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
BaiXuePrincess
Paddle
提交
f72d52e7
P
Paddle
项目概览
BaiXuePrincess
/
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
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
未验证
提交
f72d52e7
编写于
9月 22, 2021
作者:
W
Wilber
提交者:
GitHub
9月 22, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[cherry-pick] trt engine dtor when the last predictor dtor (#35881)
* cherry-pick 32842
上级
edeb0ade
变更
7
隐藏空白更改
内联
并排
Showing
7 changed file
with
116 addition
and
10 deletion
+116
-10
cmake/external/lite.cmake
cmake/external/lite.cmake
+1
-1
paddle/fluid/framework/ir/fc_fuse_pass.cc
paddle/fluid/framework/ir/fc_fuse_pass.cc
+1
-1
paddle/fluid/framework/ir/map_matmul_to_mul_pass.cc
paddle/fluid/framework/ir/map_matmul_to_mul_pass.cc
+4
-4
paddle/fluid/framework/op_desc.h
paddle/fluid/framework/op_desc.h
+1
-1
paddle/fluid/inference/api/analysis_predictor.cc
paddle/fluid/inference/api/analysis_predictor.cc
+11
-1
paddle/fluid/inference/tensorrt/op_teller.cc
paddle/fluid/inference/tensorrt/op_teller.cc
+96
-0
paddle/scripts/paddle_build.sh
paddle/scripts/paddle_build.sh
+2
-2
未找到文件。
cmake/external/lite.cmake
浏览文件 @
f72d52e7
...
...
@@ -42,7 +42,7 @@ if (NOT LITE_SOURCE_DIR OR NOT LITE_BINARY_DIR)
set
(
LITE_INSTALL_DIR
${
THIRD_PARTY_PATH
}
/install/lite
)
if
(
NOT LITE_GIT_TAG
)
set
(
LITE_GIT_TAG
d3a3a6931b6d22d504d21ba32b3ae972770e920
4
)
set
(
LITE_GIT_TAG
1c4698c6efd9a5f57a4f8369bd5b6374166f5ba
4
)
endif
()
if
(
NOT CUDA_ARCH_NAME
)
...
...
paddle/fluid/framework/ir/fc_fuse_pass.cc
浏览文件 @
f72d52e7
...
...
@@ -135,7 +135,7 @@ int FCFusePass::ApplyFCPattern(Graph* graph, bool with_relu) const {
}
// Create an FC Node.
OpDesc
desc
;
OpDesc
desc
(
mul
->
Op
()
->
Block
())
;
desc
.
SetType
(
"fc"
);
// Set inputs of fc
...
...
paddle/fluid/framework/ir/map_matmul_to_mul_pass.cc
浏览文件 @
f72d52e7
...
...
@@ -220,7 +220,7 @@ void MapMatmul2MulPass::ApplyImpl(ir::Graph* graph) const {
LOG
(
WARNING
)
<<
"Pass in op compat failed."
;
return
;
}
OpDesc
desc
;
OpDesc
desc
(
matmul_op
->
Op
()
->
Block
())
;
desc
.
SetType
(
"mul"
);
desc
.
SetInput
(
"X"
,
{
matmul_in_x
->
Name
()});
desc
.
SetInput
(
"Y"
,
{
matmul_in_y
->
Name
()});
...
...
@@ -299,7 +299,7 @@ void Squeeze2MatmulFusePass::ApplyImpl(ir::Graph* graph) const {
LOG
(
WARNING
)
<<
"Pass in op compat failed."
;
return
;
}
OpDesc
desc
;
OpDesc
desc
(
matmul_op
->
Op
()
->
Block
())
;
desc
.
SetType
(
"mul"
);
desc
.
SetInput
(
"X"
,
{
squeeze2_in_x
->
Name
()});
desc
.
SetInput
(
"Y"
,
{
matmul_in_y
->
Name
()});
...
...
@@ -441,7 +441,7 @@ void Reshape2MatmulFusePass::ApplyImpl(ir::Graph* graph) const {
LOG
(
WARNING
)
<<
"Pass in op compat failed."
;
return
;
}
OpDesc
desc
;
OpDesc
desc
(
matmul_op
->
Op
()
->
Block
())
;
desc
.
SetType
(
"mul"
);
desc
.
SetInput
(
"X"
,
{
reshape2_in_x
->
Name
()});
desc
.
SetInput
(
"Y"
,
{
matmul_in_y
->
Name
()});
...
...
@@ -526,7 +526,7 @@ void Flatten2MatmulFusePass::ApplyImpl(ir::Graph* graph) const {
LOG
(
WARNING
)
<<
"Pass in op compat failed."
;
return
;
}
OpDesc
desc
;
OpDesc
desc
(
matmul_op
->
Op
()
->
Block
())
;
desc
.
SetType
(
"mul"
);
desc
.
SetInput
(
"X"
,
{
flatten2_in_x
->
Name
()});
desc
.
SetInput
(
"Y"
,
{
matmul_in_y
->
Name
()});
...
...
paddle/fluid/framework/op_desc.h
浏览文件 @
f72d52e7
...
...
