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
16c6f70b
提交
16c6f70b
authored
1月 11, 2017
作者:
Reyhane Askari
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
GpuJoin perform and C function work inplace
上级
46eaf9df
隐藏空白字符变更
内嵌
并排
正在显示
3 个修改的文件
包含
100 行增加
和
52 行删除
+100
-52
basic_ops.py
theano/gpuarray/basic_ops.py
+74
-50
test_basic_ops.py
theano/gpuarray/tests/test_basic_ops.py
+21
-0
basic.py
theano/tensor/basic.py
+5
-2
没有找到文件。
theano/gpuarray/basic_ops.py
浏览文件 @
16c6f70b
...
...
@@ -1255,16 +1255,14 @@ class GpuJoin(HideC, Join):
# start from index 1.
self
.
view_map
=
{
0
:
[
1
+
view
]}
# def __str__(self):
# if self.view == -1:
# return "Join"
# else:
# return super(Join, self).__str__()
# def __setstate__(self, d):
# self.__dict__.update(d)
# if not hasattr(self, "view"):
# self.view = -1
def
__str__
(
self
):
if
self
.
view
==
-
1
:
return
self
.
__class__
.
__name__
else
:
return
"
%
s{
%
s}"
%
(
self
.
__class__
.
__name__
,
", "
.
join
(
"
%
s=
%
r"
%
(
p
,
getattr
(
self
,
p
))
for
p
in
self
.
__props__
))
def
make_node
(
self
,
axis
,
*
tensors
):
node
=
Join
.
make_node
(
self
,
axis
,
*
tensors
)
...
...
@@ -1296,59 +1294,85 @@ class GpuJoin(HideC, Join):
if
(
view
!=
-
1
)
and
numpy
.
all
(
[
tensor
.
shape
[
axis
]
==
0
for
tensor
in
tensors
[
0
:
view
]
+
tensors
[
view
+
1
:]]):
import
ipdb
;
ipdb
.
set_trace
()
out
[
0
]
=
tensors
[
view
]
else
:
out
[
0
]
=
pygpu
.
concatenate
(
tensors
,
axis
=
axis
,
context
=
ctx
)
.
astype
(
node
.
outputs
[
0
]
.
dtype
)
def
c_code_cache_version
(
self
):
return
return
(
2
,)
return
(
3
,)
def
c_support_code
_
(
self
):
def
c_support_code
(
self
):
return
"""
#if PY_MAJOR_VERSION >= 3
#define PyInt_AsLong PyLong_AsLong
#endif
"""
#if PY_MAJOR_VERSION >= 3
#define PyInt_AsLong PyLong_AsLong
#endif
"""
def
c_code_
(
self
,
node
,
name
,
inputs
,
out_
,
sub
):
def
c_headers
(
self
):
return
[
'<numpy_compat.h>'
]
def
c_code
(
self
,
node
,
name
,
inputs
,
out_
,
sub
):
axis
,
tensors
=
inputs
[
0
],
inputs
[
1
:]
copy_to_list
=
[]
restype
=
pygpu
.
gpuarray
.
dtype_to_typecode
(
node
.
outputs
[
0
]
.
dtype
)
for
i
,
inp
in
enumerate
(
inputs
[
1
:]):
view
=
self
.
view
non_empty_tensor
=
tensors
[
view
]
for
i
,
inp
in
enumerate
(
tensors
):
copy_to_list
.
append
(
"als[
%
s] = &
%
s->ga;"
%
(
i
,
inp
))
return
"""
const GpuArray **als = (const GpuArray **)PyMem_Malloc(sizeof(GpuArray *) *
n
=
len
(
tensors
)
fail
=
sub
[
'fail'
]
out
=
out_
[
0
]
copy_inputs_to_list
=
'
\n
'
.
