[Host] Add empty op and ceil op (#10092)

lite/kernels/host/CMakeLists.txt

@@ -124,6 +124,8 @@ add_kernel(round_compute_host Host extra SRCS round_compute.cc)
 add_kernel(temporal_shift_compute_host Host extra SRCS temporal_shift_compute.cc)
 add_kernel(pad_compute_host Host extra SRCS pad_compute.cc)
 add_kernel(bitwise_compute_host Host extra SRCS bitwise_compute.cc)
+add_kernel(empty_compute_host Host extra SRCS empty_compute.cc)
+
 
 if(LITE_BUILD_EXTRA AND LITE_WITH_x86)
   lite_cc_test(test_where_index_compute_host SRCS where_index_compute.cc)

lite/kernels/host/activation_compute.cc

@@ -173,6 +173,17 @@ void FloorCompute::Run() {
   }
 }
 
+void CeilCompute::Run() {
+  auto& param = this->Param<param_t>();
+  CHECK(param.X);
+  auto x_dims = param.X->dims();
+  auto x_data = param.X->data<float>();
+  auto output_data = param.Out->mutable_data<float>();
+  for (int i = 0; i < x_dims.production(); i++) {
+    output_data[i] = std::ceil(x_data[i]);
+  }
+}
+
 void HardSigmoidCompute::Run() {
   auto& param = this->Param<param_t>();
   CHECK(param.X);
@@ -377,6 +388,11 @@ REGISTER_LITE_KERNEL(
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kHost))})
     .Finalize();
+REGISTER_LITE_KERNEL(
+    ceil, kHost, kFloat, kNCHW, paddle::lite::kernels::host::CeilCompute, def)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kHost))})
+    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kHost))})
+    .Finalize();
 REGISTER_LITE_KERNEL(hard_sigmoid,
                      kHost,
                      kFloat,

lite/kernels/host/activation_compute.h

@@ -121,6 +121,15 @@ class FloorCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
   virtual ~FloorCompute() = default;
 };
 
+class CeilCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
+ public:
+  using param_t = operators::ActivationParam;
+
+  void Run() override;
+
+  virtual ~CeilCompute() = default;
+};
+
 class HardSigmoidCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
  public:
   using param_t = operators::ActivationParam;

lite/kernels/host/empty_compute.cc

@@ -0,0 +1,60 @@
+// Copyright (c) 2023 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/kernels/host/empty_compute.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace host {
+
+void EmptyCompute::Run() {
+  auto& param = *param_.get_mutable<param_t>();
+  auto output = param.Out;
+  auto output_dims = output->dims();
+  if (param.dtype == static_cast<int32_t>(lite::core::FluidType::BOOL)) {
+    output->set_precision(PRECISION(kBool));
+    output->template mutable_data<bool>();
+  } else if (param.dtype == static_cast<int32_t>(lite::core::FluidType::FP32)) {
+    output->set_precision(PRECISION(kFloat));
+    output->template mutable_data<float>();
+  } else if (param.dtype ==
+             static_cast<int32_t>(lite::core::FluidType::INT32)) {
+    output->set_precision(PRECISION(kInt32));
+    output->template mutable_data<int32_t>();
+  } else if (param.dtype ==
+             static_cast<int32_t>(lite::core::FluidType::INT64)) {
+    output->set_precision(PRECISION(kInt64));
+    output->template mutable_data<int64_t>();
+  } else {
+    output->set_precision(PRECISION(kInt32));
+    output->template mutable_data<int32_t>();
+  }
+
+  return;
+}
+
+}  // namespace host
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_LITE_KERNEL(
+    empty, kHost, kAny, kNCHW, paddle::lite::kernels::host::EmptyCompute, def)
+    .BindInput("ShapeTensor",
+               {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kAny))})
+    .BindInput("ShapeTensorList",
+               {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kAny))})
+    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kHost), PRECISION(kAny))})
+    .Finalize();

lite/kernels/host/empty_compute.h

@@ -0,0 +1,36 @@
+// Copyright (c) 2023 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 "lite/core/kernel.h"
+#include "lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace kernels {
+namespace host {
+
+class EmptyCompute : public KernelLite<TARGET(kHost), PRECISION(kAny)> {
+ public:
+  using param_t = operators::EmptyParam;
+
+  void Run() override;
+
+  virtual ~EmptyCompute() = default;
+};
+
+}  // namespace host
+}  // namespace kernels
+}  // namespace lite
+}  // namespace paddle

lite/operators/CMakeLists.txt

@@ -65,6 +65,8 @@ add_operator(pow_op extra SRCS pow_op.cc)
 add_operator(sign_op extra SRCS sign_op.cc)
 add_operator(rnn_op extra SRCS rnn_op.cc)
 
+add_operator(empty_op extra SRCS empty_op.cc)
+
 # 2.basic ops not used in basic models
 add_operator(negative_op extra SRCS negative_op.cc)
 add_operator(crop_op extra SRCS crop_op.cc)
@@ -212,6 +214,7 @@ add_operator(unique_with_counts_op extra SRCS unique_with_counts_op.cc)
 add_operator(unique_op extra SRCS unique_op.cc)
 add_operator(viterbi_decode extra SRCS viterbi_decode_op.cc)
 
