m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge branch 'develop' into Paddle-ReduceProd

Browse Source
shanyi15-patch-2
zhouhanqing 7 years ago
parent 732eebb286 90215b7844
commit 3ca968441d
51 changed files with 1622 additions and 547 deletions
Show Stats Download Patch File Download Diff File

2
.travis.yml
Unescape View File

@ -56,7 +56,7 @@ script:
export DEPLOY_DOCS_SH=https://raw.githubusercontent.com/PaddlePaddle/PaddlePaddle.org/master/scripts/deploy/deploy_docs.sh
export DOCS_DIR=`pwd`
cd ..
curl $DEPLOY_DOCS_SH | bash -s $CONTENT_DEC_PASSWD $TRAVIS_BRANCH $DOCS_DIR $DOCS_DIR/build/doc/v2
curl $DEPLOY_DOCS_SH | bash -s $CONTENT_DEC_PASSWD $TRAVIS_BRANCH $DOCS_DIR $DOCS_DIR/build/doc/
notifications:
email:
on_success: change

8
cmake/external/openblas.cmake vendored
Unescape View File

@ -77,7 +77,8 @@ IF(NOT ${CBLAS_FOUND})
INSTALL_DIR ${CBLAS_INSTALL_DIR}
BUILD_IN_SOURCE 1
BUILD_COMMAND ${CMAKE_MAKE_PROGRAM} ${COMMON_ARGS} ${OPTIONAL_ARGS}
INSTALL_COMMAND ${CMAKE_MAKE_PROGRAM} install NO_SHARED=1 NO_LAPACK=1 PREFIX=<INSTALL_DIR>
INSTALL_COMMAND ${CMAKE_MAKE_PROGRAM} install NO_SHARED=1 NO_LAPACK=1 PREFIX=<INSTALL_DIR>
&& rm -r ${CBLAS_INSTALL_DIR}/lib/cmake ${CBLAS_INSTALL_DIR}/lib/pkgconfig
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
)
@ -100,11 +101,6 @@ IF(NOT ${CBLAS_FOUND})
\"${CBLAS_INSTALL_DIR}/lib -> ${CMAKE_INSTALL_PREFIX}/${TMP_INSTALL_DIR}\"
)"
)
INSTALL(CODE "execute_process(
COMMAND rm -r ${CMAKE_INSTALL_PREFIX}/${TMP_INSTALL_DIR}/cmake
${CMAKE_INSTALL_PREFIX}/${TMP_INSTALL_DIR}/pkgconfig
)"
)
ENDIF()
ENDIF(NOT ${CBLAS_FOUND})

5
cmake/generic.cmake
Unescape View File

@ -186,7 +186,9 @@ function(cc_library TARGET_NAME)
add_library(${TARGET_NAME} SHARED ${cc_library_SRCS})
else()
add_library(${TARGET_NAME} STATIC ${cc_library_SRCS})
find_fluid_modules(${TARGET_NAME})
endif()
if(cc_library_DEPS)
# Don't need link libwarpctc.so
if("${cc_library_DEPS};" MATCHES "warpctc;")
@ -263,7 +265,8 @@ function(nv_library TARGET_NAME)
if (nv_library_SHARED OR nv_library_shared) # build *.so
cuda_add_library(${TARGET_NAME} SHARED ${nv_library_SRCS})
else()
cuda_add_library(${TARGET_NAME} STATIC ${nv_library_SRCS})
cuda_add_library(${TARGET_NAME} STATIC ${nv_library_SRCS})
find_fluid_modules(${TARGET_NAME})
endif()
if (nv_library_DEPS)
add_dependencies(${TARGET_NAME} ${nv_library_DEPS})

