rustyverse/rustytorch
2
0
Fork 0
Code Issues Pull Requests Actions 18 Packages Projects Releases Wiki Activity

embedded3 S2-2b: the moving path's host work 5.0 s → 0.6 s per step at 1.45 M cells, bit-identical — Problem::diagonals (the per-cell link diagonal was cells × links: 1.8 s twice per step), counting CSR + rayon in the device tables, rayon in the cut geometry builder; RTX_E3_MOVING_PROFILE laps
CI / Build CPU-Only (Explicit) (push) Failing after 4s
Details
CI / Clippy Check (push) Failing after 5s
Details
Documentation / Build User Guide (push) Successful in 5s
Details
CI / Format Check (push) Failing after 11s
Details
Documentation / Build API Documentation (push) Failing after 1m34s
Details
Performance Benchmarks / Run Benchmarks (push) Successful in 1m54s
Details
CI / Build (ubuntu-latest) (push) Failing after 2m11s
Details
CI / Build (macos-latest) (push) Canceled after 0s
Details
CI / Test (macos-latest) (push) Canceled after 0s
Details
CI / Test (ubuntu-latest) (push) Canceled after 0s
Details
CI / Python Bindings (maturin) (macos-latest) (push) Canceled after 0s
Details
CI / Python Bindings (maturin) (ubuntu-latest) (push) Canceled after 0s
Details
CI / WASM Build + Size Check (push) Canceled after 0s
Details
CI / Distributed Training Tests (push) Canceled after 0s
Details
CI / CI Success (push) Canceled after 0s
Details

Browse Source 
Co-Authored-By: Claude Fable 5.1 <[email protected]>
This commit is contained in:
Omar Sobh
2026-09-18 12:48:45 -05:00
co-authored by Claude Fable 5.1
parent 99e4a7214b
commit cdd5ad0b66
7 changed files with 251 additions and 132 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/specialized/rtx-cfd/src/solvers/incompressible/embedded3/cut.rs
+96 -90
View File
@@ -59,26 +59,32 @@ impl CutGeometry {
let g = grid;
let (nx, ny, nz, dx, dy, dz) = (g.nx, g.ny, g.nz, g.dx, g.dy, g.dz);
let n_nodes = (nx + 1) * (ny + 1) * (nz + 1);
// Every loop below is per-index with read-only inputs: rayon (S2-2b),
// the same arithmetic per entry.
use rayon::prelude::*;
let mut phi = vec![0.0; n_nodes];
let mut bound = vec![0.0; n_nodes];
for k in 0..=nz {
for j in 0..=ny {
for i in 0..=nx {
let n = Self::node(g, k, j, i);
if let Some((p, band, motion)) = prev {
let b = p.bound[n] - motion;
if b > band {
phi[n] = p.phi[n];
bound[n] = b;
continue;
}
phi.par_iter_mut()
.zip(bound.par_iter_mut())
.enumerate()
.for_each(|(n, (phi_n, bound_n))| {
let (k, j, i) = (
n / ((ny + 1) * (nx + 1)),
(n / (nx + 1)) % (ny + 1),
n % (nx + 1),
);
if let Some((p, band, motion)) = prev {
let b = p.bound[n] - motion;
if b > band {
*phi_n = p.phi[n];
*bound_n = b;
return;
}
let v = body.phi(i as f64 * dx, j as f64 * dy, k as f64 * dz, t);
phi[n] = v;
bound[n] = v.abs();
}
}
}
let v = body.phi(i as f64 * dx, j as f64 * dy, k as f64 * dz, t);
*phi_n = v;
*bound_n = v.abs();
});
let corner = |k: usize, j: usize, i: usize| phi[Self::node(g, k, j, i)];
// Face apertures: a face's two triangles along the diagonal from its
// (0, 0) to its (1, 1) corner in the face's own (a, b) order — the
@@ -113,51 +119,51 @@ impl CutGeometry {
let mut d_u = vec![0.0; (nx + 1) * ny * nz];
let mut d_v = vec![0.0; nx * (ny + 1) * nz];
let mut d_w = vec![0.0; nx * ny * (nz + 1)];
for k in 0..nz {
for j in 0..ny {
for i in 0..=nx {
// x-face at i: corners (j, k), (j+1, k), (j, k+1), (j+1, k+1)
let (q00, q10, q01, q11) = (
corner(k, j, i),
corner(k, j + 1, i),
corner(k + 1, j, i),
corner(k + 1, j + 1, i),
);
a_u[g.uface(k, j, i)] = quad_fraction(q00, q10, q01, q11);
d_u[g.uface(k, j, i)] = 0.25 * (q00 + q10 + q01 + q11);
}
}
}
for k in 0..nz {
for j in 0..=ny {
for i in 0..nx {
// y-face at j: corners (i, k), (i+1, k), (i, k+1), (i+1, k+1)
let (q00, q10, q01, q11) = (
corner(k, j, i),
corner(k, j, i + 1),
corner(k + 1, j, i),
corner(k + 1, j, i + 1),
