Home /Modules/comparison/ARITHMETIC.SPEED.002.md

ARITHMETIC.SPEED.002.md

comparison/ARITHMETIC.SPEED.002.md · 41 line(s) · markdown

Arithmetic Comparison โ€” Common Eight Cases

Suite: SCS-PERF-ARITH-COMMON-002 Recorded: 2026-11-07 Scope: The eight shared cases present in both hosted Linux datasets: signed 64-bit integer and double, each with +, -, *, /.

1. Consolidated results

Median nanoseconds per operation (ns/op). Lower is faster within a given protocol.

CaseC11C++17Java 21+Basic NativeSLVM 1SLVM 2SLVM 3SLVM 4SLVM 5SLVM 6SLVM 7SLVM 8SLVM 9SLVM 10SLVM 11
i64 +0.6410.7760.61572.10772.53772.54272.56672.57173.07472.80772.71875.77972.80572.57681.842
i64 -0.8661.1511.10072.17872.32172.27772.51872.49872.51572.65072.18575.11172.52072.34081.706
i64 *1.1521.1530.96872.11272.51972.24172.52072.21872.18772.43672.60775.32972.29772.23881.853
i64 /2.8782.8832.88772.26072.48572.41072.32372.29772.25572.31872.30975.53072.23772.33282.468
f64 +1.4391.1531.01880.55280.88180.65680.48680.57980.51880.73880.95482.88880.58180.50687.858
f64 -1.4411.1511.10779.12479.68279.70279.80979.71079.74679.55479.65681.71779.73479.60887.165
f64 *1.4411.1531.00679.76880.46579.91079.85479.97179.88979.91679.95082.72179.74980.11987.606
f64 /1.3871.2871.46982.92083.05783.37883.05782.89782.99383.16583.10384.79783.08083.18391.017

2. Comparability qualification

This is a common-coverage consolidation, not yet a fully equivalent timing comparison. It aligns the semantic cases, but the existing source measurements use different harnesses and timing boundaries:

Do not calculate cross-language or native-versus-VM speed ratios from this table. Compare C11/C++17/Java within their own dataset, and compare Basic Native/SLVM generations within theirs. Matching labels do not make the ns/op measurements interchangeable.

3. Validation and provenance

4. Next genuinely equivalent iteration

A true matched comparison must rerun C11, C++17, Java 21+, Basic Native, and Premium SLVM 1โ€“11 under one protocol: identical deterministic operand arrays, signed 64-bit and double operations, equal measured operation counts, equivalent warm-up, common checksum validation, aligned timing boundaries, medians, and sample spread. Raw samples should be retained. This report deliberately does not claim timing equivalence until that run exists.