Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into lod_tensor2
dangqingqing
8 years ago
commit
30a58b5121
Binary file not shown.
|
Before Width: | Height: | Size: 24 KiB After Width: | Height: | Size: 24 KiB |
@ -0,0 +1,109 @@
|
|||||||
|
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License. */
|
||||||
|
|
||||||
|
#include "paddle/operators/elementwise_mul_op.h"
|
||||||
|
|
||||||
|
namespace paddle {
|
||||||
|
namespace operators {
|
||||||
|
|
||||||
|
using Tensor = framework::Tensor;
|
||||||
|
|
||||||
|
class ElementWiseMulOp : public framework::OperatorWithKernel {
|
||||||
|
public:
|
||||||
|
using framework::OperatorWithKernel::OperatorWithKernel;
|
||||||
|
|
||||||
|
protected:
|
||||||
|
void InferShape(const framework::InferShapeContext &ctx) const override {
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "Input(X) should not be null");
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("Y"), "Input(Y) should not be null");
|
||||||
|
auto x_dim = ctx.Input<Tensor>("X")->dims();
|
||||||
|
auto y_dim = ctx.Input<Tensor>("Y")->dims();
|
||||||
|
PADDLE_ENFORCE_GE(x_dim.size(), y_dim.size(),
|
||||||
|
"Rank of first input must >= rank of second input.")
|
||||||
|
ctx.Output<Tensor>("Out")->Resize(x_dim);
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
class ElementWiseMulOpMaker : public framework::OpProtoAndCheckerMaker {
|
||||||
|
public:
|
||||||
|
ElementWiseMulOpMaker(framework::OpProto *proto,
|
||||||
|
framework::OpAttrChecker *op_checker)
|
||||||
|
: OpProtoAndCheckerMaker(proto, op_checker) {
|
||||||
|
AddInput("X", "The first input of elementwise mul op");
|
||||||
|
AddInput("Y", "The second input of elementwise mul op");
|
||||||
|
AddAttr<int>("axis",
|
||||||
|
R"DOC(
|
||||||
|
When shape(Y) does not equal shape(X),Y will be broadcasted
|
||||||
|
to match the shape of X and axis should be dimension index Y in X
|
||||||
|
)DOC")
|
||||||
|
.SetDefault(-1)
|
||||||
|
.EqualGreaterThan(-1);
|
||||||
|
|
||||||
|
AddOutput("Out", "The output of elementwise mul op");
|
||||||
|
AddComment(R"DOC(
|
||||||
|
Limited elementwise multiple operator.The equation is: Out = X ⊙ Y.
|
||||||
|
1. The shape of Y should be same with X or
|
||||||
|
2. Y's shape is a subset of X.
|
||||||
|
Y will be broadcasted to match the shape of X and axis should be dimension index Y in X.
|
||||||
|
example:
|
||||||
|
shape(X) = (2, 3, 4, 5), shape(Y) = (,)
|
||||||
|
shape(X) = (2, 3, 4, 5), shape(Y) = (5,)
|
||||||
|
shape(X) = (2, 3, 4, 5), shape(Y) = (4, 5)
|
||||||
|
shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1
|
||||||
|
shape(X) = (2, 3, 4, 5), shape(Y) = (2), with axis=0
|
||||||
|
)DOC");
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
class ElementWiseMulOpGrad : public framework::OperatorWithKernel {
|
||||||
|
public:
|
||||||
|
using framework::OperatorWithKernel::OperatorWithKernel;
|
||||||
|
|
||||||
|
protected:
|
||||||
|
void InferShape(const framework::InferShapeContext &ctx) const override {
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "Input(X) should not be null");
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("Y"), "Input(Y) should not be null");
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar(framework::GradVarName("Out")),
|
||||||
|
"Input(Out@GRAD) should not be null");
|
||||||
|
|
||||||
|
auto x_dims = ctx.Input<Tensor>("X")->dims();
|
||||||
|
auto y_dims = ctx.Input<Tensor>("Y")->dims();
|
||||||
|
auto out_dims = ctx.Input<Tensor>(framework::GradVarName("Out"))->dims();
|
||||||
|
auto *x_grad = ctx.Output<Tensor>(framework::GradVarName("X"));
|
||||||
|
auto *y_grad = ctx.Output<Tensor>(framework::GradVarName("Y"));
|
||||||
|
|
||||||
|
PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(),
|
||||||
|
"Rank of first input must >= rank of second input.")
