Lab / AUG 21 2026 / benchmark
1.024×, and not rounded up
The intra-in-P fallback measured 2.290 → 2.235 ms, a 1.024× movement. The difference is under twice the larger standard deviation, so it is recorded and marked as not a claim.
Benchmarking, Negative results, Measurement methodologyKiln- Machine
- arm64, `RELEASE_ARM64_T6050`, Darwin 25.6.0
- Commit
- UNKNOWN — the artifact records no commit
- EnableIntraInPFallback=False
- 2,290,059 ns (2.290 ms), stddev 28,141 ns
- EnableIntraInPFallback=True
- 2,235,369 ns (2.235 ms), stddev 23,746 ns
- Ratio
- 1.024×
- Difference
- 54.7 µs
The intra-in-P fallback measured 2,290,059 ns with it off and 2,235,369 ns with it on. That is 2.290 ms against 2.235 ms — a 1.024× movement, or 2.4% faster with the fallback enabled.
| Arm | mean (ms) | stddev (ns) |
|---|---|---|
| EnableIntraInPFallback=False | 2.290 | 28,141 |
| EnableIntraInPFallback=True | 2.235 | 23,746 |
I am reporting it as 1.024× rather than 1.03× or "about 3%" because the second decimal is the only place the effect is visible. Rounding it up would make a movement smaller than the noise look like a result.
I would not publish it as a result on its own. The difference of 54.7 µs is under twice the larger standard deviation of 28,141 ns, and `n` is not stated in the artifact. It is recorded because it is in the committed baseline, and marked as not a claim for exactly that reason.
The same artifact contains what a result looks like when the effect is bigger than the noise: `Satd4x4_Once` at 15.988 ns with intrinsics off against 11.182 ns with them on, a 1.43× speedup with a standard deviation of 0.038 ns on the fast arm.