change om to mindir for naml

official/recommend/naml/MINDlarge_config.yaml

@@ -43,14 +43,14 @@ category_embedding_dim: 112
 query_vector_dim: 208
 n_filters: 400
 window_size: 3
-checkpoint_path: ""
-batch_size: 64
+checkpoint_path: "naml/naml_large_new.ckpt"
+batch_size: 16
 
 # train option
 beta1: 0.9
 beta2: 0.999
 epsilon: 0.00000001 # 1e-8
-neg_sample: 4
+neg_sample: -1  #when training, neg_sample=4, when test, neg_sample=-1
 mixed: True
 sink_mode: True
 weight_decay: True
@@ -63,9 +63,14 @@ eval_neg_sample: -1
 
 # export option
 export_file_dir: "./"
-file_format: "AIR"
+file_format: "MINDIR"
 export_neg_sample: -1
 
+# infer option
+preprocess_path: "./"
+result_path: "./"
+label_path: "./"
+
 ---
 
 # Help description for each configuration

official/recommend/naml/README.md

@@ -53,11 +53,11 @@ You can download the dataset and put the directory in structure as follows:
 ├── naml
   ├── README.md                    # descriptions about NAML
   ├── model_utils
-  │   ├──__init__.py              // module init file
-  │   ├──config.py                // Parse arguments
-  │   ├──device_adapter.py        // Device adapter for ModelArts
-  │   ├──local_adapter.py         // Local adapter
-  │   ├──moxing_adapter.py        // Moxing adapter for ModelArts
+  │   ├──__init__.py               # module init file
+  │   ├──config.py                 # Parse arguments
+  │   ├──device_adapter.py         # Device adapter for ModelArts
+  │   ├──local_adapter.py          # Local adapter
+  │   ├──moxing_adapter.py         # Moxing adapter for ModelArts
   ├── scripts
   │   ├──run_distribute_train.sh   # shell script for distribute training
   │   ├──run_train.sh              # shell script for training
@@ -76,7 +76,8 @@ You can download the dataset and put the directory in structure as follows:
   ├── train.py                     # training script
   ├── eval.py                      # evaluation script
   ├── export.py                    # export mindir script
-  └── postprogress.py              # post process for 310 inference
+  ├── preprocess.py                # preprocess input data
+  └── postprocess.py               # post process for 310 inference
 ```
 
 ## [Training process](#contents)
@@ -200,6 +201,17 @@ python export.py --platform [PLATFORM] --checkpoint_path [CHECKPOINT_PATH] --fil
 
 - `EXPORT_FORMAT` should be in ["AIR", "MINDIR"]
 
+## [Infer on Ascend310](#contents)
+
+```shell
+# Ascend310 inference
+bash run_infer_310.sh [NEWS_MODEL] [USER_MODEL] [DEVICE_ID]
+```
+
+- `NEWS_MODEL` specifies path of news "MINDIR" OR "AIR" model.
+- `USER_MODEL` specifies path of user "MINDIR" OR "AIR" model.
+- `DEVICE_ID` is optional, default value is 0.
+
 # [Model Description](#contents)
 
 ## [Performance](#contents)
@@ -239,12 +251,12 @@ python export.py --platform [PLATFORM] --checkpoint_path [CHECKPOINT_PATH] --fil
 | ------------------- | --------------------------- |
 | Model Version       | NAML                        |
 | Resource            | Ascend 310                  |
-| Uploaded Date       | 03/13/2021 (month/day/year) |
-| MindSpore Version   | 1.2.0                       |  
+| Uploaded Date       | 12/10/2021 (month/day/year) |
+| MindSpore Version   | 1.6.0                       |  
 | Dataset             | MINDlarge                   |
-| batch_size          | 64                          |
+| batch_size          | 16                          |
 | outputs             | probability                 |
-| Accuracy            | AUC: 0.667                  |
+| Accuracy            | AUC: 0.6669                 |
 
 # [Description of Random Situation](#contents)
 

official/recommend/naml/ascend310_infer/CMakeLists.txt

-# Copyright (c) Huawei Technologies Co., Ltd. 2019. All rights reserved.
-
-# CMake lowest version requirement
-cmake_minimum_required(VERSION 3.5.1)
-
-# project information
-project(ACL_RESNET50)
-
-find_package(gflags REQUIRED)
-include_directories(${gflags_INCLUDE_DIR})
-
-# Compile options
-add_compile_options(-std=c++11 -g -O0)
-
-set(CMAKE_RUNTIME_OUTPUT_DIRECTORY  "../out")
-set(CMAKE_CXX_FLAGS_DEBUG "-fPIC -O0 -g -Wall")
-set(CMAKE_CXX_FLAGS_RELEASE "-fPIC -O0 -g -Wall")
-
-set(INC_PATH $ENV{DDK_PATH})
-
-if(NOT DEFINED ENV{DDK_PATH})
-    set(INC_PATH "/usr/local/Ascend")
-    message(STATUS "set default INC_PATH: ${INC_PATH}")
-else()
-    message(STATUS "env INC_PATH: ${INC_PATH}")
-endif()
-
-set(LIB_PATH $ENV{NPU_HOST_LIB})
-
-if(NOT DEFINED ENV{NPU_HOST_LIB})
-    set(LIB_PATH "/usr/local/Ascend/acllib/lib64/stub/")
-    message(STATUS "set default LIB_PATH: ${LIB_PATH}")
-else()
-    message(STATUS "env LIB_PATH: ${LIB_PATH}")
-endif()
-
-# Header path
-include_directories(
-    ${INC_PATH}/acllib/include/
-    ../include/
-)
-
-# add host lib path
-link_directories(
-    ${LIB_PATH}
-)
-
-# Set output directory
+cmake_minimum_required(VERSION 3.14.1)
+project(Ascend310Infer)
+add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O0 -g -std=c++17 -Werror -Wall -fPIE -Wl,--allow-shlib-undefined")
 set(PROJECT_SRC_ROOT ${CMAKE_CURRENT_LIST_DIR}/)
-
-# Set include directory and library directory
-set(ACL_LIB_DIR $ENV{ASCEND_HOME}/fwkacllib)
-set(ATLAS_ACL_LIB_DIR $ENV{ASCEND_HOME}/ascend-toolkit/latest/fwkacllib)
-# Header path
-include_directories(${ACL_LIB_DIR}/include/)
-include_directories(${ATLAS_ACL_LIB_DIR}/include/)
-include_directories(${PROJECT_SRC_ROOT}/../inc)
-
-# add host lib path
-link_directories(${ACL_LIB_DIR})
-find_library(acl libascendcl.so ${ACL_LIB_DIR}/lib64 ${ATLAS_ACL_LIB_DIR}/lib64)
-find_library(stdc libstdc++.so.6 /usr/)
-
-add_executable(main
-        ./src/utils.cpp
-        ./src/model_process.cpp
-        ./src/sample_process.cpp
-        ./src/main.cpp)
-
-target_link_libraries(main ${acl} ${stdc} gflags)
-
-install(TARGETS main DESTINATION ${CMAKE_RUNTIME_OUTPUT_DIRECTORY})
+option(MINDSPORE_PATH "mindspore install path" "")
+include_directories(${MINDSPORE_PATH})
+include_directories(${MINDSPORE_PATH}/include)
+include_directories(${PROJECT_SRC_ROOT})
+find_library(MS_LIB libmindspore.so ${MINDSPORE_PATH}/lib)
+file(GLOB_RECURSE MD_LIB ${MINDSPORE_PATH}/_c_dataengine*)
+
+add_executable(main src/main.cc src/utils.cc)
+target_link_libraries(main ${MS_LIB} ${MD_LIB} gflags)

official/recommend/naml/ascend310_infer/build.sh

@@ -13,11 +13,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ============================================================================
-
-if [ ! -d out ]; then
-  mkdir out
+if [ -d out ]; then
+    rm -rf out
 fi
+
+mkdir out
 cd out || exit
-export CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0
-cmake .. -DCMAKE_CXX_COMPILER=g++ -DCMAKE_SKIP_RPATH=TRUE
+
+if [ -f "Makefile" ]; then
+  make clean
+fi
+
+cmake .. \
+    -DMINDSPORE_PATH="`pip3.7 show mindspore-ascend | grep Location | awk '{print $2"/mindspore"}' | xargs realpath`"
 make

official/recommend/naml/ascend310_infer/inc/model_process.h

-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * 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.
- */
-
-#ifndef MINDSPORE_MODEL_ZOO_NAML_MODEL_PROCESS_H_
-#define MINDSPORE_MODEL_ZOO_NAML_MODEL_PROCESS_H_
-#pragma once
-#include <iostream>
-#include <map>
-#include <vector>
-#include <string>
-#include "./utils.h"
-#include "acl/acl.h"
-
-class ModelProcess {
- public:
-    ModelProcess() = default;
-    ModelProcess(const std::string &inputDataPath, const std::string &idFilePath, uint32_t batchSize);
-
-    ~ModelProcess();
-
-    Result LoadModelFromFileWithMem(const char *modelPath);
-
-    void Unload();
-
-    Result CreateDesc();
-
-    void DestroyDesc();
-
-    void DestroyInput();
-
-    Result CreateOutput();
-
-    void DestroyOutput();
-
-    Result Execute(uint32_t index);
-
-    void OutputModelResult();
-    Result CreateInput();
-    Result CpyFileToDevice(std::string fileName, uint32_t inputNum);
-    void CpyOutputFromDeviceToHost(uint32_t index);
-    std::map<int, void *> GetResult();
-    std::vector<uint32_t> GetOutputSize();
-    std::vector<uint32_t> GetInputSize();
-    Result ExecuteWithFile(uint32_t fileNum);
-    Result CpyDataToDevice(void *data, uint32_t len, uint32_t inputNum);
-    std::string GetInputDataPath();
-    std::string GetCostTimeInfo();
-    void DestroyResource();
-    std::vector<std::vector<void *>> ReadInputFiles(std::vector<std::vector<std::string>> inputFiles,
-                                                    size_t inputSize, std::vector<std::vector<uint32_t>> *fileSize);
-    Result ReadIdFiles();
-    Result InitResource();
-    uint32_t ReadFiles();
-
- private:
-    uint32_t modelId_;
-    std::map<double, double> costTime_map_;
-    size_t modelMemSize_;
-    size_t modelWeightSize_;
-    void *modelMemPtr_;
-    void *modelWeightPtr_;
-    uint32_t batchSize_;
-    bool loadFlag_;  // model load flag
-    aclmdlDesc *modelDesc_;
-    aclmdlDataset *input_;
-    uint32_t inputNum_;
-    std::vector<uint32_t> inputBuffSize_;
-    aclmdlDataset *output_;
-    uint32_t outputNum_;
-    std::vector<uint32_t> outputBuffSize_;
-
-    std::map<int, void *> result_;
-    std::string inputDataPath_;
-    std::string idFilePath_;
-    std::vector<std::vector<void *>> fileBuff_;
-    std::vector<std::vector<uint32_t>> fileSize_;
-    std::vector<std::vector<int>> ids_;
-};
-
-#endif

official/recommend/naml/ascend310_infer/inc/sample_process.h

-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * 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.
