Frank14f/CelerisLab
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

test: add scaled N256 Taylor-Green diagnostic

Browse Source
This commit is contained in:
Codex committed 2026-10-18 17:00:55 -07:00
1 parent eec4677db1
commit 92b8afeeed
3 files changed
+249 -25
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

tests/validation/taylor_green/assess_projected_pairs.py
+105 -14
View File
@@ -1,6 +1,7 @@
"""New-source four-case repeat of 11901a7; frozen oracle and gates are unchanged."""
import argparse
import hashlib
import json
import math
from pathlib import Path
@@ -8,16 +9,25 @@ from pathlib import Path
from assess import BASELINE_FACTOR, FLOORS
from oracle import LIMITS, TIMES, U, rectangles
from run_case import (INITIALIZATION, sample_steps, NUMERICAL_REVISION, SOURCE_SHA256,
BUILD_CAPTURE_SHA256, REFERENCE_HARNESS)
BUILD_CAPTURE_SHA256, REFERENCE_HARNESS, SCALED_DIAGNOSTIC)
# Host-generated snapshots of a135252 at the frozen two local configurations.
# Host-generated snapshots of a135252 at the frozen local configurations.
# These describe expected inputs; actual sidecars are captured before cleanup.
BUILD_SOURCES = {
128: "803177416b7599ab89cc5afd35cf4a28a673b2f4bb8e8a6a1abc0b6d77610873",
256: "17ec5e60f87b840d3d13b8778bc8a3aecf485469c5a061af46c5cc19b73396cf",
512: "4e85e38098858ec95c137d1b1e96846a58d9cf3cc242b349a3ad7c8b8e3ce0aa",
}
CASES = ((100, 128, 1, "constant"), (100, 256, 1, "constant"),
(100, 128, 2, "constant"), (100, 128, 2, "projected_linear"))
REFERENCE_REVISION = "2cf9853952e52f59bea63cf5c1d3a1cd76e53eed"
# Audited 2cf9853 scalar receipts, archived at evidence commit 90849a55cad4a0da9b72747da7bfe6eb75962a44.
REFERENCE_RECEIPTS = {
"uniform128": "724efafc267eaa59dcc59e8d4c16177b67f6c6a08fd2732a481833237bb19cee",
"uniform256": "3c312a99ebf747a135674dd481b77455a3d2c9a406ca388f30524b37dad9bbe9",
"constant128": "2ee28efd1e81b2e28b7902b6c4f95bb2c1ea4e5f0ec484638fa5ae83dca34bba",
"projected128": "ee67b1d4d90a048c0d3d1a16447c6eab3d89c94d151f791306e373f96dd8b963",
}
NUMERICS = dict(lattice_model="D2Q9", collision="SRT", store_precision="FP32",
ddf_shifting=True, streaming="double_buffer", les_enabled=False,
boundary="periodic", block_shape=[8, 8],
@@ -73,7 +83,11 @@ def valid_builds(record, n, depth):
return True
def assess(records, expected_revision=None):
def assess(records, expected_revision=None, *, diagnostic=None):
if diagnostic not in (None, SCALED_DIAGNOSTIC):
raise ValueError("Unknown diagnostic")
base_n = 256 if diagnostic else 128
cases = tuple((re, n * base_n // 128, depth, scheme) for re, n, depth, scheme in CASES)
indexed, duplicates = {}, []
for record in records:
key = tuple(record.get(k) for k in ("re_L", "n", "depth", "explosion"))
@@ -81,9 +95,11 @@ def assess(records, expected_revision=None):
duplicates.append(str(key))
else:
indexed[key] = record
missing = [str(k) for k in CASES if k not in indexed]
checks = dict(complete_matrix=not missing, no_extra_cases=set(indexed) <= set(CASES),
missing = [str(k) for k in cases if k not in indexed]
checks = dict(complete_matrix=not missing, no_extra_cases=set(indexed) <= set(cases),
no_duplicate_cases=not duplicates, finite_receipts=finite_tree(records))
if diagnostic or any("diagnostic" in r for r in records):
checks["diagnostic_identity"] = all(r.get("diagnostic") == diagnostic for r in records)
sources = [r.get("source", {}) for r in records]
checks["same_source_and_harness"] = bool(sources) and len({
json.dumps(s, sort_keys=True) for s in sources}) == 1
@@ -106,9 +122,11 @@ def assess(records, expected_revision=None):
) and len(set(devices)) == 1
enabled = [r.get("early_pressure", {}).get("enabled") for r in records]
