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
兽拳
Paddle
提交
ca84c2ca
P
Paddle
项目概览
兽拳
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
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
1
Issue
1
列表
看板
标记
里程碑
合并请求
0
合并请求
0
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
体验新版 GitCode,发现更多精彩内容 >>
提交
ca84c2ca
编写于
12月 12, 2018
作者:
S
sneaxiy
浏览文件
操作
浏览文件
下载
差异文件
merge develop
test=develop
上级
06f8aa5b
8175983e
变更
57
隐藏空白更改
内联
并排
Showing
57 changed file
with
183 addition
and
145 deletion
+183
-145
CMakeLists.txt
CMakeLists.txt
+7
-1
cmake/FindGperftools.cmake
cmake/FindGperftools.cmake
+63
-0
cmake/configure.cmake
cmake/configure.cmake
+1
-0
cmake/generic.cmake
cmake/generic.cmake
+16
-0
paddle/fluid/framework/parallel_executor.cc
paddle/fluid/framework/parallel_executor.cc
+30
-1
paddle/fluid/operators/bilinear_tensor_product_op.cu
paddle/fluid/operators/bilinear_tensor_product_op.cu
+0
-1
paddle/fluid/operators/cos_sim_op.cu
paddle/fluid/operators/cos_sim_op.cu
+0
-2
paddle/fluid/operators/crop_op.cu
paddle/fluid/operators/crop_op.cu
+0
-2
paddle/fluid/operators/dropout_op.cu
paddle/fluid/operators/dropout_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_add_op.cu
paddle/fluid/operators/elementwise/elementwise_add_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_div_op.cu
paddle/fluid/operators/elementwise/elementwise_div_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_max_op.cu
paddle/fluid/operators/elementwise/elementwise_max_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_min_op.cu
paddle/fluid/operators/elementwise/elementwise_min_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_mul_op.cu
paddle/fluid/operators/elementwise/elementwise_mul_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_pow_op.cu
paddle/fluid/operators/elementwise/elementwise_pow_op.cu
+0
-2
paddle/fluid/operators/elementwise/elementwise_sub_op.cu
paddle/fluid/operators/elementwise/elementwise_sub_op.cu
+0
-2
paddle/fluid/operators/expand_op.cu
paddle/fluid/operators/expand_op.cu
+0
-3
paddle/fluid/operators/gru_unit_op.cu
paddle/fluid/operators/gru_unit_op.cu
+0
-2
paddle/fluid/operators/hinge_loss_op.cu
paddle/fluid/operators/hinge_loss_op.cu
+0
-2
paddle/fluid/operators/huber_loss_op.cu
paddle/fluid/operators/huber_loss_op.cu
+0
-2
paddle/fluid/operators/im2sequence_op.cu
paddle/fluid/operators/im2sequence_op.cu
+0
-2
paddle/fluid/operators/isfinite_op.cu
paddle/fluid/operators/isfinite_op.cu
+0
-2
paddle/fluid/operators/l1_norm_op.cu
paddle/fluid/operators/l1_norm_op.cu
+0
-2
paddle/fluid/operators/log_loss_op.cu
paddle/fluid/operators/log_loss_op.cu
+0
-2
paddle/fluid/operators/math/context_project.cu
paddle/fluid/operators/math/context_project.cu
+0
-3
paddle/fluid/operators/math/jit_kernel_layer_norm.cc
paddle/fluid/operators/math/jit_kernel_layer_norm.cc
+8
-5
paddle/fluid/operators/math/math_function.cu
paddle/fluid/operators/math/math_function.cu
+0
-2
paddle/fluid/operators/math/sequence2batch.cu
paddle/fluid/operators/math/sequence2batch.cu
+0
-2
paddle/fluid/operators/math/softmax.cu
paddle/fluid/operators/math/softmax.cu
+0
-3
paddle/fluid/operators/mean_op.cu
paddle/fluid/operators/mean_op.cu
+0
-3
paddle/fluid/operators/optimizers/adadelta_op.cu
paddle/fluid/operators/optimizers/adadelta_op.cu
+0
-2
paddle/fluid/operators/optimizers/adagrad_op.cu
paddle/fluid/operators/optimizers/adagrad_op.cu
+0
-2
paddle/fluid/operators/optimizers/adam_op.cu
