threshold-signextend4to8

Sign extend a 4-bit signed integer to 8 bits.

Function

Copies the sign bit (MSB) to the upper 4 bits:

• Input: x3 (sign), x2, x1, x0
• Output: x3, x3, x3, x3, x3, x2, x1, x0

Truth Table (selected examples)

4-bit	signed	8-bit	signed
0000	0	00000000	0
0001	1	00000001	1
0111	7	00000111	7
1000	-8	11111000	-8
1111	-1	11111111	-1

Architecture

Single layer with 8 neurons. Each neuron passes through one input bit:

• y0-y3: direct passthrough of x0-x3
• y4-y7: passthrough of x3 (sign bit)

Output	Input	Weights	Bias
y0	x0	[0,0,0,1]	-1
y1	x1	[0,0,1,0]	-1
y2	x2	[0,1,0,0]	-1
y3	x3	[1,0,0,0]	-1
y4	x3	[1,0,0,0]	-1
y5	x3	[1,0,0,0]	-1
y6	x3	[1,0,0,0]	-1
y7	x3	[1,0,0,0]	-1

Parameters

Inputs	4
Outputs	8
Neurons	8
Layers	1
Parameters	40
Magnitude	16

Usage

from safetensors.torch import load_file
import torch

w = load_file('model.safetensors')

def signextend4to8(x3, x2, x1, x0):
    inp = torch.tensor([float(x3), float(x2), float(x1), float(x0)])
    return [int((inp * w[f'y{i}.weight']).sum() + w[f'y{i}.bias'] >= 0)
            for i in range(8)]

# Sign extend -3 (1101 in 4-bit) to 8-bit
result = signextend4to8(1, 1, 0, 1)
print(result)  # [1, 0, 1, 1, 1, 1, 1, 1] = 11111101 = -3

License

MIT