Crate sim_lib_numbers_tensor

Expand description

The n-dimensional tensor number domain: the uniform Tensor value, its domain registration and constructors (tensor, vec, mat, …), and the SpecTensor interface that specialized element-type backends plug into. Tensor shape and dtype stay canonical while the open TensorStorage contract supplies boxed host or loadable resident storage. Observation through Tensor::cell, Tensor::cells, and Tensor::materialize is checked so encoding and projection report resident readback failures.

Modules§

domains: Canonical symbols for the number-domain family.

Structs§

BoxedTensorStorage: Boxed host storage for scalar runtime values.
CpuTensorExecutor: Default host executor that delegates to the registered tensor functions.
SpecTensorDescriptor: Metadata describing one registered SpecTensor backend, surfaced as a descriptor value so the registry can advertise the specialized tensor.
SubmissionEvidence: Evidence returned after an executor has flushed accepted submissions.
Tensor: The uniform tensor value: an n-dimensional array of scalar number cells.
TensorExecutorCard: Description of one tensor executor.
TensorMeta: Tensor shape and dtype expected from an execution request.
TensorNumbersLib: Registered number-domain library that installs the numbers/tensor domain.
TensorOp: Open operation descriptor carried by a tensor request.
TensorRequest: A checked tensor execution request.
TensorSite: Eval-fabric wrapper that binds one tensor executor for a realization.
TypedTensorStorage: Host storage for typed scalar cells behind the canonical Tensor.

Enums§

TensorExecError: Error reported by tensor execution contracts.
TensorExecution: Result of submitting a tensor request to an executor.
TensorLocation: The observable placement of tensor storage.

Constants§

MAX_TENSOR_CELLS: The largest number of cells a tensor operation will materialize in one allocation. A dimension product can be far below usize::MAX and still be hopeless to allocate (a [1_000_000, 1_000_000] broadcast is 1e12 cells); this ceiling is the line past which the input is rejected rather than driven into an out-of-memory abort.

Statics§

RECIPES: Cookbook recipes for this domain, embedded at build time.

Traits§

SpecTensor: Interface a specialized element-type tensor backend implements to bridge its typed view and the uniform Tensor value.
TensorCell: A scalar cell type that can be held in typed host tensor storage.
TensorExecutor: A loadable provider that executes checked tensor requests.
TensorStorage: Storage behind the canonical Tensor value.