paddle/fluid/operators/optimizers/adam_op.cu
+0
-2
paddle/fluid/operators/optimizers/adamax_op.cu
paddle/fluid/operators/optimizers/adamax_op.cu
+0
-2
paddle/fluid/operators/optimizers/decayed_adagrad_op.cu
paddle/fluid/operators/optimizers/decayed_adagrad_op.cu
+0
-2
paddle/fluid/operators/optimizers/ftrl_op.cu
paddle/fluid/operators/optimizers/ftrl_op.cu
+0
-2
paddle/fluid/operators/optimizers/proximal_adagrad_op.cu
paddle/fluid/operators/optimizers/proximal_adagrad_op.cu
+0
-2
paddle/fluid/operators/optimizers/proximal_gd_op.cu
paddle/fluid/operators/optimizers/proximal_gd_op.cu
+0
-2
paddle/fluid/operators/optimizers/rmsprop_op.cu
paddle/fluid/operators/optimizers/rmsprop_op.cu
+0
-2
paddle/fluid/operators/pad_constant_like_op.cu
paddle/fluid/operators/pad_constant_like_op.cu
+0
-2
paddle/fluid/operators/pad_op.cu
paddle/fluid/operators/pad_op.cu
+0
-2
paddle/fluid/operators/sequence_ops/sequence_pool_op.cu
paddle/fluid/operators/sequence_ops/sequence_pool_op.cu
+0
-3
paddle/fluid/operators/sigmoid_cross_entropy_with_logits_op.cu
...e/fluid/operators/sigmoid_cross_entropy_with_logits_op.cu
+0
-2
paddle/fluid/operators/smooth_l1_loss_op.cu
paddle/fluid/operators/smooth_l1_loss_op.cu
+0
-3
paddle/fluid/operators/softmax_with_cross_entropy_op.cu
paddle/fluid/operators/softmax_with_cross_entropy_op.cu
+0
-3
paddle/fluid/operators/split_selected_rows_op.h
paddle/fluid/operators/split_selected_rows_op.h
+6
-3
paddle/fluid/operators/squared_l2_distance_op.cu
paddle/fluid/operators/squared_l2_distance_op.cu
+0
-3
paddle/fluid/operators/squared_l2_norm_op.cu
paddle/fluid/operators/squared_l2_norm_op.cu
+0
-2
paddle/fluid/operators/sum_op.cu
paddle/fluid/operators/sum_op.cu
+0
-2
paddle/fluid/platform/cuda_helper_test.cu
paddle/fluid/platform/cuda_helper_test.cu
+1
-1
paddle/fluid/platform/device_context.h
paddle/fluid/platform/device_context.h
+0
-1
paddle/fluid/platform/enforce.h
paddle/fluid/platform/enforce.h
+42
-38
paddle/fluid/platform/float16.h
paddle/fluid/platform/float16.h
+0
-3
paddle/fluid/pybind/pybind.cc
paddle/fluid/pybind/pybind.cc
+2
-0
python/paddle/fluid/__init__.py
python/paddle/fluid/__init__.py
+1
-1
python/paddle/fluid/average.py
python/paddle/fluid/average.py
+1
-0
python/paddle/fluid/tests/unittests/test_split_selected_rows_op.py
...ddle/fluid/tests/unittests/test_split_selected_rows_op.py
+5
-0
未找到文件。
CMakeLists.txt
浏览文件 @
ca84c2ca
...
...
@@ -54,7 +54,7 @@ option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON)
option
(
WITH_DOUBLE
"Compile PaddlePaddle with double precision"
OFF
)
option
(
WITH_RDMA
"Compile PaddlePaddle with RDMA support"
OFF
)
option
(
WITH_TIMER
"Compile PaddlePaddle with stats timer"
OFF
)
option
(
WITH_PROFILER
"Compile PaddlePaddle with GPU profiler"
OFF
)
option
(
WITH_PROFILER
"Compile PaddlePaddle with GPU profiler
and gperftools
"
OFF
)
option
(
WITH_DOC
"Compile PaddlePaddle with documentation"
OFF
)
option
(
WITH_COVERAGE
"Compile PaddlePaddle with code coverage"
OFF
)
option
(
COVERALLS_UPLOAD
"Package code coverage data to coveralls"
OFF
)
...
...
@@ -254,6 +254,12 @@ elseif()
set
(
WITH_ANAKIN OFF CACHE STRING
"Anakin is used in MKL only now."
FORCE
)
endif
()
if
(
WITH_PROFILER
)
find_package
(
Gperftools REQUIRED
)
include_directories
(
${
GPERFTOOLS_INCLUDE_DIR
}
)
add_definitions
(
-DWITH_GPERFTOOLS
)
endif
()
include
(
generic
)
# simplify cmake module
include
(
package
)
# set paddle packages
include
(
ccache
)
# set ccache for compilation
...