join
(
copy_to_list
)
restype
=
restype
ctx
=
sub
[
'params'
]
code
=
"""
const GpuArray **als = (const GpuArray **)PyMem_Malloc(sizeof(GpuArray *) *
%(n)
s);
if (als == NULL) {
PyErr_NoMemory();
%(fail)
s
}
%(copy_inputs_to_list)
s
Py_XDECREF(
%(out)
s);
{
int axis = PyInt_AsLong((PyObject *)
%(axis)
s);
if (axis < 0) {
if (axis == -1 && PyErr_Occurred()) {
%(fail)
s
}
axis += als[0]->nd;
if (axis < 0) {
PyErr_SetString(PyExc_IndexError, "invalid axis");
%(fail)
s
}
}
%(out)
s = pygpu_concatenate(als,
%(n)
s, axis,
%(restype)
s, (PyObject *)&PyGpuArrayType,
%(ctx)
s);
}
PyMem_Free(als);
if (
%(out)
s == NULL)
%(fail)
s
"""
%
dict
(
n
=
len
(
inputs
[
1
:]),
fail
=
sub
[
'fail'
],
out
=
out_
[
0
],
axis
=
inputs
[
0
],
copy_inputs_to_list
=
'
\n
'
.
join
(
copy_to_list
),
restype
=
restype
,
ctx
=
sub
[
'params'
])
if (als == NULL) {
PyErr_NoMemory();
%(fail)
s
}
%(copy_inputs_to_list)
s
Py_XDECREF(
%(out)
s);
{
int axis = PyInt_AsLong((PyObject *)
%(axis)
s);
if (axis < 0) {
if (axis == -1 && PyErr_Occurred()) {
%(fail)
s
}
axis += als[0]->nd;
if (axis < 0) {
PyErr_SetString(PyExc_IndexError, "invalid axis");
%(fail)
s
}
}
int tensors_lens_sum = 0"""
%
locals
()
for
inp
in
tensors
:
code
+=
""" + PyGpuArray_DIM(
%(inp)
s, axis)"""
%
locals
()
code
+=
""";
\n
tensors_lens_sum -= PyGpuArray_DIM(
%(non_empty_tensor)
s, axis);
if(
%(view)
s != -1 && tensors_lens_sum == 0){
Py_XDECREF(
%(out)
s);
Py_INCREF(
%(non_empty_tensor)
s);
%(out)
s =
%(non_empty_tensor)
s;
}
else{
%(out)
s = pygpu_concatenate(als,
%(n)
s, axis,
%(restype)
s, (PyObject *)&PyGpuArrayType,
%(ctx)
s);
}
PyMem_Free(als);
}
if (
%(out)
s == NULL)
%(fail)
s
"""
%
locals
()
return
code
gpu_join
=
GpuJoin
()
...
...
theano/gpuarray/tests/test_basic_ops.py
浏览文件 @
16c6f70b
...
...
@@ -453,3 +453,24 @@ def test_hostfromgpu_shape_i():
assert
isinstance
(
topo
[
1
]
.
op
,
theano
.
compile
.
Shape_i
)
assert
isinstance
(
topo
[
2
]
.
op
,
theano
.
tensor
.
opt
.
MakeVector
)
assert
tuple
(
f
(
cv
))
==
(
5
,
4
)
def
test_Gpujoin_inplace
():
"""Test Gpujoin to work inplace.
This function tests the case when several elements are passed to the
Gpujoin function but all except one of them are empty. In this case
Gpujoin should work inplace and the output should be the view of the
non-empty element.
"""
s
=
T
.
lscalar
()
data
=
numpy
.
array
([
3
,
4
,
5
],
dtype
=
theano
.
config
.
floatX
)
x
=
theano
.
shared
(
data
,
borrow
=
True
)
z
=
T
.
zeros
((
s
,))
join
=
GpuJoin
(
view
=
0
)
c
=
join
(
0
,
x
,
z
)
f
=
theano
.
function
([
s
],
theano
.
Out
(
c
,
borrow
=
True
))
assert
x
.
get_value
(
borrow
=
True
,
return_internal_type
=
True
)
is
f
(
0
)
assert
numpy
.
allclose
(
f
(
0
),
[
3
,
4
,
5
])
theano/tensor/basic.py
浏览文件 @
16c6f70b
...
...
@@ -3897,9 +3897,12 @@ class Join(Op):
def
__str__
(
self
):
if
self
.
view
==
-
1
:
return
"Join"
return
self
.
__class__
.
__name__
else
:
return
super
(
Join
,
self
)
.
__str__
()
return
"
%
s{
%
s}"
%
(
self
.
__class__
.
__name__
,
", "
.
join
(
"
%
s=
%
r"
%
(
p
,
getattr
(
self
,
p
))
for
p
in
self
.
__props__
))
def
__setstate__
(
self
,
d
):
self
.
__dict__
.
update
(
d
)
...
...
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