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
Crayon鑫
Paddle
提交
5cf3f898
P
Paddle
项目概览
Crayon鑫
/
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
未验证
提交
5cf3f898
编写于
6月 16, 2022
作者:
Y
Yuang Liu
提交者:
GitHub
6月 16, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[cuda graph] bug fix for cuda graph static mode (#43539)
上级
890c7315
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
4 addition
and
3 deletion
+4
-3
python/paddle/device/cuda/graphs.py
python/paddle/device/cuda/graphs.py
+4
-3
未找到文件。
python/paddle/device/cuda/graphs.py
浏览文件 @
5cf3f898
...
@@ -173,7 +173,7 @@ def construct_program_and_find_ins_outs(section, origin_program, section_idx):
...
@@ -173,7 +173,7 @@ def construct_program_and_find_ins_outs(section, origin_program, section_idx):
# This in var is generated from op outside this section
# This in var is generated from op outside this section
# Only record once for same input
# Only record once for same input
ins
.
append
(
in_name
)
ins
.
append
(
in_name
)
elif
later_ins
.
count
(
in_name
)
==
0
:
elif
later_ins
.
count
(
in_name
)
==
0
and
outs
.
count
(
in_name
)
>
0
:
# this is var is generated from op inside this section, and only will be used inside this section
# this is var is generated from op inside this section, and only will be used inside this section
outs
.
remove
(
in_name
)
outs
.
remove
(
in_name
)
for
out_name
in
op
.
output_arg_names
:
for
out_name
in
op
.
output_arg_names
:
...
@@ -248,13 +248,13 @@ def get_cuda_graph_sections(program):
...
@@ -248,13 +248,13 @@ def get_cuda_graph_sections(program):
sub_block_related
=
(
op
.
type
==
'conditional_block'
sub_block_related
=
(
op
.
type
==
'conditional_block'
or
op
.
type
==
'while'
)
or
op
.
type
==
'while'
)
if
loss_related
or
sub_block_related
:
if
loss_related
or
sub_block_related
:
#
i
f loss_related is True
#
I
f loss_related is True
# The internal section contains loss related ops,
# The internal section contains loss related ops,
# although these ops are between two cuda graph sections with same graph id,
# although these ops are between two cuda graph sections with same graph id,
# they belong to none of these two sections.
# they belong to none of these two sections.
# The loss related op should be wrapped by user explicitly.
# The loss related op should be wrapped by user explicitly.
#
i
f sub_block_related is True
#
I
f sub_block_related is True
# The internal section contains while op or conditional block op.
# The internal section contains while op or conditional block op.
# These two ops are not supported by cuda graph. Won't extend the section.
# These two ops are not supported by cuda graph. Won't extend the section.
internal_section
=
[]
internal_section
=
[]
...
@@ -274,6 +274,7 @@ def get_cuda_graph_sections(program):
...
@@ -274,6 +274,7 @@ def get_cuda_graph_sections(program):
current_section
.
append
(
internal_section
[
i
])
current_section
.
append
(
internal_section
[
i
])
current_idx
.
append
(
internal_idx
[
i
])
current_idx
.
append
(
internal_idx
[
i
])
internal_section
=
[]
internal_section
=
[]
internal_idx
=
[]
current_section
.
append
(
op
)
current_section
.
append
(
op
)
current_idx
.
append
(
idx
)
current_idx
.
append
(
idx
)
else
:
else
:
...
...
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