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
机器未来
Paddle
提交
d139f2ca
P
Paddle
项目概览
机器未来
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
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
1
Issue
1
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 >>
未验证
提交
d139f2ca
编写于
4月 04, 2018
作者:
W
Wu Yi
提交者:
GitHub
4月 04, 2018
浏览文件
操作
浏览文件
下载
差异文件
Merge pull request #9595 from typhoonzero/fix_test_sendrecv_portbind
Fix sendrecv port bind
上级
56418592
b03fa889
变更
6
隐藏空白更改
内联
并排
Showing
6 changed file
with
209 addition
and
144 deletion
+209
-144
paddle/fluid/operators/CMakeLists.txt
paddle/fluid/operators/CMakeLists.txt
+1
-0
paddle/fluid/operators/detail/grpc_server.cc
paddle/fluid/operators/detail/grpc_server.cc
+4
-2
paddle/fluid/operators/detail/grpc_server.h
paddle/fluid/operators/detail/grpc_server.h
+3
-0
paddle/fluid/operators/listen_and_serv_op.cc
paddle/fluid/operators/listen_and_serv_op.cc
+126
-137
paddle/fluid/operators/listen_and_serv_op.h
paddle/fluid/operators/listen_and_serv_op.h
+53
-0
paddle/fluid/operators/send_recv_op_test.cc
paddle/fluid/operators/send_recv_op_test.cc
+22
-5
未找到文件。
paddle/fluid/operators/CMakeLists.txt
浏览文件 @
d139f2ca
...
...
@@ -193,6 +193,7 @@ if(WITH_DISTRIBUTE)
set_source_files_properties
(
send_vars_op.cc PROPERTIES COMPILE_FLAGS
${
DISTRIBUTE_COMPILE_FLAGS
}
)
op_library
(
send_barrier_op DEPS
${
DISTRIBUTE_DEPS
}
)
set_source_files_properties
(
send_barrier_op.cc PROPERTIES COMPILE_FLAGS
${
DISTRIBUTE_COMPILE_FLAGS
}
)
set_source_files_properties
(
send_recv_op_test.cc PROPERTIES COMPILE_FLAGS
${
DISTRIBUTE_COMPILE_FLAGS
}
)
cc_test
(
test_send_recv SRCS send_recv_op_test.cc DEPS prefetch_op send_op listen_and_serv_op sum_op executor
)
else
()
set
(
DEPS_OPS
${
DEPS_OPS
}
send_op prefetch_op recv_op listen_and_serv_op send_vars_op send_barrier_op
)
...
...
paddle/fluid/operators/detail/grpc_server.cc
浏览文件 @
d139f2ca
...
...
@@ -186,7 +186,8 @@ void AsyncGRPCServer::WaitClientGet(int count) {
void
AsyncGRPCServer
::
RunSyncUpdate
()
{
::
grpc
::
ServerBuilder
builder
;
builder
.
AddListeningPort
(
address_
,
::
grpc
::
InsecureServerCredentials
());
builder
.
AddListeningPort
(
address_
,
::
grpc
::
InsecureServerCredentials
(),
&
selected_port_
);
builder
.
SetMaxSendMessageSize
(
std
::
numeric_limits
<
int
>::
max
());
builder
.
SetMaxReceiveMessageSize
(
std
::
numeric_limits
<
int
>::
max
());
builder
.
RegisterService
(
&
service_
);
...
...
@@ -196,7 +197,8 @@ void AsyncGRPCServer::RunSyncUpdate() {
cq_prefetch_
=
builder
.
AddCompletionQueue
();
server_
=
builder
.
BuildAndStart
();
LOG
(
INFO
)
<<
"Server listening on "
<<
address_
<<
std
::
endl
;
LOG
(
INFO
)
<<
"Server listening on "
<<
address_
<<
" selected port: "
<<
selected_port_
;
std
::
function
<
void
()
>
send_register
=
std
::
bind
(
&
AsyncGRPCServer
::
TryToRegisterNewSendOne
,
this
);
...
...
paddle/fluid/operators/detail/grpc_server.h
浏览文件 @
d139f2ca
...
...
@@ -63,6 +63,8 @@ class AsyncGRPCServer final {
void
SetExecutor
(
framework
::
Executor
*
executor
)
{
executor_
=
executor
;
}
int
GetSelectedPort
()
{
return
selected_port_
;
}
const
ReceivedMessage
Get
()
{
return
this
->
var_recv_queue_
.
