Merge pull request #14805 from mozga-intel/mozga-intel/element_wise_operator_ngraph

Enable element_wise_add operator for a ngraph engine

paddle/fluid/framework/ngraph_bridge.cc

@@ -26,11 +26,14 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+namespace NG_OPS = paddle::operators::ngraphs;
 std::map<std::string,
          std::function<void(const std::shared_ptr<OperatorBase>&,
                             std::shared_ptr<std::unordered_map<
                                 std::string, std::shared_ptr<ngraph::Node>>>)>>
     NgraphBridge::NG_NODE_MAP = {
+        {"elementwise_add", NG_OPS::BuildElementwiseAddNode},
+        {"elementwise_add_grad", NG_OPS::BuildElementwiseAddGradNode},
         {"fill_constant", paddle::operators::ngraphs::BuildFillConstantNode},
         {"mean", paddle::operators::ngraphs::BuildMeanNode},
         {"mean_grad", paddle::operators::ngraphs::BuildMeanGradNode},

paddle/fluid/operators/ngraph/ngraph_ops.h

@@ -22,6 +22,7 @@ limitations under the License. */
 #pragma once
 
 #include "ops/binary_unnary_op.h"
+#include "ops/elementwise_add_op.h"
 #include "ops/fill_constant_op.h"
 #include "ops/mean_op.h"
 #include "ops/mul_op.h"

paddle/fluid/operators/ngraph/ops/elementwise_add_op.h

+/*Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+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 <string>
+#include <vector>
+
+#include "ngraph/ngraph.hpp"
+#include "paddle/fluid/operators/ngraph/ops/elementwise_node.h"
+#include "paddle/fluid/platform/ngraph_helper.h"
+
+namespace paddle {
+namespace operators {
+namespace ngraphs {
+
+void BuildElementwiseAddNode(
+    const std::shared_ptr<paddle::framework::OperatorBase>& op,
+    std::shared_ptr<
+        std::unordered_map<std::string, std::shared_ptr<ngraph::Node>>>
+        ngb_node_map) {
+  BuildElementwiseBinaryNode<ngraph::op::Add>(op, ngb_node_map);
+}
+
+void BuildElementwiseAddGradNode(
+    const std::shared_ptr<paddle::framework::OperatorBase>& op,
+    std::shared_ptr<
+        std::unordered_map<std::string, std::shared_ptr<ngraph::Node>>>
+        ngb_node_map) {
+  auto op_attrs = paddle::framework::AttrReader(op->Attrs());
+  int axis = op_attrs.Get<int>("axis");
+
+  auto dout = paddle::platform::GetInputNode(op, "Out@GRAD", ngb_node_map);
+  auto y = paddle::platform::GetInputNode(op, "Y", ngb_node_map);
+  auto dout_shape = dout->get_shape();
+  auto y_shape = y->get_shape();
+
+  if (dout_shape == y_shape) {
+    paddle::platform::SetOutputNode(op, "X@GRAD", dout, ngb_node_map);
+    paddle::platform::SetOutputNode(op, "Y@GRAD", dout, ngb_node_map);
+  } else {
+    axis = (axis == -1 ? dout_shape.size() - y_shape.size() : axis);
+    paddle::platform::TrimTrailingSingularDims(&y_shape);
+    axis = (y_shape.size() == 0 ? dout_shape.size() : axis);
+
+    int pre, n, post;
+    paddle::platform::GetMidDims(dout_shape, y_shape, axis, &pre, &n, &post);
+
+    ngraph::Shape lhs_shape{};
+    lhs_shape.push_back(pre);
+    lhs_shape.push_back(n);
+    if (post != 1) {
+      lhs_shape.push_back(post);
+    }
+
+    std::vector<size_t> lhs_order(dout_shape.size());
+    std::iota(std::begin(lhs_order), std::end(lhs_order), 0);
+    auto dout_reshape = std::make_shared<ngraph::op::Reshape>(
+        dout, ngraph::AxisVector(lhs_order), lhs_shape);
+
+    ngraph::AxisSet axis_set{0};
+    if (post != 1) {
+      axis_set.insert(2);
+    }
+
+    auto dout_sum = std::make_shared<ngraph::op::Sum>(dout_reshape, axis_set);
+    auto dy = std::make_shared<ngraph::op::Reshape>(
+        dout_sum, ngraph::AxisVector{0}, y->get_shape());
+
+    paddle::platform::SetOutputNode(op, "X@GRAD", dout, ngb_node_map);
+    paddle::platform::SetOutputNode(op, "Y@GRAD", dy, ngb_node_map);
+  }
+}
+}  // namespace ngraphs
+}  // namespace operators
+}  // namespace paddle

