diff options
| author |  Bjarke Hammersholt Roune <broune@google.com> | 2017-12-15 12:29:06 -0800 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2017-12-15 12:32:40 -0800 |
| commit | 22828e18071a682fcccf12f79d75eb6d57144dba (patch) | |
| tree | 27501f7eb3621ec677b2f04091f25dc79b6d580a /tensorflow/compiler/xla/tests/test_utils.cc | |
| parent | ed24130f90c2c45db0473df3e9158d4895ce326b (diff) | |
Resolves an issue that came up with ResNet50 when using the CPU backend as a reference.
* The CPU now handles window reversal on convolutions instead of ignoring it.
* The GPU backend now reports an error on window reversal instead of ignoring it.
PiperOrigin-RevId: 179226118
Diffstat (limited to 'tensorflow/compiler/xla/tests/test_utils.cc')
| -rw-r--r-- | tensorflow/compiler/xla/tests/test_utils.cc | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/tensorflow/compiler/xla/tests/test_utils.cc b/tensorflow/compiler/xla/tests/test_utils.cc index 56859542a9..f9c62ec217 100644 --- a/tensorflow/compiler/xla/tests/test_utils.cc +++ b/tensorflow/compiler/xla/tests/test_utils.cc @@ -25,6 +25,7 @@ namespace { template <typename FloatT> void PopulateWithRandomFloatingPointData(Literal* literal) { + // TODO(b/69179121): Generate data that is less self-similar. CHECK_EQ(literal->shape().element_type(), primitive_util::NativeToPrimitiveType<FloatT>()); std::minstd_rand0 engine; |