A (per-base read-and-sum per 80-step interval), W (window + step(80) +
one read) and W1 (window + 80 x step(1)) over equal simulated duration
from the same fresh state and warm-up, 3 interleaved repeats. Bitwise
populations and per-interval total impulses (documented engineering
tolerance) are compared outside timing; health checks untimed. Records
candidate source identity (revision, dirty flag, file hashes) and does not
use or modify frozen refined-DFG manifests/results.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Separate GPU class for a shifted MRT, all-blocked, stationary-body case
without Jets: window totals vs summed per-base reads (substep-scaled
tolerance), bitwise populations vs a raw-only stepper, a single step(N)
stepper with bitwise-equal window rows and populations, step(0)
non-change, and reset after a window. SRT/Jet test stays separately
scoped. Test-only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Scale the window-vs-summed-reads tolerance by the actual absolute fine
substep contributions (last_raw after every fine hop), labelled a
conservative engineering tolerance rather than a proof. Compare
populations by finite FP32 bits plus NaN masks and reject infinities.
Test-only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FixedPatchStepper(observe_window_exchange=True), which requires
observe_raw_exchange, allocates a second per-link FP64 ledger per body
level. While a window is open, the unchanged BlockAccumulateExchange
kernel also accumulates into it after each fine hop. Small API:
start_exchange_window() zeroes and opens at the current completed step;
read_exchange_window(body_id, level=) is non-destructive and reports
impulse with rates over the integer window duration. initialize/
_initialize_raw and step failure invalidate; remap builds a new instance
with no open window; Jet amplitude updates may stay in-window. Default
disabled: no allocation and no extra launch; last_substep/last_base_step
unchanged. summarize_exchange gains window='window' with duration.
Host contract tests (mocked driver) and a GPU gate against summed
per-base reads with a reordering-bound FP64 tolerance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Simulation.get_ddf() is flat (9*ny*nx); view it as (Q, cells) before
applying the fluid mask, with a flat-DDF regression. Label A/E as
runner-like force-read workloads (values discarded, no aggregation or
validation; no macro/probe/CSV cost; not benchmark throughput) and split
status lines into CONSTRUCTED and WARM_READY. Test/doc only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Untimed physical-leaf/fluid health after warm-up and after each timed
window using the existing runner limits (finite macros, rho in [0.5,1.5],
max Mach <= 0.2); refined masks as in run_refined_dfg. Parity comparator
rejects unequal level counts and any infinity; matching NaN stays allowed
only as a storage mask. Regression tests for both. No numerical gate or
production change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Explicit stream synchronization before/after each timed loop (refined
FixedPatchStepper stream, uniform Simulation stream, in their contexts);
required --max-wall-seconds outer timeout checked between variants and in
per-step loops; untimed finite leaf/fluid health checks; force notes for
B/C/E' (no host reads; final base interval only; whole-window average with
different summation order). Parity stays A/B/C vs A and E/E' vs E only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Wall-time A/B/C/E/E' variants over equal simulated duration, reusing
run_refined_dfg setup functions. Fresh state and identical warm-up per
repeat, 3 interleaved repeats, construction/JIT, warm-up and bitwise
population/NaN-mask parity checks outside timed regions. B, C and E' are
labelled timing-only (no valid force history). No numerical or
production change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Remove MRT from the FixedPatch2D rejection matrix (TRT and other
rejections retained); convert host MRT tests to stdlib unittest; add GPU
gates for unshifted dense-vs-blocked MRT two-level roots (exact bits) and
a shifted all-blocked three-level MRT smoke checking per-level omega.
No numerical source change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Additive compile-time COLLISION_MODEL == 2 branches in BlockStep,
BlockChannelStep and patch_collide repeat the uniform collide_dispatch
MRT sequence (zero forcing, omega clamp, collide_mrt_no_force). SRT
statements, including the shifted retained-delta/direct-momentum path,
are unchanged under the alternate branch. Guards accept SRT or MRT only;
TRT, LES, non-FP32, non-double-buffer and shifted dense/mixed refinement
stay rejected. Transfers, subcycling, BFL, raw ledgers, channel closures,
shifted-Jet SRT guard and PeriodicBlockStepper are untouched. Existing
MRT operator, shifted representation and fixed kinetic rates are reused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Jet gains a last field condition='velocity'. 'pressure' raises
NotImplementedError in __post_init__ before profile/amplitude validation, so a
pressure jet never reaches surface_link_velocity, tables or kernels; unknown
values raise ValueError. Amplitude stays a lattice velocity (not a flow rate);
no pressure units, field reuse, tables or kernel changes are introduced.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit fbf6f1d122)
initialize(coarse_g=..., fine_g=..., child_g=...) raised TypeError after the
switch to *level_g. A small entry adapter (_initial_levels) maps either the
positional per-level form or the original names to one per-level tuple and
rejects mixed or incomplete forms with TypeError before any state check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Schedule test drives _pairs[i].explode/coalesce and patches _advance(index,
buffer) for the stale L1->L2 feed mutant; remap rejection test patches
_boundary_hosts entries and hierarchy.pairs instead of the read-only views;
the GPU launch-failure test patches _pairs[0].explode. Assertions unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
build_hierarchy(cfg, [], wall_aware=True) left both base wall rows ACTIVE;
the refined path already marked them OUTER. Add the regression test.
Found by native review of eeb90ac.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The constructor builds lbm.hierarchy geometry, per-level layouts, kernel
configs, boundaries and leaf masks as lists, then one _build_level per level
and one _build_pair per adjacent pair. step() runs _hop(0): a child explodes
from its parent's hop source and advances twice, then the parent gathers
halos, stages receiver-local restriction, applies BFL and steps. Launch
sequence for existing two/three-level cases is unchanged; depth >= 4 works
for fully blocked hierarchies with bodies/Jets on the finest level.
New keywords refinements/level_blocks/level_objects/level_jets (L0 included);
legacy keywords are normalised by _legacy_levels and are mutually exclusive
with them. The dense two-level path stays as an isolated branch in _hop
(fused PatchCoarseStep root, dense fine BFL). Removed _child, _middle_kind
storage, _explode/_coalesce/_ghosts/_donors/_targets/_children,
_advance_child, _blocked_fine_boundary and _remap_blocks; existing public
attributes and the private names tests still read are thin read-only views.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lbm/hierarchy.py loops the existing pair logic (FixedPatch2D/FixedPatchSet2D
for L0->L1, nest_patch/nest_patch_set for deeper pairs, _middle_config for
the per-level config chain) and returns validated per-level configs, pair
tables and composed kinds for any depth. Legacy scalar window forms are
accepted so existing public shapes are unchanged. Validates each config copy
because LBMConfig.omega is only computed by validate(). Not yet used by the
stepper.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pure move of blocked_fine_tables, blocked_kind, blocked_base_tables,
_BlockedBoundary and _prepare_blocked_boundary out of the pycuda-importing
refinement_stepper into a numpy-only module, so CPU tests can import them
directly. refinement_stepper re-exports the public names; no behaviour change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