- */
-
-#ifndef MINDSPORE_MODEL_ZOO_NAML_SAMPLE_PROCESS_H_
-#define MINDSPORE_MODEL_ZOO_NAML_SAMPLE_PROCESS_H_
-
-#pragma once
-#include <map>
-#include <memory>
-#include <mutex>
-#include <string>
-#include <vector>
-
-#include "acl/acl.h"
-#include "./utils.h"
-#include "./model_process.h"
-
-class SampleProcess {
- public:
-    SampleProcess();
-    SampleProcess(uint32_t deviceId, uint32_t threadNum);
-
-    ~SampleProcess();
-
-    Result InitResource();
-
-    Result Process(const std::vector<std::string> &omPaths,
-                   const std::vector<std::string> &inputDataPaths,
-                   const std::vector<std::string> &inputIdPaths,
-                   const std::string &browsedNewsPath,
-                   uint32_t batchSize);
-    Result CreateModelProcessInstance(const std::vector<std::string> &omPaths,
-                                      const std::vector<std::string> &inputDataPaths,
-                                      const std::vector<std::string> &inputIdPaths, uint32_t batchSize,
-                                      std::map<int, std::shared_ptr<ModelProcess>> *modelProcessContainer);
-    Result GetPred(std::map<int, std::shared_ptr<ModelProcess>> *modelProcessContainer, uint32_t fileNum);
-    int WriteResult(const std::string& imageFile, std::vector<float> result, uint32_t size);
-    std::vector<std::string> GetModelExecCostTimeInfo();
-    std::vector<std::vector<std::vector<int>>> ReadHistory(std::vector<std::string> historyFile, uint32_t batchSize);
-    Result ReadBrowsedFile(const std::string &browsedNewsPath, std::vector<std::string> userIdFiles,
-                           std::vector<std::vector<int>> *usersIds, std::vector<std::vector<int>> *candidateNewsIds);
-    uint32_t ReadBrowsedData(const std::string &browsedNewsPath);
-    void GetResult(uint32_t startPos, uint32_t endPos,
-                                  std::map<int, void *> newsEncodeResult,
-                                  std::map<int, void *> userEncodeResult);
-
- private:
-    void DestroyResource();
-
-    int32_t deviceId_;
-    aclrtContext context_;
-    aclrtStream stream_;
-    std::map<double, double> secondModelCostTime_map_;
-    std::map<double, double> thirdModelCostTime_map_;
-    std::map<double, double> totalCostTime_map_;
-    std::vector<std::vector<int>> usersIds_;
-    std::vector<std::vector<int>> candidateNewsIds_;
-    std::vector<std::string> userIdFiles_;
-    std::vector<std::string> time_cost_;
-    uint32_t threadNum_;
-    std::mutex mtx_;
-};
-
-#endif

official/recommend/naml/ascend310_infer/inc/utils.h

@@ -14,43 +14,19 @@
  * limitations under the License.
  */
 
-#ifndef MINDSPORE_MODEL_ZOO_NAML_UTILS_H_
-#define MINDSPORE_MODEL_ZOO_NAML_UTILS_H_
+#ifndef MINDSPORE_INFERENCE_UTILS_H_
+#define MINDSPORE_INFERENCE_UTILS_H_
 
-#pragma once
+#include <sys/stat.h>
 #include <dirent.h>
-#include <iostream>
 #include <vector>
 #include <string>
-
-#define INFO_LOG(fmt, args...) fprintf(stdout, "[INFO]  " fmt "\n", ##args)
-#define WARN_LOG(fmt, args...) fprintf(stdout, "[WARN]  " fmt "\n", ##args)
-#define ERROR_LOG(fmt, args...) fprintf(stdout, "[ERROR] " fmt "\n", ##args)
-
-typedef enum Result {
-    SUCCESS = 0,
-    FAILED = 1
-} Result;
-
-class Utils {
- public:
-    static void *GetDeviceBufferOfFile(std::string fileName, uint32_t *fileSize);
-
-    static void *ReadBinFile(std::string fileName, uint32_t *fileSize);
-
-    static std::vector <std::vector<std::string>> GetAllInputData(std::string dir_name);
-
-    static DIR *OpenDir(std::string dir_name);
-
-    static std::string RealPath(std::string path);
-
-    static std::vector <std::string> GetAllBins(std::string dir_name);
-
-    static Result ReadFileToVector(std::string newsIdFileName, uint32_t batchSize, std::vector<int> *newsId);
-    static Result ReadFileToVector(std::string newsIdFileName, std::vector<int> *newsId);
-    static Result ReadFileToVector(std::string newsIdFileName, uint32_t batchSize, uint32_t count,
-                                   std::vector<std::vector<int>> *newsId);
-};
-#pragma once
-
+#include <memory>
+#include "include/api/types.h"
+
+std::vector<std::string> GetAllFiles(std::string_view dirName);
+DIR *OpenDir(std::string_view dirName);
+std::string RealPath(std::string_view path);
+mindspore::MSTensor ReadFileToTensor(const std::string &file);
+int WriteResult(const std::string& imageFile, const std::vector<mindspore::MSTensor> &outputs);
 #endif

official/recommend/naml/ascend310_infer/src/main.cc

+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * 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 <sys/time.h>
+#include <gflags/gflags.h>
+#include <dirent.h>
+#include <string.h>
+#include <iostream>
+#include <algorithm>
+
+#include <string>
+#include <iosfwd>
+#include <vector>
+#include <fstream>
+#include <sstream>
+
+#include "include/api/model.h"
+#include "include/api/context.h"
+#include "include/api/types.h"
+#include "include/api/serialization.h"
+#include "include/dataset/execute.h"
+#include "include/dataset/vision.h"
+#include "inc/utils.h"
+
+using mindspore::Context;
+using mindspore::Serialization;
+using mindspore::Model;
+using mindspore::Status;
+using mindspore::MSTensor;
+using mindspore::dataset::Execute;
+using mindspore::ModelType;
+using mindspore::GraphCell;
+using mindspore::kSuccess;
+
+//*****     bs=16     *****//
+DEFINE_string(news_mindir, "./", "om model path.");
+DEFINE_string(user_mindir, "./", "om model path.");
+DEFINE_int32(batch_size, 16, "batch size");
+DEFINE_int32(device_id, 0, "device id");
+DEFINE_string(base_path, "./", "dataset base path.");
+
+std::vector<std::string> news_input0_0_files;
+auto context = std::make_shared<Context>();
+
+struct timeval total_start;
+struct timeval total_end;
+std::map<double, double> model0_cost_time;
+std::map<double, double> model1_cost_time;
+
+int InitModel(const std::string &model_path, Model *model) {
+  mindspore::Graph graph;
+  Serialization::Load(model_path, ModelType::kMindIR, &graph);
+  Status ret = model->Build(GraphCell(graph), context);
+  if (ret != kSuccess) {
+    std::cout << "ERROR: Build model failed." << std::endl;
+    return 1;
+  }
+  return 0;
+}
+
+void GetAveTime(const std::map<double, double> &costTime_map, int *inferCount, double *average) {
+  for (auto iter = costTime_map.begin(); iter != costTime_map.end(); iter++) {
+    double diff = 0.0;
+    diff = iter->second - iter->first;
+    (*average) += diff;
+    (*inferCount)++;
+  }
+  (*average) /= (*inferCount);
+  return;
+}
+
+void InitInputs(size_t input_index, const std::vector<MSTensor> &model0_inputs, std::vector<MSTensor> *inputs) {
+  std::string news_dataset_path = FLAGS_base_path + "/news_test_data";
+  auto file0_name = news_dataset_path + "/00_category_data/naml_news_" + std::to_string(input_index) + ".bin";
+  auto file1_name = news_dataset_path + "/01_subcategory_data/naml_news_" + std::to_string(input_index) + ".bin";
+  auto file2_name = news_dataset_path + "/02_title_data/naml_news_" + std::to_string(input_index) + ".bin";
+  auto file3_name = news_dataset_path + "/03_abstract_data/naml_news_" + std::to_string(input_index) + ".bin";
+
+  auto news_input0 = ReadFileToTensor(file0_name);
+  auto news_input1 = ReadFileToTensor(file1_name);
+  auto news_input2 = ReadFileToTensor(file2_name);
+  auto news_input3 = ReadFileToTensor(file3_name);
+
+  inputs->emplace_back(model0_inputs[0].Name(), model0_inputs[0].DataType(), model0_inputs[0].Shape(),
+                      news_input0.Data().get(), news_input0.DataSize());
+  inputs->emplace_back(model0_inputs[1].Name(), model0_inputs[1].DataType(), model0_inputs[1].Shape(),
+                      news_input1.Data().get(), news_input1.DataSize());
+  inputs->emplace_back(model0_inputs[2].Name(), model0_inputs[2].DataType(), model0_inputs[2].Shape(),
+                      news_input2.Data().get(), news_input2.DataSize());
+  inputs->emplace_back(model0_inputs[3].Name(), model0_inputs[3].DataType(), model0_inputs[3].Shape(),
+                      news_input3.Data().get(), news_input3.DataSize());
+  return;
+}
+
+void InitNewsDict(const std::vector<std::string> &news_input0_0_files, const std::vector<MSTensor> &model0_inputs,
+                  std::map<uint32_t, MSTensor> *news_dict, Model *model0) {
+  int count = 0;
+  for (size_t i = 0; i < news_input0_0_files.size() - 1; ++i) {
+    struct timeval start;
+    struct timeval end;
+    double startTime_ms;
+    double endTime_ms;
+    std::vector <MSTensor> inputs0;
+    std::vector <MSTensor> outputs0;
+
+    // init inputs by model0 input for each iter
+    InitInputs(i, model0_inputs, &inputs0);
+
+    // get model0 outputs
+    gettimeofday(&total_start, nullptr);
+    gettimeofday(&start, nullptr);
+    Status ret0 = model0->Predict(inputs0, &outputs0);
+    gettimeofday(&end, nullptr);
+    if (ret0 != kSuccess) {
+      std::cout << "ERROR: Predict model0 failed." << std::endl;
+      return;
+    }
+    // init news_dict
+    auto file_name = FLAGS_base_path + "/news_id_data/naml_news_" + std::to_string(count++) + ".bin";
+    auto nid = ReadFileToTensor(file_name);
+    auto nid_addr = reinterpret_cast<uint32_t *>(nid.MutableData());
+    // output0 size is 1, shape: batch_size * 400
+    auto outputs0_addr = reinterpret_cast<float *>(outputs0[0].MutableData());
+
+    for (int k = 0; k < FLAGS_batch_size; ++k) {
+      MSTensor ret("", mindspore::DataType::kNumberTypeFloat32, {400}, nullptr, sizeof(float)* 400);
+      uint32_t *addr = reinterpret_cast<uint32_t *>(ret.MutableData());
+      memcpy(addr, outputs0_addr, 400 * sizeof(float));
+      (*news_dict)[nid_addr[k]] = ret;
+      outputs0_addr += 400;
+    }
+    startTime_ms = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTime_ms = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    model0_cost_time.insert(std::pair<double, double>(startTime_ms, endTime_ms));
+  }
+  std::cout << "=========InitNewsDict end" << std::endl;
+  return;
+}
+
+void InitUserDict(const std::vector<std::string> &history_input_files, const std::map<uint32_t, MSTensor> &news_dict,
+                  std::map<uint32_t, MSTensor> *user_dict, Model *model1) {
+  int count = 0;
+  int uid_count = 0;
+  for (size_t i = 0; i < history_input_files.size() - 1; ++i) {
+    // model1, get model1 outputs
+    struct timeval start;
+    struct timeval end;
+    double startTime_ms;
+    double endTime_ms;
+    std::vector<MSTensor> outputs1;
+    int64_t size = FLAGS_batch_size * 50 * 400 * sizeof(float);
+    MSTensor buffer("", mindspore::DataType::kNumberTypeFloat32, {FLAGS_batch_size, 50, 400}, nullptr, size);
