diff options
author | A. Unique TensorFlower <gardener@tensorflow.org> | 2017-07-10 18:39:30 -0700 |
---|---|---|
committer | TensorFlower Gardener <gardener@tensorflow.org> | 2017-07-10 18:43:01 -0700 |
commit | aa239529c47c09599d1085791fc849f03efbddfe (patch) | |
tree | 0b06b956c810e02b5334397a98ab4f5364c6d648 | |
parent | 7624e8e9336e9b9f9a066292b11cf75f8fdeea05 (diff) |
[XLA] Fix bugs and incomplete comment in reduce_precision_test.cc.
We want a bitwise or, not bitwise and, to add in the sign bit. This was previously just creating zeros instead of negative test values.
PiperOrigin-RevId: 161473273
-rw-r--r-- | tensorflow/compiler/xla/tests/reduce_precision_test.cc | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/tensorflow/compiler/xla/tests/reduce_precision_test.cc b/tensorflow/compiler/xla/tests/reduce_precision_test.cc index a66c9b4487..48212dc7d1 100644 --- a/tensorflow/compiler/xla/tests/reduce_precision_test.cc +++ b/tensorflow/compiler/xla/tests/reduce_precision_test.cc @@ -214,11 +214,12 @@ XLA_TEST_P(ReducePrecisionTest, ReducePrecisionF32) { // Add positive values. input_values.push_back(tensorflow::bit_cast<float>(test_value[0])); expected_values.push_back(tensorflow::bit_cast<float>(test_value[index])); - // Add negative values. We do this in the + // Add negative values. We do this in the bitwise representation so as to + // avoid problems with NaN handling. input_values.push_back( - tensorflow::bit_cast<float>(test_value[0] & sign_bit)); + tensorflow::bit_cast<float>(test_value[0] ^ sign_bit)); expected_values.push_back( - tensorflow::bit_cast<float>(test_value[index] & sign_bit)); + tensorflow::bit_cast<float>(test_value[index] ^ sign_bit)); } // This is required for proper handling of NaN values. |