Lab / / failure
A result that did not keep its sign
qpel luma interpolation through a Weft-generated kernel measured 0.99×–1.02×. The integration was declined and the work stayed on its branch.
Negative results, Measurement methodology, SIMD and intrinsicsKiln- Machine
- UNKNOWN — not stated
- Branch
- `weft-pilot/motion-sad` (left unmerged)
- Reported speedup
- 0.99× – 1.02×
- Circulating claim that does not reproduce
- ~50% faster with Weft
I integrated a Weft-generated kernel for qpel luma interpolation into Kiln and measured it. The reported speedup was 0.99× to 1.02×.
The range straddles 1.00. A delta that does not keep its sign is not a delta, and this one does not keep its sign. The integration was rejected.
The branch that outlived the experiment
The branch `weft-pilot/motion-sad` on Kiln still carries *"EXPERIMENTAL: optional whole-grid SAD provider for the exhaustive motion search"* as its HEAD commit. The work was left in a branch, not merged, so the artifact of this experiment is a branch that was never merged rather than a diff that was reverted.
A *"~50% faster with Weft"* figure circulated around this work. It does not reproduce. The measured result is 0.99×–1.02×, and I have not published the 50% figure anywhere.
| Arm | Result |
|---|---|
| qpel via Weft kernel | 0.99×–1.02× reported speedup |
| circulating figure | ~50% faster — does not reproduce |