Matmul.splits: def of bendtorch-core@0.1.0.0
Import: import bendtorch-core@0.1.0.0/src/tensor/matmul.bend as Matmul
Lines 336-343 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=336&end_line=343

import Base

# Split by work, not by cell count: a cell costs one term per inner index.
# The build forks until a subtree holds about 2^13 terms, and at most six
# levels, 64 subtrees, each with its own copy of both inputs: an output of
# 2^depth cells of `inner` terms forks depth + log2(inner) - 13 levels. On
# medusa2 at 8 threads a product of 2^15 terms or more ran faster with every
# extra level up to six, and forking cost nothing measurable on one thread.
def splits(+depth: Nat, +inner: Nat) -> Nat:
  at_most(at_most(Nat.sub(Nat.add(depth, halvings(inner, inner)), 13n), 6n), depth)

Uses (5):
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=Nat
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=Nat.add
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=Nat.sub
  https://jend.leeeet.dev/definition.txt?path=bendtorch-core@0.1.0.0/src/tensor/matmul.bend&name=at_most
  https://jend.leeeet.dev/definition.txt?path=bendtorch-core@0.1.0.0/src/tensor/matmul.bend&name=halvings