+    uint8_t *addr = reinterpret_cast<uint8_t *>(buffer.MutableData());
+    auto file_name = FLAGS_base_path + "/users_test_data/01_history_data/naml_users_" +
+                     std::to_string(count++) + ".bin";
+    auto nid = ReadFileToTensor(file_name);
+    auto nid_addr = reinterpret_cast<uint32_t *>(nid.MutableData());
+
+    for (int j = 0; j < FLAGS_batch_size; ++j) {
+      for (int k = 0; k < 50; ++k) {
+        if (news_dict.find(nid_addr[k]) == news_dict.end()) {
+          addr += 400 * sizeof(float);
+          continue;
+        }
+        auto ms_tensor = news_dict.at(nid_addr[k]);
+        uint32_t *new_dict_data = reinterpret_cast<uint32_t *>(ms_tensor.MutableData());
+        if (addr == nullptr || new_dict_data == nullptr) {
+          std::cout << "addr is nullptr or new_dict_data is nullptr"
+                    << "src is: " << &new_dict_data << " dst is: " << &addr << std::endl;
+          return;
+        }
+        memcpy(addr, new_dict_data, 400 * sizeof(float));
+        addr += 400 * sizeof(float);
+      }
+      nid_addr += 50;
+    }
+    gettimeofday(&start, nullptr);
+    Status ret1 = model1->Predict({buffer}, &outputs1);
+    gettimeofday(&end, nullptr);
+    if (ret1 != kSuccess) {
+      std::cout << "ERROR: Predict model1 failed." << std::endl;
+      return;
+    }
+    // init user_dict
+    auto file_name1 = FLAGS_base_path + "/users_test_data/00_user_id_data/naml_users_" +
+                      std::to_string(uid_count++) + ".bin";
+    auto nid1 = ReadFileToTensor(file_name1);
+    auto nid_addr1 = reinterpret_cast<uint32_t *>(nid1.MutableData());
+    auto outputs1_addr = reinterpret_cast<float *>(outputs1[0].MutableData());
+    for (int k = 0; k < FLAGS_batch_size; ++k) {
+      MSTensor ret("", mindspore::DataType::kNumberTypeFloat32, {400}, nullptr, sizeof(float) * 400);
+      uint32_t *addr1 = reinterpret_cast<uint32_t *>(ret.MutableData());
+      memcpy(addr1, outputs1_addr, 400 * sizeof(float));
+      addr1 += 400;
+      (*user_dict)[nid_addr1[k]] = ret;
+      outputs1_addr += 400;
+    }
+    startTime_ms = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
+    endTime_ms = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
+    model1_cost_time.insert(std::pair<double, double>(startTime_ms, endTime_ms));
+  }
+  std::cout << "=========InitUserDict end" << std::endl;
+  return;
+}
+
+void InitPred(const std::map<uint32_t, MSTensor> &news_dict, const std::map<uint32_t, MSTensor> &user_dict) {
+  std::string browsed_news_path = FLAGS_base_path + "/browsed_news_test_data";
+  std::string file1_path = browsed_news_path + "/01_candidate_nid_data";
+  auto eval_candidate_news = GetAllFiles(file1_path);
+  for (size_t i = 0; i < eval_candidate_news.size() - 1; ++i) {
+    std::vector<MSTensor> pred;
+    std::string file2_path = browsed_news_path + "/00_user_id_data";
+    auto uid_file = file2_path + "/naml_browsed_news_" + std::to_string(i) + ".bin";
+    auto uid_nid = ReadFileToTensor(uid_file);
+    auto uid_nid_addr = reinterpret_cast<uint32_t *>(uid_nid.MutableData());
+    if (user_dict.find(uid_nid_addr[0]) == user_dict.end()) {
+      int rst = WriteResult(uid_file, pred);
+      if (rst != 0) {
+        std::cout << "write result failed." << std::endl;
+        return;
+      }
+      continue;
+    }
+
+    MSTensor dot2 = user_dict.at(uid_nid_addr[0]);
+    auto candidate_nid_file = file1_path + "/naml_browsed_news_" + std::to_string(i) + ".bin";
+    auto candidate_nid = ReadFileToTensor(candidate_nid_file);
+    size_t bin_size = candidate_nid.DataSize();
+    size_t browsed_news_count = bin_size / sizeof(float);
+
+    auto candidate_nid_addr = reinterpret_cast<uint32_t *>(candidate_nid.MutableData());
+    MSTensor ret("", mindspore::DataType::kNumberTypeFloat32, {static_cast<int>(browsed_news_count)},
+                 nullptr, bin_size);
+    uint8_t *addr = reinterpret_cast<uint8_t *>(ret.MutableData());
+    for (size_t j = 0; j < browsed_news_count; ++j) {
+      float sum = 0;
+      if (news_dict.find(candidate_nid_addr[j]) == news_dict.end()) {
+        addr += sizeof(float);
+        continue;
+      }
+
+      MSTensor dot1 = news_dict.at(candidate_nid_addr[j]);
+      auto dot1_addr = reinterpret_cast<float *>(dot1.MutableData());
+      auto dot2_addr = reinterpret_cast<float *>(dot2.MutableData());
+      for (int k = 0; k < 400; ++k) {
+        sum = sum + dot1_addr[k] * dot2_addr[k];
+      }
+      memcpy(addr, &sum, sizeof(float));
+      addr += sizeof(float);
+    }
+    pred.emplace_back(ret);
+    int rst = WriteResult(uid_file, pred);
+    if (rst != 0) {
+      std::cout << "write result failed." << std::endl;
+      return;
+    }
+  }
+  std::cout << "=========InitPred end" << std::endl;
+  return;
+}
+
+int main(int argc, char **argv) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (RealPath(FLAGS_news_mindir).empty()) {
+    std::cout << "Invalid mindir" << std::endl;
+    return 1;
+  }
+  auto ascend310 = std::make_shared<mindspore::Ascend310DeviceInfo>();
+  ascend310->SetDeviceID(FLAGS_device_id);
+  context->MutableDeviceInfo().push_back(ascend310);
+
+  double startTimeMs;
+  double endTimeMs;
+  std::map<double, double> total_cost_time;
+  // om -> model
+  Model model0;
+  Model model1;
+  if (InitModel(FLAGS_news_mindir, &model0) != 0 || InitModel(FLAGS_user_mindir, &model1) != 0) {
+    std::cout << "ERROR: Init model failed." << std::endl;
+    return 1;
+  }
+
+  // get model inputs
+  std::vector<MSTensor> model0_inputs = model0.GetInputs();
+  if (model0_inputs.empty()) {
+    std::cout << "Invalid model, inputs is empty." << std::endl;
+    return 1;
+  }
+
+  // get input files by bin files
+  std::string news_dataset_path = FLAGS_base_path + "/news_test_data";
+  std::string history_data_path = FLAGS_base_path + "/users_test_data/01_history_data";
+  news_input0_0_files = GetAllFiles(news_dataset_path + "/00_category_data");
+  auto history_input_files = GetAllFiles(history_data_path);
+  if (news_input0_0_files.empty() || history_input_files.empty()) {
+    std::cout << "ERROR: input data empty." << std::endl;
+    return 1;
+  }
+
+  std::map<uint32_t, MSTensor> news_dict;
+  std::map<uint32_t, MSTensor> user_dict;
+
+  InitNewsDict(news_input0_0_files, model0_inputs, &news_dict, &model0);
+  InitUserDict(history_input_files, news_dict, &user_dict, &model1);
+  InitPred(news_dict, user_dict);
+  gettimeofday(&total_end, nullptr);
+
+  startTimeMs = (1.0 * total_start.tv_sec * 1000000 + total_start.tv_usec) / 1000;
+  endTimeMs = (1.0 * total_end.tv_sec * 1000000 + total_end.tv_usec) / 1000;
+  total_cost_time.insert(std::pair<double, double>(startTimeMs, endTimeMs));
+
+  double average0 = 0.0;
+  int inferCount0 = 0;
+  double average1 = 0.0;
+  int infer_cnt = 0;
+  GetAveTime(model0_cost_time, &inferCount0, &average0);
+  GetAveTime(model1_cost_time, &infer_cnt, &average1);
+  double total_time = total_cost_time.begin()->second - total_cost_time.begin()->first;
+
+  std::stringstream timeCost0, timeCost1, totalTimeCost;
+  timeCost0 << "first model cost average time: "<< average0 << " ms of infer_count " << inferCount0 << std::endl;
+  timeCost1 << "second model cost average time: "<< average1 << " ms of infer_count " << infer_cnt << std::endl;
+  totalTimeCost << "total cost time: "<< total_time << " ms of infer_count " << infer_cnt << std::endl;
+
+  std::string fileName = "./time_Result" + std::string("/test_perform_static.txt");
+  std::ofstream fileStream(fileName.c_str(), std::ios::trunc);
+  fileStream << timeCost0.str() << std::endl;
+  fileStream << timeCost1.str() << std::endl;
+  fileStream << totalTimeCost.str() << std::endl;
+  std::cout << timeCost0.str() << std::endl;
+  std::cout << timeCost1.str() << std::endl;
+  std::cout << totalTimeCost.str() << std::endl;
+  fileStream.close();
+  model0_cost_time.clear();
+  model1_cost_time.clear();
+  total_cost_time.clear();
+  std::cout << "Execute success." << std::endl;
+  return 0;
+}

official/recommend/naml/ascend310_infer/src/main.cpp

-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * 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 <gflags/gflags.h>
-#include <string>
-#include <iostream>
-#include <memory>
-#include <fstream>
-#include "../inc/utils.h"
-#include "../inc/sample_process.h"
-
-bool g_isDevice = false;
-
-DEFINE_string(news_om_path, "../model/relu_double_geir.om", "om model path.");
-DEFINE_string(user_om_path, "../model/relu_double_geir.om", "om model path.");
-DEFINE_string(news_dataset_path, "../data", "input data dir");
-DEFINE_string(user_dataset_path, "../data", "input data dir");
-DEFINE_string(newsid_data_path, "../data", "input data dir");
-DEFINE_string(userid_data_path, "../data", "input data dir");
-DEFINE_string(browsed_news_path, "../data", "input data dir");
-DEFINE_int32(batch_size, 16, "batch size");
-DEFINE_int32(device_id, 0, "device id");
-DEFINE_int32(thread_num, 8, "thread num");
-
-int main(int argc, char** argv) {
-    gflags::ParseCommandLineFlags(&argc, &argv, true);
-    std::cout << "news OM File Path :" << FLAGS_news_om_path << std::endl;
-    std::cout << "user OM File Path :" << FLAGS_user_om_path << std::endl;
-    std::cout << "news Dataset Path :" << FLAGS_news_dataset_path << std::endl;
-    std::cout << "user Dataset Path :" << FLAGS_user_dataset_path << std::endl;
-    std::cout << "browsed_news_path Path :" << FLAGS_browsed_news_path << std::endl;
-    std::cout << "batch size :" << FLAGS_batch_size << std::endl;
-    std::cout << "device id :" << FLAGS_device_id << std::endl;
-    std::cout << "thread num :" << FLAGS_thread_num << std::endl;
-
-    std::vector<std::string> omPaths;
-    std::vector<std::string> datasetPaths;
-    std::vector<std::string> idsPaths;
-    omPaths.emplace_back(FLAGS_news_om_path);
-    omPaths.emplace_back(FLAGS_user_om_path);
-    datasetPaths.emplace_back(FLAGS_news_dataset_path);
-    datasetPaths.emplace_back(FLAGS_user_dataset_path);
-    idsPaths.emplace_back(FLAGS_newsid_data_path);
-    idsPaths.emplace_back(FLAGS_userid_data_path);
-
-    SampleProcess processSample(FLAGS_device_id, FLAGS_thread_num);
-    Result ret = processSample.InitResource();
-    if (ret != SUCCESS) {
-        ERROR_LOG("sample init resource failed");
-        return FAILED;
-    }
-
-    ret = processSample.Process(omPaths, datasetPaths, idsPaths, FLAGS_browsed_news_path, FLAGS_batch_size);
-    if (ret != SUCCESS) {
-        ERROR_LOG("sample process failed");
-        return FAILED;
-    }
-
-    std::vector<std::string> costTime = processSample.GetModelExecCostTimeInfo();
-    std::string file_name = "./time_Result" + std::string("/test_perform_static.txt");
