framework.py 83.8 KB
Newer Older
1
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# 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
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13 14
# 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.

15 16
from __future__ import print_function

Y
Yu Yang 已提交
17
import collections
X
Xin Pan 已提交
18
from collections import defaultdict
Q
qiaolongfei 已提交
19
import contextlib
P
peizhilin 已提交
20
import os
F
fengjiayi 已提交
21
import re
22
import traceback
23
import six
24

Y
Yu Yang 已提交
25
import numpy as np
26
import subprocess
Q
qiaolongfei 已提交
27

M
minqiyang 已提交
28
from .. import compat as cpt
29
from .proto import framework_pb2
30
try:
P
peizhilin 已提交
31
    if os.name == 'nt':
P
peizhilin 已提交
32
        import sys
P
peizhilin 已提交
33 34 35 36 37
        third_lib_path = os.path.abspath(os.path.dirname(
            __file__)) + os.sep + '..' + os.sep + 'libs'
        os.environ['path'] += ';' + third_lib_path
        sys.path.append(third_lib_path)

38
    from . import core
39
except ImportError as e:
P
peizhilin 已提交
40
    if os.name == 'nt':
41
        executable_path = os.path.abspath(os.path.dirname(sys.executable))
P
peizhilin 已提交
42
        raise ImportError(
43 44 45 46 47
            """NOTE: You may need to run \"set PATH=%s;%%PATH%%\"
        if you encounters \"DLL load failed\" errors. If you have python
        installed in other directory, replace \"%s\" with your own
        directory. The original error is: \n %s""" %
            (executable_path, executable_path, cpt.get_exception_message(e)))
P
peizhilin 已提交
48 49 50 51 52 53
    else:
        raise ImportError(
            """NOTE: You may need to run \"export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH\"
        if you encounters \"libmkldnn.so not found\" errors. If you have python
        installed in other directory, replace \"/usr/local/lib\" with your own
        directory. The original error is: \n""" + cpt.get_exception_message(e))
54
except Exception as e:
55
    raise e
56
from . import unique_name
Y
Yu Yang 已提交
57

58
__all__ = [
59 60 61 62
    'Program',
    'default_startup_program',
    'default_main_program',
    'program_guard',
63
    'name_scope',
64
]
Y
Yu Yang 已提交
65

Q
qiaolongfei 已提交
66 67 68 69
EMPTY_VAR_NAME = core.kEmptyVarName()
TEMP_VAR_NAME = core.kTempVarName()
GRAD_VAR_SUFFIX = core.kGradVarSuffix()
ZERO_VAR_SUFFIX = core.kZeroVarSuffix()
W
Wu Yi 已提交
70 71
CONTROL_DEP_VAR_PREFIX = core.kControlDepVarName()

72
_imperative_tracer_ = None
M
minqiyang 已提交
73
_imperative_current_expected_place_ = None
74 75 76 77 78 79 80 81 82


def _in_imperative_mode():
    return _imperative_tracer_ is not None


def _imperative_tracer():
    return _imperative_tracer_

W
Wu Yi 已提交
83

M
minqiyang 已提交
84
def _current_expected_place():
M
minqiyang 已提交
85
    return _imperative_current_expected_place_
M
minqiyang 已提交
86 87


88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126
class NameScope(object):
    def __init__(self, name="", parent=None):
        self._children = dict()
        self._name = name
        self._parent = parent

    def child(self, prefix):
        if prefix not in self._children:
            new_child = NameScope(prefix, self)
            self._children[prefix] = [new_child]
        else:
            new_child = NameScope(prefix + "_%d" % len(self._children[prefix]),
                                  self)
            self._children[prefix].append(new_child)
        return new_child

    def parent(self):
        return self._parent

    def name(self):
        return self._name


_name_scope = NameScope()


@contextlib.contextmanager
def name_scope(prefix=None):
    """
    Generate hierarchical name prefix for the operators.

    Note: This should only used for debugging and visualization purpose.
    Don't use it for serious analysis such as graph/program transformations.

    Args:
        prefix(str): prefix.

