ti-runtime

Neutral Taichi runtime for MechDSL (PlanJune14, PJ-0).

This package is the seam + primitive layer of the "Seams & Bodies" architecture:

MechDSL owns the seams + primitives (this package); algo2code generates the bodies (solvers, preconditioners, constitutive updates, time integrators) from LaTeX, injected into these seams.

It contains only stable infrastructure — nothing algorithmic is hardcoded here:

Provenance & invariants

• One-time harvest (PlanJune14 D-B) — adapted from NumerixWeave (libs/tisolvers, libs/ticonstit, apps/tifem); no ongoing sync.
• Portable output (D-D) — generated artifacts depend on ti_runtime, never on mechdsl, so they can later feed NumerixWeave / MOOSE / MFEM.
• Conventions follow MechDSL dev/design_docs/07-CONVENTIONS.md: Voigt order [xx, yy, zz, xy, xz, yz], tensorial (unscaled shears), metric diag(1,1,1,2,2,2).

Operator / solver contract

A matrix-free operator is an in-place callable apply(out, x) -> None computing out = A @ x over Taichi fields. Inject it and run a (generated) solver against it:

ctx = LinearSolveContext()
ctx.set_operator(my_tangent_matvec)        # out = K(u) @ x, matrix-free @ti.kernel
ctx.set_preconditioner(DiagonalPreconditioner(diag))
# a generated PCG body calls ctx.apply_A / ctx.apply_preconditioner + vector_ops