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
PaddlePaddle
Paddle
提交
2b1386b2
P
Paddle
项目概览
PaddlePaddle
/
Paddle
12 个月 前同步成功
通知
2292
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
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
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
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
2b1386b2
编写于
2月 07, 2020
作者:
L
LielinJiang
提交者:
GitHub
2月 07, 2020
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
optimize performance of interpolate op (#22436)
* optimize interpolate op, test=develop
上级
77dd0d97
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
74 addition
and
20 deletion
+74
-20
paddle/fluid/operators/interpolate_op.cu
paddle/fluid/operators/interpolate_op.cu
+26
-20
paddle/fluid/platform/gpu_launch_config.h
paddle/fluid/platform/gpu_launch_config.h
+48
-0
未找到文件。
paddle/fluid/operators/interpolate_op.cu
浏览文件 @
2b1386b2
...
...
@@ -9,9 +9,11 @@
See the License for the specific language governing permissions and
limitations under the License. */
#include <algorithm>
#include <string>
#include "paddle/fluid/operators/interpolate_op.h"
#include "paddle/fluid/platform/cuda_primitives.h"
#include "paddle/fluid/platform/gpu_launch_config.h"
namespace
paddle
{
namespace
operators
{
...
...
@@ -586,17 +588,18 @@ static void Interpolate2DCUDAFwd(const framework::ExecutionContext& ctx,
int
out_chw
=
c
*
out_hw
;
int
pixelNum
=
n
*
out_chw
;
int
grid_dim
=
(
pixelNum
+
512
-
1
)
/
512
;
grid_dim
=
grid_dim
>
8
?
8
:
grid_dim
;
platform
::
GpuLaunchConfig
config
=
platform
::
getGpuLaunchConfig
(
pixelNum
,
ctx
);
if
(
"nearest"
==
interp_method
)
{
KeNearestNeighborInterpFw
<
T
><<<
grid_dim
,
512
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
KeNearestNeighborInterpFw
<
T
><<<
config
.
blocks
,
config
.
threads
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
input_data
,
in_h
,
in_w
,
n
,
in_chw
,
output_data
,
out_h
,
out_w
,
n
,
out_chw
,
c
,
ratio_h
,
ratio_w
,
align_corners
,
data_layout
);
}
else
if
(
"bilinear"
==
interp_method
)
{
KeBilinearInterpFw
<
T
><<<
grid_dim
,
512
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
KeBilinearInterpFw
<
T
><<<
config
.
blocks
,
config
.
threads
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
input_data
,
in_h
,
in_w
,
n
,
in_chw
,
output_data
,
out_h
,
out_w
,
n
,
out_chw
,
c
,
ratio_h
,
ratio_w
,
align_corners
,
align_mode
,
data_layout
);
}
...
...
@@ -696,12 +699,13 @@ static void Interpolate3DCUDAFwd(const framework::ExecutionContext& ctx,
int
out_cdhw
=
c
*
out_dhw
;
int
pixelNum
=
n
*
out_cdhw
;
int
grid_dim
=
(
pixelNum
+
512
-
1
)
/
512
;
grid_dim
=
grid_dim
>
8
?
8
:
grid_dim
;
platform
::
GpuLaunchConfig
config
=
platform
::
getGpuLaunchConfig
(
pixelNum
,
ctx
);
if
(
"trilinear"
==
interp_method
)
{
KeTrilinearInterpFw
<
T
><<<
grid_dim
,
512
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
KeTrilinearInterpFw
<
T
><<<
config
.
blocks
,
config
.
threads
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
input_data
,
in_d
,
in_h
,
in_w
,
n
,
in_cdhw
,
output_data
,
out_d
,
out_h
,
out_w
,
n
,
out_cdhw
,
c
,
ratio_d
,
ratio_h
,
ratio_w
,
align_corners
,
align_mode
,
data_layout
);
...
...