35
cmake/inference_lib.cmake
Unescape View File

@ -1,9 +1,22 @@
set_property(GLOBAL PROPERTY FLUID_MODULES "")
# find all fluid modules is used for paddle fluid static library
function(find_fluid_modules TARGET_NAME)
get_filename_component(__target_path ${TARGET_NAME} ABSOLUTE)
string(FIND "${__target_path}" "fluid" pos)
if(pos GREATER 1)
get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
set(fluid_modules ${fluid_modules} ${TARGET_NAME})
set_property(GLOBAL PROPERTY FLUID_MODULES "${fluid_modules}")
endif()
endfunction(find_fluid_modules)
# make package for paddle fluid shared and static library
function(copy TARGET)
set(options "")
set(oneValueArgs "")
set(multiValueArgs SRCS DSTS DEPS)
cmake_parse_arguments(copy_lib "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
set(inference_lib_dist_dep ${TARGET} ${inference_lib_dist_dep} PARENT_SCOPE)
list(LENGTH copy_lib_SRCS copy_lib_SRCS_len)
list(LENGTH copy_lib_DSTS copy_lib_DSTS_len)
@ -42,13 +55,21 @@ copy(glog_lib
DSTS ${dst_dir} ${dst_dir}/lib
)
IF(NOT PROTOBUF_FOUND)
if(NOT PROTOBUF_FOUND)
set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/protobuf")
copy(protobuf_lib
SRCS ${PROTOBUF_INCLUDE_DIR} ${PROTOBUF_LITE_LIBRARY}
SRCS ${PROTOBUF_INCLUDE_DIR} ${PROTOBUF_LIBRARY}
DSTS ${dst_dir} ${dst_dir}/lib
)
ENDIF(NOT PROTOBUF_FOUND)
endif()
if(NOT CBLAS_FOUND)
set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/openblas")
copy(openblas_lib
SRCS ${CBLAS_INSTALL_DIR}/lib ${CBLAS_INSTALL_DIR}/include
DSTS ${dst_dir} ${dst_dir}
)
endif()
# paddle fluid module
set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
@ -66,8 +87,8 @@ copy(memory_lib
)
set(module "inference")
copy(inference_lib DEPENDS paddle_fluid_shared
SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.so
copy(inference_lib DEPS paddle_fluid_shared paddle_fluid
SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.*
DSTS ${dst_dir}/${module} ${dst_dir}/${module}
)
@ -83,6 +104,4 @@ copy(string_lib
DSTS ${dst_dir}/${module} ${dst_dir}/${module}/tinyformat
)
add_custom_target(inference_lib_dist DEPENDS
inference_lib framework_lib memory_lib platform_lib string_lib
gflags_lib glog_lib protobuf_lib eigen3_lib)
add_custom_target(inference_lib_dist DEPENDS ${inference_lib_dist_dep})

0
doc/v2/howto/optimization/pprof_1.png → doc/fluid/howto/optimization/pprof_1.png
Unescape View File

Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 344 KiB

After

Width:  |  Height:  |  Size: 344 KiB

0
doc/v2/howto/optimization/pprof_2.png → doc/fluid/howto/optimization/pprof_2.png
Unescape View File

Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 190 KiB

After

Width:  |  Height:  |  Size: 190 KiB

BIN
doc/fluid/howto/optimization/timeline.jpeg
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 69 KiB

27
doc/fluid/howto/optimization/timeline.md
Unescape View File

@ -0,0 +1,27 @@
## how to use timeline tool to do profile
1. Add `with profiler.profiler(...)` to the main training loop. After run, the code will generate a profile record file `/tmp/profile`. **Warning**: Please do not run too many batches when use profiler to record timeline information, for the profile record will grow with the batch number.
```python
with profiler.profiler('All', 'total', '/tmp/profile') as prof:
for pass_id in range(pass_num):
for batch_id, data in enumerate(train_reader()):
exe.run(fluid.default_main_program(),
feed=feeder.feed(data),
fetch_list=[],
use_program_cache=True)
...
```
1. Run `python paddle/tools/timeline.py` to process `/tmp/profile`, it will generate another
file `/tmp/timeline` by default. You can change the path by cmd parameter, please take a look at
[timeline.py](https://github.com/PaddlePaddle/Paddle/blob/develop/tools/timeline.py) for details.
1. Open chrome and visit <chrome://tracing/>, use `load` button to load the generated `timeline` file.
![chrome tracing](./tracing.jpeg)
1. The resulting timeline should be like:
![chrome timeline](./timeline.jpeg)

BIN
doc/fluid/howto/optimization/tracing.jpeg
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 30 KiB

