Struct TensorWasmKernelBlueprint 

pub struct TensorWasmKernelBlueprint {
    pub entry: String,
    pub ops: Vec<TensorWasmOp>,
    pub grid_hint: GridHint,
    pub shared_mem_bytes: u32,
}

A complete blueprint that the emitter (S12) can turn into PTX text.

Fields

entry: String

Symbolic entry name (used by load_ptx lookup).

ops: Vec<TensorWasmOp>

Ordered ops to emit.

grid_hint: GridHint

Grid hint (S11 fills in; S12 honours).

shared_mem_bytes: u32

Bytes of shared memory the kernel will request.

Implementations

impl TensorWasmKernelBlueprint

pub fn new(entry: impl Into<String>) -> Self

Construct a new blueprint.

pub fn push(self, op: TensorWasmOp) -> Self

Append an op; returns self for builder-style chaining.

pub fn with_grid(self, hint: GridHint) -> Self

Update the grid hint.

pub fn with_shared_mem(self, bytes: u32) -> Self

Update the shared-memory requirement.

pub fn is_empty(&self) -> bool

True if the blueprint contains no ops (a no-op kernel).

pub fn fingerprint(&self) -> u64

Stable 64-bit hash used as the KernelCache key by S13.

Uses blake3 over a canonical byte serialisation so the hash is deterministic across Rust versions and platforms — std’s DefaultHasher (SipHash) is explicitly documented as unstable and must not be relied on for persistence. The first 8 bytes of the 32- byte BLAKE3 digest are interpreted as a little-endian u64.

The width is intentionally kept at 64 bits: the on-disk cache format (DiskCache, [24..32) blueprint fingerprint (u64 LE)), the crate::deopt::DeoptGuard DashMap<u64, _> key, and the cross-crate OffloadedFunction.fingerprint ABI all key on this u64. Callers wanting a lower collision probability without those persistence/ABI constraints can use fingerprint128.

pub fn fingerprint128(&self) -> u128

128-bit widening of fingerprint, derived from the same canonical BLAKE3 digest.

jit HIGH (finding 2, truncation width): a 64-bit truncation of a 256-bit digest has a birthday-bound collision probability around 2^-32 across ~4 billion distinct kernels. Embedders that key kernels in a context not bound by the 64-bit on-disk/ABI format (e.g. an in-memory dedup map over a very large kernel population) should key on this 128-bit value instead, which pushes the birthday bound out to ~2^-64. The low 64 bits equal fingerprint.

Trait Implementations

impl Clone for TensorWasmKernelBlueprint

fn clone(&self) -> TensorWasmKernelBlueprint

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorWasmKernelBlueprint

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

Formats the value using the given formatter.

impl PartialEq for TensorWasmKernelBlueprint

fn eq(&self, other: &TensorWasmKernelBlueprint) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for TensorWasmKernelBlueprint