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
BaiXuePrincess
Paddle
提交
7a92e74b
P
Paddle
项目概览
BaiXuePrincess
/
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
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
未验证
提交
7a92e74b
编写于
9月 06, 2022
作者:
W
Wen Sun
提交者:
GitHub
9月 06, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Completes basic dtypes for collective api in eager mode (#45574)
上级
1137677a
变更
23
展开全部
显示空白变更内容
内联
并排
Showing
23 changed file
with
679 addition
and
221 deletion
+679
-221
paddle/fluid/distributed/collective/ProcessGroupNCCL.cc
paddle/fluid/distributed/collective/ProcessGroupNCCL.cc
+11
-2
paddle/phi/kernels/cpu/concat_kernel.cc
paddle/phi/kernels/cpu/concat_kernel.cc
+2
-0
paddle/phi/kernels/gpu/concat_kernel.cu
paddle/phi/kernels/gpu/concat_kernel.cu
+1
-0
python/paddle/distributed/collective.py
python/paddle/distributed/collective.py
+183
-196
python/paddle/fluid/tests/unittests/collective/CMakeLists.txt
...on/paddle/fluid/tests/unittests/collective/CMakeLists.txt
+26
-3
python/paddle/fluid/tests/unittests/collective/collective_alltoall_api_dygraph.py
...s/unittests/collective/collective_alltoall_api_dygraph.py
+3
-6
python/paddle/fluid/tests/unittests/collective/collective_alltoall_single_api_dygraph.py
...ests/collective/collective_alltoall_single_api_dygraph.py
+36
-0
python/paddle/fluid/tests/unittests/collective/collective_broadcast_api_dygraph.py
.../unittests/collective/collective_broadcast_api_dygraph.py
+36
-0
python/paddle/fluid/tests/unittests/collective/collective_isend_irecv_api_dygraph.py
...nittests/collective/collective_isend_irecv_api_dygraph.py
+40
-0
python/paddle/fluid/tests/unittests/collective/collective_reduce_api_dygraph.py
...sts/unittests/collective/collective_reduce_api_dygraph.py
+36
-0
python/paddle/fluid/tests/unittests/collective/collective_reduce_scatter_api_dygraph.py
...tests/collective/collective_reduce_scatter_api_dygraph.py
+37
-0
python/paddle/fluid/tests/unittests/collective/collective_scatter_api_dygraph.py
...ts/unittests/collective/collective_scatter_api_dygraph.py
+42
-0
python/paddle/fluid/tests/unittests/collective/test_collective_alltoall_api.py
...ests/unittests/collective/test_collective_alltoall_api.py
+10
-4
python/paddle/fluid/tests/unittests/collective/test_collective_alltoall_single_api.py
...ittests/collective/test_collective_alltoall_single_api.py
+39
-0
python/paddle/fluid/tests/unittests/collective/test_collective_broadcast_api.py
...sts/unittests/collective/test_collective_broadcast_api.py
+25
-0
python/paddle/fluid/tests/unittests/collective/test_collective_isend_irecv_api.py
...s/unittests/collective/test_collective_isend_irecv_api.py
+39
-0
python/paddle/fluid/tests/unittests/collective/test_collective_reduce_api.py
.../tests/unittests/collective/test_collective_reduce_api.py
+25
-0
python/paddle/fluid/tests/unittests/collective/test_collective_reduce_scatter_api.py
...nittests/collective/test_collective_reduce_scatter_api.py
+39
-0
python/paddle/fluid/tests/unittests/collective/test_collective_scatter_api.py
...tests/unittests/collective/test_collective_scatter_api.py
+25
-0
python/paddle/fluid/tests/unittests/collective/test_collective_sendrecv_api.py
...ests/unittests/collective/test_collective_sendrecv_api.py
+7
-2
python/paddle/fluid/tests/unittests/collective/testslist.csv
python/paddle/fluid/tests/unittests/collective/testslist.csv
+6
-3
python/paddle/fluid/tests/unittests/test_collective_api_base.py
.../paddle/fluid/tests/unittests/test_collective_api_base.py
+6
-0
python/paddle/tensor/manipulation.py
python/paddle/tensor/manipulation.py
+5
-5
未找到文件。
paddle/fluid/distributed/collective/ProcessGroupNCCL.cc
浏览文件 @
7a92e74b
...
