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
50cafa0b
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 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
50cafa0b
编写于
3月 19, 2021
作者:
Z
zlsh80826
提交者:
GitHub
3月 19, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
remove redundant sync, set collect/dist kernel to context stream, sub_lod memcpy opt (#31641)
上级
1d197f6c
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
10 addition
and
11 deletion
+10
-11
paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
+2
-2
paddle/fluid/operators/detection/distribute_fpn_proposals_op.cu
.../fluid/operators/detection/distribute_fpn_proposals_op.cu
+8
-8
paddle/fluid/operators/detection/generate_proposals_op.cu
paddle/fluid/operators/detection/generate_proposals_op.cu
+0
-1
未找到文件。
paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
浏览文件 @
50cafa0b
...
...
@@ -198,8 +198,8 @@ class GPUCollectFpnProposalsOpKernel : public framework::OpKernel<T> {
int
threads
=
kNumCUDAThreads
;
// get length-based lod by batch ids
GetLengthLoD
<<<
blocks
,
threads
>>>
(
real_post_num
,
out_id_data
,
length_lod_data
);
GetLengthLoD
<<<
blocks
,
threads
,
0
,
dev_ctx
.
stream
()
>>>
(
real_post_num
,
out_id_data
,
length_lod_data
);
std
::
vector
<
int
>
length_lod_cpu
(
lod_size
);
memory
::
Copy
(
platform
::
CPUPlace
(),
length_lod_cpu
.
data
(),
place
,
length_lod_data
,
sizeof
(
int
)
*
lod_size
,
dev_ctx
.
stream
());
...
...
paddle/fluid/operators/detection/distribute_fpn_proposals_op.cu
浏览文件 @
50cafa0b
...
...
@@ -131,11 +131,10 @@ class GPUDistributeFpnProposalsOpKernel : public framework::OpKernel<T> {
int
dist_blocks
=
NumBlocks
(
roi_num
);
int
threads
=
kNumCUDAThreads
;
// get target levels and sub_lod list
GPUDistFpnProposalsHelper
<
T
><<<
dist_blocks
,
threads
>>>
(
GPUDistFpnProposalsHelper
<
T
><<<
dist_blocks
,
threads
,
0
,
dev_ctx
.
stream
()
>>>
(
roi_num
,
fpn_rois
->
data
<
T
>
(),
lod_size
,
refer_level
,
refer_scale
,
max_level
,
min_level
,
roi_batch_id_list_gpu
.
data
<
int
>
(),
sub_lod_list_data
,
target_lvls_data
,
pixel_offset
);
dev_ctx
.
Wait
();
auto
place
=
BOOST_GET_CONST
(
platform
::
CUDAPlace
,
dev_ctx
.
GetPlace
());
Tensor
index_in_t
;
...
...
@@ -172,17 +171,18 @@ class GPUDistributeFpnProposalsOpKernel : public framework::OpKernel<T> {
int
start
=
0
;
auto
multi_rois_num
=
ctx
.
MultiOutput
<
Tensor
>
(
"MultiLevelRoIsNum"
);
std
::
vector
<
int
>
sub_lod_list_cpu
(
lod_size
*
num_level
);
memory
::
Copy
(
platform
::
CPUPlace
(),
sub_lod_list_cpu
.
data
(),
place
,
sub_lod_list_data
,
sizeof
(
int
)
*
lod_size
*
num_level
,
dev_ctx
.
stream
());
dev_ctx
.
Wait
();
for
(
int
i
=
0
;
i
<
num_level
;
++
i
)
{
Tensor
sub_lod
=
sub_lod_list
.
Slice
(
i
,
i
+
1
);
int
*
sub_lod_data
=
sub_lod
.
data
<
int
>
();
// transfer length-based lod to offset-based lod
std
::
vector
<
size_t
>
offset
(
1
,
0
);
std
::
vector
<
int
>
sub_lod_cpu
(
lod_size
);
memory
::
Copy
(
platform
::
CPUPlace
(),
sub_lod_cpu
.
data
(),
place
,
sub_lod_data
,
sizeof
(
int
)
*
lod_size
,
dev_ctx
.
stream
());
dev_ctx
.
Wait
();
for
(
int
j
=
0
;
j
<
lod_size
;
++
j
)
{
offset
.
emplace_back
(
offset
.
back
()
+
sub_lod_
cpu
[
j
]);
offset
.
emplace_back
(
offset
.
back
()
+
sub_lod_
list_cpu
[
i
*
lod_size
+
j
]);
}
int
sub_rois_num
=
offset
.
back
();
...
...
paddle/fluid/operators/detection/generate_proposals_op.cu
浏览文件 @
50cafa0b
...
...
@@ -198,7 +198,6 @@ class CUDAGenerateProposalsKernel : public framework::OpKernel<T> {
memory
::
Copy
(
place
,
rpn_roi_probs_data
+
num_proposals
,
place
,
scores
.
data
<
T
>
(),
sizeof
(
T
)
*
scores
.
numel
(),
dev_ctx
.
stream
());
dev_ctx
.
Wait
();
num_proposals
+=
proposals
.
dims
()[
0
];
offset
.
emplace_back
(
num_proposals
);
tmp_num
.
push_back
(
proposals
.
dims
()[
0
]);
...
...
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