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
极客李华
Paddle
提交
a5e1cf5a
P
Paddle
项目概览
极客李华
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
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
0
Issue
0
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
体验新版 GitCode,发现更多精彩内容 >>
提交
a5e1cf5a
编写于
12月 27, 2017
作者:
Y
Yang Yu
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Rename API of DeviceContext
Make them as usual names.
上级
94096ae5
变更
23
隐藏空白更改
内联
并排
Showing
23 changed file
with
59 addition
and
92 deletion
+59
-92
paddle/framework/init.cc
paddle/framework/init.cc
+1
-1
paddle/framework/operator.cc
paddle/framework/operator.cc
+2
-2
paddle/operators/array_operator.h
paddle/operators/array_operator.h
+2
-2
paddle/operators/array_to_lod_tensor_op.cc
paddle/operators/array_to_lod_tensor_op.cc
+3
-2
paddle/operators/assign_op.cc
paddle/operators/assign_op.cc
+2
-2
paddle/operators/cond_op.cc
paddle/operators/cond_op.cc
+2
-2
paddle/operators/feed_op.cc
paddle/operators/feed_op.cc
+2
-2
paddle/operators/fetch_op.cc
paddle/operators/fetch_op.cc
+2
-2
paddle/operators/fill_constant_op.cc
paddle/operators/fill_constant_op.cc
+2
-2
paddle/operators/fill_op.cc
paddle/operators/fill_op.cc
+3
-2
paddle/operators/load_op.cc
paddle/operators/load_op.cc
+2
-2
paddle/operators/lod_tensor_to_array_op.cc
paddle/operators/lod_tensor_to_array_op.cc
+3
-2
paddle/operators/merge_lod_tensor_op.cc
paddle/operators/merge_lod_tensor_op.cc
+2
-2
paddle/operators/recurrent_op.cc
paddle/operators/recurrent_op.cc
+5
-4
paddle/operators/reorder_lod_tensor_by_rank_op.cc
paddle/operators/reorder_lod_tensor_by_rank_op.cc
+2
-2
paddle/operators/save_op.cc
paddle/operators/save_op.cc
+2
-2
paddle/operators/shrink_rnn_memory_op.cc
paddle/operators/shrink_rnn_memory_op.cc
+2
-2
paddle/operators/split_lod_tensor_op.cc
paddle/operators/split_lod_tensor_op.cc
+2
-2
paddle/operators/tensor_array_read_write_op.cc
paddle/operators/tensor_array_read_write_op.cc
+6
-4
paddle/platform/device_context.cc
paddle/platform/device_context.cc
+1
-19
paddle/platform/device_context.h
paddle/platform/device_context.h
+3
-9
paddle/platform/device_context_test.cu
paddle/platform/device_context_test.cu
+7
-22
paddle/platform/nccl_test.cu
paddle/platform/nccl_test.cu
+1
-1
未找到文件。
paddle/framework/init.cc
浏览文件 @
a5e1cf5a
...
...
@@ -71,7 +71,7 @@ bool InitDevices(const std::vector<std::string> &devices) {
places
.
emplace_back
(
platform
::
CPUPlace
());
LOG
(
WARNING
)
<<
"Not specified CPU device, create CPU by Default."
;
}
platform
::
DeviceContextPool
::
Create
(
places
);
platform
::
DeviceContextPool
::
Init
(
places
);
return
true
;
}
...
...
paddle/framework/operator.cc
浏览文件 @
a5e1cf5a
...
...
@@ -388,8 +388,8 @@ void OperatorWithKernel::Run(const Scope& scope,
const
platform
::
Place
&
place
)
const
{
RuntimeInferShapeContext
infer_shape_ctx
(
*
this
,
scope
);
this
->
InferShape
(
&
infer_shape_ctx
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
dev_ctx
=
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
dev_ctx
=
pool
.
Get
(
place
);
// check if op[type] has kernel registered.
auto
&
all_op_kernels
=
AllOpKernels
();
...
...
paddle/operators/array_operator.h
浏览文件 @
a5e1cf5a
...
...
@@ -35,8 +35,8 @@ class ArrayOp : public framework::OperatorBase {
PADDLE_ENFORCE_EQ
(
i_tensor
.
numel
(),
1
);
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
size_t
offset
;
if
(
platform
::
is_gpu_place
(
i_tensor
.
place
()))
{
...
...
paddle/operators/array_to_lod_tensor_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -106,8 +106,9 @@ class ArrayToLoDTensorOp : public framework::OperatorBase {
}
auto
slice
=
out
->
Slice
(
out_offset
,
out_offset
+
len
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
CopyFrom
(
x
[
x_idx
].
Slice
(
start_offset
,
end_offset
),
place
,
dev_ctx
,
&
slice
);
...
