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
OPTHREE
Paddle
提交
9261dff4
P
Paddle
项目概览
OPTHREE
/
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
9261dff4
编写于
3月 25, 2022
作者:
Z
zn
提交者:
GitHub
3月 25, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[MLU]add allreduce max/prod/min mlu kernel (#40792)
上级
01b688c0
变更
12
隐藏空白更改
内联
并排
Showing
12 changed file
with
304 addition
and
29 deletion
+304
-29
paddle/fluid/operators/collective/c_allreduce_max_op_mlu.cc
paddle/fluid/operators/collective/c_allreduce_max_op_mlu.cc
+26
-0
paddle/fluid/operators/collective/c_allreduce_min_op_mlu.cc
paddle/fluid/operators/collective/c_allreduce_min_op_mlu.cc
+26
-0
paddle/fluid/operators/collective/c_allreduce_prod_op_mlu.cc
paddle/fluid/operators/collective/c_allreduce_prod_op_mlu.cc
+26
-0
python/paddle/fluid/tests/unittests/mlu/CMakeLists.txt
python/paddle/fluid/tests/unittests/mlu/CMakeLists.txt
+4
-1
python/paddle/fluid/tests/unittests/mlu/collective_allgather_op.py
...ddle/fluid/tests/unittests/mlu/collective_allgather_op.py
+3
-3
python/paddle/fluid/tests/unittests/mlu/collective_allreduce_op.py
...ddle/fluid/tests/unittests/mlu/collective_allreduce_op.py
+4
-4
python/paddle/fluid/tests/unittests/mlu/collective_broadcast_op.py
...ddle/fluid/tests/unittests/mlu/collective_broadcast_op.py
+2
-2
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_max.py
...luid/tests/unittests/mlu/test_collective_allreduce_max.py
+13
-12
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_min.py
...luid/tests/unittests/mlu/test_collective_allreduce_min.py
+56
-0
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_prod.py
...uid/tests/unittests/mlu/test_collective_allreduce_prod.py
+56
-0
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_sum.py
...luid/tests/unittests/mlu/test_collective_allreduce_sum.py
+56
-0
python/paddle/fluid/tests/unittests/mlu/test_collective_base_mlu.py
...dle/fluid/tests/unittests/mlu/test_collective_base_mlu.py
+32
-7
未找到文件。
paddle/fluid/operators/collective/c_allreduce_max_op_mlu.cc
0 → 100644
浏览文件 @
9261dff4
/* Copyright (c) 2022 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. */
#include "paddle/fluid/operators/collective/c_allreduce_op.h"
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OP_MLU_KERNEL
(
c_allreduce_max
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMax
,
float
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMax
,
plat
::
float16
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMax
,
int
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMax
,
int16_t
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMax
,
int8_t
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMax
,
uint8_t
>
)
paddle/fluid/operators/collective/c_allreduce_min_op_mlu.cc
0 → 100644
浏览文件 @
9261dff4
/* Copyright (c) 2022 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. */
#include "paddle/fluid/operators/collective/c_allreduce_op.h"
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OP_MLU_KERNEL
(
c_allreduce_min
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMin
,
float
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMin
,
plat
::
float16
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMin
,
int
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMin
,
int16_t
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMin
,
int8_t
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedMin
,
uint8_t
>
)
paddle/fluid/operators/collective/c_allreduce_prod_op_mlu.cc
0 → 100644
浏览文件 @
9261dff4
/* Copyright (c) 2022 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. */
#include "paddle/fluid/operators/collective/c_allreduce_op.h"
namespace
ops
=
paddle
::
operators
;
namespace
plat
=
paddle
::
platform
;
REGISTER_OP_MLU_KERNEL
(
c_allreduce_prod
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedProd
,
float
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedProd
,
plat
::
float16
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedProd
,
int
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedProd
,
int16_t
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedProd
,
int8_t
>
,
ops
::
CAllReduceOpMLUKernel
<
ops
::
kRedProd
,
uint8_t
>
)
python/paddle/fluid/tests/unittests/mlu/CMakeLists.txt
浏览文件 @
9261dff4
...
...
