ChainHash at a0116ea: same-binary refresh on Xeon and M2

The chart's ChainHash and ChainHash-128 cells on the Xeon and M2 tabs are measured at thomasahle/chainhash a0116ea. The AMD EPYC 9R14 (Zen 4) tab measures the same revision (records/zen4), so all three tabs show one revision of each function.

Since the previously published cells, the headers gained only exact-output speed changes:

The calibration header is opt-in and is not used here. Both functions are unchanged: the verification values are the repository's, 0x66672BD6 (ChainHash) and 0x1FCA728C (ChainHash-128), on every registration and host.

Each binary links two revisions of the headers side by side: a0116ea and 0a03c63 as a control. PolyXOR128 and XXH3-128 run in the same binary.

Registration Xeon bulk Xeon 1–31 B M2 bulk M2 1–31 B
chainhash.a0116ea 28.25 80.38 30.15 89.20
chainhash.0a03c63 28.28 120.14 30.20 89.16
chainhash-128.a0116ea 14.90 129.30 16.85 158.25
chainhash-128.0a03c63 14.92 146.38 16.70 159.34
polyxor-128 15.58 172.16 16.71 142.51
XXH3-128 19.94 34.11 12.55 30.97

These replace the previously published cells:

Row Xeon (bulk / 1–31 B) M2 (bulk / 1–31 B)
ChainHash 28.31 / 155.14 26.26 / 87.49
ChainHash-128 14.43 / 175.93 10.26 / 167.89

Method

Files

File What it is
speeds_chainhash_a0116ea.json Selected cells, every pass, spreads, per-length 1–31 B costs, load at launch
evidence/Xeon/, evidence/M2/ Raw --test=Speed logs (NAME.runN.txt), run metadata (NAME.runN.json), Sanity logs, gate log, build provenance
bench/ Registration sources, link script, runner and summarizer