Struct FlatVectorData 

pub struct FlatVectorData {
    pub dim: usize,
    pub doc_ids: Vec<(u32, u16)>,
    /* private fields */
}

Flat vector data for brute-force search.

Uses a single contiguous Vec<f32> instead of Vec<Vec<f32>>. Loading is a single bulk memcpy (1 allocation) instead of N separate allocations with float-by-float parsing. For 3.3M vectors at 768 dims this reduces load time from ~36s to ~1s.

Fields

dim: usize

doc_ids: Vec<(u32, u16)>

Document IDs with ordinals: (doc_id, ordinal) pairs Ordinal tracks which vector in a multi-valued field

Implementations

impl FlatVectorData

pub fn num_vectors(&self) -> usize

Number of vectors

pub fn get_vector(&self, idx: usize) -> &[f32]

Get vector at index as a &f32 slice.

Panics

Panics if idx >= num_vectors.

pub fn get_doc_id(&self, idx: usize) -> (u32, u16)

Get doc_id and ordinal at index.

pub fn vectors_as_bytes(&self) -> &[u8]

Raw flat vector storage as byte slice for bulk streaming. Returns the contiguous f32 data reinterpreted as bytes.

pub fn stream_to_writer( &self, writer: &mut dyn Write, doc_id_offset: u32, ) -> Result<()>

Stream vectors and doc_ids from this FlatVectorData to a writer. doc_id_offset is added to each doc_id for multi-segment merges.

pub fn write_binary_header( dim: usize, num_vectors: usize, writer: &mut dyn Write, ) -> Result<()>

Write the binary header (magic + dim + num_vectors) to a writer.

pub fn estimated_memory_bytes(&self) -> usize

Estimate memory usage

pub fn from_binary_bytes(data: &[u8]) -> Result<Self>

Deserialize from binary format. Single bulk memcpy for vectors.

Parses the 12-byte header, bulk-copies vectors as one contiguous Vec, and parses doc_ids. For 3.3M vectors at 768 dims this is ~1s vs ~36s with the old Vec<Vec> approach (1 allocation vs 3.3M allocations).

pub fn serialized_binary_size(index_dim: usize, num_vectors: usize) -> usize

Compute the serialized size without actually serializing.

pub fn serialize_binary_from_flat_streaming( index_dim: usize, flat_vectors: &[f32], original_dim: usize, doc_ids: &[(u32, u16)], writer: &mut dyn Write, ) -> Result<()>

Stream directly from flat f32 storage to a writer (zero-buffer serialization).

flat_vectors is contiguous storage of dim*n floats. original_dim is the dimension in flat_vectors (may differ from index_dim for MRL).

Trait Implementations

impl Clone for FlatVectorData

fn clone(&self) -> FlatVectorData

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for FlatVectorData

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

Formats the value using the given formatter.