Add vectorisation to some suspicious blend related missing parts#1579
Open
DJMcNab wants to merge 1 commit into
Open
Add vectorisation to some suspicious blend related missing parts#1579DJMcNab wants to merge 1 commit into
DJMcNab wants to merge 1 commit into
Conversation
Collaborator
|
Back then, IIRC I purposefully decided not to inline blending-related methods to reduce potential code bloat, and since I assumed that blending itself is so slow that inlining itself won't give us that much benefit. Definitely open to changing this, but yes we should try to 1) check the impact on binary size and 2) see whether it actually yields any performance improvements. Though the vectorize calls definitely make sense! I presume especially on x86 this should yield improvements. |
Member
Author
|
All of the methods I changed (in theory) wouldn't be optimised properly/at all on x86. |
Collaborator
|
Will take a look when I find the time, thanks! |
pull Bot
pushed a commit
to Mu-L/vello
that referenced
this pull request
May 22, 2026
Mostly generated with codex, but I did look at it myself and make some adjustments, so I hope it's good now. Since we do have quite a few tests for blending (both manual ones as well as via COLR), not too concerned about correctness issues here. Note that this does not address linebender#1579 yet so it's possible this won't have much effect on AVX2. However, on NEON I'm seeing 4x-5x speedups for blending now: ``` fine/blend/normal_u8_neon time: [40.096 ns 40.297 ns 40.521 ns] change: [-1.1748% +0.8265% +3.1392%] (p = 0.45 > 0.05) No change in performance detected. Found 8 outliers among 100 measurements (8.00%) 3 (3.00%) high mild 5 (5.00%) high severe fine/blend/multiply_u8_neon time: [976.39 ns 978.78 ns 981.34 ns] change: [-80.790% -80.728% -80.664%] (p = 0.00 < 0.05) Performance has improved. fine/blend/screen_u8_neon time: [1.0140 µs 1.0167 µs 1.0199 µs] change: [-80.496% -80.407% -80.320%] (p = 0.00 < 0.05) Performance has improved. Found 1 outliers among 100 measurements (1.00%) 1 (1.00%) high mild fine/blend/overlay_u8_neon time: [1.3631 µs 1.3667 µs 1.3701 µs] change: [-74.682% -74.592% -74.499%] (p = 0.00 < 0.05) Performance has improved. Found 3 outliers among 100 measurements (3.00%) 2 (2.00%) high mild 1 (1.00%) high severe fine/blend/darken_u8_neon time: [1.1359 µs 1.1385 µs 1.1412 µs] change: [-77.273% -77.197% -77.125%] (p = 0.00 < 0.05) Performance has improved. Found 3 outliers among 100 measurements (3.00%) 3 (3.00%) high mild fine/blend/lighten_u8_neon time: [1.1535 µs 1.1557 µs 1.1582 µs] change: [-77.013% -76.936% -76.857%] (p = 0.00 < 0.05) Performance has improved. Found 3 outliers among 100 measurements (3.00%) 2 (2.00%) high mild 1 (1.00%) high severe fine/blend/color_dodge_u8_neon time: [5.6951 µs 5.7070 µs 5.7195 µs] change: [+1.6232% +1.9789% +2.3529%] (p = 0.00 < 0.05) Performance has regressed. Found 4 outliers among 100 measurements (4.00%) 3 (3.00%) high mild 1 (1.00%) high severe fine/blend/color_burn_u8_neon time: [5.6208 µs 5.6334 µs 5.6466 µs] change: [+1.5668% +1.9000% +2.2646%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 100 measurements (1.00%) 1 (1.00%) high mild fine/blend/hard_light_u8_neon time: [1.3581 µs 1.3602 µs 1.3626 µs] change: [-75.426% -75.345% -75.265%] (p = 0.00 < 0.05) Performance has improved. Found 5 outliers among 100 measurements (5.00%) 4 (4.00%) high mild 1 (1.00%) high severe fine/blend/soft_light_u8_neon time: [6.0497 µs 6.0630 µs 6.0768 µs] change: [+1.2844% +1.7849% +2.2700%] (p = 0.00 < 0.05) Performance has regressed. Found 5 outliers among 100 measurements (5.00%) 3 (3.00%) high mild 2 (2.00%) high severe fine/blend/difference_u8_neon time: [1.2694 µs 1.2720 µs 1.2747 µs] change: [-75.605% -75.514% -75.423%] (p = 0.00 < 0.05) Performance has improved. Found 3 outliers among 100 measurements (3.00%) 1 (1.00%) high mild 2 (2.00%) high severe fine/blend/exclusion_u8_neon time: [1.0596 µs 1.0614 µs 1.0634 µs] change: [-80.316% -80.250% -80.184%] (p = 0.00 < 0.05) Performance has improved. Found 5 outliers among 100 measurements (5.00%) 2 (2.00%) high mild 3 (3.00%) high severe fine/blend/hue_u8_neon time: [8.5128 µs 8.5387 µs 8.5659 µs] change: [+1.5041% +1.9534% +2.4143%] (p = 0.00 < 0.05) Performance has regressed. Found 5 outliers among 100 measurements (5.00%) 5 (5.00%) high mild fine/blend/saturation_u8_neon time: [8.5693 µs 8.6052 µs 8.6431 µs] change: [+1.7844% +2.2460% +2.6872%] (p = 0.00 < 0.05) Performance has regressed. Found 3 outliers among 100 measurements (3.00%) 2 (2.00%) high mild 1 (1.00%) high severe fine/blend/color_u8_neon time: [7.5338 µs 7.5591 µs 7.5869 µs] change: [+0.7948% +1.2293% +1.6653%] (p = 0.00 < 0.05) Change within noise threshold. Found 3 outliers among 100 measurements (3.00%) 3 (3.00%) high mild fine/blend/luminosity_u8_neon time: [7.5325 µs 7.5531 µs 7.5772 µs] change: [+0.8659% +1.3864% +1.8684%] (p = 0.00 < 0.05) Change within noise threshold. Found 6 outliers among 100 measurements (6.00%) 3 (3.00%) high mild 3 (3.00%) high severe ```
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.
In writing #1578, I stumbled upon these missing instances.
I'm aware of the comment at
vello/sparse_strips/vello_cpu/src/fine/highp/mod.rs
Lines 339 to 342 in 6d23bf5
I probably won't have time to drive this further (i.e. do the benchmarking myself), so feel free to push to this as appropriate.