@@ -178,7 +178,7 @@ class OpDesc {
}
proto
::
OpDesc
desc_
;
BlockDesc
*
block_
;
// not_own
BlockDesc
*
block_
{
nullptr
}
;
// not_own
// input arg name => input variable names
VariableNameMap
inputs_
;
// output arg name => output variable names
...
...
paddle/fluid/inference/api/analysis_predictor.cc
浏览文件 @
f72d52e7
...
...
@@ -645,7 +645,17 @@ void AnalysisPredictor::OptimizeInferenceProgram() {
VLOG
(
5
)
<<
"to prepare executor"
;
ARGUMENT_CHECK_FIELD
((
&
argument_
),
ir_analyzed_program
);
inference_program_
.
reset
(
new
framework
::
ProgramDesc
(
argument_
.
ir_analyzed_program
()));
new
framework
::
ProgramDesc
(
argument_
.
ir_analyzed_program
()),
[](
framework
::
ProgramDesc
*
prog
)
{
// Note, please do NOT use any member variables, because member variables may
// have been destructed in multiple threads.
#if PADDLE_WITH_TENSORRT
paddle
::
inference
::
Singleton
<
inference
::
tensorrt
::
TRTEngineManager
>::
Global
()
.
DeleteAll
();
#endif
delete
prog
;
});
// The config and argument take a lot of storage,
// when the predictor settings are complete, we release these stores.
argument_
.
PartiallyRelease
();
...
...
paddle/fluid/inference/tensorrt/op_teller.cc
浏览文件 @
f72d52e7
...
...
@@ -159,6 +159,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
op_type
==
"relu"
||
op_type
==
"relu6"
||
op_type
==
"tanh"
||
op_type
==
"sigmoid"
)
{
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -274,6 +280,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
op_type
==
"matmul"
)
{
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
for
(
auto
&
param_name
:
desc
.
Inputs
())
{
for
(
auto
&
var_name
:
param_name
.
second
)
{
auto
*
var_desc
=
block
->
FindVar
(
var_name
);
...
...
@@ -324,6 +336,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
axis
[
0
]
==
0
&&
axis
.
size
()
==
2
)
return
false
;
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -372,6 +390,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
return
false
;
}
else
{
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
*
x_var_desc
=
block
->
FindVar
(
desc
.
Input
(
"X"
)[
0
]);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
if
(
x_shape
.
size
()
==
1
)
{
...
...
@@ -385,6 +409,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
!
with_dynamic_shape
)
return
false
;
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
index_var_name
=
desc
.
Input
(
"Index"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
...
...
@@ -428,6 +458,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
data_layout
!=
framework
::
DataLayout
::
kNCHW
)
return
false
;
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -439,6 +475,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
op_type
==
"multiclass_nms"
)
{
if
(
with_dynamic_shape
)
return
false
;
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
for
(
auto
&
param_name
:
desc
.
Inputs
())
{
for
(
auto
&
var_name
:
param_name
.
second
)
{
auto
*
var_desc
=
block
->
FindVar
(
var_name
);
...
...
@@ -598,6 +640,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
return
false
;
}
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -657,6 +705,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
}
}
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -724,6 +778,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
return
false
;
}
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
*
x_var_desc
=
block
->
FindVar
(
desc
.
Input
(
"X"
)[
0
]);
auto
*
y_var_desc
=
block
->
FindVar
(
desc
.
Input
(
"Y"
)[
0
]);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -775,6 +835,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
}
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -856,6 +922,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
}
std
::
vector
<
int64_t
>
shape
;
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
for
(
auto
&
param_name
:
desc
.
Inputs
())
{
for
(
auto
&
var_name
:
param_name
.
second
)
{
auto
*
var_desc
=
block
->
FindVar
(
var_name
);
...
...
@@ -881,6 +953,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
op_type
==
"scale"
)
{
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -892,6 +970,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
if
(
op_type
==
"swish"
)
{
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
@@ -916,6 +1000,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
}
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
*
var_desc
=
block
->
FindVar
(
desc
.
Input
(
"Alpha"
)[
0
]);
if
(
!
var_desc
)
{
VLOG
(
3
)
<<
"Variable Alpha of prelu TRT converter not found."
;
...
...
@@ -1051,6 +1141,12 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
}
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
"Developers need to check whether block_desc is passed in "
"the pass."
;
return
false
;
}
auto
x_var_name
=
desc
.
Input
(
"X"
)[
0
];
auto
*
x_var_desc
=
block
->
FindVar
(
x_var_name
);
const
auto
x_shape
=
x_var_desc
->
GetShape
();
...
...
paddle/scripts/paddle_build.sh
浏览文件 @
f72d52e7
...
...
@@ -1161,8 +1161,8 @@ function parallel_test_base_gpu() {
EOF
set
-x
# set trt_convert ut to run
30
% cases.
export
TEST_NUM_PERCENT_CASES
=
0.
3
# set trt_convert ut to run
15
% cases.
export
TEST_NUM_PERCENT_CASES
=
0.
15
precison_cases
=
""
bash
$PADDLE_ROOT
/tools/check_added_ut.sh
if
[
${
PRECISION_TEST
:-
OFF
}
==
"ON"
]
;
then
...
...
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