...
@@ -738,14 +738,23 @@ void* GetPointerByOffset(void* raw_pointer,
}
else
if
(
type
==
experimental
::
DataType
::
FLOAT64
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
double
*>
(
raw_pointer
)
+
offset
);
}
else
if
(
type
==
experimental
::
DataType
::
FLOAT16
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
int16_t
*>
(
raw_pointer
)
+
offset
);
}
else
if
(
type
==
experimental
::
DataType
::
INT32
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
int32_t
*>
(
raw_pointer
)
+
offset
);
}
else
if
(
type
==
experimental
::
DataType
::
INT64
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
int64_t
*>
(
raw_pointer
)
+
offset
);
}
else
if
(
type
==
experimental
::
DataType
::
FLOAT16
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
int16_t
*>
(
raw_pointer
)
+
}
else
if
(
type
==
experimental
::
DataType
::
INT8
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
int8_t
*>
(
raw_pointer
)
+
offset
);
}
else
if
(
type
==
experimental
::
DataType
::
UINT8
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
uint8_t
*>
(
raw_pointer
)
+
offset
);
}
else
if
(
type
==
experimental
::
DataType
::
BOOL
)
{
return
reinterpret_cast
<
void
*>
(
reinterpret_cast
<
bool
*>
(
raw_pointer
)
+
offset
);
}
else
{
PADDLE_THROW
(
platform
::
errors
::
Unimplemented
(
...
...
paddle/phi/kernels/cpu/concat_kernel.cc
浏览文件 @
7a92e74b
...
...
@@ -124,6 +124,8 @@ PD_REGISTER_KERNEL(concat,
int64_t
,
int
,
uint8_t
,
int8_t
,
phi
::
dtype
::
float16
,
phi
::
dtype
::
bfloat16
,
phi
::
dtype
::
complex
<
float
>
,
phi
::
dtype
::
complex
<
double
>
)
{}
paddle/phi/kernels/gpu/concat_kernel.cu
浏览文件 @
7a92e74b
...
...
@@ -121,6 +121,7 @@ PD_REGISTER_KERNEL(concat,
int64_t
,
int
,
uint8_t
,
int8_t
,
phi
::
dtype
::
float16
,
phi
::
dtype
::
bfloat16
,
phi
::
dtype
::
complex
<
float
>
,
...
...
python/paddle/distributed/collective.py
浏览文件 @
7a92e74b
此差异已折叠。
点击以展开。
python/paddle/fluid/tests/unittests/collective/CMakeLists.txt
浏览文件 @
7a92e74b
...
...
@@ -78,7 +78,7 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
test_collective_alltoall_api MODULES test_collective_alltoall_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:
${
PADDLE_BINARY_DIR
}
/python"
)
set_tests_properties
(
test_collective_alltoall_api
PROPERTIES TIMEOUT
"
12
0"
LABELS
"RUN_TYPE=DIST"
)
PROPERTIES TIMEOUT
"
30
0"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
bash_test_modules
(
...
...
@@ -92,6 +92,14 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
)
set_tests_properties
(
test_collective_alltoall_single PROPERTIES TIMEOUT
"350"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
test_collective_alltoall_single_api MODULES
test_collective_alltoall_single_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:
${
PADDLE_BINARY_DIR
}
/python"
)
set_tests_properties
(
test_collective_alltoall_single_api
PROPERTIES TIMEOUT
"300"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
test_collective_barrier_api MODULES test_collective_barrier_api ENVS
...
...
@@ -117,7 +125,7 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
test_collective_broadcast_api MODULES test_collective_broadcast_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:
${
PADDLE_BINARY_DIR
}
/python"
)
set_tests_properties
(
test_collective_broadcast_api
PROPERTIES TIMEOUT
"
12
0"
LABELS
"RUN_TYPE=DIST"
)
PROPERTIES TIMEOUT
"
30
0"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
...
...
