feature/inference api demo impl (#10825)

add inference api demo impl

CMakeLists.txt

@@ -58,6 +58,8 @@ option(USE_NNPACK       "Compile PaddlePaddle with NNPACK library"      OFF)
 option(WITH_DISTRIBUTE  "Compile with grpc distributed support"         OFF)
 option(USE_EIGEN_FOR_BLAS   "Use matrix multiplication in Eigen"        OFF)
 option(WITH_ARM_FP16    "Use half precision support on armv8.2-a cpu"   OFF)
+option(WITH_FAST_BUNDLE_TEST    "Bundle tests that can be run in a single process together to reduce launch overhead"   OFF)
+option(WITH_CONTRIB     "Compile the third-party contributation"        OFF)
 
 # CMAKE_BUILD_TYPE
 if(NOT CMAKE_BUILD_TYPE)
@@ -230,3 +232,7 @@ if(WITH_DOC)
     find_python_module(recommonmark REQUIRED)
     add_subdirectory(doc)
 endif()
+
+if (WITH_CONTRIB)
+    add_subdirectory(paddle/contrib)
+endif()

paddle/.gitignore

@@ -11,7 +11,6 @@ GTAGS
 *.pb.cc
 *.pb.h
 *_pb2.py
-paddle_*
 output/
 google/
 Makefile

paddle/contrib/CMakeLists.txt

+# Copyright (c) 2016 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.
+#
+
+add_subdirectory(inference)

paddle/contrib/inference/CMakeLists.txt

+# Copyright (c) 2016 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.
+#
+
+cc_library(paddle_inference_api
+    SRCS paddle_inference_api.cc
+    DEPS ${FLUID_CORE_MODULES} ${GLOB_OP_LIB})
+
+cc_test(test_paddle_inference_api
+        SRCS test_paddle_inference_api.cc
+        DEPS paddle_inference_api)

paddle/contrib/inference/paddle_inference_api.cc

+/* 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. */
+
+#include "paddle/contrib/inference/paddle_inference_api.h"

paddle/contrib/inference/paddle_inference_api.h

@@ -12,49 +12,65 @@
    See the License for the specific language governing permissions and
    limitations under the License. */
 
+/*
+ * This file contains the definition of a simple Inference API for Paddle.
+ *
+ * ATTENTION: It requires some C++ features, for lower version C++ or C, we
+ * might release another API.
+ */
+
 #pragma once
 
+#include <memory>
 #include <string>
 #include <vector>
 
 namespace paddle {
 
-class Predictor {
+struct PaddleTensor {
+  std::string name;  // variable name.
+  std::vector<int> shape;
+  std::vector<unsigned char> data;         // bytes of data.
+  size_t type{typeid(float).hash_code()};  // hash of type
+};
+
+/*
+ * A simple Inference API for Paddle. Currently this API might just be used by
+ * non-sequence scenerios.
+ * TODO(Superjomn) Prepare another API for NLP-related usages.
+ */
+class PaddlePredictor {
 public:
-  struct Attr;
-  Predictor() = default;
+  struct Config;
+  PaddlePredictor() = default;
+  PaddlePredictor(const PaddlePredictor&) = delete;
 
-  // Build the network before inference.
-  bool Init(const Attr& attr);
+  // One drived class should has such a constructor
+  // PaddlePredictor(const XConfig& config);
+  // The XConfig is a derived class of Config.
 
   // Predict an record.
-  // Arguments:
-  //   inputs: the name of the input variables.
-  //   outputs: the name of the output varaibles.
-  //   input_shapes: the shape of the input variables.
-  //   output_shapes: the shape of the output variables.
-  //   input_data: the data of the input variables.
-  //   output_data: the data of the output variables.
-  bool Run(const std::vector<std::string>& inputs,
-           const std::vector<std::string>& outputs,
-           const std::vector<std::vector<int>>& input_shapes,
-           const std::vector<std::vector<int>>& output_shapes,
-           const std::vector<std::vector<float>>& input_data,
-           std::vector<std::vector<float>>* output_data);
-
-  // Clone a predictor that share the model weights.
-  Predictor* Clone();
+  virtual bool Run(const std::vector<PaddleTensor>& inputs,
+                   std::vector<PaddleTensor>* output_data) = 0;
+
+  // Clone a predictor that share the model weights, the Cloned predictor should
+  // be thread-safe.
+  virtual std::unique_ptr<PaddlePredictor> Clone() = 0;
 
