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

Change conv_input_scale and side_input_scale from attributes to inputs for improved flexibility, in fused_conv2d_bias_activation op.

Browse Source 
PiperOrigin-RevId: 168311988
This commit is contained in:
A. Unique TensorFlower 2017-09-11 17:37:59 -07:00 committed by TensorFlower Gardener
parent 4b4e10f9c8
commit 60f15462be
3 changed files with 49 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
tensorflow/contrib/fused_conv/kernels/fused_conv2d_bias_activation_op.cc
View File

@ -41,6 +41,7 @@ limitations under the License.
namespace tensorflow {
namespace {
typedef Eigen::GpuDevice GPUDevice;
template <typename T>
@ -66,6 +67,7 @@ template <>
struct Int8x4ToInt32<int8> {
using type = int32;
};
} // namespace
// T is the element type of the conv_input, filter and side_input tensors.
// BiasType is the element type of the bias tensor, which can be different.
@ -73,9 +75,20 @@ struct Int8x4ToInt32<int8> {
template <typename Device, typename T, typename BiasType, typename ScaleType>
class FusedConv2DBiasActivationOp : public OpKernel {
public:
enum InputIndexes {
kConvInput = 0,
kFilter,
kBias,
kSideInput,
kConvInputScale,
kSideInputScale,
kNumInputs
};
explicit FusedConv2DBiasActivationOp(OpKernelConstruction* context)
: OpKernel(context) {
string data_format_str, filter_format_str;
CHECK_EQ(kNumInputs, context->num_inputs());
OP_REQUIRES_OK(context, context->GetAttr("data_format", &data_format_str));
OP_REQUIRES(context, FormatFromString(data_format_str, &data_format_),
errors::InvalidArgument("Invalid data format"));
@ -125,13 +138,6 @@ class FusedConv2DBiasActivationOp : public OpKernel {
errors::InvalidArgument("Current implementation only supports "
"RELU as the activation function."));
cudnn_use_autotune_ = CudnnUseAutotune();
float conv_input_scale_flt, side_input_scale_flt;
OP_REQUIRES_OK(context,
context->GetAttr("conv_input_scale", &conv_input_scale_flt));
OP_REQUIRES_OK(context,
context->GetAttr("side_input_scale", &side_input_scale_flt));
conv_input_scale_ = conv_input_scale_flt;
side_input_scale_ = side_input_scale_flt;
}
Status CheckShape(const Tensor& tensor, const string& tensor_name) {
@ -154,22 +160,30 @@ class FusedConv2DBiasActivationOp : public OpKernel {
void Compute(OpKernelContext* context) override {
// The conv_input tensor is one of the following formats:
// NHWC, NCHW, NCHW_VECT_C.
const Tensor& conv_input = context->input(0);
const Tensor& conv_input = context->input(kConvInput);
OP_REQUIRES_OK(context, CheckShape(conv_input, "conv_input"));
// The filter tensor is one of the following formats:
// HWIO, OIHW, OIHW_VECT_I.
const Tensor& filter = context->input(1);
const Tensor& filter = context->input(kFilter);
OP_REQUIRES_OK(context, CheckShape(filter, "filter"));
// Input bias is a 1-D tensor, with size matching output depth.
const Tensor& bias = context->input(2);
const Tensor& bias = context->input(kBias);
OP_REQUIRES_OK(context, CheckShape(bias, "conv_input"));
const Tensor& conv_input_scale_tensor = context->input(kConvInputScale);
const Tensor& side_input_scale_tensor = context->input(kSideInputScale);
auto conv_input_scale = *reinterpret_cast<const ScaleType*>(
conv_input_scale_tensor.tensor_data().data());
auto side_input_scale = *reinterpret_cast<const ScaleType*>(
side_input_scale_tensor.tensor_data().data());
// If side_input_scale != 0, then side_input is not ignored and
// has the same type and dimensions as the output.
const Tensor& side_input = context->input(3);
if (side_input_scale_ != 0) {
const Tensor& side_input = context->input(kSideInput);
if (side_input_scale != 0) {
OP_REQUIRES_OK(context, CheckShape(side_input, "side_input"));
}
@ -212,10 +226,10 @@ class FusedConv2DBiasActivationOp : public OpKernel {
return;
}
launcher_.launch(context, cudnn_use_autotune_, conv_input,
conv_input_scale_, filter, stride_rows_, stride_cols_,
eigen_padding_type_, side_input, side_input_scale_, bias,
activation_mode_, data_format_, filter_format_, output);
launcher_.launch(context, cudnn_use_autotune_, conv_input, conv_input_scale,
filter, stride_rows_, stride_cols_, eigen_padding_type_,
side_input, side_input_scale, bias, activation_mode_,
data_format_, filter_format_, output);
}
private:
@ -225,8 +239,6 @@ class FusedConv2DBiasActivationOp : public OpKernel {
ActivationMode activation_mode_;
TensorFormat data_format_;
FilterTensorFormat filter_format_;
ScaleType conv_input_scale_;
ScaleType side_input_scale_;
LaunchFusedConv2DBiasActivationOp<Device, T, BiasType, ScaleType> launcher_;
bool cudnn_use_autotune_;
@ -579,14 +591,18 @@ REGISTER_KERNEL_BUILDER(
Name("FusedConv2DBiasActivation")
.Device(DEVICE_GPU)
.TypeConstraint<float>("T")
.TypeConstraint<float>("Tbias"),
.TypeConstraint<float>("Tbias")
.HostMemory("conv_input_scale")
.HostMemory("side_input_scale"),
FusedConv2DBiasActivationOp<GPUDevice, float, float, float>);
REGISTER_KERNEL_BUILDER(
Name("FusedConv2DBiasActivation")
.Device(DEVICE_GPU)
.TypeConstraint<qint8>("T")
.TypeConstraint<float>("Tbias"),
.TypeConstraint<float>("Tbias")
.HostMemory("conv_input_scale")
.HostMemory("side_input_scale"),
FusedConv2DBiasActivationOp<GPUDevice, qint8, float, float>);
#endif // GOOGLE_CUDA

