Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
PaddleDetection
提交
fc8aedb1
P
PaddleDetection
项目概览
PaddlePaddle
/
PaddleDetection
大约 1 年 前同步成功
通知
695
Star
11112
Fork
2696
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
184
列表
看板
标记
里程碑
合并请求
40
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleDetection
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
184
Issue
184
列表
看板
标记
里程碑
合并请求
40
合并请求
40
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
提交
fc8aedb1
编写于
7月 12, 2017
作者:
X
xzl
浏览文件
操作
浏览文件
下载
差异文件
Merge branch 'develop' of
https://github.com/PaddlePaddle/Paddle
into mobilenet_gpu
上级
030a3db2
be441f7d
变更
11
显示空白变更内容
内联
并排
Showing
11 changed file
with
132 addition
and
39 deletion
+132
-39
cmake/generic.cmake
cmake/generic.cmake
+1
-1
paddle/framework/CMakeLists.txt
paddle/framework/CMakeLists.txt
+1
-1
paddle/framework/op_registry.h
paddle/framework/op_registry.h
+7
-7
paddle/framework/op_registry_test.cc
paddle/framework/op_registry_test.cc
+2
-2
paddle/framework/operator_test.cc
paddle/framework/operator_test.cc
+51
-1
paddle/platform/CMakeLists.txt
paddle/platform/CMakeLists.txt
+3
-2
paddle/platform/device_context.cc
paddle/platform/device_context.cc
+25
-8
paddle/platform/device_context.h
paddle/platform/device_context.h
+23
-12
paddle/platform/device_context_test.cc
paddle/platform/device_context_test.cc
+16
-3
paddle/platform/dynload/CMakeLists.txt
paddle/platform/dynload/CMakeLists.txt
+1
-1
paddle/pybind/pybind.cc
paddle/pybind/pybind.cc
+2
-1
未找到文件。
cmake/generic.cmake
浏览文件 @
fc8aedb1
...
@@ -93,7 +93,7 @@ include_directories(${CMAKE_CURRENT_BINARY_DIR})
...
@@ -93,7 +93,7 @@ include_directories(${CMAKE_CURRENT_BINARY_DIR})
if
(
NOT APPLE
)
if
(
NOT APPLE
)
find_package
(
Threads REQUIRED
)
find_package
(
Threads REQUIRED
)
link_libraries
(
${
CMAKE_THREAD_LIBS_INIT
}
)
link_libraries
(
${
CMAKE_THREAD_LIBS_INIT
}
)
set
(
CMAKE_CXX_LINK_EXECUTABLE
"
${
CMAKE_CXX_LINK_EXECUTABLE
}
-ldl"
)
set
(
CMAKE_CXX_LINK_EXECUTABLE
"
${
CMAKE_CXX_LINK_EXECUTABLE
}
-ldl
-lrt
"
)
endif
(
NOT APPLE
)
endif
(
NOT APPLE
)
function
(
merge_static_libs TARGET_NAME
)
function
(
merge_static_libs TARGET_NAME
)
...
...
paddle/framework/CMakeLists.txt
浏览文件 @
fc8aedb1
...
@@ -12,7 +12,7 @@ cc_test(op_proto_test SRCS op_proto_test.cc DEPS op_proto protobuf)
...
@@ -12,7 +12,7 @@ cc_test(op_proto_test SRCS op_proto_test.cc DEPS op_proto protobuf)
proto_library
(
op_desc SRCS op_desc.proto DEPS attr_type
)
proto_library
(
op_desc SRCS op_desc.proto DEPS attr_type
)
cc_test
(
op_desc_test SRCS op_desc_test.cc DEPS op_desc protobuf
)
cc_test
(
op_desc_test SRCS op_desc_test.cc DEPS op_desc protobuf
)
cc_library
(
operator SRCS operator.cc DEPS op_desc protobuf
)
cc_library
(
operator SRCS operator.cc DEPS op_desc protobuf
)
cc_test
(
operator_test SRCS operator_test.cc DEPS operator op_registry
)
cc_test
(
operator_test SRCS operator_test.cc DEPS operator op_registry
place
)
cc_library
(
op_registry SRCS op_registry.cc DEPS op_proto op_desc
)
cc_library
(
op_registry SRCS op_registry.cc DEPS op_proto op_desc
)
cc_test
(
op_registry_test SRCS op_registry_test.cc DEPS op_registry operator
)
cc_test
(
op_registry_test SRCS op_registry_test.cc DEPS op_registry operator
)
py_proto_compile
(
framework_py_proto SRCS attr_type.proto op_proto.proto op_desc.proto
)
py_proto_compile
(
framework_py_proto SRCS attr_type.proto op_proto.proto op_desc.proto
)
...