@@ -141,6 +149,13 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
set_tests_properties
(
test_collective_global_scatter
PROPERTIES TIMEOUT
"200"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
test_collective_isend_irecv_api MODULES test_collective_isend_irecv_api
ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:
${
PADDLE_BINARY_DIR
}
/python"
)
set_tests_properties
(
test_collective_isend_irecv_api
PROPERTIES TIMEOUT
"300"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
test_collective_optimizer MODULES test_collective_optimizer ENVS
...
...
@@ -186,6 +201,14 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
)
set_tests_properties
(
test_collective_reduce_scatter PROPERTIES TIMEOUT
"350"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
test_collective_reduce_scatter_api MODULES
test_collective_reduce_scatter_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:
${
PADDLE_BINARY_DIR
}
/python"
)
set_tests_properties
(
test_collective_reduce_scatter_api
PROPERTIES TIMEOUT
"300"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
test_collective_scatter MODULES test_collective_scatter ENVS
...
...
@@ -212,7 +235,7 @@ if((WITH_GPU OR WITH_ROCM) AND (LINUX))
test_collective_sendrecv_api MODULES test_collective_sendrecv_api ENVS
"http_proxy=;https_proxy=;PYTHONPATH=..:
${
PADDLE_BINARY_DIR
}
/python"
)
set_tests_properties
(
test_collective_sendrecv_api
PROPERTIES TIMEOUT
"
12
0"
LABELS
"RUN_TYPE=DIST"
)
PROPERTIES TIMEOUT
"
30
0"
LABELS
"RUN_TYPE=DIST"
)
endif
()
if
((
WITH_GPU OR WITH_ROCM
)
AND
(
LINUX
))
py_test_modules
(
...
...
python/paddle/fluid/tests/unittests/collective/collective_alltoall_api_dygraph.py
浏览文件 @
7a92e74b
...
...
@@ -45,12 +45,9 @@ class TestCollectiveAllToAllAPI(TestCollectiveAPIRunnerBase):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
tindata
=
paddle
.
split
(
tindata
,
2
,
axis
=
0
)
tout_data
=
[]
paddle
.
distributed
.
alltoall
(
tindata
,
tout_data
)
output_data
=
[]
for
data
in
tout_data
:
output_data
.
append
(
data
.
numpy
())
return
output_data
toutdata
=
[]
paddle
.
distributed
.
alltoall
(
tindata
,
toutdata
)
return
[
data
.
numpy
()
for
data
in
toutdata
]
if
__name__
==
"__main__"
:
...
...
python/paddle/fluid/tests/unittests/collective/collective_alltoall_single_api_dygraph.py
0 → 100644
浏览文件 @
7a92e74b
# 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
paddle
import
paddle.fluid
as
fluid
import
test_collective_api_base
as
test_base
class
TestCollectiveAllToAllSingleAPI
(
test_base
.
TestCollectiveAPIRunnerBase
):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
,
rank
,
indata
=
None
):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
toutdata
=
paddle
.
to_tensor
(
indata
)
paddle
.
distributed
.
alltoall_single
(
tindata
,
toutdata
)
return
[
toutdata
.
numpy
()]
if
__name__
==
"__main__"
:
test_base
.
runtime_main
(
TestCollectiveAllToAllSingleAPI
,
"alltoall"
)
python/paddle/fluid/tests/unittests/collective/collective_broadcast_api_dygraph.py
0 → 100644
浏览文件 @
7a92e74b
# 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
paddle
import
paddle.fluid
as
fluid
import
unittest
import
test_collective_api_base
as
test_base
class
TestCollectiveBroadcastAPI
(
test_base
.
TestCollectiveAPIRunnerBase
):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
,
rank
,
indata
=
None
):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
paddle
.
distributed
.
broadcast
(
tindata
,
src
=
1
)
return
[
tindata
.
numpy
()]
if
__name__
==
"__main__"
:
test_base
.
runtime_main
(
TestCollectiveBroadcastAPI
,
"broadcast"
)
python/paddle/fluid/tests/unittests/collective/collective_isend_irecv_api_dygraph.py
0 → 100644
浏览文件 @
7a92e74b
# 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
paddle
import
paddle.fluid
as
fluid
import
unittest
import
test_collective_api_base
as
test_base
class
TestCollectiveIsendIrecvAPI
(
test_base
.
