[NPU] add argamx op bridge and unit test (#2580)

test=develop

lite/kernels/npu/bridges/CMakeLists.txt

@@ -23,6 +23,7 @@ lite_cc_library(npu_bridge_square_op SRCS square_op.cc DEPS ${npu_bridge_deps})
 lite_cc_library(npu_bridge_sqrt_op SRCS sqrt_op.cc DEPS ${npu_bridge_deps})
 lite_cc_library(npu_bridge_reduce_mean_op SRCS reduce_mean_op.cc DEPS ${npu_bridge_deps})
 lite_cc_library(npu_bridge_unsqueeze_op SRCS unsqueeze_op.cc DEPS ${npu_bridge_deps})
+lite_cc_library(npu_bridge_argmax_op SRCS argmax_op.cc DEPS ${npu_bridge_deps})
 
 set(npu_bridges
         npu_bridge_registry
@@ -47,6 +48,7 @@ set(npu_bridges
         npu_bridge_sqrt_op
         npu_bridge_reduce_mean_op
         npu_bridge_unsqueeze_op
+        npu_bridge_argmax_op
         CACHE INTERNAL "npu_bridges")
 
 set(npu_bridge_test_deps ${npu_bridges} ${npu_kernels} ${ops})
@@ -72,5 +74,6 @@ lite_cc_test(test_npu_bridge_square_op SRCS square_op_test.cc test_helper.cc DEP
 lite_cc_test(test_npu_bridge_sqrt_op SRCS sqrt_op_test.cc test_helper.cc DEPS ${npu_bridge_test_deps})
 lite_cc_test(test_npu_bridge_reduce_mean_op SRCS reduce_mean_op_test.cc test_helper.cc DEPS ${npu_bridge_test_deps})
 lite_cc_test(test_npu_bridge_unsqueeze_op SRCS unsqueeze_op_test.cc test_helper.cc DEPS ${npu_bridge_test_deps})
+lite_cc_test(test_npu_bridge_argmax_op SRCS argmax_op_test.cc test_helper.cc DEPS ${npu_bridge_test_deps})
 
 message(STATUS "+++++ npu_bridges: ${npu_bridges}")

lite/kernels/npu/bridges/argmax_op.cc

+// Copyright (c) 2019 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.
+
+#include "lite/backends/npu/builder.h"
+#include "lite/kernels/npu/bridges/registry.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace npu {
+namespace bridges {
+
+node_map_type ArgmaxConverter(const std::shared_ptr<lite::OpLite> argmax_op,
+                              const node_map_type& inputs_map) {
+  auto scope = argmax_op->scope();
+  auto op_info = argmax_op->op_info();
+  auto op_type = op_info->Type();
+  auto unique_op_type = lite::npu::UniqueName(op_type);
+  LOG(INFO) << "[NPU] Converting " + op_type + "...";
+
+  int axis = op_info->GetAttr<int64_t>("axis");
+
+  std::shared_ptr<ge::op::ArgMax> argmax_node =
+      std::make_shared<ge::op::ArgMax>(unique_op_type);
+
+  auto x_var_name = op_info->Input("X").front();
+
+  CHECK(inputs_map.count(x_var_name));
+  argmax_node->set_input_x1(*inputs_map.at(x_var_name));
+  lite::npu::OpList::Global().add(inputs_map.at(x_var_name));
+  lite::npu::OpList::Global().add(argmax_node);
+
+  Tensor x2_t;
+  x2_t.Resize(std::vector<int64_t>{1});
+  auto x2_t_data = x2_t.mutable_data<int>();
+  x2_t_data[0] = axis;
+
+  auto x2 = std::make_shared<ge::op::Const>(unique_op_type + "/axis");
+  x2->set_attr_value(lite::npu::CvtTensor(&x2_t));
+  argmax_node->set_input_x2(*x2);
+  lite::npu::OpList::Global().add(x2);
+
+  //  argmax_node->set_attr_axis(axis);
+  // argmax only support output_type==int32
+  // argmax_node->set_attr_output_type(3);
+
+  node_map_type outputs_map;
+  outputs_map[op_info->Output("Out").front()] = argmax_node;
+  return outputs_map;
+}
+
+}  // namespace bridges
+}  // namespace npu
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_NPU_BRIDGE(arg_max,
+                    paddle::lite::kernels::npu::bridges::ArgmaxConverter);