Functions§

active_tensor_executor: Returns the tensor executor currently bound in the active environment.
add_op_symbol: Open operation symbol for element-wise tensor addition.
bounded_element_count: The number of cells in a tensor of the given shape, failing closed both when the dimension product overflows usize (via checked_element_count) and when it exceeds MAX_TENSOR_CELLS.
build_scalar_tensor_value: Builds a rank-0 scalar tensor wrapping a single scalar number value.
build_tensor_value: Builds a tensor Value of the given shape from row-major data cells.
cast_op_symbol: Open operation symbol for explicit tensor casts.
cast_symbol: Returns the qualified function symbol for explicit tensor dtype conversion.
cast_tensor: Casts a tensor into target_dtype.
cast_tensor_value: Casts a tensor value into target_dtype and returns it as a runtime value.
checked_element_count: The number of cells in a tensor of the given shape, failing closed when the dimension product overflows usize instead of wrapping (release) or panicking (debug).
cos_op_symbol: Open operation symbol for element-wise cosine.
div_op_symbol: Open operation symbol for element-wise tensor division.
dot_op_symbol: Open operation symbol for vector dot product.
element_count: The number of cells in a tensor of the given shape. An empty shape is a scalar (one cell). This is the one home for the element_count helper that the generic, broadcast, linalg, and every typed tensor crate re-grew.
execute_tensor_binary_op: Runs a binary tensor operation through the active executor, or CPU when no executor is bound in the environment.
execute_tensor_dot: Runs a vector dot product through the active executor.
execute_tensor_matmul: Runs vector or matrix multiplication through the active executor.
execute_tensor_norm: Runs a Euclidean whole-tensor norm through the active executor.
execute_tensor_reduction: Runs a whole-tensor sum, min, or max through the active executor.
execute_tensor_request: Executes one tensor request through the active executor, defaulting to CPU.
execute_tensor_transcendental: Runs an f32/f64/half element-wise transcendental through the active executor.
execute_tensor_transpose: Runs a rank-2 transpose through the active executor.
execute_tensor_unary_op: Runs a unary tensor operation through the active executor, or CPU when no executor is bound in the environment.
exp_op_symbol: Open operation symbol for element-wise exponential.
flatten_tensor_scalar_cells: Compatibility view of a tensor’s row-major scalar cells as a shared flat slice.
index_op_symbol: Open operation symbol for indexing a tensor.
map_op_symbol: Open operation symbol for mapping a callable over a tensor.
mat_op_symbol: Open operation symbol for constructing a matrix tensor.
matmul_exec_op_symbol: Open operation symbol for matrix multiplication.
max_op_symbol: Open operation symbol for reducing all tensor cells to the maximum value.
min_op_symbol: Open operation symbol for reducing all tensor cells to the minimum value.
mul_op_symbol: Open operation symbol for element-wise tensor multiplication.
neg_op_symbol: Open operation symbol for element-wise tensor negation.
norm_op_symbol: Open operation symbol for Euclidean norm over all tensor cells.
number_domain: The symbol naming the tensor number domain (numbers/tensor).
number_literal_for_tensor_cell: Extracts the canonical NumberLiteral of a scalar tensor cell value, or None if the value is not a number. Shared backing for the typed literal-cell parsers below.
parse_bf16_literal_cell: Parses a tensor cell as a bfloat16 value, returning None unless it is a number in the numbers/bf16 domain whose canonical f32-form text parses cleanly.
parse_complex_literal_cell: Parses a tensor cell as a (real, imaginary) pair, returning None unless it is a number in the numbers/complex domain whose canonical a+bi form parses cleanly.
parse_f16_literal_cell: Parses a tensor cell as an IEEE half-precision value, returning None unless it is a number in the numbers/f16 domain whose canonical f32-form text parses cleanly.
parse_f32_literal_cell: Parses a tensor cell as an f32, returning None unless it is a number in the numbers/f32 domain whose canonical form parses cleanly.
parse_f64_literal_cell: Parses a tensor cell as an f64, returning None unless it is a number in the numbers/f64 domain whose canonical form parses cleanly.
parse_i64_literal_cell: Parses a tensor cell as an i64, returning None unless it is a number in the numbers/i64 domain whose canonical form parses cleanly.
parse_rational_literal_cell: Parses a tensor cell as a (numerator, denominator) rational pair, returning None unless it is a number in the numbers/rational domain whose canonical num/den form parses cleanly.
pow_op_symbol: Open operation symbol for element-wise tensor exponentiation.
rem_op_symbol: Open operation symbol for element-wise tensor remainder.
reshape_op_symbol: Open operation symbol for reshaping a tensor.
scalar_op_symbol: Open operation symbol for constructing a scalar tensor.
sin_op_symbol: Open operation symbol for element-wise sine.
slice_op_symbol: Open operation symbol for slicing a tensor.
spec_tensor_descriptor_value: Encodes a SpecTensorDescriptor as a registry descriptor table value with kind, symbol, dtype, implementation, and storage entries.
spec_tensor_symbol: Builds a descriptor symbol (numbers/tensor-spec/<name>) for a specialized tensor backend.
sqrt_op_symbol: Open operation symbol for element-wise square root.
sub_op_symbol: Open operation symbol for element-wise tensor subtraction.
sum_op_symbol: Open operation symbol for reducing all tensor cells with addition.
tensor_dtype: The shared element number domain (dtype) of a tensor’s cells.
tensor_execute_capability: Capability required by TensorSite when a request asks for tensor execution authority.
tensor_executor_math_op_symbols: Returns the operation symbols accepted by executor math providers.
tensor_executor_symbol: Symbol bound in a TensorSite child environment to the active executor.
tensor_op_symbol: Open operation symbol for constructing a tensor from shape, dtype, and cells.
tensor_site_symbol: Symbol naming the local tensor execution site exported by the tensor lib.
tensor_value_class_symbol: The symbol naming the tensor value class (numbers/Tensor) under which tensor values register and reconstruct.
tensor_value_ref: Borrows the Tensor backing a value, or None if it is not a tensor.
transpose_exec_op_symbol: Open operation symbol for matrix transpose.
vec_op_symbol: Open operation symbol for constructing a vector tensor.