...
cmake/FindGperftools.cmake
0 → 100644
浏览文件 @
ca84c2ca
# Tries to find Gperftools.
#
# Usage of this module as follows:
#
# find_package(Gperftools)
#
# Variables used by this module, they can change the default behaviour and need
# to be set before calling find_package:
#
# Gperftools_ROOT_DIR Set this variable to the root installation of
# Gperftools if the module has problems finding
# the proper installation path.
#
# Variables defined by this module:
#
# GPERFTOOLS_FOUND System has Gperftools libs/headers
# GPERFTOOLS_LIBRARIES The Gperftools libraries (tcmalloc & profiler)
# GPERFTOOLS_INCLUDE_DIR The location of Gperftools headers
find_library
(
GPERFTOOLS_TCMALLOC
NAMES tcmalloc
HINTS
${
Gperftools_ROOT_DIR
}
/lib
)
find_library
(
GPERFTOOLS_PROFILER
NAMES profiler
HINTS
${
Gperftools_ROOT_DIR
}
/lib
)
find_library
(
GPERFTOOLS_TCMALLOC_AND_PROFILER
NAMES tcmalloc_and_profiler
HINTS
${
Gperftools_ROOT_DIR
}
/lib
)
find_path
(
GPERFTOOLS_INCLUDE_DIR
NAMES gperftools/heap-profiler.h
HINTS
${
Gperftools_ROOT_DIR
}
/include
)
set
(
GPERFTOOLS_LIBRARIES
${
GPERFTOOLS_TCMALLOC_AND_PROFILER
}
)
include
(
FindPackageHandleStandardArgs
)
find_package_handle_standard_args
(
Gperftools
DEFAULT_MSG
GPERFTOOLS_LIBRARIES
GPERFTOOLS_INCLUDE_DIR
)
mark_as_advanced
(
Gperftools_ROOT_DIR
GPERFTOOLS_TCMALLOC
GPERFTOOLS_PROFILER
GPERFTOOLS_TCMALLOC_AND_PROFILER
GPERFTOOLS_LIBRARIES
GPERFTOOLS_INCLUDE_DIR
)
# create IMPORTED targets
if
(
Gperftools_FOUND AND NOT TARGET gperftools::tcmalloc
)
add_library
(
gperftools::tcmalloc UNKNOWN IMPORTED
)
set_target_properties
(
gperftools::tcmalloc PROPERTIES
IMPORTED_LOCATION
${
GPERFTOOLS_TCMALLOC
}
INTERFACE_INCLUDE_DIRECTORIES
"
${
GPERFTOOLS_INCLUDE_DIR
}
"
)
add_library
(
gperftools::profiler UNKNOWN IMPORTED
)
set_target_properties
(
gperftools::profiler PROPERTIES
IMPORTED_LOCATION
${
GPERFTOOLS_PROFILER
}
INTERFACE_INCLUDE_DIRECTORIES
"
${
GPERFTOOLS_INCLUDE_DIR
}
"
)
endif
()
cmake/configure.cmake
浏览文件 @
ca84c2ca
...
...
@@ -86,6 +86,7 @@ endif(NOT WITH_GOLANG)
if
(
WITH_GPU
)
add_definitions
(
-DPADDLE_WITH_CUDA
)
add_definitions
(
-DEIGEN_USE_GPU
)
FIND_PACKAGE
(
CUDA REQUIRED
)
...
...
cmake/generic.cmake
浏览文件 @
ca84c2ca
...
...
@@ -110,6 +110,14 @@ function(find_fluid_modules TARGET_NAME)
endif
()
endfunction
(
find_fluid_modules
)
function
(
common_link TARGET_NAME
)
if
(
WITH_PROFILER
)
target_link_libraries
(
${
TARGET_NAME
}
gperftools::profiler
)
endif
()
endfunction
()
# find all third_party modules is used for paddle static library
# for reduce the dependency when building the inference libs.
set_property
(
GLOBAL PROPERTY FLUID_THIRD_PARTY
)
...
...
@@ -274,6 +282,7 @@ function(cc_library TARGET_NAME)
endif
()
target_link_libraries
(
${
TARGET_NAME
}
${
cc_library_DEPS
}
)
add_dependencies
(
${
TARGET_NAME
}
${
cc_library_DEPS
}
)
common_link
(
${
TARGET_NAME
}
)
endif
()
# cpplint code style
...
...
@@ -340,6 +349,7 @@ function(cc_binary TARGET_NAME)
if
(
cc_binary_DEPS
)
target_link_libraries
(
${
TARGET_NAME
}
${
cc_binary_DEPS
}
)
add_dependencies
(
${
TARGET_NAME
}
${
cc_binary_DEPS
}
)
common_link
(
${
TARGET_NAME
}
)
endif
()
endfunction
(
cc_binary
)
...
...