|
||||||
|
|
||||||
|
if (x_grad) {
|
||||||
|
x_grad->Resize(x_dims);
|
||||||
|
}
|
||||||
|
|
||||||
|
if (y_grad) {
|
||||||
|
y_grad->Resize(y_dims);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
} // namespace operators
|
||||||
|
} // namespace paddle
|
||||||
|
|
||||||
|
namespace ops = paddle::operators;
|
||||||
|
REGISTER_OP(elementwise_mul, ops::ElementWiseMulOp, ops::ElementWiseMulOpMaker,
|
||||||
|
elementwise_mul_grad, ops::ElementWiseMulOpGrad);
|
||||||
|
REGISTER_OP_CPU_KERNEL(
|
||||||
|
elementwise_mul,
|
||||||
|
ops::ElementWiseMulKernel<paddle::platform::CPUPlace, float>);
|
||||||
|
REGISTER_OP_CPU_KERNEL(
|
||||||
|
elementwise_mul_grad,
|
||||||
|
ops::ElementWiseMulGradKernel<paddle::platform::CPUPlace, float>);
|
||||||
@ -0,0 +1,25 @@
|
|||||||
|
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License. */
|
||||||
|
|
||||||
|
#define EIGEN_USE_GPU
|
||||||
|
#include "paddle/operators/elementwise_mul_op.h"
|
||||||
|
|
||||||
|
namespace ops = paddle::operators;
|
||||||
|
|
||||||
|
REGISTER_OP_GPU_KERNEL(
|
||||||
|
elementwise_mul,
|
||||||
|
ops::ElementWiseMulKernel<paddle::platform::GPUPlace, float>);
|
||||||
|
REGISTER_OP_GPU_KERNEL(
|
||||||
|
elementwise_mul_grad,
|
||||||
|
ops::ElementWiseMulGradKernel<paddle::platform::GPUPlace, float>);
|
||||||
@ -0,0 +1,185 @@
|
|||||||
|
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License. */
|
||||||
|
|
||||||
|
#pragma once
|
||||||
|
#include <iostream>
|
||||||
|
#include "paddle/framework/eigen.h"
|
||||||
|
#include "paddle/framework/op_registry.h"
|
||||||
|
#include "paddle/operators/math/math_function.h"
|
||||||
|
|
||||||
|
namespace paddle {
|
||||||
|
namespace operators {
|
||||||
|
/*
|
||||||
|
* Out = X ⊙ Y
|
||||||
|
* 1. shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1
|
||||||
|
* pre=2, n=3*4, post=5
|
||||||
|
* 2. shape(X) = (2, 3, 4, 5), shape(Y) = (4,5)
|
||||||
|
* pre=2*3, n=4*5, post=1
|
||||||
|
*/
|
||||||
|
|
||||||
|
inline void get_mid_dims(const framework::DDim& x_dims,
|
||||||
|
const framework::DDim& y_dims, const int axis,
|
||||||
|
int& pre, int& n, int& post) {
|
||||||
|
pre = 1;
|
||||||
|
n = 1;
|
||||||
|
post = 1;
|
||||||
|
for (int i = 0; i < axis; ++i) {
|
||||||
|
pre *= x_dims[i];
|
||||||
|
}
|
||||||
|
|
||||||
|
for (int i = 0; i < y_dims.size(); ++i) {
|
||||||
|
PADDLE_ENFORCE_EQ(x_dims[i + axis], y_dims[i],
|
||||||
|
"Broadcast dimension mismatch.");
|
||||||
|
n *= y_dims[i];
|
||||||
|
}
|
||||||
|
|
||||||
|
for (int i = axis + y_dims.size(); i < x_dims.size(); ++i) {
|
||||||
|
post *= x_dims[i];
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
template <typename Place, typename T>
|
||||||
|
class ElementWiseMulKernel : public framework::OpKernel {
|
||||||
|
public:
|
||||||
|
void Compute(const framework::ExecutionContext& ctx) const override {
|
||||||
|
using Tensor = framework::Tensor;
|
||||||
|
|
||||||
|
auto* x = ctx.Input<Tensor>("X");
|
||||||
|
auto* y = ctx.Input<Tensor>("Y");
|
||||||
|
auto* z = ctx.Output<Tensor>("Out");
|
||||||
|
z->mutable_data<T>(ctx.GetPlace());
|
||||||
|
|
||||||
|
auto x_e = framework::EigenVector<T>::Flatten(*x);
|
||||||
|
auto y_e = framework::EigenVector<T>::Flatten(*y);
|
||||||
|
auto z_e = framework::EigenVector<T>::Flatten(*z);
|
||||||
|
|
||||||
|
auto x_dims = x->dims();
|
||||||
|
auto y_dims = y->dims();
|
||||||
|
PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(),
|
||||||
|
"Rank of first input must >= rank of second input.")