...
paddle/framework/op_registry.h
浏览文件 @
fc8aedb1
...
@@ -147,13 +147,13 @@ class OpRegisterHelper {
...
@@ -147,13 +147,13 @@ class OpRegisterHelper {
}
}
};
};
#define REGISTER_OP(
__op_class, __op_maker_class, __op_type)
\
#define REGISTER_OP(
type, op_class, op_maker_class)
\
class
__op_class##Register {
\
class
op_class##Register {
\
private: \
private: \
const static OpRegisterHelper<
__op_class, __op_maker_class> reg;
\
const static OpRegisterHelper<
op_class, op_maker_class> reg;
\
}; \
}; \
const OpRegisterHelper<
__op_class, __op_maker_class>
\
const OpRegisterHelper<
op_class, op_maker_class> op_class##Register::reg(
\
__op_class##Register::reg(#__op_type);
#type)
}
// namespace framework
}
// namespace framework
}
// namespace paddle
}
// namespace paddle
paddle/framework/op_registry_test.cc
浏览文件 @
fc8aedb1
...
@@ -26,7 +26,7 @@ class CosineOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
...
@@ -26,7 +26,7 @@ class CosineOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
}
}
};
};
REGISTER_OP
(
CosineOp
,
CosineOpProtoAndCheckerMaker
,
cos_sim
)
REGISTER_OP
(
cos_sim
,
CosineOp
,
CosineOpProtoAndCheckerMaker
);
class
MyTestOp
:
public
OperatorBase
{
class
MyTestOp
:
public
OperatorBase
{
public:
public:
...
@@ -53,7 +53,7 @@ class MyTestOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
...
@@ -53,7 +53,7 @@ class MyTestOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
}
}
};
};
REGISTER_OP
(
MyTestOp
,
MyTestOpProtoAndCheckerMaker
,
my_test_op
)
REGISTER_OP
(
my_test_op
,
MyTestOp
,
MyTestOpProtoAndCheckerMaker
);
}
// namespace framework
}
// namespace framework
}
// namespace paddle
}
// namespace paddle
...
...
paddle/framework/operator_test.cc
浏览文件 @
fc8aedb1
...
@@ -45,7 +45,7 @@ class OperatorTestProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
...
@@ -45,7 +45,7 @@ class OperatorTestProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
}
}
};
};
REGISTER_OP
(
OperatorTest
,
OperatorTestProtoAndCheckerMaker
,
test_operator
)
REGISTER_OP
(
test_operator
,
OperatorTest
,
OperatorTestProtoAndCheckerMaker
);
TEST
(
OperatorBase
,
all
)
{
TEST
(
OperatorBase
,
all
)
{
OpDesc
op_desc
;
OpDesc
op_desc
;
...
@@ -69,5 +69,55 @@ TEST(OperatorBase, all) {
...
@@ -69,5 +69,55 @@ TEST(OperatorBase, all) {
delete
op
;
delete
op
;
}
}
class
OpKernelTestProtoAndCheckerMaker
:
public
OpProtoAndCheckerMaker
{
public:
OpKernelTestProtoAndCheckerMaker
(
OpProto
*
proto
,
OpAttrChecker
*
op_checker
)
:
OpProtoAndCheckerMaker
(
proto
,
op_checker
)
{
AddInput
(
"input"
,
"input of test op"
);
AddOutput
(
"output"
,
"output of test op"
);
AddAttr
<
float
>
(
"scale"
,
"scale of cosine op"
)
.
SetDefault
(
1.0
)
.
LargerThan
(
0.0
);
AddType
(
"test_operator"
);
AddComment
(
"This is test op"
);
}
};
class
OpWithKernelTest
:
public
OperatorWithKernel
{
public:
void
InferShape
(
const
std
::
shared_ptr
<
Scope
>&
scope
)
const
override
{}
};
class
CPUKernelTest
:
public
OpKernel
{
public:
void
Compute
(
const
KernelContext
&
context
)
const
{
float
scale
=
context
.
op_
.