-    std::ofstream file_stream(file_name.c_str(), std::ios::trunc);
-    for (auto cost : costTime) {
-        std::cout << cost << std::endl;
-        file_stream << cost << std::endl;
-    }
-
-    file_stream.close();
-
-    INFO_LOG("execute sample success");
-    return SUCCESS;
-}

official/recommend/naml/ascend310_infer/src/model_process.cpp

-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * 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 <sys/time.h>
-#include <iostream>
-#include <fstream>
-#include <map>
-#include <sstream>
-#include <algorithm>
-#include "../inc/utils.h"
-#include "../inc/model_process.h"
-
-extern bool g_isDevice;
-
-ModelProcess::ModelProcess(const std::string &inputDataPath, const std::string &idFilePath, uint32_t batchSize):
-    modelId_(0),
-    modelMemSize_(0),
-    modelWeightSize_(0),
-    modelMemPtr_(nullptr),
-    modelWeightPtr_(nullptr),
-    loadFlag_(false),
-    modelDesc_(nullptr),
-    output_(nullptr),
-    inputDataPath_(inputDataPath),
-    input_(nullptr),
-    batchSize_(batchSize),
-    idFilePath_(idFilePath),
-    inputNum_(0),
-    outputNum_(0) {}
-
-ModelProcess::~ModelProcess() {
-    Unload();
-    DestroyResource();
-}
-
-Result ModelProcess::InitResource() {
-    Result ret = CreateDesc();
-    if (ret != SUCCESS) {
-        ERROR_LOG("create model description failed");
-        return FAILED;
-    }
-
-    ret = CreateOutput();
-    if (ret != SUCCESS) {
-        ERROR_LOG("create model output failed");
-        return FAILED;
-    }
-
-    ret = CreateInput();
-    if (ret != SUCCESS) {
-        ERROR_LOG("create model input failed");
-        return FAILED;
-    }
-
-    ret = ReadIdFiles();
-    if (ret != SUCCESS) {
-        ERROR_LOG("read id files failed");
-        return FAILED;
-    }
-}
-void ModelProcess::DestroyResource() {
-    DestroyDesc();
-    DestroyInput();
-    DestroyOutput();
-    return;
-}
-
-Result ModelProcess::LoadModelFromFileWithMem(const char *modelPath) {
-    if (loadFlag_) {
-        ERROR_LOG("has already loaded a model");
-        return FAILED;
-    }
-
-    aclError ret = aclmdlQuerySize(modelPath, &modelMemSize_, &modelWeightSize_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("query model failed, model file is %s", modelPath);
-        return FAILED;
-    }
-
-    ret = aclrtMalloc(&modelMemPtr_, modelMemSize_, ACL_MEM_MALLOC_HUGE_FIRST);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("malloc buffer for mem failed, require size is %zu", modelMemSize_);
-        return FAILED;
-    }
-
-    ret = aclrtMalloc(&modelWeightPtr_, modelWeightSize_, ACL_MEM_MALLOC_HUGE_FIRST);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("malloc buffer for weight failed, require size is %zu", modelWeightSize_);
-        return FAILED;
-    }
-
-    ret = aclmdlLoadFromFileWithMem(modelPath, &modelId_, modelMemPtr_,
-        modelMemSize_, modelWeightPtr_, modelWeightSize_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("load model from file failed, model file is %s", modelPath);
-        return FAILED;
-    }
-
-    loadFlag_ = true;
-    INFO_LOG("load model %s success", modelPath);
-    return SUCCESS;
-}
-
-Result ModelProcess::CreateDesc() {
-    modelDesc_ = aclmdlCreateDesc();
-    if (modelDesc_ == nullptr) {
-        ERROR_LOG("create model description failed");
-        return FAILED;
-    }
-
-    aclError ret = aclmdlGetDesc(modelDesc_, modelId_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("get model description failed");
-        return FAILED;
-    }
-
-    INFO_LOG("create model description success");
-
-    return SUCCESS;
-}
-
-void ModelProcess::DestroyDesc() {
-    if (modelDesc_ != nullptr) {
-        (void)aclmdlDestroyDesc(modelDesc_);
-        modelDesc_ = nullptr;
-    }
-}
-
-Result ModelProcess::CreateInput() {
-    if (modelDesc_ == nullptr) {
-        ERROR_LOG("no model description, create output failed");
-        return FAILED;
-    }
-
-    input_ = aclmdlCreateDataset();
-    if (input_ == nullptr) {
-        ERROR_LOG("can't create dataset, create input failed");
-        return FAILED;
-    }
-
-    size_t inputSize = aclmdlGetNumInputs(modelDesc_);
-    inputNum_ = inputSize;
-    for (size_t i = 0; i < inputSize; ++i) {
-        size_t buffer_size = aclmdlGetInputSizeByIndex(modelDesc_, i);
-        inputBuffSize_.emplace_back(buffer_size);
-
-        void *inputBuffer = nullptr;
-        aclError ret = aclrtMalloc(&inputBuffer, buffer_size, ACL_MEM_MALLOC_NORMAL_ONLY);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("can't malloc buffer, size is %zu, create input failed", buffer_size);
-            return FAILED;
-        }
-
-        aclDataBuffer* inputData = aclCreateDataBuffer(inputBuffer, buffer_size);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("can't create data buffer, create input failed");
-            aclrtFree(inputBuffer);
-            return FAILED;
-        }
-
-        ret = aclmdlAddDatasetBuffer(input_, inputData);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("can't add data buffer, create output failed");
-            aclrtFree(inputBuffer);
-            aclDestroyDataBuffer(inputData);
-            return FAILED;
-        }
-    }
-
-    INFO_LOG("create model input success");
-    return SUCCESS;
-}
-
-void ModelProcess::DestroyInput() {
-    if (input_ == nullptr) {
-        return;
-    }
-
-    for (size_t i = 0; i < aclmdlGetDatasetNumBuffers(input_); ++i) {
-        aclDataBuffer* dataBuffer = aclmdlGetDatasetBuffer(input_, i);
-        aclDestroyDataBuffer(dataBuffer);
-    }
-    aclmdlDestroyDataset(input_);
-    input_ = nullptr;
-}
-
-Result ModelProcess::CreateOutput() {
-    if (modelDesc_ == nullptr) {
-        ERROR_LOG("no model description, create output failed");
-        return FAILED;
-    }
-
-    output_ = aclmdlCreateDataset();
-    if (output_ == nullptr) {
-        ERROR_LOG("can't create dataset, create output failed");
-        return FAILED;
-    }
-
-    size_t outputSize = aclmdlGetNumOutputs(modelDesc_);
-    outputNum_ = outputSize;
-    for (size_t i = 0; i < outputSize; ++i) {
-        size_t buffer_size = aclmdlGetOutputSizeByIndex(modelDesc_, i);
-        outputBuffSize_.emplace_back(buffer_size);
-
-        void *outputBuffer = nullptr;
-        aclError ret = aclrtMalloc(&outputBuffer, buffer_size, ACL_MEM_MALLOC_NORMAL_ONLY);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("can't malloc buffer, size is %zu, create output failed", buffer_size);
-            return FAILED;
-        }
-
-        aclDataBuffer* outputData = aclCreateDataBuffer(outputBuffer, buffer_size);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("can't create data buffer, create output failed");
-            aclrtFree(outputBuffer);
-            return FAILED;
-        }
-
-        ret = aclmdlAddDatasetBuffer(output_, outputData);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("can't add data buffer, create output failed");
-            aclrtFree(outputBuffer);
-            aclDestroyDataBuffer(outputData);
-            return FAILED;
-        }
-    }
-
-    INFO_LOG("create model output success");
-    return SUCCESS;
-}
-
-void ModelProcess::OutputModelResult() {
-    for (size_t i = 0; i < aclmdlGetDatasetNumBuffers(output_); ++i) {
-        aclDataBuffer* dataBuffer = aclmdlGetDatasetBuffer(output_, i);
-        void* data = aclGetDataBufferAddr(dataBuffer);
-        uint32_t len = aclGetDataBufferSizeV2(dataBuffer);
-
-        void *outHostData = NULL;
-        aclError ret = ACL_ERROR_NONE;
-        float *outData = NULL;
-        if (!g_isDevice) {
-            ret = aclrtMallocHost(&outHostData, len);
-            if (ret != ACL_ERROR_NONE) {
-                ERROR_LOG("aclrtMallocHost failed, ret[%d]", ret);
-                return;
-            }
-
-            ret = aclrtMemcpy(outHostData, len, data, len, ACL_MEMCPY_DEVICE_TO_HOST);
-            if (ret != ACL_ERROR_NONE) {
-                ERROR_LOG("aclrtMemcpy failed, ret[%d]", ret);
-                (void)aclrtFreeHost(outHostData);
-                return;
-            }
-
-            outData = reinterpret_cast<float *>(outHostData);
-        } else {
-            outData = reinterpret_cast<float *>(data);
-        }
-        std::map<float, unsigned int, std::greater<float> > resultMap;
-        for (unsigned int j = 0; j < len / sizeof(float); ++j) {
-            resultMap[*outData] = j;
-            outData++;
-        }
-
-        int cnt = 0;
-        for (auto it = resultMap.begin(); it != resultMap.end(); ++it) {
-            // print top 5
-            if (++cnt > 5) {
-                break;
-            }
-
-            INFO_LOG("top %d: index[%d] value[%lf]", cnt, it->second, it->first);
-        }
-        if (!g_isDevice) {
-            ret = aclrtFreeHost(outHostData);
-            if (ret != ACL_ERROR_NONE) {
-                ERROR_LOG("aclrtFreeHost failed, ret[%d]", ret);
-                return;
-            }
-        }
-    }
-
-    INFO_LOG("output data success");
-    return;
-}
-
-void ModelProcess::DestroyOutput() {
-    if (output_ == nullptr) {
-        return;
-    }
-
-    for (size_t i = 0; i < aclmdlGetDatasetNumBuffers(output_); ++i) {
-        aclDataBuffer* dataBuffer = aclmdlGetDatasetBuffer(output_, i);
-        void* data = aclGetDataBufferAddr(dataBuffer);
-        (void)aclrtFree(data);
-        (void)aclDestroyDataBuffer(dataBuffer);
-    }
-
-    (void)aclmdlDestroyDataset(output_);
-    output_ = nullptr;
-}
-
-Result ModelProcess::CpyFileToDevice(std::string fileName, uint32_t inputNum) {
-    uint32_t inputHostBuffSize = 0;
-    void *inputHostBuff = Utils::ReadBinFile(fileName, &inputHostBuffSize);
-    if (inputHostBuff == nullptr) {
-        return FAILED;
-    }
-    aclDataBuffer *inBufferDev = aclmdlGetDatasetBuffer(input_, inputNum);
-    void *p_batchDst = aclGetDataBufferAddr(inBufferDev);
-    aclrtMemset(p_batchDst, inputHostBuffSize, 0, inputHostBuffSize);
-    aclError ret = aclrtMemcpy(p_batchDst, inputHostBuffSize, inputHostBuff, inputHostBuffSize,
-                               ACL_MEMCPY_HOST_TO_DEVICE);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("memcpy failed. device buffer size is %u, input host buffer size is %u",
-                  inputHostBuffSize, inputHostBuffSize);
-        aclrtFreeHost(inputHostBuff);
-        return FAILED;
-    }
-    aclrtFreeHost(inputHostBuff);
-    return SUCCESS;
-}
-
-Result ModelProcess::CpyDataToDevice(void *data, uint32_t len, uint32_t inputNum) {
-    if (len != inputBuffSize_[inputNum]) {
-        return FAILED;
-    }
