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
机器未来
Paddle
提交
cb124156
P
Paddle
项目概览
机器未来
/
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
未验证
提交
cb124156
编写于
4月 02, 2022
作者:
L
Leo Chen
提交者:
GitHub
4月 02, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[new-exec] support to enable mkldnn by flags (#41274)
上级
0fe2001a
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
21 addition
and
3 deletion
+21
-3
paddle/fluid/framework/new_executor/interpretercore.cc
paddle/fluid/framework/new_executor/interpretercore.cc
+8
-3
paddle/fluid/framework/new_executor/interpretercore_util.cc
paddle/fluid/framework/new_executor/interpretercore_util.cc
+13
-0
未找到文件。
paddle/fluid/framework/new_executor/interpretercore.cc
浏览文件 @
cb124156
...
@@ -425,13 +425,18 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
...
@@ -425,13 +425,18 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
:
global_scope_
->
GetMutableScope
();
:
global_scope_
->
GetMutableScope
();
auto
op_with_kernel
=
dynamic_cast
<
const
framework
::
OperatorWithKernel
*>
(
op
);
auto
op_with_kernel
=
dynamic_cast
<
const
framework
::
OperatorWithKernel
*>
(
op
);
{
{
// If it is OperatorBase, InferShape do nothing.
if
(
op_with_kernel
!=
nullptr
)
{
if
(
op_with_kernel
!=
nullptr
)
{
platform
::
RecordEvent
infershape_event
(
platform
::
RecordEvent
infershape_event
(
"infer_shape"
,
platform
::
TracerEventType
::
OperatorInner
,
1
,
"infer_shape"
,
platform
::
TracerEventType
::
OperatorInner
,
1
,
platform
::
EventRole
::
kInnerOp
);
platform
::
EventRole
::
kInnerOp
);
// If it is OperatorBase, InferShape do nothing.
op_with_kernel
->
Info
().
infer_shape_
(
// see OperatorWithKernel::RunImpl in operator.cc for why
instr_node
.
InnerInferShapeContext
().
get
());
if
(
!
(
op_with_kernel
->
HasAttr
(
kAllKernelsMustComputeRuntimeShape
)
&&
op_with_kernel
->
Attr
<
bool
>
(
kAllKernelsMustComputeRuntimeShape
)))
{
op_with_kernel
->
Info
().
infer_shape_
(
instr_node
.
InnerInferShapeContext
().
get
());
}
}
}
}
}
...
...
paddle/fluid/framework/new_executor/interpretercore_util.cc
浏览文件 @
cb124156
...
@@ -29,6 +29,8 @@ PADDLE_DEFINE_EXPORTED_bool(
...
@@ -29,6 +29,8 @@ PADDLE_DEFINE_EXPORTED_bool(
new_executor_sequential_run
,
false
,
new_executor_sequential_run
,
false
,
"Enable sequential execution for standalone executor, used for debug"
);
"Enable sequential execution for standalone executor, used for debug"
);
DECLARE_bool
(
use_mkldnn
);
namespace
paddle
{
namespace
paddle
{
namespace
framework
{
namespace
framework
{
namespace
interpreter
{
namespace
interpreter
{
...
@@ -192,6 +194,7 @@ void create_all_ops(const framework::BlockDesc& block,
...
@@ -192,6 +194,7 @@ void create_all_ops(const framework::BlockDesc& block,
const
VariableNameMap
&
inputs_names
=
op
->
Inputs
();
const
VariableNameMap
&
inputs_names
=
op
->
Inputs
();
const
VariableNameMap
&
outputs_names
=
op
->
Outputs
();
const
VariableNameMap
&
outputs_names
=
op
->
Outputs
();
AttributeMap
op_attr_map
=
op
->
GetAttrMap
();
AttributeMap
op_attr_map
=
op
->
GetAttrMap
();
if
(
info
.
Checker
()
!=
nullptr
)
{
if
(
info
.
Checker
()
!=
nullptr
)
{
...
@@ -199,6 +202,16 @@ void create_all_ops(const framework::BlockDesc& block,
...
@@ -199,6 +202,16 @@ void create_all_ops(const framework::BlockDesc& block,
}
}
auto
op_base
=
auto
op_base
=
info
.
Creator
()(
op
->
Type
(),
inputs_names
,
outputs_names
,
op_attr_map
);
info
.
Creator
()(
op
->
Type
(),
inputs_names
,
outputs_names
,
op_attr_map
);
#ifdef PADDLE_WITH_MKLDNN
if
(
FLAGS_use_mkldnn
)
{
if
(
op
->
HasAttr
(
"use_mkldnn"
))
{
VLOG
(
4
)
<<
"Set use_mkldnn=True for "
<<
op_base
->
Type
();
op_base
->
SetAttr
(
"use_mkldnn"
,
true
);
}
}
#endif
ops
->
emplace_back
(
std
::
unique_ptr
<
OperatorBase
>
(
op_base
));
ops
->
emplace_back
(
std
::
unique_ptr
<
OperatorBase
>
(
op_base
));
}
}
}
}
...
...
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