Matmul.Order: type of bendtorch-core@0.1.0.0
Import: import bendtorch-core@0.1.0.0/src/tensor/matmul.bend as Matmul
Lines 410-425 of bendtorch-core@0.1.0.0/src/tensor/matmul.bend: https://jend.leeeet.dev/read.txt?path=bendtorch-core@0.1.0.0/src/tensor/matmul.bend&start_line=410&end_line=425

import Base
import bendtorch-core@0.1.0.0/src/tensor/parallel.bend as Parallel

# Products that read an operand in place. `x @ W^T` and `g^T @ x` read a
# transposed operand; instead of building the transpose they read the
# operand at the transposed index. An operand of a product is laid out in
# one of two orders: `Kept`, as `[outer, inner]`, where term k of output row
# (or column) `outer` sits at outer * inner + k; or `Swapped`, as
# `[inner, outer]`, where it sits at k * outers + outer. `matmul` is the
# product of a `Kept` left and a `Swapped` right operand, `matmul_nt` of two
# `Kept` operands and `matmul_tn` of two `Swapped` ones. Cell (row, column)
# adds the inner terms in increasing k from +0.0, as `natural_dot` does. The
# cells are built by `product_build` below: it forks as `Parallel.build`
# forks, and below the forks `product_tree` builds them as `Parallel.tree`
# does, except that eight cells below the count and in one output row run
# as one loop with eight accumulators (`row_run`).
type Order is Data:
  Kept{}
  Swapped{}

Uses (0):