ClaudeandClaude Opus 5.5
2aadffe183
Each physical link belongs once to the block owning its fluid receiver core; the reflected source population is written into that block's own core/halo copy (no halo copy between BFL and pull). Rejects repeated links and repeated local writers; one row per link so force counts once. Documents stale remote solid-core slots vs unique core ownership, and that the 4-cell refinement halo is separate. Static geometry only. Tests: address equivalence incl. cross-block q<.5 donors, and static BFL+pull bit-equal to the dense operator on fluid cells. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
248 lines
11 KiB
Python
248 lines
11 KiB
Python
"""CPU block layout contracts: ownership, halo pairs, packing and tagged pull."""
|
|
import sys
|
|
import unittest
|
|
|
|
import numpy as np
|
|
|
|
from CelerisLab.body.geometry.base import CutLink
|
|
from CelerisLab.body.geometry.circle import CircleGeometry
|
|
from CelerisLab.lbm.blocks import CORNER, FACE, NO_DONOR, NQ, BlockLayout2D
|
|
from CelerisLab.lbm.descriptors import D2Q9_EX, D2Q9_EY
|
|
|
|
CASES = ( # (ny, nx, by, bx, halo)
|
|
(6, 8, 3, 4, 1), # 2x2 split
|
|
(6, 12, 3, 4, 1), # 2x3 split
|
|
(8, 8, 4, 4, 2), # wider halo
|
|
(4, 8, 4, 4, 1), # one block along y
|
|
)
|
|
PERIODIC = ((True, True), (False, False), (True, False), (False, True))
|
|
|
|
|
|
def tags(ny, nx):
|
|
"""Unique, exactly representable value per dense population slot."""
|
|
return np.arange(NQ * ny * nx, dtype=np.float64).reshape(NQ, ny, nx)
|
|
|
|
|
|
def dense_pull(a, periodic):
|
|
"""f[q](x) = a[q](x - c_q); NaN where a non-periodic exterior is crossed."""
|
|
px, py = periodic
|
|
_, ny, nx = a.shape
|
|
gy, gx = np.mgrid[0:ny, 0:nx]
|
|
out = np.empty_like(a)
|
|
for q in range(NQ):
|
|
sy, sx = gy - D2Q9_EY[q], gx - D2Q9_EX[q]
|
|
valid = ((py | ((sy >= 0) & (sy < ny))) & (px | ((sx >= 0) & (sx < nx))))
|
|
out[q] = np.where(valid, a[q, sy % ny, sx % nx], np.nan)
|
|
return out
|
|
|
|
|
|
def blocked_pull(layout, blocked):
|
|
"""Exchange, then pull every core cell from its own block storage only."""
|
|
layout.exchange(blocked)
|
|
out = np.full_like(blocked, np.nan)
|
|
(by, bx), h = layout.block, layout.halo
|
|
for q in range(NQ):
|
|
ey, ex = D2Q9_EY[q], D2Q9_EX[q]
|
|
out[:, q, h:h + by, h:h + bx] = blocked[:, q, h - ey:h - ey + by, h - ex:h - ex + bx]
|
|
return out
|
|
|
|
|
|
def opp(q):
|
|
return q + 1 if q % 2 else q - 1
|
|
|
|
|
|
def reflect(q, toward, other):
|
|
"""Static linear Bouzidi stand-in; equivalence holds for any read->write map."""
|
|
if q < .5:
|
|
return 2 * q * toward + (1 - 2 * q) * other
|
|
return toward / (2 * q) + (2 * q - 1) / (2 * q) * other
|
|
|
|
|
|
def dense_bfl(a, links):
|
|
out = a.copy()
|
|
_, ny, nx = a.shape
|
|
for link in links:
|
|
y, x = divmod(link.fluid_idx, nx)
|
|
q, d = link.q, link.dir
|
|
if q < .5 and link.fallback == 0:
|
|
other = a[d, (y - D2Q9_EY[d]) % ny, (x - D2Q9_EX[d]) % nx]
|
|
elif q < .5:
|
|
other = 0.0
|
|
else:
|
|
other = a[opp(d), y, x]
|
|
out[opp(d), (y + D2Q9_EY[d]) % ny, (x + D2Q9_EX[d]) % nx] = reflect(q, a[d, y, x], other)
|
|
return out
|
|
|
|
|
|
def blocked_bfl(layout, blocked, links, mapped):
|
|
"""Reads and writes only the receiver block's storage; no exchange inside."""
|
|
flat = blocked.ravel()
|
|
source = flat.copy()
|
|
for link, receiver, donor, write in zip(links, mapped.receiver, mapped.donor, mapped.write):
|
|
q, d = link.q, link.dir
|
|
toward = source[int(layout.slot(receiver, d))]
|
|
if donor != NO_DONOR:
|
|
other = source[int(layout.slot(donor, d))]
|
|
elif q < .5:
|
|
other = 0.0
|
|
else:
|
|
other = source[int(layout.slot(receiver, opp(d)))]
|
|
flat[int(write)] = reflect(q, toward, other)
|
|
|
|
|
|
def seam_circle(n=16):
|
|
"""Circle on the shared corner of a 2x2 split (integer-node frame)."""