Pop
();
}
void
Push
(
const
std
::
string
&
msg_name
)
{
...
...
@@ -111,6 +113,7 @@ class AsyncGRPCServer final {
int
prefetch_blk_id_
;
framework
::
ProgramDesc
*
program_
;
framework
::
Executor
*
executor_
;
int
selected_port_
;
};
};
// namespace detail
...
...
paddle/fluid/operators/listen_and_serv_op.cc
浏览文件 @
d139f2ca
...
...
@@ -12,20 +12,14 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include <stdint.h>
#include <ostream>
#include <thread>
#include "paddle/fluid/framework/executor.h"
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/threadpool.h"
#include "paddle/fluid/operators/detail/grpc_server.h"
#include "paddle/fluid/operators/listen_and_serv_op.h"
namespace
paddle
{
namespace
operators
{
constexpr
char
kOptimizeBlock
[]
=
"OptimizeBlock"
;
void
RunServer
(
std
::
shared_ptr
<
detail
::
AsyncGRPCServer
>
service
)
{
service
->
RunSyncUpdate
();
VLOG
(
4
)
<<
"RunServer thread end"
;
...
...
@@ -66,143 +60,138 @@ static void ParallelExecuteBlocks(
for
(
size_t
i
=
0
;
i
<
fs
.
size
();
++
i
)
fs
[
i
].
wait
();
}
class
ListenAndServOp
:
public
framework
::
OperatorBase
{
public:
ListenAndServOp
(
const
std
::
string
&
type
,
const
framework
::
VariableNameMap
&
inputs
,
const
framework
::
VariableNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
)
{
if
(
!
rpc_service_
)
{
std
::
string
endpoint
=
Attr
<
std
::
string
>
(
"endpoint"
);
rpc_service_
.
reset
(
new
detail
::
AsyncGRPCServer
(
endpoint
));
server_thread_
.
reset
(
new
std
::
thread
(
RunServer
,
rpc_service_
));
}
}
ListenAndServOp
::
ListenAndServOp
(
const
std
::
string
&
type
,
const
framework
::
VariableNameMap
&
inputs
,
const
framework
::
VariableNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
)
:
OperatorBase
(
type
,
inputs
,
outputs
,
attrs
)
{}
void
Stop
()
override
{
rpc_service_
->
Push
(
LISTEN_TERMINATE_MESSAGE
);
server_thread_
->
join
();
int
ListenAndServOp
::
GetSelectedPort
()
{
return
rpc_service_
->
GetSelectedPort
();
}
void
ListenAndServOp
::
Stop
()
{
rpc_service_
->
Push
(
LISTEN_TERMINATE_MESSAGE
);
server_thread_
->
join
();
}
void
ListenAndServOp
::
RunImpl
(
const
framework
::
Scope
&
scope
,
const
platform
::
Place
&
dev_place
)
const
{
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
dev_place
);
framework
::
Scope
&
recv_scope
=
scope
.
NewScope
();
if
(
!
rpc_service_
)
{
std
::
string
endpoint
=
Attr
<
std
::
string
>
(
"endpoint"
);
rpc_service_
.
reset
(
new
detail
::
AsyncGRPCServer
(
endpoint
));
}
void
RunImpl
(
const
framework
::
Scope
&
scope
,
const
platform
::
Place
&
dev_place
)
const
override
{
platform
::
DeviceContextPool
&
pool
=
platform
::
DeviceContextPool
::
Instance
();
auto
&
dev_ctx
=
*
pool
.
Get
(
dev_place
);
framework
::
Scope
&
recv_scope
=
scope
.
NewScope
();
// FIXME(Yancey1989): initialize rpc server with lazy mode.
rpc_service_
->
SetScope
(
&
recv_scope
);
rpc_service_
->
SetDevCtx
(
&
dev_ctx
);
auto
ins
=
Inputs
(
"X"
);
auto
fan_in
=
Attr
<
int
>
(
"Fanin"
);
auto
*
block
=
Attr
<
framework
::
BlockDesc
*>
(
kOptimizeBlock
);
auto
*
program
=
block
->
Program
();
size_t
num_blocks
=
program
->
Size
();
PADDLE_ENFORCE_GE
(
num_blocks
,
2
,
"server program should have at least 2 blocks"
);
framework
::
Executor
executor
(
dev_place
);
std
::
vector
<
int
>
block_list
;
for
(
size_t
blkid
=
1
;
blkid
<
num_blocks
;
++
blkid
)
block_list
.
push_back
(
blkid
);
auto
prepared
=
executor
.