    Examples:
        .. code-block:: python
T
Tink_Y 已提交
127

128 129 130 131
          with name_scope("encoder"):
             ...
          with name_scope("decoder"):
             ...
T
Tink_Y 已提交
132 133
          with name_scope("attention"):
             ...
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152
    """
    # TODO(panyx0718): Only [0-9a-z].
    assert prefix, "namescope prefix cannot be empty."
    global _name_scope
    _name_scope = _name_scope.child(prefix)
    yield
    _name_scope = _name_scope.parent()


def _full_name_scope():
    global _name_scope
    scope = _name_scope
    name = ""
    while scope:
        name = scope.name() + "/" + name
        scope = scope.parent()
    return name


W
Wu Yi 已提交
153 154 155
def generate_control_dev_var_name():
    import random
    return CONTROL_DEP_VAR_PREFIX + "@" + str(random.random())
Q
qiaolongfei 已提交
156 157 158 159


def grad_var_name(var_name):
    """
160 161
    Returns:
        str: gradient name for a certain var name
Q
qiaolongfei 已提交
162 163 164
    """
    return var_name + GRAD_VAR_SUFFIX

Y
Yu Yang 已提交
165

166
def convert_np_dtype_to_dtype_(np_dtype):
167 168
    """
    Convert the data type in numpy to the data type in Paddle
169

170
    Args:
171
        np_dtype(np.dtype): the data type in numpy.
172

173 174
    Returns:
        core.VarDesc.VarType: the data type in Paddle.
175 176

    """
177 178
    dtype = np.dtype(np_dtype)
    if dtype == np.float32:
179
        return core.VarDesc.VarType.FP32
180
    elif dtype == np.float64:
181
        return core.VarDesc.VarType.FP64
182
    elif dtype == np.float16:
183
        return core.VarDesc.VarType.FP16
184
    elif dtype == np.int32:
185
        return core.VarDesc.VarType.INT32
186
    elif dtype == np.int16:
187
        return core.VarDesc.VarType.INT16
188
    elif dtype == np.int64:
189
        return core.VarDesc.VarType.INT64
190
    elif dtype == np.bool:
191
        return core.VarDesc.VarType.BOOL
192 193
    elif dtype == np.uint16:
        return core.VarDesc.VarType.INT16
194 195
    elif dtype == np.uint8:
        return core.VarDesc.VarType.UINT8
Q
qingqing01 已提交
196 197
    elif dtype == np.int8:
        return core.VarDesc.VarType.INT8
198
    else:
M
minqiyang 已提交
199
        raise ValueError("Not supported numpy dtype %s" % dtype)
200 201 202


def dtype_is_floating(dtype):
203 204 205
    """
    Check the data type is floating or not.
    Args:
206
        dtype(np.dtype|core.VarDesc.VarType): data type.
207 208 209 210 211
            Could be numpy format or Paddle format

    Returns(bool): True if data type is a float value

    """
212
    if not isinstance(dtype, core.VarDesc.VarType):
213 214
        dtype = convert_np_dtype_to_dtype_(dtype)

215 216 217 218
    return dtype in [
        core.VarDesc.VarType.FP16, core.VarDesc.VarType.FP32,
        core.VarDesc.VarType.FP64
    ]
219 220


Y
Yang Yang(Tony) 已提交
221
def _debug_string_(proto, throw_on_error=True):
222 223 224 225 226 227 228 229 230 231 232
    """
    Get the debug string of a protobuf message. The message could be not
    initialized.
    Args:
        proto(google.protobuf.message.Message): The protobuf message
        throw_on_error(bool): True if raise an error when the protobuf message
            is not initialized.

    Returns(str): The debug string of the protobuf message

    """
Y
Yu Yang 已提交
233
    error_fields = list()
Y
Yang Yang(Tony) 已提交
234
    if not proto.IsInitialized(error_fields) and throw_on_error:
C
caoying03 已提交
235 236
        raise ValueError("{0} are not initialized.\nThe message is {1}:\n".
                         format(error_fields, proto))
Y
Yu Yang 已提交
237 238 239
    return proto.__str__()


X
Xin Pan 已提交
240
class Variable(object):
241
    """
242 243 244
    In Fluid, every input and output of an operator is a variable. In most
    cases, variables are used for holding different kinds of data or training
    labels. A variable belongs to a block. All variable has its own name and
245
    two variables in different blocks could have the same name.
246

247 248
    There are many kinds of variables. Each kind of them has its own attributes
    and usages. Please reference the framework.proto for details.
249

250
    Most of a Variable's member variables can be setted to be None. It mean
251
    it is not available or will be specified later.
252 253

