Struct FlatVectorData 

pub struct FlatVectorData;

Flat vector binary format helpers for writing.

Binary format v3:

[magic(u32)][dim(u32)][num_vectors(u32)][quant_type(u8)][padding(3)]
[vectors: N×dim×element_size]
[doc_ids: N×(u32+u16)]

element_size is determined by quant_type: f32=4, f16=2, uint8=1. Packed binary vectors instead use dim / 8 bytes per row. LazyFlatVectorData retains a zero-copy view of the packed document/ordinal map and accesses vector payloads lazily via range reads (mmap on native files).

Implementations

impl FlatVectorData

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

Write the binary header to a writer.

pub fn serialized_binary_size( dim: usize, num_vectors: usize, quant: DenseVectorQuantization, ) -> Result<usize>

Compute the serialized size without actually serializing.

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

Stream from flat f32 storage to a writer, quantizing on write.

flat_vectors is contiguous storage of dim*n f32 floats. Vectors are quantized to the specified format before writing.

pub fn serialize_binary_from_bits_streaming( dim_bits: usize, packed_vectors: &[u8], doc_ids: &[(u32, u16)], writer: &mut dyn Write, ) -> Result<()>

Stream packed binary vectors (pre-packed bytes) to a writer.

packed_vectors is contiguous storage of num_vectors * byte_len bytes. dim_bits is the number of bits (dimensions).

pub fn write_raw_vector_bytes( raw_bytes: &[u8], writer: &mut dyn Write, ) -> Result<()>

Write raw pre-quantized vector bytes to a writer (for merger streaming).

raw_bytes is already in the target quantized format.