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

fix an initialization problem in MklFusedBatchNorm

Browse Source
This commit is contained in:
Guozhong Zhuang 2020-03-24 16:43:40 -07:00
parent c5ff38aebf
commit 920f7e6264
1 changed files with 2 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -1008,7 +1008,7 @@ class MklFusedBatchNormOp : public OpKernel {
tf_shape_scale, mkl_shape_saved_mean);
DCHECK(*saved_mean_tensor);
// Set NAN mean value in case of empty input tensor
// Set 0 mean value in case of empty input tensor
auto saved_mean_data = (*saved_mean_tensor)->flat<U>().data();
std::fill_n(saved_mean_data, num_elements, static_cast<U>(0));
@ -1019,7 +1019,7 @@ class MklFusedBatchNormOp : public OpKernel {
mkl_shape_saved_variance);
DCHECK(*saved_variance_tensor);
// Set NAN variance value in case of empty input tensor
// Set 0 variance value in case of empty input tensor
auto saved_variance_data = (*saved_variance_tensor)->flat<U>().data();
std::fill_n(saved_variance_data, num_elements, static_cast<U>(0));
@ -1346,16 +1346,12 @@ class MklFusedBatchNormGradOp : public OpKernel {
mkl_shape_p.SetMklTensor(false);
AllocateOutputSetMklShape(context, kP1Index, &p1_tensor, TensorShape({}),
mkl_shape_p);
#ifndef ENABLE_MKLDNN_V1
std::fill_n(p1_tensor->flat<U>().data(), p1_tensor->shape().num_elements(),
static_cast<U>(0));
#endif // !ENABLE_MKLDNN_V1
AllocateOutputSetMklShape(context, kP2Index, &p2_tensor, TensorShape({}),
mkl_shape_p);
#ifndef ENABLE_MKLDNN_V1
std::fill_n(p2_tensor->flat<U>().data(), p2_tensor->shape().num_elements(),
static_cast<U>(0));
#endif // !ENABLE_MKLDNN_V1
}
memory::dims GetMeanVarianceDims() { return memory::dims({1, depth_}); }