|
|
geometry = CircleGeometry(n / 2 - .5, n / 2 - .5, 3.3)
|
|
links = geometry.build_cut_links(n, n)
|
|
gy, gx = np.mgrid[0:n, 0:n]
|
|
solid = (gx - (n / 2 - .5)) ** 2 + (gy - (n / 2 - .5)) ** 2 < 3.3 ** 2
|
|
return links, solid
|
|
|
|
|
|
class BlockLayoutTests(unittest.TestCase):
|
|
def test_cpu_import_boundary(self):
|
|
self.assertFalse(any(name == "pycuda" or name.startswith("pycuda.") for name in sys.modules))
|
|
|
|
def test_origins_and_unique_core_ownership(self):
|
|
layout = BlockLayout2D.build(6, 8, 3, 4)
|
|
np.testing.assert_array_equal(layout.origins, [[0, 0], [0, 4], [3, 0], [3, 4]])
|
|
self.assertEqual(layout.blocked_shape, (4, NQ, 5, 6))
|
|
for ny, nx, by, bx, h in CASES:
|
|
layout = BlockLayout2D.build(ny, nx, by, bx, halo=h)
|
|
ly, lx = layout.local_shape
|
|
core = [b * ly * lx + (y + h) * lx + x + h
|
|
for b in range(layout.nblocks) for y in range(by) for x in range(bx)]
|
|
self.assertEqual(sorted(layout.owner.ravel().tolist()), sorted(core))
|
|
for y in range(ny):
|
|
for x in range(nx):
|
|
b = (y // by) * (nx // bx) + x // bx
|
|
self.assertEqual(layout.block_cell(b, y, x), layout.owner[y, x])
|
|
|
|
def test_halo_pairs_are_owner_core_to_unique_halo(self):
|
|
for ny, nx, by, bx, h in CASES:
|
|
for periodic in PERIODIC:
|
|
layout = BlockLayout2D.build(ny, nx, by, bx, halo=h, periodic=periodic)
|
|
core = set(layout.owner.ravel().tolist())
|
|
dst = layout.halo_dst.tolist()
|
|
self.assertEqual(dst, sorted(dst))
|
|
self.assertEqual(len(dst), len(set(dst)))
|
|
self.assertTrue(core.isdisjoint(dst))
|
|
self.assertTrue(set(layout.halo_src.tolist()) <= core)
|
|
self.assertTrue(set(layout.halo_kind.tolist()) <= {FACE, CORNER})
|
|
full = BlockLayout2D.build(6, 8, 3, 4)
|
|
per_block = np.bincount(full.halo_kind, minlength=2) // full.nblocks
|
|
np.testing.assert_array_equal(per_block, [2 * (3 + 4), 4])
|
|
closed = BlockLayout2D.build(6, 8, 3, 4, periodic=(False, False))
|
|
# Each 2x2 block keeps one inner face column, one face row and one corner.
|
|
self.assertEqual(len(closed.halo_dst), 4 * (3 + 4 + 1))
|
|
|
|
def test_pack_round_trip_and_exterior_stays_invalid(self):
|
|
for ny, nx, by, bx, h in CASES:
|
|
for periodic in PERIODIC:
|
|
layout = BlockLayout2D.build(ny, nx, by, bx, halo=h, periodic=periodic)
|
|
a = tags(ny, nx)
|
|
blocked = layout.to_blocked(a)
|
|
np.testing.assert_array_equal(layout.to_dense(blocked), a)
|
|
filled = np.isfinite(blocked[:, 0]).sum()
|
|
self.assertEqual(filled, ny * nx + len(layout.halo_dst))
|
|
flat = blocked.ravel()
|
|
for q in (0, 5):
|
|
np.testing.assert_array_equal(flat[layout.slot(layout.halo_dst, q).astype(np.intp)],
|
|
flat[layout.slot(layout.halo_src, q).astype(np.intp)])
|
|
|
|
def test_tagged_pull_matches_dense_pull(self):
|
|
for ny, nx, by, bx, h in CASES:
|
|
for periodic in PERIODIC:
|
|
with self.subTest(case=(ny, nx, by, bx, h), periodic=periodic):
|
|
layout = BlockLayout2D.build(ny, nx, by, bx, halo=h, periodic=periodic)
|
|
a = tags(ny, nx)
|
|
got = layout.to_dense(blocked_pull(layout, layout.to_blocked(a)))
|
|
np.testing.assert_array_equal(got, dense_pull(a, periodic))
|
|
|
|
def test_block_cell_resolves_periodic_halo_images(self):
|
|
layout = BlockLayout2D.build(6, 8, 3, 4)