|
||||||
|
|
||||||
|
if (x_dims == y_dims || product(y_dims) == 1) {
|
||||||
|
z_e.device(ctx.GetEigenDevice<Place>()) = x_e * y_e;
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
|
int axis = ctx.Attr<int>("axis");
|
||||||
|
axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
|
||||||
|
PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(),
|
||||||
|
"Axis should be in range [0, x_dims)");
|
||||||
|
|
||||||
|
int pre, n, post;
|
||||||
|
get_mid_dims(x_dims, y_dims, axis, pre, n, post);
|
||||||
|
if (post == 1) {
|
||||||
|
auto y_bcast = y_e.reshape(Eigen::DSizes<int, 2>(1, n))
|
||||||
|
.broadcast(Eigen::DSizes<int, 2>(pre, 1))
|
||||||
|
.reshape(Eigen::DSizes<int, 1>(x_e.size()));
|
||||||
|
z_e.device(ctx.GetEigenDevice<Place>()) = x_e * y_bcast;
|
||||||
|
return;
|
||||||
|
} else {
|
||||||
|
auto y_bcast = y_e.reshape(Eigen::DSizes<int, 3>(1, n, 1))
|
||||||
|
.broadcast(Eigen::DSizes<int, 3>(pre, 1, post))
|
||||||
|
.reshape(Eigen::DSizes<int, 1>(x_e.size()));
|
||||||
|
z_e.device(ctx.GetEigenDevice<Place>()) = x_e * y_bcast;
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
template <typename Place, typename T>
|
||||||
|
class ElementWiseMulGradKernel : public framework::OpKernel {
|
||||||
|
public:
|
||||||
|
void Compute(const framework::ExecutionContext& ctx) const override {
|
||||||
|
using Tensor = framework::Tensor;
|
||||||
|
|
||||||
|
auto* x = ctx.Input<Tensor>("X");
|
||||||
|
auto* y = ctx.Input<Tensor>("Y");
|
||||||
|
auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
|
||||||
|
|
||||||
|
auto x_e = framework::EigenVector<T>::Flatten(*x);
|
||||||
|
auto y_e = framework::EigenVector<T>::Flatten(*y);
|
||||||
|
auto dout_e = framework::EigenVector<T>::Flatten(*dout);
|
||||||
|
|
||||||
|
auto x_dims = x->dims();
|
||||||
|
auto y_dims = y->dims();
|
||||||
|
|
||||||
|
auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
|
||||||
|
auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
|
||||||
|
if (dx) {
|
||||||
|
dx->mutable_data<T>(ctx.GetPlace());
|
||||||
|
}
|
||||||
|
if (dy) {
|
||||||
|
dy->mutable_data<T>(ctx.GetPlace());
|
||||||
|
}
|
||||||
|
|
||||||
|
if (x_dims == y_dims || product(y_dims) == 1) {
|
||||||
|
if (dx) {
|
||||||
|
auto dx_e = framework::EigenVector<T>::Flatten(*dx);
|
||||||
|
dx_e.device(ctx.GetEigenDevice<Place>()) = dout_e * y_e;
|
||||||
|
}
|
||||||
|
|
||||||
|
if (dy) {
|
||||||
|
auto dy_e = framework::EigenVector<T>::Flatten(*dy);
|
||||||
|
dy_e.device(ctx.GetEigenDevice<Place>()) = x_e * dout_e;
|
||||||
|
}
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
|
int axis = ctx.Attr<int>("axis");
|
||||||
|
axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
|
||||||
|
|
||||||
|
int pre, n, post;
|
||||||
|
get_mid_dims(x_dims, y_dims, axis, pre, n, post);
|
||||||
|
|
||||||
|
// TODO(gongweibao): wrap reshape to a function.
