未验证 提交 bb11cbc2 编写于 作者: W wangchaochaohu 提交者: GitHub

[API2.0] add Device api (set_device and get_device)(#26103)

上级 30e1083e
......@@ -1320,6 +1320,8 @@ All parameter, weight, gradient are variables in Paddle.
#endif
})
#ifdef PADDLE_WITH_CUDA
.def("get_device_id",
[](const platform::CUDAPlace &self) { return self.GetDeviceId(); })
.def("_type", &PlaceIndex<platform::CUDAPlace>)
.def("_equals", &IsSamePlace<platform::CUDAPlace, platform::Place>)
.def("_equals", &IsSamePlace<platform::CUDAPlace, platform::CUDAPlace>)
......
......@@ -42,6 +42,7 @@ import paddle.nn
import paddle.distributed.fleet
import paddle.optimizer
import paddle.metric
import paddle.device
import paddle.incubate.complex as complex
# TODO: define alias in tensor and framework directory
......@@ -231,6 +232,9 @@ from .tensor.stat import reduce_mean #DEFINE_ALIAS
from .tensor.stat import std #DEFINE_ALIAS
from .tensor.stat import var #DEFINE_ALIAS
from .fluid.data import data
from .device import set_device
from .device import get_device
# from .tensor.tensor import Tensor #DEFINE_ALIAS
# from .tensor.tensor import LoDTensor #DEFINE_ALIAS
# from .tensor.tensor import LoDTensorArray #DEFINE_ALIAS
......
......@@ -12,11 +12,83 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# TODO: define the functions to manipulate devices
# __all__ = ['cpu_places',
# 'CPUPlace',
# 'cuda_pinned_places',
# 'cuda_places',
# 'CUDAPinnedPlace',
# 'CUDAPlace',
# 'is_compiled_with_cuda']
from paddle.fluid import core
from paddle.fluid import framework
import re
__all__ = [
'set_device',
'get_device'
# 'cpu_places',
# 'CPUPlace',
# 'cuda_pinned_places',
# 'cuda_places',
# 'CUDAPinnedPlace',
# 'CUDAPlace',
# 'is_compiled_with_cuda'
]
def set_device(device):
"""
Paddle supports running calculations on various types of devices, including CPU and GPU.
They are represented by string identifiers. This function can specify the global device
which the OP will run.
Parameters:
device(str): This parameter determines the specific running device.
It can be ``cpu`` or ``gpu:0``. When ``device`` is ``cpu``, the
program is running on the cpu. When ``device`` is ``gpu``, the
program is running ont the gpu.
Examples:
.. code-block:: python
import paddle
paddle.enable_imperative()
paddle.fluid.dygraph.set_device("gpu:0")
x1 = paddle.ones(name='x1', shape=[1, 2], dtype='int32')
x2 = paddle.zeros(name='x2', shape=[1, 2], dtype='int32')
data = paddle.stack([x1,x2], axis=1)
"""
lower_device = device.lower()
if lower_device == 'cpu':
place = core.CPUPlace()
framework._set_expected_place(place)
else:
avaliable_device = ((lower_device == 'cpu') or
re.match(r'gpu:\d+', lower_device))
if not avaliable_device:
raise ValueError(
"The device must be a string which is like 'cpu' or 'gpu:0'")
device_info_list = device.split(':', 1)
device_id = device_info_list[1]
device_id = int(device_id)
place = core.CUDAPlace(device_id)
framework._set_expected_place(place)
def get_device():
"""
This funciton can get the current global device of the program is running.
It's a string which is like 'cpu' and 'gpu:0'. if the global device is not
set, it will return a string which is 'gpu:0' when cuda is avaliable or it
will return a string which is 'cpu' when cuda is not avaliable.
Examples:
.. code-block:: python
import paddle
paddle.enable_imperative()
device = paddle.fluid.dygraph.get_device()
"""
device = ''
place = framework._current_expected_place()
if isinstance(place, core.CPUPlace):
device = 'cpu'
elif isinstance(place, core.CUDAPlace):
device_id = place.get_device_id()
device = 'gpu:' + str(device_id)
return device
......@@ -26,13 +26,8 @@ import objgraph
from ..data_feeder import convert_dtype
__all__ = [
'no_grad',
'grad',
'guard',
'enable_dygraph',
'disable_dygraph',
'enabled',
'to_variable',
'no_grad', 'grad', 'guard', 'enable_dygraph', 'disable_dygraph', 'enabled',
'to_variable'
]
......@@ -285,12 +280,11 @@ def guard(place=None):
tracer = Tracer()
VarBase = core.VarBase
if place is None:
if core.is_compiled_with_cuda():
place = core.CUDAPlace(0)
else:
place = core.CPUPlace()
tracer._expected_place = place
if place is not None:
expected_place = place
else:
expected_place = framework._current_expected_place()
tracer._expected_place = expected_place
with framework.program_guard(train, startup):
with framework.unique_name.guard():
......
