viva_tensor/lns

LNS - Log-Number System

Efficient arithmetic for resonance and probabilistic computing.

Represents numbers as their logarithms: x => log(|x|) + sign. Multiplication becomes addition: log(a*b) = log(a) + log(b). Division becomes subtraction: log(a/b) = log(a) - log(b). Power becomes multiplication: log(x^p) = p * log(x).

Used for:

High-dynamic range calculations
Chain multiplications (probability chains)
“Resonance” calculations (heavy multiplication/exponentiation)

Precision is traded for speed (up to 8x faster than FP32 FMA).

Types

LogNumber

</>

LNS Tensor reference (storage in f32 log-domain)

pub type LogNumber =
  ffi.LnsTensorRef

Values

div

</>

pub fn div(
  a: ffi.LnsTensorRef,
  b: ffi.LnsTensorRef,
) -> ffi.LnsTensorRef

Divide two LNS tensors (Subtraction in log-domain)

from_tensor

</>

pub fn from_tensor(ref: ffi.NativeTensorRef) -> ffi.LnsTensorRef

Convert standard f64 tensor to LNS format

mul

</>

pub fn mul(
  a: ffi.LnsTensorRef,
  b: ffi.LnsTensorRef,
) -> ffi.LnsTensorRef

Multiply two LNS tensors (Addition in log-domain)

sqrt

</>

pub fn sqrt(a: ffi.LnsTensorRef) -> ffi.LnsTensorRef

Square root of LNS tensor (Bit shift / division by 2 in log-domain)

to_tensor

</>

pub fn to_tensor(lns: ffi.LnsTensorRef) -> ffi.NativeTensorRef

Convert LNS tensor back to standard f64 tensor