def run_partitioned

stelliferous@0.0.2.0 · kernel_pipeline.bend:191

import Base
import stelliferous@0.0.2.0/kernel_tile.bend as Tile
import stelliferous@0.0.2.0/storage_buffer.bend as Storage
import stelliferous@0.0.2.0/kernel_window.bend as KernelWindow
import stelliferous@0.0.2.0/math_activation.bend as Activations

191def run_partitioned(~custom: F32 -> F32,activation: Activations.Activation,depth: U32,work: Work,input: Storage.Buffer<F32>,weights: Storage.Buffer<Tile.Tile8>,bias: Storage.Buffer<F32>) -> Storage.Buffer<F32>:
192  Work{+rows,+h,w,k,s,p,ow,+positions,+inner} = work
193  +padded = KernelWindow.padded(positions)
194  +limit = depth_limit(padded)
195  finish(partition(~custom,U32.to_nat(U32.min(depth,limit)),Task{Work{rows,h,w,k,s,p,ow,positions,inner},activation},0,padded,InputWindow{input,0,h},weights,bias),rows,positions)