Livt packages

Livt.ML

Version 1.6.1

Fixed-size quantized inference, tensor storage, attention, and transformer building blocks for Livt.

Published files (132)

src/linear/Int8BroadcastMac64x8Primitive.lvt

namespace Livt.ML.Linear

/**
 * Portable pipelined primitive that broadcasts one activation over 64 int8 weights.
 */
@Opaque
component Int8BroadcastMac64x8Primitive
{
	// Command and operand inputs held stable until acknowledgment.
	public request: in logic
	public clear: in logic
	public activation: in int
	public weights: in logic[512]

	// Selects one accumulator for result readback.
	public resultIndex: in logic[6]

	// Native worker completion and selected accumulator result.
	public acknowledge: out logic
	public result: out int
}