---
license: mit
tags:
- pytorch
- safetensors
- threshold-logic
- neuromorphic
---

# threshold-clz4

4-bit count leading zeros.

## Function

clz4(a3, a2, a1, a0) = number of leading zeros from MSB

## Truth Table (selected rows)

| Input | First 1 | CLZ | Output |
|-------|---------|-----|--------|
| 1xxx | bit 3 | 0 | 000 |
| 01xx | bit 2 | 1 | 001 |
| 001x | bit 1 | 2 | 010 |
| 0001 | bit 0 | 3 | 011 |
| 0000 | none | 4 | 100 |

## Architecture

```
Layer 1: Priority detection from MSB
  has3 = a3 (MSB is set, clz=0)
  has2_first = a2 AND NOT(a3) (clz=1)
  has1_first = a1 AND NOT(a2) AND NOT(a3) (clz=2)
  has0_first = a0 AND NOT(a1) AND NOT(a2) AND NOT(a3) (clz=3)
  all_zero = NOT(any bit) (clz=4)

Layer 2: Binary encoding
  y0 = has2_first OR has0_first (clz is 1 or 3)
  y1 = has1_first OR has0_first (clz is 2 or 3)
  y2 = all_zero (clz is 4)
```

## Parameters

| | |
|---|---|
| Inputs | 4 |
| Outputs | 3 |
| Neurons | 8 |
| Layers | 2 |
| Parameters | 39 |
| Magnitude | 26 |

## Usage

```python
from safetensors.torch import load_file
# See model.py for full implementation

# clz4(1, 0, 0, 0) = [0, 0, 0] = 0 (MSB is set)
# clz4(0, 0, 1, 0) = [0, 1, 0] = 2 (two leading zeros)
# clz4(0, 0, 0, 0) = [1, 0, 0] = 4 (all zeros)
```

## License

MIT