GetAttr
<
float
>
(
"scale"
);
ASSERT_NEAR
(
scale
,
3.14
,
1e-5
);
std
::
cout
<<
"this is cpu kernel"
<<
std
::
endl
;
std
::
cout
<<
context
.
op_
.
DebugString
()
<<
std
::
endl
;
}
};
REGISTER_OP
(
op_with_kernel
,
OpWithKernelTest
,
OpKernelTestProtoAndCheckerMaker
);
REGISTER_OP_KERNEL
(
op_with_kernel
,
platform
::
CPUPlace
,
CPUKernelTest
);
TEST
(
OpKernel
,
all
)
{
OpDesc
op_desc
;
op_desc
.
set_type
(
"op_with_kernel"
);
*
op_desc
.
mutable_inputs
()
->
Add
()
=
"IN1"
;
*
op_desc
.
mutable_outputs
()
->
Add
()
=
"OUT1"
;
auto
attr
=
op_desc
.
mutable_attrs
()
->
Add
();
attr
->
set_name
(
"scale"
);
attr
->
set_type
(
paddle
::
framework
::
AttrType
::
FLOAT
);
attr
->
set_f
(
3.14
);
platform
::
CPUDeviceContext
cpu_device_context
;
auto
scope
=
std
::
make_shared
<
Scope
>
();
OperatorBase
*
op
=
paddle
::
framework
::
OpRegistry
::
CreateOp
(
op_desc
);
op
->
Run
(
scope
,
cpu_device_context
);
delete
op
;
}
}
// namespace framework
}
// namespace framework
}
// namespace paddle
}
// namespace paddle
\ No newline at end of file
paddle/platform/CMakeLists.txt
浏览文件 @
fc8aedb1
add_subdirectory
(
dynload
)
add_subdirectory
(
dynload
)
nv_test
(
cuda_test SRCS cuda_test.cu
DEPS dyload_cuda
)
nv_test
(
cuda_test SRCS cuda_test.cu
)
cc_library
(
place SRCS place.cc
)
cc_library
(
place SRCS place.cc
)
cc_test
(
place_test SRCS place_test.cc DEPS place glog gflags
)
cc_test
(
place_test SRCS place_test.cc DEPS place glog gflags
)
IF
(
WITH_GPU
)
IF
(
WITH_GPU
)
set
(
GPU_CTX_DEPS dy
load_cuda dynamic_loader
)
set
(
GPU_CTX_DEPS dy
nload_cuda dynamic_loader
)
ELSE
()
ELSE
()
set
(
GPU_CTX_DEPS
)
set
(
GPU_CTX_DEPS
)
ENDIF
()
ENDIF
()
...
...
paddle/platform/device_context.cc
浏览文件 @
fc8aedb1
#include <paddle/platform/device_context.h>
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/platform/device_context.h"
namespace
paddle
{
namespace
paddle
{
namespace
platform
{
namespace
platform
{
namespace
dynload
{
namespace
dummy
{
// Make DeviceContext A library.
int
DUMMY_VAR_FOR_DEV_CTX
=
0
;
}
// namespace dummy
template
<
>
}
// namespace dynload
Eigen
::
DefaultDevice
*
DeviceContext
::
get_eigen_device
<
Eigen
::
DefaultDevice
>
()
{
return
reinterpret_cast
<
CPUDeviceContext
*>
(
this
)
->
eigen_device
();
}
#ifndef PADDLE_ONLY_CPU
template
<
>
Eigen
::
GpuDevice
*
DeviceContext
::
get_eigen_device
<
Eigen
::
GpuDevice
>
()
{
return
reinterpret_cast
<
CUDADeviceContext
*>
(
this
)
->
eigen_device
();
}
#endif
}
// namespace platform
}
// namespace platform
}
// namespace paddle
}
// namespace paddle
paddle/platform/device_context.h
浏览文件 @
fc8aedb1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
...
@@ -23,6 +20,7 @@ limitations under the License. */
...
@@ -23,6 +20,7 @@ limitations under the License. */
#define EIGEN_USE_GPU
#define EIGEN_USE_GPU
#endif
#endif
#include <paddle/platform/place.h>
#include <paddle/platform/place.h>
#include <memory>
#include <unsupported/Eigen/CXX11/Tensor>
#include <unsupported/Eigen/CXX11/Tensor>
namespace
paddle
{
namespace
paddle
{
...
