Skip to content
项目
群组
代码片段
帮助
当前项目
正在载入...
登录 / 注册
切换导航面板
P
pytensor
项目
项目
详情
活动
周期分析
仓库
仓库
文件
提交
分支
标签
贡献者
图表
比较
统计图
议题
0
议题
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
CI / CD
CI / CD
流水线
作业
日程
统计图
Wiki
Wiki
代码片段
代码片段
成员
成员
折叠边栏
关闭边栏
活动
图像
聊天
创建新问题
作业
提交
问题看板
Open sidebar
testgroup
pytensor
Commits
acc19605
提交
acc19605
authored
8月 20, 2014
作者:
Frederic
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
pep8
上级
0776ceb1
隐藏空白字符变更
内嵌
并排
正在显示
1 个修改的文件
包含
25 行增加
和
12 行删除
+25
-12
profiling.py
theano/compile/profiling.py
+25
-12
没有找到文件。
theano/compile/profiling.py
浏览文件 @
acc19605
...
@@ -672,12 +672,12 @@ class ProfileStats(object):
...
@@ -672,12 +672,12 @@ class ProfileStats(object):
dependencies
=
fgraph
.
profile
.
dependencies
dependencies
=
fgraph
.
profile
.
dependencies
# two data structure used to mimic Python gc
# two data structure used to mimic Python gc
viewed_by
=
{}
# {var1: [vars that view var1]}
viewed_by
=
{}
# {var1: [vars that view var1]}
# The len of the list is the value of python ref count. But we use a list, not just the ref count value.
# The len of the list is the value of python ref count. But we use a list, not just the ref count value.
# This is more safe to help detect potential bug in the algo
# This is more safe to help detect potential bug in the algo
for
var
in
fgraph
.
variables
:
for
var
in
fgraph
.
variables
:
viewed_by
[
var
]
=
[]
viewed_by
[
var
]
=
[]
view_of
=
{}
# {var1: original var viewed by var1}
view_of
=
{}
# {var1: original var viewed by var1}
# The orignal mean that we don't keep trac of all the intermediate relationship in the view.
# The orignal mean that we don't keep trac of all the intermediate relationship in the view.
for
node
in
order
:
for
node
in
order
:
...
@@ -695,40 +695,53 @@ class ProfileStats(object):
...
@@ -695,40 +695,53 @@ class ProfileStats(object):
node_memory_saved_by_view
+=
v
node_memory_saved_by_view
+=
v
idx
+=
1
idx
+=
1
# Update the Python emulating dicts and add the memory allocated by the node
# Update the Python emulating dicts and add the memory
# allocated by the node
idx2
=
0
idx2
=
0
for
out
in
node
.
outputs
:
for
out
in
node
.
outputs
:
if
(
dmap
and
idx2
in
dmap
)
or
(
vmap
and
idx2
in
vmap
):
if
(
dmap
and
idx2
in
dmap
)
or
(
vmap
and
idx2
in
vmap
):
# This is needed for destroy_map in case it return a partial view that is destroyed.
# This is needed for destroy_map in case it
# So the output could be different then the input.
# return a partial view that is destroyed. So
# the output could be different then the
# input.
for
ins
in
node
.
inputs
:
for
ins
in
node
.
inputs
:
assert
isinstance
(
ins
,
theano
.
Variable
)
assert
isinstance
(
ins
,
theano
.
Variable
)
view_of
[
out
]
=
view_of
.
get
(
ins
,
ins
)
# This get make that we keep trac of view only again the original
# we keep trac of view only again the original
view_of
[
out
]
=
view_of
.
get
(
ins
,
ins
)
viewed_by
[
ins
]
.
append
(
out
)
viewed_by
[
ins
]
.
append
(
out
)
else
:
else
:
running_memory_size
+=
var_mem
[
out
]
running_memory_size
+=
var_mem
[
out
]
node_memory_size
+=
var_mem
[
out
]
node_memory_size
+=
var_mem
[
out
]
idx2
+=
1
idx2
+=
1
running_max_memory_size
=
max
(
running_max_memory_size
,
running_memory_size
)
running_max_memory_size
=
max
(
running_max_memory_size
,
running_memory_size
)
# Mimic the combination of Theano and Python gc
# Mimic the combination of Theano and Python gc
for
ins
in
node
.
inputs
:
for
ins
in
node
.
inputs
:
assert
not
(
ins
in
view_of
and
viewed_by
[
ins
])
assert
not
(
ins
in
view_of
and
viewed_by
[
ins
])
# we keep trac of the original var, so this shouldn't happen
# we trac the original var, so this shouldn't happen
if
dependencies
[
ins
]
and
ins
not
in
fgraph
.
outputs
and
ins
.
owner
:
if
(
dependencies
[
ins
]
and
ins
not
in
fgraph
.
outputs
and
ins
.
owner
):
if
ins
not
in
view_of
and
not
viewed_by
.
get
(
ins
,
[]):
if
ins
not
in
view_of
and
not
viewed_by
.
get
(
ins
,
[]):
running_memory_size
-=
var_mem
[
ins
]
running_memory_size
-=
var_mem
[
ins
]
elif
ins
in
view_of
:
elif
ins
in
view_of
:
origin
=
view_of
[
ins
]
origin
=
view_of
[
ins
]
viewed_by
[
origin
]
.
remove
(
ins
)
viewed_by
[
origin
]
.
remove
(
ins
)
if
not
viewed_by
[
origin
]
and
origin
not
in
fgraph
.
inputs
:
if
(
not
viewed_by
[
origin
]
and
origin
not
in
fgraph
.
inputs
):
running_memory_size
-=
var_mem
[
origin
]
running_memory_size
-=
var_mem
[
origin
]
else
:
else
:
# ins is viewed_by something else, so its memory isn't freed
# ins is viewed_by something else, so its
# memory isn't freed
pass
pass
return
[
node_memory_size
,
running_memory_size
,
running_max_memory_size
,
node_memory_saved_by_inplace
,
node_memory_saved_by_view
]
return
[
node_memory_size
,
running_memory_size
,
running_max_memory_size
,
node_memory_saved_by_inplace
,
node_memory_saved_by_view
]
def
count_minimum_peak
(
node_list
,
fgraph
,
nodes_mem
):
def
count_minimum_peak
(
node_list
,
fgraph
,
nodes_mem
):
global
mem_count
,
mem_bound
,
max_mem_count
global
mem_count
,
mem_bound
,
max_mem_count
...
...
编写
预览
Markdown
格式
0%
重试
或
添加新文件
添加附件
取消
您添加了
0
人
到此讨论。请谨慎行事。
请先完成此评论的编辑!
取消
请
注册
或者
登录
后发表评论