sots-re/verify/fpu-cw/cw_census.py
alex 849c5069fc lane F: x87 precision sensitivity measured; STATE_CHECKSUM 3.5 closed
Seven End Turns from ref-turn2.sav on VM140, six control words, whole-state
checksum on every post-turn autosave.

53-bit and 64-bit x87 give byte-identical state across all 35,394 leaves, so
an x64/SSE port computing in IEEE double has NO double-rounding budget to
preserve and floats=bits is free. Two settings do move state, each reproduced
on a repeat run:

  0x007f (24-bit)   Sys[112 "Gamma Cephei"]/Pop2/PopG/PopC 540000000 -> 540000002
  0x1a7f (round-up) Flt[34 "Beta Fleet"]/Pos/.[0] and /Pos/.[2], 1 ULP each

So the port must hold intermediates at 53 bits and use round-to-nearest --
both SSE defaults, now measured rather than assumed, each with a named
regression witness.

The briefed triple was under-powered: 0x027f is 53-bit (it differs from 0x127f
only in bit 12, infinity control, ignored since the 387) and 0x137f is 64-bit,
not a rounding change. Run as written all three come back identical, and that
would have "proved" something false on both axes that matter.

Evidence the forced word actually held: read-back at each force site plus 38
independent in-pipeline hook samples per run spanning turn phases 4, 6 and 8,
all reading the forced value. Mars::Application::Run calls
_controlfp(0x50000,0x3070300) at 0x0089f606 every frame, which is 0x127f, so
forcing at StrategyClient::EndTurn is wiped before the turn runs;
StrategyServer::BeginProcessTurn is the point that works.

Also re-confirms the End-Turn determinism oracle on engine cef889e:
bb4fd9ac... / 978041ac... unchanged.

New tools: verify/fpu-cw/cw_census.py, verify/fpu-cw/trace_bitdiff.py (the
latter exists because under a forced 24-bit word the CRT's own %g rendering
degrades, so trace text is not a valid comparison surface).
2026-09-08 05:07:53 -04:00

72 lines
2.5 KiB
Python

#!/usr/bin/env python3
"""Census of the `fpu_cw` field the shim's template hooks record on every call.
Each hook snapshots the x87 control word as it finds it, so a run's trace carries independent
samples of the word from *inside* the turn pipeline -- phase 4 (MoveFleet), phase 6
(ServerSystem::ProcessTurn) and phase 8 (ComputeBudget / ProcessResearch). That is the evidence
that a control word forced at the turn gate actually held for the duration of the turn, rather
than being silently restored partway through (which would produce identical saves and a
completely false "nothing depends on precision" conclusion).
Usage: cw_census.py <shim.trace.jsonl> [...]
"""
import gzip
import json
import sys
from collections import Counter, defaultdict
def _open(path):
op = gzip.open if path.endswith(".gz") else open
return op(path, "rt", encoding="utf-8", errors="replace")
def census(path):
per_hook = defaultdict(Counter)
first_last = {}
n = 0
with _open(path) as f:
for line in f:
line = line.strip()
if not line.startswith("{"):
continue
try:
rec = json.loads(line)
except json.JSONDecodeError:
continue
hook = rec.get("hook")
if not hook:
continue
n += 1
for arg in rec.get("args", []) or []:
if arg.get("n") == "fpu_cw":
cw = int(arg["v"])
per_hook[hook][cw] += 1
key = (hook, cw)
if key not in first_last:
first_last[key] = [rec.get("call_id"), rec.get("call_id")]
else:
first_last[key][1] = rec.get("call_id")
return n, per_hook, first_last
def main():
for path in sys.argv[1:]:
n, per_hook, first_last = census(path)
print(f"== {path} ({n} hook records)")
total = Counter()
for hook in sorted(per_hook):
for cw, count in sorted(per_hook[hook].items()):
lo, hi = first_last[(hook, cw)]
total[cw] += count
print(f" {hook:<42} cw=0x{cw:04x} x{count:<5} call_id {lo}..{hi}")
if not per_hook:
print(" (no hook recorded an fpu_cw field)")
else:
vals = ", ".join(f"0x{cw:04x} x{c}" for cw, c in sorted(total.items()))
print(f" ALL SAMPLES: {vals}")
print()
if __name__ == "__main__":
main()