+
 # for content-dnn specific
 add_operator(search_aligned_mat_mul_op extra SRCS search_aligned_mat_mul_op.cc)
 add_operator(search_seq_fc_op extra SRCS search_seq_fc_op.cc)

lite/operators/activation_ops.cc

@@ -121,6 +121,7 @@ REGISTER_LITE_OP(log, paddle::lite::operators::ActivationOp);
 REGISTER_LITE_OP(exp, paddle::lite::operators::ActivationOp);
 REGISTER_LITE_OP(abs, paddle::lite::operators::ActivationOp);
 REGISTER_LITE_OP(floor, paddle::lite::operators::ActivationOp);
+REGISTER_LITE_OP(ceil, paddle::lite::operators::ActivationOp);
 REGISTER_LITE_OP(hard_sigmoid, paddle::lite::operators::ActivationOp);
 REGISTER_LITE_OP(sqrt, paddle::lite::operators::ActivationOp);
 REGISTER_LITE_OP(rsqrt, paddle::lite::operators::ActivationOp);

lite/operators/empty_op.cc

@@ -0,0 +1,89 @@
+// Copyright (c) 2023 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/empty_op.h"
+#include "lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+bool EmptyOp::CheckShape() const {
+  CHECK_OR_FALSE(param_.Out);
+  return true;
+}
+
+bool EmptyOp::InferShapeImpl() const {
+  std::vector<int64_t> OutShape;
+  auto ShapeTensor = param_.ShapeTensor;
+  auto ShapeTensorList = param_.ShapeTensorList;
+  if (ShapeTensor != nullptr) {
+    auto ShapeTensorData = ShapeTensor->data<int>();
+    for (int i = 0; i < ShapeTensor->numel(); i++) {
+      OutShape.push_back(ShapeTensorData[i]);
+    }
+  } else if (!ShapeTensorList.empty()) {
+    for (size_t i = 0; i < ShapeTensorList.size(); i++) {
+      OutShape.push_back(ShapeTensorList[i]->data<int>()[0]);
+    }
+  } else if (!param_.shape.empty()) {
+    OutShape = param_.shape;
+  } else {
+    LOG(FATAL) << "no valid out_shape. Must set one of shape_tensor, or "
+                  "shape_tensor_list, or shape.";
+  }
+
+  param_.Out->Resize(OutShape);
+  return true;
+}
+
+bool EmptyOp::AttachImpl(const cpp::OpDesc& opdesc, lite::Scope* scope) {
+  if (opdesc.HasInput("ShapeTensor") && !opdesc.Input("ShapeTensor").empty()) {
+    param_.ShapeTensor =
+        scope->FindMutableTensor(opdesc.Input("ShapeTensor").front());
+  }
+  param_.ShapeTensorList.clear();
+  if (opdesc.HasInput("ShapeTensorList") &&
+      !opdesc.Input("ShapeTensorList").empty()) {
+    for (auto name : opdesc.Input("ShapeTensorList")) {
+      param_.ShapeTensorList.push_back(
+          GetMutableVar<lite::Tensor>(scope, name));
+    }
+  }
+  if (opdesc.HasAttr("shape")) {
+    auto type = opdesc.GetAttrType("shape");
+    if (type == OpAttrType::INTS) {  // paddle1.0 shape type is ints
+      auto shape = opdesc.GetAttr<std::vector<int32_t>>("shape");
+      param_.shape.resize(shape.size());
+      for (int i = 0; i < shape.size(); i++) {
+        param_.shape[i] = shape[i];
+      }
+    } else {
+      param_.shape = opdesc.GetAttr<std::vector<int64_t>>("shape");
+    }
+  }
+  param_.Out = scope->FindMutableTensor(opdesc.Output("Out").front());
+  CHECK(param_.Out) << "Output(Out) of EmptyOp should not be null.";
+  if (opdesc.HasAttr("dtype")) {
+    param_.dtype = opdesc.GetAttr<int>("dtype");
+  }
+
+  return true;
+}
+
+}  // namespace operators
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_LITE_OP(empty, paddle::lite::operators::EmptyOp);

lite/operators/empty_op.h

@@ -0,0 +1,44 @@
+// Copyright (c) 2023 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 "lite/core/op_lite.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+class EmptyOp : public OpLite {
+ public:
+  EmptyOp() {}
+  explicit EmptyOp(const std::string &op_type) : OpLite(op_type) {}
+
+  bool CheckShape() const override;
+
+  bool InferShapeImpl() const override;
+
+  bool AttachImpl(const cpp::OpDesc &opdesc, lite::Scope *scope) override;
+
+  void AttachKernel(KernelBase *kernel) override { kernel->SetParam(param_); }
+  std::string DebugString() const override { return "empty"; }
+
+ protected:
+  mutable EmptyParam param_;
+};
+
+}  // namespace operators
+}  // namespace lite
+}  // namespace paddle

lite/operators/op_params.h

@@ -2633,6 +2633,14 @@ struct TemporalShiftParam : ParamBase {
   std::string data_format{"NCHW"};
 };
 
+struct EmptyParam : ParamBase {
+  lite::Tensor* ShapeTensor{nullptr};
+  std::vector<lite::Tensor*> ShapeTensorList{};
+  std::vector<int64_t> shape{};
+  int dtype{static_cast<int>(VarDescAPI::VarDataType::FP32)};
+  lite::Tensor* Out{};
+};
+
 struct ViterbiDecodeParam : ParamBase {
   const lite::Tensor* input{};
   const lite::Tensor* length{};