experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Avoid dividing by zero in xla/literal_test.

Browse Source 
According to the standard, this is undefined behaviour, and MSVC implements it
as such.

PiperOrigin-RevId: 294264326
Change-Id: Ib5f64a5b54347b7dd401f89206da2f3de45de150
This commit is contained in:
Gunhan Gulsoy 2020-02-10 11:25:48 -08:00 committed by TensorFlower Gardener
parent 1a1c4f0f35
commit 02208d6dbb
1 changed files with 2 additions and 2 deletions

4
tensorflow/compiler/xla/literal_test.cc
View File

@ -15,6 +15,7 @@ limitations under the License.
#include "tensorflow/compiler/xla/literal.h"
#include <limits>
#include <vector>
#include "absl/base/casts.h"
@ -2009,8 +2010,7 @@ TEST_F(LiteralUtilTest, IsEqualAt) {
EXPECT_TRUE(c3.IsEqualAt({}, val_double));
EXPECT_TRUE(c3.IsEqualAt({}, val_integral));
EXPECT_TRUE(c3.IsEqualAt({}, val_complex));
double val_inf = 1. / 0;
EXPECT_FALSE(c3.IsEqualAt({}, val_inf));
EXPECT_FALSE(c3.IsEqualAt({}, std::numeric_limits<double>::infinity()));
complex128 val_true_complex = {10, 3};
complex64 val_smaller_complex = {10, 3};
Literal c4 = LiteralUtil::CreateR0<complex128>(val_true_complex);