@@ -362,6 +372,7 @@ function(cc_test TARGET_NAME)
target_link_libraries
(
${
TARGET_NAME
}
${
win32_deps
}
)
endif
(
WIN32
)
add_dependencies
(
${
TARGET_NAME
}
${
cc_test_DEPS
}
paddle_gtest_main lod_tensor memory gtest gflags glog
)
common_link
(
${
TARGET_NAME
}
)
add_test
(
NAME
${
TARGET_NAME
}
COMMAND
${
TARGET_NAME
}
${
cc_test_ARGS
}
WORKING_DIRECTORY
${
CMAKE_CURRENT_BINARY_DIR
}
)
...
...
@@ -420,6 +431,7 @@ function(nv_binary TARGET_NAME)
if
(
nv_binary_DEPS
)
target_link_libraries
(
${
TARGET_NAME
}
${
nv_binary_DEPS
}
)
add_dependencies
(
${
TARGET_NAME
}
${
nv_binary_DEPS
}
)
common_link
(
${
TARGET_NAME
}
)
endif
()
endif
()
endfunction
(
nv_binary
)
...
...
@@ -433,6 +445,7 @@ function(nv_test TARGET_NAME)
cuda_add_executable
(
${
TARGET_NAME
}
${
nv_test_SRCS
}
)
target_link_libraries
(
${
TARGET_NAME
}
${
nv_test_DEPS
}
paddle_gtest_main lod_tensor memory gtest gflags glog
)
add_dependencies
(
${
TARGET_NAME
}
${
nv_test_DEPS
}
paddle_gtest_main lod_tensor memory gtest gflags glog
)
common_link
(
${
TARGET_NAME
}
)
add_test
(
${
TARGET_NAME
}
${
TARGET_NAME
}
)
if
(
nv_test_SERIAL
)
set_property
(
TEST
${
TARGET_NAME
}
PROPERTY RUN_SERIAL 1
)
...
...
@@ -499,6 +512,7 @@ function(hip_binary TARGET_NAME)
if
(
hip_binary_DEPS
)
target_link_libraries
(
${
TARGET_NAME
}
${
hip_binary_DEPS
}
)
add_dependencies
(
${
TARGET_NAME
}
${
hip_binary_DEPS
}
)
common_link
(
${
TARGET_NAME
}
)
endif
()
endif
()
endfunction
(
hip_binary
)
...
...
@@ -518,6 +532,7 @@ function(hip_test TARGET_NAME)
set_target_properties
(
${
TARGET_NAME
}
PROPERTIES LINKER_LANGUAGE HIP
)
target_link_libraries
(
${
TARGET_NAME
}
${
hip_test_DEPS
}
paddle_gtest_main memory gtest gflags
)
add_dependencies
(
${
TARGET_NAME
}
${
hip_test_DEPS
}
paddle_gtest_main memory gtest gflags
)
common_link
(
${
TARGET_NAME
}
)
add_test
(
${
TARGET_NAME
}
${
TARGET_NAME
}
)
endif
()
endfunction
(
hip_test
)
...
...
@@ -560,6 +575,7 @@ function(go_library TARGET_NAME)
endif
()
if
(
go_library_DEPS
)
add_dependencies
(
${
TARGET_NAME
}
${
go_library_DEPS
}
)
common_link
(
${
TARGET_NAME
}
)
endif
(
go_library_DEPS
)
# The "source file" of the library is `${dummyfile}` which never
...
...
paddle/fluid/framework/parallel_executor.cc
浏览文件 @
ca84c2ca
...
...
@@ -31,13 +31,36 @@ limitations under the License. */
#include "paddle/fluid/framework/details/threaded_ssa_graph_executor.h"
#include "paddle/fluid/platform/profiler.h"
#ifdef WITH_GPERFTOOLS
#include "gperftools/profiler.h"
#endif
DEFINE_string
(
pe_profile_fname
,
""
,
"Profiler filename for PE, which generated by gperftools."
"Only valid when compiled `WITH_PRIFILER=ON`. Empty if disable."
);
namespace
paddle
{
namespace
framework
{
static
std
::
once_flag
gProfileOnce
;
#ifdef WITH_GPERFTOOLS
static
bool
gProfileStarted
=
false
;
#endif
class
ParallelExecutorPrivate
{
public:
explicit
ParallelExecutorPrivate
(
const
std
::
vector
<
platform
::
Place
>
&
places
)
:
places_
(
places
)
{}
:
places_
(
places
)
{
if
(
!
FLAGS_pe_profile_fname
.
empty
())
{
std
::
call_once
(
gProfileOnce
,
[]
{
#ifdef WITH_GPERFTOOLS
ProfilerStart
(
FLAGS_pe_profile_fname
.
c_str
());
gProfileStarted
=
true
;
#else
LOG
(
WARNING
)
<<
"Paddle is not compiled with gperftools. "
"FLAGS_pe_profile_fname will be ignored"
;
#endif
});
}
}
~
ParallelExecutorPrivate
()
{
if
(
own_local_scope_
)
{
...