30
doc/fluid/read_source.md
Unescape View File

@ -2,17 +2,17 @@
Examples: https://github.com/PaddlePaddle/Paddle/tree/develop/python/paddle/fluid/tests/book
Core: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/framework
Core: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid/framework
Operator: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators
Operator: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid/operators
Memory: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/memory
Memory: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid/memory
Platform: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/platform
Platform: https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid/platform
# Compile Time
The following **defines** the NN. The definition goes into this [protocol buffer](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto).
The following **defines** the NN. The definition goes into this [protocol buffer](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/framework.proto).
```python
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
@ -29,10 +29,10 @@ sgd_optimizer.minimize(avg_cost)
- Variables: `x`, `y`, `y_predict`, `cost` and `avg_cost`. [Python](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/framework.py#)
- Layers: `fluid.layers.data`, `fluid.layers.fc` and `fluid.layers.mean` are layers. [Python](https://github.com/PaddlePaddle/Paddle/tree/develop/python/paddle/fluid/layers)
- Every Layer has one or more operators and variables/parameters
- All the operators are defined at [`paddle/operators/`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators). Other worth-looking files:
- Base class: [`paddle/framework/operator.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h)
- Operator Registration: [`paddle/framework/op_registry.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/op_registry.h)
- Operator Lookup: [`paddle/framework/op_info.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/op_info.h)
- All the operators are defined at [`paddle/fluid/operators/`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid/operators). Other worth-looking files:
- Base class: [`paddle/fluid/framework/operator.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/operator.h)
- Operator Registration: [`paddle/fluid/framework/op_registry.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/op_registry.h)
- Operator Lookup: [`paddle/fluid/framework/op_info.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/op_info.h)
- Optimizer: `fluid.optimizer.SGD`. It does the following
- Add backward operators. [[Python](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/backward.py)]
- Add optimizer operators. [[Python](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/optimizer.py)]
@ -55,13 +55,13 @@ exe.run(fluid.default_main_program(),
fetch_list=[avg_cost])
```
- Place: `place`. one of CPU, GPU or FPGA. [C++](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/place.h)
- The device handle are at [paddle/platform/device_context.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/device_context.h)
- Executor: `fluid.Executor(place)`. [[Python](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/executor.py), [C++](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/executor.cc)]
- Place: `place`. one of CPU, GPU or FPGA. [C++](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/platform/place.h)
- The device handle are at [paddle/fluid/platform/device_context.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/platform/device_context.h)
- Executor: `fluid.Executor(place)`. [[Python](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/executor.py), [C++](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/executor.cc)]
- Feeds the data: `feed=feeder.feed(data)`
- Evaluates all the operators
- Fetches the result: `fetch_list=[avg_cost]`
- Other worth looking files:
- Scope: [paddle/framework/scope.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/scope.h). Where all the variables live
- Variable: [paddle/framework/variable.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/variable.h). Where all the data (most likely tensors) live
- Tensor: [paddle/framework/tensor.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/tensor.h). Where we allocate memory through [`paddle/memory/`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/memory)
- Scope: [paddle/fluid/framework/scope.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/scope.h). Where all the variables live
- Variable: [paddle/fluid/framework/variable.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/variable.h). Where all the data (most likely tensors) live
- Tensor: [paddle/fluid/framework/tensor.h](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/framework/tensor.h). Where we allocate memory through [`paddle/fluid/memory/`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/fluid/memory)

2
doc/v2/build_and_install/pip_install_cn.rst
Unescape View File

@ -39,7 +39,7 @@ PaddlePaddle可以使用常用的Python包管理工具
"cpu_avx_mkl", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddle.tgz>`_"
"cpu_avx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "暂无"
"cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "暂无"
"cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddle.tgz>`_"
"cuda7.5_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddle.tgz>`_"
"cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddle.tgz>`_"
"cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddle.tgz>`_"

2
doc/v2/build_and_install/pip_install_en.rst
Unescape View File

@ -42,7 +42,7 @@ If the links below shows up the login form, just click "Log in as guest" to star
"cpu_avx_mkl", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddle.tgz>`_"
"cpu_avx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "Not Available"
"cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "Not Available"
"cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddle.tgz>`_"
"cuda7.5_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddle.tgz>`_"
"cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddle.tgz>`_"
"cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_", "`paddle.tgz <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddle.tgz>`_"

26
doc/v2/howto/index_cn.rst
Unescape View File

@ -1,11 +1,37 @@
进阶使用
========
PaddlePaddle支持用户灵活地设置各种命令行参数，以实现对模型训练或预测流程的控制。使用方式请参考：
.. toctree::
:maxdepth: 1
cmd_parameter/index_cn.rst
PaddlePaddle支持在fabric集群、MPI集群、kubernetes集群上分布式训练任务，具体环境配置和使用说明请参考：
.. toctree::
:maxdepth: 1
cluster/index_cn.rst
PaddlePaddle提供了用于预测的C-API，关于C-API的使用，我们提供了如下指南:
.. toctree::
:maxdepth: 1
capi/index_cn.rst
PaddlePaddle支持多种灵活和高效的循环神经网络，具体配置使用方式请参考：
.. toctree::
:maxdepth: 1
rnn/index_cn.rst
关于如何使用内置的定时工具、nvprof 或 nvvp 来运行性能分析和调优，请参考：
.. toctree::
:maxdepth: 1
optimization/gpu_profiling_cn.rst

