A static step's probe gives every node its principal stresses, largest
first, beside its stress and von Mises: the eigenvalues of the same
nodal stress, from the same function the principal stress contours use.
An oracle quantity reads one number at a JSON pointer, so a benchmark
whose target is a principal stress (NAFEMS LE6) needs it in a reply.
Closes#574. Part of #97.
The thin-walled cylinder with a hemispherical cap under internal
pressure as four axisymmetric quad8 blocks: axial stress on the outer
surface at D 25.283.
The NAFEMS target, 25.86, is the axisymmetric thin-shell solution; our
solid model converges to 25.27, the value ESRD reports for its converged
axisymmetric solid. So the NAFEMS quantity keeps a 1% tolerance and is
a known failure until there is an axisymmetric shell element, and a
second quantity checks the model against the solid reference (pass,
0.05%).
Part of #97.
The Z-section cantilever under an end torque as five hex20 blocks (two
flanges, two corner blocks, the web; two elements through every wall):
axial stress at A, on the mid-surface of the lower flange at its free
edge, -109.45 against -108 (2%).
The solid model converges to about -109.7, 1.6% beyond the thin-walled
reference, as ESRD's converged p-version solid (-109.5) does. The README
also records the MITC4 mid-surface shell model of the same problem,
which converges to about -111.3 (3.1%), and why the case is the solid.
Part of #97.
LE2, the cylindrical shell bending patch test, in its two load cases
(edge moment; pressure with the balancing edge traction) on an 8 × 8
MITC4 patch: tangential stress at E 59.936 and 59.904 against 60.
LE3, the hemispherical shell with point loads, on a 32 × 32
longitude-latitude mesh (planar quad4 cells, tri3 at the pole): radial
displacement at A 0.18418 against 0.185.
Both meshes are written node by node from the generator each README
lists, with the refinement that supports its tolerance. Targets are
transcribed from the reproductions of TNSB Rev. 3 in the Abaqus 2016
Benchmarks Guide and the ESRD StressCheck NAFEMS guide.
Part of #97.
The results reply named the fluid's 79.69 °C cell as the maximum while the
app showed the face held at 80 °C. It now names that face, by body and
face as for wall heat, so a script reads what the legend shows.
Part of #157.
The coloured obstacle surface samples the fluid beside each face, and
the solid cells carry no temperature, so a face held at 80 °C took the
mean of the fluid cells around it, which mixes in the cooler ones further
out: the heated-flow sample's bracket looked 50 °C beside 79.7 °C air. A
face whose thermal condition holds it at a temperature is now drawn at
that value, and the legend's range reaches it (the fluid never quite
does). Adiabatic and flux faces keep showing the fluid beside them.
Part of #157.
A region's corners are its interior angles below 135 degrees, and an
O-grid's 45-degree cuts meet straight sides at exactly that angle. The
computed angle then rounds either way with the last bits of its points,
which differ between platforms' sin and cos and between tessellations:
one ulp in three points of the slot test's arcs, Windows' sin against
glibc's, turned a mapped region into a triangle and the whole blocking
into a different one with a corner scaled Jacobian of 0.375 for 0.707.
Angles within 1e-9 rad of the corner and reflex thresholds now count as
on them, and on a threshold is smooth (neither a corner nor reflex),
which keeps the slot's intended O-grid of 9 regions and 848 quads on
Linux, macOS and Windows. The region classification and the O-grid's
own corner tests share the rule, the slot test pins the blocking, and
its arc helper uses the mesher's portable trig.
Part of #556.
Demonstrate temperature slices and signed face-average wall heat with cool air passing an isothermal bracket. Register the original CPU-rendered gallery thumbnail and document the coarse transient setup.
Part of #157.
Route saved temperature through slices, obstacle surfaces, streamlines and both result views, with availability and model units shared by commands and the UI. Show signed obstacle heat histories per complete face area at the selected frame, leaving unrecorded times uncoloured and retaining the saved slice settings.
Part of #157.
The adapter selection built its wgpu instance with default backend
options, so WGPU_DX12_COMPILER had no effect: on DX12 there was no way to
choose between FXC and DXC when measuring them (#561: FXC takes ~8.5 s
to create the GPU solver's pipelines and runs it at half of Vulkan's
speed, DXC 0.5 s and full speed). The options now come from wgpu's own
environment variables. staticdxc is refused with an error: it needs
wgpu's static-dxc feature, which FieldCraft does not build, and wgpu
would panic on it.
Part of #561.