...
@@ -341,6 +364,12 @@ void ParallelExecutor::BCastParamsToDevices(
void
ParallelExecutor
::
Run
(
const
std
::
vector
<
std
::
string
>
&
fetch_tensors
,
const
std
::
string
&
fetched_var_name
)
{
#ifdef WITH_GPERFTOOLS
if
(
gProfileStarted
)
{
ProfilerFlush
();
}
#endif
platform
::
RecordBlock
b
(
0
);
if
(
member_
->
HasGarbageCollectors
())
{
member_
->
ResetRuntimeReferenceCount
(
fetch_tensors
,
fetched_var_name
);
...
...
paddle/fluid/operators/bilinear_tensor_product_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -12,7 +12,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/bilinear_tensor_product_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/cos_sim_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/cos_sim_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/crop_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/crop_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/dropout_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include <thrust/device_ptr.h>
#include <thrust/iterator/counting_iterator.h>
#include <thrust/random.h>
...
...
paddle/fluid/operators/elementwise/elementwise_add_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
#include "paddle/fluid/platform/float16.h"
...
...
paddle/fluid/operators/elementwise/elementwise_div_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_div_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/elementwise/elementwise_max_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_max_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/elementwise/elementwise_min_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_min_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/elementwise/elementwise_mul_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_mul_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/elementwise/elementwise_pow_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -8,8 +8,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_pow_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/elementwise/elementwise_sub_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/elementwise/elementwise_sub_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/expand_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/expand_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/gru_unit_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/gru_unit_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/hinge_loss_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/hinge_loss_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/huber_loss_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/huber_loss_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/im2sequence_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@
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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/im2sequence_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/isfinite_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@
// 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.
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/isfinite_op.h"
#include "paddle/fluid/platform/float16.h"
...
...
paddle/fluid/operators/l1_norm_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/l1_norm_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/log_loss_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/log_loss_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/math/context_project.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/math/context_project.h"
namespace
paddle
{
...
...
paddle/fluid/operators/math/jit_kernel_layer_norm.cc
浏览文件 @
ca84c2ca
...
...
@@ -79,16 +79,16 @@ class LayerNormKernelImpl : public LayerNormKernel<T> {
}
};
#define INTRIAVX_FLOAT(isa,
block)
\
#define INTRIAVX_FLOAT(isa,
jit_block)
\
template <> \
LayerNormKernelImpl<float, isa,
block>::LayerNormKernelImpl(int right)
\
LayerNormKernelImpl<float, isa,
jit_block>::LayerNormKernelImpl(int right)
\
: LayerNormKernel<float>() { \
this->num_ = right; \
this->rest_ = this->num_ % YMM_FLOAT_BLOCK; \
this->end_ = this->num_ - this->rest_; \
} \
template <> \
void LayerNormKernelImpl<float,
platform::avx, block>::Compute(
\
void LayerNormKernelImpl<float,
isa, jit_block>::Compute(
\
float* x, float* out, float* mean, float* var, const float* scale, \
const float* bias, int height, const float epsilon) const { \
__m256 sum; \
...
...
@@ -97,6 +97,7 @@ class LayerNormKernelImpl : public LayerNormKernel<T> {
__m256 tmp; \
size_t offset; \
size_t j; \
size_t block = YMM_FLOAT_BLOCK; \
__m256 reverse_num_vec = \
_mm256_div_ps(_mm256_set1_ps(1.0), _mm256_set1_ps(this->num_)); \
__m256 epsilon_vec = _mm256_set1_ps(epsilon); \
...
...
@@ -221,12 +222,14 @@ INTRIAVX_FLOAT(platform::avx, kEQ8);
INTRIAVX_FLOAT
(
platform
::
avx
,
kGT8LT16
);
INTRIAVX_FLOAT
(
platform
::
avx
,
kEQ16
);
INTRIAVX_FLOAT
(
platform
::
avx
,
kGT16
);
#endif
#ifdef __AVX2__
INTRIAVX_FLOAT
(
platform
::
avx2
,
kEQ8
);
INTRIAVX_FLOAT
(
platform
::
avx2
,
kGT8LT16
);
INTRIAVX_FLOAT
(
platform
::
avx2
,
kEQ16
);
INTRIAVX_FLOAT
(
platform
::
avx2
,
kGT16
);
INTRIAVX_FLOAT
(
platform
::
avx512f
,
kEQ8
);
INTRIAVX_FLOAT
(
platform
::
avx512f
,
kGT8LT16
);
INTRIAVX_FLOAT
(
platform
::
avx512f
,
kEQ16
);
INTRIAVX_FLOAT
(
platform
::
avx512f
,
kGT16
);
#endif
#undef INTRIAVX_FLOAT
...