    Args:
254
        block(Block): The block that the variable belongs to.
255 256
        type(core.VarDesc.VarType): Variable type. Please reference the
            framework.proto for details.
257 258
        name(str|None): The name of the variable. If setted None, it will be
            generated automatically. Default: None
259
        shape(tuple|list|None): The shape of the variable. -1 means the batch size.
260
            Some kinds of variable do not contain shape, just set it to None.
261 262 263
            Default: None
        dtype(np.dtype|core.VarDesc.VarType|str|None): The data type of variable.
            Default: None
264
        lod_level (int|None): The level of lod tensor. 0 means it is not a time
265
            series data.
266
            Default: None
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
        capacity (int|None): The capacity of Channel variable. Ignored for other
            types. Default: None
        persistable (bool|None): True if the variable is persistable. A persistable
            variable will not be deleted after an iteration ending. Defaults: None.
        error_clip (BaseErrorClipAttr|None): The error clip attributes of the
            corresponding gradient variable. Default: None
        stop_gradient (bool): True if the variable will stop to calculate its
            gradients when backward. Default: False.
        is_data (bool): True if the variable is an input data. Default: False

    Notes:
        The constructor of Variable should not be invoked directly. Please
        use `Block.create_var` to create a variable.

    Examples:
        .. code-block:: python

            cur_program = Program()
            cur_block = cur_program.current_block()
            new_variable = cur_block.create_var(name="X",
                                                shape=[-1, 23, 48],
                                                dtype='float32')
289 290
    """

Y
Yu Yang 已提交
291 292
    def __init__(self,
                 block,
Y
Yu Yang 已提交
293
                 type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
294 295 296 297
                 name=None,
                 shape=None,
                 dtype=None,
                 lod_level=None,
298
                 capacity=None,
Q
QI JUN 已提交
299
                 persistable=None,
F
fengjiayi 已提交
300
                 error_clip=None,
Y
Yu Yang 已提交
301
                 stop_gradient=False,
F
fengjiayi 已提交
302
                 is_data=False,
Y
Yu Yang 已提交
303
                 **kwargs):
Y
Yu Yang 已提交
304
        self.block = block
F
fengjiayi 已提交
305
        self.error_clip = error_clip
Y
Yu Yang 已提交
306 307

        if name is None:
Y
Yu Yang 已提交
308
            name = unique_name.generate('_generated_var')
D
Dong Zhihong 已提交
309
        is_new_var = False
M
minqiyang 已提交
310
        name = cpt.to_text(name)
M
minqiyang 已提交
311
        self.desc = self.block.desc.find_var(cpt.to_bytes(name))
D
Dong Zhihong 已提交
312 313

        if self.desc is None:
M
minqiyang 已提交
314
            self.desc = self.block.desc.var(cpt.to_bytes(name))
Y
Yu Yang 已提交
315
            is_new_var = True
Y
Yu Yang 已提交
316

Y
Yu Yang 已提交
317 318 319 320 321 322 323 324
        if is_new_var:
            self.desc.set_type(type)
        elif self.desc.type() != type:
            raise ValueError("Variable {0} has been created before. The "
                             "previous type is {1}; the new type is {2}. They"
                             " are not matched".format(self.name,
                                                       self.desc.type(), type))

Y
Yu Yang 已提交
325
        if shape is not None:
Y
Yu Yang 已提交
326
            if is_new_var:
327
                self.desc.set_shape(shape)
Y
Yu Yang 已提交
328 329 330 331 332 333 334 335
            else:
                old_shape = self.shape
                shape = tuple(shape)
                if shape != old_shape:
                    raise ValueError(
                        "Variable {0} has been created before. the previous "
                        "shape is {1}; the new shape is {2}. They are not "
                        "matched.".format(self.name, old_shape, shape))
Y
Yu Yang 已提交
336
        if dtype is not None:
337
            if not isinstance(dtype, core.VarDesc.VarType):
338
                dtype = convert_np_dtype_to_dtype_(dtype)
Y
Yu Yang 已提交
339
            if is_new_var:
F
fengjiayi 已提交
340
                self.desc.set_dtype(dtype)
Y
Yu Yang 已提交
341
            else:
F
fengjiayi 已提交
342
                old_dtype = self.dtype
Q
QI JUN 已提交
343
                if dtype != old_dtype:
Y
Yu Yang 已提交
344 345 346 347 348
                    raise ValueError("Variable {0} has been created before. "
                                     "The previous data type is {1}; the new "
                                     "data type is {2}. They are not "
                                     "matched.".format(self.name, old_dtype,
                                                       dtype))
Y
Yu Yang 已提交
349 350