-    aclDataBuffer *inBufferDev = aclmdlGetDatasetBuffer(input_, inputNum);
-    void *p_batchDst = aclGetDataBufferAddr(inBufferDev);
-    aclrtMemset(p_batchDst, len, 0, len);
-    aclError ret = aclrtMemcpy(p_batchDst, len, data, len, ACL_MEMCPY_HOST_TO_DEVICE);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("memcpy failed. device buffer size is %u, input host buffer size is %u",
-                  len, len);
-        return FAILED;
-    }
-    return SUCCESS;
-}
-
-void ModelProcess::CpyOutputFromDeviceToHost(uint32_t index) {
-    size_t outputNum = aclmdlGetDatasetNumBuffers(output_);
-
-    for (size_t i = 0; i < outputNum; ++i) {
-        aclDataBuffer* dataBuffer = aclmdlGetDatasetBuffer(output_, i);
-        void* data = aclGetDataBufferAddr(dataBuffer);
-        uint32_t bufferSize = aclGetDataBufferSizeV2(dataBuffer);
-
-        void* outHostData = NULL;
-        aclError ret = ACL_ERROR_NONE;
-        ret = aclrtMallocHost(&outHostData, bufferSize);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("aclrtMallocHost failed, ret[%d]", ret);
-            return;
-        }
-        ret = aclrtMemcpy(outHostData, bufferSize, data, bufferSize, ACL_MEMCPY_DEVICE_TO_HOST);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("aclrtMemcpy failed, ret[%d]", ret);
-            (void)aclrtFreeHost(outHostData);
-            return;
-        }
-
-        uint32_t len = (uint32_t)bufferSize / batchSize_;
-        for (size_t j = 0; j < batchSize_; j++) {
-            result_.emplace(ids_[index][j], reinterpret_cast<uint8_t *>(outHostData) + (j * len));
-        }
-    }
-    return;
-}
-
-std::vector<std::vector<void *>> ModelProcess::ReadInputFiles(std::vector<std::vector<std::string>> inputFiles,
-                                                              size_t inputSize,
-                                                              std::vector<std::vector<uint32_t>> *fileSize) {
-    size_t fileNum = inputFiles[0].size();
-    std::vector<std::vector<void *>> buff(fileNum);
-    if (inputFiles.size() != inputSize) {
-        std::cout << "the num of input file is incorrect" << std::endl;
-        return buff;
-    }
-
-    void *inputHostBuff = nullptr;
-    uint32_t inputHostBuffSize = 0;
-    for (int i = 0; i < inputSize; ++i) {
-        for (int j = 0; j < fileNum; ++j) {
-            inputHostBuff = Utils::ReadBinFile(inputFiles[i][j], &inputHostBuffSize);
-            buff[i].emplace_back(inputHostBuff);
-            (*fileSize)[i].emplace_back(inputHostBuffSize);
-        }
-    }
-
-    return buff;
-}
-
-Result ModelProcess::ReadIdFiles() {
-    std::vector<std::string> idFiles = Utils::GetAllBins(idFilePath_);
-
-    for (int i = 0; i < idFiles.size(); ++i) {
-        std::vector<int> ids;
-        Utils::ReadFileToVector(idFiles[i], batchSize_, &ids);
-        ids_.emplace_back(ids);
-    }
-    return SUCCESS;
-}
-
-uint32_t ModelProcess::ReadFiles() {
-    size_t inputSize = aclmdlGetNumInputs(modelDesc_);
-    std::vector<std::vector<uint32_t>> fileSize(inputSize);
-    std::vector<std::vector<std::string>> inputFiles = Utils::GetAllInputData(inputDataPath_);
-
-    fileBuff_ = ReadInputFiles(inputFiles, inputSize, &fileSize);
-    uint32_t fileNum = inputFiles[0].size();
-    fileSize_ = fileSize;
-    return fileNum;
-}
-
-Result ModelProcess::ExecuteWithFile(uint32_t fileNum) {
-    for (size_t index = 0; index < fileNum; ++index) {
-        struct timeval start;
-        struct timeval end;
-        double startTime_ms;
-        double endTime_ms;
-        gettimeofday(&start, NULL);
-        void *picDevBuffer = nullptr;
-
-        for (auto i = 0; i < inputNum_; ++i) {
-            CpyDataToDevice(fileBuff_[i][index], fileSize_[i][index], i);
-        }
-
-        aclError ret = aclmdlExecute(modelId_, input_, output_);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("execute model failed, modelId is %u", modelId_);
-            return FAILED;
-        }
-
-        CpyOutputFromDeviceToHost(index);
-        gettimeofday(&end, NULL);
-        startTime_ms = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
-        endTime_ms = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
-        costTime_map_.insert(std::pair<double, double>(startTime_ms, endTime_ms));
-    }
-    return SUCCESS;
-}
-
-Result ModelProcess::Execute(uint32_t index) {
-    struct timeval start;
-    struct timeval end;
-    double startTime_ms;
-    double endTime_ms;
-
-    gettimeofday(&start, NULL);
-    aclError ret = aclmdlExecute(modelId_, input_, output_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("execute model failed, modelId is %u", modelId_);
-        return FAILED;
-    }
-
-    CpyOutputFromDeviceToHost(index);
-    gettimeofday(&end, NULL);
-    startTime_ms = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
-    endTime_ms = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
-    costTime_map_.insert(std::pair<double, double>(startTime_ms, endTime_ms));
-    return SUCCESS;
-}
-
-std::map<int, void *> ModelProcess::GetResult() {
-    return result_;
-}
-
-void ModelProcess::Unload() {
-    if (!loadFlag_) {
-        WARN_LOG("no model had been loaded, unload failed");
-        return;
-    }
-
-    aclError ret = aclmdlUnload(modelId_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("unload model failed, modelId is %u", modelId_);
-    }
-
-    if (modelDesc_ != nullptr) {
-        (void)aclmdlDestroyDesc(modelDesc_);
-        modelDesc_ = nullptr;
-    }
-
-    if (modelMemPtr_ != nullptr) {
-        aclrtFree(modelMemPtr_);
-        modelMemPtr_ = nullptr;
-        modelMemSize_ = 0;
-    }
-
-    if (modelWeightPtr_ != nullptr) {
-        aclrtFree(modelWeightPtr_);
-        modelWeightPtr_ = nullptr;
-        modelWeightSize_ = 0;
-    }
-
-    loadFlag_ = false;
-    INFO_LOG("unload model success, modelId is %u", modelId_);
-}
-
-std::vector<uint32_t> ModelProcess::GetInputSize() {
-    return inputBuffSize_;
-}
-
-std::vector<uint32_t> ModelProcess::GetOutputSize() {
-    return outputBuffSize_;
-}
-
-std::string ModelProcess::GetInputDataPath() {
-    return inputDataPath_;
-}
-
-std::string ModelProcess::GetCostTimeInfo() {
-    double average = 0.0;
-    int infer_cnt = 0;
-
-    for (auto iter = costTime_map_.begin(); iter != costTime_map_.end(); iter++) {
-        double diff = 0.0;
-        diff = iter->second - iter->first;
-        average += diff;
-        infer_cnt++;
-    }
-    average = average / infer_cnt;
-
-    std::stringstream timeCost;
-    timeCost << "first model latency "<< average << " ms; count " << infer_cnt << std::endl;
-
-    return timeCost.str();
-}

official/recommend/naml/ascend310_infer/src/sample_process.cpp

-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * 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 <dirent.h>
-#include <sys/time.h>
-#include <time.h>
-#include <string>
-#include <iostream>
-#include <vector>
-#include <map>
-#include <queue>
-#include <unordered_map>
-#include <iterator>
-#include <thread>
-#include <sstream>
-
-#include "acl/acl.h"
-#include "../inc/utils.h"
-#include "../inc/model_process.h"
-#include "../inc/sample_process.h"
-
-extern bool g_isDevice;
-
-SampleProcess::SampleProcess() :deviceId_(0), context_(nullptr), stream_(nullptr), threadNum_(0) {}
-
-SampleProcess::SampleProcess(uint32_t deviceId, uint32_t threadNum):
-    deviceId_(deviceId),
-    threadNum_(threadNum),
-    context_(nullptr),
-    stream_(nullptr) {}
-
-SampleProcess::~SampleProcess() {
-    DestroyResource();
-}
-
-Result SampleProcess::InitResource() {
-    aclError ret = aclInit(nullptr);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("acl init failed");
-        return FAILED;
-    }
-    INFO_LOG("acl init success");
-
-    ret = aclrtSetDevice(deviceId_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("acl open device %d failed", deviceId_);
-        return FAILED;
-    }
-    INFO_LOG("open device %d success", deviceId_);
-
-    // create context (set current)
-    ret = aclrtCreateContext(&context_, deviceId_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("acl create context failed");
-        return FAILED;
-    }
-    INFO_LOG("create context success");
-
-    // create stream
-    ret = aclrtCreateStream(&stream_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("acl create stream failed");
-        return FAILED;
-    }
-    INFO_LOG("create stream success");
-
-    // get run mode
-    aclrtRunMode runMode;
-    ret = aclrtGetRunMode(&runMode);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("acl get run mode failed");
-        return FAILED;
-    }
-    g_isDevice = (runMode == ACL_DEVICE);
-    INFO_LOG("get run mode success");
-
-    return SUCCESS;
-}
-
-Result SampleProcess::CreateModelProcessInstance(const std::vector<std::string> &omPaths,
-                                                 const std::vector<std::string> &inputDataPaths,
-                                                 const std::vector<std::string> &inputIdPaths,
-                                                 uint32_t batchSize,
-                                                 std::map<int, std::shared_ptr<ModelProcess>> *modelProcessContainer) {
-    for (int i = 0; i < omPaths.size(); ++i) {
-        std::cout << "om_path : " << omPaths[i]  << std::endl;
-        auto processModel = std::make_shared<ModelProcess>(inputDataPaths[i], inputIdPaths[i], batchSize);
-        Result ret = processModel->LoadModelFromFileWithMem(omPaths[i].c_str());
-        if (ret != SUCCESS) {
-            ERROR_LOG("load model from file failed");
-            return FAILED;
-        }
-
-        ret = processModel->InitResource();
-        if (ret != SUCCESS) {
-            ERROR_LOG("create model description failed");
-            return FAILED;
-        }
-        modelProcessContainer->emplace(i, processModel);
-    }
-
-    return SUCCESS;
-}
-
-std::vector<std::vector<std::vector<int>>> SampleProcess::ReadHistory(std::vector<std::string> historyFile,
-                                                                      uint32_t batchSize) {
-    std::vector<std::vector<std::vector<int>>> allHistory;
-    for (auto &file : historyFile) {
-        std::vector<std::vector<int>> history(batchSize, std::vector<int>(50));
-        Utils::ReadFileToVector(file, batchSize, 50, &history);
-        allHistory.emplace_back(history);
-    }
-
-    return allHistory;
-}
-
-Result SampleProcess::Process(const std::vector<std::string> &omPaths,
-                              const std::vector<std::string> &inputDataPaths,
-                              const std::vector<std::string> &inputIdPaths,
