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
wmsofts
Paddle
提交
83c85f34
P
Paddle
项目概览
wmsofts
/
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
83c85f34
编写于
8月 14, 2018
作者:
Z
Zhaolong Xing
提交者:
GitHub
8月 14, 2018
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #12598 from NHZlX/add_tensorrt_softmax
Add tensorrt softmax
上级
1e1974c9
29ad9794
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
120 addition
and
9 deletion
+120
-9
paddle/fluid/inference/analysis/analyzer.cc
paddle/fluid/inference/analysis/analyzer.cc
+1
-1
paddle/fluid/inference/tensorrt/convert/CMakeLists.txt
paddle/fluid/inference/tensorrt/convert/CMakeLists.txt
+4
-1
paddle/fluid/inference/tensorrt/convert/softmax_op.cc
paddle/fluid/inference/tensorrt/convert/softmax_op.cc
+49
-0
paddle/fluid/inference/tensorrt/convert/test_softmax_op.cc
paddle/fluid/inference/tensorrt/convert/test_softmax_op.cc
+49
-0
paddle/fluid/inference/tensorrt/convert/ut_helper.h
paddle/fluid/inference/tensorrt/convert/ut_helper.h
+17
-7
未找到文件。
paddle/fluid/inference/analysis/analyzer.cc
浏览文件 @
83c85f34
...
...
@@ -44,7 +44,7 @@ class DfgPassManagerImpl final : public DfgPassManager {
if
(
FLAGS_inference_analysis_enable_tensorrt_subgraph_engine
)
{
auto
trt_teller
=
[
&
](
const
Node
*
node
)
{
std
::
unordered_set
<
std
::
string
>
teller_set
(
{
"elementwise_add"
,
"mul"
,
"conv2d"
,
"pool2d"
,
"relu"
});
{
"elementwise_add"
,
"mul"
,
"conv2d"
,
"pool2d"
,
"relu"
,
"softmax"
});
if
(
!
node
->
IsFunction
())
return
false
;
const
auto
*
func
=
static_cast
<
const
Function
*>
(
node
);
...
...
paddle/fluid/inference/tensorrt/convert/CMakeLists.txt
浏览文件 @
83c85f34
# Add TRT tests
nv_library
(
tensorrt_converter
SRCS mul_op.cc conv2d_op.cc fc_op.cc pool2d_op.cc elementwise_op.cc
activation_op.cc
activation_op.cc
softmax_op.cc
DEPS tensorrt_engine operator scope framework_proto op_registry
)
nv_test
(
test_op_converter SRCS test_op_converter.cc DEPS
...
...
@@ -21,3 +21,6 @@ nv_test(test_trt_pool2d_op SRCS test_pool2d_op.cc pool2d_op.cc
nv_test
(
test_trt_elementwise_op SRCS test_elementwise_op.cc elementwise_op.cc
DEPS
${
FLUID_CORE_MODULES
}
tensorrt_engine elementwise_add_op SERIAL
)
nv_test
(
test_trt_softmax_op SRCS test_softmax_op.cc softmax_op.cc
DEPS
${
FLUID_CORE_MODULES
}
tensorrt_engine softmax_op SERIAL
)
paddle/fluid/inference/tensorrt/convert/softmax_op.cc
0 → 100644
浏览文件 @
83c85f34
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
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/fluid/inference/tensorrt/convert/op_converter.h"
namespace
paddle
{
namespace
inference
{
namespace
tensorrt
{
/*
* SoftMaxOp, ISoftMaxLayer in TRT. This Layer doesn't has weights.
*/
class
SoftMaxOpConverter
:
public
OpConverter
{
public:
void
operator
()(
const
framework
::
proto
::
OpDesc
&
op
,
const
framework
::
Scope
&
scope
,
bool
test_mode
)
override
{
VLOG
(
4
)
<<
"convert a fluid softmax op to tensorrt softmax layer without bias"
;
framework
::
OpDesc
op_desc
(
op
,
nullptr
);
// Declare inputs
auto
*
input1
=
engine_
->
GetITensor
(
op_desc
.
Input
(
"X"
)[
0
]);
auto
*
layer
=
TRT_ENGINE_ADD_LAYER
(
engine_
,
SoftMax
,
*
const_cast
<
nvinfer1
::
ITensor
*>
(
input1
));
auto
output_name
=
op_desc
.