        if lod_level is not None:
Y
Yu Yang 已提交
351
            if is_new_var:
352
                self.desc.set_lod_level(lod_level)
Y
Yu Yang 已提交
353 354 355 356 357 358 359
            else:
                if lod_level != self.lod_level:
                    raise ValueError("Variable {0} has been created before. "
                                     "The previous lod_level is {1}; the new "
                                     "lod_level is {2}. They are not "
                                     "matched".format(self.name, self.lod_level,
                                                      lod_level))
360 361 362 363 364 365 366 367 368 369 370
        if persistable is not None:
            if is_new_var:
                self.desc.set_persistable(persistable)
            else:
                if persistable != self.persistable:
                    raise ValueError(
                        "Variable {0} has been created before."
                        "The previous persistable is {1}; the new "
                        "persistable is {2}. They are not matched".format(
                            self.name, self.persistable, persistable))

371 372 373 374 375 376 377 378
        if capacity is not None:
            if is_new_var:
                self.desc.set_capacity(capacity)
            else:
                # TODO(abhinavarora) : Compare with set capacity once,
                # get_capacity is implemented
                pass

Y
Yu Yang 已提交
379
        self.block.vars[name] = self
Y
Yu Yang 已提交
380
        self.op = None
M
minqiyang 已提交
381
        self.stop_gradient = stop_gradient
F
fengjiayi 已提交
382
        self.is_data = is_data
X
Xin Pan 已提交
383
        if _in_imperative_mode():
M
minqiyang 已提交
384 385 386
            self._ivar = kwargs.get("ivar", None)
            if not self._ivar:
                self._ivar = core.VarBase()
X
Xin Pan 已提交
387
            self._ivar.desc = self.desc
388
            self._ivar.stop_gradient = stop_gradient
Y
Yu Yang 已提交
389

390
    def _numpy(self):
M
minqiyang 已提交
391
        new_ivar = self._ivar._copy_to(core.CPUPlace(), True)
P
Paddle CI 已提交
392
        return np.array(new_ivar.value().get_tensor())
393 394

    def _backward(self):
X
Xin Pan 已提交
395
        self._ivar._run_backward()
396 397

    def _gradient(self):
M
minqiyang 已提交
398
        return np.array(self._ivar._grad_value())
399

X
Xin Pan 已提交
400 401
    def _clear_gradient(self):
        self._ivar._clear_gradient()
X
Xin Pan 已提交
402

403
    def __str__(self):
Y
Yang Yang(Tony) 已提交
404 405
        return self.to_string(True)

F
update  
fengjiayi 已提交
406
    def to_string(self, throw_on_error, with_details=False):
407 408 409 410
        """
        Get debug string.

        Args:
411 412
            throw_on_error(bool): True if raise an exception when self is
                not initialized.
F
update  
fengjiayi 已提交
413
            with_details(bool): more details about variables and parameters
414 415
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False;
416

417 418
        Returns:
            str: The debug string.
419
        """
F
update  
fengjiayi 已提交
420 421
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
422
        protostr = self.desc.serialize_to_string()
423
        proto = framework_pb2.VarDesc.FromString(six.binary_type(protostr))
F
update  
fengjiayi 已提交
424 425 426 427
        res_str = _debug_string_(proto, throw_on_error)
        if with_details:
            additional_attr = ("error_clip", "stop_gradient")
            for attr_name in additional_attr:
428 429
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
430
        return res_str
431 432 433

    __repr__ = __str__

W
Wu Yi 已提交
434
    def _set_desc(self, input):
435 436 437 438 439 440 441 442 443
        """
        Set the variable description.

        Args:
            input(core.VarDesc): The new VarDesc.

        Returns:
            None
        """
444 445
        self.desc = input

446 447
    @property
    def _stop_gradient(self):
M
minqiyang 已提交
448 449 450 451
        if _in_imperative_mode():
            return self._ivar.stop_gradient
        else:
            return self.stop_gradient
452 453 454

    @_stop_gradient.setter
    def _stop_gradient(self, s):
M
minqiyang 已提交
455 456 457
        if _in_imperative_mode():
            self._ivar.stop_gradient = s
        self.stop_gradient = s
458

459 460 461 462
    @property
    def persistable(self):
        return self.desc.persistable()

Y
Yu Yang 已提交
463 464 465 466
    @persistable.setter
    def persistable(self, p):
        self.desc.set_persistable(p)

Y
Yu Yang 已提交
467 468
    @property
    def name(self):
M
minqiyang 已提交
469
        return cpt.to_text(self.desc.name())
Y
Yu Yang 已提交
470

