Merge develop

Author: sneaxiy

CMakeLists.txt

@@ -69,6 +69,7 @@ option(WITH_ANAKIN      "Compile with Anakin library"                   OFF)
 option(WITH_GRPC     "Use grpc as the default rpc framework"            ${WITH_DISTRIBUTE})
 option(WITH_BRPC_RDMA     "Use brpc rdma as the rpc protocal"           OFF)
 option(WITH_INFERENCE    "Compile fluid inference library"              ON)
+option(WITH_INFERENCE_API_TEST   "Test fluid inference high-level api interface"  OFF)
 option(WITH_SYSTEM_BLAS   "Use system blas library"           OFF)
 option(PY_VERSION       "Compile PaddlePaddle with python3 support"     ${PY_VERSION})
 

cmake/tensorrt.cmake

@@ -18,7 +18,7 @@ find_library(TENSORRT_LIBRARY NAMES libnvinfer.so libnvinfer.a
 if(TENSORRT_INCLUDE_DIR AND TENSORRT_LIBRARY)
   if(WITH_DSO)
     set(TENSORRT_FOUND ON)
-  endif(WITH DSO)
+  endif(WITH_DSO)
 else()
     set(TENSORRT_FOUND OFF)
 endif()

doc/fluid/dev/versioning_en.md

@@ -0,0 +1,66 @@
+# Versioning (Work In Progress)
+
+
+PaddlePaddle framework follows Semantic Versioning 2.0 (semver).
+Each release has version of the following format: MAJOR.MINOR.PATCH
+(e.g. 1.2.0). Some key points:
+
+
+ * Major version number change can result in backward-incompatible changes. Codes working in old version don’t necessarily work in the new version. In addition, data, such as program model and checkpointed parameters, generated by the previous major version might not work in the new version. Tools will be attempted to be built to help the release migration.
+
+ * Minor version number change always maintain backward compatibility. It normally contains compatible improvements and bug fixes.
+
+ * Patch number change is for bug fixes.
+
+ * Violation of the policy are considered as bugs and should be fixed.
+
+### What is Covered
+
+* All public documented Python APIs, excluding those live in the contrib namespace.
+
+### What is Not Covered
+
+* If an API’s implementation has bugs, we reserve the rights to fix the bugs and change the behavior.
+
+* The Python APIs in contrib namespace.
+
+* The Python function and classes that start with ‘_’.
+
+* The offline tools.
+
+* The data generated by the framework, such as serialized Program model file and checkpointed variables, are subject to different versioning scheme described below.
+
+* C++ Inference APIs. (To be covered)
+
+
+## Data
+
+
+Data refers to the artifacts generated by the framework. Here, we specifically mean model Program file and the checkpointed variables.
+
+
+
+* Backward Compatibility: User sometimes generates Data at PaddlePaddle version 1.1 and expects it to be consumed by PaddlePaddle version 1.2.
+  This can happen when an new online system wants to serve an old model trained previously.
+
+
+
+* Forward Compatibility: User sometimes generates Data at PaddlePaddle version 1.2 and expects it to be consumed by PaddlePaddle version 1.1.
+  The can happen when an new successful research model want to be served by an old online system that is not frequently upgraded.
+
+
+
+### Versioning
+
+Data version. Data is assigned an integer version number. Version is increased when incompatible change is introduced.
+
+PaddlePaddle framework has an interval of Data version that it supports. PadlePaddle framework within the same major version (semver) cannot drop support of lower version of Data. Hence, a minor version change cannot drop support of Data version.
+
+
+For example, For PaddlePaddle version 1.1, it supports Program version 3 to 5. Later, Program version is increased from 5 to 6 due to addition of an attribute. As a result PaddlePaddle version 1.1 won’t be able to consume it. PaddlePaddle 1.2 should support Program version 3 to 6. PaddlePaddle can only drop support for Program version 3 until PaddlePaddle version 2.0.
+
+
+
+### Known Issues
+
+Currently, forward compatibility for new Data version is best-effort.