24
doc/v2/howto/rnn/index_cn.rst
Unescape View File

@ -1,10 +1,34 @@
RNN模型
===========
循环神经网络（RNN）是对序列数据建模的重要工具。PaddlePaddle提供了灵活的接口以支持复杂循环神经网络的构建。
这里将分为以下四个部分详细介绍如何使用PaddlePaddle搭建循环神经网络。
第一部分由浅入深的展示了使用PaddlePaddle搭建循环神经网络的全貌：首先以简单的循环神经网络（vanilla RNN）为例，
说明如何封装配置循环神经网络组件；然后更进一步的通过序列到序列（sequence to sequence）模型，逐步讲解如何构建完整而复杂的循环神经网络模型。
.. toctree::
:maxdepth: 1
rnn_config_cn.rst
Recurrent Group是PaddlePaddle中实现复杂循环神经网络的关键，第二部分阐述了PaddlePaddle中Recurrent Group的相关概念和原理，
对Recurrent Group接口进行了详细说明。另外，对双层RNN（对应的输入为双层序列）及Recurrent Group在其中的使用进行了介绍。
.. toctree::
:maxdepth: 1
recurrent_group_cn.md
第三部分对双层序列进行了解释说明，列出了PaddlePaddle中支持双层序列作为输入的Layer，并对其使用进行了逐一介绍。
.. toctree::
:maxdepth: 1
hierarchical_layer_cn.rst
第四部分以PaddlePaddle的双层RNN单元测试中的网络配置为示例，辅以效果相同的单层RNN网络配置作为对比，讲解了多种情况下双层RNN的使用。
.. toctree::
:maxdepth: 1
hrnn_rnn_api_compare_cn.rst

1
paddle/fluid/framework/data_type_transform.cc
Unescape View File

@ -53,6 +53,7 @@ struct CastDataType {
auto* context = static_cast<const platform::CUDADeviceContext*>(ctx_);
trans(*context, in_begin, in_end, out_begin,
CastDataTypeFunctor<InType, OutType>());
context->Wait();
#endif
} else {
PADDLE_THROW("Unsupported place!");

24
paddle/fluid/framework/data_type_transform_test.cc
Unescape View File

@ -50,13 +50,13 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_fp32, kernel_fp64, in, &out);
double* out_data_double = out.data<double>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_double[i], static_cast<double>(i / 3));
EXPECT_EQ(out_data_double[i], static_cast<double>(i / 3));
}
TransDataType(kernel_fp32, kernel_int32, in, &out);
int* out_data_int = out.data<int>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_int[i], static_cast<int>(i / 3));
EXPECT_EQ(out_data_int[i], static_cast<int>(i / 3));
}
}
@ -76,31 +76,31 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_fp16, kernel_fp32, in, &out);
float* out_data_float = out.data<float>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_float[i], static_cast<float>(ptr[i]));
EXPECT_EQ(out_data_float[i], static_cast<float>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_fp64, in, &out);
double* out_data_double = out.data<double>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_double[i], static_cast<double>(ptr[i]));
EXPECT_EQ(out_data_double[i], static_cast<double>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_int32, in, &out);
int* out_data_int = out.data<int>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_int[i], static_cast<int>(ptr[i]));
EXPECT_EQ(out_data_int[i], static_cast<int>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_int64, in, &out);
int64_t* out_data_int64 = out.data<int64_t>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_int64[i], static_cast<int64_t>(ptr[i]));
EXPECT_EQ(out_data_int64[i], static_cast<int64_t>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_bool, in, &out);
bool* out_data_bool = out.data<bool>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_bool[i], static_cast<bool>(ptr[i]));
EXPECT_EQ(out_data_bool[i], static_cast<bool>(ptr[i]));
}
// transform float to float16
@ -112,7 +112,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_fp32, kernel_fp16, in, &out);
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_float[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_float[i]).x);
}
// transform double to float16
@ -124,7 +124,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_fp64, kernel_fp16, in, &out);
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_double[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_double[i]).x);
}
// transform int to float16
@ -136,7 +136,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_int32, kernel_fp16, in, &out);
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_int[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_int[i]).x);
}
// transform int64 to float16
@ -148,7 +148,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_int64, kernel_fp16, in, &out);
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_int64[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_int64[i]).x);
}
// transform bool to float16
@ -160,7 +160,7 @@ TEST(DataTypeTransform, CPUTransform) {
TransDataType(kernel_bool, kernel_fp16, in, &out);
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_bool[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_bool[i]).x);
}
}
}