...
paddle/fluid/operators/math/math_function.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include <vector>
#include "paddle/fluid/framework/data_type.h"
#include "paddle/fluid/operators/math/blas.h"
...
...
paddle/fluid/operators/math/sequence2batch.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/math/sequence2batch.h"
namespace
paddle
{
...
...
paddle/fluid/operators/math/softmax.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include <vector>
#include "paddle/fluid/operators/math/math_function.h"
...
...
paddle/fluid/operators/mean_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/mean_op.h"
#include "paddle/fluid/platform/float16.h"
...
...
paddle/fluid/operators/optimizers/adadelta_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/adadelta_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/adagrad_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/selected_rows_functor.h"
#include "paddle/fluid/operators/optimizers/adagrad_op.h"
...
...
paddle/fluid/operators/optimizers/adam_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/adam_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/adamax_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/adamax_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/decayed_adagrad_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/decayed_adagrad_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/ftrl_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -10,8 +10,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/ftrl_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/proximal_adagrad_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -10,8 +10,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/proximal_adagrad_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/proximal_gd_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -10,8 +10,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/proximal_gd_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/optimizers/rmsprop_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/optimizers/rmsprop_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/pad_constant_like_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/pad_constant_like_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/pad_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/pad_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/sequence_ops/sequence_pool_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/sequence_ops/sequence_pool_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/sigmoid_cross_entropy_with_logits_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/sigmoid_cross_entropy_with_logits_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/smooth_l1_loss_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/smooth_l1_loss_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/softmax_with_cross_entropy_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include <cub/cub.cuh>
#include "paddle/fluid/operators/math/cross_entropy.h"
#include "paddle/fluid/operators/softmax_with_cross_entropy_op.h"
...
...
paddle/fluid/operators/split_selected_rows_op.h
浏览文件 @
ca84c2ca
...
...
@@ -72,10 +72,11 @@ class SplitSelectedRowsOpKernel : public framework::OpKernel<T> {
for
(
size_t
i
=
0
;
i
<
outs_rows_idx
.
size
();
++
i
)
{
auto
rows_idx
=
outs_rows_idx
[
i
];
outs
[
i
]
->
set_height
(
height_sections
[
i
]);
auto
dims
=
x
->
GetCompleteDims
();
dims
[
0
]
=
rows_idx
.
size
();
outs
[
i
]
->
mutable_value
()
->
mutable_data
<
T
>
(
dims
,
x
->
place
());
outs
[
i
]
->
mutable_rows
()
->
clear
();
if
(
rows_idx
.
size
()
>
0
)
{
auto
dims
=
x
->
GetCompleteDims
();
dims
[
0
]
=
rows_idx
.
size
();
outs
[
i
]
->
mutable_value
()
->
mutable_data
<
T
>
(
dims
,
x
->
place
());
for
(
auto
idx
:
rows_idx
)
{
outs
[
i
]
->
mutable_rows
()
->
push_back
(
idx
-
abs_sections
[
i
]);
}
...
...
@@ -98,6 +99,8 @@ class SplitSelectedRowsOpKernel : public framework::OpKernel<T> {
}
}
}
PADDLE_ENFORCE_EQ
(
rows_idx
.
size
(),
outs
[
i
]
->
rows
().
size
(),
"rows should has the same size with tensor dim 0"
);
}
}
};
...
...
paddle/fluid/operators/squared_l2_distance_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,9 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/squared_l2_distance_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/squared_l2_norm_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -11,8 +11,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/squared_l2_norm_op.h"
namespace
ops
=
paddle
::
operators
;
...
...
paddle/fluid/operators/sum_op.cu
浏览文件 @
ca84c2ca
...
...
@@ -8,8 +8,6 @@ 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. */
#define EIGEN_USE_GPU
#include "paddle/fluid/operators/sum_op.h"
#include "paddle/fluid/platform/float16.h"
...
...
paddle/fluid/platform/cuda_helper_test.cu
浏览文件 @
ca84c2ca
...
...
@@ -93,7 +93,7 @@ TEST(CudaAtomic, float16) {
// unalignment of uint8
void
TestUnalign
(
size_t
num
,
const
int
shift_bit
)
{
PADDLE_ENFORCE
(
num
%
2
==
0
,
"must be a multiple of 2"
);
ASSERT_EQ
(
num
%
2
,
0
);
float16
*
in1
,
*
in2
,
*
out
;
float16
*
d_in1
,
*
d_in2
;
size_t
size
=
sizeof
(
uint8_t
)
*
(
num
+
shift_bit
);
...
...
paddle/fluid/platform/device_context.h
浏览文件 @
ca84c2ca
...
...