lite/kernels/npu/bridges/argmax_op_test.cc

+// Copyright (c) 2019 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.
+
+#include "lite/operators/argmax_op.h"
+#include <gtest/gtest.h>
+#include <cmath>
+#include "lite/core/op_registry.h"
+#include "lite/kernels/npu/bridges/registry.h"
+#include "lite/kernels/npu/bridges/test_helper.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace npu {
+namespace bridges {
+
+template <typename dtype>
+void argmax_ref(const std::shared_ptr<operators::ArgmaxOpLite> op) {
+  Scope* scope = op->scope();
+  const OpInfo* op_info = op->op_info();
+
+  auto x = scope->FindTensor("x");
+  auto out = scope->FindMutableTensor("out_ref");
+  int axis = op_info->GetAttr<int64_t>("axis");
+  auto x_dims = x->dims();
+  if (axis < 0) {
+    axis += x_dims.size();
+  }
+  auto y_shape = x_dims.Vectorize();
+  y_shape.erase(y_shape.begin() + axis);
+  out->Resize(y_shape);
+  auto out_dims = out->dims();
+
+  auto x_data = x->data<dtype>();
+  auto out_data = out->mutable_data<dtype>();
+
+  const int size = x_dims[axis];
+  const int in_channel = x_dims.count(axis, x_dims.size());
+  const int out_channel = out_dims.count(axis, out_dims.size());
+  const int in_stride = x_dims.count(axis + 1, x_dims.size());
+  const int out_stride = x_dims.count(0, axis);
+
+  for (int n = 0; n < out_stride; n++) {
+    for (int k = 0; k < in_stride; k++) {
+      const float* in_ptr = x_data + n * in_channel + k;
+      std::vector<std::pair<float, int>> vec;
+      vec.resize(size);
+      for (int i = 0; i < size; i++) {
+        vec[i] = std::make_pair(in_ptr[i * in_stride], i);
+      }
+      // sort
+      std::partial_sort(vec.begin(),
+                        vec.begin() + 1,
+                        vec.end(),
+                        std::greater<std::pair<float, int>>());
+
+      // out
+      dtype* out_ptr = out_data + n * out_channel + k;
+      *out_ptr = vec[0].second;
+    }
+  }
+}
+
+void test_argmax(const std::vector<int64_t>& input_shape, int axis) {
+  // prepare input&output variables
+  Scope scope;
+  std::string x_var_name = "x";
+  std::string out_var_name = "out";
+  std::string out_ref_var_name = "out_ref";
+  auto* x = scope.NewTensor(x_var_name);
+  auto* out = scope.NewTensor(out_var_name);
+  auto* out_ref = scope.NewTensor(out_ref_var_name);
+  x->Resize(input_shape);
+
+  // initialize input&output data
+  FillTensor<float>(x);
+
+  // initialize op desc
+  cpp::OpDesc opdesc;
+  opdesc.SetType("arg_max");
+  opdesc.SetInput("X", {x_var_name});
+  opdesc.SetOutput("Out", {out_var_name});
+  opdesc.SetAttr("axis", static_cast<int64_t>(axis));
+
+  // create and convert op to NPU model, then run it on NPU
+  auto op = CreateOp<operators::ArgmaxOpLite>(opdesc, &scope);
+  LauchOp(op, {x_var_name}, {out_var_name});
+
+  // execute reference implementation and save to output tensor
+  argmax_ref<float>(op);
+
+  // compare results
+  auto* out_data = out->mutable_data<int>();
+  auto* out_ref_data = out_ref->mutable_data<float>();
+  for (int i = 0; i < out->dims().production(); i++) {
+    EXPECT_NEAR(out_data[i], out_ref_data[i], 1e-2);
+  }
+}
+
+TEST(NPUBridges, argmax) {
+  test_argmax({1, 2, 3, 4}, 1);
+  test_argmax({1, 2, 3, 4}, 2);
+  test_argmax({1, 2, 3, 4}, 3);
+}
+
+}  // namespace bridges
+}  // namespace npu
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle
+
+USE_LITE_OP(arg_max);
+USE_NPU_BRIDGE(arg_max);

lite/kernels/npu/bridges/paddle_use_npu_bridges.h

@@ -25,6 +25,7 @@ USE_NPU_BRIDGE(leaky_relu);
 USE_NPU_BRIDGE(softsign);
 USE_NPU_BRIDGE(hard_sigmoid);
 
+USE_NPU_BRIDGE(arg_max);
 USE_NPU_BRIDGE(batch_norm);
 USE_NPU_BRIDGE(concat);
 USE_NPU_BRIDGE(conv2d);