Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into fix_python35_CI_random_fail

Author: velconia

CMakeLists.txt

@@ -213,9 +213,11 @@ include(configure)          # add paddle env configuration
 if(WITH_GPU)
     include(cuda)
     include(tensorrt)
+endif()
+if(WITH_MKL OR WITH_MKLML)
     include(external/anakin)
 elseif()
-    set(WITH_ANAKIN OFF CACHE STRING "Anakin is used in GPU only now." FORCE)
+    set(WITH_ANAKIN OFF CACHE STRING "Anakin is used in MKL only now." FORCE)
 endif()
 
 include(generic)            # simplify cmake module

cmake/external/anakin.cmake

@@ -16,16 +16,6 @@ set(ANAKIN_LIBRARY     ${ANAKIN_INSTALL_DIR})
 set(ANAKIN_SHARED_LIB  ${ANAKIN_LIBRARY}/libanakin.so)
 set(ANAKIN_SABER_LIB   ${ANAKIN_LIBRARY}/libanakin_saber_common.so)
 
-# TODO(luotao): ANAKIN_MODLE_URL etc will move to demo ci later.
-set(INFERENCE_URL "http://paddle-inference-dist.bj.bcebos.com")
-set(ANAKIN_MODLE_URL "${INFERENCE_URL}/mobilenet_v2.anakin.bin")
-set(ANAKIN_RNN_MODLE_URL "${INFERENCE_URL}/anakin_test%2Fditu_rnn.anakin2.model.bin")
-set(ANAKIN_RNN_DATA_URL "${INFERENCE_URL}/anakin_test%2Fditu_rnn_data.txt")
-execute_process(COMMAND bash -c "mkdir -p ${ANAKIN_SOURCE_DIR}")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_MODLE_URL} -N")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_MODLE_URL} -N")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_DATA_URL} -N")
-
 include_directories(${ANAKIN_INCLUDE})
 include_directories(${ANAKIN_INCLUDE}/saber/)
 include_directories(${ANAKIN_INCLUDE}/saber/core/)
@@ -48,6 +38,11 @@ set(ANAKIN_COMPILE_EXTRA_FLAGS
     -Wno-reorder
     -Wno-error=cpp)
 
+if(WITH_GPU)
+    set(CMAKE_ARGS_PREFIX -DUSE_GPU_PLACE=YES -DCUDNN_ROOT=${CUDNN_ROOT} -DCUDNN_INCLUDE_DIR=${CUDNN_INCLUDE_DIR})
+else()
+    set(CMAKE_ARGS_PREFIX -DUSE_GPU_PLACE=NO)
+endif()
 ExternalProject_Add(
     extern_anakin
     ${EXTERNAL_PROJECT_LOG_ARGS}
@@ -56,13 +51,11 @@ ExternalProject_Add(
     GIT_TAG             "9424277cf9ae180a14aff09560d3cd60a49c76d2"
     PREFIX              ${ANAKIN_SOURCE_DIR}
     UPDATE_COMMAND      ""
-    CMAKE_ARGS          -DUSE_GPU_PLACE=YES
+    CMAKE_ARGS          ${CMAKE_ARGS_PREFIX}
                         -DUSE_X86_PLACE=YES
                         -DBUILD_WITH_UNIT_TEST=NO
                         -DPROTOBUF_ROOT=${THIRD_PARTY_PATH}/install/protobuf
                         -DMKLML_ROOT=${THIRD_PARTY_PATH}/install/mklml
-                        -DCUDNN_ROOT=${CUDNN_ROOT}
-                        -DCUDNN_INCLUDE_DIR=${CUDNN_INCLUDE_DIR}
                         -DENABLE_OP_TIMER=${ANAKIN_ENABLE_OP_TIMER}
                         ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS    -DCMAKE_INSTALL_PREFIX:PATH=${ANAKIN_INSTALL_DIR}

cmake/inference_lib.cmake

@@ -145,7 +145,7 @@ copy(memory_lib
 set(inference_deps paddle_fluid_shared paddle_fluid)
 
 set(module "inference/api")
-if (WITH_ANAKIN AND WITH_GPU)
+if (WITH_ANAKIN AND WITH_MKL)
     copy(anakin_inference_lib DEPS paddle_inference_api inference_anakin_api
         SRCS
         ${PADDLE_BINARY_DIR}/paddle/fluid/inference/api/libinference_anakin_api* # compiled anakin api

doc/fluid/api/layers.rst

@@ -822,6 +822,14 @@ pad
 ..  autofunction:: paddle.fluid.layers.pad
     :noindex:
 
