Struct TurboCodeWireV1 

pub struct TurboCodeWireV1;

Encoder/decoder for TurboCode wire format v1.

Implementations

impl TurboCodeWireV1

pub fn encode(code: &TurboCode, profile: &TurboQuantizer) -> Result<Vec<u8>>

Encode a validated TurboCode using the supplied quantizer profile.

pub fn decode(bytes: &[u8], profile: &TurboQuantizer) -> Result<TurboCode>

Decode and validate TurboCode wire bytes against the supplied profile.

pub fn encode_batch( codes: &[TurboCode], profile: &TurboQuantizer, ) -> Result<Vec<u8>>

Encode a batch of TurboCodes using the supplied shared quantizer profile. Convenience wrapper that uses f32 (lossless) radii.

pub fn encode_batch_with_radii( codes: &[TurboCode], profile: &TurboQuantizer, radii_codec: RadiusCodecProfileV1, ) -> Result<Vec<u8>>

Encode a batch of TurboCodes using the supplied shared quantizer profile and the specified radii codec.

For a batch of N vectors at dim D, projections P, bits B, the layout is:

offset size field

---

0 4 magic “TQB1” 4 2 version (LE u16) 6 2 rotation flag (LE u16) 8 1 variant (= 2 for BATCH) 9 1 reserved (= 0) 10 4 dim (LE u32) 14 1 polar_bits (u8) 15 1 radii_codec (0=f32, 1=BlockLogU8) <– NEW 16 2 reserved (= 0) 18 4 qjl_projections (LE u32) 22 8 seed (LE u64) 30 4 n_vectors (LE u32) 34 4 vector_payload_len (LE u32) — size of EACH per-vector payload 38 … vector 0 payload (vector_payload_len bytes) … … vector 1 payload … … …

Per-vector payload depends on radii_codec:

• f32 (lossless): [radii: 4*N bytes] [angles: A bytes] [signs: S bytes]
• BlockLogU8 (lossy): [radii: 1*N + 8 bytes (min/max)] [angles] [signs]

pub fn decode_batch( bytes: &[u8], profile: &TurboQuantizer, ) -> Result<Vec<TurboCode>>

Decode a batched TQB1 payload into a Vec.

impl TurboCodeWireV1

pub fn parse_header(bytes: &[u8]) -> Result<TurboCodeWireHeader>

Parse just the 44-byte wire header. This is the public entry point for callers that want to extract the quantizer profile from the wire format without validating against a specific quantizer instance.