Module simd 

SIMD vector operations for embedding similarity and compressed-vector search.

Dispatch uses the best supported target kernel with scalar fallbacks; WASM SIMD128 is selected at compile time. The public SIMD surface is unstable.

See docs/simd.md for the kernel family, dispatch, and quantization design.

Structs

BinaryVector

Unstable: binary quantization format and struct layout are under active design.

Int4Params

Unstable: INT4 quantization internals; scale/bias scheme may change.

Int4Vector

Unstable: INT4 quantization format is under active design; struct layout may change.

PreparedQueryWithMeta

A prepared query with caller-provided normalization metadata.

QuantizationParams

Unstable: INT8 quantization parameters; scale/bias scheme may change.

QuantizedVector

Unstable: INT8 quantized vector; struct layout and invariants may change.

SimdConfig

Unstable: SIMD dispatch internals; fields may be added as new ISAs are supported.

Enums

NormalizationHint

Caller assertion that a vector is L2-unit-normalized (norm ≈ 1).

PreparedQuery

Unstable: pre-quantized query for repeated same-tier distance computation.

QuantizationTier

Unstable: tier design is under active iteration; tier boundaries may change.

QuantizedData

Unstable: unified quantized data container; variants may change with tier redesign.

Functions

approximate_cosine_distance

Unstable: quantizes an f32 query and computes tiered cosine distance.

approximate_cosine_distance_prepared

Unstable: computes prepared cosine distance in [0, 2] for matching tiers.

approximate_cosine_distance_prepared_with_meta

Computes prepared cosine distance; hints are accepted but do not select a separate code path.

approximate_dot_product

Unstable: approximate tiered dot-product dispatch.

approximate_dot_product_prepared

Unstable: computes a prepared dot product for matching non-binary tiers.

approximate_int4_batch_prepared

Computes distances from one prepared INT4 query without re-quantizing it.

approximate_int4_batch_prepared_into

Writes prepared INT4 distances into a reusable buffer, clearing it on error.

approximate_int8_batch_prepared

Computes distances from one prepared INT8 query without re-quantizing it.

approximate_int8_batch_prepared_into

Writes prepared INT8 distances into a reusable buffer, clearing it on error.

batch_approximate_cosine_distance_prepared

Computes distances from one prepared query to all stored vectors.

batch_approximate_cosine_distance_prepared_into

Writes prepared-query distances into a reusable buffer, clearing it on error.

batch_cosine_one_vs_many

Unstable: one-query/many-candidate cosine similarity.

batch_cosine_similarity

Unstable: batched cosine dispatch; callers supply normalization knowledge.

batch_dot_product

Unstable: batched dot-product dispatch with a same-query fast path.

cosine_similarity

Computes cosine similarity, returning 0.0 for a mismatch, empty input, or zero norm.

cosine_similarity_fused

Unstable: fused single-pass cosine similarity.

cosine_similarity_i8

Unstable: SIMD INT8 cosine similarity; norm storage approach may change.

cosine_similarity_pre_normalized

Cosine similarity against a query whose norm the caller already knows.

dot_product

Computes the float dot product, returning 0.0 for a dimensional mismatch.

dot_product_batch4

Compute dot product of one query against 4 candidates simultaneously.

dot_product_i8

Unstable: computes an approximate float dot product, returning 0.0 for a mismatch.

dot_product_i8_raw

Unstable: computes an unscaled raw INT8 dot product, returning 0.0 for a mismatch.

euclidean_distance

Computes Euclidean (L2) distance, returning f32::MAX for a dimensional mismatch.

is_unit_norm

Returns true when the squared norm of v is within 1e-4 of 1.0.

manhattan_distance

Computes Manhattan (L1) distance, the sum of absolute per-element differences.

normalize

Unstable: SIMD dispatch layer; use lattice_embed::utils::normalize for the stable wrapper.

prepare_query

Prepare a query vector for repeated distance computation against a homogeneous tier.

prepare_query_with_norm

Prepare a query annotated with the given normalization hint.

query_norm

The query-side norm for cosine_similarity_pre_normalized.

resolved_dot_product_batch4_kernel

Resolve the best available batch-4 f32 dot-product kernel once and return it.

resolved_dot_product_kernel

Resolve the best available f32 dot-product kernel once and return it.

resolved_i8_dot_kernel

Return the cached INT8 dot-product kernel for tight loops.

simd_config

Unstable: SIMD dispatch internal; shape may change as new backends are added.

squared_euclidean_distance

Computes squared Euclidean distance without a square root.

try_approximate_cosine_distance_prepared

Alias for approximate_cosine_distance_prepared retained for compatibility.

try_approximate_dot_product_prepared

Alias for approximate_dot_product_prepared retained for compatibility.

Type Aliases

DotBatch4Kernel

SIMD kernel type for batch-4 f32 dot product.

DotKernel

SIMD kernel function pointer type for f32 dot product.

I8DotKernel

INT8 dot-product kernel function pointer type.