TestCollectiveAPIRunnerBase
):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
,
rank
,
indata
=
None
):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
if
rank
==
0
:
task
=
paddle
.
distributed
.
isend
(
tindata
,
dst
=
1
)
else
:
task
=
paddle
.
distributed
.
irecv
(
tindata
,
src
=
0
)
task
.
wait
()
return
[
tindata
.
numpy
()]
if
__name__
==
"__main__"
:
test_base
.
runtime_main
(
TestCollectiveIsendIrecvAPI
,
"sendrecv"
)
python/paddle/fluid/tests/unittests/collective/collective_reduce_api_dygraph.py
0 → 100644
浏览文件 @
7a92e74b
# 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
paddle
import
paddle.fluid
as
fluid
import
unittest
import
test_collective_api_base
as
test_base
class
TestCollectiveReduceAPI
(
test_base
.
TestCollectiveAPIRunnerBase
):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
,
rank
,
indata
=
None
):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
paddle
.
distributed
.
reduce
(
tindata
,
dst
=
0
)
return
[
tindata
.
numpy
()]
if
__name__
==
"__main__"
:
test_base
.
runtime_main
(
TestCollectiveReduceAPI
,
"reduce"
)
python/paddle/fluid/tests/unittests/collective/collective_reduce_scatter_api_dygraph.py
0 → 100644
浏览文件 @
7a92e74b
# 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
paddle
import
paddle.fluid
as
fluid
import
unittest
import
test_collective_api_base
as
test_base
class
TestCollectiveReduceScatterAPI
(
test_base
.
TestCollectiveAPIRunnerBase
):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
,
rank
,
indata
=
None
):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
subdata1
,
subdata2
=
paddle
.
split
(
tindata
,
2
,
axis
=
0
)
paddle
.
distributed
.
reduce_scatter
(
subdata1
,
[
subdata1
,
subdata2
])
return
[
subdata1
.
numpy
()]
if
__name__
==
"__main__"
:
test_base
.
runtime_main
(
TestCollectiveReduceScatterAPI
,
"reduce_scatter"
)
python/paddle/fluid/tests/unittests/collective/collective_scatter_api_dygraph.py
0 → 100644
浏览文件 @
7a92e74b
# 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
paddle
import
paddle.fluid
as
fluid
import
unittest
import
test_collective_api_base
as
test_base
class
TestCollectiveScatterAPI
(
test_base
.
TestCollectiveAPIRunnerBase
):
def
__init__
(
self
):
self
.
global_ring_id
=
0
def
get_model
(
self
,
main_prog
,
startup_program
,
rank
,
indata
=
None
):
with
fluid
.
program_guard
(
main_prog
,
startup_program
):
tindata
=
paddle
.
to_tensor
(
indata
)
subdata1
,
subdata2
=
paddle
.
split
(
tindata
,
2
,
axis
=
0
)
if
rank
==
0
:
paddle
.
distributed
.
scatter
(
subdata1
,
src
=
1
)
else
:
paddle
.
distributed
.
scatter
(
subdata1
,
tensor_list
=
[
subdata1
,
subdata2
],
src
=
1
)
return
[
subdata1
.
numpy
()]
if
__name__
==
"__main__"
:
test_base
.
runtime_main
(
TestCollectiveScatterAPI
,
"scatter"
)
python/paddle/fluid/tests/unittests/collective/test_collective_alltoall_api.py
浏览文件 @
7a92e74b
...
...
@@ -31,10 +31,16 @@ class TestCollectiveAllToAllAPI(TestDistBase):
self
.
check_with_place
(
"collective_alltoall_api.py"
,
"alltoall"
,
"nccl"
)
def
test_alltoall_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_alltoall_api_dygraph.py"
,
"alltoall"
,
"nccl"
,
static_mode
=
"0"
)
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
...
...
python/paddle/fluid/tests/unittests/collective/test_collective_alltoall_single_api.py
0 → 100644
浏览文件 @
7a92e74b
# 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.
import
unittest
import
paddle
import
test_collective_api_base
as
test_base
class
TestCollectiveAllToAllSingleAPI
(
test_base
.