T
typhoonzero 已提交
471 472 473 474
    @name.setter
    def name(self, new_name):
        self.desc.set_name(new_name)

Y
Yu Yang 已提交
475 476 477
    @property
    def shape(self):
        # convert to tuple, make it as same as numpy API.
478
        return tuple(self.desc.shape())
Y
Yu Yang 已提交
479 480

    @property
F
fengjiayi 已提交
481 482
    def dtype(self):
        return self.desc.dtype()
Y
Yu Yang 已提交
483 484 485

    @property
    def lod_level(self):
486
        return self.desc.lod_level()
Y
Yu Yang 已提交
487

Y
Yu Yang 已提交
488 489 490 491
    @property
    def type(self):
        return self.desc.type()

W
Wu Yi 已提交
492
    def _set_error_clip(self, error_clip):
493 494 495 496 497 498 499 500 501
        """
        Set the error_clip.

        Args:
            error_clip(BaseErrorClipAttr) : The new error_clip.

        Returns:
            None
        """
502 503
        self.error_clip = error_clip

Y
Yu Yang 已提交
504

F
fengjiayi 已提交
505 506 507
def get_all_op_protos():
    """
    Get all registered op proto from PaddlePaddle C++ end.
508

509 510
    Returns:
       list: list of OpProto.
F
fengjiayi 已提交
511 512 513 514
    """
    protostrs = core.get_all_op_protos()
    ret_values = []
    for pbstr in protostrs:
515
        op_proto = framework_pb2.OpProto.FromString(six.binary_type(pbstr))
F
fengjiayi 已提交
516 517 518 519 520
        ret_values.append(op_proto)
    return ret_values


class OpProtoHolder(object):
521 522 523 524
    """
    A global variable to hold all OpProtos from C++ as a map
    """

F
fengjiayi 已提交
525 526 527 528 529 530 531 532 533
    @classmethod
    def instance(cls):
        if not hasattr(cls, '_instance'):
            cls._instance = cls()
        return cls._instance

    def __init__(self):
        assert not hasattr(
            self.__class__,
534
            '_instance'), 'Please use `instance()` to get OpProtoHolder object!'
F
fengjiayi 已提交
535 536 537 538 539 540
        op_protos = get_all_op_protos()
        self.op_proto_map = {}
        for proto in op_protos:
            self.op_proto_map[proto.type] = proto

    def get_op_proto(self, type):
541 542 543 544 545 546 547 548
        """
        Get OpProto by a type string.
        Args:
            type(str): The type that operator registered in C++ side.

        Returns(framework_pb2.OpProto): The OpProto

        """
Y
Yu Yang 已提交
549 550
        if type not in self.op_proto_map:
            raise ValueError("Operator \"%s\" has not been registered." % type)
F
fengjiayi 已提交
551 552
        return self.op_proto_map[type]

553 554 555 556
    @staticmethod
    def generated_op_attr_names():
        return {
            core.op_proto_and_checker_maker.kOpRoleAttrName(),
S
sneaxiy 已提交
557
            core.op_proto_and_checker_maker.kOpRoleVarAttrName(),
558
            core.op_proto_and_checker_maker.kOpNameScopeAttrName()
559 560
        }

F
fengjiayi 已提交
561

X
Xin Pan 已提交
562
class Operator(object):
563
    """
564 565 566 567 568 569 570
    In Fluid, all the operation are represented by Operator, and Operator
    is regarded as a build in an instruction of a Block. Users can use the
    build in instructions to describe their neural network.

    Args:
        block(Block): The block has the current operator.
        desc(core.OpDesc): The protobuf description of Operator.
C
chengduoZH 已提交
571
        type(str): The type of operator. Default None.
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591
        inputs(dict): The input of this Operator. it is a dictionary, for every
            element, key is the input parameter name, and value is a list of
            variables. Default None.
        outputs(dict): The output of this Operator. it is a dictionary, for
            every element, key is the input parameter name, and value is a list
            of variables. Default None.
        attrs(dict): The attributes of this Operator. it is a dictionary, for
            every element, key is attribute name, and value is the attribute value.
            The attribute type should be as same as the type registered in C++ side.
            Default None.

    Returns:
        Operator: The initialized Operator.

    Raises:
        ValueError: If the passed input, output and attrs doesn't match the
            initializing Operator's that registered in C++ side.

