment.
For testing, since we want to explicitly test the fallback, I think the safest option would be to enable the force_support_fallback feature and then use fallback, like before. This way, we can be sure that it will always use that.
Collaborator
There was a problem hiding this comment.
(This is also why you are getting the CI failures in WASM)
Member
Author
There was a problem hiding this comment.
vello_sparse_tests now requires that feature, and I've reverted the tests that explicitly requested fallback before to do that again.
Member
Author
There was a problem hiding this comment.
For vello_sparse_tests that is, maybe we also want that for tests in vello_cpu and vello_hybrid? Not sure. They currently request baseline.
Collaborator
There was a problem hiding this comment.
I think it's fine for to leave as is for those. The visreg tests are the ones where it actually matters.
Member
Author
There was a problem hiding this comment.
vello_cpu seems to mostly have been constructing Level::try_detect().unwrap_or(Level::fallback()) in its tests. vello_hybrid had a lot of explicit Level::fallback().
LaurenzV
approved these changes
Feb 20, 2026
tomcur
added a commit
to tomcur/vello
that referenced
this pull request
Feb 20, 2026
Part One of fixing the performance regression caused by bumping to `fearless_simd` 0.4 in linebender#1462. I have a Part 2 that completely fixes the regression, but that probably requires more discussion, whereas the changes here will probably be uncontroversial. Relative to `main` before bumping `fearless_simd`, this now benches as as follows on my x86 machine (i7-13700k). ``` render_strips/Ghostscript_Tiger_simd time: [214.62 µs 214.97 µs 215.34 µs] change: [+10.312% +10.793% +11.168%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 50 measurements (2.00%) 1 (2.00%) high mild render_strips/paris-30k_simd time: [24.577 ms 24.668 ms 24.763 ms] change: [+4.6786% +5.7653% +6.7171%] (p = 0.00 < 0.05) Performance has regressed. ```
tomcur
added a commit
to tomcur/vello
that referenced
this pull request
Feb 20, 2026
Part One of fixing the performance regression caused by bumping to `fearless_simd` 0.4 in linebender#1462. I have a Part 2 that completely fixes the regression, but that probably requires more discussion, whereas the changes here will probably be uncontroversial. Relative to `main` before bumping `fearless_simd`, this now benches as as follows on my x86 machine (i7-13700k). ``` render_strips/Ghostscript_Tiger_simd time: [214.62 µs 214.97 µs 215.34 µs] change: [+10.312% +10.793% +11.168%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 50 measurements (2.00%) 1 (2.00%) high mild render_strips/paris-30k_simd time: [24.577 ms 24.668 ms 24.763 ms] change: [+4.6786% +5.7653% +6.7171%] (p = 0.00 < 0.05) Performance has regressed. ```
github-merge-queue bot
pushed a commit
that referenced
this pull request
Feb 21, 2026
…min` semantics) (#1463) Part One relaxes `f32x4::min_precise` to `f32x4::min`, fixing part of the regression caused by bumping to `fearless_simd` 0.4 in #1462. I have a Part 2 that completely fixes the regression, but that probably requires more discussion, whereas the changes here will probably be uncontroversial. Relative to `main` before bumping `fearless_simd`, this now benches as as follows on my x86 machine (i7-13700k). ``` render_strips/Ghostscript_Tiger_simd time: [214.62 µs 214.97 µs 215.34 µs] change: [+10.312% +10.793% +11.168%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 50 measurements (2.00%) 1 (2.00%) high mild render_strips/paris-30k_simd time: [24.577 ms 24.668 ms 24.763 ms] change: [+4.6786% +5.7653% +6.7171%] (p = 0.00 < 0.05) Performance has regressed. ```
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
2 participants
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.
This brings a nice little boost to flattening, but badly regresses analytic AA in strip rendering.
I've traced the regression back to
max/max_precisesemantics introduced here: linebender/fearless_simd#136.I'll open a PR based on top of this one to fix the issue.
Benchmarks:
Flattening
Strip generation