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
urhero
Paddle
提交
4efdebc6
P
Paddle
项目概览
urhero
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
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
0
Issue
0
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 >>
未验证
提交
4efdebc6
编写于
2月 28, 2019
作者:
T
Tao Luo
提交者:
GitHub
2月 28, 2019
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #15931 from yihuaxu/develop_
2c5c7b2a
_gelu_mkl_opt
Optimize gelu operation with mkl erf
上级
e5f9d3a4
73967886
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
59 addition
and
2 deletion
+59
-2
cmake/external/mklml.cmake
cmake/external/mklml.cmake
+4
-2
paddle/fluid/operators/activation_op.h
paddle/fluid/operators/activation_op.h
+22
-0
paddle/fluid/operators/math/blas.h
paddle/fluid/operators/math/blas.h
+8
-0
paddle/fluid/operators/math/blas_impl.h
paddle/fluid/operators/math/blas_impl.h
+23
-0
paddle/fluid/platform/dynload/mklml.h
paddle/fluid/platform/dynload/mklml.h
+2
-0
未找到文件。
cmake/external/mklml.cmake
浏览文件 @
4efdebc6
...
...
@@ -39,8 +39,10 @@ IF(WIN32)
SET
(
MKLML_IOMP_LIB
${
MKLML_LIB_DIR
}
/libiomp5md.lib
)
SET
(
MKLML_SHARED_LIB
${
MKLML_LIB_DIR
}
/mklml.dll
)
SET
(
MKLML_SHARED_IOMP_LIB
${
MKLML_LIB_DIR
}
/libiomp5md.dll
)
ELSE
()
SET
(
MKLML_VER
"mklml_lnx_
${
TIME_VERSION
}
"
CACHE STRING
""
FORCE
)
ELSE
()
#TODO(intel-huying):
# Now enable Erf function in mklml library temporarily, it will be updated as offical version later.
SET
(
MKLML_VER
"Glibc225_vsErf_mklml_lnx_
${
TIME_VERSION
}
"
CACHE STRING
""
FORCE
)
SET
(
MKLML_URL
"http://paddlepaddledeps.cdn.bcebos.com/
${
MKLML_VER
}
.tgz"
CACHE STRING
""
FORCE
)
SET
(
MKLML_LIB
${
MKLML_LIB_DIR
}
/libmklml_intel.so
)
SET
(
MKLML_IOMP_LIB
${
MKLML_LIB_DIR
}
/libiomp5.so
)
...
...
paddle/fluid/operators/activation_op.h
浏览文件 @
4efdebc6
...
...
@@ -11,6 +11,7 @@ limitations under the License. */
#pragma once
#include <glog/logging.h>
#include <algorithm>
#include <string>
#include <unordered_set>
#include <utility>
...
...
@@ -24,6 +25,7 @@ limitations under the License. */
#include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/operators/detail/safe_ref.h"
#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/platform/float16.h"
#ifdef PADDLE_WITH_MKLDNN
...
...
@@ -300,8 +302,28 @@ template <typename T>
struct
GeluFunctor
:
public
BaseActivationFunctor
<
T
>
{
template
<
typename
Device
,
typename
X
,
typename
Out
>
void
operator
()(
Device
d
,
X
x
,
Out
out
)
const
{
// Because the execute or device context can not be deliver here, it keep the
// marco for NVCC.
#if defined(PADDLE_WITH_MKLML) && !defined(_WIN32) && !defined(__APPLE__) && \
!defined(__OSX__) && !defined(PADDLE_WITH_CUDA)
auto
x_data
=
x
.
data
();
auto
out_data
=
out
.
data
();
int
n
=
std
::
min
(
x
.
size
(),
out
.
