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
5ee08ced
提交
5ee08ced
authored
8月 01, 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
Something wrong with the mem_freed
上级
adc26d0e
隐藏空白字符变更
内嵌
并排
正在显示
2 个修改的文件
包含
50 行增加
和
78 行删除
+50
-78
profiling.py
theano/compile/profiling.py
+47
-78
vm.py
theano/gof/vm.py
+3
-0
没有找到文件。
theano/compile/profiling.py
浏览文件 @
5ee08ced
...
...
@@ -25,7 +25,7 @@ import numpy
import
theano
from
theano.gof
import
graph
from
theano.gof
.vm
import
compute_gc_dependencies
from
theano.gof
import
vm
from
theano.configparser
import
AddConfigVar
,
BoolParam
,
IntParam
...
...
@@ -692,19 +692,16 @@ class ProfileStats(object):
return
[
node_memory_size
,
running_memory_size
,
running_max_memory_size
,
node_memory_saved_by_inplace
,
node_memory_saved_by_view
]
# count the minimum peak
best_order
=
[]
minimum_peak
=
0
max_minimum_peak
=
0
def
count_minimum_peak
(
node_list
,
fgraph
,
nodes_mem
):
global
maybe_executed
global
mem_count
,
mem_bound
global
maybe_executed
,
mem_count
,
mem_bound
order
=
[]
min_order
=
[]
node_list
=
list
(
node_list
)
min_mem
=
sys
.
maxint
current_mem
=
0
check_len
=
len
(
node_list
)
def
check_node_state
(
node
):
"""
...
...
@@ -737,21 +734,48 @@ class ProfileStats(object):
executables_nodes
.
add
(
c
)
mem_count
=
0
mem_bound
=
0
mem_bound
=
numpy
.
inf
# dependencies = {}
dependencies
=
fgraph
.
profile
.
dependencies
print
dependencies
# for node in node_list[0].inputs[0]:
# dependencies[node] = []
# if val.owner and val.clients:
# ls = []
# for c in val.clients:
# if c[0] is not 'output':
# ls += c[0].outputs
# dependencies[val] += ls
def
min_memory_generator
(
executables_nodes
):
global
mem_count
,
mem_bound
for
node
in
executables_nodes
:
new_exec_nodes
=
executables_nodes
.
copy
()
new_exec_nodes
.
remove
(
node
)
mem
=
count_node_memory
(
node
)
mem_count
+=
mem
if
mem_bound
:
# check if at this time, mem_current and mem_bound
# if higher use 'continue'
if
mem_count
>
mem_bound
:
mem_count
-=
mem
continue
mem_created
=
sum
(
nodes_mem
[
node
])
mem_count
+=
mem_created
# Add memory created
# dependencies = {}
mem_freed
=
0
for
val
in
node
.
outputs
:
if
(
dependencies
[
val
]
and
val
.
owner
and
val
not
in
fgraph
.
outputs
):
if
all
(
compute_map
[
v
][
0
]
for
v
in
dependencies
[
val
]):
mem_freed
+=
var_mem
[
val
]
print
mem_freed
mem_count
-=
mem_freed
# Reduce memory freed
if
mem_count
>
mem_bound
:
mem_count
-=
mem_created
mem_count
+=
mem_freed
continue
for
var
in
node
.
outputs
:
compute_map
[
var
][
0
]
=
1
for
var
in
node
.
outputs
:
...
...
@@ -760,77 +784,22 @@ class ProfileStats(object):
new_exec_nodes
.
add
(
c
)
if
not
new_exec_nodes
:
yield
[
node
]
if
not
mem_bound
:
# initial the mem_bound
mem_bound
=
mem_count
elif
mem_current
<
mem_bound
:
if
mem_count
<
mem_bound
:
# update the mem_bound
mem_bound
=
mem_c
urre
nt
mem_bound
=
mem_c
ou
nt
else
:
for
p
in
min_memory_generator
(
new_exec_nodes
):
yield
[
node
]
+
p
mem_count
-=
mem
mem_count
-=
mem_created
mem_count
+=
mem_freed
# reset
for
var
in
node
.
outputs
:
compute_map
[
var
][
0
]
=
0
def
count_node_memory
(
node
):
dependencies
=
compute_gc_dependencies
(
node
)
mem
=
0
for
val
in
node
.
inputs
:
if
(
dependencies
[
val
]
and
val
.
owner
and
val
not
in
fgraph
.
outputs
):
mem
+=
node
.
inputs
.
index
(
val
)
return
mem
def
count_min_memory
(
order
,
thunk_old_storage
,
nodes_mem
):
running_memory_size
=
0
running_max_memory_size
=
0
node_idx
=
0
for
node
in
order
:
val
=
nodes_mem
[
node
]
dmap
=
getattr
(
node
.
op
,
'destroy_map'
,
None
)
vmap
=
getattr
(
node
.
op
,
'view_map'
,
None
)
idx
=
0
for
v
in
val
:
# TODO check the op returned a view
if
dmap
and
idx
in
dmap
:
continue
elif
vmap
and
idx
in
vmap
:
continue
elif
not
isinstance
(
v
,
str
):
running_memory_size
+=
v
idx
+=
1
if
running_memory_size
>
running_max_memory_size
:
running_max_memory_size
=
running_memory_size
old_storage
=
thunk_old_storage
[
node_idx
]
for
old_s
in
old_storage
:
old_v
=
var_mem
[
node
.
inputs
[
old_s
]]
if
not
isinstance
(
old_v
,
str
):
running_memory_size
-=
old_v
node_idx
+=
1
return
running_max_memory_size
for
order
in
min_memory_generator
(
executables_nodes
):
post_thunk_old_storage
=
[]
computed
,
last_user
=
theano
.
gof
.
link
.
gc_helper
(
order
)
for
node
in
order
:
post_thunk_old_storage
.
append
([
input_idx
for
input_idx
,
input
in
enumerate
(
node
.
inputs
)
if
(
input
in
computed
)
and
(
input
not
in
fgraph
.
outputs
)
and
node
==
last_user
[
input
]])
current_mem
=
count_min_memory
(
order
,
post_thunk_old_storage
,
nodes_mem
)
if
current_mem
<
min_mem
:
min_mem
=
current_mem
min_order
=
order
return
min_order
,
mem_bound
continue
return
mem_bound
for
fgraph
,
nodes_mem
in
fct_memory
.
iteritems
():
# Sum of the size of all variables in bytes
...
...
@@ -882,7 +851,7 @@ class ProfileStats(object):
node_list
=
fgraph
.
apply_nodes
_
,
minimum_peak
=
count_minimum_peak
(
node_list
,
fgraph
,
nodes_mem
)
minimum_peak
=
count_minimum_peak
(
node_list
,
fgraph
,
nodes_mem
)
# for the best order, we dont use it now
max_minimum_peak
=
max
(
max_minimum_peak
,
minimum_peak
)
...
...
theano/gof/vm.py
浏览文件 @
5ee08ced
...
...
@@ -147,6 +147,9 @@ class VM(object):
if
hasattr
(
self
,
'node_cleared_order'
):
profile
.
node_cleared_order
=
self
.
node_cleared_order
[:]
if
hasattr
(
self
,
'dependencies'
):
profile
.
dependencies
=
self
.
dependencies
.
copy
()
# clear the timer info out of the buffers
for
i
in
xrange
(
len
(
self
.
call_times
)):
self
.
call_times
[
i
]
=
0.0
...
...
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