@@ -21,7 +21,10 @@ if (WITH_MLU)
bash_test_modules
(
test_launch_nproc_mlu START_BASH test_launch_nproc_mlu.sh ENVS PADDLE_BINARY_DIR=
${
PADDLE_BINARY_DIR
}
)
bash_test_modules
(
test_c_comm_init_op_mlu START_BASH test_c_comm_init_op_mlu.sh ENVS PADDLE_BINARY_DIR=
${
PADDLE_BINARY_DIR
}
)
set_tests_properties
(
test_collective_broadcast PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_sum PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_max PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_min PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_prod PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allgather PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_broadcast_api_mlu PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_api_mlu PROPERTIES TIMEOUT 120
)
...
...
python/paddle/fluid/tests/unittests/mlu/collective_allgather_op.py
浏览文件 @
9261dff4
...
...
@@ -41,14 +41,14 @@ class TestCollectiveAllgather(TestCollectiveRunnerBase):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
):
def
get_model
(
self
,
main_prog
,
startup_program
,
col_type
):
ring_id
=
0
nranks
=
2
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
layers
.
data
(
name
=
"tindata"
,
shape
=
[
10
,
1000
],
dtype
=
'float32'
)
toutdata
=
main_prog
.
current_block
().
create_var
(
name
=
"outofgather"
,
name
=
"outof
all
gather"
,
dtype
=
'float32'
,
type
=
core
.
VarDesc
.
VarType
.
LOD_TENSOR
,
persistable
=
False
,
...
...
@@ -68,4 +68,4 @@ class TestCollectiveAllgather(TestCollectiveRunnerBase):
if
__name__
==
"__main__"
:
runtime_main
(
TestCollectiveAllgather
,
"allgather"
,
0
)
runtime_main
(
TestCollectiveAllgather
)
python/paddle/fluid/tests/unittests/mlu/collective_allreduce_op.py
浏览文件 @
9261dff4
...
...
@@ -42,19 +42,19 @@ class TestCollectiveAllreduce(TestCollectiveRunnerBase):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
):
def
get_model
(
self
,
main_prog
,
startup_program
,
col_type
):
ring_id
=
0
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
layers
.
data
(
name
=
"tindata"
,
shape
=
[
10
,
1000
],
dtype
=
'float32'
)
toutdata
=
main_prog
.
current_block
().
create_var
(
name
=
"outof
allreduce"
,
name
=
"outof
"
+
col_type
,
dtype
=
'float32'
,
type
=
core
.
VarDesc
.
VarType
.
LOD_TENSOR
,
persistable
=
False
,
stop_gradient
=
False
)
main_prog
.
global_block
().
append_op
(
type
=
"c_
allreduce_sum"
,
type
=
"c_
"
+
col_type
,
inputs
=
{
'X'
:
tindata
},
attrs
=
{
'ring_id'
:
ring_id
},
outputs
=
{
'Out'
:
toutdata
})
...
...
@@ -67,4 +67,4 @@ class TestCollectiveAllreduce(TestCollectiveRunnerBase):
if
__name__
==
"__main__"
:
runtime_main
(
TestCollectiveAllreduce
,
"allreduce"
,
0
)
runtime_main
(
TestCollectiveAllreduce
)
python/paddle/fluid/tests/unittests/mlu/collective_broadcast_op.py
100644 → 100755
浏览文件 @
9261dff4
...
...
@@ -42,7 +42,7 @@ class TestCollectiveBroadcast(TestCollectiveRunnerBase):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
):
def
get_model
(
self
,
main_prog
,
startup_program
,
col_type
):
ring_id
=
0
rootid
=
1
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
...
...
@@ -69,4 +69,4 @@ class TestCollectiveBroadcast(TestCollectiveRunnerBase):
if
__name__
==
"__main__"
:
runtime_main
(
TestCollectiveBroadcast
,
"broadcast"
,
0
)
runtime_main
(
TestCollectiveBroadcast
)
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce.py
→
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce
_max
.py
浏览文件 @
9261dff4
...
...
@@ -27,27 +27,28 @@ class TestCAllreduceOp(TestDistBase):
def
_setup_config
(
self
):
pass
def
test_allreduce_fp32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce"
,
def
test_allreduce_
max_
fp32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce
_max
"
,
"float32"
)
def
test_allreduce_fp16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce"
,
def
test_allreduce_
max_
fp16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce
_max
"
,
"float16"
)
def
test_allreduce_int32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce"
,
def
test_allreduce_
max_
int32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce
_max
"
,
"int32"
)
def
test_allreduce_int16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce"
,
def
test_allreduce_
max_
int16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce
_max
"
,
"int16"
)
def
test_allreduce_int8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce"
,
"int8"
)
def
test_allreduce_max_int8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_max"
,
"int8"
)
def
test_allreduce_uint8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce"
,
def
test_allreduce_
max_
uint8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce
_max
"
,
"uint8"
)
...
