move contrib/inference to paddle/fluid/inference/api

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/paddle_inference_api_anakin_engine.cc → paddle/fluid/inference/api/api_anakin_engine.cc

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/contrib/inference/paddle_inference_api_anakin_engine.h"
 #include <cuda.h>
+#include "paddle/contrib/inference/paddle_inference_api_anakin_engine.h"
 
 namespace paddle {
 
@@ -47,8 +47,7 @@ bool PaddleInferenceAnakinPredictor::Run(
     }
     auto d_tensor_in_p = executor_.get_in(input.name);
     float *d_data_p = d_tensor_in_p->mutable_data();
-    if (cudaMemcpy(d_data_p,
-                   static_cast<float *>(input.data.data()),
+    if (cudaMemcpy(d_data_p, static_cast<float *>(input.data.data()),
                    d_tensor_in_p->valid_size() * sizeof(float),
                    cudaMemcpyHostToDevice) != 0) {
       LOG(ERROR) << "copy data from CPU to GPU error";
@@ -70,8 +69,7 @@ bool PaddleInferenceAnakinPredictor::Run(
       output.data.Resize(tensor->valid_size() * sizeof(float));
     }
     // Copy data from GPU -> CPU
-    if (cudaMemcpy(output.data.data(),
-                   tensor->mutable_data(),
+    if (cudaMemcpy(output.data.data(), tensor->mutable_data(),
                    tensor->valid_size() * sizeof(float),
                    cudaMemcpyDeviceToHost) != 0) {
       LOG(ERROR) << "copy data from GPU to CPU error";
@@ -106,9 +104,8 @@ std::unique_ptr<PaddlePredictor> PaddleInferenceAnakinPredictor::Clone() {
 
 // A factory to help create difference predictor.
 template <>
-std::unique_ptr<PaddlePredictor>
-CreatePaddlePredictor<AnakinConfig, PaddleEngineKind::kAnakin>(
-    const AnakinConfig &config) {
+std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<
+    AnakinConfig, PaddleEngineKind::kAnakin>(const AnakinConfig &config) {
   VLOG(3) << "Anakin Predictor create.";
   std::unique_ptr<PaddlePredictor> x(
       new PaddleInferenceAnakinPredictor(config));

paddle/contrib/inference/paddle_inference_api_anakin_engine.h → paddle/fluid/inference/api/api_anakin_engine.h

@@ -48,8 +48,7 @@ class PaddleInferenceAnakinPredictor : public PaddlePredictor {
  private:
   bool Init(const AnakinConfig& config);
 
-  anakin::graph::Graph<anakin::NV,
-                       anakin::saber::AK_FLOAT,
+  anakin::graph::Graph<anakin::NV, anakin::saber::AK_FLOAT,
                        anakin::Precision::FP32>
       graph_;
   anakin::Net<anakin::NV, anakin::saber::AK_FLOAT, anakin::Precision::FP32>

paddle/contrib/inference/paddle_inference_api_impl.cc → paddle/fluid/inference/api/api_impl.cc

@@ -77,8 +77,8 @@ bool NativePaddlePredictor::Init(
   if (!config_.model_dir.empty()) {
     // Parameters are saved in separate files sited in
     // the specified `dirname`.
-    inference_program_ = paddle::inference::Load(
-        executor_.get(), scope_.get(), config_.model_dir);
+    inference_program_ = paddle::inference::Load(executor_.get(), scope_.get(),
+                                                 config_.model_dir);
   } else if (!config_.prog_file.empty() && !config_.param_file.empty()) {
     // All parameters are saved in a single file.
     // The file names should be consistent with that used
@@ -91,8 +91,8 @@ bool NativePaddlePredictor::Init(
   }
 
   ctx_ = executor_->Prepare(*inference_program_, 0);
-  executor_->CreateVariables(
-      *inference_program_, sub_scope_ ? sub_scope_ : scope_.get(), 0);
+  executor_->CreateVariables(*inference_program_,
+                             sub_scope_ ? sub_scope_ : scope_.get(), 0);
 
   // Get the feed_target_names and fetch_target_names
   feed_target_names_ = inference_program_->GetFeedTargetNames();
@@ -134,10 +134,8 @@ bool NativePaddlePredictor::Run(const std::vector<PaddleTensor> &inputs,
   // if share variables, we need not create variables
   VLOG(4) << "Run prepared context";
   executor_->RunPreparedContext(
-      ctx_.get(),
-      sub_scope_ != nullptr ? sub_scope_ : scope_.get(),
-      &feed_targets,
-      &fetch_targets,
+      ctx_.get(), sub_scope_ != nullptr ? sub_scope_ : scope_.get(),
+      &feed_targets, &fetch_targets,
       false /* don't create variable eatch time */);
   VLOG(4) << "Finish prepared context";
   if (!GetFetch(fetchs, output_data)) {
@@ -181,8 +179,7 @@ bool NativePaddlePredictor::SetFeed(const std::vector<PaddleTensor> &inputs,
     }
 