    Notes:
        The constructor of operator should not be invoked directly. Use
W
Wu Yi 已提交
592
        Block.append_op or Block._prepend_op instead.
593 594 595 596 597 598 599 600 601 602

    Examples:
        .. code-block:: python

            cur_program = Program()
            cur_block = cur_program.current_block()
            # var1 += var2 + var3
            cur_block.append_op(type="sum",
                                inputs={"X": [var1, var2, var3]},
                                outputs={"Out": [var1]})
603
    """
604 605 606
    OP_WITHOUT_KERNEL_SET = {
        'feed', 'fetch', 'save', 'load', 'recurrent', 'go',
        'rnn_memory_helper_grad', 'conditional_block', 'while', 'send', 'recv',
X
Xin Pan 已提交
607 608
        'listen_and_serv', 'save_combine', 'load_combine', 'ncclInit', 'select',
        'checkpoint_notify', 'gen_nccl_id'
609
    }
610

Y
Yu Yang 已提交
611 612
    def __init__(self,
                 block,
Y
Yu Yang 已提交
613
                 desc,
Y
Yu Yang 已提交
614 615 616
                 type=None,
                 inputs=None,
                 outputs=None,
M
minqiyang 已提交
617
                 attrs=None):
Y
Yu Yang 已提交
618
        self.block = block
Y
Yu Yang 已提交
619
        self.desc = desc
G
gongweibao 已提交
620 621 622 623 624
        # note: not add self.attrs here:
        # https://github.com/PaddlePaddle/Paddle/pull/12583#pullrequestreview-145093173
        op_attrs = attrs
        if op_attrs is None:
            op_attrs = dict()
Y
yuyang18 已提交
625 626 627 628
        del attrs

        op_maker = core.op_proto_and_checker_maker

G
gongweibao 已提交
629 630
        if op_maker.kOpRoleAttrName() not in op_attrs:
            op_attrs[op_maker.kOpRoleAttrName()] = self.block.program.op_role
Y
yuyang18 已提交
631 632 633

        role_var_name = op_maker.kOpRoleVarAttrName()
        if len(self.block.program.
G
gongweibao 已提交
634 635
               op_role_var) != 0 and role_var_name not in op_attrs:
            op_attrs[role_var_name] = self.block.program.op_role_var
Y
yuyang18 已提交
636

G
gongweibao 已提交
637 638
        if role_var_name in op_attrs and len(op_attrs[role_var_name]) == 0:
            del op_attrs[role_var_name]
Y
yuyang18 已提交
639

F
fengjiayi 已提交
640 641 642 643 644
        if len(self.desc.type()) != 0:
            return
        if type is None:
            raise ValueError(
                "`type` to initilized an Operator can not be None.")
P
peizhilin 已提交
645 646 647 648 649
        else:
            callstack_var_name = op_maker.kOpCreationCallstackAttrName()
            op_attrs[callstack_var_name] = list(
                reversed(traceback.format_stack()))[1:]

F
Update  
fengjiayi 已提交
650
        self.desc.set_type(type)
F
fengjiayi 已提交
651
        proto = OpProtoHolder.instance().get_op_proto(type)
652

653 654 655
        namescope_var_name = op_maker.kOpNameScopeAttrName()
        op_attrs[namescope_var_name] = _full_name_scope()

Y
Yang Yang(Tony) 已提交
656 657
        def find_name(var_list, name):
            for var_name in var_list:
Q
Qiao Longfei 已提交
658
                if var_list[var_name] is not None and var_name == name:
Y
Yang Yang(Tony) 已提交
659 660
                    return True
            return False
Q
QI JUN 已提交
661

Y
Yang Yang(Tony) 已提交
662 663 664 665 666 667 668
        if inputs is not None:
            for in_proto in proto.inputs:
                found = find_name(inputs, in_proto.name)
                assert found or in_proto.dispensable, "Input {} not found".format(
                    in_proto.name)

                if found:
669 670 671 672
                    in_args = inputs[in_proto.name]
                    if not isinstance(in_args, list):
                        in_args = [in_args]
                    if not in_proto.duplicable and len(in_args) > 1:
Y
Yang Yang(Tony) 已提交
673 674
                        raise ValueError(
                            "Input %s expects only one input, but %d are given."
675 676 677
                            % (in_proto.name, len(in_args)))
                    in_arg_names = []
                    for arg in in_args:
678
                        if isinstance(arg, six.string_types):
Y
Yang Yu 已提交
679
                            in_arg_names.append(arg)
680 681
                        elif isinstance(arg, six.binary_type):
                            in_arg_names.append(arg.decode())
Y
Yang Yu 已提交
682
                        else:
M
minqiyang 已提交
683
                            in_arg_names.append(cpt.to_text(arg.name))
684
                    self.desc.set_input(in_proto.name, in_arg_names)
Y
Yang Yang(Tony) 已提交
685 686
                else:
                    self.desc.set_input(in_proto.name, [])
F
Update  
fengjiayi 已提交
687