|
|
ly, lx = layout.local_shape
|
|
# Block 0 sees x=7 (wrapped) in its west halo and y=5 in its south halo.
|
|
self.assertEqual(layout.block_cell(0, 0, 7), 1 * lx + 0)
|
|
self.assertEqual(layout.block_cell(0, 5, 0), 0 * lx + 1)
|
|
closed = BlockLayout2D.build(6, 8, 3, 4, periodic=(False, False))
|
|
with self.assertRaises(ValueError):
|
|
closed.block_cell(0, 0, 7)
|
|
with self.assertRaises(ValueError):
|
|
layout.block_cell(0, 4, 0) # two rows past the one-cell halo
|
|
|
|
def test_invalid_inputs_rejected(self):
|
|
for args, kw in (((6, 8, 4, 4), {}), ((6, 8, 3, 4), {"halo": 0}),
|
|
((6, 8, 3, 4), {"halo": 4}), ((6, 8, 3, 4), {"periodic": (1, 1)}),
|
|
((6, 8, 3, True), {}), ((6, 8.0, 3, 4), {})):
|
|
with self.assertRaises(ValueError):
|
|
BlockLayout2D.build(*args, **kw)
|
|
layout = BlockLayout2D.build(6, 8, 3, 4)
|
|
with self.assertRaises(ValueError):
|
|
layout.to_blocked(np.zeros((NQ, 6, 8), dtype=np.int64))
|
|
with self.assertRaises(ValueError):
|
|
layout.to_dense(np.zeros((4, NQ, 5, 5)))
|
|
|
|
def test_cut_links_are_receiver_local_and_unique(self):
|
|
links, _ = seam_circle()
|
|
for periodic in ((True, True), (False, False)):
|
|
layout = BlockLayout2D.build(16, 16, 8, 8, periodic=periodic)
|
|
mapped = layout.map_cut_links(links)
|
|
self.assertEqual(len(mapped.write), len(links)) # one row, one force per link
|
|
self.assertEqual(len(set(mapped.write.tolist())), len(links))
|
|
self.assertEqual(len(set(mapped.block.tolist())), 4) # seams are crossed
|
|
cells = int(np.prod(layout.local_shape))
|
|
core = set(layout.owner.ravel().tolist())
|
|
a = tags(16, 16)
|
|
flat = layout.to_blocked(a).ravel()
|
|
halo_reads = 0
|
|
for link, b, r, d, w in zip(links, mapped.block, mapped.receiver,
|
|
mapped.donor, mapped.write):
|
|
y, x = divmod(link.fluid_idx, 16)
|
|
self.assertEqual(r, layout.owner[y, x])
|
|
self.assertEqual(int(r) // cells, b)
|
|
self.assertEqual(int(w) // (NQ * cells), b)
|
|
sy, sx = y + D2Q9_EY[link.dir], x + D2Q9_EX[link.dir]
|
|
self.assertEqual(flat[int(w)], a[opp(link.dir), sy, sx])
|
|
if d != NO_DONOR:
|
|
self.assertEqual(int(d) // cells, b)
|
|
self.assertEqual(flat[int(layout.slot(d, link.dir))],
|
|
a[link.dir, y - D2Q9_EY[link.dir], x - D2Q9_EX[link.dir]])
|
|
halo_reads += int(d) not in core
|
|
self.assertGreater(halo_reads, 0) # some q<.5 donors come from a neighbor's core
|
|
|
|
def test_static_bfl_pull_matches_dense_on_fluid_cells(self):
|
|
links, solid = seam_circle()
|
|
a = tags(16, 16) / 7.0
|
|
expected = dense_pull(dense_bfl(a, links), (True, True))
|
|
layout = BlockLayout2D.build(16, 16, 8, 8)
|
|
blocked = layout.to_blocked(a) # gather halos
|
|
blocked_bfl(layout, blocked, links, layout.map_cut_links(links))
|
|
# Pull core cells without another halo copy (it would undo halo writes).
|
|
out = np.full_like(blocked, np.nan)
|
|
(by, bx), h = layout.block, layout.halo
|
|
for q in range(NQ):
|
|
ey, ex = D2Q9_EY[q], D2Q9_EX[q]
|
|
out[:, q, h:h + by, h:h + bx] = blocked[:, q, h - ey:h - ey + by, h - ex:h - ex + bx]
|
|
got = layout.to_dense(out)
|
|
np.testing.assert_array_equal(got[:, ~solid], expected[:, ~solid])
|
|
|
|
def test_cut_link_rejections(self):
|
|
layout = BlockLayout2D.build(16, 16, 8, 8, periodic=(False, False))
|
|
link = CutLink(fluid_idx=5 * 16 + 5, dir=1, q=.6, body_id=0)
|
|
with self.assertRaises(ValueError):
|
|
layout.map_cut_links([link, link])
|
|
with self.assertRaises(ValueError):
|
|
layout.map_cut_links([CutLink(fluid_idx=15, dir=1, q=.6, body_id=0)])
|
|
with self.assertRaises(ValueError):
|
|
layout.map_cut_links([CutLink(fluid_idx=5 * 16 + 5, dir=0, q=.6, body_id=0)])
|
|
|
|
def test_tables_are_readonly(self):
|
|
layout = BlockLayout2D.build(6, 8, 3, 4)
|
|
for array in (layout.origins, layout.owner, layout.halo_dst, layout.halo_src, layout.halo_kind):
|
|
self.assertFalse(array.flags.writeable)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|