The collide and stream passes each moved the whole population field
through two buffers (304 bytes per cell update). The AA pattern (Bailey
et al. 2009; Wittmann et al. 2013) keeps one buffer: even steps collide
each cell in place with its directions swapped, odd steps read a cell's
incoming populations at its neighbours and write the outgoing ones where
the next step reads them, so each thread writes the addresses it read
and streaming happens inside the collision (152 bytes). A cell's
coordinates are computed once and neighbours wrap with a compare and an
add. The per-link wall rules gather their sources before a scatter
writes them, so no link overwrites another's post-collision value, and
the momentum exchange is taken in the gather of the step whose forces
are read back.
Options::streaming picks the layout; Auto, what a solve uses, streams
in place, which measured faster on every device tried: 1.5x on an RTX
4090, 1.85x on the GB10 and 1.45x on an M4 Max at periodic 128^3.
Options::profile times each kernel with timestamp queries for the
throughput test, which now runs both layouts. The verification tests
run in both layouts, with two new ones comparing fields, mass, forces
and heat with the CPU after odd and even step counts across submission
boundaries.
Part of #562.
The open boundaries summed the tangential and outgoing populations as
deviations from their weights plus the weights (2/3 and 1/6), in single
precision; at low speed the weights round away the deviations' last
digits. A low-speed case through an inlet, an outlet and a slip face
came out 2.26e-5 of the largest speed from the CPU under DXC, against
the 2e-5 the tests allow. The deviations are now summed first and the
total weight, exactly 1, added once: the same Zou-He rule, so results
change only at the round-off level.
Part of #562.
The D3Q7 scalar lattice steps on the device after the flow's step, in
its own A–B pair of buffers: BGK collision with the velocity the
momentum collision leaves in the moments buffer, pull streaming with
the links' rules read through each group's thermal condition
(anti-bounce-back for a held temperature, bounce-back for an adiabatic
face, a flux added to the bounced population), inlets at their held
temperature and outlets copying their neighbour. The populations are
stored as deviations from w_q T₀ so single precision keeps the digits
of the differences. The heat each link exchanged goes into the fourth
component of the momentum-exchange contributions and the group sums;
the buoyant acceleration −β g θ is read per cell in the collision, the
open boundaries and the moments. A temperature that stops being a
number ends the chunk with Error::Unstable like the flow's checks.
Flow exposes what the backend needs to build the device state (the
thermal settings and populations).
Verification against the CPU on lavapipe: plates held at 20 and 21
(temperatures within 1.4e-5 of the range, the wall heats within 6e-5
and both against k ΔT A / H), a heated channel with a parabolic inlet
carrying its temperature, a flux wall, a held wall and an outlet
(temperature 3e-7, heat 1e-7) and a buoyant cavity at Ra 10³
(velocity 1.3e-6, temperature 6e-7, heat 2.4e-6). The tests now fail
on a shader that does not compile instead of skipping; only a missing
device skips.
Part of #157.
A fluid domain takes `thermal` settings (the initial temperature, a
thermal condition per box face and per obstacle face, gravity with the
expansion coefficient for buoyancy) in model.fluid_domain.create and
.edit, and model.fluid_domain.thermal_boundary sets one boundary's
condition as model.fluid_domain.boundary does for the flow. A material
a domain with heat uses keeps needing its conductivity and specific
heat when it is replaced.
solve.fluid's summary adds the thermal relaxation time, the fluid's
temperature range and the heat into every wall (final and the mean
over the last quarter of the history); solve.probe gives the
temperature in the cells at points and, with `heat`, the walls' heat
histories. The form, the catalogue and docs/features.md follow.
Part of #157.
A D3Q7 passive scalar rides on the D3Q19 flow: it is advected by the
cell velocity, diffuses with the fluid's thermal diffusivity (the
relaxation time follows from k / (rho c)), and feeds back into the
momentum through a per-cell Boussinesq force when the domain has an
expansion coefficient and gravity. Walls and obstacle faces hold a
temperature (anti-bounce-back), a heat flux (added to the bounced
population) or are adiabatic (bounce-back); velocity inlets carry the
held temperature in, pressure outlets copy their neighbour.
The heat crossing each face group is accumulated during streaming and
reported in physical units (W) in the result's heat histories, next to
the force histories; frames gain a temperature field and the stats
report the thermal relaxation time.
Closed-form checks: linear conduction between held plates, a flux wall
against q H / k with the wall's heat balance, an inlet temperature
carried through a plug flow, and a periodic blob that conserves its
total and returns after one lap. The de Vahl Davis cavity (Ra 1e3 and
1e4 against the published Nusselt numbers) is there as an ignored test
because it takes minutes.
The GPU backend declines domains with heat for now and asks for the
CPU; its kernels follow.
Part of #157.
A fluid domain may carry thermal settings: the initial temperature,
gravity and the fluid's expansion coefficient for buoyancy, and a thermal
condition (adiabatic, held temperature, heat flux) on each box face and on
named obstacle faces, adiabatic unless told otherwise. The material then
needs an isotropic conductivity and a specific heat. Boundaries are set
one at a time like the flow ones.