paddle/fluid/API.spec

@@ -116,6 +116,7 @@ paddle.fluid.layers.beam_search_decode ArgSpec(args=['ids', 'scores', 'beam_size
 paddle.fluid.layers.conv2d_transpose ArgSpec(args=['input', 'num_filters', 'output_size', 'filter_size', 'padding', 'stride', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, None, 0, 1, 1, None, None, None, True, None, None))
 paddle.fluid.layers.conv3d_transpose ArgSpec(args=['input', 'num_filters', 'output_size', 'filter_size', 'padding', 'stride', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name'], varargs=None, keywords=None, defaults=(None, None, 0, 1, 1, None, None, None, True, None, None))
 paddle.fluid.layers.sequence_expand ArgSpec(args=['x', 'y', 'ref_level', 'name'], varargs=None, keywords=None, defaults=(-1, None))
+paddle.fluid.layers.sequence_expand_as ArgSpec(args=['x', 'y', 'name'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.sequence_pad ArgSpec(args=['x', 'pad_value', 'maxlen'], varargs=None, keywords=None, defaults=(None,))
 paddle.fluid.layers.lstm_unit ArgSpec(args=['x_t', 'hidden_t_prev', 'cell_t_prev', 'forget_bias', 'param_attr', 'bias_attr', 'name'], varargs=None, keywords=None, defaults=(0.0, None, None, None))
 paddle.fluid.layers.reduce_sum ArgSpec(args=['input', 'dim', 'keep_dim', 'name'], varargs=None, keywords=None, defaults=(None, False, None))

paddle/fluid/framework/details/multi_devices_graph_pass.cc

@@ -127,6 +127,9 @@ static const char kLocalScopes[] = "local_scopes";
 static const char kStrategy[] = "strategy";
 
 void MultiDevSSAGraphBuilder::Init() const {
+  all_vars_.clear();
+  balance_vars_.clear();
+
   loss_var_name_ = Get<const std::string>(kLossVarName);
   places_ = Get<const std::vector<platform::Place>>(kPlaces);
   local_scopes_ = Get<const std::vector<Scope *>>(kLocalScopes);

paddle/fluid/framework/details/multi_devices_graph_pass.h

@@ -40,12 +40,6 @@ class MultiDevSSAGraphBuilder : public ir::Pass {
                          size_t device_id) const;
   void Init() const;
 
- private:
-  mutable std::string loss_var_name_;
-  mutable std::vector<platform::Place> places_;
-  mutable std::vector<Scope *> local_scopes_;
-  mutable std::unordered_set<std::string> grad_names_;
-
 #ifdef PADDLE_WITH_CUDA
   mutable platform::NCCLContextMap *nccl_ctxs_;
 #endif
@@ -95,13 +89,17 @@ class MultiDevSSAGraphBuilder : public ir::Pass {
   size_t GetAppropriateDeviceID(
       const std::vector<std::string> &var_names) const;
 
- private:
+  void SetCommunicationContext(OpHandleBase *op_handle,
+                               const platform::Place &p) const;
+
+  mutable std::string loss_var_name_;
+  mutable std::vector<platform::Place> places_;
+  mutable std::vector<Scope *> local_scopes_;
+  mutable std::unordered_set<std::string> grad_names_;
+
   mutable BuildStrategy strategy_;
   mutable std::unordered_map<std::string, VarDesc *> all_vars_;
   mutable std::vector<int64_t> balance_vars_;
-
-  void SetCommunicationContext(OpHandleBase *op_handle,
-                               const platform::Place &p) const;
 };
 }  // namespace details
 }  // namespace framework

paddle/fluid/framework/parallel_executor.cc

@@ -233,30 +233,9 @@ ParallelExecutor::ParallelExecutor(
 
 void ParallelExecutor::BCastParamsToDevices(
     const std::unordered_set<std::string> &vars) const {
-  // the initializing bcast, all vars would be bcast from device(0),
-  // otherwise
-  // bcast from the specified device.
-  bool initializing = member_->executor_ ? false : true;
+  // the initializing bcast, all vars would be bcast from device(0).
   for (auto &var : vars) {
-    int var_dev_id = -1;
-    if (member_->executor_) {
-      auto &sharded_var_device =
-          member_->executor_->Graph().Get<details::ShardedVarDevice>(
-              details::kShardedVarDevice);
-      if (sharded_var_device.find(var) != sharded_var_device.end()) {
-        var_dev_id = sharded_var_device.at(var);
-      }
-    }
-
-    if (!initializing && var_dev_id == -1) continue;
-
-    framework::Variable *main_var = nullptr;
-    if (initializing) {
-      main_var = member_->local_scopes_[0]->FindVar(var);
-    } else {
-      main_var = member_->local_scopes_[var_dev_id]->FindVar(var);
-    }
-
+    framework::Variable *main_var = member_->local_scopes_[0]->FindVar(var);
     if (main_var == nullptr || !main_var->IsType<LoDTensor>()) {
       continue;
     }
@@ -272,8 +251,7 @@ void ParallelExecutor::BCastParamsToDevices(
         auto place = member_->places_[i];
         void *buffer;
 
