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

CMakeLists.txt

@@ -36,6 +36,7 @@ include(simd)
 
 ################################ Configurations #######################################
 option(WITH_GPU         "Compile PaddlePaddle with NVIDIA GPU"          ${CUDA_FOUND})
+option(WITH_AMD_GPU     "Compile PaddlePaddle with AMD GPU"             OFF)
 option(WITH_AVX         "Compile PaddlePaddle with AVX intrinsics"      ${AVX_FOUND})
 option(WITH_MKL         "Compile PaddlePaddle with MKL support."        ${AVX_FOUND})
 option(WITH_DSO         "Compile PaddlePaddle with dynamic linked CUDA" ON)
@@ -180,6 +181,11 @@ if(WITH_GPU)
     include(cuda)
 endif(WITH_GPU)
 
+if(WITH_AMD_GPU)
+    find_package(HIP)
+    include(hip)
+endif(WITH_AMD_GPU)
+
 if(WITH_MKLML)
     list(APPEND EXTERNAL_LIBS ${MKLML_IOMP_LIB})
 endif()

benchmark/cluster/vgg16/vgg16_fluid.py

@@ -18,12 +18,13 @@ import sys
 import time
 import numpy as np
 import paddle.v2 as paddle
-import paddle.v2.fluid as fluid
-import paddle.v2.fluid.core as core
-import paddle.v2.fluid.profiler as profiler
+import paddle.fluid as fluid
+import paddle.fluid.core as core
+import paddle.fluid.profiler as profiler
 import argparse
 import functools
 import os
+from paddle.fluid import debuger
 
 
 def str2bool(v):
@@ -182,28 +183,27 @@ def main():
             start_time = time.time()
             num_samples = 0
             train_pass_acc.reset()
-            with profiler.profiler("CPU", 'total') as prof:
-                for batch_id, data in enumerate(train_reader()):
-                    ts = time.time()
-                    img_data = np.array(
-                        map(lambda x: x[0].reshape(data_shape), data)).astype(
-                            "float32")
-                    y_data = np.array(map(lambda x: x[1], data)).astype("int64")
-                    y_data = y_data.reshape([-1, 1])
-
-                    loss, acc, b_size = exe.run(
-                        trainer_prog,
-                        feed={"pixel": img_data,
-                              "label": y_data},
-                        fetch_list=[avg_cost, batch_acc, batch_size])
-                    iters += 1
-                    num_samples += len(data)
-                    train_pass_acc.add(value=acc, weight=b_size)
-                    print(
-                        "Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, Speed = %.2f img/s"
-                        % (pass_id, iters, loss, acc,
-                           len(data) / (time.time() - ts))
-                    )  # The accuracy is the accumulation of batches, but not the current batch.
+            for batch_id, data in enumerate(train_reader()):
+                ts = time.time()
+                img_data = np.array(
+                    map(lambda x: x[0].reshape(data_shape), data)).astype(
+                        "float32")
+                y_data = np.array(map(lambda x: x[1], data)).astype("int64")
+                y_data = y_data.reshape([-1, 1])
+
+                loss, acc, b_size = exe.run(
+                    trainer_prog,
+                    feed={"pixel": img_data,
+                          "label": y_data},
+                    fetch_list=[avg_cost, batch_acc, batch_size])
+                iters += 1
+                num_samples += len(data)
+                train_pass_acc.add(value=acc, weight=b_size)
+                print(
+                    "Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, Speed = %.2f img/s"
+                    % (pass_id, iters, loss, acc,
+                       len(data) / (time.time() - ts))
+                )  # The accuracy is the accumulation of batches, but not the current batch.
 
             pass_elapsed = time.time() - start_time
             pass_train_acc = train_pass_acc.eval()
@@ -254,9 +254,7 @@ def main():
             pserver_prog = t.get_pserver_program(current_endpoint)
             pserver_startup = t.get_startup_program(current_endpoint,
                                                     pserver_prog)
-            print("starting server side startup")
             exe.run(pserver_startup)
-            print("starting parameter server...")
             exe.run(pserver_prog)
         elif training_role == "TRAINER":
             # Parameter initialization

benchmark/cluster/vgg16/vgg16_tf.py

@@ -292,14 +292,18 @@ def run_benchmark(cluster_spec, server):
         return np.mean(test_accs)
 
     config = tf.ConfigProto(
-        intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
+        intra_op_parallelism_threads=1,
+        inter_op_parallelism_threads=1,
+        log_device_placement=True)
     config.gpu_options.allow_growth = True
 
     hooks = [tf.train.StopAtStepHook(last_step=1000000)]
 
     with tf.train.MonitoredTrainingSession(
-            master=server.target, is_chief=(args.task_index == 0),
-            hooks=hooks) as sess:
+            master=server.target,
+            is_chief=(args.task_index == 0),
+            hooks=hooks,
+            config=config) as sess:
         iters, num_samples, start_time = 0, 0, 0.0
         for pass_id in range(args.num_passes):
             # train

cmake/configure.cmake

@@ -57,11 +57,7 @@ if(NOT WITH_GOLANG)
     add_definitions(-DPADDLE_WITHOUT_GOLANG)
 endif(NOT WITH_GOLANG)
 