Prepare
(
*
program
,
block_list
);
prepared
.
insert
(
prepared
.
begin
(),
std
::
shared_ptr
<
framework
::
ExecutorPrepareContext
>
(
nullptr
));
// TODO(qiao) set proper fields for table lookup and update
rpc_service_
->
SetExecutor
(
&
executor
);
rpc_service_
->
SetPrefetchBlkdId
(
0
);
rpc_service_
->
SetProgram
(
program
);
// TODO(typhoonzero): change this to a while_op for every cluster-batch.
bool
exit_flag
=
false
;
// Record received sparse variables, so that
// we could reset those after execute optimize program
std
::
vector
<
framework
::
Variable
*>
sparse_vars
;
while
(
!
exit_flag
)
{
// Get from multiple trainers, we don't care about the order in which
// the gradients arrives, just add suffix 0~n and merge the gradient.
rpc_service_
->
SetCond
(
0
);
size_t
recv_var_cnt
=
0
;
int
batch_barrier
=
0
;
while
(
batch_barrier
!=
fan_in
)
{
const
detail
::
ReceivedMessage
v
=
rpc_service_
->
Get
();
auto
recv_var_name
=
v
.
first
;
if
(
recv_var_name
==
LISTEN_TERMINATE_MESSAGE
)
{
LOG
(
INFO
)
<<
"received terminate message and exit"
;
exit_flag
=
true
;
break
;
}
else
if
(
recv_var_name
==
BATCH_BARRIER_MESSAGE
)
{
VLOG
(
3
)
<<
"recv batch barrier message"
;
batch_barrier
++
;
continue
;
}
else
{
VLOG
(
3
)
<<
"received grad: "
<<
recv_var_name
;
recv_var_cnt
++
;
auto
var
=
v
.
second
->
GetVar
();
if
(
var
==
nullptr
)
{
LOG
(
ERROR
)
<<
"Can not find server side var: "
<<
recv_var_name
;
PADDLE_THROW
(
"Can not find server side var"
);
}
if
(
var
->
IsType
<
framework
::
SelectedRows
>
())
{
sparse_vars
.
push_back
(
var
);
}
}
}
if
(
exit_flag
)
{
rpc_service_
->
SetCond
(
1
);
rpc_service_
->
ShutDown
();
auto
ins
=
Inputs
(
"X"
);
auto
fan_in
=
Attr
<
int
>
(
"Fanin"
);
auto
*
block
=
Attr
<
framework
::
BlockDesc
*>
(
kOptimizeBlock
);
auto
*
program
=
block
->
Program
();
size_t
num_blocks
=
program
->
Size
();
PADDLE_ENFORCE_GE
(
num_blocks
,
2
,
"server program should have at least 2 blocks"
);
framework
::
Executor
executor
(
dev_place
);
std
::
vector
<
int
>
block_list
;
for
(
size_t
blkid
=
1
;
blkid
<
num_blocks
;
++
blkid
)
{
block_list
.
push_back
(
blkid
);
}
auto
prepared
=
executor
.
Prepare
(
*
program
,
block_list
);
// Insert placeholder for block0 which holds current op itself.
prepared
.
insert
(
prepared
.
begin
(),
std
::
shared_ptr
<
framework
::
ExecutorPrepareContext
>
(
nullptr
));
rpc_service_
->
SetScope
(
&
recv_scope
);
rpc_service_
->
SetDevCtx
(
&
dev_ctx
);
// TODO(qiao) set proper fields for table lookup and update
rpc_service_
->
SetExecutor
(
&
executor
);
rpc_service_
->
SetPrefetchBlkdId
(
0
);
rpc_service_
->
SetProgram
(
program
);
// start the server listening after all member initialized.
server_thread_
.
reset
(
new
std
::
thread
(
RunServer
,
rpc_service_
));
// FIXME(typhoonzero): do we need to wait until the server port is ready?
sleep
(
5
);
// TODO(typhoonzero): change this to a while_op for every cluster-batch.
bool
exit_flag
=
false
;
// Record received sparse variables, so that
// we could reset those after execute optimize program
std
::
vector
<
framework
::
Variable
*>
sparse_vars
;
while
(
!