...
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_min.py
0 → 100644
浏览文件 @
9261dff4
# Copyright (c) 2022 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.
from
__future__
import
print_function
import
sys
import
unittest
import
numpy
as
np
import
paddle
from
test_collective_base_mlu
import
TestDistBase
paddle
.
enable_static
()
class
TestCAllreduceOp
(
TestDistBase
):
def
_setup_config
(
self
):
pass
def
test_allreduce_min_fp32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_min"
,
"float32"
)
def
test_allreduce_min_fp16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_min"
,
"float16"
)
def
test_allreduce_min_int32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_min"
,
"int32"
)
def
test_allreduce_min_int16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_min"
,
"int16"
)
def
test_allreduce_min_int8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_min"
,
"int8"
)
def
test_allreduce_min_uint8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_min"
,
"uint8"
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_prod.py
0 → 100644
浏览文件 @
9261dff4
# Copyright (c) 2022 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.
from
__future__
import
print_function
import
sys
import
unittest
import
numpy
as
np
import
paddle
from
test_collective_base_mlu
import
TestDistBase
paddle
.
enable_static
()
class
TestCAllreduceOp
(
TestDistBase
):
def
_setup_config
(
self
):
pass
def
test_allreduce_prod_fp32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_prod"
,
"float32"
)
def
test_allreduce_prod_fp16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_prod"
,
"float16"
)
def
test_allreduce_prod_int32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_prod"
,
"int32"
)
def
test_allreduce_prod_int16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_prod"
,
"int16"
)
def
test_allreduce_prod_int8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_prod"
,
"int8"
)
def
test_allreduce_prod_uint8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_prod"
,
"uint8"
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/mlu/test_collective_allreduce_sum.py
0 → 100644
浏览文件 @
9261dff4
# Copyright (c) 2022 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.
from
__future__
import
print_function
import
sys
import
unittest
import
numpy
as
np
import
paddle
from
test_collective_base_mlu
import
TestDistBase
paddle
.
enable_static
()
class
TestCAllreduceOp
(
TestDistBase
):
def
_setup_config
(
self
):
pass
def
test_allreduce_sum_fp32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_sum"
,
"float32"
)
def
test_allreduce_sum_fp16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_sum"
,
"float16"
)
def
test_allreduce_sum_int32
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_sum"
,
"int32"
)
def
test_allreduce_sum_int16
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_sum"
,
"int16"
)
def
test_allreduce_sum_int8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_sum"
,
"int8"
)
def
test_allreduce_sum_uint8
(
self
):
self
.
check_with_place
(
"collective_allreduce_op.py"
,
"allreduce_sum"
,
"uint8"
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/mlu/test_collective_base_mlu.py
浏览文件 @
9261dff4
...
...
@@ -53,7 +53,7 @@ def DataTypeCast(date_type):
class
TestCollectiveRunnerBase
(
object
):
def
get_model
(
self
,
train_prog
,
startup_prog
):
def
get_model
(
self
,
train_prog
,
startup_prog
,
col_type
):
raise
NotImplementedError
(
"get model should be implemented by child class."
)
...
...
@@ -129,7 +129,7 @@ class TestCollectiveRunnerBase(object):
self
.
initCommunicator
(
startup_prog
,
rank
,
nranks
,
True
,
current_endpoint
,
endpoints
)
self
.
rank
=
rank
result
=
self
.
get_model
(
train_prog
,
startup_prog
)
result
=
self
.
get_model
(
train_prog
,
startup_prog
,
args
[
"col_type"
]
)
device_id
=
int
(
os
.
getenv
(
"FLAGS_selected_mlus"
,
"0"
))
place
=
fluid
.
MLUPlace
(
device_id
)
exe
=
fluid
.
Executor
(
place
)
...
...
