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
TonyTonyFun
Paddle
提交
6e90ba1b
P
Paddle
项目概览
TonyTonyFun
/
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
6e90ba1b
编写于
5月 12, 2022
作者:
Z
zhupengyang
提交者:
GitHub
5月 12, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add exp,log trt converter (#42655)
上级
5914b18a
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
256 addition
and
6 deletion
+256
-6
paddle/fluid/inference/api/analysis_predictor.cc
paddle/fluid/inference/api/analysis_predictor.cc
+2
-0
paddle/fluid/inference/tensorrt/convert/CMakeLists.txt
paddle/fluid/inference/tensorrt/convert/CMakeLists.txt
+33
-5
paddle/fluid/inference/tensorrt/convert/unary_op.cc
paddle/fluid/inference/tensorrt/convert/unary_op.cc
+84
-0
paddle/fluid/inference/tensorrt/op_teller.cc
paddle/fluid/inference/tensorrt/op_teller.cc
+5
-1
python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_unary.py
...id/tests/unittests/ir/inference/test_trt_convert_unary.py
+132
-0
未找到文件。
paddle/fluid/inference/api/analysis_predictor.cc
浏览文件 @
6e90ba1b
...
...
@@ -1744,6 +1744,8 @@ USE_TRT_CONVERTER(flatten_contiguous_range);
USE_TRT_CONVERTER
(
matmul
);
USE_TRT_CONVERTER
(
conv2d
);
USE_TRT_CONVERTER
(
relu
);
USE_TRT_CONVERTER
(
exp
);
USE_TRT_CONVERTER
(
log
);
USE_TRT_CONVERTER
(
sigmoid
);
USE_TRT_CONVERTER
(
tanh
);
USE_TRT_CONVERTER
(
fc
);
...
...
paddle/fluid/inference/tensorrt/convert/CMakeLists.txt
浏览文件 @
6e90ba1b
# Add TRT tests
nv_library
(
tensorrt_converter
SRCS matmul_op.cc conv2d_op.cc fc_op.cc pool2d_op.cc elementwise_op.cc
batch_norm_op.cc activation_op.cc softmax_op.cc concat_op.cc dropout_op.cc group_norm_op.cc
pad_op.cc split_op.cc prelu_op.cc leaky_relu_op.cc gelu_op.cc layer_norm_op.cc multihead_matmul_op.cc
shuffle_channel_op.cc swish_op.cc instance_norm_op.cc stack_op.cc transpose_op.cc flatten_op.cc flatten_contiguous_range_op.cc
emb_eltwise_layernorm.cc skip_layernorm.cc scale_op.cc slice_op.cc hard_sigmoid_op.cc hard_swish_op.cc clip_op.cc
SRCS matmul_op.cc
conv2d_op.cc
fc_op.cc
pool2d_op.cc
elementwise_op.cc
batch_norm_op.cc
activation_op.cc
unary_op.cc
softmax_op.cc
concat_op.cc
dropout_op.cc
group_norm_op.cc
pad_op.cc
split_op.cc
prelu_op.cc
leaky_relu_op.cc
gelu_op.cc
layer_norm_op.cc
multihead_matmul_op.cc
shuffle_channel_op.cc
swish_op.cc
instance_norm_op.cc
stack_op.cc
transpose_op.cc
flatten_op.cc
flatten_contiguous_range_op.cc
emb_eltwise_layernorm.cc
skip_layernorm.cc
scale_op.cc
slice_op.cc
hard_sigmoid_op.cc
hard_swish_op.cc
clip_op.cc
gather_op.cc
anchor_generator_op.cc
yolo_box_op.cc
...
...
paddle/fluid/inference/tensorrt/convert/unary_op.cc
0 → 100644
浏览文件 @
6e90ba1b
/* Copyright (c) 2022 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 <NvInfer.h>
#include <string>
#include "glog/logging.h"
#include "paddle/fluid/framework/op_desc.h"
#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
#include "paddle/fluid/inference/tensorrt/engine.h"
#include "paddle/fluid/inference/tensorrt/helper.h"
#include "paddle/fluid/platform/enforce.h"
namespace
paddle
{
namespace
framework
{
class
Scope
;
namespace
proto
{
class
OpDesc
;
}
// namespace proto
}
// namespace framework
}
// namespace paddle
namespace
paddle
{
namespace
inference
{
namespace
tensorrt
{
class
UnaryOpConverter
:
public
OpConverter
{
public:
UnaryOpConverter
()
{}
void
operator
()(
const
framework
::
proto
::
OpDesc
&
op
,
const
framework
::
Scope
&
scope
,
bool
test_mode
)
override
{
// Here the two nullptr looks strange, that's because the
// framework::OpDesc's constructor is strange.
framework
::
OpDesc
op_desc
(
op
,
nullptr
);
VLOG
(
3
)
<<
"convert a fluid unary op to tensorrt unary layer whose "
"type is "
<<
op_type_
;
nvinfer1
::
ITensor
*
input_tensor
=
engine_
->
GetITensor
(
op_desc
.