33
paddle/fluid/framework/data_type_transform_test.cu
Unescape View File

@ -49,15 +49,16 @@ TEST(DataTypeTransform, GPUTransform) {
float arr[6] = {0, 1, 2, 3, 4, 5};
int data_number = sizeof(arr) / sizeof(arr[0]);
memcpy(in_ptr, arr, sizeof(arr));
TensorCopy(in, gpu_place, context, &in_gpu);
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
TransDataType(kernel_fp32, kernel_fp64, in_gpu, &out_gpu);
TensorCopy(out_gpu, cpu_place, context, &out);
context.Wait();
double* out_data_double = out.data<double>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_double[i], static_cast<double>(arr[i]));
EXPECT_EQ(out_data_double[i], static_cast<double>(arr[i]));
}
TransDataType(kernel_fp32, kernel_int32, in_gpu, &out_gpu);
@ -66,7 +67,7 @@ TEST(DataTypeTransform, GPUTransform) {
int* out_data_int = out.data<int>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_int[i], static_cast<int>(arr[i]));
EXPECT_EQ(out_data_int[i], static_cast<int>(arr[i]));
}
}
@ -83,6 +84,7 @@ TEST(DataTypeTransform, GPUTransform) {
int data_number = sizeof(arr) / sizeof(arr[0]);
memcpy(ptr, arr, sizeof(arr));
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
// transform from float16 to other data types
TransDataType(kernel_fp16, kernel_fp32, in_gpu, &out_gpu);
@ -91,7 +93,7 @@ TEST(DataTypeTransform, GPUTransform) {
float* out_data_float = out.data<float>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_float[i], static_cast<float>(ptr[i]));
EXPECT_EQ(out_data_float[i], static_cast<float>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_fp64, in_gpu, &out_gpu);
@ -100,7 +102,7 @@ TEST(DataTypeTransform, GPUTransform) {
double* out_data_double = out.data<double>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_double[i], static_cast<double>(ptr[i]));
EXPECT_EQ(out_data_double[i], static_cast<double>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_int32, in_gpu, &out_gpu);
@ -109,7 +111,7 @@ TEST(DataTypeTransform, GPUTransform) {
int* out_data_int = out.data<int>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_int[i], static_cast<int>(ptr[i]));
EXPECT_EQ(out_data_int[i], static_cast<int>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_int64, in_gpu, &out_gpu);
@ -118,7 +120,7 @@ TEST(DataTypeTransform, GPUTransform) {
int64_t* out_data_int64 = out.data<int64_t>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_int64[i], static_cast<int64_t>(ptr[i]));
EXPECT_EQ(out_data_int64[i], static_cast<int64_t>(ptr[i]));
}
TransDataType(kernel_fp16, kernel_bool, in_gpu, &out_gpu);
@ -127,7 +129,7 @@ TEST(DataTypeTransform, GPUTransform) {
bool* out_data_bool = out.data<bool>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(out_data_bool[i], static_cast<bool>(ptr[i]));
EXPECT_EQ(out_data_bool[i], static_cast<bool>(ptr[i]));
}
// transform float to float16
@ -137,13 +139,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
TransDataType(kernel_fp32, kernel_fp16, in_gpu, &out_gpu);
TensorCopy(out_gpu, cpu_place, context, &out);
context.Wait();
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_float[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_float[i]).x);
}
// transform double to float16
@ -154,13 +157,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
TransDataType(kernel_fp64, kernel_fp16, in_gpu, &out_gpu);
TensorCopy(out_gpu, cpu_place, context, &out);
context.Wait();
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_double[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_double[i]).x);
}
// transform int to float16
@ -170,13 +174,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
TransDataType(kernel_int32, kernel_fp16, in_gpu, &out_gpu);
TensorCopy(out_gpu, cpu_place, context, &out);
context.Wait();
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_int[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_int[i]).x);
}
// transform int64 to float16
@ -187,13 +192,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
TransDataType(kernel_int64, kernel_fp16, in_gpu, &out_gpu);
TensorCopy(out_gpu, cpu_place, context, &out);
context.Wait();
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_int64[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_int64[i]).x);
}
// transform bool to float16
@ -203,13 +209,14 @@ TEST(DataTypeTransform, GPUTransform) {
}
TensorCopy(in, gpu_place, context, &in_gpu);
context.Wait();
TransDataType(kernel_bool, kernel_fp16, in_gpu, &out_gpu);
TensorCopy(out_gpu, cpu_place, context, &out);
context.Wait();
ptr = out.data<float16>();
for (int i = 0; i < data_number; ++i) {
ASSERT_EQ(ptr[i].x, static_cast<float16>(in_data_bool[i]).x);
EXPECT_EQ(ptr[i].x, static_cast<float16>(in_data_bool[i]).x);
}
}
}

3
paddle/fluid/framework/executor.cc
Unescape View File

@ -125,8 +125,9 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
for (auto& op_desc : block.AllOps()) {
auto op = paddle::framework::OpRegistry::CreateOp(*op_desc);
VLOG(3) << place_ << " " << op->DebugStringEx(local_scope);
VLOG(4) << place_ << " " << op->DebugStringEx(local_scope);
op->Run(*local_scope, place_);
VLOG(3) << place_ << " " << op->DebugStringEx(local_scope);
if (FLAGS_benchmark) {
VLOG(2) << "Memory used after operator " + op->Type() + " running: "