checks["same_early_sampling"] = bool(enabled) and all(type(v) is bool for v in enabled) and len(set(enabled)) == 1
if diagnostic:
checks["diagnostic_early_sampling"] = all(v is True for v in enabled)
shared_valid = all(checks.values())
complete, case_valid, early = {}, {}, []
for key in CASES:
for key in cases:
if key not in indexed:
continue
r = indexed[key]
@@ -135,7 +153,7 @@ def assess(records, expected_revision=None):
(("module", "__init__.py"), ("compiler", "cuda/compiler_v2.py"), ("kernel_root", "lbm/kernels")))
expected_patch = [] if depth == 1 else rectangles(n, depth)
checks[label + " patch"] = json.dumps(r.get("refinements")) == json.dumps(expected_patch)
cells = [n*n] if depth == 1 else [8704, 30720]
cells = [n*n] if depth == 1 else [17*n*n//32, 15*n*n//8]
hops = [16*n * 2**l for l in range(depth)]
work = r.get("workload", {})
checks[label + " workload"] = (
@@ -168,19 +186,22 @@ def assess(records, expected_revision=None):
comparisons, scheme_results = [], {}
for scheme in ("constant", "projected_linear"):
key = (100, 128, 2, scheme)
key = (100, base_n, 2, scheme)
for metric, floor in FLOORS.items():
amr, coarse, fine = (worst(k, metric) for k in (key, CASES[0], CASES[1]))
amr, coarse, fine = (worst(k, metric) for k in (key, cases[0], cases[1]))
budget = None if coarse is None else max(BASELINE_FACTOR * coarse, floor)
budget = budget if finite(budget) else None
ratio = None if amr is None or fine is None else amr / max(fine, 1e-30)
passed = amr is not None and budget is not None and amr <= budget
checks[f"baseline {scheme}/{metric}"] = passed
comparisons.append(dict(explosion=scheme, metric=metric, amr=amr,
uniform128=coarse, baseline_budget=budget, passed=passed,
uniform256=fine, amr_to_uniform256=ratio if finite(ratio) else None))
baseline_budget=budget, passed=passed,
**{f"uniform{base_n}": coarse, f"uniform{2*base_n}": fine,
f"amr_to_uniform{2*base_n}": ratio if finite(ratio) else None}))
if diagnostic:
comparisons[-1]["floor_dominates"] = None if coarse is None else floor >= BASELINE_FACTOR * coarse
baseline_passed = all(checks[f"baseline {scheme}/{metric}"] for metric in FLOORS)
valid = shared_valid and all(case_valid.get(k, False) for k in (CASES[0], CASES[1], key))
valid = shared_valid and all(case_valid.get(k, False) for k in (cases[0], cases[1], key))
scheme_results[scheme] = dict(passed=valid and baseline_passed,
valid_case_and_controls=valid, baseline_passed=baseline_passed)
return dict(passed=all(checks.values()), missing=missing, duplicates=duplicates, checks=checks,
@@ -190,19 +211,89 @@ def assess(records, expected_revision=None):
floors=FLOORS, baseline_factor=BASELINE_FACTOR,
main_statistic="maximum of four post-initial samples at t*=.12,.24,.36,.48",
early_statistic="sampled pressure maximum only, t*=k*.001875, k=0..64; not an all-step or continuous maximum",
finest_control="uniform256 and AMR/ uniform256 ratios are descriptive, not acceptance gates",
finest_control=f"uniform{2*base_n} and AMR/ uniform{2*base_n} ratios are descriptive, not acceptance gates",
scope="2D periodic Re100 smooth-decay diagnostic; no new algorithm, convergence-order or general AMR qualification")
def load_reference(paths):
"""Use the four audited historical files verbatim, never relabel them fresh controls."""
blobs = [path.read_bytes() for path in paths]
hashes = [hashlib.sha256(blob).hexdigest() for blob in blobs]
if len(hashes) != 4 or set(hashes) != set(REFERENCE_RECEIPTS.values()):
raise ValueError("Historical N128 receipt hashes differ from the audited 2cf9853 evidence")
return [json.loads(blob) for blob in blobs]
def resolution_trend(records, reference, expected_revision):
"""Signed changes are descriptive; they never replace the frozen acceptance budgets."""