-                              const std::string &browsedNewsPath,
-                              uint32_t batchSize) {
-    INFO_LOG("Start do sample process");
-    struct timeval totalStart;
-    struct timeval totalEnd;
-    std::map<int, std::shared_ptr<ModelProcess>> modelProcessContainer;
-
-    CreateModelProcessInstance(omPaths, inputDataPaths, inputIdPaths, batchSize, &modelProcessContainer);
-
-    uint32_t fileNum = modelProcessContainer[0]->ReadFiles();
-    std::string historyDir = modelProcessContainer[1]->GetInputDataPath() + "/00_history_data";
-    std::vector<std::string> historyFile = Utils::GetAllBins(historyDir);
-
-    size_t historySize = historyFile.size();
-    std::vector<std::vector<std::vector<int>>> allHistory = ReadHistory(historyFile, batchSize);
-
-    uint32_t browsedFileNum = ReadBrowsedData(browsedNewsPath);
-
-    gettimeofday(&totalStart, NULL);
-    modelProcessContainer[0]->ExecuteWithFile(fileNum);
-
-    std::map<int, void *> result = modelProcessContainer[0]->GetResult();
-
-    std::vector<uint32_t> model1OutputBuffSize = modelProcessContainer[0]->GetOutputSize();
-    std::vector<uint32_t> inputBuffSize = modelProcessContainer[1]->GetInputSize();
-
-    uint32_t singleDatsSize = model1OutputBuffSize[0] / batchSize;
-    void* browedNews = NULL;
-    aclrtMallocHost(&browedNews, inputBuffSize[0]);
-
-    struct timeval start;
-    struct timeval end;
-    double startTime_ms;
-    double endTime_ms;
-    for (int i = 0; i < historySize; ++i) {
-        gettimeofday(&start, NULL);
-        for (int j = 0; j < 16; ++j) {
-            for (int k = 0; k < 50; ++k) {
-                auto it = result.find(allHistory[i][j][k]);
-                if (it != result.end()) {
-                    aclrtMemcpy(reinterpret_cast<uint8_t *>(browedNews) + (j * 50 + k) * singleDatsSize, singleDatsSize,
-                                result[allHistory[i][j][k]], singleDatsSize, ACL_MEMCPY_HOST_TO_HOST);
-                }
-            }
-        }
-        modelProcessContainer[1]->CpyDataToDevice(browedNews, inputBuffSize[0], 0);
-        modelProcessContainer[1]->Execute(i);
-        gettimeofday(&end, NULL);
-        startTime_ms = (1.0 * start.tv_sec * 1000000 + start.tv_usec) / 1000;
-        endTime_ms = (1.0 * end.tv_sec * 1000000 + end.tv_usec) / 1000;
-        secondModelCostTime_map_.insert(std::pair<double, double>(startTime_ms, endTime_ms));
-    }
-
-    GetPred(&modelProcessContainer, browsedFileNum);
-    gettimeofday(&totalEnd, NULL);
-    startTime_ms = (1.0 * totalStart.tv_sec * 1000000 + totalStart.tv_usec) / 1000;
-    endTime_ms = (1.0 * totalEnd.tv_sec * 1000000 + totalEnd.tv_usec) / 1000;
-    totalCostTime_map_.insert(std::pair<double, double>(startTime_ms, endTime_ms));
-    time_cost_.clear();
-    time_cost_.emplace_back(modelProcessContainer[0]->GetCostTimeInfo());
-    aclrtFreeHost(browedNews);
-    return SUCCESS;
-}
-
-Result SampleProcess::ReadBrowsedFile(const std::string &browsedNewsPath,
-                                      std::vector<std::string> userIdFiles,
-                                      std::vector<std::vector<int>> *usersIds,
-                                      std::vector<std::vector<int>> *candidateNewsIds) {
-    std::vector<int> candidateNewsId;
-    std::vector<int> usersId;
-    for (auto file : userIdFiles) {
-        candidateNewsId.clear();
-        usersId.clear();
-        std::size_t pos = file.rfind("/");
-        std::string name = file.substr(pos);
-
-        std::string newsIdFileName = browsedNewsPath + "/01_candidate_nid_data" + name;
-
-        Utils::ReadFileToVector(file, &usersId);
-        Utils::ReadFileToVector(newsIdFileName, &candidateNewsId);
-
-        usersIds->emplace_back(usersId);
-        candidateNewsIds->emplace_back(candidateNewsId);
-    }
-    return SUCCESS;
-}
-
-uint32_t SampleProcess::ReadBrowsedData(const std::string &browsedNewsPath) {
-    userIdFiles_ = Utils::GetAllBins(browsedNewsPath + "/00_user_id_data");
-    ReadBrowsedFile(browsedNewsPath, userIdFiles_, &usersIds_, &candidateNewsIds_);
-    uint32_t fileNum = userIdFiles_.size();
-
-    return fileNum;
-}
-
-Result SampleProcess::GetPred(std::map<int, std::shared_ptr<ModelProcess>> *modelProcessContainer,
-                              uint32_t fileNum) {
-    std::map<int, void *> newsEncodeResult = (*modelProcessContainer)[0]->GetResult();
-    std::map<int, void *> userEncodeResult = (*modelProcessContainer)[1]->GetResult();
-
-    uint32_t perThreadNum = fileNum / threadNum_;
-    std::vector<std::thread> threads;
-
-    for (int i = 0; i < threadNum_; ++i) {
-        if (i != threadNum_ - 1) {
-            threads.emplace_back(std::thread(&SampleProcess::GetResult, this,
-                                             i * perThreadNum, (i + 1) * perThreadNum,
-                                             newsEncodeResult,
-                                             userEncodeResult));
-        } else {
-            threads.emplace_back(std::thread(&SampleProcess::GetResult, this,
-                                             i * perThreadNum,
-                                             fileNum,
-                                             newsEncodeResult,
-                                             userEncodeResult));
-        }
-    }
-    for (int i = 0; i < threads.size(); ++i) {
-        threads[i].join();
-    }
-
-    return SUCCESS;
-}
-void SampleProcess::GetResult(uint32_t startPos, uint32_t endPos,
-                              std::map<int, void *> newsEncodeResult,
-                              std::map<int, void *> userEncodeResult) {
-    for (int i = startPos; i < endPos; ++i) {
-        std::vector<std::vector<float>> newsCandidate;
-        std::vector<float> userEncodeIds(400);
-        for (int j = 0; j < candidateNewsIds_[i].size(); ++j) {
-            std::vector<float> newsResults(400);
-            float *newsResult = reinterpret_cast<float *>(newsEncodeResult[candidateNewsIds_[i][j]]);
-            std::copy(newsResult, newsResult + 400, newsResults.begin());
-            newsCandidate.emplace_back(newsResults);
-        }
-        float *userResult = reinterpret_cast<float *>(userEncodeResult[usersIds_[i][0]]);
-        std::copy(userResult, userResult + 400, userEncodeIds.begin());
-
-        std::vector<float> predResult;
-        for (int j = 0; j < newsCandidate.size(); ++j) {
-            float dotMulResult = 0;
-            for (int k = 0; k < 400; ++k) {
-                dotMulResult += newsCandidate[j][k] * userEncodeIds[k];
-            }
-            predResult.emplace_back(dotMulResult);
-        }
-        mtx_.lock();
-        WriteResult(userIdFiles_[i], predResult, predResult.size() * 4);
-        mtx_.unlock();
-    }
-
-    return;
-}
-
-int SampleProcess::WriteResult(const std::string& imageFile, std::vector<float> result, uint32_t size) {
-    std::string homePath = "./result_Files/";
-    std::size_t pos = imageFile.rfind("/");
-    std::string name = imageFile.substr(pos);
-    for (size_t i = 0; i < 1; ++i) {
-        std::string outFileName = homePath + "/" + name;
-        try {
-            FILE *outputFile = fopen(outFileName.c_str(), "wb");
-            fwrite(static_cast<void *>(&result[0]), size, sizeof(char), outputFile);
-            fclose(outputFile);
-            outputFile = nullptr;
-        } catch (std::exception &e) {
-            std::cout << "write result file " << outFileName << " failed, error info: " << e.what() << std::endl;
-            return FAILED;
-        }
-    }
-    return SUCCESS;
-}
-
-void SampleProcess::DestroyResource() {
-    aclError ret;
-    if (stream_ != nullptr) {
-        ret = aclrtDestroyStream(stream_);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("destroy stream failed");
-        }
-        stream_ = nullptr;
-    }
-    INFO_LOG("end to destroy stream");
-
-    if (context_ != nullptr) {
-        ret = aclrtDestroyContext(context_);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("destroy context failed");
-        }
-        context_ = nullptr;
-    }
-    INFO_LOG("end to destroy context");
-
-    ret = aclFinalize();
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("finalize acl failed");
-    }
-    INFO_LOG("end to finalize acl");
-
-    ret = aclrtResetDevice(deviceId_);
-    if (ret != ACL_ERROR_NONE) {
-        ERROR_LOG("reset device failed");
-    }
-    INFO_LOG("end to reset device is %d", deviceId_);
-}
-
-std::vector<std::string> SampleProcess::GetModelExecCostTimeInfo() {
-    double secondModelAverage = 0.0;
-    int infer_cnt = 0;
-
-    for (auto iter = secondModelCostTime_map_.begin(); iter != secondModelCostTime_map_.end(); iter++) {
-        double diff = 0.0;
-        diff = iter->second - iter->first;
-        secondModelAverage += diff;
-        infer_cnt++;
-    }
-    secondModelAverage = secondModelAverage / infer_cnt;
-    std::stringstream timeCost;
-    timeCost << "second model inference cost average time: "<< secondModelAverage <<
-        " ms of infer_count " << infer_cnt << std::endl;
-    time_cost_.emplace_back(timeCost.str());
-
-    double totalCostTime = totalCostTime_map_.begin()->second - totalCostTime_map_.begin()->first;
-    std::stringstream totalTimeCost;
-    totalTimeCost << "total inference cost time: "<< totalCostTime << " ms; count " << infer_cnt << std::endl;
-    time_cost_.emplace_back(totalTimeCost.str());
-
-    return time_cost_;
-}

official/recommend/naml/ascend310_infer/src/utils.cc

+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * 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 <fstream>
+#include <algorithm>
+#include <iostream>
+#include "inc/utils.h"
+
+using mindspore::MSTensor;
+using mindspore::DataType;
+
+std::vector<std::string> GetAllFiles(std::string_view dirName) {
+  struct dirent *filename;
+  DIR *dir = OpenDir(dirName);
+  if (dir == nullptr) {
+    return {};
+  }
+  std::vector<std::string> res;
+  while ((filename = readdir(dir)) != nullptr) {
+    std::string dName = std::string(filename->d_name);
+    if (dName == "." || dName == ".." || filename->d_type != DT_REG) {
+      continue;
+    }
+    res.emplace_back(std::string(dirName) + "/" + filename->d_name);
+  }
+  std::sort(res.begin(), res.end());
+  for (auto &f : res) {
+    std::cout << "image file: " << f << std::endl;
+  }
+  return res;
+}
+
+int WriteResult(const std::string& imageFile, const std::vector<MSTensor> &outputs) {
+  std::string homePath = "./result_Files";
+  const int INVALID_POINTER = -1;
+  const int ERROR = -2;
+  for (size_t i = 0; i < outputs.size(); ++i) {
+    size_t outputSize;
+    std::shared_ptr<const void> netOutput;