+.. _api_fluid_layers_pad_constant_like:
+
+pad_constant_like
+---
+
+..  autofunction:: paddle.fluid.layers.pad_constant_like
+    :noindex:
+
 .. _api_fluid_layers_label_smooth:
 
 label_smooth
@@ -1145,6 +1153,14 @@ sigmoid
 ..  autofunction:: paddle.fluid.layers.sigmoid
     :noindex:
 
+.. _api_fluid_layers_hsigmoid:
+
+hsigmoid
+-------
+
+..  autofunction:: paddle.fluid.layers.hsigmoid
+    :noindex:
+
 .. _api_fluid_layers_logsigmoid:
 
 logsigmoid

doc/survey/dynamic_graph.md

@@ -30,7 +30,7 @@ x = Variable(randn(20, 1)))
 label = Variable(randint(1))
 W_1, W_2 = Variable(randn(20, 20)), Variable(randn(10, 20))
 h = matmul(W_1, x)
-pred = matmul(W_2, x)
+pred = matmul(W_2, h)
 loss = softmax(pred, label)
 loss.backward()
 ```

paddle/fluid/framework/ir/CMakeLists.txt

@@ -1,14 +1,21 @@
 set(pass_file ${PADDLE_BINARY_DIR}/paddle/fluid/inference/api/paddle_inference_pass.h)
 file(WRITE ${pass_file} "// Generated by the paddle/fluid/framework/ir/CMakeLists.txt.  DO NOT EDIT!\n\n")
 file(APPEND ${pass_file} "\#include \"paddle/fluid/framework/ir/pass.h\"\n")
-function(pass_library TARGET)
+
+
+# Usage: pass_library(target inference) will append to paddle_inference_pass.h
+function(pass_library TARGET DEST)
     set(options "")
     set(oneValueArgs "")
     set(multiValueArgs SRCS DEPS)
     cmake_parse_arguments(op_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
-    cc_library(${TARGET} SRCS ${TARGET}.cc DEPS graph_pattern_detector pass)
-    file(APPEND ${pass_file} "USE_PASS(${TARGET});\n")
-    set(PASS_LIBRARY ${TARGET} ${PASS_LIBRARY} PARENT_SCOPE)
+    cc_library(${TARGET} SRCS ${TARGET}.cc DEPS graph_pattern_detector pass ${op_library_DEPS})
+    # add more DEST here, such as train, dist and collect USE_PASS into a file automatically.
+    if (${DEST} STREQUAL "base" OR ${DEST} STREQUAL "inference")
+        message(STATUS "add pass ${TARGET} ${DEST}")
+        file(APPEND ${pass_file} "USE_PASS(${TARGET});\n")
+        set(PASS_LIBRARY ${TARGET} ${PASS_LIBRARY} PARENT_SCOPE)
+    endif()
 endfunction()
 
 cc_library(node SRCS node.cc DEPS proto_desc)
@@ -18,13 +25,13 @@ cc_library(pass SRCS pass.cc DEPS graph node graph_helper)
 cc_library(graph_traits SRCS graph_traits.cc DEPS graph)
 cc_library(graph_pattern_detector SRCS graph_pattern_detector.cc DEPS graph graph_helper graph_traits)
 
-pass_library(graph_to_program_pass)
-pass_library(graph_viz_pass)
-pass_library(fc_fuse_pass)
-pass_library(attention_lstm_fuse_pass)
-pass_library(infer_clean_graph_pass)
-pass_library(fc_lstm_fuse_pass)
-pass_library(seq_concat_fc_fuse_pass)
+pass_library(graph_to_program_pass base)
+pass_library(graph_viz_pass base)
+pass_library(fc_fuse_pass inference)
+pass_library(attention_lstm_fuse_pass inference)
+pass_library(infer_clean_graph_pass inference)
+pass_library(fc_lstm_fuse_pass inference)
+pass_library(seq_concat_fc_fuse_pass inference)
 set(GLOB_PASS_LIB ${PASS_LIBRARY} CACHE INTERNAL "Global PASS library")
 
 cc_test(pass_test SRCS pass_test.cc DEPS graph pass graph_helper)

paddle/fluid/inference/analysis/CMakeLists.txt

@@ -100,12 +100,17 @@ inference_analysis_test(test_analyzer_lac SRCS analyzer_lac_tester.cc
 
 