old = assess(reference, REFERENCE_REVISION)
new = assess(records, expected_revision, diagnostic=SCALED_DIAGNOSTIC)
# Numerical screen failures still have useful raw errors; only malformed inputs block trends.
valid = all(v for result in (old, new) for k, v in result["checks"].items()
if not k.startswith("baseline ") and not k.endswith(" absolute"))
valid = valid and all(maximum(r["rows"], m) is not None for r in reference + records for m in LIMITS)
oracle = reference[0]["source"]["harness_sha256"].get("oracle.py") if valid else None
valid = valid and bool(oracle) and records[0]["source"]["harness_sha256"].get("oracle.py") == oracle
if not valid:
return dict(valid=False, per_time=[], worst_post_initial=[], uniform256_bridge=[])
key = lambda r: (r["re_L"], r["n"], r["depth"], r["explosion"])
before, after = {key(r): r for r in reference}, {key(r): r for r in records}
def change(role, metric, low, high, **extra):
ratio = high / low if low else None
percent = 100*(ratio-1) if ratio is not None else None
return dict(role=role, metric=metric, n128_campaign_error=low, n256_campaign_error=high,
signed_change=high-low, new_to_old=ratio if finite(ratio) else None,
ratio_unavailable_reason=None if finite(ratio) else "zero historical error" if not low else "overflow",
change_percent=percent if finite(percent) else None, **extra)
per_time, worst, bridge = [], [], []
for old_key, role in zip(CASES, ("uniform_root", "uniform_fine", "constant", "projected_linear")):
re, n, depth, scheme = old_key
low, high = before[old_key], after[(re, 2*n, depth, scheme)]
for metric in LIMITS:
per_time.extend(change(role, metric, a[metric], b[metric], tstar=a["tstar"])
for a, b in zip(low["rows"], high["rows"]))
worst.append(change(role, metric, maximum(low["rows"][1:], metric),
maximum(high["rows"][1:], metric)))
for a, b in zip(before[CASES[1]]["rows"], after[CASES[1]]["rows"]):
bridge.extend(dict(metric=m, tstar=a["tstar"], historical=a[m], fresh=b[m], signed_change=b[m]-a[m])
for m in LIMITS)
low, high = before[CASES[1]], after[CASES[1]]
bridge_identity = dict(main_rows_identical=low["rows"] == high["rows"],
early_pressure_rows_identical=low["early_pressure"]["rows"] == high["early_pressure"]["rows"])
same_runtime = reference[0]["runtime_identity"] == records[0]["runtime_identity"]
return dict(valid=True, reference_revision=REFERENCE_REVISION, reference_receipt_sha256=REFERENCE_RECEIPTS,
reference_scheme_results=old["scheme_results"], per_time=per_time, worst_post_initial=worst,
uniform256_bridge=bridge, uniform256_bridge_identity=bridge_identity,
same_recorded_runtime=same_runtime,
trend_interpretation_ready=all(bridge_identity.values()) and same_runtime,
scope="Historical same-production/oracle reference versus fresh controls; same-placement two-resolution trend only, not convergence order or phase robustness")
if __name__ == "__main__":
p = argparse.ArgumentParser(description=__doc__)
p.add_argument("files", type=Path, nargs="+")
p.add_argument("--output", type=Path, required=True)
p.add_argument("--expected-revision", required=True)
p.add_argument("--diagnostic", choices=[SCALED_DIAGNOSTIC])
p.add_argument("--reference-files", type=Path, nargs=4)
args = p.parse_args()
if args.output.exists():
p.error("Output already exists; choose a new path to retain prior evidence")
result = assess([json.loads(path.read_text()) for path in args.files], args.expected_revision)
if bool(args.diagnostic) != bool(args.reference_files):
p.error("--diagnostic scaled_n256 requires --reference-files; references are diagnostic-only")
records = [json.loads(path.read_text()) for path in args.files]
result = assess(records, args.expected_revision, diagnostic=args.diagnostic)
if args.diagnostic:
try:
reference = load_reference(args.reference_files)
except ValueError as exc:
p.error(str(exc))
result["diagnostic"] = args.diagnostic
result["resolution_trend"] = resolution_trend(records, reference, args.expected_revision)
result["checks"]["valid_historical_reference_and_trend"] = result["resolution_trend"]["valid"]
result["passed"] = all(result["checks"].values())
args.output.write_text(json.dumps(result, indent=2, allow_nan=False) + "\n")
print(json.dumps(result, indent=2, allow_nan=False))
raise SystemExit(0 if result["passed"] else 1)