|
||||||
|
if (post == 1) {
|
||||||
|
auto y_e_bcast = y_e.reshape(Eigen::DSizes<int, 2>(1, n))
|
||||||
|
.broadcast(Eigen::DSizes<int, 2>(pre, 1))
|
||||||
|
.reshape(Eigen::DSizes<int, 1>(x_e.size()));
|
||||||
|
if (dx) {
|
||||||
|
auto dx_e = framework::EigenVector<T>::Flatten(*dx);
|
||||||
|
dx_e.device(ctx.GetEigenDevice<Place>()) = dout_e * y_e_bcast;
|
||||||
|
}
|
||||||
|
|
||||||
|
if (dy) {
|
||||||
|
auto dy_e = framework::EigenVector<T>::Flatten(*dy);
|
||||||
|
dy_e.device(ctx.GetEigenDevice<Place>()) =
|
||||||
|
(x_e * dout_e)
|
||||||
|
.reshape(Eigen::DSizes<int, 2>(pre, n))
|
||||||
|
.sum(Eigen::array<int, 1>{{0}});
|
||||||
|
}
|
||||||
|
return;
|
||||||
|
} else {
|
||||||
|
auto y_e_bcast = y_e.reshape(Eigen::DSizes<int, 3>(1, n, 1))
|
||||||
|
.broadcast(Eigen::DSizes<int, 3>(pre, 1, post))
|
||||||
|
.reshape(Eigen::DSizes<int, 1>(x_e.size()));
|
||||||
|
if (dx) {
|
||||||
|
auto dx_e = framework::EigenVector<T>::Flatten(*dx);
|
||||||
|
dx_e.device(ctx.GetEigenDevice<Place>()) = dout_e * y_e_bcast;
|
||||||
|
}
|
||||||
|
|
||||||
|
if (dy) {
|
||||||
|
auto dy_e = framework::EigenVector<T>::Flatten(*dy);
|
||||||
|
dy_e.device(ctx.GetEigenDevice<Place>()) =
|
||||||
|
(x_e * dout_e)
|
||||||
|
.reshape(Eigen::DSizes<int, 3>(pre, n, post))
|
||||||
|
.sum(Eigen::array<int, 2>{{0, 2}});
|
||||||
|
}
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
} // namespace operators
|
||||||
|
} // namespace paddle
|
||||||
@ -0,0 +1,112 @@
|
|||||||
|
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License. */
|
||||||
|
|
||||||
|
#include "paddle/operators/pad_op.h"
|
||||||
|
|
||||||
|
namespace paddle {
|
||||||
|
namespace operators {
|
||||||
|
|
||||||
|
using framework::Tensor;
|
||||||
|
|
||||||
|
class PadOp : public framework::OperatorWithKernel {
|
||||||
|
public:
|
||||||
|
using framework::OperatorWithKernel::OperatorWithKernel;
|
||||||
|
|
||||||
|
protected:
|
||||||
|
void InferShape(const framework::InferShapeContext &ctx) const override {
|
||||||
|
auto x_dim = ctx.Input<Tensor>("X")->dims();
|
||||||
|
auto paddings = Attr<std::vector<int>>("paddings");
|
||||||
|
PADDLE_ENFORCE_EQ(x_dim.size() * 2, int64_t(paddings.size()),
|
||||||
|
"Size of paddings should be equal to 2 * dimension size "
|
||||||
|
"of input tensor.");
|
||||||
|
std::vector<int64_t> out_dims(x_dim.size());
|
||||||
|
for (int i = 0; i < x_dim.size(); ++i) {
|
||||||
|
out_dims[i] = x_dim[i] + paddings[i * 2] + paddings[i * 2 + 1];
|
||||||
|
}
|
||||||
|
ctx.Output<Tensor>("Out")->Resize(framework::make_ddim(out_dims));
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
class PadOpMaker : public framework::OpProtoAndCheckerMaker {
|
||||||
|
public:
|
||||||
|
PadOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
|
||||||
|
: OpProtoAndCheckerMaker(proto, op_checker) {
|
||||||
|
AddInput("X",
|
||||||
|
"The input of pad op. "
|
||||||
|
"The input should be a k-D tensor(k > 0 and k < 7)");
|
||||||
|
AddOutput("Out",
|
||||||
|
"The output of pad op."
|
||||||
|
"A tensor with the same shape as X.")