Input
(
"X"
)[
0
]);
auto
op_pair
=
ops
.
find
(
op_type_
);
nvinfer1
::
IUnaryLayer
*
layer
=
TRT_ENGINE_ADD_LAYER
(
engine_
,
Unary
,
*
input_tensor
,
op_pair
->
second
);
auto
output_name
=
op_desc
.
Output
(
"Out"
)[
0
];
RreplenishLayerAndOutput
(
layer
,
op_type_
,
{
output_name
},
test_mode
);
}
protected:
std
::
string
op_type_
;
static
const
std
::
unordered_map
<
std
::
string
,
nvinfer1
::
UnaryOperation
>
ops
;
};
const
std
::
unordered_map
<
std
::
string
,
nvinfer1
::
UnaryOperation
>
UnaryOpConverter
::
ops
=
{
{
"exp"
,
nvinfer1
::
UnaryOperation
::
kEXP
},
{
"log"
,
nvinfer1
::
UnaryOperation
::
kLOG
},
};
class
ExpOpConverter
:
public
UnaryOpConverter
{
public:
ExpOpConverter
()
{
op_type_
=
"exp"
;
}
};
class
LogOpConverter
:
public
UnaryOpConverter
{
public:
LogOpConverter
()
{
op_type_
=
"log"
;
}
};
}
// namespace tensorrt
}
// namespace inference
}
// namespace paddle
REGISTER_TRT_OP_CONVERTER
(
exp
,
ExpOpConverter
);
REGISTER_TRT_OP_CONVERTER
(
log
,
LogOpConverter
);
paddle/fluid/inference/tensorrt/op_teller.cc
浏览文件 @
6e90ba1b
...
...
@@ -65,6 +65,8 @@ struct SimpleOpTypeSetTeller : public Teller {
"conv2d_fusion"
,
"pool2d"
,
"relu"
,
"exp"
,
"log"
,
"softmax"
,
"sigmoid"
,
"hard_swish"
,
...
...
@@ -128,6 +130,8 @@ struct SimpleOpTypeSetTeller : public Teller {
"conv2d_fusion"
,
"pool2d"
,
"relu"
,
"exp"
,
"log"
,
"softmax"
,
"sigmoid"
,
"hard_swish"
,
...
...
@@ -200,7 +204,7 @@ bool OpTeller::Tell(const framework::ir::Node* node, bool use_no_calib_int8,
for
(
auto
&
teller
:
tellers_
)
{
if
(
op_type
==
"relu"
||
op_type
==
"relu6"
||
op_type
==
"tanh"
||
op_type
==
"sigmoid"
)
{
op_type
==
"sigmoid"
||
op_type
==
"exp"
||
op_type
==
"log"
)
{
auto
*
block
=
desc
.
Block
();
if
(
block
==
nullptr
)
{
VLOG
(
3
)
<<
"The block desc is nullptr, we can't continue to analyze. "
...
...
python/paddle/fluid/tests/unittests/ir/inference/test_trt_convert_unary.py
0 → 100644
浏览文件 @
6e90ba1b
# Copyright (c) 2022 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.
from
trt_layer_auto_scan_test
import
TrtLayerAutoScanTest
,
SkipReasons
from
program_config
import
TensorConfig
,
ProgramConfig
import
unittest
import
numpy
as
np
import
paddle.inference
as
paddle_infer
from
functools
import
partial
from
typing
import
Optional
,
List
,
Callable
,
Dict
,
Any
,
Set
class
TrtConvertActivationTest
(
TrtLayerAutoScanTest
):
def
is_program_valid
(
self
,
program_config
:
ProgramConfig
)
->
bool
:
return
True
def
sample_program_configs
(
self
):
def
generate_input1
(
dims
,
batch
,
attrs
:
List
[
Dict
[
str
,
Any
]]):
if
dims
==
1
:
return
np
.