paddle/fluid/operators/ngraph/ops/elementwise_binary_prepare_node.h

+/*Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+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 <string>
+#include <vector>
+
+#include "ngraph/ngraph.hpp"
+#include "paddle/fluid/platform/ngraph_helper.h"
+
+namespace paddle {
+namespace operators {
+namespace ngraphs {
+
+ngraph::NodeVector ElementwiseBinaryNodePrepare(
+    const std::shared_ptr<paddle::framework::OperatorBase>& op,
+    std::shared_ptr<
+        std::unordered_map<std::string, std::shared_ptr<ngraph::Node>>>
+        ngb_node_map) {
+  auto op_attrs = paddle::framework::AttrReader(op->Attrs());
+  int axis = op_attrs.Get<int>("axis");
+  auto lhs = paddle::platform::GetInputNode(op, "X", ngb_node_map);
+  auto rhs = paddle::platform::GetInputNode(op, "Y", ngb_node_map);
+
+  auto lhs_shape = lhs->get_shape();
+  auto rhs_shape = rhs->get_shape();
+
+  PADDLE_ENFORCE_GE(lhs_shape.size(), rhs_shape.size(),
+                    "Rank of first input must >= rank of second input.");
+  if (lhs_shape == rhs_shape) {
+    return ngraph::NodeVector{lhs, rhs};
+  }
+  axis = (axis == -1 ? lhs_shape.size() - rhs_shape.size() : axis);
+  PADDLE_ENFORCE(axis >= 0 && axis < (int)(lhs_shape.size()),
+                 "Axis should be in range [0, lhs_shape)");
+  paddle::platform::TrimTrailingSingularDims(&rhs_shape);
+  axis = (rhs_shape.size() == 0) ? lhs_shape.size() : axis;
+
+  int pre, n, post;
+  paddle::platform::GetMidDims(lhs_shape, rhs_shape, axis, &pre, &n, &post);
+
+  ngraph::Shape l_shape{};
+  l_shape.push_back(pre);
+  l_shape.push_back(n);
+  l_shape.push_back(post);
+
+  std::vector<size_t> rhs_order(rhs->get_shape().size());
+  std::iota(std::begin(rhs_order), std::end(rhs_order), 0);
+  ngraph::Shape r_shape{};
+  r_shape.push_back(n);
+  auto rhs_reshape = std::make_shared<ngraph::op::Reshape>(
+      rhs, ngraph::AxisVector(rhs_order), r_shape);
+  auto rhs_bcast = std::make_shared<ngraph::op::Broadcast>(
+      rhs_reshape, l_shape, ngraph::AxisSet{0, 2});
+  std::vector<size_t> bcast_order(rhs_bcast->get_shape().size());
+  std::iota(std::begin(bcast_order), std::end(bcast_order), 0);
+  std::shared_ptr<ngraph::Node> rhs_bcast_reshape =
+      std::make_shared<ngraph::op::Reshape>(
+          rhs_bcast, ngraph::AxisVector(bcast_order), lhs_shape);
+  return ngraph::NodeVector{lhs, rhs_bcast_reshape};
+}
+}  // namespace ngraphs
+}  // namespace operators
+}  // namespace paddle