...
paddle/operators/assign_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -82,8 +82,8 @@ class AssignOp : public framework::OperatorBase {
out
!=
nullptr
,
"The Output(Out) should not be null if the Input(X) is set."
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
VisitVarType
(
*
x
,
AssignFunctor
(
out
,
dev_ctx
));
}
...
...
paddle/operators/cond_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -195,8 +195,8 @@ void CondOp::MergeDataFromSubnet(const framework::Scope& scope,
void
CondOp
::
Run
(
const
Scope
&
scope
,
const
platform
::
Place
&
place
)
const
{
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
PrepareDataForSubnet
(
scope
,
dev_ctx
);
std
::
vector
<
framework
::
Scope
*>&
sub_scopes
=
GetSubScopes
(
scope
);
...
...
paddle/operators/feed_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -49,8 +49,8 @@ class FeedOp : public framework::OperatorBase {
auto
*
out_item
=
out_var
->
GetMutable
<
framework
::
FeedFetchType
>
();
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
CopyFrom
(
feed_item
,
place
,
dev_ctx
,
out_item
);
out_item
->
set_lod
(
feed_item
.
lod
());
...
...
paddle/operators/fetch_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -52,8 +52,8 @@ class FetchOp : public framework::OperatorBase {
// FIXME(yuyang18): Should we assume the fetch operator always generate
// CPU outputs?
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
CopyFrom
(
src_item
,
platform
::
CPUPlace
(),
dev_ctx
,
&
dst_item
);
dev_ctx
.
Wait
();
...
...
paddle/operators/fill_constant_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -49,8 +49,8 @@ class FillConstantOp : public framework::OperatorBase {
out
.
mutable_data
(
dev_place
,
framework
::
ToTypeIndex
(
data_type
));
}
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
dev_place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
dev_place
);
math
::
set_constant
(
dev_ctx
,
&
out
,
value
);
}
};
...
...
paddle/operators/fill_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -69,8 +69,9 @@ class FillOp : public framework::OperatorBase {
if
(
!
force_cpu
&&
platform
::
is_gpu_place
(
place
))
{
// Copy tensor to out
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
CopyFrom
(
tensor
,
place
,
dev_ctx
,
&
out
);
}
}
...
...
paddle/operators/load_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -40,8 +40,8 @@ class LoadOp : public framework::OperatorBase {
auto
*
tensor
=
out_var
->
GetMutable
<
framework
::
LoDTensor
>
();
framework
::
DeserializeFromStream
(
fin
,
tensor
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
if
(
platform
::
is_gpu_place
(
place
))
{
// copy CPU to GPU
...
...
paddle/operators/lod_tensor_to_array_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -88,8 +88,9 @@ class LoDTensorToArrayOp : public framework::OperatorBase {
auto
slice
=
out
[
i
].
Slice
(
static_cast
<
int
>
(
offset
),
static_cast
<
int
>
(
offset
+
len
));
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
CopyFrom
(
x
.
Slice
(
static_cast
<
int
>
(
each_range
.
begin
),
static_cast
<
int
>
(
each_range
.
end
)),
...
...
paddle/operators/merge_lod_tensor_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -30,8 +30,8 @@ class MergeLoDTensorOp : public framework::OperatorBase {
void
Run
(
const
framework
::
Scope
&
scope
,
const
platform
::
Place
&
dev_place
)
const
override
{
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
dev_place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
dev_place
);
auto
&
x
=
scope
.
FindVar
(
Input
(
"X"
))
->
Get
<
framework
::
LoDTensor
>
();
auto
&
mask
=
scope
.
FindVar
(
Input
(
"Mask"
))
->
Get
<
framework
::
LoDTensor
>
();
...
...
paddle/operators/recurrent_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -272,8 +272,9 @@ class RecurrentOp : public RecurrentBase {
false
/*create_local_scope*/
);
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
// Copy inside::output -> outside::output
// outside::output[seq_offset: seq_offset + 1] = inside::output
...
...
@@ -326,8 +327,8 @@ class RecurrentGradOp : public RecurrentBase {
auto
*
program
=
block
->
Program
();
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
for
(
size_t
step_id
=
0
;
step_id
<
seq_len
;
++
step_id
)
{
size_t
seq_offset
=
reverse
?
step_id
:
seq_len
-
step_id
-
1
;
...
...
paddle/operators/reorder_lod_tensor_by_rank_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -131,8 +131,8 @@ class ReorderLoDTensorByRankTableBase : public framework::OperatorBase {
auto
x_sliced
=
x
.
Slice
(
x_offset
,
x_offset
+
len
);
auto
out_sliced
=
out
->
Slice
(
out_offset
,
out_offset
+
len
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
CopyFrom
(
x_sliced
,
out_sliced
.
place
(),
dev_ctx
,
&
out_sliced
);
out_offset
+=
len
;
return
out_offset
;
...