1
paddle/fluid/framework/tensor_util.cc
Unescape View File

@ -187,7 +187,6 @@ bool TensorContainsInf(const framework::Tensor& tensor) {
void TensorToStream(std::ostream& os, const Tensor& tensor,
const platform::DeviceContext& dev_ctx) {
// TODO(typhoonzero): serialize to ostream
{ // the 1st field, uint32_t version
constexpr uint32_t version = 0;
os.write(reinterpret_cast<const char*>(&version), sizeof(version));

3
paddle/fluid/inference/CMakeLists.txt
Unescape View File

@ -5,7 +5,8 @@ cc_library(paddle_fluid_api
DEPS ${FLUID_CORE_MODULES} ${GLOB_OP_LIB})
# Create static library
cc_library(paddle_fluid DEPS paddle_fluid_api ${FLUID_CORE_MODULES} ${GLOB_OP_LIB})
get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
cc_library(paddle_fluid DEPS ${fluid_modules})
# Create shared library
cc_library(paddle_fluid_shared SHARED

15
paddle/fluid/inference/io.cc
Unescape View File

@ -22,14 +22,14 @@ namespace paddle {
namespace inference {
void ReadBinaryFile(const std::string& filename, std::string& contents) {
VLOG(3) << "loading model from " << filename;
std::ifstream inputfs(filename, std::ios::in | std::ios::binary);
inputfs.seekg(0, std::ios::end);
std::ifstream fin(filename, std::ios::in | std::ios::binary);
PADDLE_ENFORCE(static_cast<bool>(fin), "Cannot open file %s", filename);
fin.seekg(0, std::ios::end);
contents.clear();
contents.resize(inputfs.tellg());
inputfs.seekg(0, std::ios::beg);
inputfs.read(&contents[0], contents.size());
inputfs.close();
contents.resize(fin.tellg());
fin.seekg(0, std::ios::beg);
fin.read(&contents[0], contents.size());
fin.close();
}
bool IsPersistable(const framework::VarDesc* var) {
@ -97,6 +97,7 @@ std::unique_ptr<framework::ProgramDesc> Load(framework::Executor& executor,
const std::string& dirname) {
std::string model_filename = dirname + "/__model__";
std::string program_desc_str;
VLOG(3) << "loading model from " << model_filename;
ReadBinaryFile(model_filename, program_desc_str);
std::unique_ptr<framework::ProgramDesc> main_program(

19
paddle/fluid/inference/tests/book/test_inference_image_classification.cc
Unescape View File

@ -17,10 +17,13 @@ limitations under the License. */
#include "paddle/fluid/inference/tests/test_helper.h"
DEFINE_string(dirname, "", "Directory of the inference model.");
DEFINE_int32(batch_size, 1, "Batch size of input data");
DEFINE_int32(repeat, 1, "Running the inference program repeat times");
TEST(inference, image_classification) {
if (FLAGS_dirname.empty()) {
LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model";
if (FLAGS_dirname.empty() || FLAGS_batch_size < 1 || FLAGS_repeat < 1) {
LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model "
"--batch_size=1 --repeat=1";
}
LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl;
@ -29,13 +32,11 @@ TEST(inference, image_classification) {
// 0. Call `paddle::framework::InitDevices()` initialize all the devices
// In unittests, this is done in paddle/testing/paddle_gtest_main.cc
int64_t batch_size = 1;
paddle::framework::LoDTensor input;
// Use normilized image pixels as input data,
// which should be in the range [0.0, 1.0].
SetupTensor<float>(input,
{batch_size, 3, 32, 32},
{FLAGS_batch_size, 3, 32, 32},
static_cast<float>(0),
static_cast<float>(1));
std::vector<paddle::framework::LoDTensor*> cpu_feeds;
@ -46,7 +47,9 @@ TEST(inference, image_classification) {
cpu_fetchs1.push_back(&output1);
// Run inference on CPU
TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
LOG(INFO) << "--- CPU Runs: ---";
TestInference<paddle::platform::CPUPlace>(
dirname, cpu_feeds, cpu_fetchs1, FLAGS_repeat);
LOG(INFO) << output1.dims();
#ifdef PADDLE_WITH_CUDA
@ -55,7 +58,9 @@ TEST(inference, image_classification) {
cpu_fetchs2.push_back(&output2);
// Run inference on CUDA GPU
TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
LOG(INFO) << "--- GPU Runs: ---";
TestInference<paddle::platform::CUDAPlace>(
dirname, cpu_feeds, cpu_fetchs2, FLAGS_repeat);
LOG(INFO) << output2.dims();
CheckError<float>(output1, output2);

83
paddle/fluid/inference/tests/book/test_inference_recognize_digits.cc
Unescape View File