-if(NOT WITH_GPU)
-    add_definitions(-DHPPL_STUB_FUNC)
-
-    list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
-else()
+if(WITH_GPU)
     add_definitions(-DPADDLE_WITH_CUDA)
 
     FIND_PACKAGE(CUDA REQUIRED)
@@ -84,7 +80,14 @@ else()
     # Include cuda and cudnn
     include_directories(${CUDNN_INCLUDE_DIR})
     include_directories(${CUDA_TOOLKIT_INCLUDE})
-endif(NOT WITH_GPU)
+elseif(WITH_AMD_GPU)
+    add_definitions(-DPADDLE_WITH_HIP)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D__HIP_PLATFORM_HCC__")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D__HIP_PLATFORM_HCC__")
+else()
+    add_definitions(-DHPPL_STUB_FUNC)
+    list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
+endif()
 
 if (WITH_MKLML AND MKLML_IOMP_LIB)
     message(STATUS "Enable Intel OpenMP with ${MKLML_IOMP_LIB}")

cmake/external/boost.cmake

@@ -24,7 +24,7 @@ set(BOOST_PROJECT       "extern_boost")
 # So we use 1.41.0 here.
 set(BOOST_VER           "1.41.0")
 set(BOOST_TAR           "boost_1_41_0")
-set(BOOST_URL           "http://paddlepaddledeps.s3-website-us-west-1.amazonaws.com/${BOOST_TAR}.tar.gz")
+set(BOOST_URL           "http://paddlepaddledeps.bj.bcebos.com/${BOOST_TAR}.tar.gz")
 set(BOOST_SOURCES_DIR ${THIRD_PARTY_PATH}/boost)
 set(BOOST_DOWNLOAD_DIR  "${BOOST_SOURCES_DIR}/src/${BOOST_PROJECT}")
 set(BOOST_INCLUDE_DIR "${BOOST_DOWNLOAD_DIR}/${BOOST_TAR}" CACHE PATH "boost include directory." FORCE)

cmake/external/eigen.cmake

@@ -4,18 +4,33 @@ SET(EIGEN_SOURCE_DIR ${THIRD_PARTY_PATH}/eigen3)
 SET(EIGEN_INCLUDE_DIR ${EIGEN_SOURCE_DIR}/src/extern_eigen3)
 INCLUDE_DIRECTORIES(${EIGEN_INCLUDE_DIR})
 
-ExternalProject_Add(
-    extern_eigen3
-    ${EXTERNAL_PROJECT_LOG_ARGS}
-    GIT_REPOSITORY  "https://github.com/RLovelett/eigen.git"
-    GIT_TAG         70661066beef694cadf6c304d0d07e0758825c10
-    PREFIX          ${EIGEN_SOURCE_DIR}
-    UPDATE_COMMAND  ""
-    CONFIGURE_COMMAND ""
-    BUILD_COMMAND     ""
-    INSTALL_COMMAND   ""
-    TEST_COMMAND      ""
-)
+if(WITH_AMD_GPU)
+    ExternalProject_Add(
+        extern_eigen3
+        ${EXTERNAL_PROJECT_LOG_ARGS}
+        GIT_REPOSITORY  "https://github.com/sabreshao/hipeigen.git"
+        GIT_TAG         0cba03ff9f8f9f70bbd92ac5857b031aa8fed6f9
+        PREFIX          ${EIGEN_SOURCE_DIR}
+        UPDATE_COMMAND  ""
+        CONFIGURE_COMMAND ""
+        BUILD_COMMAND     ""
+        INSTALL_COMMAND   ""
+        TEST_COMMAND      ""
+    )
+else()
+    ExternalProject_Add(
+        extern_eigen3
+        ${EXTERNAL_PROJECT_LOG_ARGS}
+        GIT_REPOSITORY  "https://github.com/RLovelett/eigen.git"
+        GIT_TAG         70661066beef694cadf6c304d0d07e0758825c10
+        PREFIX          ${EIGEN_SOURCE_DIR}
+        UPDATE_COMMAND  ""
+        CONFIGURE_COMMAND ""
+        BUILD_COMMAND     ""
+        INSTALL_COMMAND   ""
+        TEST_COMMAND      ""
+    )
+endif()
 
 if (${CMAKE_VERSION} VERSION_LESS "3.3.0")
     set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/eigen3_dummy.c)

cmake/generic.cmake

@@ -317,6 +317,82 @@ function(nv_test TARGET_NAME)
   endif()
 endfunction(nv_test)
 