   // Destroy the Predictor.
-  ~Predictor();
+  virtual ~PaddlePredictor() {}
+
+  friend std::unique_ptr<PaddlePredictor> CreatePaddlePredictor(
+      const PaddlePredictor::Config& config);
 
-  struct Attr {
+  // The common configs for all the predictors.
+  struct Config {
     enum class EngineKind;
 
     std::string model_dir;      // path to the model directory.
     bool enable_engine{false};  // Enable to execute (part of) the model on
     // third-party engines.
-    EngineKind engine_kind{Attr::EngineKind::kNone};
+    EngineKind engine_kind{Config::EngineKind::kNone};
 
     enum class EngineKind {
       kNone = -1,          // Use the native Fluid facility.
@@ -66,4 +82,8 @@ public:
   };
 };
 
+// A factory to help create difference predictor.
+template <typename ConfigT>
+std::unique_ptr<PaddlePredictor> CreatePaddlePredictor(const ConfigT& config);
+
 }  // namespace paddle

paddle/contrib/inference/test_paddle_inference_api.cc

+/* 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. */
+
+#include "paddle/contrib/inference/paddle_inference_api.h"
+
+#include <glog/logging.h>
+#include <gtest/gtest.h>
+
+namespace paddle {
+
+/*
+ * Do not use this, just a demo indicating how to customize a config for a
+ * specific predictor.
+ */
+struct DemoConfig : public PaddlePredictor::Config {
+  float other_config;
+};
+
+/*
+ * Do not use this, just a demo indicating how to customize a Predictor.
+ */
+class DemoPredictor : public PaddlePredictor {
+public:
+  explicit DemoPredictor(const DemoConfig &config) {
+    LOG(INFO) << "I get other_config " << config.other_config;
+  }
+  bool Run(const std::vector<PaddleTensor> &inputs,
+           std::vector<PaddleTensor> *output_data) override {
+    LOG(INFO) << "Run";
+    return false;
+  }
+
+  std::unique_ptr<PaddlePredictor> Clone() override { return nullptr; }
+
+  ~DemoPredictor() override {}
+};
+
+template <>
+std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<DemoConfig>(
+    const DemoConfig &config) {
+  std::unique_ptr<PaddlePredictor> x(new DemoPredictor(config));
+  return x;
+}
+
+TEST(paddle_inference_api, demo) {
+  DemoConfig config;
+  config.other_config = 1.7;
+  auto predictor = CreatePaddlePredictor(config);
+  std::vector<PaddleTensor> outputs;
+  predictor->Run({}, &outputs);
+}
+
+}  // namespace paddle

paddle/scripts/paddle_build.sh

@@ -104,6 +104,8 @@ function cmake_gen() {
         -DCMAKE_MODULE_PATH=/opt/rocm/hip/cmake
         -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
         -DWITH_FLUID_ONLY=${WITH_FLUID_ONLY:-OFF}
+        -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
+        -DWITH_CONTRIB=ON
     ========================================
 EOF
     # Disable UNITTEST_USE_VIRTUALENV in docker because
@@ -129,7 +131,8 @@ EOF
         -DWITH_FAST_BUNDLE_TEST=ON \
         -DCMAKE_MODULE_PATH=/opt/rocm/hip/cmake \
         -DWITH_FLUID_ONLY=${WITH_FLUID_ONLY:-OFF} \
-        -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
+        -DCMAKE_EXPORT_COMPILE_COMMANDS=ON \
+        -DWITH_CONTRIB=ON
 }
 
 function abort(){