lutinterp 0.1.0

Generate C lookup-table interpolation code from Python callables

lutInterp

Generate C lookup-table interpolation code from Python callables — using integer-only arithmetic suitable for embedded targets (no floating point at runtime).

Installation

pip install lutinterp

What it does

Given a Python function, CInterpolator produces:

• A C coefficient array (int8_t, int16_t, or int32_t) scaled to maximally fill the chosen integer type
• A C function that approximates the original using only integer arithmetic and bit-shifts — no division, no floating point

The x-axis is parameterized with power-of-2 exponents so that table lookup is a single right-shift instruction.

Usage

import numpy as np
from lutinterp import CInterpolator

# Approximate sqrt over [0, 4096] with a linear LUT
lut = CInterpolator(
    np.sqrt,
    typ="linear",
    coefResExp=16,       # store coefficients as int16_t
    xSupportPointsExp=4, # 2^4 = 16 segments
    xRangeExp=12,        # input range [0, 2^12] = [0, 4096]
    xRangeSign="pos",
)

print(lut)              # prints C code to stdout
lut.print_to_file("sqrt_lut.c")  # or save to a file

Cubic spline

lut = CInterpolator(
    np.sqrt,
    typ="cubic",
    coefResExp=16,
    xSupportPointsExp=3,
    xRangeExp=12,
    xRangeSign="pos",
)

Visualise error

lut.plot_error()   # relative error vs exact function
lut.plot_func()    # overlay: linear / cubic / pchip / exact

Key parameters

Parameter	Description
typ	'linear' or 'cubic'
coefResExp	Bit-width of coefficients (8, 16, or 32)
xSupportPointsExp	log2 of the number of LUT segments
xRangeExp	log2 of the input range
xRangeSign	'pos' → [0, 2^n], 'neg' → [2^n, 0], 'center' → [-2^n, +2^n]

License

MIT — see LICENSE.