15
tensorflow/contrib/fused_conv/ops/fused_conv2d_bias_activation_op.cc
View File

@ -42,11 +42,11 @@ REGISTER_OP("FusedConv2DBiasActivation")
.Input("filter: T")
.Input("bias: Tbias")
.Input("side_input: T")
.Input("conv_input_scale: float")
.Input("side_input_scale: float")
.Output("output: T")
.Attr("T: {float, half, qint8}")
.Attr("Tbias: {float, half}")
.Attr("conv_input_scale: float = 1.0")
.Attr("side_input_scale: float = 0.0")
.Attr("strides: list(int)")
.Attr(GetPaddingAttrString())
.Attr("data_format: {'NHWC', 'NCHW', 'NCHW_VECT_C'} = 'NHWC'")
@ -97,6 +97,11 @@ REGISTER_OP("FusedConv2DBiasActivation")
TF_RETURN_IF_ERROR(c->Merge(side_input_shape, c->output(0), &unused));
}
// Check that conv_input_scale and side_input_scale are scalar tensors.
ShapeHandle unused;
TF_RETURN_IF_ERROR(c->WithRank(c->input(4), 0, &unused));
TF_RETURN_IF_ERROR(c->WithRank(c->input(5), 0, &unused));
return Status::OK();
})
.Doc(R"doc(
@ -117,15 +122,15 @@ REGISTER_OP("FusedConv2DBiasActivation")
side_input: A tensor with format as specified by `data_format` (see below).
This tensor will be ignored and can be [] if side_input_scale == 0.
Otherwise, the size of each dimension must match the `output` tensor.
conv_input_scale: scalar float value to be multiplied by `conv_input`.
(conceptually.. in reality it is applied after convolution).
side_input_scale: scalar float value to be multiplied by `side_input`.
output: A tensor with format as specified by `data_format` (see below).
The dimension sizes are determined automatically based on other inputs
and attributes.
T: The element data type of `conv_input`, `side_input` and `output` tensors.
Note: must match with the `data_format`.
Tbias: The element data type of `bias`.
conv_input_scale: scalar float value to be multiplied by `conv_input`.
(conceptually.. in reality it is applied after convolution).
side_input_scale: scalar float value to be multiplied by `side_input`.
strides: 1-D tensor of length 4. The stride of the sliding window for each
dimension of `input`. The dimension order is determined by the value of
`data_format`, see below for details.

6
tensorflow/contrib/fused_conv/python/ops/fused_conv2d_bias_activation_op.py
View File

@ -97,11 +97,11 @@ def fused_conv2d_bias_activation(conv_input,
conv_input,
filter,
bias,
side_input,
conv_input_scale,
side_input_scale,
padding=padding,
strides=strides,
conv_input_scale=conv_input_scale,
side_input_scale=side_input_scale,
side_input=side_input,
activation_mode=activation_mode,
data_format=data_format,
filter_format=filter_format,