diff --git a/CMakeLists.txt b/CMakeLists.txt index c7d743e193e7d32dbc0b56f3bcb05b6c61f85f1d..b174831109372cb014741d63032fa6a470e74042 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -36,8 +36,8 @@ include(simd) ################################ Configurations ####################################### option(WITH_GPU "Compile PaddlePaddle with NVIDIA GPU" ${CUDA_FOUND}) option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND}) -option(WITH_MKLDNN "Compile PaddlePaddle with mkl-dnn support." OFF) -option(WITH_MKLML "Compile PaddlePaddle with mklml package." OFF) +option(WITH_MKLDNN "Compile PaddlePaddle with mkl-dnn support." ${AVX_FOUND}) +option(WITH_MKLML "Compile PaddlePaddle with mklml package." ${AVX_FOUND}) option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON) option(WITH_TESTING "Compile PaddlePaddle with unit testing" ON) option(WITH_SWIG_PY "Compile PaddlePaddle with inference api" ON) diff --git a/cmake/configure.cmake b/cmake/configure.cmake index 69220e03fe8e337205f31cb1f45e3e19ae4f5d1e..2ac098954647d37e26ac2499e0675dae39910edc 100644 --- a/cmake/configure.cmake +++ b/cmake/configure.cmake @@ -74,8 +74,6 @@ if(WITH_MKLDNN) set(OPENMP_FLAGS "-fopenmp") set(CMAKE_C_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS}) set(CMAKE_CXX_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS}) - set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -L${MKLDNN_IOMP_DIR} -liomp5 -Wl,--as-needed") - set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -L${MKLDNN_IOMP_DIR} -liomp5 -Wl,--as-needed") set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OPENMP_FLAGS}") set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OPENMP_FLAGS}") else() diff --git a/cmake/cpplint.cmake b/cmake/cpplint.cmake index e50530411cc74392091c8026fa012ec7631f7f6b..5184f0815faac005b3dff1015395235f4e19d65b 100644 --- a/cmake/cpplint.cmake +++ b/cmake/cpplint.cmake @@ -42,29 +42,21 @@ macro(add_style_check_target TARGET_NAME) if(WITH_STYLE_CHECK) set(SOURCES_LIST ${ARGN}) list(REMOVE_DUPLICATES SOURCES_LIST) - list(SORT SOURCES_LIST) - foreach(filename ${SOURCES_LIST}) - set(LINT ON) foreach(pattern ${IGNORE_PATTERN}) if(filename MATCHES ${pattern}) - message(STATUS "DROP LINT ${filename}") - set(LINT OFF) + list(REMOVE_ITEM SOURCES_LIST ${filename}) endif() endforeach() - if(LINT MATCHES ON) - # cpplint code style - get_filename_component(base_filename ${filename} NAME) - set(CUR_GEN ${CMAKE_CURRENT_BINARY_DIR}/${base_filename}.cpplint) - add_custom_command(OUTPUT ${CUR_GEN} PRE_BUILD - COMMAND "${PYTHON_EXECUTABLE}" "${PROJ_ROOT}/paddle/scripts/cpplint.py" - "--filter=${STYLE_FILTER}" - "--write-success=${CUR_GEN}" ${filename} - DEPENDS ${filename} ${PROJ_ROOT}/paddle/scripts/cpplint.py - WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) - add_custom_target(${base_filename}.cpplint DEPENDS ${CUR_GEN}) - add_dependencies(${TARGET_NAME} ${base_filename}.cpplint) - endif() endforeach() + + if(SOURCES_LIST) + add_custom_command(TARGET ${TARGET_NAME} POST_BUILD + COMMAND "${PYTHON_EXECUTABLE}" "${PROJ_ROOT}/paddle/scripts/cpplint.py" + "--filter=${STYLE_FILTER}" + ${SOURCES_LIST} + COMMENT "cpplint: Checking source code style" + WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) + endif() endif() endmacro() diff --git a/cmake/external/any.cmake b/cmake/external/any.cmake index 45e3764e8482a4cfc8ee72fe4d79f04a3c9b74fa..5d2f7219b2007493916a39e839d647a9d0046c9f 100644 --- a/cmake/external/any.cmake +++ b/cmake/external/any.cmake @@ -7,7 +7,7 @@ INCLUDE_DIRECTORIES(${ANY_SOURCE_DIR}/src/extern_lib_any) ExternalProject_Add( extern_lib_any ${EXTERNAL_PROJECT_LOG_ARGS} - GIT_REPOSITORY "https://github.com/thelink2012/any.git" + GIT_REPOSITORY "https://github.com/PaddlePaddle/any.git" GIT_TAG "8fef1e93710a0edf8d7658999e284a1142c4c020" PREFIX ${ANY_SOURCE_DIR} UPDATE_COMMAND "" diff --git a/cmake/external/gflags.cmake b/cmake/external/gflags.cmake index a0d0a892c4b3cc3743ac725f3cd90444f18abf34..16e5bef4cdb8d6513de51838e3c3c8398dbad60d 100644 --- a/cmake/external/gflags.cmake +++ b/cmake/external/gflags.cmake @@ -28,7 +28,14 @@ INCLUDE_DIRECTORIES(${GFLAGS_INCLUDE_DIR}) ExternalProject_Add( extern_gflags ${EXTERNAL_PROJECT_LOG_ARGS} - GIT_REPOSITORY "https://github.com/gflags/gflags.git" + # TODO(yiwang): The annoying warnings mentioned in + # https://github.com/PaddlePaddle/Paddle/issues/3277 are caused by + # gflags. I fired a PR https://github.com/gflags/gflags/pull/230 + # to fix it. Before it gets accepted by the gflags team, we use + # my personal fork, which contains above fix, temporarily. Let's + # change this back to the official Github repo once my PR is + # merged. + GIT_REPOSITORY "https://github.com/wangkuiyi/gflags.git" PREFIX ${GFLAGS_SOURCES_DIR} UPDATE_COMMAND "" CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER} diff --git a/cmake/external/openblas.cmake b/cmake/external/openblas.cmake index 60a1041936437775e0994157b8ffcb7c52b7ab87..db09232c0e69016bf18c1d981e4620e9e804ff7c 100644 --- a/cmake/external/openblas.cmake +++ b/cmake/external/openblas.cmake @@ -69,8 +69,13 @@ ENDIF(NOT ${CBLAS_FOUND}) MESSAGE(STATUS "BLAS library: ${CBLAS_LIBRARIES}") INCLUDE_DIRECTORIES(${CBLAS_INC_DIR}) -ADD_LIBRARY(cblas STATIC IMPORTED) -SET_PROPERTY(TARGET cblas PROPERTY IMPORTED_LOCATION ${CBLAS_LIBRARIES}) +# FIXME(gangliao): generate cblas target to track all high performance +# linear algebra libraries for cc_library(xxx SRCS xxx.c