|
||||||
|
.NotInGradient();
|
||||||
|
AddComment(R"DOC(
|
||||||
|
Pad input into output, as specified by paddings and pad_value. The input should be a k-D tensor(k > 0 and k < 7). As an example:
|
||||||
|
|
||||||
|
Given:
|
||||||
|
|
||||||
|
X = [[1, 2],
|
||||||
|
[3, 4]]
|
||||||
|
|
||||||
|
and
|
||||||
|
|
||||||
|
paddings = [0, 1, 1, 2]
|
||||||
|
|
||||||
|
and
|
||||||
|
|
||||||
|
pad_value = 0
|
||||||
|
|
||||||
|
then we get
|
||||||
|
|
||||||
|
Out = [[0, 1, 2, 0, 0]
|
||||||
|
[0, 3, 4, 0, 0]
|
||||||
|
[0, 0, 0, 0, 0]]
|
||||||
|
)DOC");
|
||||||
|
AddAttr<std::vector<int>>(
|
||||||
|
"paddings",
|
||||||
|
"A list<int> to describes padding rules for each dimension."
|
||||||
|
" For 2-D image tensor, paddings=[0, 1, 2, 3] means"
|
||||||
|
" padding 0 row to top, 1 row to bottom, 2 columns to left"
|
||||||
|
" and 3 columns to right.Size of paddings should be equal to"
|
||||||
|
" 2 * dimension size of input tensor.");
|
||||||
|
AddAttr<float>("pad_value",
|
||||||
|
"(float) default to 0; "
|
||||||
|
"The value to fill padded areas.")
|
||||||
|
.SetDefault(0.0f);
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
class PadOpGrad : public framework::OperatorWithKernel {
|
||||||
|
public:
|
||||||
|
using framework::OperatorWithKernel::OperatorWithKernel;
|
||||||
|
|
||||||
|
protected:
|
||||||
|
void InferShape(const framework::InferShapeContext &ctx) const override {
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "Input(X) should not be null");
|
||||||
|
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar(framework::GradVarName("Out")),
|
||||||
|
"Input(Out@GRAD) should not be null");
|
||||||
|
auto x_dims = ctx.Input<Tensor>("X")->dims();
|
||||||
|
auto *x_grad = ctx.Output<Tensor>(framework::GradVarName("X"));
|
||||||
|
if (x_grad != nullptr) {
|
||||||
|
x_grad->Resize(x_dims);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
} // namespace operators
|
||||||
|
} // namespace paddle
|
||||||
|
|
||||||
|
namespace ops = paddle::operators;
|
||||||
|
REGISTER_OP(pad, ops::PadOp, ops::PadOpMaker, pad_grad, ops::PadOpGrad);
|
||||||
|
REGISTER_OP_CPU_KERNEL(pad, ops::PadKernel<paddle::platform::CPUPlace, float>);
|
||||||
|
REGISTER_OP_CPU_KERNEL(pad_grad,
|
||||||
|
ops::PadGradKernel<paddle::platform::CPUPlace, float>);
|
||||||
@ -0,0 +1,21 @@
|
|||||||
|
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License. */
|
||||||
|
|
||||||
|
#define EIGEN_USE_GPU
|
||||||
|
#include "paddle/operators/pad_op.h"
|
||||||
|
|
||||||
|
namespace ops = paddle::operators;
|
||||||
|
REGISTER_OP_GPU_KERNEL(pad, ops::PadKernel<paddle::platform::GPUPlace, float>);
|
||||||
|
REGISTER_OP_GPU_KERNEL(pad_grad,
|
||||||
|
ops::PadGradKernel<paddle::platform::GPUPlace, float>);
|
||||||
@ -0,0 +1,132 @@
|
|||||||
|
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||||
|
|
||||||
|
Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
you may not use this file except in compliance with the License.