ones
([
32
]).
astype
(
np
.
float32
)
elif
dims
==
2
:
return
np
.
ones
([
3
,
32
]).
astype
(
np
.
float32
)
elif
dims
==
3
:
return
np
.
ones
([
3
,
32
,
32
]).
astype
(
np
.
float32
)
else
:
return
np
.
ones
([
batch
,
3
,
32
,
32
]).
astype
(
np
.
float32
)
for
dims
in
[
1
,
2
,
3
,
4
]:
for
batch
in
[
1
,
4
]:
for
op_type
in
[
"exp"
,
"log"
]:
self
.
dims
=
dims
dics
=
[{}]
ops_config
=
[{
"op_type"
:
op_type
,
"op_inputs"
:
{
"X"
:
[
"input_data"
]
},
"op_outputs"
:
{
"Out"
:
[
"output_data"
]
},
"op_attrs"
:
dics
[
0
]
}]
ops
=
self
.
generate_op_config
(
ops_config
)
program_config
=
ProgramConfig
(
ops
=
ops
,
weights
=
{},
inputs
=
{
"input_data"
:
TensorConfig
(
data_gen
=
partial
(
generate_input1
,
dims
,
batch
,
dics
))
},
outputs
=
[
"output_data"
])
yield
program_config
def
sample_predictor_configs
(
self
,
program_config
)
->
(
paddle_infer
.
Config
,
List
[
int
],
float
):
def
generate_dynamic_shape
(
attrs
):
if
self
.
dims
==
1
:
self
.
dynamic_shape
.
min_input_shape
=
{
"input_data"
:
[
1
]}
self
.
dynamic_shape
.
max_input_shape
=
{
"input_data"
:
[
64
]}
self
.
dynamic_shape
.
opt_input_shape
=
{
"input_data"
:
[
32
]}
elif
self
.
dims
==
2
:
self
.
dynamic_shape
.
min_input_shape
=
{
"input_data"
:
[
1
,
16
]}
self
.
dynamic_shape
.
max_input_shape
=
{
"input_data"
:
[
4
,
32
]}
self
.
dynamic_shape
.
opt_input_shape
=
{
"input_data"
:
[
3
,
32
]}
elif
self
.
dims
==
3
:
self
.
dynamic_shape
.
min_input_shape
=
{
"input_data"
:
[
1
,
16
,
16
]}
self
.
dynamic_shape
.
max_input_shape
=
{
"input_data"
:
[
4
,
32
,
32
]}
self
.
dynamic_shape
.
opt_input_shape
=
{
"input_data"
:
[
3
,
32
,
32
]}
else
:
self
.
dynamic_shape
.
min_input_shape
=
{
"input_data"
:
[
1
,
3
,
16
,
16
]
}
self
.
dynamic_shape
.
max_input_shape
=
{
"input_data"
:
[
4
,
3
,
32
,
32
]
}
self
.
dynamic_shape
.
opt_input_shape
=
{
"input_data"
:
[
1
,
3
,
32
,
32
]
}
def
clear_dynamic_shape
():
self
.
dynamic_shape
.
min_input_shape
=
{}
self
.
dynamic_shape
.
max_input_shape
=
{}
self
.
dynamic_shape
.
opt_input_shape
=
{}
def
generate_trt_nodes_num
(
attrs
,
dynamic_shape
):
if
self
.
dims
==
1
:
return
0
,
3
return
1
,
2
attrs
=
[
program_config
.
ops
[
i
].
attrs
for
i
in
range
(
len
(
program_config
.
ops
))
]
# for static_shape
clear_dynamic_shape
()
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Float32
yield
self
.
create_inference_config
(),
generate_trt_nodes_num
(
attrs
,
False
),
1e-5
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Half
yield
self
.
create_inference_config
(),
generate_trt_nodes_num
(
attrs
,
False
),
1e-5
# for dynamic_shape
generate_dynamic_shape
(
attrs
)
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Float32
yield
self
.
create_inference_config
(),
generate_trt_nodes_num
(
attrs
,
True
),
1e-5
self
.
trt_param
.
precision
=
paddle_infer
.
PrecisionType
.
Half
yield
self
.
create_inference_config
(),
generate_trt_nodes_num
(
attrs
,
True
),
1e-5
def
test
(
self
):
self
.
run_test
()
if
__name__
==
"__main__"
:
unittest
.
main
()
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