@@ -787,17 +791,18 @@ static void Interpolate2DCUDABwd(const framework::ExecutionContext& ctx,
int
out_chw
=
c
*
out_hw
;
int
pixelNum
=
n
*
out_chw
;
int
grid_dim
=
(
pixelNum
+
512
-
1
)
/
512
;
grid_dim
=
grid_dim
>
8
?
8
:
grid_dim
;
platform
::
GpuLaunchConfig
config
=
platform
::
getGpuLaunchConfig
(
pixelNum
,
ctx
);
if
(
"nearest"
==
interp_method
)
{
KeNearestNeighborInterpBw
<
T
><<<
grid_dim
,
512
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
KeNearestNeighborInterpBw
<
T
><<<
config
.
blocks
,
config
.
threads
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
input_grad_data
,
in_h
,
in_w
,
n
,
in_chw
,
output_grad_data
,
out_h
,
out_w
,
n
,
out_chw
,
c
,
ratio_h
,
ratio_w
,
align_corners
,
data_layout
);
}
else
if
(
"bilinear"
==
interp_method
)
{
KeBilinearInterpBw
<
T
><<<
grid_dim
,
512
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
KeBilinearInterpBw
<
T
><<<
config
.
blocks
,
config
.
threads
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
input_grad_data
,
in_h
,
in_w
,
n
,
in_chw
,
output_grad_data
,
out_h
,
out_w
,
n
,
out_chw
,
c
,
ratio_h
,
ratio_w
,
align_corners
,
align_mode
,
data_layout
);
...
...
@@ -892,12 +897,13 @@ static void Interpolate3DCUDABwd(const framework::ExecutionContext& ctx,
int
out_cdhw
=
c
*
out_dhw
;
int
pixelNum
=
n
*
out_cdhw
;
int
grid_dim
=
(
pixelNum
+
512
-
1
)
/
512
;
grid_dim
=
grid_dim
>
8
?
8
:
grid_dim
;
platform
::
GpuLaunchConfig
config
=
platform
::
getGpuLaunchConfig
(
pixelNum
,
ctx
);
if
(
"trilinear"
==
interp_method
)
{
KeTrilinearInterpBw
<
T
><<<
grid_dim
,
512
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
KeTrilinearInterpBw
<
T
><<<
config
.
blocks
,
config
.
threads
,
0
,
ctx
.
cuda_device_context
().
stream
()
>>>
(
input_grad_data
,
in_d
,
in_h
,
in_w
,
n
,
in_cdhw
,
output_grad_data
,
out_d
,
out_h
,
out_w
,
n
,
out_cdhw
,
c
,
ratio_d
,
ratio_h
,
ratio_w
,
align_corners
,
align_mode
,
data_layout
);
...
...
paddle/fluid/platform/gpu_launch_config.h
0 → 100644
浏览文件 @
2b1386b2
/* Copyright (c) 2020 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. */
#pragma once
#include <algorithm>
#include "paddle/fluid/platform/cuda_primitives.h"
namespace
paddle
{
namespace
platform
{
struct
GpuLaunchConfig
{
// Number of threads per block.
int
threads
;
// Number of blocks for GPU kernel launch.
int
blocks
;
GpuLaunchConfig
(
int
threads
,
int
blocks
)
:
threads
(
threads
),
blocks
(
blocks
)
{}
};
inline
GpuLaunchConfig
getGpuLaunchConfig
(
const
int
N
,
const
framework
::
ExecutionContext
&
ctx
)
{
int
threads
=
std
::
min
(
1024
,
ctx
.
cuda_device_context
().
GetMaxThreadsPerBlock
());
int
physical_thread_count
=
std
::
min
(
ctx
.
cuda_device_context
().
GetMaxPhysicalThreadCount
(),
N
);
int
blocks
=
std
::
min
((
physical_thread_count
+
threads
-
1
)
/
threads
,
ctx
.
cuda_device_context
().
GetSMCount
());
GpuLaunchConfig
config
(
threads
,
blocks
);
return
config
;
}
}
// namespace platform
}
// namespace paddle
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