etrommer / torch-approx

GPU-accelerated Neural Network layers using Approximate Multiplications for PyTorch
https://etrommer.de/torch-approx
MIT License
6 stars 3 forks source link

Support inline compilation of C approximate functions #17

Open etrommer opened 8 months ago

etrommer commented 8 months ago

Add the feature to replace the LUT operation with an inlined C function that performs operand transformations according to the logic of a given AM.

This will be helpful in benchmarking 12-Bit and 16-Bit AMs where a LUT would be too large.