@@ -32,14 +30,27 @@ class DeviceContext {
...
@@ -32,14 +30,27 @@ class DeviceContext {
public:
public:
virtual
~
DeviceContext
()
{}
virtual
~
DeviceContext
()
{}
virtual
Place
GetPlace
()
const
=
0
;
virtual
Place
GetPlace
()
const
=
0
;
template
<
typename
DeviceType
>
DeviceType
*
get_eigen_device
();
};
};
class
CPUDeviceContext
:
public
DeviceContext
{
class
CPUDeviceContext
:
public
DeviceContext
{
public:
public:
Eigen
::
DefaultDevice
*
eigen_device
()
{
if
(
!
eigen_device_
)
{
eigen_device_
.
reset
(
new
Eigen
::
DefaultDevice
());
}
return
eigen_device_
.
get
();
}
Place
GetPlace
()
const
override
{
Place
GetPlace
()
const
override
{
Place
retv
=
CPUPlace
();
Place
retv
=
CPUPlace
();
return
retv
;
return
retv
;
}
}
private:
std
::
unique_ptr
<
Eigen
::
DefaultDevice
>
eigen_device_
;
};
};
#ifndef PADDLE_ONLY_CPU
#ifndef PADDLE_ONLY_CPU
...
@@ -64,8 +75,8 @@ class CUDADeviceContext : public DeviceContext {
...
@@ -64,8 +75,8 @@ class CUDADeviceContext : public DeviceContext {
GPUPlaceGuard
guard
(
gpu_place_
);
GPUPlaceGuard
guard
(
gpu_place_
);
paddle
::
platform
::
throw_on_error
(
cudaStreamCreate
(
&
stream_
),
paddle
::
platform
::
throw_on_error
(
cudaStreamCreate
(
&
stream_
),
"cudaStreamCreate failed"
);
"cudaStreamCreate failed"
);
eigen_stream_
=
new
Eigen
::
CudaStreamDevice
(
&
stream_
);
eigen_stream_
.
reset
(
new
Eigen
::
CudaStreamDevice
(
&
stream_
)
);
eigen_device_
=
new
Eigen
::
GpuDevice
(
eigen_stream_
);
eigen_device_
.
reset
(
new
Eigen
::
GpuDevice
(
eigen_stream_
.
get
())
);
}
}
Place
GetPlace
()
const
override
{
Place
GetPlace
()
const
override
{
...
@@ -80,7 +91,7 @@ class CUDADeviceContext : public DeviceContext {
...
@@ -80,7 +91,7 @@ class CUDADeviceContext : public DeviceContext {
cudaStream_t
stream
()
{
return
stream_
;
}
cudaStream_t
stream
()
{
return
stream_
;
}
Eigen
::
GpuDevice
eigen_device
()
{
return
*
eigen_device_
;
}
Eigen
::
GpuDevice
*
eigen_device
()
{
return
eigen_device_
.
get
()
;
}
cublasHandle_t
cublas_handle
()
{
cublasHandle_t
cublas_handle
()
{
if
(
!
blas_handle_
)
{
if
(
!
blas_handle_
)
{
...
@@ -145,10 +156,8 @@ class CUDADeviceContext : public DeviceContext {
...
@@ -145,10 +156,8 @@ class CUDADeviceContext : public DeviceContext {
rand_generator_
)
==
CURAND_STATUS_SUCCESS
,
rand_generator_
)
==
CURAND_STATUS_SUCCESS
,
"curandDestroyGenerator failed"
);
"curandDestroyGenerator failed"
);
}
}
eigen_stream_
.
reset
();
delete
eigen_stream_
;
eigen_device_
.
reset
();
delete
eigen_device_
;
paddle
::
platform
::
throw_on_error
(
cudaStreamDestroy
(
stream_
),
paddle
::
platform
::
throw_on_error
(
cudaStreamDestroy
(
stream_
),
"cudaStreamDestroy failed"
);
"cudaStreamDestroy failed"
);
}
}
...
@@ -157,8 +166,8 @@ class CUDADeviceContext : public DeviceContext {
...