Part of #157.
`FluidStats` gained `backend` in #535 and the fixture of the flow drawing
tests arrived with #550; merged one after the other, the fixture lacks the
field and `cargo test -p fieldcraft-engine` no longer compiles on develop.
Part of #157.
The window built a deformed copy of the model for every results plot,
which drops the geometry surfaces, so after a fluid solve the bracket
vanished and nothing else was drawn.
A fluid step's plot keeps the part as it is and draws the engine's flow
display through it (the same scene as the headless pictures): the
field on a slice, the part coloured by the flow beside it and
streamlines. The legend shows the field with its bands and the cells
of its extremes (Max/Min mark and turn the view to them); the results
panel swaps the deformation rows for the slice's normal and position,
the obstacles' colours, the streamlines and a frame slider, all
running the results.* commands. The status line describes the new
commands.
Part of #193.
After solve.fluid the results view had nothing to draw: a fluid step's
fields are per cell, the view only knew nodal fields, and the deformed
display it built dropped the geometry surfaces, so even the part
vanished.
A fluid step now has its own display (`engine::flow`, on render's
flow display data), shared by the headless pictures and the window:
the part drawn as it is, its faces coloured by the field in the fluid
beside them, the field on a slice through the domain and streamlines
seeded on the inlet face upstream of the obstacles, coloured by speed
when the legend shows velocity. It is first shown with velocity
magnitude on a slice through the middle, normal to the horizontal axis
across the flow.
New commands: results.slice {step, field: velocity|pressure, axis:
x|y|z|none, position, frame}, results.surface {step, show} and
results.streamlines {step, show, count, max_steps, seeds | line,
frame}, which replies with what it traced. results.contour takes
velocity and pressure for fluid steps (a new `pressure` field) with
the usual bands and legend clamp, and results.frame walks a fluid
step's frames. Streamlines leave the coming-soon list. Hostile
parameters are refused, and the counts are capped.
Part of #193.
A fluid step's result is a field per cell of a regular grid, not per
node, so the contour machinery had nothing to draw it with and the
viewport showed nothing after a fluid solve.
Add `render::flow`, CPU display data for such fields with no solver
dependency (render stays at L1): an axis-aligned slice through the
cells as triangles carrying cell-centred values (interpolated between
the two layers either side, outer centres moved to the domain faces,
cells inside obstacles left out), the field sampled at a surface
beside the fluid cells for colouring an obstacle, and streamlines
traced through the trilinear velocity field by the midpoint rule,
seeded on a lattice upstream of the obstacles. Seeds, steps and the
total number of points are capped so hostile counts cost a bounded
time. `contour::banded` fans band pieces into scene triangles for the
callers.
Part of #193.
Every cell did an atomicMax on the same word; a workgroup reduction first
leaves one per 256 cells. Same result, less contention on a single
address.
Part of #529.
The CPU stays the default, so every existing result and fixture is
bit-identical; "gpu" steps on the device FIELDCRAFT_GPU picks through the
registered backend, and is an error when the build has none. The desktop
app and the CLI (hence the MCP server) register fieldcraft-lbm-gpu at
start; the summary names what stepped. Hostile values are errors.
Part of #529.
A new L7 crate: the CPU scheme's collision (TRT with Guo forcing), the
per-link streaming rules (bounce-back with Ladd's term, slip reflection,
open placeholders), the Zou–He boundaries and the momentum exchange per
force group as wgpu compute shaders, stepping the flow fieldcraft_lbm's
Flow prepared, through its Backend trait: populations stay on the device,
a chunk of steps is one submission, and only frames, forces and the
stability check come back. Populations are stored as deviations from
their weights, which keeps single precision honest: the verification
cases (Poiseuille with Λ = 3/16, Couette with wall forces, Taylor–Green,
a sphere's drag, a Zou–He channel) agree with the CPU to 2–5e-6 in
velocity, 1e-6 in density and 5e-5 in force, with those tolerances stated
in the tests. The rules are applied one thread per link rather than in a
per-cell loop, which lavapipe's JIT miscompiles (and which diverges on
real GPUs anyway). Native only for now; the browser registers nothing.
Part of #529.
run() drives a Backend in chunks of steps (to the next frame, history sample
or progress check) instead of the CPU Flow directly, so a device backend
registered by an application can step the same prepared flow: Flow keeps
building the grid, links and open cells and exposes them read-only, and
implements the trait itself. solve_fluid_on picks the CPU or the registered
device backend; solve_fluid stays the CPU and is unchanged in its results.
Part of #529.
Christian Svensson