diff options
| author |  Justin Lebar <jlebar@google.com> | 2018-04-23 17:16:55 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2018-04-23 17:19:22 -0700 |
| commit | 80fc661853f9a0844faf95eb68438dc85a5879e3 (patch) | |
| tree | e7328a367c4f4db472063f96c983b70e63aa3f26 /tensorflow/core/kernels/fused_batch_norm_op.cc | |
| parent | ecd837fd0ab69cf54d920eae3b1c73602be6c626 (diff) | |
Use tensorflow::se instead of perftools::gputools for StreamExecutor.
PiperOrigin-RevId: 194010749
Diffstat (limited to 'tensorflow/core/kernels/fused_batch_norm_op.cc')
| -rw-r--r-- | tensorflow/core/kernels/fused_batch_norm_op.cc | 22 |
1 files changed, 11 insertions, 11 deletions
diff --git a/tensorflow/core/kernels/fused_batch_norm_op.cc b/tensorflow/core/kernels/fused_batch_norm_op.cc index 9b4dca8511..f99dd643f7 100644 --- a/tensorflow/core/kernels/fused_batch_norm_op.cc +++ b/tensorflow/core/kernels/fused_batch_norm_op.cc @@ -251,7 +251,7 @@ struct FusedBatchNorm<GPUDevice, T, U> { Tensor x_maybe_transformed = x; Tensor x_transformed; Tensor y_transformed; - perftools::gputools::DeviceMemory<T> y_ptr; + se::DeviceMemory<T> y_ptr; if (tensor_format == FORMAT_NCHW) { y_ptr = StreamExecutorUtil::AsDeviceMemory<T>(*y); @@ -279,19 +279,19 @@ struct FusedBatchNorm<GPUDevice, T, U> { return; } - perftools::gputools::dnn::BatchDescriptor x_desc; + se::dnn::BatchDescriptor x_desc; x_desc.set_count(batch_size) .set_feature_map_count(channels) .set_height(height) .set_width(width) - .set_layout(perftools::gputools::dnn::DataLayout::kBatchDepthYX); + .set_layout(se::dnn::DataLayout::kBatchDepthYX); - perftools::gputools::dnn::BatchDescriptor scale_offset_desc; + se::dnn::BatchDescriptor scale_offset_desc; scale_offset_desc.set_count(1) .set_feature_map_count(channels) .set_height(1) .set_width(1) - .set_layout(perftools::gputools::dnn::DataLayout::kBatchDepthYX); + .set_layout(se::dnn::DataLayout::kBatchDepthYX); auto x_ptr = StreamExecutorUtil::AsDeviceMemory<T>(x_maybe_transformed); auto scale_ptr = StreamExecutorUtil::AsDeviceMemory<U>(scale); @@ -308,7 +308,7 @@ struct FusedBatchNorm<GPUDevice, T, U> { StreamExecutorUtil::AsDeviceMemory<U>(*saved_inv_var); GPUDevice d = context->eigen_device<GPUDevice>(); - using perftools::gputools::DeviceMemory; + using se::DeviceMemory; Tensor inv_var; OP_REQUIRES_OK( context, context->allocate_temp(DataTypeToEnum<U>::value, @@ -390,7 +390,7 @@ struct FusedBatchNormGrad<GPUDevice, T, U> { // Outputs Tensor x_backprop_transformed; - perftools::gputools::DeviceMemory<T> x_backprop_ptr; + se::DeviceMemory<T> x_backprop_ptr; if (tensor_format == FORMAT_NCHW) { x_backprop_ptr = StreamExecutorUtil::AsDeviceMemory<T>(*x_backprop); @@ -433,19 +433,19 @@ struct FusedBatchNormGrad<GPUDevice, T, U> { return; } - perftools::gputools::dnn::BatchDescriptor x_desc; + se::dnn::BatchDescriptor x_desc; x_desc.set_count(batch_size) .set_feature_map_count(channels) .set_height(height) .set_width(width) - .set_layout(perftools::gputools::dnn::DataLayout::kBatchDepthYX); + .set_layout(se::dnn::DataLayout::kBatchDepthYX); - perftools::gputools::dnn::BatchDescriptor scale_offset_desc; + se::dnn::BatchDescriptor scale_offset_desc; scale_offset_desc.set_count(1) .set_feature_map_count(channels) .set_height(1) .set_width(1) - .set_layout(perftools::gputools::dnn::DataLayout::kBatchDepthYX); + .set_layout(se::dnn::DataLayout::kBatchDepthYX); auto y_backprop_ptr = StreamExecutorUtil::AsDeviceMemory<T>(y_backprop_maybe_transformed); |