@ -17,10 +17,13 @@ limitations under the License. */
#include "paddle/fluid/inference/tests/test_helper.h"
DEFINE_string(dirname, "", "Directory of the inference model.");
DEFINE_int32(batch_size, 1, "Batch size of input data");
DEFINE_int32(repeat, 1, "Running the inference program repeat times");
TEST(inference, recognize_digits) {
if (FLAGS_dirname.empty()) {
LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model";
if (FLAGS_dirname.empty() || FLAGS_batch_size < 1 || FLAGS_repeat < 1) {
LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model "
"--batch_size=1 --repeat=1";
}
LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl;
@ -29,77 +32,39 @@ TEST(inference, recognize_digits) {
// 0. Call `paddle::framework::InitDevices()` initialize all the devices
// In unittests, this is done in paddle/testing/paddle_gtest_main.cc
int64_t batch_size = 1;
paddle::framework::LoDTensor input;
// Use normilized image pixels as input data,
// which should be in the range [-1.0, 1.0].
SetupTensor<float>(input,
{batch_size, 1, 28, 28},
{FLAGS_batch_size, 1, 28, 28},
static_cast<float>(-1),
static_cast<float>(1));
std::vector<paddle::framework::LoDTensor*> cpu_feeds;
cpu_feeds.push_back(&input);
paddle::framework::LoDTensor output1;
std::vector<paddle::framework::LoDTensor*> cpu_fetchs1;
cpu_fetchs1.push_back(&output1);
for (auto is_combined : {false, true}) {
paddle::framework::LoDTensor output1;
std::vector<paddle::framework::LoDTensor*> cpu_fetchs1;
cpu_fetchs1.push_back(&output1);
// Run inference on CPU
TestInference<paddle::platform::CPUPlace>(dirname, cpu_feeds, cpu_fetchs1);
LOG(INFO) << output1.dims();
// Run inference on CPU
LOG(INFO) << "--- CPU Runs: is_combined=" << is_combined << " ---";
TestInference<paddle::platform::CPUPlace>(
dirname, cpu_feeds, cpu_fetchs1, FLAGS_repeat, is_combined);
LOG(INFO) << output1.dims();
#ifdef PADDLE_WITH_CUDA
paddle::framework::LoDTensor output2;
std::vector<paddle::framework::LoDTensor*> cpu_fetchs2;
cpu_fetchs2.push_back(&output2);
paddle::framework::LoDTensor output2;
std::vector<paddle::framework::LoDTensor*> cpu_fetchs2;
cpu_fetchs2.push_back(&output2);
// Run inference on CUDA GPU
TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
LOG(INFO) << output2.dims();
// Run inference on CUDA GPU
LOG(INFO) << "--- GPU Runs: is_combined=" << is_combined << " ---";
TestInference<paddle::platform::CUDAPlace>(
dirname, cpu_feeds, cpu_fetchs2, FLAGS_repeat, is_combined);
LOG(INFO) << output2.dims();
CheckError<float>(output1, output2);
CheckError<float>(output1, output2);
#endif
}
TEST(inference, recognize_digits_combine) {
if (FLAGS_dirname.empty()) {
LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model";
}
LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl;
std::string dirname = FLAGS_dirname;
// 0. Call `paddle::framework::InitDevices()` initialize all the devices
// In unittests, this is done in paddle/testing/paddle_gtest_main.cc
paddle::framework::LoDTensor input;
// Use normilized image pixels as input data,
// which should be in the range [-1.0, 1.0].
SetupTensor<float>(
input, {1, 1, 28, 28}, static_cast<float>(-1), static_cast<float>(1));
std::vector<paddle::framework::LoDTensor*> cpu_feeds;
cpu_feeds.push_back(&input);
paddle::framework::LoDTensor output1;
std::vector<paddle::framework::LoDTensor*> cpu_fetchs1;
cpu_fetchs1.push_back(&output1);
// Run inference on CPU
TestInference<paddle::platform::CPUPlace, true>(
dirname, cpu_feeds, cpu_fetchs1);
LOG(INFO) << output1.dims();
#ifdef PADDLE_WITH_CUDA
paddle::framework::LoDTensor output2;
std::vector<paddle::framework::LoDTensor*> cpu_fetchs2;
cpu_fetchs2.push_back(&output2);
// Run inference on CUDA GPU
TestInference<paddle::platform::CUDAPlace, true>(
dirname, cpu_feeds, cpu_fetchs2);
LOG(INFO) << output2.dims();
CheckError<float>(output1, output2);
#endif
}

