def reorder_ns

stelliferous@0.0.2.0 · kernel_planning.bend:58

import Base
import stelliferous@0.0.2.0/kernel_cost.bend as Cost

58# The weight reorder writes complete eight-channel tiles, including zero lanes.
59def reorder_ns(+work: Work,config: Cost.Configuration) -> F32:
60  Work{positions,+inner,+rows,owners,geometry,activation} = work
61  Cost.Configuration{mac,byte_ns,operation_ns,grain,wake} = config
62  ((U32.to_f32(((rows + 7) / 8 * 8 : U32)) * U32.to_f32(inner) * 4.0 + payload_float(weight_bytes(work))) * byte_ns : F32)