[BradyPlanden]

Description

Adds Symbol.to_rust(graph, rust_symbols=None), which lowers a discretised expression tree into a pybamm.rust.ExprGraph. Each class implements _to_rust, and a subtree shared within the tree is lowered once, cached on its id. Nothing calls it yet; #5960 uses it to compile a model's functions.

• Leaves: Scalar, Time, InputParameter (at its expected width), and StateVector/StateVectorDot, with several slices concatenated.
• Array: lowers as dense values, or as a dense matrix when it has more than one column. A sparse array stays a CSR matrix as the left operand of a matrix product, copied into canonical order so that each row sums in ascending column order, as CasADi does. scipy's stored order, which can be unsorted, left residual noise near roundoff that stalled IDA on the DFN.
• Operators:
  • arithmetic, Power, Inner, Minimum/Maximum, Modulo, Hypot, the Heavisides, Equality and MatrixMultiplication. Modulo truncates like CasADi's fmod, so for a negative operand it matches CasADi rather than evaluate's floored modulo. At a tie, Minimum/Maximum and the max/min reductions take the left operand's or first entry's derivative, as Symbol.jac does, where CasADi splits it 0.5/0.5;
  • Negate, AbsoluteValue, Sign, Floor, Ceiling and Index;
  • NumpyConcatenation, DomainConcatenation, SparseStack, Conditional and VectorField;
  • ExplicitTimeIntegral, DiscreteTimeSum and NotConstant lower their child.
• Functions:
  • the 13 SpecificFunctions, and Arcsinh2 in the same sign-and-hypot form as CasADi;
  • RegPower through the _decomposition that evaluate and CasADi share (fix: make CasADi's RegPower and uniform-grid interpolants match evaluate #5955), so for exponents 0.5 and 2 with delta > 0 Rust agrees with both bitwise; with delta = 0 all three take power(base, exponent).
  • A generic Function cannot be lowered, so this relies on fix: keep a function's class when copying it without simplification #5954 keeping a copied function's class.
• Interpolants:
  • 1D linear, with one interpolant per column for vector-valued data. A repeated interior knot lowers as a step, so step profiles solve on Rust; a table Rust refuses, such as one repeating its first knot, raises pybamm.ModelError naming convert_to_format = 'casadi';
  • 1D cubic and pchip, from the same PPoly coefficients evaluate uses;
  • 2D and 3D linear and cubic.
  • extrapolate=False is ignored, as on CasADi.

Unsupported nodes raise when they are lowered:

• TypeError, naming the class and convert_to_format = 'casadi', for KroneckerProduct, TensorProduct, Transpose, Magnitude, Component, Norm, spatial operators and a generic Function;
• NotImplementedError for a strided Index, 1D interpolant data with more than two dimensions, and interpolants in four or more dimensions.

Type of change

New feature, under ## Features in CHANGELOG.md.

Key checklist

• No style issues: pre-commit run
• Unit tests pass (N), as do the integration tests (N)
• Tests added: 152 Python test functions.
  • Lowering:
    • each node lowered and evaluated against evaluate;
    • RegPower bitwise equal across numpy, CasADi MX and SX, and Rust, for exponents 0.5 and 2 at scales 1 and 33133;
    • interpolants in 1D to 3D;
    • Jacobians against CasADi;
    • the errors for unsupported nodes, and step interpolants and refused tables;
    • RegPower with delta = 0, and Modulo with negative operands, against CasADi;
    • the cross-graph guard;
    • max/min subgradients, and abs, sign, Minimum, Maximum and the reductions at ties.
  • Parity: Rust against CasADi on toy models, and on the SPM rhs and named outputs, at rtol 1e-8 to 1e-12.
  • Numerics: multi-slice state vectors, a sparse product summed in column order, a repeated 1+1D DomainConcatenation, and dense-row colouring of the SPMe "voltage as a state" Jacobian.
  • Trajectories and pickling: trajectory methods on lists, float32 time grids and a strided array, and pickled graphs keeping input indices and widths.
  • Fixtures: dual_backend_compare is in tests/conftest_rust.py; feat(rust): evaluate rust models through compiled evaluators #5960 adds rust_backend there, once "rust" is a format.

Rust execution engine stack: proposal #5667, showcase #5732

Each PR is one commit on the one above it.

#	PR	Adds
0	#5954	Function.create_copy keeps the function's class
0b	#5955	CasADi's RegPower and uniform-grid interpolant lookups match evaluate
1	#5956	pybamm-core: expression graph, typed IR, interpreters, forward-mode AD
2	#5957	pybamm-core: CompiledModel, coloured Jacobians, the C ABI
3	#5958	pybamm.rust bindings, built into the pybamm wheel
4	#5959	Symbol.to_rust ← this PR
5	#5960	convert_to_format = "rust" for ScipySolver, AlgebraicSolver and EIS
6	#5961	pybammsolvers: IDAKLU over the C ABI
7	#5962	IDAKLUSolver on rust models
8	#5963	variables and sensitivities read in Rust
9	#5964	pybamm-core: diffsol DAE solver
10	#5965	pybamm.DiffsolSolver
11	#5966	"rust" as the default convert_to_format

[codecov]

Codecov Report

❌ Patch coverage is 90.80882% with 25 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (stack/03-pybamm-rust-packaging@2a667be). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...s/pybamm/src/pybamm/expression_tree/interpolant.py	87.95%	10 Missing ⚠️
...ges/pybamm/src/pybamm/expression_tree/functions.py	89.13%	5 Missing ⚠️
...ybamm/src/pybamm/expression_tree/concatenations.py	88.23%	2 Missing ⚠️
...mm/src/pybamm/expression_tree/discrete_time_sum.py	50.00%	2 Missing ⚠️
...bamm/src/pybamm/expression_tree/unary_operators.py	93.10%	2 Missing ⚠️
...ackages/pybamm/src/pybamm/expression_tree/array.py	95.23%	1 Missing ⚠️
...amm/src/pybamm/expression_tree/binary_operators.py	96.96%	1 Missing ⚠️
.../pybamm/src/pybamm/expression_tree/state_vector.py	90.00%	1 Missing ⚠️
...ckages/pybamm/src/pybamm/expression_tree/symbol.py	94.11%	1 Missing ⚠️

Additional details and impacted files

@@                        Coverage Diff                        @@
##             stack/03-pybamm-rust-packaging    #5959   +/-   ##
=================================================================
  Coverage                                  ?   98.18%           
=================================================================
  Files                                     ?      351           
  Lines                                     ?    36399           
  Branches                                  ?        0           
=================================================================
  Hits                                      ?    35738           
  Misses                                    ?      661           
  Partials                                  ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Bencher Report

Project	PyBaMM
Branch	stack/04-to-rust-lowering
Testbed	bare-metal

⚠️ WARNING: Truncated view!

The full continuous benchmarking report exceeds the maximum length allowed on this platform.

🐰 View full continuous benchmarking report in Bencher