* Make fake argument generation see through ReducePrecision and Convert ops when
authorBjarke Hammersholt Roune <broune@google.com>
Sat, 6 Jan 2018 00:56:14 +0000 (16:56 -0800)
committerTensorFlower Gardener <gardener@tensorflow.org>
Sat, 6 Jan 2018 01:00:14 +0000 (17:00 -0800)
commit6ed75e60c192c487a955bad155d0bf478135e7a5
tree4e7e0a4add0e05b2c9d1e7207fbb610a34921a9f
parent9f462b27e6df975b1e402c9926d740dfdb9f977a
 * Make fake argument generation see through ReducePrecision and Convert ops when
     determining constraints.
 * Generate both positive and negative numbers as a work-around for the CPU reduce
     implementation having poor numerical stability.
 * You can now VLOG hlo_runner to see the results.

PiperOrigin-RevId: 180993147
tensorflow/compiler/xla/service/hlo_runner.cc
tensorflow/compiler/xla/tests/test_utils.cc