bitshifter / glam-rs / 32665367086
99%

DEFAULT BRANCH: main
Ran
Jobs 4
Files 153
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 99.106% (-0.01%) from 99.119%
32665367086

push

github

web-flow 
ci: gungraun baseline comparisons and expanded benchmark coverage (#788)

## What

Expands the gungraun benchmark suite and adds CI infrastructure for
instruction-count regression testing against committed baselines.

### Benchmark coverage (from 36 to 66 benchmarks)

- New `bench_affine3a` and `bench_vec3` groups
- New `Quat` benches: `dot`, `from_axis_angle`, `from_euler`, `inverse`,
`mul_vec3`, `lerp`, `slerp`
- New `Mat4` benches: `from_quat`, `from_scale_rotation_translation`,
`transform_point3`, `transform_vector3`; `Mat3A::from_quat`
- New `Vec3A` benches: `lerp`, `normalize_or_zero`; `Vec4::mul_vec4`
- Branchy functions get one bench per code path with inputs chosen far
from branch thresholds: `Quat::slerp` (orthogonal / negative-dot /
nearly-parallel), `Quat::lerp` (positive / negative dot), `Vec3A::slerp`
(general / anti-parallel / parallel), `Vec3A::normalize_or_zero` (zero /
nonzero)
- Cache simulation is disabled (`cache-sim=no`): for these one-shot
micro-benchmarks the cache metrics only contained noise from cold
instruction fetches (dependent on linker layout, not the measured code),
which also made Estimated Cycles unreliable. Instruction counts are
deterministic and are now the tracked metric.

### Baselines and CI

- Baselines are committed in `benches/gungraun-baselines`, tracked per
backend: `sse2` and `scalar_math` on x86_64 Linux, `neon` on aarch64
Linux (valgrind doesn't support macOS, and counts are only comparable
per toolchain and triple).
- New ci tool command: `cargo run --release -p ci -- bench` compares
against baselines with `--callgrind-limits=ir=0%`; `--save` regenerates
the baselines for the host's backends (and prunes stale ones).
- New `Bench` workflow checks PRs on `ubuntu-latest` (sse2, scalar_math)
and `ubuntu-24.04-arm` (neon).
- New manually dispatched `Update gungraun baselines` workflow
regenerates all backends and opens a PR, for use after intentional perf
changes and toolchain bumps. The bench toolchain is ... (continued)

91337 of 92161 relevant lines covered (99.11%)

126567.0 hits per line

Coverage Regressions

Lines Coverage ∆ File
10
94.08
 -2.04% src/f32/coresimd/quat.rs
5
73.68
 -13.16% src/coresimd.rs
3
95.95
 -2.03% src/bool/coresimd/bvec4a.rs
3
99.19
 -0.81% src/f32/coresimd/mat2.rs
3
98.0
 -0.5% src/f32/coresimd/mat3a.rs
3
99.05
 -0.29% src/u16/u16vec2.rs
Jobs
ID Job ID Ran Files Coverage
1 native_simd-macos-latest - 32665367086.1 121
99.01
 GitHub Action Run
2 scalar_math-ubuntu-latest - 32665367086.2 120
99.04
 GitHub Action Run
3 native_simd-ubuntu-latest - 32665367086.3 121
99.02
 GitHub Action Run
4 core_simd-ubuntu-latest - 32665367086.4 120
98.96
 GitHub Action Run
Source Files on build 32665367086