);
a_v[g.vface(k, j, i)] = quad_fraction(q00, q10, q01, q11);
d_v[g.vface(k, j, i)] = 0.25 * (q00 + q10 + q01 + q11);
}
}
}
for k in 0..=nz {
for j in 0..ny {
for i in 0..nx {
// z-face at k: corners (i, j), (i+1, j), (i, j+1), (i+1, j+1)
let (q00, q10, q01, q11) = (
corner(k, j, i),
corner(k, j, i + 1),
corner(k, j + 1, i),
corner(k, j + 1, i + 1),
);
a_w[g.wface(k, j, i)] = quad_fraction(q00, q10, q01, q11);
d_w[g.wface(k, j, i)] = 0.25 * (q00 + q10 + q01 + q11);
}
}
}
a_u.par_iter_mut()
.zip(d_u.par_iter_mut())
.enumerate()
.for_each(|(f, (a, d))| {
// x-face at i: corners (j, k), (j+1, k), (j, k+1), (j+1, k+1)
let (k, j, i) = (f / (ny * (nx + 1)), (f / (nx + 1)) % ny, f % (nx + 1));
let (q00, q10, q01, q11) = (
corner(k, j, i),
corner(k, j + 1, i),
corner(k + 1, j, i),
corner(k + 1, j + 1, i),
);
*a = quad_fraction(q00, q10, q01, q11);
*d = 0.25 * (q00 + q10 + q01 + q11);
});
a_v.par_iter_mut()
.zip(d_v.par_iter_mut())
.enumerate()
.for_each(|(f, (a, d))| {
// y-face at j: corners (i, k), (i+1, k), (i, k+1), (i+1, k+1)
let (k, j, i) = (f / ((ny + 1) * nx), (f / nx) % (ny + 1), f % nx);
let (q00, q10, q01, q11) = (
corner(k, j, i),
corner(k, j, i + 1),
corner(k + 1, j, i),
corner(k + 1, j, i + 1),
);
*a = quad_fraction(q00, q10, q01, q11);
*d = 0.25 * (q00 + q10 + q01 + q11);
});
a_w.par_iter_mut()
.zip(d_w.par_iter_mut())
.enumerate()
.for_each(|(f, (a, d))| {
// z-face at k: corners (i, j), (i+1, j), (i, j+1), (i+1, j+1)
let (k, j, i) = (f / (ny * nx), (f / nx) % ny, f % nx);
let (q00, q10, q01, q11) = (
corner(k, j, i),
corner(k, j, i + 1),
corner(k, j + 1, i),
corner(k, j + 1, i + 1),
);
*a = quad_fraction(q00, q10, q01, q11);
*d = 0.25 * (q00 + q10 + q01 + q11);
});
// Cell volumes by the Kuhn split: the six tetrahedra around the
// diagonal (0,0,0)–(1,1,1) in unit-cube coordinates.
let mut vol = vec![0.0; g.cells()];
@@ -170,36 +176,36 @@ impl CutGeometry {
[[0, 0, 0], [0, 0, 1], [1, 0, 1], [1, 1, 1]],
[[0, 0, 0], [0, 0, 1], [0, 1, 1], [1, 1, 1]],
];
for k in 0..nz {
for j in 0..ny {
for i in 0..nx {
let mut fluid = 0.0;
for tet in &KUHN {
let pts: Vec<[f64; 3]> = tet
.iter()
.map(|c| [c[0] as f64, c[1] as f64, c[2] as f64])
.collect();
let vals: Vec<f64> = tet
.iter()
.map(|c| corner(k + c[2], j + c[1], i + c[0]))
.collect();
fluid += tet_fluid_volume(&pts, &vals);
}
// The six tets fill the unit cube (volume 1).
let idx = g.cell(k, j, i);
vol[idx] = fluid;
let ax = dy * dz;
let ay = dx * dz;
let az = dx * dy;
// Outward normals of the cell's faces times their fluid area,
// summed; the wall closes the fluid part of the cell.
let sx = (a_u[g.uface(k, j, i + 1)] - a_u[g.uface(k, j, i)]) * ax;
let sy = (a_v[g.vface(k, j + 1, i)] - a_v[g.vface(k, j, i)]) * ay;
let sz = (a_w[g.wface(k + 1, j, i)] - a_w[g.wface(k, j, i)]) * az;
wall[idx] = [-sx, -sy, -sz];
}
let cell_fluid = |k: usize, j: usize, i: usize| -> f64 {
let mut fluid = 0.0;
for tet in &KUHN {
let pts: Vec<[f64; 3]> = tet
.iter()
.map(|c| [c[0] as f64, c[1] as f64, c[2] as f64])
.collect();
let vals: Vec<f64> = tet
.iter()
.map(|c| corner(k + c[2], j + c[1], i + c[0]))
.collect();
fluid += tet_fluid_volume(&pts, &vals);
}
}
fluid
};
let (ax, ay, az) = (dy * dz, dx * dz, dx * dy);
vol.par_iter_mut()
.zip(wall.par_iter_mut())
.enumerate()
.for_each(|(idx, (v, w))| {
let (k, j, i) = (idx / (ny * nx), (idx / nx) % ny, idx % nx);
// The six tets fill the unit cube (volume 1).
*v = cell_fluid(k, j, i);
// Outward normals of the cell's faces times their fluid area,
// summed; the wall closes the fluid part of the cell.
let sx = (a_u[g.uface(k, j, i + 1)] - a_u[g.uface(k, j, i)]) * ax;
let sy = (a_v[g.vface(k, j + 1, i)] - a_v[g.vface(k, j, i)]) * ay;
let sz = (a_w[g.wface(k + 1, j, i)] - a_w[g.wface(k, j, i)]) * az;
*w = [-sx, -sy, -sz];
});
Self {
grid,
phi,