README.md

Loom Example

S. Sharify, A. Delmas, P. Judd, K. Siu, and A. Moshovos Loom: exploiting weight and activation precisions to accelerate convolutional neural networks

Input Parameters Description

The following parameters are valid for this architecture:

Name	Data Type	Description	Valid Options	Default
lanes	uint32	Number of concurrent multiplications per PE	Positive Number	16
columns	uint32	Number of columns/windows in the tile	Positive number	16
rows	uint32	Number of rows/filters in the tile	Positive number	16
tiles	uint32	Number of tiles	Positive number	16
pe_width	uint32	PE input bit-width	Positive number	16
group_size	uint32	Number of columns/rows per group	Positive number	1
pe_serial_bits	uint32	Number of serial activations bits per PE	Positive Number	1
minor_bit	bool	Calculate also the minor bit for dynamic precisions	True-False	false
dynamic_weights	bool	Use dynamic precision for the weights	True-False	False

Example batch files:

• Loom_example: Performs Loom simulation and calculates potentials