Merge branch 'develop' into all_data

Author: luotao1

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

paddle/fluid/inference/analysis/CMakeLists.txt

@@ -6,6 +6,7 @@ cc_library(analysis SRCS pass_manager.cc node.cc data_flow_graph.cc graph_traits
   analyzer.cc
   helper.cc
   # passes
+  analysis_pass.cc
   fluid_to_data_flow_graph_pass.cc
   data_flow_graph_to_fluid_pass.cc
   dfg_graphviz_draw_pass.cc
@@ -99,12 +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/pass.cc renamed to paddle/fluid/inference/analysis/analysis_pass.cc

@@ -12,4 +12,4 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/fluid/inference/analysis/pass.h"
+#include "paddle/fluid/inference/analysis/analysis_pass.h"

paddle/fluid/inference/analysis/pass.h renamed to paddle/fluid/inference/analysis/analysis_pass.h

@@ -28,10 +28,10 @@ namespace paddle {
 namespace inference {
 namespace analysis {
 
-class Pass {
+class AnalysisPass {
  public:
-  Pass() = default;
-  virtual ~Pass() = default;
+  AnalysisPass() = default;
+  virtual ~AnalysisPass() = default;
   // Mutable Pass.
   virtual bool Initialize(Argument *argument) { return false; }
   // Readonly Pass.
@@ -42,53 +42,25 @@ class Pass {
   virtual bool Finalize() { return false; }
 
   // Get a Pass appropriate to print the Node this pass operates on.
-  virtual Pass *CreatePrinterPass(std::ostream &os,
-                                  const std::string &banner) const {
+  virtual AnalysisPass *CreatePrinterPass(std::ostream &os,
+                                          const std::string &banner) const {
     return nullptr;
   }
 
   // Create a debugger Pass that draw the DFG by graphviz toolkit.
-  virtual Pass *CreateGraphvizDebugerPass() const { return nullptr; }
+  virtual AnalysisPass *CreateGraphvizDebugerPass() const { return nullptr; }
 
-  virtual void Run() { LOG(FATAL) << "not valid"; }
-  // Run on a single Node.
-  virtual void Run(Node *x) { LOG(FATAL) << "not valid"; }
-  // Run on a single Function.
-  virtual void Run(Function *x) { LOG(FATAL) << "not valid"; }
-  // Run on a single FunctionBlock.
-  virtual void Run(FunctionBlock *x) { LOG(FATAL) << "not valid"; }
   // Run on a single DataFlowGraph.
-  virtual void Run(DataFlowGraph *x) { LOG(FATAL) << "not valid"; }
+  virtual void Run(DataFlowGraph *x) = 0;
 
   // Human-readable short representation.
   virtual std::string repr() const = 0;
   // Human-readable long description.
   virtual std::string description() const { return "No DOC"; }
 };
 
-// NodePass process on any Node types.
-class NodePass : public Pass {
- public:
-  virtual void Run(Node *node) = 0;
-};
-
-// NodePass process on any Function node types.
-class FunctionPass : public Pass {
- public:
-  virtual void Run(Function *node) = 0;
-};
-
-// NodePass process on any FunctionBlock node types.
-class FunctionBlockPass : public Pass {
- public:
-  virtual void Run(FunctionBlock *node) = 0;
-};
-
 // GraphPass processes on any GraphType.
-class DataFlowGraphPass : public Pass {
- public:
-  virtual void Run(DataFlowGraph *graph) = 0;
-};
+class DataFlowGraphPass : public AnalysisPass {};
 
 }  // namespace analysis
 }  // namespace inference

paddle/fluid/inference/analysis/analyzer.cc

@@ -15,6 +15,7 @@
 #include "paddle/fluid/inference/analysis/analyzer.h"
 #include <string>
 #include <vector>
+
 #include "paddle/fluid/inference/analysis/data_flow_graph_to_fluid_pass.h"
 #include "paddle/fluid/inference/analysis/dfg_graphviz_draw_pass.h"
 #include "paddle/fluid/inference/analysis/fluid_to_data_flow_graph_pass.h"
@@ -58,7 +59,7 @@ class DfgPassManagerImpl final : public DfgPassManager {
   std::string description() const override { return "DFG pass manager."; }
 
  private:
-  void AddPass(const std::string& name, Pass* pass) {
+  void AddPass(const std::string& name, AnalysisPass* pass) {
     VLOG(3) << "Adding pass " << name;
     Register(name, pass);
     AddGraphvizDebugerPass(pass);
@@ -87,7 +88,7 @@ class DfgPassManagerImpl final : public DfgPassManager {
   }
 
   // Add the graphviz debuger pass if the parent pass has one.
-  void AddGraphvizDebugerPass(Pass* pass) {
+  void AddGraphvizDebugerPass(AnalysisPass* pass) {
     auto* debuger_pass = pass->CreateGraphvizDebugerPass();
     if (debuger_pass) {
       Register(debuger_pass->repr(), debuger_pass);

paddle/fluid/inference/analysis/analyzer.h

@@ -36,8 +36,11 @@ limitations under the License. */
  */
 
 #include <gflags/gflags.h>
+#include <string>
+#include <vector>
+
+#include "paddle/fluid/inference/analysis/analysis_pass.h"
 #include "paddle/fluid/inference/analysis/flags.h"
-#include "paddle/fluid/inference/analysis/pass.h"
 #include "paddle/fluid/inference/analysis/pass_manager.h"
 
 namespace paddle {

paddle/fluid/inference/analysis/analyzer_text_classification_tester.cc

@@ -16,6 +16,7 @@
 #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"
@@ -27,43 +28,70 @@ 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 {
 
-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}});
+struct DataReader {
+  explicit DataReader(const std::string &path)
+      : file(new std::ifstream(path)) {}
+
+  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;
+
+    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(batch_size, FLAGS_repeat, 1, 0, sum / FLAGS_repeat);