+    netOutput = outputs[i].Data();
+    outputSize = outputs[i].DataSize();
+    int pos = imageFile.rfind('/');
+    std::string fileName(imageFile, pos + 1);
+    fileName.replace(fileName.find('.'), fileName.size() - fileName.find('.'), '_' + std::to_string(i) + ".bin");
+    std::string outFileName = homePath + "/" + fileName;
+    FILE * outputFile = fopen(outFileName.c_str(), "wb");
+    if (outputFile == nullptr) {
+        std::cout << "open result file " << outFileName << " failed" << std::endl;
+        return INVALID_POINTER;
+    }
+    size_t size = fwrite(netOutput.get(), sizeof(char), outputSize, outputFile);
+    if (size != outputSize) {
+        fclose(outputFile);
+        outputFile = nullptr;
+        std::cout << "write result file " << outFileName << " failed, write size[" << size <<
+            "] is smaller than output size[" << outputSize << "], maybe the disk is full." << std::endl;
+        return ERROR;
+    }
+    fclose(outputFile);
+    outputFile = nullptr;
+  }
+  return 0;
+}
+
+mindspore::MSTensor ReadFileToTensor(const std::string &file) {
+  if (file.empty()) {
+    std::cout << "Pointer file is nullptr" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  std::ifstream ifs(file);
+  if (!ifs.good()) {
+    std::cout << "File: " << file << " is not exist" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  if (!ifs.is_open()) {
+    std::cout << "File: " << file << "open failed" << std::endl;
+    return mindspore::MSTensor();
+  }
+
+  ifs.seekg(0, std::ios::end);
+  size_t size = ifs.tellg();
+  mindspore::MSTensor buffer(file, mindspore::DataType::kNumberTypeUInt8, {static_cast<int64_t>(size)}, nullptr, size);
+
+  ifs.seekg(0, std::ios::beg);
+  ifs.read(reinterpret_cast<char *>(buffer.MutableData()), size);
+  ifs.close();
+
+  return buffer;
+}
+
+DIR *OpenDir(std::string_view dirName) {
+  if (dirName.empty()) {
+    std::cout << " dirName is null ! " << std::endl;
+    return nullptr;
+  }
+  std::string realPath = RealPath(dirName);
+  struct stat s;
+  lstat(realPath.c_str(), &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dirName is not a valid directory !" << std::endl;
+    return nullptr;
+  }
+  DIR *dir;
+  dir = opendir(realPath.c_str());
+  if (dir == nullptr) {
+    std::cout << "Can not open dir " << dirName << std::endl;
+    return nullptr;
+  }
+  std::cout << "Successfully opened the dir " << dirName << std::endl;
+  return dir;
+}
+
+std::string RealPath(std::string_view path) {
+  char realPathMem[PATH_MAX] = {0};
+  char *realPathRet = nullptr;
+  realPathRet = realpath(path.data(), realPathMem);
+
+  if (realPathRet == nullptr) {
+    std::cout << "File: " << path << " is not exist.";
+    return "";
+  }
+
+  std::string realPath(realPathMem);
+  std::cout << path << " realpath is: " << realPath << std::endl;
+  return realPath;
+}
+

official/recommend/naml/ascend310_infer/src/utils.cpp

-/**
- * Copyright 2021 Huawei Technologies Co., Ltd
- *
- * 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 <sys/stat.h>
-#include <iostream>
-#include <fstream>
-#include <cstring>
-#include <vector>
-#include <algorithm>
-#include "acl/acl.h"
-#include "../inc/utils.h"
-
-extern bool g_isDevice;
-
-void* Utils::ReadBinFile(std::string fileName, uint32_t *fileSize) {
-    struct stat sBuf;
-    int fileStatus = stat(fileName.data(), &sBuf);
-    if (fileStatus == -1) {
-        ERROR_LOG("failed to get file");
-        return nullptr;
-    }
-    if (S_ISREG(sBuf.st_mode) == 0) {
-        ERROR_LOG("%s is not a file, please enter a file", fileName.c_str());
-        return nullptr;
-    }
-
-    std::ifstream binFile(fileName, std::ifstream::binary);
-    if (binFile.is_open() == false) {
-        ERROR_LOG("open file %s failed", fileName.c_str());
-        return nullptr;
-    }
-
-    binFile.seekg(0, binFile.end);
-    uint32_t binFileBufferLen = binFile.tellg();
-    if (binFileBufferLen == 0) {
-        ERROR_LOG("binfile is empty, filename is %s", fileName.c_str());
-        binFile.close();
-        return nullptr;
-    }
-
-    binFile.seekg(0, binFile.beg);
-
-    void* binFileBufferData = nullptr;
-    if (!g_isDevice) {
-        aclrtMallocHost(&binFileBufferData, binFileBufferLen);
-        if (binFileBufferData == nullptr) {
-            ERROR_LOG("malloc binFileBufferData failed");
-            binFile.close();
-            return nullptr;
-        }
-    } else {
-        aclError ret = aclrtMalloc(&binFileBufferData, binFileBufferLen, ACL_MEM_MALLOC_NORMAL_ONLY);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("malloc device buffer failed. size is %u", binFileBufferLen);
-            binFile.close();
-            return nullptr;
-        }
-    }
-    binFile.read(static_cast<char *>(binFileBufferData), binFileBufferLen);
-    binFile.close();
-    *fileSize = binFileBufferLen;
-    return binFileBufferData;
-}
-
-void* Utils::GetDeviceBufferOfFile(std::string fileName, uint32_t *fileSize) {
-    uint32_t inputHostBuffSize = 0;
-    void* inputHostBuff = Utils::ReadBinFile(fileName, &inputHostBuffSize);
-    if (inputHostBuff == nullptr) {
-        return nullptr;
-    }
-    if (!g_isDevice) {
-        void *inBufferDev = nullptr;
-        uint32_t inBufferSize = inputHostBuffSize;
-        aclError ret = aclrtMalloc(&inBufferDev, inBufferSize, ACL_MEM_MALLOC_NORMAL_ONLY);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("malloc device buffer failed. size is %u", inBufferSize);
-            aclrtFreeHost(inputHostBuff);
-            return nullptr;
-        }
-
-        ret = aclrtMemcpy(inBufferDev, inBufferSize, inputHostBuff, inputHostBuffSize, ACL_MEMCPY_HOST_TO_DEVICE);
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("memcpy failed. device buffer size is %u, input host buffer size is %u",
-                inBufferSize, inputHostBuffSize);
-            aclrtFree(inBufferDev);
-            aclrtFreeHost(inputHostBuff);
-            return nullptr;
-        }
-        aclrtFreeHost(inputHostBuff);
-        *fileSize = inBufferSize;
-        return inBufferDev;
-    } else {
-        *fileSize = inputHostBuffSize;
-        return inputHostBuff;
-    }
-}
-
-std::vector<std::vector<std::string>> Utils::GetAllInputData(std::string dir_name) {
-    DIR *dir = OpenDir(dir_name);
-    if (dir == nullptr) {
-        return {};
-    }
-    struct dirent *filename;
-
-    std::vector<std::string> sub_dirs;
-    while ((filename = readdir(dir)) != nullptr) {
-        std::string d_name = std::string(filename->d_name);
-        if (d_name == "." || d_name == ".." || d_name.empty() || d_name[0] != '0') {
-            continue;
-        }
-
-        std::string dir_path = RealPath(std::string(dir_name) + "/" + filename->d_name);
-        struct stat s;
-        lstat(dir_path.c_str(), &s);
-        if (!S_ISDIR(s.st_mode)) {
-            continue;
-        }
-
-        sub_dirs.emplace_back(dir_path);
-    }
-    std::sort(sub_dirs.begin(), sub_dirs.end());
-
-    std::vector<std::vector<std::string>> result(sub_dirs.size());
-
-    std::transform(sub_dirs.begin(), sub_dirs.end(), result.begin(), GetAllBins);
-
-    return result;
-}
-
-DIR *Utils::OpenDir(std::string dir_name) {
-    // check the parameter !
-    if (dir_name.empty()) {
-        std::cout << " dir_name is null ! " << std::endl;
-        return nullptr;
-    }
-
-    std::string real_path = RealPath(dir_name);
-
-    // check if dir_name is a valid dir
-    struct stat s;
-    lstat(real_path.c_str(), &s);
-    if (!S_ISDIR(s.st_mode)) {
-        std::cout << "dir_name is not a valid directory !" << std::endl;
-        return nullptr;
-    }
-
-    DIR *dir = opendir(real_path.c_str());
-    if (dir == nullptr) {
-        std::cout << "Can not open dir " << dir_name << std::endl;
-        return nullptr;
-    }
-    std::cout << "Successfully opened the dir " << dir_name << std::endl;
-    return dir;
-}
-
-std::string Utils::RealPath(std::string path) {
-    char real_path_mem[PATH_MAX] = {0};
-    char *real_path_ret = nullptr;
-    real_path_ret = realpath(path.data(), real_path_mem);
-    if (real_path_ret == nullptr) {
-        std::cout << "File: " << path << " is not exist.";
-        return "";
-    }
-
-    std::string real_path(real_path_mem);
-    std::cout << path << " realpath is: " << real_path << std::endl;
-    return real_path;
-}
-
-std::vector<std::string> Utils::GetAllBins(std::string dir_name) {
-    struct dirent *filename;
-    DIR *dir = OpenDir(dir_name);
-    if (dir == nullptr) {
-        return {};
-    }
-
-    std::vector<std::string> res;
-    while ((filename = readdir(dir)) != nullptr) {
-        std::string d_name = std::string(filename->d_name);
-        if (d_name == "." || d_name == ".." || d_name.size() <= 3 || d_name.substr(d_name.size() - 4) != ".bin" ||
-            filename->d_type != DT_REG) {
-            continue;
-        }
-        res.emplace_back(std::string(dir_name) + "/" + filename->d_name);
-    }
-
-    std::sort(res.begin(), res.end());
-
-    return res;
-}
-
-Result Utils::ReadFileToVector(std::string newsIdFileName, std::vector<int> *newsId) {
-    int id;
-
-    std::ifstream in(newsIdFileName, std::ios::in | std::ios::binary);
-    while (in.read(reinterpret_cast<char *>(&id), sizeof(id))) {
-        newsId->emplace_back(id);
-    }
-    in.close();
-
-    return SUCCESS;
-}
-
-Result Utils::ReadFileToVector(std::string newsIdFileName, uint32_t batchSize, std::vector<int> *newsId) {
-    int id;
-
-    std::ifstream in(newsIdFileName, std::ios::in | std::ios::binary);
-    for (int i = 0; i < batchSize; ++i) {
-        in.read(reinterpret_cast<char *>(&id), sizeof(id));
-        newsId->emplace_back(id);
-    }
-    in.close();
-
-    return SUCCESS;
-}
-
-Result Utils::ReadFileToVector(std::string fileName, uint32_t batchSize,
-                               uint32_t count, std::vector<std::vector<int>> *newsId) {
-    int id;
-
-    std::ifstream in(fileName, std::ios::in | std::ios::binary);
-    for (int i = 0; i < batchSize; ++i) {
-        for (int j = 0; j < count; ++j) {
-            in.read(reinterpret_cast<char *>(&id), sizeof(id));
-            (*newsId)[i][j] = id;
-        }
-    }
-    in.close();
-
-    return SUCCESS;
-}

official/recommend/naml/postprocess.py

@@ -15,16 +15,11 @@
 
 """Evaluation for NAML"""
 import os
-import argparse
+from model_utils.config import config
 import numpy as np
 
 from sklearn.metrics import roc_auc_score
 
-parser = argparse.ArgumentParser(description="")
-parser.add_argument("--result_path", type=str, default="", help="Device id")
-parser.add_argument("--label_path", type=str, default="", help="output file name.")