......@@ -31,6 +31,7 @@ from .. import compat as cpt
from .trainer_factory import TrainerFactory
from .trainer_factory import FetchHandlerMonitor
import copy
from . import framework
from .incubate.checkpoint import auto_checkpoint as acp
__all__ = ['Executor', 'global_scope', 'scope_guard']
......@@ -544,10 +545,8 @@ class Executor(object):
def __init__(self, place=None):
if place is None:
if core.is_compiled_with_cuda():
self.place = core.CUDAPlace(0)
else:
self.place = core.CPUPlace()
expected_place = framework._current_expected_place()
self.place = expected_place
else:
self.place = place
self.program_caches = dict()
......
......@@ -64,7 +64,7 @@ ZERO_VAR_SUFFIX = core.kZeroVarSuffix()
CONTROL_DEP_VAR_PREFIX = core.kControlDepVarName()
_dygraph_tracer_ = None
_dygraph_current_expected_place_ = None
_global_expected_place_ = None
_current_device = None
global_prog_seed = 0
......@@ -247,7 +247,26 @@ def _dygraph_tracer():
def _current_expected_place():
return _dygraph_current_expected_place_
global _global_expected_place_
if _global_expected_place_ is None:
if core.is_compiled_with_cuda():
_global_expected_place_ = core.CUDAPlace(0)
else:
_global_expected_place_ = core.CPUPlace()
return _global_expected_place_
def _set_dygraph_tracer_expected_place(place):
global _dygraph_tracer_
if _dygraph_tracer_ is not None:
_dygraph_tracer_._expected_place = place
def _set_expected_place(place):
global _global_expected_place_
_global_expected_place_ = place
_set_dygraph_tracer_expected_place(place)
# TODO(zhiqiu): remove this function.
......@@ -5417,14 +5436,14 @@ def _dygraph_guard(tracer):
@signature_safe_contextmanager
def _dygraph_place_guard(place):
global _dygraph_current_expected_place_
tmp_place = _dygraph_current_expected_place_
_dygraph_current_expected_place_ = place
global _global_expected_place_
tmp_place = _global_expected_place_
_global_expected_place_ = place
try:
yield
finally:
_dygraph_current_expected_place_ = tmp_place
_global_expected_place_ = tmp_place
def load_op_library(lib_filename):
......
# Copyright (c) 2020 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 unittest
from op_test import OpTest
import numpy as np
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.framework as framework
import warnings
import paddle
class TestStaticDeviceManage(unittest.TestCase):
def test_cpu_device(self):
paddle.set_device('cpu')
out1 = paddle.zeros(shape=[1, 3], dtype='float32')
out2 = paddle.ones(shape=[1, 3], dtype='float32')
out3 = paddle.concat(x=[out1, out2], axis=0)
exe = paddle.fluid.Executor()
exe.run(paddle.fluid.default_startup_program())
res = exe.run(fetch_list=[out3])
device = paddle.get_device()
self.assertEqual(isinstance(exe.place, core.CPUPlace), True)
self.assertEqual(device, "cpu")
def test_gpu_device(self):
if core.is_compiled_with_cuda():
out1 = paddle.zeros(shape=[1, 3], dtype='float32')
out2 = paddle.ones(shape=[1, 3], dtype='float32')
out3 = paddle.concat(x=[out1, out2], axis=0)
paddle.set_device('gpu:0')
exe = paddle.fluid.Executor()
exe.run(paddle.fluid.default_startup_program())
res = exe.run(fetch_list=[out3])
device = paddle.get_device()
self.assertEqual(isinstance(exe.place, core.CUDAPlace), True)
self.assertEqual(device, "gpu:0")
class TestImperativeDeviceManage(unittest.TestCase):
def test_cpu(self):
with fluid.dygraph.guard():
paddle.set_device('cpu')
out1 = paddle.zeros(shape=[1, 3], dtype='float32')
out2 = paddle.ones(shape=[1, 3], dtype='float32')
out3 = paddle.concat(x=[out1, out2], axis=0)
device = paddle.get_device()
self.assertEqual(
isinstance(framework._current_expected_place(), core.CPUPlace),
True)
self.assertEqual(device, "cpu")
def test_gpu(self):
if core.is_compiled_with_cuda():
with fluid.dygraph.guard():
paddle.set_device('gpu:0')
out1 = paddle.zeros(shape=[1, 3], dtype='float32')
out2 = paddle.ones(shape=[1, 3], dtype='float32')
out3 = paddle.concat(x=[out1, out2], axis=0)
device = paddle.get_device()
self.assertEqual(
isinstance(framework._current_expected_place(),
core.CUDAPlace), True)
self.assertEqual(device, "gpu:0")
if __name__ == '__main__':
unittest.main()
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册