exit_flag
)
{
// Get from multiple trainers, we don't care about the order in which
// the gradients arrives, just add suffix 0~n and merge the gradient.
rpc_service_
->
SetCond
(
0
);
size_t
recv_var_cnt
=
0
;
int
batch_barrier
=
0
;
while
(
batch_barrier
!=
fan_in
)
{
const
detail
::
ReceivedMessage
v
=
rpc_service_
->
Get
();
auto
recv_var_name
=
v
.
first
;
if
(
recv_var_name
==
LISTEN_TERMINATE_MESSAGE
)
{
LOG
(
INFO
)
<<
"received terminate message and exit"
;
exit_flag
=
true
;
break
;
}
// NOTE: if is_gpu_place, CUDA kernels are laugched by multiple threads
// and this will still work.
// The optimize blocks which have the same parent ID would run parallel
// TODO(Yancey1989): need to use ParallelExecutor for future
int32_t
last_parent_blkid
=
program
->
Block
(
1
).
Parent
();
std
::
vector
<
size_t
>
parallel_blkids
;
parallel_blkids
.
push_back
(
1
);
double
ts
=
detail
::
GetTimestamp
();
for
(
size_t
blkid
=
2
;
blkid
<
num_blocks
;
++
blkid
)
{
if
(
program
->
Block
(
blkid
).
Parent
()
!=
last_parent_blkid
)
{
for
(
size_t
idx
:
parallel_blkids
)
VLOG
(
3
)
<<
idx
;
ParallelExecuteBlocks
(
parallel_blkids
,
&
executor
,
prepared
,
program
,
&
recv_scope
);
parallel_blkids
.
clear
();
last_parent_blkid
=
program
->
Block
(
blkid
).
Parent
();
}
else
if
(
recv_var_name
==
BATCH_BARRIER_MESSAGE
)
{
VLOG
(
3
)
<<
"recv batch barrier message"
;
batch_barrier
++
;
continue
;
}
else
{
VLOG
(
3
)
<<
"received grad: "
<<
recv_var_name
;
recv_var_cnt
++
;
auto
var
=
v
.
second
->
GetVar
();
if
(
var
==
nullptr
)
{
LOG
(
ERROR
)
<<
"Can not find server side var: "
<<
recv_var_name
;
PADDLE_THROW
(
"Can not find server side var"
);
}
if
(
var
->
IsType
<
framework
::
SelectedRows
>
())
{
sparse_vars
.
push_back
(
var
);
}
parallel_blkids
.
push_back
(
blkid
);
}
ParallelExecuteBlocks
(
parallel_blkids
,
&
executor
,
prepared
,
program
,
&
recv_scope
);
VLOG
(
3
)
<<
"run all blocks spent "
<<
detail
::
GetTimestamp
()
-
ts
<<
"(ms)"
;
// Reset the received sparse variables, the sum operator would not
// sum the input sparse variables which rows is empty at the next
// mini-batch.
// TODO(Yancey1989): move the reset action into an operator, we couldn't
// have any hide logic in the operator.
for
(
auto
&
var
:
sparse_vars
)
{
var
->
GetMutable
<
framework
::
SelectedRows
>
()
->
mutable_rows
()
->
clear
();
}
}
if
(
exit_flag
)
{
rpc_service_
->
SetCond
(
1
);
// NOTE: does not consider barrier request retry in here, we may use
// global barrier id to resolve this.
rpc_service_
->
WaitClientGet
(
fan_in
);
sparse_vars
.
clear
();
}
// while(true)
}
rpc_service_
->
ShutDown
();
break
;
}
protected:
std
::
shared_ptr
<
detail
::
AsyncGRPCServer
>
rpc_service_
;
std
::
shared_ptr
<
std
::
thread
>
server_thread_
;
};
// NOTE: if is_gpu_place, CUDA kernels are laugched by multiple threads
// and this will still work.
// The optimize blocks which have the same parent ID would run parallel
// TODO(Yancey1989): need to use ParallelExecutor for future
int32_t
last_parent_blkid
=
program
->
Block
(
1
).
Parent
();
std
::
vector
<
size_t
>
parallel_blkids
;
parallel_blkids
.
push_back
(
1
);
double
ts
=
detail
::
GetTimestamp
();
for
(
size_t
blkid
=
2
;
blkid
<
num_blocks
;
++
blkid
)
{
if
(
program
->
Block
(
blkid
).