78
paddle/fluid/inference/tests/test_helper.h
Unescape View File

@ -15,6 +15,7 @@ limitations under the License. */
#include <time.h>
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/inference/io.h"
#include "paddle/fluid/platform/profiler.h"
template <typename T>
void SetupTensor(paddle::framework::LoDTensor& input,
@ -87,31 +88,60 @@ void CheckError(paddle::framework::LoDTensor& output1,
EXPECT_EQ(count, 0U) << "There are " << count << " different elements.";
}
template <typename Place, bool IsCombined = false>
template <typename Place>
void TestInference(const std::string& dirname,
const std::vector<paddle::framework::LoDTensor*>& cpu_feeds,
std::vector<paddle::framework::LoDTensor*>& cpu_fetchs) {
std::vector<paddle::framework::LoDTensor*>& cpu_fetchs,
const int repeat = 1,
const bool is_combined = false) {
// 1. Define place, executor, scope
auto place = Place();
auto executor = paddle::framework::Executor(place);
auto* scope = new paddle::framework::Scope();
// Profile the performance
paddle::platform::ProfilerState state;
if (paddle::platform::is_cpu_place(place)) {
state = paddle::platform::ProfilerState::kCPU;
} else {
#ifdef PADDLE_WITH_CUDA
state = paddle::platform::ProfilerState::kCUDA;
// The default device_id of paddle::platform::CUDAPlace is 0.
// Users can get the device_id using:
// int device_id = place.GetDeviceId();
paddle::platform::SetDeviceId(0);
#else
PADDLE_THROW("'CUDAPlace' is not supported in CPU only device.");
#endif
}
// Enable the profiler
paddle::platform::EnableProfiler(state);
// 2. Initialize the inference_program and load parameters
std::unique_ptr<paddle::framework::ProgramDesc> inference_program;
if (IsCombined) {
// All parameters are saved in a single file.
// Hard-coding the file names of program and parameters in unittest.
// The file names should be consistent with that used in Python API
// `fluid.io.save_inference_model`.
std::string prog_filename = "__model_combined__";
std::string param_filename = "__params_combined__";
inference_program = paddle::inference::Load(executor,
*scope,
dirname + "/" + prog_filename,
dirname + "/" + param_filename);
} else {
// Parameters are saved in separate files sited in the specified `dirname`.
inference_program = paddle::inference::Load(executor, *scope, dirname);
{
paddle::platform::RecordEvent record_event(
"init_program",
paddle::platform::DeviceContextPool::Instance().Get(place));
if (is_combined) {
// All parameters are saved in a single file.
// Hard-coding the file names of program and parameters in unittest.
// The file names should be consistent with that used in Python API
// `fluid.io.save_inference_model`.
std::string prog_filename = "__model_combined__";
std::string param_filename = "__params_combined__";
inference_program =
paddle::inference::Load(executor,
*scope,
dirname + "/" + prog_filename,
dirname + "/" + param_filename);
} else {
// Parameters are saved in separate files sited in the specified
// `dirname`.
inference_program = paddle::inference::Load(executor, *scope, dirname);
}
}
// 3. Get the feed_target_names and fetch_target_names
@ -134,7 +164,21 @@ void TestInference(const std::string& dirname,
}
// 6. Run the inference program
executor.Run(*inference_program, scope, feed_targets, fetch_targets);
{
// Run repeat times to profile the performance
for (int i = 0; i < repeat; ++i) {
paddle::platform::RecordEvent record_event(
"run_inference",
paddle::platform::DeviceContextPool::Instance().Get(place));
executor.Run(*inference_program, scope, feed_targets, fetch_targets);
}
}
// Disable the profiler and print the timing information
paddle::platform::DisableProfiler(paddle::platform::EventSortingKey::kDefault,
"profiler.txt");
paddle::platform::ResetProfiler();
delete scope;
}

5
paddle/fluid/operators/detail/CMakeLists.txt
Unescape View File

@ -1,3 +1,6 @@
if(WITH_DISTRIBUTE)
grpc_library(sendrecvop_grpc SRCS sendrecvop_utils.cc grpc_client.cc grpc_server.cc PROTO send_recv.proto DEPS lod_tensor selected_rows)
grpc_library(sendrecvop_grpc SRCS bytebuffer_stream.cc sendrecvop_utils.cc grpc_client.cc grpc_server.cc PROTO send_recv.proto DEPS lod_tensor selected_rows)
set(DISTRIBUTE_COMPILE_FLAGS "-Wno-non-virtual-dtor -Wno-error=non-virtual-dtor -Wno-error=delete-non-virtual-dtor")
set_source_files_properties(test_serde.cc PROPERTIES COMPILE_FLAGS ${DISTRIBUTE_COMPILE_FLAGS})
cc_test(serde_test SRCS test_serde.cc DEPS grpc++_unsecure grpc_unsecure gpr cares zlib protobuf sendrecvop_grpc)
endif()

Some files were not shown because too many files have changed in this diff Show More

Write Preview
Loading…
Cancel
Save