TestDistBase
):
def
_setup_config
(
self
):
pass
def
test_alltooall_single_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_alltoall_single_api_dygraph.py"
,
"alltoall"
,
"nccl"
,
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/collective/test_collective_broadcast_api.py
浏览文件 @
7a92e74b
...
...
@@ -35,6 +35,31 @@ class TestCollectiveBroadcastAPI(TestDistBase):
self
.
check_with_place
(
"collective_broadcast_api.py"
,
"broadcast"
,
"gloo"
,
"0"
)
def
test_broadcast_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_broadcast_api_dygraph.py"
,
"broadcast"
,
"nccl"
,
static_mode
=
"0"
,
dtype
=
dtype
)
def
test_broadcast_gloo_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_broadcast_api_dygraph.py"
,
"broadcast"
,
"gloo"
,
"0"
,
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/collective/test_collective_isend_irecv_api.py
0 → 100644
浏览文件 @
7a92e74b
# 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.
import
unittest
import
paddle
import
test_collective_api_base
as
test_base
class
TestCollectiveIsendIrecvAPI
(
test_base
.
TestDistBase
):
def
_setup_config
(
self
):
pass
def
test_isend_irecv_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_isend_irecv_api_dygraph.py"
,
"sendrecv"
,
"nccl"
,
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/collective/test_collective_reduce_api.py
浏览文件 @
7a92e74b
...
...
@@ -38,6 +38,31 @@ class TestCollectiveReduceAPI(TestDistBase):
def
test_reduce_gloo
(
self
):
self
.
check_with_place
(
"collective_reduce_api.py"
,
"reduce"
,
"gloo"
,
"1"
)
def
test_reduce_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_reduce_api_dygraph.py"
,
"reduce"
,
"nccl"
,
static_mode
=
"0"
,
dtype
=
dtype
)
def
test_reduce_gloo_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_reduce_api_dygraph.py"
,
"reduce"
,
"gloo"
,
"1"
,
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/collective/test_collective_reduce_scatter_api.py
0 → 100644
浏览文件 @
7a92e74b
# 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.
import
unittest
import
paddle
import
test_collective_api_base
as
test_base
class
TestCollectiveReduceScatterAPI
(
test_base
.
TestDistBase
):
def
_setup_config
(
self
):
pass
def
test_reduce_scatter_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_reduce_scatter_api_dygraph.py"
,
"reduce_scatter"
,
"nccl"
,
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/collective/test_collective_scatter_api.py
浏览文件 @
7a92e74b
...
...
@@ -34,6 +34,31 @@ class TestCollectiveScatterAPI(TestDistBase):
def
test_scatter_nccl
(
self
):
self
.
check_with_place
(
"collective_scatter_api.py"
,
"scatter"
,
"nccl"
)
def
test_scatter_nccl_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_scatter_api_dygraph.py"
,
"scatter"
,
"nccl"
,
static_mode
=
"0"
,
dtype
=
dtype
)
def
test_scatter_gloo_dygraph
(
self
):
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_scatter_api_dygraph.py"
,
"scatter"
,
"gloo"
,
"4"
,
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/collective/test_collective_sendrecv_api.py
浏览文件 @
7a92e74b
...
...
@@ -33,11 +33,16 @@ class TestCollectiveSendRecvAPI(TestDistBase):
# "nccl")
def
test_sendrecv_nccl_dygraph
(
self
):
if
paddle
.
fluid
.
core
.
is_compiled_with_cuda
():
dtypes_to_test
=
[
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
,
'int8'
,
'uint8'
,
'bool'
]
for
dtype
in
dtypes_to_test
:
self
.
check_with_place
(
"collective_sendrecv_api_dygraph.py"
,
"sendrecv"
,
"nccl"
,
static_mode
=
'0'
)
static_mode
=
"0"
,
dtype
=
dtype
)
if
__name__
==
'__main__'
:
...
...
python/paddle/fluid/tests/unittests/collective/testslist.csv
浏览文件 @
7a92e74b
...
...
