| --- |
| license: mit |
| tags: |
| - pytorch |
| - safetensors |
| - threshold-logic |
| - neuromorphic |
| --- |
| |
| # threshold-clz8 |
|
|
| 8-bit count leading zeros. |
|
|
| ## Function |
|
|
| clz8(a7, a6, a5, a4, a3, a2, a1, a0) = number of leading zeros from MSB (0-8) |
|
|
| ## Truth Table (selected rows) |
|
|
| | Input | First 1 | CLZ | Output | |
| |-------|---------|-----|--------| |
| | 1xxxxxxx | bit 7 | 0 | 0000 | |
| | 01xxxxxx | bit 6 | 1 | 0001 | |
| | 001xxxxx | bit 5 | 2 | 0010 | |
| | 0001xxxx | bit 4 | 3 | 0011 | |
| | 00001xxx | bit 3 | 4 | 0100 | |
| | 000001xx | bit 2 | 5 | 0101 | |
| | 0000001x | bit 1 | 6 | 0110 | |
| | 00000001 | bit 0 | 7 | 0111 | |
| | 00000000 | none | 8 | 1000 | |
|
|
| ## Architecture |
|
|
| ``` |
| Layer 1: Priority detection from MSB (9 neurons) |
| has7 = a7 (MSB is set, clz=0) |
| has6_first = a6 AND NOT(a7) (clz=1) |
| has5_first = a5 AND NOT(a6) AND NOT(a7) (clz=2) |
| ... |
| has0_first = a0 AND NOT(a1..a7) (clz=7) |
| all_zero = NOT(any bit) (clz=8) |
| |
| Layer 2: Binary encoding (4 neurons) |
| y0 = has6_first OR has4_first OR has2_first OR has0_first |
| y1 = has5_first OR has4_first OR has1_first OR has0_first |
| y2 = has3_first OR has2_first OR has1_first OR has0_first |
| y3 = all_zero |
| ``` |
|
|
| ## Parameters |
|
|
| | | | |
| |---|---| |
| | Inputs | 8 | |
| | Outputs | 4 | |
| | Neurons | 13 | |
| | Layers | 2 | |
| | Parameters | 117 | |
| | Magnitude | 69 | |
|
|
| ## Usage |
|
|
| ```python |
| from safetensors.torch import load_file |
| # See model.py for full implementation |
| |
| # clz8(1,0,0,0,0,0,0,0) = [0,0,0,0] = 0 (MSB set) |
| # clz8(0,0,0,0,1,0,0,0) = [0,1,0,0] = 4 (four leading zeros) |
| # clz8(0,0,0,0,0,0,0,0) = [1,0,0,0] = 8 (all zeros) |
| ``` |
|
|
| ## License |
|
|
| MIT |
|
|