 set(TEXT_CLASSIFICATION_MODEL_URL "http://paddle-inference-dist.bj.bcebos.com/text-classification-Senta.tar.gz")
+set(TEXT_CLASSIFICATION_DATA_URL "http://paddle-inference-dist.bj.bcebos.com/text_classification_data.txt.tar.gz")
 set(TEXT_CLASSIFICATION_INSTALL_DIR "${THIRD_PARTY_PATH}/inference_demo/text_classification" CACHE PATH "Text Classification model and data root." FORCE)
 
 if (NOT EXISTS ${TEXT_CLASSIFICATION_INSTALL_DIR} AND WITH_TESTING AND WITH_INFERENCE)
   inference_download_and_uncompress(${TEXT_CLASSIFICATION_INSTALL_DIR} ${TEXT_CLASSIFICATION_MODEL_URL} "text-classification-Senta.tar.gz")
+  inference_download_and_uncompress(${TEXT_CLASSIFICATION_INSTALL_DIR} ${TEXT_CLASSIFICATION_DATA_URL} "text_classification_data.txt.tar.gz")
 endif()
 
 inference_analysis_test(test_text_classification SRCS analyzer_text_classification_tester.cc
     EXTRA_DEPS paddle_inference_api paddle_fluid_api analysis_predictor
-    ARGS --infer_model=${TEXT_CLASSIFICATION_INSTALL_DIR}/text-classification-Senta)
+    ARGS --infer_model=${TEXT_CLASSIFICATION_INSTALL_DIR}/text-classification-Senta
+         --infer_data=${TEXT_CLASSIFICATION_INSTALL_DIR}/data.txt
+         --topn=1 # Just run top 1 batch.
+    )

paddle/fluid/inference/analysis/analyzer_lac_tester.cc

@@ -114,12 +114,6 @@ void GetOneBatch(std::vector<PaddleTensor> *input_slots, DataRecord *data,
   PADDLE_ENFORCE_EQ(batch_size, static_cast<int>(one_batch.lod.size() - 1));
   input_slots->assign({input_tensor});
 }
-static void PrintTime(const double latency, const int bs, const int repeat) {
-  LOG(INFO) << "===========profile result===========";
-  LOG(INFO) << "batch_size: " << bs << ", repeat: " << repeat
-            << ", avg latency: " << latency / repeat << "ms";
-  LOG(INFO) << "=====================================";
-}
 void BenchAllData(const std::string &model_path, const std::string &data_file,
                   const int batch_size, const int repeat) {
   NativeConfig config;
@@ -145,7 +139,7 @@ void BenchAllData(const std::string &model_path, const std::string &data_file,
       sum += timer.toc();
     }
   }
-  PrintTime(sum, batch_size, repeat);
+  PrintTime(batch_size, repeat, 1, 0, sum / repeat);
 }
 const int64_t lac_ref_data[] = {24, 25, 25, 25, 38, 30, 31, 14, 15, 44, 24, 25,
                                 25, 25, 25, 25, 44, 24, 25, 25, 25, 36, 42, 43,
@@ -176,7 +170,7 @@ void TestLACPrediction(const std::string &model_path,
   for (int i = 0; i < repeat; i++) {
     predictor->Run(input_slots, &outputs_slots);
   }
-  PrintTime(timer.toc(), batch_size, repeat);
+  PrintTime(batch_size, repeat, 1, 0, timer.toc() / repeat);
   EXPECT_EQ(outputs_slots.size(), 1UL);
   auto &out = outputs_slots[0];
   size_t size = std::accumulate(out.shape.begin(), out.shape.end(), 1,

paddle/fluid/inference/analysis/analyzer_ner_tester.cc

@@ -25,6 +25,7 @@ DEFINE_string(infer_model, "", "model path");
 DEFINE_string(infer_data, "", "data path");
 DEFINE_int32(batch_size, 10, "batch size.");
 DEFINE_int32(repeat, 1, "Running the inference program repeat times.");
+DEFINE_bool(test_all_data, false, "Test the all dataset in data file.");
 
 namespace paddle {
 namespace inference {
@@ -35,6 +36,7 @@ struct DataRecord {