     // TODO(panyx0718): Init LoDTensor from existing memcpy to save a copy.
-    std::memcpy(static_cast<void *>(input_ptr),
-                inputs[i].data.data(),
+    std::memcpy(static_cast<void *>(input_ptr), inputs[i].data.data(),
                 inputs[i].data.length());
     feeds->push_back(input);
   }
@@ -232,8 +229,7 @@ bool NativePaddlePredictor::GetFetch(
         size_t start = lod[0][j - 1] * common_dim;
         size_t end = lod[0][j] * common_dim;
         if (end > start) {
-          std::copy(output_ptr + start,
-                    output_ptr + end,
+          std::copy(output_ptr + start, output_ptr + end,
                     data.begin() + (j - 1) * max_dim * common_dim);
         }
       }
@@ -257,15 +253,13 @@ bool NativePaddlePredictor::GetFetch(
 }
 
 template <>
-std::unique_ptr<PaddlePredictor>
-CreatePaddlePredictor<NativeConfig, PaddleEngineKind::kNative>(
-    const NativeConfig &config) {
+std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<
+    NativeConfig, PaddleEngineKind::kNative>(const NativeConfig &config) {
   VLOG(3) << "create NativePaddlePredictor";
   if (config.use_gpu) {
     // 1. GPU memeroy
     PADDLE_ENFORCE_GT(
-        config.fraction_of_gpu_memory,
-        0.f,
+        config.fraction_of_gpu_memory, 0.f,
         "fraction_of_gpu_memory in the config should be set to range (0., 1.]");
     PADDLE_ENFORCE_GE(config.device, 0, "Invalid device id %d", config.device);
     std::vector<std::string> flags;

paddle/contrib/inference/paddle_inference_api_tensorrt_subgraph_engine.cc → paddle/fluid/inference/api/api_tensorrt_subgraph_engine.cc

@@ -77,8 +77,8 @@ class TensorRTSubgraphPredictor : public NativePaddlePredictor {
     ctx_ = executor_->Prepare(*inference_program_, 0);
 
     VLOG(5) << "to create variables";
-    executor_->CreateVariables(
-        *inference_program_, sub_scope_ ? sub_scope_ : scope_.get(), 0);
+    executor_->CreateVariables(*inference_program_,
+                               sub_scope_ ? sub_scope_ : scope_.get(), 0);
 
     // Get the feed_target_names and fetch_target_names
     feed_target_names_ = inference_program_->GetFeedTargetNames();
@@ -98,8 +98,7 @@ CreatePaddlePredictor<TensorRTConfig, PaddleEngineKind::kAutoMixedTensorRT>(
   if (config.use_gpu) {
     // 1. GPU memeroy
     PADDLE_ENFORCE_GT(
-        config.fraction_of_gpu_memory,
-        0.f,
+        config.fraction_of_gpu_memory, 0.f,
         "fraction_of_gpu_memory in the config should be set to range (0., 1.]");
     PADDLE_ENFORCE_GE(config.device, 0, "Invalid device id %d", config.device);
     std::vector<std::string> flags;

paddle/contrib/inference/demo_ci/simple_on_word2vec.cc → paddle/fluid/inference/api/demo_ci/simple_on_word2vec.cc

@@ -63,8 +63,8 @@ void Main(bool use_gpu) {
     PADDLE_ENFORCE(outputs.size(), 1UL);
     // Check the output buffer size and result of each tid.
     PADDLE_ENFORCE(outputs.front().data.length(), 33168UL);
-    float result[5] = {
-        0.00129761, 0.00151112, 0.000423564, 0.00108815, 0.000932706};
+    float result[5] = {0.00129761, 0.00151112, 0.000423564, 0.00108815,
+                       0.000932706};
     const size_t num_elements = outputs.front().data.length() / sizeof(float);
     // The outputs' buffers are in CPU memory.
     for (size_t i = 0; i < std::min(5UL, num_elements); i++) {
@@ -107,8 +107,8 @@ void MainThreads(int num_threads, bool use_gpu) {
         PADDLE_ENFORCE(outputs.size(), 1UL);
         // Check the output buffer size and result of each tid.
         PADDLE_ENFORCE(outputs.front().data.length(), 33168UL);
-        float result[5] = {
-            0.00129761, 0.00151112, 0.000423564, 0.00108815, 0.000932706};
+        float result[5] = {0.00129761, 0.00151112, 0.000423564, 0.00108815,
+                           0.000932706};
         const size_t num_elements =
             outputs.front().data.length() / sizeof(float);
         // The outputs' buffers are in CPU memory.

paddle/contrib/inference/demo_ci/utils.h → paddle/fluid/inference/api/demo_ci/utils.h

@@ -21,8 +21,7 @@
 namespace paddle {
 namespace demo {
 
-static void split(const std::string& str,
-                  char sep,
+static void split(const std::string& str, char sep,
                   std::vector<std::string>* pieces) {
   pieces->clear();
   if (str.empty()) {

paddle/contrib/inference/demo_ci/vis_demo.cc → paddle/fluid/inference/api/demo_ci/vis_demo.cc

@@ -29,8 +29,7 @@ DECLARE_double(fraction_of_gpu_memory_to_use);
 DEFINE_string(modeldir, "", "Directory of the inference model.");
 DEFINE_string(refer, "", "path to reference result for comparison.");
 DEFINE_string(
-    data,
-    "",
+    data, "",
     "path of data; each line is a record, format is "
     "'<space splitted floats as data>\t<space splitted ints as shape'");
 DEFINE_bool(use_gpu, false, "Whether use gpu.");

paddle/contrib/inference/test_paddle_inference_api_impl.cc → paddle/fluid/inference/api/test_api_impl.cc

@@ -121,8 +121,8 @@ void MainImageClassification(bool use_gpu) {
   // which should be in the range [0.0, 1.0].
   feed_target_shapes[0][0] = batch_size;
   framework::DDim input_dims = framework::make_ddim(feed_target_shapes[0]);
-  SetupTensor<float>(
-      &input, input_dims, static_cast<float>(0), static_cast<float>(1));
+  SetupTensor<float>(&input, input_dims, static_cast<float>(0),
+                     static_cast<float>(1));
   std::vector<framework::LoDTensor*> cpu_feeds;
   cpu_feeds.push_back(&input);