Output
(
"Out"
)[
0
];
engine_
->
SetITensor
(
output_name
,
layer
->
getOutput
(
0
));
if
(
test_mode
)
{
engine_
->
DeclareOutput
(
output_name
);
}
}
};
}
// namespace tensorrt
}
// namespace inference
}
// namespace paddle
USE_OP
(
softmax
);
REGISTER_TRT_OP_CONVERTER
(
softmax
,
SoftMaxOpConverter
);
paddle/fluid/inference/tensorrt/convert/test_softmax_op.cc
0 → 100644
浏览文件 @
83c85f34
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
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 <gtest/gtest.h>
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/inference/tensorrt/convert/ut_helper.h"
namespace
paddle
{
namespace
inference
{
namespace
tensorrt
{
TEST
(
SoftMaxOpConverter
,
main
)
{
framework
::
Scope
scope
;
std
::
unordered_set
<
std
::
string
>
parameters
;
TRTConvertValidation
validator
(
8
,
parameters
,
scope
,
1000
);
std
::
vector
<
int
>
tensor_shape
{
8
,
10
};
validator
.
DeclInputVar
(
"softmax-X"
,
tensor_shape
,
nvinfer1
::
DimsCHW
(
10
,
1
,
1
));
validator
.
DeclOutputVar
(
"softmax-Out"
,
nvinfer1
::
DimsCHW
(
10
,
1
,
1
));
// Prepare Op description
framework
::
OpDesc
desc
;
desc
.
SetType
(
"softmax"
);
desc
.
SetInput
(
"X"
,
{
"softmax-X"
});
desc
.
SetOutput
(
"Out"
,
{
"softmax-Out"
});
LOG
(
INFO
)
<<
"set OP"
;
validator
.
SetOp
(
*
desc
.
Proto
());
LOG
(
INFO
)
<<
"execute"
;
validator
.
Execute
(
3
);
}
}
// namespace tensorrt
}
// namespace inference
}
// namespace paddle
USE_OP
(
softmax
);
paddle/fluid/inference/tensorrt/convert/ut_helper.h
浏览文件 @
83c85f34
...
...
@@ -79,6 +79,12 @@ class TRTConvertValidation {
}
// Declare a Variable as input with random initialization.
void
DeclInputVar
(
const
std
::
string
&
name
,
const
std
::
vector
<
int
>
tensor_dims
,
const
nvinfer1
::
Dims
&
trt_dims
)
{
DeclVar
(
name
,
tensor_dims
);
engine_
->
DeclareInput
(
name
,
nvinfer1
::
DataType
::
kFLOAT
,
trt_dims
);
}
void
DeclInputVar
(
const
std
::
string
&
name
,
const
nvinfer1
::
Dims
&
dims
)
{
DeclVar
(
name
,
dims
);
// Declare TRT inputs.
...
...
@@ -94,12 +100,18 @@ class TRTConvertValidation {
DeclVar
(
name
,
dims
);
}
// Declare a variable in a fluid Scope.
void
DeclVar
(
const
std
::
string
&
name
,
const
nvinfer1
::
Dims
&
dims
,
bool
is_param
=
false
)
{
void
DeclVar
(
const
std
::
string
&
name
,
const
std
::
vector
<
int
>
dim_vec
)
{
platform
::
CPUPlace
place
;
platform
::
CPUDeviceContext
ctx
(
place
);
auto
*
x
=
scope_
.
Var
(
name
);
auto
*
x_tensor
=
x
->
GetMutable
<
framework
::
LoDTensor
>
();
x_tensor
->
Resize
(
framework
::
make_ddim
(
dim_vec
));
RandomizeTensor
(
x_tensor
,
place
,
ctx
);
}
// Declare a variable in a fluid Scope.
void
DeclVar
(
const
std
::
string
&
name
,
const
nvinfer1
::
Dims
&
dims
,
bool
is_param
=
false
)
{
// Init Fluid tensor.
std
::
vector
<
int
>
dim_vec
(
dims
.
d
,
dims
.
d
+
dims
.
nbDims
);
// There is no batchsize in ITensor's shape, but We should add it to
...
...
@@ -107,10 +119,8 @@ class TRTConvertValidation {
// if_add_batch_ flag is true, add the max batchsize to dim_vec.
if
(
is_param
!=
true
&&
if_add_batch_
==
true
)
dim_vec
.
insert
(
dim_vec
.
begin
(),
max_batch_size_
);
auto
*
x
=
scope_
.
Var
(
name
);
auto
*
x_tensor
=
x
->
GetMutable
<
framework
::
LoDTensor
>
();
x_tensor
->
Resize
(
framework
::
make_ddim
(
dim_vec
));
RandomizeTensor
(
x_tensor
,
place
,
ctx
);
DeclVar
(
name
,
dim_vec
);
}
void
SetOp
(
const
framework
::
proto
::
OpDesc
&
desc
)
{
...
...
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