diff options
author | Benjamin Kramer <kramerb@google.com> | 2018-06-06 11:27:25 -0700 |
---|---|---|
committer | TensorFlower Gardener <gardener@tensorflow.org> | 2018-06-06 11:32:26 -0700 |
commit | 6aeb1fdc53fb2a7df61e2544ce92243b6b43ad02 (patch) | |
tree | 3794eed022bdf7761a06ab144aadb9391de205fd /SECURITY.md | |
parent | 879fc3440495d9388754cb7d1878caf034d03d61 (diff) |
[XLA:GPU] Allow intermediate outputs for reduce input fusions.
This generalizes the emitter to allow pretty much arbitrary multi-output fusion
as long as the shapes match the input of the reduce(s). The idea is that
multi-output fusion can move intermediate inputs into the same fusion so they
don't have to be re-read by the reduce.
PiperOrigin-RevId: 199497832
Diffstat (limited to 'SECURITY.md')
0 files changed, 0 insertions, 0 deletions