+function(hip_library TARGET_NAME)
+  if (WITH_AMD_GPU)
+    set(options STATIC static SHARED shared)
+    set(oneValueArgs "")
+    set(multiValueArgs SRCS DEPS)
+    cmake_parse_arguments(hip_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+    set(_sources ${hip_library_SRCS})
+    HIP_PREPARE_TARGET_COMMANDS(${TARGET_NAME} OBJ _generated_files _source_files ${_sources} HIPCC_OPTIONS ${_hipcc_options} HCC_OPTIONS ${_hcc_options} NVCC_OPTIONS ${_nvcc_options})
+    if(_source_files)
+      list(REMOVE_ITEM _sources ${_source_files})
+    endif()
+    if(hip_library_SRCS)
+      if (hip_library_SHARED OR hip_library_shared) # build *.so
+        add_library(${TARGET_NAME} SHARED ${_cmake_options} ${_generated_files} ${_sources})
+        set_target_properties(${TARGET_NAME} PROPERTIES LINKER_LANGUAGE HIP)
+      else()
+        add_library(${TARGET_NAME} STATIC ${_cmake_options} ${_generated_files} ${_sources})
+        set_target_properties(${TARGET_NAME} PROPERTIES LINKER_LANGUAGE CXX)
+        target_link_libraries(${TARGET_NAME} /opt/rocm/hip/lib/libhip_hcc.so /opt/rocm/hip/lib/libhip_device.a)
+	find_fluid_modules(${TARGET_NAME})
+      endif()
+      if (hip_library_DEPS)
+	add_dependencies(${TARGET_NAME} ${hip_library_DEPS})
+	target_link_libraries(${TARGET_NAME} ${hip_library_DEPS})
+      endif()
+      # cpplint code style
+      foreach(source_file ${hip_library_SRCS})
+	string(REGEX REPLACE "\\.[^.]*$" "" source ${source_file})
+	if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${source}.h)
+	  list(APPEND hip_library_HEADERS ${CMAKE_CURRENT_SOURCE_DIR}/${source}.h)
+	endif()
+      endforeach()
+      add_style_check_target(${TARGET_NAME} ${hip_library_SRCS} ${hip_library_HEADERS})
+    else(hip_library_SRCS)
+      if (hip_library_DEPS)
+	merge_static_libs(${TARGET_NAME} ${hip_library_DEPS})
+      else()
+	message(FATAL "Please specify source file or library in nv_library.")
+      endif()
+    endif(hip_library_SRCS)
+  endif()
+endfunction(hip_library)
+
+function(hip_binary TARGET_NAME)
+  if (WITH_AMD_GPU)
+    set(options "")
+    set(oneValueArgs "")
+    set(multiValueArgs SRCS DEPS)
+    cmake_parse_arguments(hip_binary "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+    hip_add_executable(${TARGET_NAME} ${hip_binary_SRCS})
+    if(hip_binary_DEPS)
+      target_link_libraries(${TARGET_NAME} ${hip_binary_DEPS})
+      add_dependencies(${TARGET_NAME} ${hip_binary_DEPS})
+    endif()
+  endif()
+endfunction(hip_binary)
+
+function(hip_test TARGET_NAME)
+  if (WITH_AMD_GPU AND WITH_TESTING)
+    set(options "")
+    set(oneValueArgs "")
+    set(multiValueArgs SRCS DEPS)
+    cmake_parse_arguments(hip_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+    set(_sources ${hip_test_SRCS})
+    HIP_PREPARE_TARGET_COMMANDS(${TARGET_NAME} OBJ _generated_files _source_files ${_sources} HIPCC_OPTIONS ${_hipcc_options} HCC_OPTIONS ${_hcc_options} NVCC_OPTIONS ${_nvcc_options})
+    if(_source_files)
+      list(REMOVE_ITEM _sources ${_source_files})
+    endif()
+    add_executable(${TARGET_NAME} ${_cmake_options} ${_generated_files} ${_sources})
+    set_target_properties(${TARGET_NAME} PROPERTIES LINKER_LANGUAGE HIP)
+    target_link_libraries(${TARGET_NAME} ${hip_test_DEPS} paddle_gtest_main paddle_memory gtest gflags)
+    add_dependencies(${TARGET_NAME} ${hip_test_DEPS} paddle_gtest_main paddle_memory gtest gflags)
+    add_test(${TARGET_NAME} ${TARGET_NAME})
+  endif()
+endfunction(hip_test)
+
 function(go_library TARGET_NAME)
   set(options STATIC static SHARED shared)
   set(oneValueArgs "")