   std::vector<size_t> lod;  // two inputs have the same lod info.
   size_t batch_iter{0};
   size_t batch_size{1};
+  size_t num_samples;  // total number of samples
   DataRecord() = default;
   explicit DataRecord(const std::string &path, int batch_size = 1)
       : batch_size(batch_size) {
@@ -81,6 +83,7 @@ struct DataRecord {
       word_data_all.push_back(std::move(word_data));
       mention_data_all.push_back(std::move(mention_data));
     }
+    num_samples = num_lines;
   }
 };
 
@@ -120,21 +123,38 @@ void TestChineseNERPrediction() {
   auto predictor =
       CreatePaddlePredictor<NativeConfig, PaddleEngineKind::kNative>(config);
   std::vector<PaddleTensor> input_slots;
-  DataRecord data(FLAGS_infer_data, FLAGS_batch_size);
+  std::vector<PaddleTensor> outputs;
+  Timer timer;
+
+  if (FLAGS_test_all_data) {
+    LOG(INFO) << "test all data";
+    double sum = 0;
+    size_t num_samples;
+    for (int i = 0; i < FLAGS_repeat; i++) {
+      DataRecord data(FLAGS_infer_data, FLAGS_batch_size);
+      num_samples = data.num_samples;
+      for (size_t bid = 0; bid < num_samples; ++bid) {
+        PrepareInputs(&input_slots, &data, FLAGS_batch_size);
+        timer.tic();
+        predictor->Run(input_slots, &outputs);
+        sum += timer.toc();
+      }
+    }
+    LOG(INFO) << "total number of samples: " << num_samples;
+    PrintTime(FLAGS_batch_size, FLAGS_repeat, 1, 0, sum / FLAGS_repeat);
+    LOG(INFO) << "average latency of each sample: "
+              << sum / FLAGS_repeat / num_samples;
+    return;
+  }
   // Prepare inputs.
+  DataRecord data(FLAGS_infer_data, FLAGS_batch_size);
   PrepareInputs(&input_slots, &data, FLAGS_batch_size);
-  std::vector<PaddleTensor> outputs;
 
-  Timer timer;
   timer.tic();
   for (int i = 0; i < FLAGS_repeat; i++) {
     predictor->Run(input_slots, &outputs);
   }
-  LOG(INFO) << "===========profile result===========";
-  LOG(INFO) << "batch_size: " << FLAGS_batch_size
-            << ", repeat: " << FLAGS_repeat
-            << ", latency: " << timer.toc() / FLAGS_repeat << "ms";
-  LOG(INFO) << "=====================================";
+  PrintTime(FLAGS_batch_size, FLAGS_repeat, 1, 0, timer.toc() / FLAGS_repeat);
 
   PADDLE_ENFORCE(outputs.size(), 1UL);
   auto &out = outputs[0];

paddle/fluid/inference/analysis/analyzer_text_classification_tester.cc

@@ -16,8 +16,10 @@
 #include <gflags/gflags.h>
 #include <glog/logging.h>  // use glog instead of PADDLE_ENFORCE to avoid importing other paddle header files.
 #include <gtest/gtest.h>
+#include <fstream>
 #include "paddle/fluid/framework/ir/pass.h"
 #include "paddle/fluid/inference/analysis/ut_helper.h"
+#include "paddle/fluid/inference/api/helper.h"
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #include "paddle/fluid/inference/api/paddle_inference_pass.h"
 #include "paddle/fluid/inference/api/timer.h"
@@ -26,60 +28,72 @@ DEFINE_string(infer_model, "", "Directory of the inference model.");
 DEFINE_string(infer_data, "", "Path of the dataset.");
 DEFINE_int32(batch_size, 1, "batch size.");
 DEFINE_int32(repeat, 1, "How many times to repeat run.");
+DEFINE_int32(topn, -1, "Run top n batches of data to save time");
 
 namespace paddle {
+namespace inference {
 
-template <typename T>
