Struct CpuI8MatmulSquare 

pub struct CpuI8MatmulSquare<const DIM: usize>;

Parametric square DIM × DIM i8 × i8 → i16 matmul.

Determinism: per ADR-030’s per-axis substitution-determinism note, the integer-arithmetic CPU impl preserves bit-identity across targets. DIM is the square dimension; for non-square / non-integer / variable-shape tensor compute the wiki’s pattern is to compose this axis kernel through verbs over partition_product! (per ADR-033/044) — the axis layer fixes the atom shape.

HostBounds discipline

DIM is unconstrained at the axis level per Wiki ADR-018. The application’s HostBounds selection declares the ceiling: a CpuI8MatmulSquare<DIM> instantiation requires the application’s B to satisfy 2 * DIM * DIM <= B::AXIS_OUTPUT_BYTES_MAX per ADR-037. Specific DIM values (4, 8, 16, 32, 64, …) are picked by the application from its declared bounds, not by this crate.

Trait Implementations

impl<const DIM: usize> AxisExtension for CpuI8MatmulSquare<DIM>

const AXIS_ADDRESS: &'static str = <CpuI8MatmulSquare<DIM> as TensorAxis>::AXIS_ADDRESS

ADR-017 content address of this axis trait. The SDK macro derives this from the trait name and method signatures.

const MAX_OUTPUT_BYTES: usize = <CpuI8MatmulSquare<DIM> as TensorAxis>::MAX_OUTPUT_BYTES

Maximum bytes any kernel of this axis returns.

fn dispatch_kernel( kernel_id: u32, input: &[u8], out: &mut [u8], ) -> Result<usize, ShapeViolation>

Dispatch the kernel identified by kernel_id against the evaluated input bytes. The implementation copies the kernel’s output into out and returns the written length.

impl<const DIM: usize> Clone for CpuI8MatmulSquare<DIM>

fn clone(&self) -> CpuI8MatmulSquare<DIM>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<const DIM: usize> Debug for CpuI8MatmulSquare<DIM>

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<const DIM: usize> Default for CpuI8MatmulSquare<DIM>

fn default() -> CpuI8MatmulSquare<DIM>

Returns the “default value” for a type.

impl<const DIM: usize> SubstrateTermBody for CpuI8MatmulSquare<DIM>

fn body_arena() -> &'static [Term]

The Term arena the kernel decomposes to. Empty slice signals a primitive-fast-path axis whose body the implementation may evaluate through dispatch_kernel directly per ADR-055’s optional fast-path.

impl<const DIM: usize> TensorAxis for CpuI8MatmulSquare<DIM>

const AXIS_ADDRESS: &'static str = "https://uor.foundation/axis/TensorAxis/CpuI8MatmulSquare"

const MAX_OUTPUT_BYTES: usize

Per-impl axis output ceiling. The application’s HostBounds::AXIS_OUTPUT_BYTES_MAX (ADR-037) is checked against this value at dispatch; the axis impl carries no substrate-arbitrary cap of its own.

fn matmul(input: &[u8], out: &mut [u8]) -> Result<usize, ShapeViolation>

Multiply two row-major DIM × DIM i8 matrices into a DIM × DIM i16 product (saturating). Input is A || B (2 * DIM * DIM bytes); output is 2 * DIM * DIM bytes.

impl<const DIM: usize> Copy for CpuI8MatmulSquare<DIM>