size
());
std
::
memset
(
out_data
,
0
,
n
*
sizeof
(
T
));
math
::
CBlas
<
T
>::
AXPY
(
n
,
static_cast
<
T
>
(
M_SQRT1_2
),
x_data
,
1
,
out_data
,
1
);
math
::
CBlas
<
T
>::
VMERF
(
n
,
out_data
,
out_data
,
VML_LA
);
for
(
int
i
=
0
;
i
<
n
;
i
++
)
{
out_data
[
i
]
+=
static_cast
<
T
>
(
1
);
}
math
::
CBlas
<
T
>::
VMUL
(
n
,
x_data
,
out_data
,
out_data
);
for
(
int
i
=
0
;
i
<
n
;
i
++
)
{
out_data
[
i
]
*=
static_cast
<
T
>
(
0.5
);
}
#else
auto
temp
=
(
x
*
static_cast
<
T
>
(
M_SQRT1_2
)).
erf
();
out
.
device
(
d
)
=
x
*
static_cast
<
T
>
(
0.5
)
*
(
static_cast
<
T
>
(
1
)
+
temp
);
#endif
}
};
...
...
paddle/fluid/operators/math/blas.h
浏览文件 @
4efdebc6
...
...
@@ -184,6 +184,9 @@ class Blas {
template
<
typename
T
>
void
VINV
(
int
n
,
const
T
*
a
,
T
*
y
)
const
;
template
<
typename
T
>
void
VMERF
(
int
n
,
const
T
*
a
,
T
*
y
,
int64_t
mode
)
const
;
private:
const
DeviceContext
&
context_
;
};
...
...
@@ -290,6 +293,11 @@ class BlasT : private Blas<DeviceContext> {
Base
()
->
template
VINV
<
T
>(
args
...);
}
template
<
typename
...
ARGS
>
void
VMERF
(
ARGS
...
args
)
const
{
Base
()
->
template
VMERF
<
T
>(
args
...);
}
private:
const
Blas
<
DeviceContext
>*
Base
()
const
{
return
static_cast
<
const
Blas
<
DeviceContext
>*>
(
this
);
...
...
paddle/fluid/operators/math/blas_impl.h
浏览文件 @
4efdebc6
...
...
@@ -123,6 +123,11 @@ struct CBlas<float> {
static
void
VINV
(
ARGS
...
args
)
{
platform
::
dynload
::
vsInv
(
args
...);
}
template
<
typename
...
ARGS
>
static
void
VMERF
(
ARGS
...
args
)
{
platform
::
dynload
::
vmsErf
(
args
...);
}
};
template
<
>
...
...
@@ -223,6 +228,11 @@ struct CBlas<double> {
static
void
VINV
(
ARGS
...
args
)
{
platform
::
dynload
::
vdInv
(
args
...);
}
template
<
typename
...
ARGS
>
static
void
VMERF
(
ARGS
...
args
)
{
platform
::
dynload
::
vmdErf
(
args
...);
}
};
#else
...
...
@@ -625,6 +635,19 @@ void Blas<DeviceContext>::VINV(int n, const T *a, T *y) const {
#endif
}
template
<
>
template
<
typename
T
>
void
Blas
<
platform
::
CPUDeviceContext
>::
VMERF
(
int
n
,
const
T
*
a
,
T
*
y
,
int64_t
mode
)
const
{
#ifdef PADDLE_WITH_MKLML
CBlas
<
T
>::
VMERF
(
n
,
a
,
y
,
mode
);
#else
for
(
int
i
=
0
;
i
<
n
;
++
i
)
{
y
[
i
]
=
std
::
erf
(
a
[
i
]);
}
#endif
}
}
// namespace math
}
// namespace operators
}
// namespace paddle
paddle/fluid/platform/dynload/mklml.h
浏览文件 @
4efdebc6
...
...
@@ -86,6 +86,8 @@ extern void* mklml_dso_handle;
__macro(vdPowx); \
__macro(vsInv); \
__macro(vdInv); \
__macro(vmsErf); \
__macro(vmdErf); \
__macro(MKL_Set_Num_Threads)
MKLML_ROUTINE_EACH
(
DECLARE_DYNAMIC_LOAD_MKLML_WRAP
);
...
...
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