@@ -21,7 +21,6 @@ limitations under the License. */
#include "paddle/fluid/platform/dynload/cublas.h"
#include "paddle/fluid/platform/dynload/cudnn.h"
#include "paddle/fluid/platform/gpu_info.h"
#define EIGEN_USE_GPU
#endif
#ifdef PADDLE_WITH_MKLDNN
...
...
paddle/fluid/platform/enforce.h
浏览文件 @
ca84c2ca
...
...
@@ -62,45 +62,54 @@ inline std::string demangle(std::string name) { return name; }
#endif
struct
EnforceNotMet
:
public
std
::
exception
{
std
::
exception_ptr
exp_
;
std
::
string
err_str_
;
EnforceNotMet
(
std
::
exception_ptr
e
,
const
char
*
f
,
int
l
)
:
exp_
(
e
)
{
static
constexpr
int
TRACE_STACK_LIMIT
=
100
;
EnforceNotMet
(
std
::
exception_ptr
e
,
const
char
*
f
,
int
l
)
{
try
{
std
::
rethrow_exception
(
exp_
);
}
catch
(
const
std
::
exception
&
exp
)
{
std
::
ostringstream
sout
;
std
::
rethrow_exception
(
e
);
}
catch
(
std
::
exception
&
e
)
{
Init
(
e
.
what
(),
f
,
l
);
}
}
sout
<<
string
::
Sprintf
(
"%s at [%s:%d]"
,
exp
.
what
(),
f
,
l
)
<<
std
::
endl
;
sout
<<
"PaddlePaddle Call Stacks: "
<<
std
::
endl
;
template
<
typename
...
ARGS
>
EnforceNotMet
(
const
char
*
f
,
int
l
,
ARGS
...
args
)
{
Init
(
string
::
Sprintf
(
args
...),
f
,
l
);
}
const
char
*
what
()
const
noexcept
override
{
return
err_str_
.
c_str
();
}
private:
template
<
typename
StrType
>
inline
void
Init
(
StrType
what
,
const
char
*
f
,
int
l
)
{
static
constexpr
int
TRACE_STACK_LIMIT
=
100
;
std
::
ostringstream
sout
;
sout
<<
string
::
Sprintf
(
"%s at [%s:%d]"
,
what
,
f
,
l
)
<<
std
::
endl
;
sout
<<
"PaddlePaddle Call Stacks: "
<<
std
::
endl
;
#if !defined(_WIN32)
void
*
call_stack
[
TRACE_STACK_LIMIT
];
auto
size
=
backtrace
(
call_stack
,
TRACE_STACK_LIMIT
);
auto
symbols
=
backtrace_symbols
(
call_stack
,
size
);
Dl_info
info
;
for
(
int
i
=
0
;
i
<
size
;
++
i
)
{
if
(
dladdr
(
call_stack
[
i
],
&
info
)
&&
info
.
dli_sname
)
{
auto
demangled
=
demangle
(
info
.
dli_sname
);
auto
addr_offset
=
static_cast
<
char
*>
(
call_stack
[
i
])
-
static_cast
<
char
*>
(
info
.
dli_saddr
);
sout
<<
string
::
Sprintf
(
"%-3d %*0p %s + %zd
\n
"
,
i
,
2
+
sizeof
(
void
*
)
*
2
,
call_stack
[
i
],
demangled
,
addr_offset
);
}
else
{
sout
<<
string
::
Sprintf
(
"%-3d %*0p
\n
"
,
i
,
2
+
sizeof
(
void
*
)
*
2
,
call_stack
[
i
]);
}
void
*
call_stack
[
TRACE_STACK_LIMIT
];
auto
size
=
backtrace
(
call_stack
,
TRACE_STACK_LIMIT
);
auto
symbols
=
backtrace_symbols
(
call_stack
,
size
);
Dl_info
info
;
for
(
int
i
=
0
;
i
<
size
;
++
i
)
{
if
(
dladdr
(
call_stack
[
i
],
&
info
)
&&
info
.
dli_sname
)
{
auto
demangled
=
demangle
(
info
.
dli_sname
);
auto
addr_offset
=
static_cast
<
char
*>
(
call_stack
[
i
])
-
static_cast
<
char
*>
(
info
.
dli_saddr
);
sout
<<
string
::
Sprintf
(
"%-3d %*0p %s + %zd
\n
"
,
i
,
2
+
sizeof
(
void
*
)
*
2
,
call_stack
[
i
],
demangled
,
addr_offset
);
}
else
{
sout
<<
string
::
Sprintf
(
"%-3d %*0p
\n
"
,
i
,
2
+
sizeof
(
void
*
)
*
2
,
call_stack
[
i
]);
}
free
(
symbols
);
}
free
(
symbols
);
#else
sout
<<
"Windows not support stack backtrace yet."