@@ -143,7 +143,7 @@ class TestCollectiveRunnerBase(object):
sys
.
stdout
.
buffer
.
write
(
pickle
.
dumps
(
out
))
def
runtime_main
(
test_class
,
col_type
,
sub_type
):
def
runtime_main
(
test_class
):
args
=
{}
model
=
test_class
()
args
[
"deviceid"
]
=
os
.
getenv
(
"FLAGS_selected_mlus"
)
...
...
@@ -151,7 +151,7 @@ def runtime_main(test_class, col_type, sub_type):
args
[
"trainernum"
]
=
int
(
os
.
getenv
(
"PADDLE_TRAINERS_NUM"
))
args
[
"endpoints"
]
=
os
.
getenv
(
'PADDLE_TRAINER_ENDPOINTS'
)
args
[
"currentendpoint"
]
=
os
.
getenv
(
"PADDLE_CURRENT_ENDPOINT"
)
args
[
"col_type"
]
=
col_type
args
[
"col_type"
]
=
os
.
getenv
(
"COL_TYPE"
)
args
[
"data_type"
]
=
os
.
getenv
(
"DATA_TYPE"
)
model
.
run_trainer
(
args
)
...
...
@@ -185,7 +185,7 @@ class TestDistBase(unittest.TestCase):
def
_run_cluster
(
self
,
model_file
,
envs
):
worker_endpoints
=
self
.
_ps_endpoints
.
split
(
","
)
w0_ep
,
w1_ep
=
worker_endpoints
#print("w0_ep:",w0_ep," w1_ep:",w1_ep)
env0
=
{
"FLAGS_selected_mlus"
:
"0"
,
"PADDLE_TRAINER_ID"
:
"0"
,
...
...
@@ -209,7 +209,7 @@ class TestDistBase(unittest.TestCase):
tr1_cmd
=
tr_cmd
%
(
self
.
_python_interp
,
model_file
)
tr0_pipe
=
open
(
"/tmp/tr0_err.log"
,
"wb"
)
tr1_pipe
=
open
(
"/tmp/tr1_err.log"
,
"wb"
)
#print(tr0_cmd)
tr0_proc
=
subprocess
.
Popen
(
tr0_cmd
.
strip
().
split
(),
stdout
=
subprocess
.
PIPE
,
...
...
@@ -246,6 +246,7 @@ class TestDistBase(unittest.TestCase):
"LD_PRELOAD"
:
os
.
getenv
(
"LD_PRELOAD"
,
""
),
"GLOG_v"
:
"3"
,
"DATA_TYPE"
:
data_type
,
"COL_TYPE"
:
col_type
,
}
required_envs
.
update
(
need_envs
)
if
check_error_log
:
...
...
@@ -262,7 +263,7 @@ class TestDistBase(unittest.TestCase):
need_result
=
input2
self
.
assertTrue
(
np
.
allclose
(
tr0_out
,
need_result
))
self
.
assertTrue
(
np
.
allclose
(
tr1_out
,
need_result
))
elif
col_type
==
"allreduce"
:
elif
col_type
==
"allreduce
_sum
"
:
need_result
=
input1
+
input2
self
.
assertTrue
(
np
.
allclose
(
...
...
@@ -270,6 +271,30 @@ class TestDistBase(unittest.TestCase):
self
.
assertTrue
(
np
.
allclose
(
tr1_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
elif
col_type
==
"allreduce_prod"
:
need_result
=
input1
*
input2
self
.
assertTrue
(
np
.
allclose
(
tr0_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
self
.
assertTrue
(
np
.
allclose
(
tr1_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
elif
col_type
==
"allreduce_max"
:
need_result
=
np
.
maximum
(
input1
,
input2
)
self
.
assertTrue
(
np
.
allclose
(
tr0_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
self
.
assertTrue
(
np
.
allclose
(
tr1_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
elif
col_type
==
"allreduce_min"
:
need_result
=
np
.
minimum
(
input1
,
input2
)
self
.
assertTrue
(
np
.
allclose
(
tr0_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
self
.
assertTrue
(
np
.
allclose
(
tr1_out
,
need_result
,
rtol
=
1e-05
,
atol
=
1e-05
))
elif
col_type
==
"allgather"
:
need_result
=
np
.
vstack
((
input1
,
input2
))
self
.
assertTrue
(
np
.
allclose
(
tr0_out
,
need_result
))
...
...
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