...
paddle/operators/save_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -91,8 +91,8 @@ class SaveOp : public framework::OperatorBase {
auto
&
tensor
=
var
->
Get
<
framework
::
LoDTensor
>
();
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
SerializeToStream
(
fout
,
tensor
,
dev_ctx
);
}
...
...
paddle/operators/shrink_rnn_memory_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -106,8 +106,8 @@ class ShrinkRNNMemoryGradOp : public ArrayOp {
dx_tensor
.
mutable_data
(
x_tensor
.
place
(),
x_tensor
.
type
());
// get device context from pool
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
if
(
dout_var
==
nullptr
)
{
// dx_tensor fill zero
math
::
set_constant
(
dev_ctx
,
&
dx_tensor
,
0.0
f
);
...
...
paddle/operators/split_lod_tensor_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -45,8 +45,8 @@ class SplitLoDTensorOp : public framework::OperatorBase {
auto
&
x_lod
=
x
.
lod
();
auto
&
mask_dim
=
mask
.
dims
();
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
dev_place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
dev_place
);
std
::
unique_ptr
<
framework
::
LoDTensor
>
cpu_mask
{
new
framework
::
LoDTensor
()};
if
(
platform
::
is_cpu_place
(
mask
.
place
()))
{
...
...
paddle/operators/tensor_array_read_write_op.cc
浏览文件 @
a5e1cf5a
...
...
@@ -40,8 +40,9 @@ class WriteToArrayOp : public ArrayOp {
if
(
x_tensor
.
memory_size
()
>
0
)
{
auto
*
out_tensor
=
&
out
->
at
(
offset
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
CopyFrom
(
x_tensor
,
place
,
dev_ctx
,
out_tensor
);
out_tensor
->
set_lod
(
x_tensor
.
lod
());
...
...
@@ -132,8 +133,9 @@ class ReadFromArrayOp : public ArrayOp {
auto
*
out_tensor
=
out
->
GetMutable
<
framework
::
LoDTensor
>
();
size_t
offset
=
GetOffset
(
scope
,
place
);
if
(
offset
<
x_array
.
size
())
{
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Get
();
auto
&
dev_ctx
=
*
pool
.
Borrow
(
place
);
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
place
);
framework
::
CopyFrom
(
x_array
[
offset
],
place
,
dev_ctx
,
out_tensor
);
out_tensor
->
set_lod
(
x_array
[
offset
].
lod
());
}
else
{
...
...
paddle/platform/device_context.cc
浏览文件 @
a5e1cf5a
...
...
@@ -17,7 +17,7 @@ namespace platform {
DeviceContextPool
*
DeviceContextPool
::
pool
=
nullptr
;
const
platform
::
DeviceContext
*
DeviceContextPool
::
Borrow
(
const
platform
::
DeviceContext
*
DeviceContextPool
::
Get
(
const
platform
::
Place
&
place
)
{
auto
it
=
device_contexts_
.
find
(
place
);
if
(
it
==
device_contexts_
.
end
())
{
...
...
@@ -28,24 +28,6 @@ const platform::DeviceContext* DeviceContextPool::Borrow(
return
it
->
second
;
}
std
::
vector
<
const
platform
::
DeviceContext
*>
DeviceContextPool
::
Borrow
(
const
std
::
vector
<
platform
::
Place
>&
places
)
{
PADDLE_ENFORCE_GT
(
places
.
size
(),
0
);
PADDLE_ENFORCE_LE
(
places
.
size
(),
device_contexts_
.
size
());
std
::
vector
<
const
platform
::
DeviceContext
*>
borrowed_contexts
;
for
(
auto
&
place
:
places
)
{
auto
it
=
device_contexts_
.
find
(
place
);
if
(
it
!=
device_contexts_
.
end
())
{
borrowed_contexts
.
emplace_back
(
it
->
second
);
}
else
{
PADDLE_THROW
(
"'Place' is not supported, Please re-compile with WITH_GPU "
"option"
);
}
}
return
borrowed_contexts
;
}
DeviceContextPool
::
DeviceContextPool
(
const
std
::
vector
<
platform
::
Place
>&
places
)
{
PADDLE_ENFORCE_GT
(
places
.
size
(),
0
);
...
...
paddle/platform/device_context.h
浏览文件 @
a5e1cf5a
...
...
@@ -109,13 +109,13 @@ class DeviceContextPool {
public:
explicit
DeviceContextPool
(
const
std
::
vector
<
platform
::
Place
>&
places
);
static
DeviceContextPool
&
Get
()
{
static
DeviceContextPool
&
Instance
()
{
PADDLE_ENFORCE_NOT_NULL
(
pool
,
"Need to Create DeviceContextPool first!"
