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
OPTHREE
Paddle
提交
fdbdef0e
P
Paddle
项目概览
OPTHREE
/
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
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
fdbdef0e
编写于
6月 02, 2021
作者:
P
Pei Yang
提交者:
GitHub
6月 02, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix conv2d_transpose trt bugs (#33242)
上级
29dc439a
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
36 addition
and
7 deletion
+36
-7
paddle/fluid/inference/tensorrt/convert/conv2d_op.cc
paddle/fluid/inference/tensorrt/convert/conv2d_op.cc
+12
-7
python/paddle/fluid/tests/unittests/ir/inference/test_trt_conv_pass.py
.../fluid/tests/unittests/ir/inference/test_trt_conv_pass.py
+24
-0
未找到文件。
paddle/fluid/inference/tensorrt/convert/conv2d_op.cc
浏览文件 @
fdbdef0e
...
...
@@ -103,11 +103,18 @@ void ConvertConv2d(TensorRTEngine* engine, const framework::proto::OpDesc& op,
TensorRTEngine
::
Weight
bias
{
nvinfer1
::
DataType
::
kFLOAT
,
static_cast
<
void
*>
(
bias_data
),
bias_size
};
auto
*
layer
=
fadd_layer
(
const_cast
<
nvinfer1
::
ITensor
*>
(
X
),
n_output
,
n_input
,
nv_ksize
,
weight
,
bias
);
PADDLE_ENFORCE_NOT_NULL
(
layer
,
platform
::
errors
::
Fatal
(
"TensorRT create conv2d"
" layer error."
));
// In conv2d_transpose and depthwise_conv2d_transpose,
// output channels = filter_dims[1] * groups
auto
*
layer
=
(
op_desc
.
Type
()
==
"conv2d_transpose"
||
op_desc
.
Type
()
==
"depthwise_conv2d_transpose"
)
?
fadd_layer
(
const_cast
<
nvinfer1
::
ITensor
*>
(
X
),
n_input
*
groups
,
nv_ksize
,
weight
,
bias
)
:
fadd_layer
(
const_cast
<
nvinfer1
::
ITensor
*>
(
X
),
n_output
,
nv_ksize
,
weight
,
bias
);
PADDLE_ENFORCE_NOT_NULL
(
layer
,
platform
::
errors
::
Fatal
(
"TensorRT create conv2d/conv2d_transpose"
" layer failed."
));
layer
->
setStride
(
nv_strides
);
layer
->
setPadding
(
nv_paddings
);
layer
->
setNbGroups
(
groups
);
...
...
@@ -134,7 +141,6 @@ class Conv2dOpConverter : public OpConverter {
ConvertConv2d
(
engine_
,
op
,
scope
,
test_mode
,
[
&
](
nvinfer1
::
ITensor
*
inputs
,
int
n_output
,
/* Conv output maps */
int
n_input
,
/* Conv input maps */
nvinfer1
::
DimsHW
&
ksize
,
TensorRTEngine
::
Weight
&
weight
,
TensorRTEngine
::
Weight
&
bias
)
->
nvinfer1
::
IConvolutionLayer
*
{
auto
*
layer
=
...
...
@@ -156,7 +162,6 @@ class Deconv2dOpConverter : public OpConverter {
ConvertConv2d
(
engine_
,
op
,
scope
,
test_mode
,
[
&
](
nvinfer1
::
ITensor
*
inputs
,
int
n_output
,
/* Deconv input maps */
int
n_input
,
/* Deconv output maps */
nvinfer1
::
DimsHW
&
ksize
,
TensorRTEngine
::
Weight
&
weight
,
TensorRTEngine
::
Weight
&
bias
)
->
nvinfer1
::
IDeconvolutionLayer
*
{
auto
*
layer
=
...
...
python/paddle/fluid/tests/unittests/ir/inference/test_trt_conv_pass.py
浏览文件 @
fdbdef0e
...
...
@@ -36,6 +36,7 @@ class TensorRTSubgraphPassConvTest(InferencePassTest):
groups
=
self
.
conv_groups
,
padding
=
self
.
conv_padding
,
bias_attr
=
False
,
use_cudnn
=
self
.
use_cudnn
,
act
=
None
)
self
.
feeds
=
{
"data"
:
np
.
random
.
random
([
1
,
6
,
64
,
64
]).
astype
(
"float32"
),
...
...
@@ -50,6 +51,7 @@ class TensorRTSubgraphPassConvTest(InferencePassTest):
self
.
conv_filter_size
=
6
self
.
conv_groups
=
3
self
.
conv_padding
=
[
1
,
1
]
self
.
use_cudnn
=
True
def
test_check_output
(
self
):
if
core
.
is_compiled_with_cuda
():
...
...
@@ -65,6 +67,7 @@ class TensorRTSubgraphPassConvValidPaddingTest(TensorRTSubgraphPassConvTest):
self
.
conv_filter_size
=
6
self
.
conv_groups
=
3
self
.
conv_padding
=
'VALID'
self
.
use_cudnn
=
True
class
TensorRTSubgraphPassConvSamePaddingTest
(
InferencePassTest
):
...
...
@@ -73,6 +76,7 @@ class TensorRTSubgraphPassConvSamePaddingTest(InferencePassTest):
self
.
conv_filter_size
=
6
self
.
conv_groups
=
3
self
.
conv_padding
=
'SAME'
self
.
use_cudnn
=
True
class
TensorRTSubgraphPassDepthwiseConvTest
(
TensorRTSubgraphPassConvTest
):
...
...
@@ -81,6 +85,16 @@ class TensorRTSubgraphPassDepthwiseConvTest(TensorRTSubgraphPassConvTest):
self
.
conv_filter_size
=
6
self
.
conv_groups
=
6
self
.
conv_padding
=
[
1
,
1
]
self
.
use_cudnn
=
False
class
TensorRTSubgraphPassDepthwiseConv2Test
(
TensorRTSubgraphPassConvTest
):
def
set_params
(
self
):
self
.
conv_num_filters
=
12
self
.
conv_filter_size
=
6
self
.
conv_groups
=
6
self
.
conv_padding
=
[
1
,
1
]
self
.
use_cudnn
=
False
class
TensorRTSubgraphPassConvTransposeTest
(
InferencePassTest
):
...
...
@@ -151,6 +165,16 @@ class TensorRTSubgraphPassConvTransposeMultiGroupTest(
self
.
use_cudnn
=
True
class
TensorRTSubgraphPassConvTranspose2Test
(
TensorRTSubgraphPassConvTransposeTest
):
def
set_params
(
self
):
self
.
conv_num_filters
=
12
self
.
conv_filter_size
=
4
self
.
conv_groups
=
6
self
.
conv_padding
=
[
1
,
1
]
self
.
use_cudnn
=
False
class
TensorRTSubgraphPassDepthwiseConvTransposeTest
(
TensorRTSubgraphPassConvTransposeTest
):
def
set_params
(
self
):
...
...
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