DEPS cblas) +SET(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/cblas_dummy.c) +FILE(WRITE ${dummyfile} "const char * dummy = \"${dummyfile}\";") +ADD_LIBRARY(cblas STATIC ${dummyfile}) +TARGET_LINK_LIBRARIES(cblas ${CBLAS_LIBRARIES}) + IF(NOT ${CBLAS_FOUND}) ADD_DEPENDENCIES(cblas extern_openblas) LIST(APPEND external_project_dependencies cblas) diff --git a/cmake/flags.cmake b/cmake/flags.cmake index d00a9bb3a30cfb16623e073414088059481c3e1a..e26d8d9df386e65137aa83cc60a43bfeabf7a4a6 100644 --- a/cmake/flags.cmake +++ b/cmake/flags.cmake @@ -115,7 +115,7 @@ set(COMMON_FLAGS -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs - -Wno-error=parentheses-equality # Warnings in Pybind11 + -Wno-error=parentheses-equality # Warnings in pybind11 ) set(GPU_COMMON_FLAGS @@ -195,6 +195,7 @@ endif() # Modern gpu architectures: Pascal if (CUDA_VERSION VERSION_GREATER "8.0" OR CUDA_VERSION VERSION_EQUAL "8.0") list(APPEND __arch_flags " -gencode arch=compute_60,code=sm_60") + list(APPEND CUDA_NVCC_FLAGS --expt-relaxed-constexpr) endif() # Custom gpu architecture diff --git a/cmake/generic.cmake b/cmake/generic.cmake index 41b9b5928958ae31799c396a8d77fd7cff557905..957c20bcf603f2f264b4658f63ac0eec438f12b1 100644 --- a/cmake/generic.cmake +++ b/cmake/generic.cmake @@ -403,3 +403,16 @@ function(py_proto_compile TARGET_NAME) protobuf_generate_python(py_srcs ${py_proto_compile_SRCS}) add_custom_target(${TARGET_NAME} ALL DEPENDS ${py_srcs}) endfunction() + +function(py_test TARGET_NAME) + if(WITH_TESTING) + set(options STATIC static SHARED shared) + set(oneValueArgs "") + set(multiValueArgs SRCS DEPS) + cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) + add_test(NAME ${TARGET_NAME} + COMMAND env PYTHONPATH=${PADDLE_PYTHON_PACKAGE_DIR} + python2 ${py_test_SRCS} + WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) + endif() +endfunction() diff --git a/doc/design/mkldnn/README.MD b/doc/design/mkldnn/README.MD new file mode 100644 index 0000000000000000000000000000000000000000..e956994431fbb43438c56dcd96ad8313cf516090 --- /dev/null +++ b/doc/design/mkldnn/README.MD @@ -0,0 +1,110 @@ +# Intel® MKL-DNN on PaddlePaddle: Design Doc + +我们计划将Intel深度神经网络数学库(**MKL-DNN**\[[1](#references)\])集成到PaddlePaddle，充分展现英特尔平台的优势，有效提升PaddlePaddle在英特尔架构上的性能。 + +我们短期内的基本目标是： + +- 完成常用layer的MKL-DNN实现。 +- 完成常见深度神经网络VGG，GoogLeNet 和 ResNet的MKL-DNN实现。 + + +## Contents + +- [Overview](#overview) +- [Actions](#actions) + - [CMake](#cmake) + - [Layers](#layers) + - [Activations](#activations) + - [Unit Tests](#unit-tests) + - [Protobuf Messages](#protobuf-messages) + - [Python API](#python-api) + - [Demos](#demos) + - [Benchmarking](#benchmarking) + - [Others](#others) +- [Design Concerns](#design-concerns) + +## Overview + +我们会把MKL-DNN作为第三方库集成进PaddlePaddle，整体框架图 +
+
+Figure 1. PaddlePaddle on IA. +
+ +## Actions +我们把集成方案大致分为了如下几个方面。 + +### CMake +我们会在`CMakeLists.txt`中会添加`WITH_MKLDNN`的选项，当设置这个值为`ON`的时候会启用编译MKL-DNN功能。同时会自动开启OpenMP用于提高MKL-DNN的性能。 + +同时，我们会引入`WITH_MKLML`选项，用于选择是否使用MKL-DNN自带的MKLML安装包。这个安装包可以独立于MKL-DNN使用，但是建议在开启MKL-DNN的同时也打开MKLML的开关，这样才能发挥最好的性能。 + +所以，我们会在`cmake/external`目录新建`mkldnn.cmake`和`mklml.cmake`文件，它们会在编译PaddlePaddle的时候下载对应的软件包，并放到PaddlePaddle的third party目录中。 + +**备注**：当`WITH_MKLML=ON`的时候，会优先使用这个包作为PaddlePaddle的CBLAS和LAPACK库，所以会稍微改动`cmake/cblas.cmake`中的逻辑。 + +### Layers +所有MKL-DNN相关的C++ layers，都会按照PaddlePaddle的目录结构存放在 +`paddle/gserver/layers`中，并且文件名都会一以*Mkldnn*开头。 + +所有MKL-DNN的layers都会继承于一个叫做`MkldnnLayer`的父类，该父类继承于PaddlePaddle的基类`Layer`。 + +### Activations +由于在PaddlePaddle中，激活函数是独立于layer概念的，所以会在`paddle/gserver/activations`目录下添加一个`MkldnnActivation.h`文件定义一些用于MKL-DNN的接口，实现方法还是会在`ActivationFunction.cpp`文件。 + +### Unit Tests +会在`paddle/gserver/test`目录下添加`test_Mkldnn.cpp`和`MkldnnTester.*`用于MKL-DNN的测试。 + +Activation的测试，计划在PaddlePaddle原有的测试文件上直接添加新的测试type。 + +### Protobuf Messages +根据具体layer的需求可能会在`proto/ModelConfig.proto`里面添加必要的选项。 + +### Python API +目前只考虑**v1 API**。 + +计划在`python/paddle/trainer/config_parser.py`里面添加`use_mkldnn`这个选择，方便用户选择使用MKL-DNN的layers。 + +具体实现方式比如： + +```python +use_mkldnn = bool(int(g_command_config_args.get("use_mkldnn", 0))) +if use_mkldnn + self.layer_type = mkldnn_* +``` + +所有MKL-DNN的layer type会以*mkldnn_*开头，以示区分。 + +并且可能在`python/paddle/trainer_config_helper`目录下的`activations.py `和`layers.py`里面添加必要的MKL-DNN的接口。 + +### Demos + +会在`v1_api_demo`目录下添加一个`mkldnn`的文件夹，里面放入一些用于MKL-DNN测试的demo脚本。 + +### Benchmarking +会考虑添加部分逻辑在`benchmark/paddle/image/run.sh`，添加使用MKL-DNN的测试。 + +### Others +1. 如果在使用MKL-DNN的情况下，会把CPU的Buffer对齐为64。 +2. 深入PaddlePaddle，寻找有没有其他可以优化的可能，进一步优化。比如可能会用OpenMP改进SGD的更新性能。 + +## Design Concerns + +为了更好的符合PaddlePaddle的代码风格\[[2](#references)\]，同时又尽可能少的牺牲MKL-DNN的性能\[[3](#references)\]。 + +我们总结出一些特别需要注意的点： + +1. 使用**deviceId_**。为了尽可能少的在父类Layer中添加变量或者函数，我们决定使用已有的`deviceId_`变量来区分layer的属性，定义`-2`为`MkldnnLayer`特有的设备ID。 +2. 重写父类Layer的**init**函数，修改`deviceId_`为`-2`，代表这个layer是用于跑在MKL-DNN的环境下。 +3. 创建`MkldnnMatrix`，用于管理MKL-DNN会用到的相关memory函数、接口以及会用的到格式信息。 +4. 创建`MkldnnBase`，定义一些除了layer和memory相关的类和函数。包括MKL-DNN会用到`MkldnnStream`和`CpuEngine`，和未来可能还会用到`FPGAEngine`等。 +5. 在**Argument**里添加两个`MkldnnMatrixPtr`，取名为`mkldnnValue`和`mkldnnGrad`，用于存放`MkldnnLayer`会用到的memory buffer。 并且添加函数cvt(会修改为一个更加合适的函数名)，用于处理"CPU device"和"MKL-DNN device"之间memory的相互转化。 +6. 在父类`Layer`中的`getOutput`函数中添加一段逻辑，用于判断`deviceId`，并针对device在MKL-DNN和CPU之间不统一的情况，做一个前期转换。 也就是调用`Argument`的cvt函数把output统一到需要的device上。 +7. 在原来的`FLAGS`中添加一个`use_mkldnn`的flag，用于选择是否使用MKL-DNN的相关功能。 + +## References + +1. [Intel Math Kernel Library for Deep Neural Networks (Intel MKL-DNN)](https://github.com/01org/mkl-dnn "Intel MKL-DNN") +2. [原来的方案](https://github.com/PaddlePaddle/Paddle/pull/3096)会引入**nextLayer**的信息。但是在PaddlePaddle中，无论是重构前的layer还是重构后的op，都不会想要知道next layer/op的信息。 +3. MKL-DNN的高性能格式与PaddlePaddle原有的`NCHW`不同(PaddlePaddle中的CUDNN部分使用的也是`NCHW`，所以不存在这个问题)，所以需要引入一个转换方法，并且只需要在必要的时候转换这种格式，才能更好的发挥MKL-DNN的性能。 + diff --git a/doc/design/mkldnn/image/overview.png b/doc/design/mkldnn/image/overview.png new file mode 100644 index 0000000000000000000000000000000000000000..84b455c28230703599a2529f014cfbb222138fef Binary files /dev/null and b/doc/design/mkldnn/image/overview.png differ diff --git a/paddle/api/test/CMakeLists.txt b/paddle/api/test/CMakeLists.txt index f3b1c2c4d438b5d3e776ef27ce8f8b78f710f2ab..761aeb5b174105edece8880a9f5012c13a63fd11 100644 --- a/paddle/api/test/CMakeLists.txt +++ b/paddle/api/test/CMakeLists.txt @@ -1,2 +1,6 @@ -add_python_test(test_swig_api - testArguments.py testGradientMachine.py testMatrix.py testVector.py testTrain.py testTrainer.py) +py_test(testTrain SRCS testTrain.py) +py_test(testMatrix SRCS testMatrix.py) +py_test(testVector SRCS testVector.py) +py_test(testTrainer SRCS testTrainer.py) +py_test(testArguments SRCS testArguments.py) +py_test(testGradientMachine SRCS testGradientMachine.py) diff --git a/paddle/framework/operator.cc b/paddle/framework/operator.cc index d3fbb1e96199b87887cd313d3cf31a230c9b8a34..10ed1f9e32201d887b10ee7ccab8f44573ed1290 100644 --- a/paddle/framework/operator.cc +++ b/paddle/framework/operator.cc @@ -22,14 +22,14 @@ namespace framework { template <> Eigen::DefaultDevice& ExecutionContext::GetEigenDevice< platform::CPUPlace, Eigen::DefaultDevice>() const { - return *device_context_.get_eigen_device(); + return *device_context_->get_eigen_device(); } #ifndef PADDLE_ONLY_CPU template <> Eigen::GpuDevice& ExecutionContext::GetEigenDevice() const { - return *device_context_.get_eigen_device(); + return *device_context_->get_eigen_device(); } #endif diff --git a/paddle/framework/operator.h b/paddle/framework/operator.h index 02707b8a9996fca80c96d76240e5e89cbb6f451e..9672492d1c2c6a82c37e0a840a4ca9c111de06d8 100644 --- a/paddle/framework/operator.h +++ b/paddle/framework/operator.h @@ -174,7 +174,11 @@ class OperatorContext { template T* Output(const size_t index) const { auto var = OutputVar(index); - PADDLE_ENFORCE(var != nullptr, "Output(%d) should not be nullptr", index); + PADDLE_ENFORCE( + var != nullptr, + "Output(%d) not be nullptr, which means variable [%s] does not " + "exist in scope", + index, op_.outputs_[index]); return var->GetMutable(); } @@ -252,7 +256,7 @@ struct EigenDeviceConverter { class ExecutionContext : public OperatorContext { public: ExecutionContext(const OperatorBase* op, const Scope& scope, - const platform::DeviceContext& device_context) + const platform::DeviceContext* device_context) : OperatorContext(op, scope), device_context_(device_context) {} template GetPlace(); } - const platform::DeviceContext& device_context_; + const platform::DeviceContext* device_context_; }; class OpKernel { @@ -311,7 +315,7 @@ class OperatorWithKernel : public OperatorBase { void Run(const Scope& scope, const platform::DeviceContext& dev_ctx) const final { auto& opKernel = AllOpKernels().at(type_).at(OpKernelKey(dev_ctx)); - opKernel->Compute(ExecutionContext(this, scope, dev_ctx)); + opKernel->Compute(ExecutionContext(this, scope, &dev_ctx)); } static std::unordered_map& diff --git a/paddle/function/nnpack/NNPACKConvOp.cpp b/paddle/function/nnpack/NNPACKConvOp.cpp index f0ec77a5d00333993427fb8d0bc938c884e50c95..00d048eb216baf37c875c870a31cfd55a97f2974 100644 --- a/paddle/function/nnpack/NNPACKConvOp.cpp +++ b/paddle/function/nnpack/NNPACKConvOp.cpp @@ -49,9 +49,7 @@ class NNPACKConvFunction : public ConvFunctionBase { public: void init(const FuncConfig& config) override { ConvFunctionBase::init(config); - CHECK_EQ(groups_, (size_t)1); algorithm_ = get_nnp_convolution_algorithm(config.get("algo")); - // algorithm_ = nnp_convolution_algorithm_auto; transform_strategy_ = nnp_convolution_transform_strategy_compute; nnp_status status = nnp_initialize(); CHECK_EQ(status, nnp_status_success); @@ -67,8 +65,7 @@ public: } } - virtual void check(const BufferArgs& inputs, - const BufferArgs& outputs) override { + void check(const BufferArgs& inputs, const BufferArgs& outputs) override { const TensorShape& input = inputs[0].shape(); const TensorShape& filter = inputs[1].shape(); const TensorShape& output = outputs[0].shape(); @@ -91,8 +88,8 @@ public: size_t filterHeight = getFilterHeight(filter); size_t filterWidth = getFilterWidth(filter); size_t outputChannels = output[1]; - // size_t outputHeight = output[2]; - // size_t outputWidth = output[3]; + size_t outputHeight = output[2]; + size_t outputWidth = output[3]; nnp_size inputSize = {.width = inputWidth, .height = inputHeight}; nnp_padding padding = {.top = (size_t)paddingH(), @@ -171,49 +168,58 @@ public: } } + size_t inputOffset = inputChannels / groups_ * inputHeight * inputWidth; + size_t outputOffset = outputChannels / groups_ * outputHeight * outputWidth; + size_t filterOffset = filter.getElements() / groups_; + if (batchSize == 1) { - nnp_status status = - nnp_convolution_inference(algorithm_, - transform_strategy_, - inputChannels, - outputChannels, - inputSize, - padding, - kernelSize, - outputSubsampling, - inputData, - filterData, - nullptr, /* bias */ - outputData, - bufferPtr, - sizePtr, - nnp_activation_identity, - nullptr, - threadpool_, /* threadpool */ - nullptr); - CHECK_EQ(status, nnp_status_success); + for (size_t g = 0; g < groups_; g++) { + nnp_status status = + nnp_convolution_inference(algorithm_, + transform_strategy_, + inputChannels / groups_, + outputChannels / groups_, + inputSize, + padding, + kernelSize, + outputSubsampling, + inputData + inputOffset * g, + filterData + filterOffset * g, + nullptr, /* bias */ + outputData + outputOffset * g, + bufferPtr, + sizePtr, + nnp_activation_identity, + nullptr, + threadpool_, /* threadpool */ + nullptr); + CHECK_EQ(status, nnp_status_success); + } } else { - // only supports stride = 1 - CHECK_EQ(strideH(), 1); - CHECK_EQ(strideW(), 1); - nnp_status status = nnp_convolution_output(algorithm_, - batchSize, - inputChannels, - outputChannels, - inputSize, - padding, - kernelSize, - inputData, - filterData, - nullptr, /* bias */ - outputData, - bufferPtr, - sizePtr, - nnp_activation_identity, - nullptr, - threadpool_, /* threadpool */ - nullptr); - CHECK_EQ(status, nnp_status_success); + for (size_t g = 0; g < groups_; g++) { + // only supports stride = 1 + CHECK_EQ(strideH(), 1); + CHECK_EQ(strideW(), 1); + nnp_status status = + nnp_convolution_output(algorithm_, + batchSize, + inputChannels / groups_, + outputChannels / groups_, + inputSize, + padding, + kernelSize, + inputData + inputOffset * g, + filterData + filterOffset * g, + nullptr, /* bias */ + outputData + outputOffset * g, + bufferPtr, + sizePtr, + nnp_activation_identity, + nullptr, + threadpool_, /* threadpool */ + nullptr); + CHECK_EQ(status, nnp_status_success); + } } } diff --git a/paddle/gserver/layers/ExpandConvLayer.cpp b/paddle/gserver/layers/ExpandConvLayer.cpp index 783e02e47cb91e28eb88b079f1e94439d34fa775..0ece2799318ea5ecc91f97f71289d4d07246dcaa 100644 --- a/paddle/gserver/layers/ExpandConvLayer.cpp +++ b/paddle/gserver/layers/ExpandConvLayer.cpp @@ -57,8 +57,7 @@ bool ExpandConvLayer::init(const LayerMap &layerMap, convGradFilterType = "GemmConvGradFilter"; } - if (FLAGS_use_nnpack) { - CHECK_EQ(isDeconv_, false); + if (FLAGS_use_nnpack && !isDeconv_) { createFunction(forward_, "NNPACKConv", FuncConfig() diff --git a/paddle/gserver/tests/CMakeLists.txt b/paddle/gserver/tests/CMakeLists.txt index 4546d12a903084e7a746b967c39d67a0ade4c0cd..5511ab6b8bb05108e76cc0913264d864d2fecf5b 100644 --- a/paddle/gserver/tests/CMakeLists.txt +++ b/paddle/gserver/tests/CMakeLists.txt @@ -1,10 +1,5 @@ # gserver pacakge unittests -file(GLOB_RECURSE GSERVER_HEADER RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "*.h") -file(GLOB_RECURSE GSERVER_SOURCES RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "*.cpp") -add_style_check_target(paddle_gserver ${GSERVER_SOURCES}) -add_style_check_target(paddle_gserver ${GSERVER_HEADER}) - ################### test_ProtoDataProvider ############ add_unittest_without_exec(test_ProtoDataProvider test_ProtoDataProvider.cpp) diff --git a/paddle/operators/add_op.cc b/paddle/operators/add_op.cc index 7fbdd84a391c7d0048fca473f7318561df50daa2..d4c05ed483ca56a31dd8ee4d81b54551ae6da0d1 100644 --- a/paddle/operators/add_op.cc +++ b/paddle/operators/add_op.cc @@ -20,8 +20,8 @@ namespace operators { class AddOp : public OperatorWithKernel { protected: void InferShape(const InferShapeContext &ctx) const override { - PADDLE_ENFORCE(ctx.InputSize() == 2, "Input size of AddOp must be two"); - PADDLE_ENFORCE(ctx.OutputSize() == 1, "Output size of AddOp must be one"); + PADDLE_ENFORCE_EQ(ctx.InputSize(), 2); + PADDLE_ENFORCE_EQ(ctx.OutputSize(), 1); PADDLE_ENFORCE(ctx.InputVar(0) != nullptr && ctx.InputVar(1) != nullptr, "Inputs of AddOp must all be set"); PADDLE_ENFORCE(ctx.OutputVar(0) != nullptr, diff --git a/paddle/operators/mul_op.cc b/paddle/operators/mul_op.cc index f41e95e9db494109925fb600ec6bbd47edf6cc74..ccab9a994cc7aa9e389bd259e4c7365a06e93aa1 100644 --- a/paddle/operators/mul_op.cc +++ b/paddle/operators/mul_op.cc @@ -23,12 +23,16 @@ class MulOp : public OperatorWithKernel { PADDLE_ENFORCE(ctx.InputSize() == 2, "The mul op must take two inputs"); auto dim0 = ctx.Input(0)->dims(); auto dim1 = ctx.Input(1)->dims(); - PADDLE_ENFORCE(dim0.size() == 2 && dim1.size() == 2, - "The input of mul op must be matrix"); - PADDLE_ENFORCE( - dim0[1] == dim1[0], + PADDLE_ENFORCE_EQ(dim0.size(), 2, + "input X(%s) should be a tensor with 2 dims, a matrix", + ctx.op_.Input("X")); + PADDLE_ENFORCE_EQ(dim1.size(), 2, + "input Y(%s) should be a tensor with 2 dims, a matrix", + ctx.op_.Input("Y")); + PADDLE_ENFORCE_EQ( + dim0[1], dim1[0], "First matrix's width must be equal with second matrix's height."); - PADDLE_ENFORCE(ctx.OutputSize() == 1, "The mul op must take one output"); + PADDLE_ENFORCE_EQ(ctx.OutputSize(), 1, "The mul op takes only one output"); ctx.Output(0)->Resize({dim0[0], dim1[1]}); } }; diff --git a/paddle/operators/recurrent_op.cc b/paddle/operators/recurrent_op.cc index 389d4323950269b81912a7213ff64872aafb410f..5e9c15ca0e6a7c56611a0fadda6c3c0839f309e6 100644 --- a/paddle/operators/recurrent_op.cc +++ b/paddle/operators/recurrent_op.cc @@ -36,6 +36,7 @@ void RecurrentAlgorithm::InferShape(const Scope& scope) const { InitMemories(step_scopes[0], true /*infer_shape_mode*/); Variable* net = scope.FindVar(arg_->step_net); PADDLE_ENFORCE(net != nullptr, "failed to get step net"); + for (size_t i = 0; i < seq_len_; i++) { if (i > 