Y
Yu Yang 已提交
688
        if outputs is not None:
689
            for m in proto.outputs:
Q
qingqing01 已提交
690 691 692 693 694 695
                if (m.name not in outputs) and m.dispensable:
                    continue
                if not ((m.name in outputs) or m.dispensable):
                    raise ValueError(
                        ("Incorrect setting for output(s) of "
                         "operator \"%s\", should set: [%s].") % (type, m.name))
F
fengjiayi 已提交
696
            for out_proto in proto.outputs:
Q
qingqing01 已提交
697 698
                if out_proto.name not in outputs:
                    continue
699 700 701 702
                out_args = outputs[out_proto.name]
                if not isinstance(out_args, list):
                    out_args = [out_args]
                if not out_proto.duplicable and len(out_args) > 1:
F
Update  
fengjiayi 已提交
703 704
                    raise ValueError(
                        "Output %s expects only one output, but %d are given." %
705 706 707
                        (out_proto.name, len(out_args)))
                out_arg_names = []
                for arg in out_args:
M
minqiyang 已提交
708
                    out_arg_names.append(cpt.to_text(arg.name))
709 710
                    arg.op = self
                self.desc.set_output(out_proto.name, out_arg_names)
F
Update  
fengjiayi 已提交
711

G
gongweibao 已提交
712 713
        if op_attrs is not None:
            if not isinstance(op_attrs, dict):
714
                raise TypeError("'attrs' should be a dict.")
F
fengjiayi 已提交
715
            for attr in proto.attrs:
F
Update  
fengjiayi 已提交
716
                attr_name = attr.name
G
gongweibao 已提交
717
                if (attr_name not in op_attrs) or (op_attrs[attr_name] is None):
F
Update  
fengjiayi 已提交
718
                    continue
G
gongweibao 已提交
719
                attr_val = op_attrs[attr_name]
G
gongweibao 已提交
720 721
                self._update_desc_attr(attr_name, attr_val)

722
        self.desc.check_attrs()
M
minqiyang 已提交
723

W
Wu Yi 已提交
724
        if self._has_kernel(type):
Q
QI JUN 已提交
725
            self.desc.infer_var_type(self.block.desc)
Y
Yu Yang 已提交
726
            self.desc.infer_shape(self.block.desc)
M
minqiyang 已提交
727

X
Xin Pan 已提交
728 729 730
        if _in_imperative_mode():
            self.iop = core.OpBase()
            self.iop.desc = self.desc
X
Xin Pan 已提交
731
            self.inputs = defaultdict(list)
X
Xin Pan 已提交
732
            if inputs is not None:
X
Xin Pan 已提交
733 734 735 736 737 738
                for k, v in six.iteritems(inputs):
                    if isinstance(v, Variable):
                        self.inputs[k].append(v._ivar)
                    elif isinstance(v, list) or isinstance(v, tuple):
                        self.inputs[k].extend([var._ivar for var in v])
            self.outputs = defaultdict(list)
X
Xin Pan 已提交
739
            if outputs is not None:
X
Xin Pan 已提交
740 741 742 743 744
                for k, v in six.iteritems(outputs):
                    if isinstance(v, Variable):
                        self.outputs[k].append(v._ivar)
                    elif isinstance(v, list) or isinstance(v, tuple):
                        self.outputs[k].extend([var._ivar for var in v])
F
fengjiayi 已提交
745

W
Wu Yi 已提交
746
    def _has_kernel(self, op_type):
747 748
        return op_type not in self.OP_WITHOUT_KERNEL_SET

Y
Yang Yang(Tony) 已提交
749
    def to_string(self, throw_on_error):
750
        """
751 752
        Get debug string.

