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
bad7ae3d
提交
bad7ae3d
authored
11月 13, 2014
作者:
Roy Xue
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Fix comments and PEP8 reformat
上级
6032d56b
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
31 行增加
和
15 行删除
+31
-15
profiling.py
theano/compile/profiling.py
+31
-15
没有找到文件。
theano/compile/profiling.py
浏览文件 @
bad7ae3d
...
...
@@ -648,7 +648,7 @@ class ProfileStats(object):
v
=
out
.
type
.
get_size
(
sh
)
sum_dense
+=
v
else
:
v
=
"Unknown"
v
=
'Variable not created'
var_mem
[
out
]
=
v
fct_memory
[
node
.
fgraph
][
node
]
.
append
(
v
)
...
...
@@ -694,8 +694,8 @@ class ProfileStats(object):
node_memory_size
=
[
0
,
0
]
running_memory_size
=
[
0
,
0
]
running_max_memory_size
=
[
0
,
0
]
node_memory_saved_by_view
=
[
0
,
0
]
node_memory_saved_by_inplace
=
[
0
,
0
]
node_memory_saved_by_view
=
0
node_memory_saved_by_inplace
=
0
# This take only the inputs/outputs dependencies.
dependencies
=
fgraph
.
profile
.
dependencies
...
...
@@ -724,22 +724,18 @@ class ProfileStats(object):
for
v
in
val
:
# TODO check the op returned a view
if
isinstance
(
v
.
type
,
theano
.
sandbox
.
cuda
.
CudaNdarrayType
):
cg
=
1
else
:
cg
=
0
if
dmap
and
idx
in
dmap
:
node_memory_saved_by_inplace
[
cg
]
+=
v
node_memory_saved_by_inplace
+=
v
# TODO check the op returned a view
elif
vmap
and
idx
in
vmap
:
node_memory_saved_by_view
[
cg
]
+=
v
node_memory_saved_by_view
+=
v
idx
+=
1
# Update the Python emulating dicts and add the memory
# allocated by the node
idx2
=
0
for
out
in
node
.
outputs
:
if
isinstance
(
v
.
type
,
theano
.
sandbox
.
cuda
.
CudaNdarrayType
):
if
isinstance
(
out
.
type
,
theano
.
sandbox
.
cuda
.
CudaNdarrayType
):
cg
=
1
else
:
cg
=
0
...
...
@@ -771,18 +767,18 @@ class ProfileStats(object):
idx2
+=
1
running_max_memory_size
[
0
]
=
max
(
running_max_memory_size
[
0
],
running_memory_size
[
0
])
running_memory_size
[
0
])
running_max_memory_size
[
1
]
=
max
(
running_max_memory_size
[
1
],
running_memory_size
[
1
])
running_memory_size
[
1
])
# Mimic the combination of Theano and Python gc
for
ins
in
node
.
inputs
:
assert
not
(
ins
in
view_of
and
viewed_by
[
ins
])
# we trac the original var, so this shouldn't happen
if
isinstance
(
v
.
type
,
theano
.
sandbox
.
cuda
.
CudaNdarrayType
):
if
isinstance
(
ins
.
type
,
theano
.
sandbox
.
cuda
.
CudaNdarrayType
):
cg
=
1
else
:
cg
=
2
cg
=
0
if
(
dependencies
[
ins
]
and
ins
not
in
fgraph
.
outputs
and
ins
.
owner
and
...
...
@@ -800,7 +796,7 @@ class ProfileStats(object):
# ins is viewed_by something else, so its
# memory isn't freed
pass
return
[
node_memory_size
,
running_memory_size
,
running_max_memory_size
,
node_memory_saved_by_inplace
,
node_memory_saved_by_view
]
...
...
@@ -1050,7 +1046,27 @@ class ProfileStats(object):
print
>>
file
,
"---"
# print >> file, " Max if no gc, inplace and view: %dKB" % int(
# round(max_sum_size / 1024))
print
>>
file
,
"----CPU Memory Profile"
print
>>
file
,
" Max if no gc (allow_gc=False):
%
dKB (
%
dKB)"
%
(
int
(
round
(
new_max_node_memory_size
/
1024.
)),
int
(
round
(
max_node_memory_size
/
1024.
)))
print
>>
file
,
" Max if linker=cvm(default):
%
dKB (
%
dKB)"
%
(
int
(
round
(
new_max_running_max_memory_size
/
1024.
)),
int
(
round
(
max_running_max_memory_size
/
1024.
)))
if
min_max_peak
:
print
>>
file
,
" Minimum peak from all valid apply node order is
%
dKB(took
%.3
fs to compute)"
%
(
int
(
round
(
min_max_peak
/
1024.
)),
min_peak_time
)
print
>>
file
,
" Memory saved if views are used:
%
dKB (
%
dKB)"
%
(
int
(
round
(
new_max_node_memory_saved_by_view
/
1024.
)),
int
(
round
(
max_node_memory_saved_by_view
/
1024.
)))
print
>>
file
,
" Memory saved if inplace ops are used:
%
dKB (
%
dKB)"
%
\
(
int
(
round
(
new_max_node_memory_saved_by_inplace
/
1024.
)),
int
(
round
(
max_node_memory_saved_by_inplace
/
1024.
)))
print
>>
file
,
" Memory saved if gc is enabled:
%
dKB (
%
dKB)"
%
(
int
(
round
(
new_max_node_memory_size
-
new_max_running_max_memory_size
)
/
1024.
),
int
(
round
(
max_node_memory_size
-
max_running_max_memory_size
)
/
1024.
))
print
>>
file
,
"----GPU Memory Profile"
print
>>
file
,
" Max if no gc (allow_gc=False):
%
dKB (
%
dKB)"
%
(
int
(
round
(
new_max_node_memory_size
/
1024.
)),
int
(
round
(
max_node_memory_size
/
1024.
)))
...
...
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