0) { rnn::LinkMemories(step_scopes, arg_->memories, i, -1, @@ -56,6 +57,7 @@ void RecurrentAlgorithm::Run(const Scope& scope, Variable* net = scope.FindVar(arg_->step_net); for (size_t step_id = 0; step_id < seq_len_; step_id++) { + // create output alias variables if (step_id > 0) { rnn::LinkMemories(step_scopes, arg_->memories, step_id, -1, false /*infer_shape_mode*/); @@ -67,22 +69,31 @@ void RecurrentAlgorithm::Run(const Scope& scope, } void RecurrentAlgorithm::CreateScopes(const Scope& scope) const { - // TODO(xxx) Only two scopes are needed for inference, this case will be + // TODO(superjom) Only two scopes are needed for inference, this case will be // supported later. - auto step_scopes = - scope.FindVar(arg_->step_scopes)->GetMutable>(); + auto step_scopes_var = scope.FindVar(arg_->step_scopes); + PADDLE_ENFORCE(step_scopes_var != nullptr, ""); + auto step_scopes = step_scopes_var->GetMutable>(); + + // Now all variables in scope must be created outside of op. + auto net_var = scope.FindVar(arg_->step_net); + PADDLE_ENFORCE(net_var != nullptr, "no stepnet called %s in scope", + arg_->step_net); + auto net_op = net_var->GetMutable(); + PADDLE_ENFORCE(!net_op->outputs_.empty(), "net_op has no outputs"); if (seq_len_ > step_scopes->size()) { for (size_t i = step_scopes->size(); i < seq_len_; ++i) { auto& step_scope = scope.NewScope(); - // Now all variables in scope must be created outside of op. - auto net_op = scope.FindVar(arg_->step_net)->GetMutable(); + // create step net's temp inputs for (auto& input : net_op->inputs_) { // the weight are located in parent scope - if (!step_scope.FindVar(input)) step_scope.NewVar(input); + if (!step_scope.FindVar(input)) + step_scope.NewVar(input)->GetMutable(); } - for (auto& output : net_op->outputs_) { + // create stepnet's outputs + for (const auto& output : net_op->outputs_) { step_scope.NewVar(output); } step_scopes->emplace_back(&step_scope); @@ -100,6 +111,7 @@ void RecurrentAlgorithm::InitMemories(Scope* step_scope, Tensor* boot_mem = step_scope->FindVar(attr.boot_var)->GetMutable(); if (infer_shape_mode) { pre_mem->Resize(boot_mem->dims()); + PADDLE_ENFORCE_EQ(pre_mem->dims().size(), 2); } else { pre_mem->ShareDataWith(*boot_mem); } diff --git a/paddle/operators/rnn/recurrent_op_utils.cc b/paddle/operators/rnn/recurrent_op_utils.cc index 43c97ba29f637828d717ac82516769deff52c7da..32c6c2dd4efa85359b4e95471e8ba09e56afec57 100644 --- a/paddle/operators/rnn/recurrent_op_utils.cc +++ b/paddle/operators/rnn/recurrent_op_utils.cc @@ -53,11 +53,13 @@ void ConcatOutputs(const std::vector& step_scopes, PADDLE_ENFORCE(output_var != nullptr, "output link [%s] is not in scope.", outlinks[i].external); Tensor* output = output_var->GetMutable(); + if (infer_shape_mode) { - fmw::DDim step_dims = step_scopes[0] - ->FindVar(outlinks[i].internal) - ->GetMutable() - ->dims(); + auto step_scope_var = step_scopes[0]->FindVar(outlinks[i].internal); + PADDLE_ENFORCE(step_scope_var != nullptr, "%s not in scope", + outlinks[i].internal); + fmw::DDim step_dims = + step_scope_var->template GetMutable()->dims(); std::vector dims_vec = vectorize(step_dims); dims_vec.insert(dims_vec.begin(), seq_len); output->Resize(fmw::make_ddim(dims_vec)); @@ -79,14 +81,15 @@ void LinkMemories(const std::vector& scopes, const std::vector& memories, const size_t step_id, const int offset, bool infer_shape_mode) { - PADDLE_ENFORCE(step_id < scopes.size(), - "step [%d] is out of range of step scopes' size [%d]", step_id, - scopes.size()); - PADDLE_ENFORCE(static_cast(step_id) + offset >= 0, - "offset [%d] must be large than -[%d]", offset, step_id); - PADDLE_ENFORCE(step_id + offset < scopes.size(), - "offset [%d] is out of range, it must be less than (%d - %d)", - offset, scopes.size(), step_id); + PADDLE_ENFORCE_LT(step_id, scopes.size(), + "step [%d] is out of range of step scopes' size [%d]", + step_id, scopes.size()); + PADDLE_ENFORCE_GE(static_cast(step_id) + offset, 0, + "offset [%d] must be large than -[%d]", offset, step_id); + PADDLE_ENFORCE_LT( + step_id + offset, scopes.size(), + "offset [%d] is out of range, it must be less than (%d - %d)", offset, + scopes.size(), step_id); auto scope = scopes[step_id]; auto linked_scope = scopes[step_id + offset]; for (auto& attr : memories) { diff --git a/paddle/operators/sigmoid_op.cc b/paddle/operators/sigmoid_op.cc index 9d201eb93a2c0e34dd8e6869e97b43c4e278596e..1eb795faa858796f7a34aa495b43d043fdb5dd43 100644 --- a/paddle/operators/sigmoid_op.cc +++ b/paddle/operators/sigmoid_op.cc @@ -37,10 +37,8 @@ class SigmoidOpMaker : public OpProtoAndCheckerMaker { class SigmoidOpGrad : public OperatorWithKernel { protected: - void InferShape(const InferShapeContext &ctx) const override {} - std::string DebugString() const override { - LOG(INFO) << "SigmoidGrad"; - return ""; + void InferShape(const InferShapeContext &ctx) const override { + ctx.Output(0)->Resize(ctx.Input(0)->dims()); } }; @@ -51,3 +49,5 @@ REGISTER_OP(sigmoid, ops::SigmoidOp, ops::SigmoidOpMaker); REGISTER_GRADIENT_OP(sigmoid, sigmoid_grad, ops::SigmoidOpGrad); REGISTER_OP_CPU_KERNEL(sigmoid, ops::SigmoidKernel); +REGISTER_OP_CPU_KERNEL(sigmoid_grad, + ops::SigmoidGradKernel); diff --git a/paddle/operators/sigmoid_op.cu b/paddle/operators/sigmoid_op.cu index 2123b17e4b5e90c22c2d6e9177f2a8956f8a4ac9..e80ba081f2ff805664cf92f3cb47e9ad51889058 100644 --- a/paddle/operators/sigmoid_op.cu +++ b/paddle/operators/sigmoid_op.cu @@ -16,3 +16,5 @@ #include "paddle/operators/sigmoid_op.h" REGISTER_OP_GPU_KERNEL(sigmoid, ops::SigmoidKernel); +REGISTER_OP_GPU_KERNEL(sigmoid_grad, + ops::SigmoidGradKernel); diff --git a/paddle/operators/sigmoid_op.h b/paddle/operators/sigmoid_op.h index eb473920a5f866825b52ecb946653ccead7000ea..d513261e74423ce93a50eaaaec1c7d5fadb8f4a8 100644 --- a/paddle/operators/sigmoid_op.h +++ b/paddle/operators/sigmoid_op.h @@ -27,6 +27,7 @@ class SigmoidKernel : public OpKernel { auto output = context.Output(0); output->mutable_data(context.GetPlace()); + // The clipping is used in Paddle's raw implenmention auto X = EigenVector::Flatten(*input); auto Y = EigenVector::Flatten(*output); auto place = context.GetEigenDevice(); @@ -34,5 +35,23 @@ class SigmoidKernel : public OpKernel { Y.device(place) = 1.0 / (1.0 + (-1.0 * X).exp()); } }; + +template +class SigmoidGradKernel : public OpKernel { + public: + void Compute(const ExecutionContext& context) const override { + auto Y_t = context.Input("Y"); + auto dY_t = context.Input(framework::GradVarName("Y")); + auto dX_t = context.Output(framework::GradVarName("X")); + + dX_t->mutable_data(context.GetPlace()); + + auto dX = EigenVector::Flatten(*dX_t); + auto Y = EigenVector::Flatten(*Y_t); + auto dY = EigenVector::Flatten(*dY_t); + dX.device(context.GetEigenDevice()) = dY * Y * (1. - Y); + } +}; + } // namespace operators } // namespace paddle diff --git a/python/paddle/v2/framework/tests/CMakeLists.txt b/python/paddle/v2/framework/tests/CMakeLists.txt index ee6726c31f278331df362616dbc1b6e6d9a8dbee..7eec37678815587b451008eef587b23bcb9beeaf 100644 --- a/python/paddle/v2/framework/tests/CMakeLists.txt +++ b/python/paddle/v2/framework/tests/CMakeLists.txt @@ -1,18 +1,23 @@ -add_python_test(test_framework - test_protobuf.py - test_scope.py - test_default_scope_funcs.py - test_op_creation_methods.py - test_net.py - test_tensor.py - test_fc_op.py - test_add_two_op.py - test_sgd_op.py - test_mul_op.py - test_mean_op.py - test_sigmoid_op.py - test_softmax_op.py - test_rowwise_add_op.py - test_random_op.py - test_network.py - gradient_checker.py) +py_test(test_net SRCS test_net.py) + +py_test(test_fc_op SRCS test_fc_op.py) +py_test(test_scope SRCS test_scope.py) + +py_test(test_tensor SRCS test_tensor.py) +py_test(test_mul_op SRCS test_mul_op.py) + +py_test(test_network SRCS test_network.py) +py_test(test_mean_op SRCS test_mean_op.py) + +py_test(test_protobuf SRCS test_protobuf.py) + +py_test(test_add_two_op SRCS test_add_two_op.py) +py_test(test_sigmoid_op SRCS test_sigmoid_op.py) +py_test(test_softmax_op SRCS test_softmax_op.py) + +py_test(gradient_checker SRCS gradient_checker.py) + +py_test(test_rowwise_add_op SRCS test_rowwise_add_op.py) + +py_test(test_default_scope_funcs SRCS test_default_scope_funcs.py) +py_test(test_op_creation_methods SRCS test_op_creation_methods.py) diff --git a/python/paddle/v2/framework/tests/op_test_util.py b/python/paddle/v2/framework/tests/op_test_util.py index 9ee66c2c5103811519c3a2c28653536f97009161..e6bc7d8a9b5ddd4582a5ef8a47cb63a7e5911892 100644 --- a/python/paddle/v2/framework/tests/op_test_util.py +++ b/python/paddle/v2/framework/tests/op_test_util.py @@ -33,23 +33,28 @@ class OpTestMeta(type): for place in places: for in_name in func.all_input_args: - if hasattr(self, in_name): + if hasattr(self, "inputs") and in_name in self.inputs: kwargs[in_name] = in_name var = scope.new_var(in_name).get_tensor() - arr = getattr(self, in_name) + arr = self.inputs[in_name] var.set_dims(arr.shape) var.set(arr, place) else: kwargs[in_name] = "@EMPTY@" for out_name in func.all_output_args: - if hasattr(self, out_name): - kwargs[out_name] = out_name - scope.new_var(out_name).get_tensor() + if not hasattr(self, "outputs"): + raise ValueError( + "The test op must set self.outputs dict.") + if out_name not in self.outputs: + raise ValueError("The %s is not in self.outputs dict." % + (out_name)) + kwargs[out_name] = out_name + scope.new_var(out_name).get_tensor() for attr_name in func.all_attr_args: - if hasattr(self, attr_name): - kwargs[attr_name] = getattr(self, attr_name) + if hasattr(self, "attrs") and attr_name in self.attrs: + kwargs[attr_name] = self.attrs[attr_name] op = func(**kwargs) @@ -60,7 +65,7 @@ class OpTestMeta(type): for out_name in func.all_output_args: actual = numpy.array(scope.find_var(out_name).get_tensor()) - expect = getattr(self, out_name) + expect = self.outputs[out_name] numpy.isclose(actual, expect) obj.test_all = test_all diff --git a/python/paddle/v2/framework/tests/test_add_two_op.py b/python/paddle/v2/framework/tests/test_add_two_op.py index 6e6643201bf361fce1bad7de10b2562f0525e00a..8ef48f4727b0af46a696c6f463045d98e7a08800 100644 --- a/python/paddle/v2/framework/tests/test_add_two_op.py +++ b/python/paddle/v2/framework/tests/test_add_two_op.py @@ -12,9 +12,11 @@ class TestAddOp(unittest.TestCase): def setUp(self): self.type = "add_two" - self.X = numpy.random.random((102, 105)).astype("float32") - self.Y = numpy.random.random((102, 105)).astype("float32") - self.Out = self.X + self.Y + self.inputs = { + 'X': numpy.random.random((102, 105)).astype("float32"), + 'Y': numpy.random.random((102, 105)).astype("float32") + } + self.outputs = {'Out': self.inputs['X'] + self.inputs['Y']} class TestAddGradOp(unittest.TestCase): diff --git a/python/paddle/v2/framework/tests/test_cross_entropy_op.py b/python/paddle/v2/framework/tests/test_cross_entropy_op.py index 