-        if ((initializing && i == 0) ||
-            (!initializing && static_cast<int>(i) == var_dev_id)) {
+        if (i == 0) {
           buffer = const_cast<void *>(main_tensor.data<void>());
         } else {
           auto local_scope = member_->local_scopes_[i];
@@ -290,29 +268,18 @@ void ParallelExecutor::BCastParamsToDevices(
         platform::NCCLGroupGuard guard;
         for (size_t i = 0; i < member_->places_.size(); ++i) {
           auto &nccl_ctx = member_->nccl_ctxs_->at(member_->places_[i]);
-          if (initializing) {
-            platform::dynload::ncclBcast(buffers[i], numel, data_type, 0,
-                                         nccl_ctx.comm_, nccl_ctx.stream());
-          } else {
-            if (var_dev_id >= 0) {
-              platform::dynload::ncclBcast(buffers[i], numel, data_type,
-                                           var_dev_id, nccl_ctx.comm_,
-                                           nccl_ctx.stream());
-            }
-          }
+          platform::dynload::ncclBcast(buffers[i], numel, data_type, 0,
+                                       nccl_ctx.comm_, nccl_ctx.stream());
         }
         member_->nccl_ctxs_->WaitAll();
       }
-
 #else
       PADDLE_THROW("Not compiled with CUDA");
 #endif
     } else {
       platform::CPUPlace cpu;
       for (size_t i = 0; i < member_->places_.size(); ++i) {
-        if ((initializing && i == 0) ||
-            (!initializing && static_cast<int>(i) == var_dev_id))
-          continue;
+        if (i == 0) continue;
 
         auto local_scope = member_->local_scopes_[i];
         auto *t = local_scope->Var(var)->GetMutable<LoDTensor>();

paddle/fluid/framework/parallel_executor.h

@@ -72,9 +72,9 @@ class ParallelExecutor {
   void Run(const std::vector<std::string> &fetch_tensors,
            const std::string &fetched_var_name);
 
+ private:
   void BCastParamsToDevices(const std::unordered_set<std::string> &vars) const;
 
- private:
   ParallelExecutorPrivate *member_;
 
 #ifdef PADDLE_WITH_CUDA

paddle/fluid/inference/CMakeLists.txt

@@ -17,9 +17,7 @@ get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
 # paddle_fluid_origin exclude inference api interface
 cc_library(paddle_fluid_origin DEPS ${fluid_modules} paddle_fluid_api)
 
-#if(APPLE)
-  add_subdirectory(api)
-#endif()
+add_subdirectory(api)
 
 # Create static library
 cc_library(paddle_fluid DEPS ${fluid_modules} paddle_fluid_api paddle_inference_api analysis_predictor)
@@ -57,5 +55,7 @@ endif()
 if(WITH_TESTING)
   # tests/book depends the models that generated by python/paddle/fluid/tests/book
   add_subdirectory(tests/book)
-  add_subdirectory(tests/api)
+  if(WITH_INFERENCE_API_TEST)
+    add_subdirectory(tests/api)
+  endif()  
 endif()

paddle/fluid/inference/api/CMakeLists.txt

@@ -69,25 +69,4 @@ if (WITH_ANAKIN AND WITH_MKL) # only needed in CI
     endfunction()
     anakin_target(inference_anakin_api)
     anakin_target(inference_anakin_api_shared)
-    if (WITH_TESTING)
-        # TODO(luotao): ANAKIN_MODLE_URL etc will move to demo ci later.
-        set(INFERENCE_URL "http://paddle-inference-dist.bj.bcebos.com")
-        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_RNN_MODLE_URL} -N")
-        execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_DATA_URL} -N")
-        if(WITH_GPU)
-            set(anakin_test_extra_deps dynload_cuda)
-            set(ANAKIN_MODLE_URL "${INFERENCE_URL}/mobilenet_v2.anakin.bin")
-            execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_MODLE_URL} -N")
-            cc_test(api_anakin_engine_tester SRCS api_anakin_engine_tester.cc 
-                    ARGS --model=${ANAKIN_SOURCE_DIR}/mobilenet_v2.anakin.bin
-                    DEPS inference_anakin_api_shared ${anakin_test_extra_deps} SERIAL)
-        endif()
-        cc_test(api_anakin_engine_rnn_tester SRCS api_anakin_engine_rnn_tester.cc 
-                ARGS --model=${ANAKIN_SOURCE_DIR}/anakin_test%2Fditu_rnn.anakin2.model.bin 
-                     --datapath=${ANAKIN_SOURCE_DIR}/anakin_test%2Fditu_rnn_data.txt
-                DEPS inference_anakin_api_shared ${anakin_test_extra_deps} SERIAL)
-    endif(WITH_TESTING)
 endif()