-std::string to_string(const std::vector<T> &vec) {
-  std::stringstream ss;
-  for (const auto &c : vec) {
-    ss << c << " ";
-  }
-  return ss.str();
-}
+struct DataReader {
+  explicit DataReader(const std::string &path)
+      : file(new std::ifstream(path)) {}
 
-void PrintTime(const double latency, const int bs, const int repeat) {
-  LOG(INFO) << "===========profile result===========";
-  LOG(INFO) << "batch_size: " << bs << ", repeat: " << repeat
-            << ", avg latency: " << latency / repeat << "ms";
-  LOG(INFO) << "=====================================";
-}
+  bool NextBatch(PaddleTensor *tensor, int batch_size) {
+    PADDLE_ENFORCE_EQ(batch_size, 1);
+    std::string line;
+    tensor->lod.clear();
+    tensor->lod.emplace_back(std::vector<size_t>({0}));
+    std::vector<int64_t> data;
 
-void Main(int batch_size) {
-  // Three sequence inputs.
-  std::vector<PaddleTensor> input_slots(1);
-  // one batch starts
-  // data --
-  int64_t data0[] = {0, 1, 2};
-  for (auto &input : input_slots) {
-    input.data.Reset(data0, sizeof(data0));
-    input.shape = std::vector<int>({3, 1});
-    // dtype --
-    input.dtype = PaddleDType::INT64;
-    // LoD --
-    input.lod = std::vector<std::vector<size_t>>({{0, 3}});
+    for (int i = 0; i < batch_size; i++) {
+      if (!std::getline(*file, line)) return false;
+      inference::split_to_int64(line, ' ', &data);
+    }
+    tensor->lod.front().push_back(data.size());
+
+    tensor->data.Resize(data.size() * sizeof(int64_t));
+    memcpy(tensor->data.data(), data.data(), data.size() * sizeof(int64_t));
+    tensor->shape.clear();
+    tensor->shape.push_back(data.size());
+    tensor->shape.push_back(1);
+    return true;
   }
 
+  std::unique_ptr<std::ifstream> file;
+};
+
+void Main(int batch_size) {
   // shape --
   // Create Predictor --
   AnalysisConfig config;
   config.model_dir = FLAGS_infer_model;
   config.use_gpu = false;
   config.enable_ir_optim = true;
-  config.ir_passes.push_back("fc_lstm_fuse_pass");
   auto predictor =
       CreatePaddlePredictor<AnalysisConfig, PaddleEngineKind::kAnalysis>(
           config);
 
+  std::vector<PaddleTensor> input_slots(1);
+  // one batch starts
+  // data --
+  auto &input = input_slots[0];
+  input.dtype = PaddleDType::INT64;
+
   inference::Timer timer;
   double sum = 0;
   std::vector<PaddleTensor> output_slots;
-  for (int i = 0; i < FLAGS_repeat; i++) {
-    timer.tic();
-    CHECK(predictor->Run(input_slots, &output_slots));
-    sum += timer.toc();
+
+  int num_batches = 0;
+  for (int t = 0; t < FLAGS_repeat; t++) {
+    DataReader reader(FLAGS_infer_data);
+    while (reader.NextBatch(&input, FLAGS_batch_size)) {
+      if (FLAGS_topn > 0 && num_batches > FLAGS_topn) break;
+      timer.tic();
+      CHECK(predictor->Run(input_slots, &output_slots));
+      sum += timer.toc();
+      ++num_batches;
+    }
   }
-  PrintTime(sum, batch_size, FLAGS_repeat);
+  PrintTime(batch_size, FLAGS_repeat, 1, 0, sum / FLAGS_repeat);
 
   // Get output
   LOG(INFO) << "get outputs " << output_slots.size();
@@ -100,4 +114,5 @@ void Main(int batch_size) {
 
 TEST(text_classification, basic) { Main(FLAGS_batch_size); }
 
+}  // namespace inference
 }  // namespace paddle