6d022f6bc0be60dbf2f796780a969bff0e8bfded..b26e25d58b59bd1cb16e9ba2a1cccd27799b15f2 100644 --- a/python/paddle/v2/framework/tests/test_cross_entropy_op.py +++ b/python/paddle/v2/framework/tests/test_cross_entropy_op.py @@ -7,15 +7,17 @@ class TestSGD(unittest.TestCase): __metaclass__ = OpTestMeta def setUp(self): + # TODO this unit test is not passed self.type = "onehot_cross_entropy" batch_size = 100 class_num = 10 - self.X = numpy.random.random((batch_size, class_num)).astype("float32") - self.label = 5 * numpy.ones(batch_size).astype("int32") + X = numpy.random.random((batch_size, class_num)).astype("float32") + label = 5 * numpy.ones(batch_size).astype("int32") + self.inputs = {'X': X, 'label': label} Y = [] for i in range(0, batch_size): - Y.append(-numpy.log(self.X[i][self.label[i]])) - self.Y = numpy.array(Y).astype("float32") + Y.append(-numpy.log(X[i][label[i]])) + self.outputs = {'Y': numpy.array(Y).astype("float32")} # TODO(superjom) add gradient check diff --git a/python/paddle/v2/framework/tests/test_mean_op.py b/python/paddle/v2/framework/tests/test_mean_op.py index 78fff1eeff998109a51ea662f963a102eff49d3a..b5d52b90567bcd0c9f376147145d8638049f7bab 100644 --- a/python/paddle/v2/framework/tests/test_mean_op.py +++ b/python/paddle/v2/framework/tests/test_mean_op.py @@ -8,8 +8,8 @@ class TestMeanOp(unittest.TestCase): def setUp(self): self.type = "mean" - self.X = np.random.random((32, 784)).astype("float32") - self.Out = np.mean(self.X) + self.inputs = {'X': np.random.random((32, 784)).astype("float32")} + self.outputs = {'Out': np.mean(self.inputs['X'])} if __name__ == '__main__': diff --git a/python/paddle/v2/framework/tests/test_mul_op.py b/python/paddle/v2/framework/tests/test_mul_op.py index e1ac66d3a4d23d617f7c5a4d97d070b2660954c8..ec0ac99156a546dd3fb7b27778032bece38ab5a9 100644 --- a/python/paddle/v2/framework/tests/test_mul_op.py +++ b/python/paddle/v2/framework/tests/test_mul_op.py @@ -8,9 +8,11 @@ class TestMulOp(unittest.TestCase): def setUp(self): self.type = "mul" - self.X = np.random.random((32, 84)).astype("float32") - self.Y = np.random.random((84, 100)).astype("float32") - self.Out = np.dot(self.X, self.Y) + self.inputs = { + 'X': np.random.random((32, 84)).astype("float32"), + 'Y': np.random.random((84, 100)).astype("float32") + } + self.outputs = {'Out': np.dot(self.inputs['X'], self.inputs['Y'])} if __name__ == '__main__': diff --git a/python/paddle/v2/framework/tests/test_recurrent_op.py b/python/paddle/v2/framework/tests/test_recurrent_op.py index 0457e3f16a709140180ce433c1d56d146f0b6974..5c77c477b347f4713e4af2a8cb462b243d7a779c 100644 --- a/python/paddle/v2/framework/tests/test_recurrent_op.py +++ b/python/paddle/v2/framework/tests/test_recurrent_op.py @@ -1,3 +1,4 @@ +import logging import paddle.v2.framework.core as core import unittest import numpy as np @@ -7,10 +8,9 @@ ops = creation.op_creations def create_tensor(scope, name, shape): - tensor = scope.create_var(name).get_tensor() + tensor = scope.new_var(name).get_tensor() tensor.set_dims(shape) - tensor.alloc_float() - tensor.set(np.random.random(shape)) + tensor.set(np.random.random(shape), core.CPUPlace()) return tensor @@ -31,40 +31,36 @@ class TestRNN(unittest.TestCase): - h ''' + input_dim = 30 + batch_size = 50 + weight_dim = 15 + sent_len = 11 + def init(self): - input_dim = 30 - batch_size = 50 - weight_dim = 15 - - self.scope = core.Scope(None) - - # create vars - create_tensor(self.scope, "x", [batch_size, input_dim]) - create_tensor(self.scope, "W", [input_dim, weight_dim]) - create_tensor(self.scope, "U", [weight_dim, weight_dim]) - create_tensor(self.scope, "h_boot", [batch_size, weight_dim]) - - x_alias = "x@alias" - y_alias = "y@alias" - memory = "h@alias" - prememory = "h@pre" - output = "rnn_out" - output_alias = "rnn_out@alias" - - # create step net - stepnet_var = self.scope.create_var("stepnet") - stepnet = stepnet_var.get_net() - # stepnet = core.Net.create() - x_fc_op = ops.fc(X=x_alias, W="W", Y="Wx") - h_fc_op = ops.fc(X=prememory, W="U", Y="Uh") - sum_op = ops.add_two(X="Wx", Y="Uh", Out="sum") - sig_op = ops.sigmoid(X="sum", Y=memory) - stepnet.add_op(x_fc_op) - stepnet.add_op(h_fc_op) - stepnet.add_op(sum_op) - stepnet.add_op(sig_op) - stepnet.complete_add_op(True) + self.scope = core.Scope() + + self.create_global_variables() + self.create_step_net() + rnn_op = self.create_rnn_op() + ctx = core.DeviceContext.create(core.CPUPlace()) + print 'infer_shape' + rnn_op.infer_shape(self.scope) + + rnn_op.run(self.scope, ctx) + + def create_global_variables(self): + # create inlink + create_tensor(self.scope, "x", + [self.sent_len, self.batch_size, self.input_dim]) + create_tensor(self.scope, "W", [self.input_dim, self.input_dim]) + create_tensor(self.scope, "U", [self.input_dim, self.input_dim]) + create_tensor(self.scope, "h_boot", [self.batch_size, self.input_dim]) + self.scope.new_var("step_scopes") + self.scope.new_var("h@alias") + self.scope.new_var("h") + + def create_rnn_op(self): # create RNNOp rnnop = ops.recurrent_op( # inputs @@ -72,17 +68,27 @@ class TestRNN(unittest.TestCase): boot_memories=["h_boot"], step_net="stepnet", # outputs - outlinks=[output], + outlinks=["h"], step_scopes="step_scopes", # attributes inlink_alias=["x@alias"], - outlink_alias=[output_alias], - pre_memories=[prememory], - memories=[memory]) + outlink_alias=["h@alias"], + pre_memories=["h@pre"], + memories=["h@alias"]) + return rnnop + + def create_step_net(self): + var = self.scope.new_var("stepnet") + stepnet = var.get_net() - ctx = core.DeviceContext.cpu_context() - rnnop.infer_shape(self.scope) - rnnop.run(self.scope, ctx) + x_fc_op = ops.fc(X="x@alias", W="W", Y="Wx") + h_fc_op = ops.fc(X="h@pre", W="U", Y="Uh") + sum_op = ops.add_two(X="Wx", Y="Uh", Out="sum") + sig_op = ops.sigmoid(X="sum", Y="h@alias") + + for op in [x_fc_op, h_fc_op, sum_op, sig_op]: + stepnet.add_op(op) + stepnet.complete_add_op(True) def test_recurrent(self): self.init() diff --git a/python/paddle/v2/framework/tests/test_rowwise_add_op.py b/python/paddle/v2/framework/tests/test_rowwise_add_op.py index 04abc14ee198fe4e2307e009c696a2b40ec271b6..f8521eb517057fbeb104b28af7da4fffe54f37de 100644 --- a/python/paddle/v2/framework/tests/test_rowwise_add_op.py +++ b/python/paddle/v2/framework/tests/test_rowwise_add_op.py @@ -8,9 +8,11 @@ class TestRowwiseAddOp(unittest.TestCase): def setUp(self): self.type = "rowwise_add" - self.X = np.random.random((32, 84)).astype("float32") - self.b = np.random.random(84).astype("float32") - self.Out = np.add(self.X, self.b) + self.inputs = { + 'X': np.random.random((32, 84)).astype("float32"), + 'b': np.random.random(84).astype("float32") + } + self.outputs = {'Out': np.add(self.inputs['X'], self.inputs['b'])} if __name__ == '__main__': diff --git a/python/paddle/v2/framework/tests/test_sgd_op.py b/python/paddle/v2/framework/tests/test_sgd_op.py index ca03cc11abe2ceb31b33a87797aa752943dd2a7d..e5f9ef865e84f1a78e28884ad7e2e758f9ca8054 100644 --- a/python/paddle/v2/framework/tests/test_sgd_op.py +++ b/python/paddle/v2/framework/tests/test_sgd_op.py @@ -8,10 +8,13 @@ class TestSGD(unittest.TestCase): def setUp(self): self.type = "sgd" - self.param = numpy.random.random((102, 105)).astype("float32") - self.grad = numpy.random.random((102, 105)).astype("float32") - self.learning_rate = 0.1 - self.param_out = self.param - self.learning_rate * self.grad + w = numpy.random.random((102, 105)).astype("float32") + g = numpy.random.random((102, 105)).astype("float32") + lr = 0.1 + + self.inputs = {'param': w, 'grad': g} + self.attrs = {'learning_rate': lr} + self.outputs = {'param_out': w - lr * g} if __name__ == "__main__": diff --git a/python/paddle/v2/framework/tests/test_sigmoid_op.py b/python/paddle/v2/framework/tests/test_sigmoid_op.py index 50044a122f1d66dd54a24f6cce76074a60ee2262..2a57a41ed8b718fd420062ba68e853a4861b7359 100644 --- a/python/paddle/v2/framework/tests/test_sigmoid_op.py +++ b/python/paddle/v2/framework/tests/test_sigmoid_op.py @@ -8,9 +8,12 @@ class TestSigmoidOp(unittest.TestCase): def setUp(self): self.type = "sigmoid" - self.X = np.random.random((32, 100)).astype("float32") - self.Y = 1 / (1 + np.exp(-self.X)) + self.inputs = {'X': np.random.random((32, 100)).astype("float32")} + self.outputs = {'Y': 1 / (1 + np.exp(-self.inputs['X']))} +#class TestSigmoidGradOp(unittest.TestCase): +#TODO(qingqing) add unit test + if __name__ == '__main__': unittest.main() diff --git a/python/paddle/v2/framework/tests/test_softmax_op.py b/python/paddle/v2/framework/tests/test_softmax_op.py index c80888128781d98e4ed30d845a30b39121f66459..98ca8ddc860c3825411b02b2f6ed612db46a18d7 100644 --- a/python/paddle/v2/framework/tests/test_softmax_op.py +++ b/python/paddle/v2/framework/tests/test_softmax_op.py @@ -19,8 +19,10 @@ class TestSoftmaxOp(unittest.TestCase): def setUp(self): self.type = "softmax" - self.X = np.random.random((32, 100)).astype("float32") - self.Y = np.apply_along_axis(stable_softmax, 1, self.X) + self.inputs = {'X': np.random.random((32, 100)).astype("float32")} + self.outputs = { + 'Y': np.apply_along_axis(stable_softmax, 1, self.inputs['X']) + } class TestSoftmaxGradOp(unittest.TestCase): diff --git a/python/paddle/v2/plot/tests/CMakeLists.txt b/python/paddle/v2/plot/tests/CMakeLists.txt index da5cd764889b48a3af8461a2793d948aa609d6c1..4b6c1c80969182ccf6e0189b18bade8758bbbc30 100644 --- a/python/paddle/v2/plot/tests/CMakeLists.txt +++ b/python/paddle/v2/plot/tests/CMakeLists.txt @@ -1,5 +1,5 @@ if (NOT APPLE) # The Mac OS X backend will not be able to function correctly if Python is # not installed as a framework. - add_python_test(test_ploter test_ploter.py) + py_test(test_ploter SRCS test_ploter.py) endif() diff --git a/python/paddle/v2/reader/tests/CMakeLists.txt b/python/paddle/v2/reader/tests/CMakeLists.txt index 6a1d337b232c7a849a8793894bf16d26d609d3dd..107d5912e1567e0c8721987a281272c7feb51e63 100644 --- a/python/paddle/v2/reader/tests/CMakeLists.txt +++ b/python/paddle/v2/reader/tests/CMakeLists.txt @@ -1 +1,2 @@ -add_python_test(reader_tests creator_test.py decorator_test.py) +py_test(creator_test SRCS creator_test.py) +py_test(decorator_test SRCS decorator_test.py) diff --git a/python/paddle/v2/tests/CMakeLists.txt b/python/paddle/v2/tests/CMakeLists.txt index 058f22befd0657d06ff130ace55fe7322148213d..b7791559594321a85f41b508b69efeb077d69595 100644 --- a/python/paddle/v2/tests/CMakeLists.txt +++ b/python/paddle/v2/tests/CMakeLists.txt @@ -1,2 +1,7 @@ -add_python_test(test_v2_api test_data_feeder.py test_op.py test_parameters.py -test_layer.py test_rnn_layer.py test_topology.py test_image.py) +py_test(test_op SRCS test_op.py) +py_test(test_image SRCS test_image.py) +py_test(test_layer SRCS test_layer.py) +py_test(test_topology SRCS test_topology.py) +py_test(test_rnn_layer SRCS test_rnn_layer.py) +py_test(test_parameters SRCS test_parameters.py) +py_test(test_data_feeder SRCS test_data_feeder.py)