Struct TurboQuantData 

pub struct TurboQuantData { /* private fields */ }

TurboQuant array data.

TurboQuant is a lossy vector quantization encoding for Vector extension arrays. It stores quantized coordinate codes for unit-norm vectors, along with shared codebook centroids and the parameters of the current structured rotation.

Norms should be stored externally in the L2Denorm ScalarFnArray wrapper.

See the module docs for algorithmic details.

Note that degenerate TurboQuant arrays have zero rows and bit_width == 0, with all slots empty.

Implementations

impl TurboQuantData

pub fn try_new( dimension: u32, bit_width: u8, num_rounds: u8, ) -> VortexResult<Self>

Build a TurboQuantData with validation.

Errors

Returns an error if:

• dimension is less than MIN_DIMENSION.
• bit_width is greater than MAX_BIT_WIDTH.

pub unsafe fn new_unchecked( dimension: u32, bit_width: u8, num_rounds: u8, ) -> Self

Build a TurboQuantData without validation.

Safety

The caller must ensure:

• dimension is >= MIN_DIMENSION.
• bit_width is in the range [0, MAX_BIT_WIDTH].
• num_rounds is >= 1 (or 0 for degenerate empty arrays).

Violating these invariants may produce incorrect results during decompression.

pub fn validate( dtype: &DType, codes: &ArrayRef, centroids: &ArrayRef, rotation_signs: &ArrayRef, ) -> VortexResult<()>

Validates the components that would be used to create a TurboQuantData.

This function checks all the invariants required by new_unchecked.

pub fn dimension(&self) -> u32

The vector dimension d, as stored in the Vector extension dtype’s FixedSizeList storage.

pub fn bit_width(&self) -> u8

MSE bits per coordinate (1-MAX_BIT_WIDTH for non-empty arrays, 0 for degenerate empty arrays).

pub fn num_rounds(&self) -> u8

The number of sign-diagonal + WHT rounds in the structured rotation.

pub fn padded_dim(&self) -> u32

Padded dimension (next power of 2 >= dimension).

The current Walsh-Hadamard-based structured rotation requires power-of-2 input, so non-power-of-2 dimensions are zero-padded to this value.

Trait Implementations

impl ArrayEq for TurboQuantData

fn array_eq(&self, other: &Self, _precision: Precision) -> bool

impl ArrayHash for TurboQuantData

fn array_hash<H: Hasher>(&self, state: &mut H, _precision: Precision)

impl Clone for TurboQuantData

fn clone(&self) -> TurboQuantData

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TurboQuantData

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

Formats the value using the given formatter.

impl Display for TurboQuantData

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

Formats the value using the given formatter.