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

Segmentation Fault 11 thrown by tensorflow::HistogramFixedWidthFunctor::Compute in histogram_op.cc

Browse Source 
Occasionally tensorflow::HistogramFixedWidthFunctor::Compute triggers a
segmentation fault 11. It happens when the argument i of index_to_bin() is
negative, which is caused by casting a big (x-a)/step int64 value that exceeds
int32 range to int32. Switching the order of cwiseMin and cast fixes the
problem.

PiperOrigin-RevId: 257647164
This commit is contained in:
A. Unique TensorFlower 2019-07-11 11:27:39 -07:00 committed by TensorFlower Gardener
parent d251e8c923
commit d12c671f01
1 changed files with 6 additions and 2 deletions

8
tensorflow/core/kernels/histogram_op.cc
View File

@ -48,18 +48,22 @@ struct HistogramFixedWidthFunctor<CPUDevice, T, Tout> {
const double step = static_cast<double>(value_range(1) - value_range(0)) /
static_cast<double>(nbins);
const double nbins_minus_1 = static_cast<double>(nbins - 1);
// The calculation is done by finding the slot of each value in `values`.
// With [a, b]:
// step = (b - a) / nbins
// (x - a) / step
// , then the entries are mapped to output.
// Bug fix: Switch the order of cwiseMin and int32-casting to avoid
// producing a negative index when casting an big int64 number to int32
index_to_bin.device(d) =
((values.cwiseMax(value_range(0)) - values.constant(value_range(0)))
.template cast<double>() /
step)
.template cast<int32>()
.cwiseMin(nbins - 1);
.cwiseMin(nbins_minus_1)
.template cast<int32>();
out.setZero();
for (int32 i = 0; i < index_to_bin.size(); i++) {