|
||||||
|
You may obtain a copy of the License at
|
||||||
|
|
||||||
|
http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
|
||||||
|
Unless required by applicable law or agreed to in writing, software
|
||||||
|
distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
See the License for the specific language governing permissions and
|
||||||
|
limitations under the License. */
|
||||||
|
|
||||||
|
#pragma once
|
||||||
|
|
||||||
|
#include "paddle/framework/eigen.h"
|
||||||
|
#include "paddle/framework/op_registry.h"
|
||||||
|
|
||||||
|
namespace paddle {
|
||||||
|
namespace operators {
|
||||||
|
|
||||||
|
using Tensor = framework::Tensor;
|
||||||
|
|
||||||
|
template <typename T, size_t D, int MajorType = Eigen::RowMajor,
|
||||||
|
typename IndexType = Eigen::DenseIndex>
|
||||||
|
using EigenTensor = framework::EigenTensor<T, D, MajorType, IndexType>;
|
||||||
|
|
||||||
|
template <typename Place, typename T, size_t D>
|
||||||
|
void PadFunction(const framework::ExecutionContext& context) {
|
||||||
|
auto pads = context.Attr<std::vector<int>>("paddings");
|
||||||
|
Eigen::array<std::pair<int, int>, D> paddings;
|
||||||
|
for (size_t i = 0; i < paddings.size(); ++i) {
|
||||||
|
paddings[i].first = pads[i * 2];
|
||||||
|
paddings[i].second = pads[i * 2 + 1];
|
||||||
|
}
|
||||||
|
T pad_value = context.Attr<T>("pad_value");
|
||||||
|
|
||||||
|
auto* x = context.Input<Tensor>("X");
|
||||||
|
auto* out = context.Output<Tensor>("Out");
|
||||||
|
out->mutable_data<T>(context.GetPlace());
|
||||||
|
|
||||||
|
auto x_tensor = EigenTensor<T, D>::From(*x);
|
||||||
|
auto out_tensor = EigenTensor<T, D>::From(*out);
|
||||||
|
auto place = context.GetEigenDevice<Place>();
|
||||||
|
out_tensor.device(place) = x_tensor.pad(paddings, pad_value);
|
||||||
|
}
|
||||||
|
|
||||||
|
template <typename Place, typename T>
|
||||||
|
class PadKernel : public framework::OpKernel {
|
||||||
|
public:
|
||||||
|
void Compute(const framework::ExecutionContext& context) const override {
|
||||||
|
int rank = context.Input<Tensor>("X")->dims().size();
|
||||||
|
switch (rank) {
|
||||||
|
case 1:
|
||||||
|
PadFunction<Place, T, 1>(context);
|
||||||
|
break;
|
||||||
|
case 2:
|
||||||
|
PadFunction<Place, T, 2>(context);
|
||||||
|
break;
|
||||||
|
case 3:
|
||||||
|
PadFunction<Place, T, 3>(context);
|
||||||
|
break;
|
||||||
|
case 4:
|
||||||
|
PadFunction<Place, T, 4>(context);
|
||||||
|
break;
|
||||||
|
case 5:
|
||||||
|
PadFunction<Place, T, 5>(context);
|
||||||
|
break;
|
||||||
|
case 6:
|
||||||
|
PadFunction<Place, T, 6>(context);
|
||||||
|
break;
|
||||||
|
default:
|
||||||
|
PADDLE_THROW(
|
||||||
|
"PadOp only support tensors with no more than 6 dimensions.");
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
template <typename Place, typename T, size_t D>
|
||||||
|
void PadGradFunction(const framework::ExecutionContext& context) {
|
||||||
|
auto pads = context.Attr<std::vector<int>>("paddings");
|
||||||
|
Eigen::array<std::pair<int, int>, D> paddings;
|
||||||
|
for (size_t i = 0; i < paddings.size(); ++i) {
|
||||||
|
paddings[i].first = -pads[i * 2];
|
||||||
|
paddings[i].second = -pads[i * 2 + 1];
|
||||||
|
}
|
||||||
|
auto* d_out = context.Input<Tensor>(framework::GradVarName("Out"));
|
||||||
|
auto* d_x = context.Output<Tensor>(framework::GradVarName("X"));
|
||||||
|
if (d_x != nullptr) {
|
||||||
|
d_x->mutable_data<T>(context.GetPlace());
|
||||||
|
auto d_x_tensor = EigenTensor<T, D>::From(*d_x);