cmake/hip.cmake

@@ -0,0 +1,43 @@
+if(NOT WITH_AMD_GPU)
+    return()
+endif()
+
+include_directories("/opt/rocm/include")
+include_directories("/opt/rocm/hipblas/include")
+include_directories("/opt/rocm/hiprand/include")
+include_directories("/opt/rocm/rocrand/include")
+include_directories("/opt/rocm/rccl/include")
+include_directories("/opt/rocm/thrust")
+
+list(APPEND EXTERNAL_LIBS "-L/opt/rocm/lib/ -lhip_hcc")
+
+set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -fPIC -DPADDLE_WITH_HIP -std=c++14" )
+
+if(WITH_DSO)
+  set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -DPADDLE_USE_DSO")
+endif(WITH_DSO)
+
+if(WITH_DOUBLE)
+  set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -DPADDLE_TYPE_DOUBLE")
+endif(WITH_DOUBLE)
+
+if(WITH_TESTING)
+  set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -DPADDLE_WITH_TESTING")
+endif(WITH_TESTING)
+
+if(CMAKE_BUILD_TYPE  STREQUAL "Debug")
+    list(APPEND HIP_HCC_FLAGS  ${CMAKE_CXX_FLAGS_DEBUG})
+elseif(CMAKE_BUILD_TYPE  STREQUAL "RelWithDebInfo")
+    list(APPEND HIP_HCC_FLAGS  ${CMAKE_CXX_FLAGS_RELWITHDEBINFO})
+elseif(CMAKE_BUILD_TYPE  STREQUAL "MinSizeRel")
+    list(APPEND HIP_HCC_FLAGS  ${CMAKE_CXX_FLAGS_MINSIZEREL})
+endif()
+
+if("x${HCC_HOME}" STREQUAL "x")
+  set(HCC_HOME "/opt/rocm/hcc")
+endif()
+
+set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
+set(CMAKE_HIP_CREATE_SHARED_LIBRARY "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES> -shared")
+set(CMAKE_HIP_CREATE_SHARED_MODULE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES> -shared")
+

doc/CMakeLists.txt

@@ -1 +1,2 @@
 add_subdirectory(v2)
+add_subdirectory(fluid)

doc/fluid/CMakeLists.txt

@@ -0,0 +1,49 @@
+if(NOT DEFINED SPHINX_THEME)
+    set(SPHINX_THEME default)
+endif()
+
+if(NOT DEFINED SPHINX_THEME_DIR)
+    set(SPHINX_THEME_DIR)
+endif()
+
+# configured documentation tools and intermediate build results
+set(BINARY_BUILD_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_build")
+
+# Sphinx cache with pickled ReST documents
+set(SPHINX_CACHE_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_doctrees")
+
+# HTML output director
+set(SPHINX_HTML_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/html")
+
+configure_file(
+    "${CMAKE_CURRENT_SOURCE_DIR}/../templates/conf.py.en.in"
+    "${BINARY_BUILD_DIR_EN}/conf.py"
+    @ONLY)
+
+sphinx_add_target(paddle_fluid_docs
+                  html
+                  ${BINARY_BUILD_DIR_EN}
+                  ${SPHINX_CACHE_DIR_EN}
+                  ${CMAKE_CURRENT_SOURCE_DIR}
+                  ${SPHINX_HTML_DIR_EN})
+
+# configured documentation tools and intermediate build results
+set(BINARY_BUILD_DIR_CN "${CMAKE_CURRENT_BINARY_DIR}/cn/_build")
+
+# Sphinx cache with pickled ReST documents
+set(SPHINX_CACHE_DIR_CN "${CMAKE_CURRENT_BINARY_DIR}/cn/_doctrees")
+
+# HTML output directory
+set(SPHINX_HTML_DIR_CN "${CMAKE_CURRENT_BINARY_DIR}/cn/html")
+
+configure_file(
+    "${CMAKE_CURRENT_SOURCE_DIR}/../templates/conf.py.cn.in"
+    "${BINARY_BUILD_DIR_CN}/conf.py"
+    @ONLY)
+
+sphinx_add_target(paddle_fluid_docs_cn
+                  html
+                  ${BINARY_BUILD_DIR_CN}
+                  ${SPHINX_CACHE_DIR_CN}
+                  ${CMAKE_CURRENT_SOURCE_DIR}
+                  ${SPHINX_HTML_DIR_CN})