Parent
()
!=
last_parent_blkid
)
{
ParallelExecuteBlocks
(
parallel_blkids
,
&
executor
,
prepared
,
program
,
&
recv_scope
);
parallel_blkids
.
clear
();
last_parent_blkid
=
program
->
Block
(
blkid
).
Parent
();
}
parallel_blkids
.
push_back
(
blkid
);
}
ParallelExecuteBlocks
(
parallel_blkids
,
&
executor
,
prepared
,
program
,
&
recv_scope
);
VLOG
(
2
)
<<
"run all blocks spent "
<<
detail
::
GetTimestamp
()
-
ts
<<
"(ms)"
;
// Reset the received sparse variables, the sum operator would not
// sum the input sparse variables which rows is empty at the next
// mini-batch.
// TODO(Yancey1989): move the reset action into an operator, we couldn't
// have any hide logic in the operator.
for
(
auto
&
var
:
sparse_vars
)
{
var
->
GetMutable
<
framework
::
SelectedRows
>
()
->
mutable_rows
()
->
clear
();
}
rpc_service_
->
SetCond
(
1
);
// FIXME(typhoonzero): use another condition to sync wait clients get.
rpc_service_
->
WaitClientGet
(
fan_in
);
sparse_vars
.
clear
();
}
// while(true)
}
class
ListenAndServOpMaker
:
public
framework
::
OpProtoAndCheckerMaker
{
public:
...
...
paddle/fluid/operators/listen_and_serv_op.h
0 → 100644
浏览文件 @
d139f2ca
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include <stdint.h>
#include <ostream>
#include "paddle/fluid/framework/executor.h"
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/threadpool.h"
#include "paddle/fluid/operators/detail/grpc_server.h"
namespace
paddle
{
namespace
operators
{
constexpr
char
kOptimizeBlock
[]
=
"OptimizeBlock"
;
void
RunServer
(
std
::
shared_ptr
<
detail
::
AsyncGRPCServer
>
service
);
class
ListenAndServOp
:
public
framework
::
OperatorBase
{
public:
ListenAndServOp
(
const
std
::
string
&
type
,
const
framework
::
VariableNameMap
&
inputs
,
const
framework
::
VariableNameMap
&
outputs
,
const
framework
::
AttributeMap
&
attrs
);
int
GetSelectedPort
();
void
Stop
()
override
;
void
RunImpl
(
const
framework
::
Scope
&
scope
,
const
platform
::
Place
&
dev_place
)
const
override
;
protected:
mutable
std
::
shared_ptr
<
detail
::
AsyncGRPCServer
>
rpc_service_
;
mutable
std
::
shared_ptr
<
std
::
thread
>
server_thread_
;
};
}
// namespace operators
}
// namespace paddle
paddle/fluid/operators/send_recv_op_test.cc
浏览文件 @
d139f2ca
...
...
@@ -20,6 +20,7 @@ limitations under the License. */
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/operator.h"
#include "paddle/fluid/framework/program_desc.h"
#include "paddle/fluid/operators/listen_and_serv_op.h"
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/selected_rows_functor.h"
#include "paddle/fluid/string/printf.h"
...
...
@@ -34,6 +35,7 @@ namespace m = paddle::operators::math;
// global for simplicity.
std
::
unique_ptr
<
f
::
OperatorBase
>
listen_and_serv_op
;
int
selected_port
;
void
InitTensorsInScope
(
f
::
Scope
&
scope
,
p
::
CPUPlace
&
place
)
{
p
::
CPUDeviceContext
ctx
(
place
);
...
...
@@ -128,14 +130,16 @@ void StartServerNet(bool is_sparse) {
AddOp
(
"sum"
,
{{
"X"
,
{
"x0"
,
"x1"
}}},
{{
"Out"
,
{
"Out"
}}},
{},
optimize_block
);
f
::
AttributeMap
attrs
;
attrs
.
insert
({
"endpoint"
,
std
::
string
(
"127.0.0.1:
6174
"
)});
attrs
.
insert
({
"endpoint"
,
std
::
string
(
"127.0.0.1:
0
"
)});
attrs
.
insert
({
"Fanin"
,
1
});
attrs
.
insert
({
"ParamList"
,
std
::
vector
<
std
::
string
>
({
"Out"
})});
attrs
.