paddle/fluid/operators/ngraph/ops/elementwise_node.h

+/*Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+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 <string>
+
+#include "ngraph/ngraph.hpp"
+#include "paddle/fluid/operators/ngraph/ops/elementwise_binary_prepare_node.h"
+#include "paddle/fluid/platform/ngraph_helper.h"
+
+namespace paddle {
+namespace operators {
+namespace ngraphs {
+
+template <typename T>
+void BuildElementwiseBinaryNode(
+    const std::shared_ptr<paddle::framework::OperatorBase>& op,
+    std::shared_ptr<
+        std::unordered_map<std::string, std::shared_ptr<ngraph::Node>>>
+        ngb_node_map) {
+  auto nodes = ElementwiseBinaryNodePrepare(op, ngb_node_map);
+  std::shared_ptr<ngraph::Node>& x = nodes.at(0);
+  std::shared_ptr<ngraph::Node>& y = nodes.at(1);
+
+  if (x->get_element_type() != y->get_element_type()) {
+    y = std::make_shared<ngraph::op::Convert>(y, x->get_element_type());
+  }
+  auto out = std::make_shared<T>(x, y);
+  paddle::platform::SetOutputNode(op, "Out", out, ngb_node_map);
+}
+
+template <typename T>
+void BuildElementwiseCompareNode(
+    const std::shared_ptr<paddle::framework::OperatorBase>& op,
+    std::shared_ptr<
+        std::unordered_map<std::string, std::shared_ptr<ngraph::Node>>>
+        ngb_node_map) {
+  auto nodes = ElementwiseBinaryNodePrepare(op, ngb_node_map);
+  std::shared_ptr<ngraph::Node>& x = nodes.at(0);
+  std::shared_ptr<ngraph::Node>& y = nodes.at(1);
+
+  if (x->get_element_type() != y->get_element_type()) {
+    x = std::make_shared<ngraph::op::Convert>(x, ngraph::element::f64);
+    y = std::make_shared<ngraph::op::Convert>(y, ngraph::element::f64);
+  }
+  auto out = std::make_shared<T>(x, y);
+  paddle::platform::SetOutputNode(op, "Out", out, ngb_node_map);
+}
+}  // namespace ngraphs
+}  // namespace operators
+}  // namespace paddle

paddle/fluid/platform/ngraph_helper.h

@@ -23,7 +23,7 @@ limitations under the License. */
 namespace paddle {
 namespace platform {
 
-static ngraph::Shape FlattenTo2d(ngraph::Shape sh, int num) {
+ngraph::Shape FlattenTo2d(ngraph::Shape sh, int num) {
   auto x1 = std::accumulate(std::begin(sh), std::begin(sh) + num, 1,
                             std::multiplies<size_t>());
   auto x2 = std::accumulate(std::begin(sh) + num, std::end(sh), 1,
@@ -33,15 +33,15 @@ static ngraph::Shape FlattenTo2d(ngraph::Shape sh, int num) {
   return ngraph::Shape{x1_l, x2_l};
 }
 
-static std::shared_ptr<ngraph::Node> NgReshaper(
-    std::shared_ptr<ngraph::Node> input, ngraph::Shape shape) {
+std::shared_ptr<ngraph::Node> NgReshaper(std::shared_ptr<ngraph::Node> input,
+                                         ngraph::Shape shape) {
   std::vector<size_t> input_order(input->get_shape().size());
   std::iota(std::begin(input_order), std::end(input_order), 0);
   return std::make_shared<ngraph::op::Reshape>(
       input, ngraph::AxisVector(input_order), shape);
 }
 
-static std::shared_ptr<ngraph::Node> GetNode(
+std::shared_ptr<ngraph::Node> GetNode(
     const std::shared_ptr<paddle::framework::OperatorBase>& op,
     const std::string prm, const paddle::framework::VariableNameMap& var_map,
     std::shared_ptr<
@@ -57,7 +57,7 @@ static std::shared_ptr<ngraph::Node> GetNode(
   }
 }
 
-static std::shared_ptr<ngraph::Node> GetInputNode(
+std::shared_ptr<ngraph::Node> GetInputNode(
     const std::shared_ptr<paddle::framework::OperatorBase>& op,
     const std::string prm,
     std::shared_ptr<
@@ -66,7 +66,7 @@ static std::shared_ptr<ngraph::Node> GetInputNode(
   return GetNode(op, prm, op->Inputs(), ngb_node_map);
 }
 