@@ -157,8 +166,8 @@ class CUDADeviceContext : public DeviceContext {
GPUPlace
gpu_place_
;
GPUPlace
gpu_place_
;
cudaStream_t
stream_
;
cudaStream_t
stream_
;
Eigen
::
CudaStreamDevice
*
eigen_stream_
;
std
::
unique_ptr
<
Eigen
::
CudaStreamDevice
>
eigen_stream_
;
Eigen
::
GpuDevice
*
eigen_device_
;
std
::
unique_ptr
<
Eigen
::
GpuDevice
>
eigen_device_
;
cublasHandle_t
blas_handle_
{
nullptr
};
cublasHandle_t
blas_handle_
{
nullptr
};
...
@@ -167,6 +176,8 @@ class CUDADeviceContext : public DeviceContext {
...
@@ -167,6 +176,8 @@ class CUDADeviceContext : public DeviceContext {
int
random_seed_
;
int
random_seed_
;
curandGenerator_t
rand_generator_
{
nullptr
};
curandGenerator_t
rand_generator_
{
nullptr
};
};
};
#endif
#endif
}
// namespace platform
}
// namespace platform
}
// namespace paddle
}
// namespace paddle
paddle/platform/device_context_test.cc
浏览文件 @
fc8aedb1
...
@@ -15,13 +15,26 @@ limitations under the License. */
...
@@ -15,13 +15,26 @@ limitations under the License. */
#include "paddle/platform/device_context.h"
#include "paddle/platform/device_context.h"
#include "gtest/gtest.h"
#include "gtest/gtest.h"
TEST
(
CUDADeviceContext
,
Init
)
{
using
DEVICE_GPU
=
Eigen
::
GpuDevice
;
TEST
(
Device
,
Init
)
{
int
count
=
paddle
::
platform
::
GetDeviceCount
();
for
(
int
i
=
0
;
i
<
count
;
i
++
)
{
paddle
::
platform
::
DeviceContext
*
device_context
=
new
paddle
::
platform
::
CUDADeviceContext
(
i
);
Eigen
::
GpuDevice
*
gpu_device
=
device_context
->
template
get_eigen_device
<
DEVICE_GPU
>();
ASSERT_NE
(
nullptr
,
gpu_device
);
delete
device_context
;
}
}
TEST
(
Device
,
CUDADeviceContext
)
{
int
count
=
paddle
::
platform
::
GetDeviceCount
();
int
count
=
paddle
::
platform
::
GetDeviceCount
();
for
(
int
i
=
0
;
i
<
count
;
i
++
)
{
for
(
int
i
=
0
;
i
<
count
;
i
++
)
{
paddle
::
platform
::
CUDADeviceContext
*
device_context
=
paddle
::
platform
::
CUDADeviceContext
*
device_context
=
new
paddle
::
platform
::
CUDADeviceContext
(
i
);
new
paddle
::
platform
::
CUDADeviceContext
(
i
);
Eigen
::
GpuDevice
gpu_device
=
device_context
->
eigen_device
();
Eigen
::
GpuDevice
*
gpu_device
=
device_context
->
eigen_device
();
ASSERT_NE
(
nullptr
,
gpu_device
.
stream
()
);
ASSERT_NE
(
nullptr
,
gpu_device
);
cudnnHandle_t
cudnn_handle
=
device_context
->
cudnn_handle
();
cudnnHandle_t
cudnn_handle
=
device_context
->
cudnn_handle
();
ASSERT_NE
(
nullptr
,
cudnn_handle
);
ASSERT_NE
(
nullptr
,
cudnn_handle
);
cublasHandle_t
cublas_handle
=
device_context
->
cublas_handle
();
cublasHandle_t
cublas_handle
=
device_context
->
cublas_handle
();
...
...
paddle/platform/dynload/CMakeLists.txt
浏览文件 @
fc8aedb1
cc_library
(
dynamic_loader SRCS dynamic_loader.cc DEPS glog gflags
)
cc_library
(
dynamic_loader SRCS dynamic_loader.cc DEPS glog gflags
)
nv_library
(
dyload_cuda SRCS cublas.cc cudnn.cc curand.cc
)
nv_library
(
dy
n
load_cuda SRCS cublas.cc cudnn.cc curand.cc
)
paddle/pybind/pybind.cc
浏览文件 @
fc8aedb1
...
@@ -12,6 +12,7 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
...
@@ -12,6 +12,7 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
See the License for the specific language governing permissions and
limitations under the License. */
limitations under the License. */
#include <Python.h>
#include <paddle/framework/scope.h>
#include <paddle/framework/scope.h>
#include <pybind11/pybind11.h>
#include <pybind11/pybind11.h>
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录