|
||||||
|
auto d_out_tensor = EigenTensor<T, D>::From(*d_out);
|
||||||
|
auto place = context.GetEigenDevice<Place>();
|
||||||
|
d_x_tensor.device(place) = d_out_tensor.pad(paddings, 0);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
template <typename Place, typename T>
|
||||||
|
class PadGradKernel : public framework::OpKernel {
|
||||||
|
public:
|
||||||
|
void Compute(const framework::ExecutionContext& context) const override {
|
||||||
|
size_t rank =
|
||||||
|
context.Input<Tensor>(framework::GradVarName("Out"))->dims().size();
|
||||||
|
switch (rank) {
|
||||||
|
case 1:
|
||||||
|
PadGradFunction<Place, T, 1>(context);
|
||||||
|
break;
|
||||||
|
case 2:
|
||||||
|
PadGradFunction<Place, T, 2>(context);
|
||||||
|
break;
|
||||||
|
case 3:
|
||||||
|
PadGradFunction<Place, T, 3>(context);
|
||||||
|
break;
|
||||||
|
case 4:
|
||||||
|
PadGradFunction<Place, T, 4>(context);
|
||||||
|
break;
|
||||||
|
case 5:
|
||||||
|
PadGradFunction<Place, T, 5>(context);
|
||||||
|
break;
|
||||||
|
case 6:
|
||||||
|
PadGradFunction<Place, T, 6>(context);
|
||||||
|
break;
|
||||||
|
default:
|
||||||
|
PADDLE_THROW(
|
||||||
|
"PadOp only support tensors with no more than 6 dimensions.");
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
} // namespace operators
|
||||||
|
} // namespace paddle
|
||||||
@ -0,0 +1,157 @@
|
|||||||
|
import unittest
|
||||||
|
import numpy as np
|
||||||
|
from op_test import OpTest
|
||||||
|
|
||||||
|
|
||||||
|
class TestElementwiseMulOp_Matrix(OpTest):
|
||||||
|
def setUp(self):
|
||||||
|
self.op_type = "elementwise_mul"
|
||||||
|
""" Warning
|
||||||
|
CPU gradient check error!
|
||||||
|
'X': np.random.random((32,84)).astype("float32"),
|
||||||
|
'Y': np.random.random((32,84)).astype("float32")
|
||||||
|
"""
|
||||||
|
self.inputs = {
|
||||||
|
'X': np.random.uniform(0.1, 1, [13, 17]).astype("float32"),
|
||||||
|
'Y': np.random.uniform(0.1, 1, [13, 17]).astype("float32")
|
||||||
|
}
|
||||||
|
self.outputs = {'Out': np.multiply(self.inputs['X'], self.inputs['Y'])}
|
||||||
|
|
||||||
|
def test_check_output(self):
|
||||||
|
self.check_output()
|
||||||
|
|
||||||
|
def test_check_grad_normal(self):
|
||||||
|
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.1)
|
||||||
|
|
||||||
|
def test_check_grad_ingore_x(self):
|
||||||
|
self.check_grad(
|
||||||
|
['Y'], 'Out', max_relative_error=0.1, no_grad_set=set("X"))
|
||||||
|
|
||||||
|
def test_check_grad_ingore_y(self):
|
||||||
|
self.check_grad(
|
||||||
|
['X'], 'Out', max_relative_error=0.1, no_grad_set=set('Y'))
|
||||||
|
|
||||||
|
|
||||||
|
class TestElementwiseMulOp_Vector(OpTest):
|
||||||
|
def setUp(self):
|
||||||
|
self.op_type = "elementwise_mul"
|
||||||
|
self.inputs = {
|
||||||
|
'X': np.random.random((32, )).astype("float32"),
|
||||||
|
'Y': np.random.random((32, )).astype("float32")
|
||||||
|
}
|
||||||
|
self.outputs = {'Out': np.multiply(self.inputs['X'], self.inputs['Y'])}
|
||||||
|
|
||||||
|
def test_check_output(self):
|
||||||
|
self.check_output()
|
||||||
|
|
||||||
|
def test_check_grad_normal(self):
|
||||||
|
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.1)
|
||||||
|
|
||||||
|
def test_check_grad_ingore_x(self):
|
||||||
|
self.check_grad(
|
||||||
|
['Y'], 'Out', max_relative_error=0.1, no_grad_set=set("X"))
|
||||||
|
|
||||||
|
def test_check_grad_ingore_y(self):
|
||||||
|
self.check_grad(
|
||||||
|
['X'], 'Out', max_relative_error=0.1, no_grad_set=set('Y'))
|
||||||
|
|
||||||