-static std::shared_ptr<ngraph::Node> GetOutputNode(
+std::shared_ptr<ngraph::Node> GetOutputNode(
     const std::shared_ptr<paddle::framework::OperatorBase>& op,
     const std::string prm,
     std::shared_ptr<
@@ -75,7 +75,7 @@ static std::shared_ptr<ngraph::Node> GetOutputNode(
   return GetNode(op, prm, op->Outputs(), ngb_node_map);
 }
 
-static void SetOutputNode(
+void SetOutputNode(
     const std::shared_ptr<paddle::framework::OperatorBase>& op,
     const std::string prm, std::shared_ptr<ngraph::Node> node,
     std::shared_ptr<
@@ -91,14 +91,45 @@ static void SetOutputNode(
   }
 }
 
-static bool HasOutput(
-    const std::shared_ptr<paddle::framework::OperatorBase>& op,
-    const std::string prm) {
+bool HasOutput(const std::shared_ptr<paddle::framework::OperatorBase>& op,
+               const std::string prm) {
   auto& outputs = op->Outputs();
   if (outputs.find(prm) == outputs.end()) return false;
   return outputs.at(prm).size() > 0;
 }
 
+inline void GetMidDims(const ngraph::Shape& x_shape,
+                       const ngraph::Shape& y_shape, int axis, int* pre, int* n,
+                       int* post) {
+  *pre = 1;
+  *n = 1;
+  *post = 1;
+  for (int i = 0; i < axis; ++i) {
+    (*pre) *= x_shape[i];
+  }
+
+  for (size_t i = 0; i < y_shape.size(); ++i) {
+    PADDLE_ENFORCE_EQ(x_shape[i + axis], y_shape[i],
+                      "Broadcast dimension mismatch.");
+    (*n) *= y_shape[i];
+  }
+
+  for (size_t i = axis + y_shape.size(); i < x_shape.size(); ++i) {
+    (*post) *= x_shape[i];
+  }
+}
+
+inline void TrimTrailingSingularDims(ngraph::Shape* shape) {
+  // Remove trailing dimensions of size 1 for y
+  auto actual_shape_size = shape->size();
+  for (; actual_shape_size != 0; --actual_shape_size) {
+    if ((*shape)[actual_shape_size - 1] != 1) {
+      break;
+    } else {
+      shape->pop_back();
+    }
+  }
+}
 }  // namespace platform
 }  // namespace paddle
 

python/paddle/fluid/tests/unittests/ngraph/test_elementwise_add_ngraph_op.py

+#	Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+from __future__ import print_function
+import unittest
+from paddle.fluid.tests.unittests.test_elementwise_add_op import *
+
+
+class TestNGRAPHElementwiseAddOp(TestElementwiseAddOp):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_scalar(TestElementwiseAddOp_scalar):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_scalar, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_scalar2(TestElementwiseAddOp_scalar2):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_scalar2, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_Vector(TestElementwiseAddOp_Vector):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_Vector, self).init_input_output()
+
+
+class TesNGRAPHtElementwiseAddOp_broadcast_0(TestElementwiseAddOp_broadcast_0):
+    def init_input_output(self):
+        super(TesNGRAPHtElementwiseAddOp_broadcast_0, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_broadcast_1(TestElementwiseAddOp_broadcast_1):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_broadcast_1, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_broadcast_2(TestElementwiseAddOp_broadcast_2):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_broadcast_2, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_broadcast_3(TestElementwiseAddOp_broadcast_3):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_broadcast_3, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_broadcast_4(TestElementwiseAddOp_broadcast_4):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_broadcast_4, self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_rowwise_add_0(
+        TestElementwiseAddOp_rowwise_add_0):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_rowwise_add_0,
+              self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_rowwise_add_1(
+        TestElementwiseAddOp_rowwise_add_1):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_rowwise_add_1,
+              self).init_input_output()
+
+
+class TestNGRAPHElementwiseAddOp_channelwise_add(
+        TestElementwiseAddOp_channelwise_add):
+    def init_input_output(self):
+        super(TestNGRAPHElementwiseAddOp_channelwise_add,
+              self).init_input_output()
+
+
+if __name__ == '__main__':
+    unittest.main()