insert
({
"GradList"
,
std
::
vector
<
std
::
string
>
({
"x1"
})});
attrs
.
insert
({
"OptimizeBlock"
,
optimize_block
});
listen_and_serv_op
=
f
::
OpRegistry
::
CreateOp
(
"listen_and_serv"
,
{{
"X"
,
{
"x1"
}}},
{},
attrs
);
LOG
(
INFO
)
<<
"selected port before run "
<<
selected_port
;
listen_and_serv_op
->
Run
(
scope
,
place
);
LOG
(
INFO
)
<<
"server exit"
;
}
TEST
(
SendRecvOp
,
CPUDense
)
{
...
...
@@ -149,12 +153,19 @@ TEST(SendRecvOp, CPUDense) {
scope
.
Var
(
"RPC_CLIENT_VAR"
);
f
::
AttributeMap
attrs
;
attrs
.
insert
({
"endpoints"
,
std
::
vector
<
std
::
string
>
({
"127.0.0.1:6174"
})});
attrs
.
insert
({
"epmap"
,
std
::
vector
<
std
::
string
>
({
"127.0.0.1:6174"
})});
selected_port
=
static_cast
<
paddle
::
operators
::
ListenAndServOp
*>
(
listen_and_serv_op
.
get
())
->
GetSelectedPort
();
LOG
(
INFO
)
<<
"selected port "
<<
selected_port
;
std
::
string
endpoint
=
paddle
::
string
::
Sprintf
(
"127.0.0.1:%d"
,
selected_port
);
attrs
.
insert
({
"endpoints"
,
std
::
vector
<
std
::
string
>
({
endpoint
})});
attrs
.
insert
({
"epmap"
,
std
::
vector
<
std
::
string
>
({
endpoint
})});
auto
send_op
=
f
::
OpRegistry
::
CreateOp
(
"send"
,
{{
"X"
,
{
"x1"
}}},
{{
"Out"
,
{
"Out"
}},
{
"RPCClient"
,
{
"RPC_CLIENT_VAR"
}}},
attrs
);
LOG
(
INFO
)
<<
"before run "
<<
endpoint
;
send_op
->
Run
(
scope
,
place
);
LOG
(
INFO
)
<<
"end run"
;
auto
in_var
=
scope
.
Var
(
"x1"
);
auto
tensor
=
in_var
->
GetMutable
<
f
::
LoDTensor
>
();
...
...
@@ -167,6 +178,7 @@ TEST(SendRecvOp, CPUDense) {
for
(
int64_t
i
=
0
;
i
<
target
->
numel
();
++
i
)
{
EXPECT_EQ
(
expected
[
i
]
*
2
,
actual
[
i
]);
}
LOG
(
INFO
)
<<
"before stop"
;
listen_and_serv_op
->
Stop
();
server_thread
.
join
();
listen_and_serv_op
.
reset
(
nullptr
);
...
...
@@ -182,8 +194,13 @@ TEST(SendRecvOp, CPUSparse) {
InitSelectedRowsInScope
(
scope
,
place
);
scope
.
Var
(
"RPC_CLIENT_VAR"
);
f
::
AttributeMap
attrs
;
attrs
.
insert
({
"endpoints"
,
std
::
vector
<
std
::
string
>
({
"127.0.0.1:6174"
})});
attrs
.
insert
({
"epmap"
,
std
::
vector
<
std
::
string
>
({
"127.0.0.1:6174"
})});
selected_port
=
static_cast
<
paddle
::
operators
::
ListenAndServOp
*>
(
listen_and_serv_op
.
get
())
->
GetSelectedPort
();
LOG
(
INFO
)
<<
"selected port "
<<
selected_port
;
std
::
string
endpoint
=
paddle
::
string
::
Sprintf
(
"127.0.0.1:%d"
,
selected_port
);
attrs
.
insert
({
"endpoints"
,
std
::
vector
<
std
::
string
>
({
endpoint
})});
attrs
.
insert
({
"epmap"
,
std
::
vector
<
std
::
string
>
({
endpoint
})});
auto
send_op
=
f
::
OpRegistry
::
CreateOp
(
"send"
,
{{
"X"
,
{
"x1"
}}},
{{
"Out"
,
{
"Out"
}},
{
"RPCClient"
,
{
"RPC_CLIENT_VAR"
}}},
attrs
);
...
...
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