|
|
||||||
|
class TestElementwiseMulOp_broadcast_0(OpTest):
|
||||||
|
def setUp(self):
|
||||||
|
self.op_type = "elementwise_mul"
|
||||||
|
self.inputs = {
|
||||||
|
'X': np.random.rand(2, 3, 4).astype(np.float32),
|
||||||
|
'Y': np.random.rand(2).astype(np.float32)
|
||||||
|
}
|
||||||
|
|
||||||
|
self.attrs = {'axis': 0}
|
||||||
|
self.outputs = {
|
||||||
|
'Out': self.inputs['X'] * self.inputs['Y'].reshape(2, 1, 1)
|
||||||
|
}
|
||||||
|
|
||||||
|
def test_check_output(self):
|
||||||
|
self.check_output()
|
||||||
|
|
||||||
|
def test_check_grad_normal(self):
|
||||||
|
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.1)
|
||||||
|
|
||||||
|
def test_check_grad_ingore_x(self):
|
||||||
|
self.check_grad(
|
||||||
|
['Y'], 'Out', max_relative_error=0.1, no_grad_set=set("X"))
|
||||||
|
|
||||||
|
def test_check_grad_ingore_y(self):
|
||||||
|
self.check_grad(
|
||||||
|
['X'], 'Out', max_relative_error=0.1, no_grad_set=set('Y'))
|
||||||
|
|
||||||
|
|
||||||
|
class TestElementwiseMulOp_broadcast_1(OpTest):
|
||||||
|
def setUp(self):
|
||||||
|
self.op_type = "elementwise_mul"
|
||||||
|
self.inputs = {
|
||||||
|
'X': np.random.rand(2, 3, 4).astype(np.float32),
|
||||||
|
'Y': np.random.rand(3).astype(np.float32)
|
||||||
|
}
|
||||||
|
|
||||||
|
self.attrs = {'axis': 1}
|
||||||
|
self.outputs = {
|
||||||
|
'Out': self.inputs['X'] * self.inputs['Y'].reshape(1, 3, 1)
|
||||||
|
}
|
||||||
|
|
||||||
|
def test_check_output(self):
|
||||||
|
self.check_output()
|
||||||
|
|
||||||
|
def test_check_grad_normal(self):
|
||||||
|
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.1)
|
||||||
|
|
||||||
|
def test_check_grad_ingore_x(self):
|
||||||
|
self.check_grad(
|
||||||
|
['Y'], 'Out', max_relative_error=0.1, no_grad_set=set("X"))
|
||||||
|
|
||||||
|
def test_check_grad_ingore_y(self):
|
||||||
|
self.check_grad(
|
||||||
|
['X'], 'Out', max_relative_error=0.1, no_grad_set=set('Y'))
|
||||||
|
|
||||||
|
|
||||||
|
class TestElementwiseMulOp_broadcast_2(OpTest):
|
||||||
|
def setUp(self):
|
||||||
|
self.op_type = "elementwise_mul"
|
||||||
|
self.inputs = {
|
||||||
|
'X': np.random.rand(2, 3, 4).astype(np.float32),
|
||||||
|
'Y': np.random.rand(4).astype(np.float32)
|
||||||
|
}
|
||||||
|
|
||||||
|
self.outputs = {
|
||||||
|
'Out': self.inputs['X'] * self.inputs['Y'].reshape(1, 1, 4)
|
||||||
|
}
|
||||||
|
|
||||||
|
def test_check_output(self):
|
||||||
|
self.check_output()
|
||||||
|
|
||||||
|
def test_check_grad_normal(self):
|
||||||
|
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.1)
|
||||||
|
|
||||||
|
def test_check_grad_ingore_x(self):
|
||||||
|
self.check_grad(
|
||||||
|
['Y'], 'Out', max_relative_error=0.1, no_grad_set=set("X"))
|
||||||
|
|
||||||
|
def test_check_grad_ingore_y(self):
|
||||||
|
self.check_grad(
|
||||||
|
['X'], 'Out', max_relative_error=0.1, no_grad_set=set('Y'))
|
||||||
|
|
||||||
|
|
||||||
|
class TestElementwiseMulOp_broadcast_3(OpTest):
|
||||||
|
def setUp(self):
|
||||||
|
self.op_type = "elementwise_mul"
|
||||||
|
self.inputs = {
|
||||||
|
'X': np.random.rand(2, 3, 4, 5).astype(np.float32),
|
||||||
|
'Y': np.random.rand(3, 4).astype(np.float32)
|
||||||
|
}
|
||||||
|
|
||||||
|
self.attrs = {'axis': 1}
|
||||||
|
self.outputs = {
|
||||||
|
'Out': self.inputs['X'] * self.inputs['Y'].reshape(1, 3, 4, 1)
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == '__main__':
|
||||||
|
unittest.main()
|
||||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue