def row_block_run

bendtorch-core@0.1.0.0 · src/tensor/matmul.bend:568

import Base

568# Cells base .. base + 7, all in the row of cell base, from fuel terms.
569def row_block_run(fuel: Nat, +lo: Order, +ro: Order, +inner: Nat, +rows: Nat, +columns: Nat, +base: Nat,
570  la: Array<F32>, ra: Array<F32>) -> Array<F32> & (Array<F32> & Array<F32>):
571  match fuel:
572    case 0n:
573      eight_leaves_of(Eight{0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0}, (la, ra))
574    case 1n+rest:
575      eight_leaves(row_run(rest, stride(lo, rows), stride(ro, columns), read_offset(lo, inner, rows, row_of(columns, base), 0n),
576        read_offset(ro, inner, columns, column_of(columns, base), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(1n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(2n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(3n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(4n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(5n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(6n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(7n, base)), 0n),
577        Eight{0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0}, Array.get(F32, la, U32.from_nat(read_offset(lo, inner, rows, row_of(columns, base), 0n))),
578        read_eight(ra, read_offset(ro, inner, columns, column_of(columns, base), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(1n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(2n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(3n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(4n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(5n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(6n, base)), 0n), read_offset(ro, inner, columns, column_of(columns, Nat.add(7n, base)), 0n))))