def dense_baseline

stelliferous@0.0.2.0 · kernel_band_planning.bend:112

import Base
import stelliferous@0.0.2.0/kernel_cost.bend as Cost
import stelliferous@0.0.2.0/kernel_planning.bend as KernelPlanning
import stelliferous@0.0.2.0/tensor_band.bend as Bands
import stelliferous@0.0.2.0/tensor_band_window.bend as BandWindow
import stelliferous@0.0.2.0/kernel_window.bend as KernelWindow

112def dense_baseline(+work: KernelPlanning.Work,window: BandWindow.Geometry,+config: Cost.Configuration) -> F32:
113  match work window config:
114    case KernelPlanning.Work{+positions,inner,+rows,owners,geometry,activation} BandWindow.Geometry{Bands.Geometry{+batch,channels,height,width},kernel,stride,padding,oh,ow} Cost.Configuration{mac,+byte_ns,operation_ns,grain,wake}:
115      +one = KernelPlanning.estimate(0,work,config,KernelPlanning.movement(0n,0,KernelWindow.padded(positions),KernelPlanning.root_rows(work),work,config),1)
116      +output_size = (rows * positions : U32)
117      +images = U32.to_f32(batch)
118      Bool.pick(F32,U32.is_le(batch,1),one,
119        (reorder_cost(work,config) + images * (one - reorder_cost(work,config) + transfer((channels * height * width : U32),byte_ns) + U32.to_f32(output_size) * 4.0 * byte_ns) +
120          KernelPlanning.payload_float(Nat.mul(4n,KernelPlanning.capacity((batch * output_size : U32)))) * byte_ns + (images - 1.0) * clone_cost(work,config) : F32))