-args = parser.parse_args()
-
 def AUC(y_true, y_pred):
     return roc_auc_score(y_true, y_pred)
 
@@ -79,15 +74,23 @@ class NAMLMetric:
 
 def get_metric(result_path, label_path, metric):
     """get accuracy"""
-    result_files = os.listdir(result_path)
-    for file in result_files:
-        result_file = os.path.join(result_path, file)
+    label_list = os.listdir(label_path)
+    for file in label_list:
+        f = file.split(".bin")[0] + "_0.bin"
+        result_file = os.path.join(result_path, f)
+        if not os.path.exists(result_file):
+            print("exclude file:", file)
+            continue
         pred = np.fromfile(result_file, dtype=np.float32)
+        if pred.size == 0:
+            print("exclude file:", file)
+            continue
 
         label_file = os.path.join(label_path, file)
         label = np.fromfile(label_file, dtype=np.int32)
 
         if np.nan in pred:
+            print("exclude file:", file)
             continue
         metric.update(pred, label)
 
@@ -96,4 +99,4 @@ def get_metric(result_path, label_path, metric):
 
 if __name__ == "__main__":
     naml_metric = NAMLMetric()
-    get_metric(args.result_path, args.label_path, naml_metric)
+    get_metric(config.result_path, config.label_path, naml_metric)

official/recommend/naml/preprocess.py

+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""Preprocess NAML."""
+import os
+
+from model_utils.config import config
+from src.dataset import MINDPreprocess
+from src.dataset import create_eval_dataset, EvalNews, EvalUsers, EvalCandidateNews
+
+def export_bin():
+    '''pre process function.'''
+    config.phase = "eval"
+    config.neg_sample = config.eval_neg_sample
+    config.embedding_file = os.path.join(config.dataset_path, config.embedding_file)
+    config.word_dict_path = os.path.join(config.dataset_path, config.word_dict_path)
+    config.category_dict_path = os.path.join(config.dataset_path, config.category_dict_path)
+    config.subcategory_dict_path = os.path.join(config.dataset_path, config.subcategory_dict_path)
+    config.uid2index_path = os.path.join(config.dataset_path, config.uid2index_path)
+    config.train_dataset_path = os.path.join(config.dataset_path, config.train_dataset_path)
+    config.eval_dataset_path = os.path.join(config.dataset_path, config.eval_dataset_path)
+    args = config
+    mindpreprocess = MINDPreprocess(vars(args), dataset_path=args.eval_dataset_path)
+    base_path = args.preprocess_path
+
+    data_dir = base_path + '/news_test_data/'
+    news_id_folder = base_path + "/news_id_data/"
+    if not os.path.exists(news_id_folder):
+        os.makedirs(news_id_folder)
+    print("======== news_id_folder is ", news_id_folder, flush=True)
+
+    category_folder = os.path.join(data_dir, "00_category_data")
+    if not os.path.exists(category_folder):
+        os.makedirs(category_folder)
+    print("======== category_folder is ", category_folder, flush=True)
+
+    subcategory_folder = os.path.join(data_dir, "01_subcategory_data")
+    if not os.path.exists(subcategory_folder):
+        os.makedirs(subcategory_folder)
+
+    title_folder = os.path.join(data_dir, "02_title_data")
+    if not os.path.exists(title_folder):
+        os.makedirs(title_folder)
+
+    abstract_folder = os.path.join(data_dir, "03_abstract_data")
+    if not os.path.exists(abstract_folder):
+        os.makedirs(abstract_folder)
+    dataset = create_eval_dataset(mindpreprocess, EvalNews, batch_size=args.batch_size)
+    iterator = dataset.create_dict_iterator(output_numpy=True)
+    idx = 0
+    for idx, data in enumerate(iterator):
+        news_id = data["news_id"]
+        category = data["category"]
+        subcategory = data["subcategory"]
+        title = data["title"]
+        abstract = data["abstract"]
+        file_name = "naml_news_" + str(idx) + ".bin"
+        news_id_file_path = os.path.join(news_id_folder, file_name)
+        news_id.tofile(news_id_file_path)
+        category_file_path = os.path.join(category_folder, file_name)
+        category.tofile(category_file_path)
+        subcategory_file_path = os.path.join(subcategory_folder, file_name)
+        subcategory.tofile(subcategory_file_path)
+        title_file_path = os.path.join(title_folder, file_name)
+        title.tofile(title_file_path)
+        abstract_file_path = os.path.join(abstract_folder, file_name)
+        abstract.tofile(abstract_file_path)
+
+    data_dir = base_path + '/users_test_data/'
+    user_id_folder = os.path.join(data_dir, "00_user_id_data")
+    print("======== user_id_folder is ", user_id_folder)
+    if not os.path.exists(user_id_folder):
+        os.makedirs(user_id_folder)
+
+    history_folder = os.path.join(data_dir, "01_history_data")
+    if not os.path.exists(history_folder):
+        os.makedirs(history_folder)
+
+    dataset = create_eval_dataset(mindpreprocess, EvalUsers, batch_size=args.batch_size)
+    iterator = dataset.create_dict_iterator(output_numpy=True)
+
+    for idx, data in enumerate(iterator):
+        user_id = data["uid"]
+        history = data["history"]
+        file_name = "naml_users_" + str(idx) + ".bin"
+        user_id_file_path = os.path.join(user_id_folder, file_name)
+        user_id.tofile(user_id_file_path)
+        history_file_path = os.path.join(history_folder, file_name)
+        history.tofile(history_file_path)
+    data_dir = base_path + '/browsed_news_test_data/'
+    user_id_folder = os.path.join(data_dir, "00_user_id_data")
+    if not os.path.exists(user_id_folder):
+        os.makedirs(user_id_folder)
+
+    candidate_nid_folder = os.path.join(data_dir, "01_candidate_nid_data")
+    if not os.path.exists(candidate_nid_folder):
+        os.makedirs(candidate_nid_folder)
+
+    labels_folder = os.path.join(data_dir, "02_labels_data")
+    print("======== labels_folder is ", labels_folder)
+    if not os.path.exists(labels_folder):
+        os.makedirs(labels_folder)
+    dataset = create_eval_dataset(mindpreprocess, EvalCandidateNews, batch_size=args.batch_size)
+    iterator = dataset.create_dict_iterator(output_numpy=True)
+    for idx, data in enumerate(iterator):
+    # 'uid', 'candidate_nid', 'labels'
+        uid = data["uid"]
+        candidate_nid = data["candidate_nid"]
+        labels = data["labels"]
+        file_name = "naml_browsed_news_" + str(idx) + ".bin"
+        user_id_file_path = os.path.join(user_id_folder, file_name)
+        uid.tofile(user_id_file_path)
+        candidate_nid_file_path = os.path.join(candidate_nid_folder, file_name)
+        candidate_nid.tofile(candidate_nid_file_path)
+        labels_file_path = os.path.join(labels_folder, file_name)
+        labels.tofile(labels_file_path)
+
+if __name__ == "__main__":
+    export_bin()

official/recommend/naml/script/run_infer_310.sh

@@ -14,8 +14,8 @@
 # limitations under the License.
 # ============================================================================
 
-if [[ $# -lt 8 || $# -gt 9 ]]; then 
-    echo "Usage: sh run_infer_310.sh [NEWS_MODEL] [USER_MODEL] [NEWS_DATASET_PATH] [USER_DATASET_PATH] [NEWS_ID_PATH] [USER_ID_PATH] [BROWSED_NEWS_PATH] [SOC_VERSION] [DEVICE_ID]
+if [[ $# -lt 2 || $# -gt 3 ]]; then
+    echo "Usage: bash run_infer_310.sh [NEWS_MODEL] [USER_MODEL] [DEVICE_ID]
     DEVICE_ID is optional, it can be set by environment variable device_id, otherwise the value is zero"
 exit 1
 fi
@@ -30,16 +30,10 @@ get_real_path(){
 
 news_model=$(get_real_path $1)
 user_model=$(get_real_path $2)
-news_dataset_path=$(get_real_path $3)
-user_dataset_path=$(get_real_path $4)
-news_id_path=$(get_real_path $5)
-user_id_path=$(get_real_path $6)
-browsed_news_path=$(get_real_path $7)
-soc_version=$8
 
-if [ $# == 9 ]; then
-    device_id=$9
-elif [ $# == 8 ]; then
+if [ $# == 3 ]; then
+    device_id=$3
+elif [ $# == 2 ]; then
     if [ -z $device_id ]; then
         device_id=0
     else
@@ -49,11 +43,6 @@ fi
 
 echo $news_model
 echo $user_model
-echo $news_dataset_path
-echo $news_id_path
-echo $user_id_path
-echo $browsed_news_path
-echo $soc_version
 echo $device_id
 
 export ASCEND_HOME=/usr/local/Ascend/
@@ -70,10 +59,17 @@ else
     export ASCEND_OPP_PATH=$ASCEND_HOME/opp
 fi
 
-function air_to_om()
+function preprocess_data()
 {
-    atc --framework=1 --model=$news_model --output=news_encoder --soc_version=$soc_version
-    atc --framework=1 --model=$user_model --output=user_encoder --soc_version=$soc_version
+    if [ -d preprocess_Result ]; then
+        rm -rf ./preprocess_Result
+    fi
+    mkdir preprocess_Result
+    python3.7 ../preprocess.py --preprocess_path=./preprocess_Result &> preprocess.log
+    if [ $? -ne 0 ]; then
+        echo "preprocess data failed"
+        exit 1
+    fi
 }
 
 function compile_app()
@@ -101,7 +97,7 @@ function infer()
     fi
     mkdir result_Files
     mkdir time_Result
-    ../ascend310_infer/out/main --news_om_path news_encoder.om  --user_om_path user_encoder.om --news_dataset_path $news_dataset_path --user_dataset_path $user_dataset_path --newsid_data_path $news_id_path --userid_data_path $user_id_path --browsed_news_path $browsed_news_path &> infer.log
+    ../ascend310_infer/out/main --news_mindir=$news_model --user_mindir=$user_model --device_id=$device_id --base_path=./preprocess_Result &> infer.log
     if [ $? -ne 0 ]; then
         echo "execute inference failed"
         exit 1
@@ -110,14 +106,14 @@ function infer()
 
 function cal_acc()
 {
-    python ../postprocess.py --result_path=./result_Files --label_path=$browsed_news_path/02_labels_data  &> acc.log
+    python ../postprocess.py --result_path=./result_Files --label_path=./preprocess_Result/browsed_news_test_data/02_labels_data  &> acc.log
     if [ $? -ne 0 ]; then
         echo "calculate accuracy failed"
         exit 1
     fi
 }
 
-air_to_om
+preprocess_data
 compile_app
 infer
 cal_acc