;
sout
<<
"Windows not support stack backtrace yet."
;
#endif
err_str_
=
sout
.
str
();
}
err_str_
=
sout
.
str
();
}
const
char
*
what
()
const
noexcept
{
return
err_str_
.
c_str
();
}
};
struct
EOFException
:
public
std
::
exception
{
...
...
@@ -242,13 +251,8 @@ inline void throw_on_error(T e) {
throw_on_error
(
e
,
""
);
}
#define PADDLE_THROW(...) \
do { \
throw ::paddle::platform::EnforceNotMet( \
std::make_exception_ptr( \
std::runtime_error(paddle::string::Sprintf(__VA_ARGS__))), \
__FILE__, __LINE__); \
} while (false)
#define PADDLE_THROW(...) \
throw ::paddle::platform::EnforceNotMet(__FILE__, __LINE__, __VA_ARGS__)
#ifndef REPLACE_ENFORCE_GLOG
#define PADDLE_ENFORCE(...) \
...
...
paddle/fluid/platform/float16.h
浏览文件 @
ca84c2ca
...
...
@@ -71,9 +71,6 @@ struct float16;
}
// namespace platform
}
// namespace paddle
// NOTE():
// Do not move the eigen.h header, otherwise the eigen_vector<bool> will failed.
#include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/platform/hostdevice.h"
#include "unsupported/Eigen/CXX11/Tensor"
...
...
paddle/fluid/pybind/pybind.cc
浏览文件 @
ca84c2ca
...
...
@@ -336,6 +336,8 @@ PYBIND11_MODULE(core, m) {
.
def
(
"get_tensor"
,
[](
SelectedRows
&
self
)
{
return
self
.
mutable_value
();
},
py
::
return_value_policy
::
reference
)
.
def
(
"numel"
,
[](
SelectedRows
&
self
)
->
int64_t
{
return
self
.
value
().
numel
();
})
.
def
(
"set_height"
,
&
SelectedRows
::
set_height
)
.
def
(
"height"
,
&
SelectedRows
::
height
)
.
def
(
"set_rows"
,
...
...
python/paddle/fluid/__init__.py
浏览文件 @
ca84c2ca
...
...
@@ -128,7 +128,7 @@ def __bootstrap__():
'free_idle_memory'
,
'paddle_num_threads'
,
"dist_threadpool_size"
,
'eager_delete_tensor_gb'
,
'fast_eager_deletion_mode'
,
'allocator_strategy'
,
'reader_queue_speed_test_mode'
,
'print_sub_graph_dir'
'print_sub_graph_dir'
,
'pe_profile_fname'
]
if
'Darwin'
not
in
sysstr
:
read_env_flags
.
append
(
'use_pinned_memory'
)
...
...
python/paddle/fluid/average.py
浏览文件 @
ca84c2ca
...
...
@@ -48,6 +48,7 @@ class WeightedAverage(object):
Examples:
.. code-block:: python
avg = fluid.average.WeightedAverage()
avg.add(value=2.0, weight=1)
avg.add(value=4.0, weight=2)
...
...
python/paddle/fluid/tests/unittests/test_split_selected_rows_op.py
浏览文件 @
ca84c2ca
...
...
@@ -63,6 +63,7 @@ class TestSpliteSelectedRows(unittest.TestCase):
# expected output selected rows
expected_out0_rows
=
[
0
,
4
]
expected_out1_rows
=
[
0
,
2
]
expected_out2_rows
=
[]
expected_out4_rows
=
[
0
]
op
=
Operator
(
...
...
@@ -75,6 +76,7 @@ class TestSpliteSelectedRows(unittest.TestCase):
self
.
assertEqual
(
outs
[
0
].
rows
(),
expected_out0_rows
)
self
.
assertEqual
(
outs
[
1
].
rows
(),
expected_out1_rows
)
self
.
assertEqual
(
outs
[
2
].
rows
(),
expected_out2_rows
)
self
.
assertEqual
(
outs
[
4
].
rows
(),
expected_out4_rows
)
self
.
assertEqual
(
outs
[
0
].
height
(),
height_sections
[
0
])
...
...
@@ -84,6 +86,9 @@ class TestSpliteSelectedRows(unittest.TestCase):
self
.
assertAlmostEqual
(
4.0
,
np
.
array
(
outs
[
1
].
get_tensor
())[
1
,
1
])
self
.
assertAlmostEqual
(
8.0
,
np
.
array
(
outs
[
4
].
get_tensor
())[
0
,
1
])
self
.
assertEqual
(
outs
[
2
].
numel
(),
0
)
self
.
assertEqual
(
outs
[
3
].
numel
(),
0
)
def
check_grad_with_place
(
self
,
place
):
scope
=
core
.
Scope
()
height
=
10
...
...
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