BENCHMARKS

Numbers, not vibes.

Real output from sabas-os-benchmarks via JMH — 2 forks, 3 warmup + 8 measurement iterations at 500ms each. All throughput figures in million ops/sec.

AMD Ryzen 9 5900HX - 16 threads OpenJDK 25 GraalVM CE JMH 1.37 Linux 6.9 - bare metal Throughput mode - M ops/s

Layout Engine

Hit-test, deep tree

LayoutBenchmark.hitTestDeepTree

35.4

M ops/s

+/- 0.45

Recompute layout, deep tree

LayoutBenchmark.layoutDeepTree

10.4

M ops/s

+/- 1.68

CPU Rasteriser - SIMD (Vector API)

Opaque fill, 2560 px scanline

CpuRasterBenchmark.simdFillOpaque2560

6.63

M ops/s

+/- 0.24

Alpha blend, 2560 px scanline

CpuRasterBenchmark.simdBlendScanline2560

0.25

M ops/s

+/- 0.01

EventBus - Publish Throughput

Publish, 1 subscriber

ConfigAndEventBusBenchmark.publishOneSub

3.08

M ops/s

+/- 0.57

Publish, 10 subscribers

ConfigAndEventBusBenchmark.publishTenSubs

0.33

M ops/s

+/- 0.08

PTY Transport - 64 KB chunks

Read, 64 KB chunk

PtyTransportBenchmark.readChunk64k

0.40

M ops/s

+/- 0.01

Write, 64 KB chunk

PtyTransportBenchmark.writeChunk64k

0.10

M ops/s

+/- 0.02