doc/fluid/build_and_install/index_cn.rst

@@ -0,0 +1,2 @@
+安装与使用
+------------

doc/fluid/build_and_install/index_en.rst

@@ -0,0 +1,2 @@
+Build and Install
+------------

doc/fluid/design/dist_train/distributed_lookup_table_design.md

@@ -26,7 +26,7 @@ lookup of rows.
 The following figure illustrates the multiplication of x with two
 non-zero elements, or say, two symbols, and a lookup table W:
 
-![lookup table](./lookup_table.png)
+![lookup table](./src/lookup_table.png)
 
 ### The Backward Algorithm
 
@@ -42,7 +42,7 @@ or some more sophisticated algorithms that rely on both W' and W:
 $$W = f(W, W')$$
 
 The following figure illustrates the backward pass of the lookup
-operator: ![lookup table training](./lookup_table_training.png)
+operator: ![lookup table training](./src/lookup_table_training.png)
 
 ## Distributed Storage Service
 

doc/fluid/design/dynamic_rnn/rnn.md

@@ -5,7 +5,7 @@ This document describes the RNN (Recurrent Neural Network) operator and how it i
 ## RNN Algorithm Implementation
 
 <p align="center">
-<img src="./images/rnn.jpg"/>
+<img src="./rnn.jpg"/>
 </p>
 
 The above diagram shows an RNN unrolled into a full network.
@@ -22,7 +22,7 @@ There are several important concepts here:
 There could be local variables defined in each step-net.  PaddlePaddle runtime realizes these variables in *step-scopes* which are created for each step.
 
 <p align="center">
-<img src="./images/rnn.png"/><br/>
+<img src="./rnn.png"/><br/>
 Figure 2 illustrates the RNN's data flow
 </p>
 
@@ -49,7 +49,7 @@ or copy the memory value of the previous step to the current ex-memory variable.
 
 ### Usage in Python
 
-For more information on Block, please refer to the [design doc](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/block.md).
+For more information on Block, please refer to the [design doc](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/fluid/design/concepts/block.md).
 
 We can define an RNN's step-net using a Block:
 
@@ -93,7 +93,7 @@ For example, we could have a 2-level RNN, where the top level corresponds to par
 The following figure illustrates feeding in text into the lower level, one sentence at a step, and the feeding in step outputs to the top level. The final top level output is about the whole text.
 
 <p align="center">
-<img src="./images/2_level_rnn.png"/>
+<img src="./2_level_rnn.png"/>
 </p>
 
 ```python
@@ -149,5 +149,5 @@ If the `output_all_steps` is set to False, it will only output the final time st
 
 
 <p align="center">
-<img src="images/rnn_2level_data.png"/>
+<img src="./rnn_2level_data.png"/>
 </p>

doc/fluid/design/index_cn.rst

@@ -0,0 +1,2 @@
+设计思想
+------------

doc/fluid/design/index_en.rst

@@ -0,0 +1,2 @@
+Design
+------------

doc/fluid/design/motivation/fluid.md

@@ -103,7 +103,7 @@ In computability theory, a system of data-manipulation rules, such as a programm
 
 There are two ways to execute a Fluid program.  When a program is executed, it creates a protobuf message [`ProgramDesc`](https://github.com/PaddlePaddle/Paddle/blob/a91efdde6910ce92a78e3aa7157412c4c88d9ee8/paddle/framework/framework.proto#L145) that describes the process and is conceptually like an [abstract syntax tree](https://en.wikipedia.org/wiki/Abstract_syntax_tree).
 
-There is a C++ class [`Executor`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/executor.h), which runs a `ProgramDesc`, similar to how an interpreter runs a Python program.
+There is a C++ class [`Executor`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/executor.h), which runs a `ProgramDesc`, similar to how an interpreter runs a Python program.
 
 Fluid is moving towards the direction of a compiler, which is explain in [fluid_compiler.md](fluid_compiler.md).
 

doc/fluid/dev/index_cn.rst

@@ -0,0 +1,2 @@
+开发标准
+------------

doc/fluid/dev/index_en.rst

@@ -0,0 +1,4 @@
+Development
+------------
+
+This is Development page

doc/fluid/faq/index_cn.rst

@@ -0,0 +1,2 @@
+FAQ
+------------

doc/fluid/faq/index_en.rst

@@ -0,0 +1,2 @@
+FAQ
+------------