);
return
*
pool
;
}
/*! \brief Create should only called by Init function */
static
DeviceContextPool
&
Create
(
const
std
::
vector
<
platform
::
Place
>&
places
)
{
static
DeviceContextPool
&
Init
(
const
std
::
vector
<
platform
::
Place
>&
places
)
{
if
(
pool
==
nullptr
)
{
pool
=
new
DeviceContextPool
(
places
);
}
...
...
@@ -123,13 +123,7 @@ class DeviceContextPool {
}
/*! \brief Return handle of single device context. */
const
platform
::
DeviceContext
*
Borrow
(
const
platform
::
Place
&
place
);
/*! \brief Return handle of multi-device context. */
std
::
vector
<
const
platform
::
DeviceContext
*>
Borrow
(
const
std
::
vector
<
platform
::
Place
>&
places
);
~
DeviceContextPool
()
{}
const
platform
::
DeviceContext
*
Get
(
const
platform
::
Place
&
place
);
private:
static
DeviceContextPool
*
pool
;
...
...
paddle/platform/device_context_test.cu
浏览文件 @
a5e1cf5a
...
...
@@ -71,35 +71,20 @@ TEST(Device, DeviceContextPool) {
using
paddle
::
platform
::
CPUPlace
;
using
paddle
::
platform
::
CUDAPlace
;
DeviceContextPool
&
pool
=
DeviceContextPool
::
Get
();
auto
cpu_dev_ctx1
=
pool
.
Borrow
(
CPUPlace
());
auto
cpu_dev_ctx2
=
pool
.
Borrow
(
CPUPlace
());
EXPECT_TRUE
(
cpu_dev_ctx2
==
cpu_dev_ctx1
);
DeviceContextPool
&
pool
=
DeviceContextPool
::
Instance
();
auto
cpu_dev_ctx1
=
pool
.
Get
(
CPUPlace
());
auto
cpu_dev_ctx2
=
pool
.
Get
(
CPUPlace
());
ASSERT_EQ
(
cpu_dev_ctx2
,
cpu_dev_ctx1
);
std
::
vector
<
Place
>
gpu_places
;
int
count
=
paddle
::
platform
::
GetCUDADeviceCount
();
for
(
int
i
=
0
;
i
<
count
;
++
i
)
{
gpu_places
.
emplace_back
(
CUDAPlace
(
i
));
}
auto
dev_ctxs
=
pool
.
Borrow
(
gpu_places
);
for
(
size_t
i
=
0
;
i
<
dev_ctxs
.
size
();
++
i
)
{
auto
*
dev_ctx
=
static_cast
<
const
CUDADeviceContext
*>
(
dev_ctxs
[
i
]);
// check same as CUDAPlace(i)
CUDAPlace
place
=
boost
::
get
<
CUDAPlace
>
(
dev_ctx
->
GetPlace
());
EXPECT_EQ
(
place
.
GetDeviceId
(),
static_cast
<
int
>
(
i
));
auto
dev_ctx
=
pool
.
Get
(
CUDAPlace
(
i
));
ASSERT_NE
(
dev_ctx
,
nullptr
);
}
}
int
main
(
int
argc
,
char
**
argv
)
{
int
dev_count
=
paddle
::
platform
::
GetCUDADeviceCount
();
if
(
dev_count
<=
1
)
{
LOG
(
WARNING
)
<<
"Cannot test multi-gpu DeviceContextPool, because the CUDA "
"device count is "
<<
dev_count
;
return
0
;
}
std
::
vector
<
paddle
::
platform
::
Place
>
places
;
places
.
emplace_back
(
paddle
::
platform
::
CPUPlace
());
...
...
@@ -109,7 +94,7 @@ int main(int argc, char** argv) {
}
VLOG
(
0
)
<<
" DeviceCount "
<<
count
;
paddle
::
platform
::
DeviceContextPool
::
Create
(
places
);
paddle
::
platform
::
DeviceContextPool
::
Init
(
places
);
testing
::
InitGoogleTest
(
&
argc
,
argv
);
return
RUN_ALL_TESTS
();
...
...
paddle/platform/nccl_test.cu
浏览文件 @
a5e1cf5a
...
...
@@ -144,7 +144,7 @@ int main(int argc, char** argv) {
}
VLOG
(
0
)
<<
" DeviceCount "
<<
count
;
paddle
::
platform
::
DeviceContextPool
::
Create
(
places
);
paddle
::
platform
::
DeviceContextPool
::
Init
(
places
);
testing
::
InitGoogleTest
(
&
argc
,
argv
);
return
RUN_ALL_TESTS
();
...
...
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