@@ -8,23 +8,26 @@ test_collective_split_embedding,linux,rocm;gpu,300,DIST,../dist_test.sh,2,,PYTHO
test_collective_allgather_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_allgather_object_api,linux,gpu;rocm,120,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_allreduce_api,linux,gpu;rocm,120,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_alltoall_api,linux,gpu;rocm,
12
0,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_alltoall_api,linux,gpu;rocm,
30
0,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_alltoall_single,linux,gpu;rocm,350,DIST,../dist_test.sh,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_alltoall_single_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_barrier_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_batch_isend_irecv,linux,gpu;rocm,350,DIST,../dist_test.sh,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_broadcast_api,linux,gpu;rocm,
12
0,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_broadcast_api,linux,gpu;rocm,
30
0,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_cpu_barrier_with_gloo,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_global_gather,linux,gpu;rocm,200,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_global_scatter,linux,gpu;rocm,200,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_isend_irecv_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_optimizer,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_process_group,linux,gpu;rocm,350,DIST,../dist_test.sh,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_reduce,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_reduce_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_reduce_scatter,linux,gpu;rocm,350,DIST,../dist_test.sh,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_reduce_scatter_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_scatter,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_scatter_api,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_sendrecv,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_sendrecv_api,linux,gpu;rocm,
12
0,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_sendrecv_api,linux,gpu;rocm,
30
0,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_split_col_linear,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_split_embedding_none_divisible,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
test_collective_split_row_linear,linux,gpu;rocm,300,DIST,test_runner.py,2,,http_proxy=;https_proxy=;PYTHONPATH=..,
...
...
python/paddle/fluid/tests/unittests/test_collective_api_base.py
浏览文件 @
7a92e74b
...
...
@@ -335,6 +335,12 @@ class TestDistBase(unittest.TestCase):
need_result2
=
need_result
[
need_result
.
shape
[
0
]
//
2
:]
np
.
testing
.
assert_allclose
(
tr0_out
[
0
],
need_result1
,
rtol
=
1e-05
)
np
.
testing
.
assert_allclose
(
tr1_out
[
0
],
need_result2
,
rtol
=
1e-05
)
elif
col_type
==
"reduce_scatter"
:
need_result
=
input1
+
input2
need_result1
=
need_result
[
0
:
need_result
.
shape
[
0
]
//
2
]
need_result2
=
need_result
[
need_result
.
shape
[
0
]
//
2
:]
np
.
testing
.
assert_allclose
(
tr0_out
[
0
],
need_result1
,
rtol
=
1e-05
)
np
.
testing
.
assert_allclose
(
tr1_out
[
0
],
need_result2
,
rtol
=
1e-05
)
elif
col_type
==
"allreduce"
:
need_result
=
input1
+
input2
np
.
testing
.
assert_allclose
(
tr0_out
[
0
],
...
...
python/paddle/tensor/manipulation.py
浏览文件 @
7a92e74b
...
...
@@ -1015,7 +1015,7 @@ def concat(x, axis=0, name=None):
Args:
x (list|tuple): ``x`` is a Tensor list or Tensor tuple which is with data type bool, float16,
float32, float64, int32, int64, uint8. All the Tensors in ``x`` must have same data type.
float32, float64, int32, int64,
int8,
uint8. All the Tensors in ``x`` must have same data type.
axis (int|Tensor, optional): Specify the axis to operate on the input Tensors.
It's a scalar with data type int or a Tensor with shape [1] and data type int32
or int64. The effective range is [-R, R), where R is Rank(x). When ``axis < 0``,
...
...
@@ -1073,10 +1073,10 @@ def concat(x, axis=0, name=None):
check_type
(
input
,
'input'
,
(
list
,
tuple
,
Variable
),
'concat'
)
if
not
isinstance
(
input
,
Variable
):
for
id
,
x
in
enumerate
(
input
):
check_variable_and_dtype
(
x
,
'input['
+
str
(
id
)
+
']
'
,
[
'bool'
,
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64'
],
'concat'
)
check_variable_and_dtype
(
x
,
'input['
+
str
(
id
)
+
']'
,
[
'bool'
,
'float16'
,
'float32'
,
'float64'
,
'int32'
,
'int64
'
,
'int8'
,
'unit8'
],
'concat'
)
if
x
.
dtype
!=
input
[
0
].
dtype
:
raise
TypeError
(
"All the Tensors in the input must have the same data type."
...
...
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