753
        Args:
754 755
            throw_on_error(bool): Whether to raise exception if self is not
                initialized.
756

757 758
        Returns:
            str: The debug string.
759 760

        """
761
        protostr = self.desc.serialize_to_string()
762
        proto = framework_pb2.OpDesc.FromString(six.binary_type(protostr))
Y
Yang Yang(Tony) 已提交
763 764 765 766
        return _debug_string_(proto, throw_on_error)

    def __str__(self):
        return self.to_string(True)
767 768 769

    __repr__ = __str__

F
fengjiayi 已提交
770 771 772 773 774
    @property
    def type(self):
        return self.desc.type()

    def input(self, name):
775
        """
776
        Get the input arguments according to the input parameter name.
777

778 779
        Args:
            name(str): The input parameter name.
780

781 782 783
        Returns:
            list: return the list of argument names that associated with \
                the specific parameter name.
784
        """
F
fengjiayi 已提交
785 786
        return self.desc.input(name)

W
Wu Yi 已提交
787
    def _rename_input(self, old_name, new_name):
788 789 790 791 792 793 794 795 796 797
        """
        Rename the `old_name` to `new_name`.

        Args:
            old_name(str): The old name of the Operator's input.
            new_name(str): The new name of the Operator's input.

        Returns:
            None
        """
W
Wu Yi 已提交
798
        self.desc._rename_input(old_name, new_name)
T
typhoonzero 已提交
799

W
Wu Yi 已提交
800
    def _rename_output(self, old_name, new_name):
801 802 803 804 805 806 807 808 809 810
        """
        Rename the `old_name` to `new_name`.

        Args:
            old_name(str): The old name of the Operator's output.
            new_name(str): The new name of the Operator's output.

        Returns:
            None
        """
W
Wu Yi 已提交
811
        self.desc._rename_output(old_name, new_name)
T
typhoonzero 已提交
812

F
fengjiayi 已提交
813 814 815 816
    @property
    def input_names(self):
        return self.desc.input_names()

T
typhoonzero 已提交
817 818 819 820 821 822 823 824
    @property
    def input_arg_names(self):
        return self.desc.input_arg_names()

    @property
    def output_arg_names(self):
        return self.desc.output_arg_names()

F
fengjiayi 已提交
825
    def output(self, name):
826
        """
827
        Get output arguments by the output parameter name.
828

829 830
        Args:
            name(str): The output parameter name.
831

832 833 834
        Returns:
            list: return the list of argument names associated with \
                the specific parameter name.
835
        """
F
fengjiayi 已提交
836 837 838 839 840 841
        return self.desc.output(name)

    @property
    def output_names(self):
        return self.desc.output_names()

842 843 844 845 846 847 848 849
    @property
    def idx(self):
        for i, op in enumerate(self.block.ops):
            if op == self:
                return i
        raise ValueError(
            "Can't find op itself in it's block. It could be a bug of Paddle.")

F
fengjiayi 已提交
850
    def has_attr(self, name):
851
        """
852 853
        Whether this Operator has the attribute with name or not.

854
        Args:
855
            name(str): the attribute name.
856

857 858
        Returns:
            bool: True if has this attribute.
859 860

        """
F
fengjiayi 已提交
861 862 863
        return self.desc.has_attr(name)

    def attr_type(self, name):
864
        """
865
        Get the type of attribute by attribute's name.
866

867 868
        Args:
            name(str): the attribute name.
869

870 871
        Returns:
            core.AttrType: the attribute type.
872
        """
F
fengjiayi 已提交
873 874
        return self.desc.attr_type(name)

W
Wu Yi 已提交
875
    def _set_attr(self, name, val):
876 877 878 879 880 881 882 883 884 885
        """
        Set the value of attribute by attribute's name.

        Args:
            name(str): the attribute name.
            val(bool|int|str|float|list): the value of the attribute.

        Raises:
            ValueError: If the type of value doesn't match with desc.attr_type(name).
        """
G
gongweibao 已提交
886 887 888 889 890 891 892 893 894 895 896 897 898
        self._update_desc_attr(name, val)

    def _update_desc_attr(self, name, val):
        """
        Update the value of desc's attribute by attribute's name.

        Args:
            name(str): the attribute name.
            val(bool|int|str|float|list): the value of the attribute.

        Raises:
            ValueError: If the type of value doesn't match with desc.attr_type(name).
        """
Q
Qiyang Min 已提交
899 900
        if isinstance(val, Block):
            self.desc.set_block_attr(name, val.desc)
Y
Yancey1989 已提交
901 902
        elif isinstance(val, list) and val and all(
                isinstance(v, Block) for v in val):
903
            self.desc.set_blocks_attr(name, [v.desc for v in val])
Q
Qiyang Min 已提交
904 905 906 907
        elif isinstance(val, core.BlockDesc) or \
                isinstance(val, core.ProgramDesc):
            self.desc.set_serialized_attr(name, val.serialize_to_string())
        else:
W
Wu Yi 已提交
908
            self.desc._set_attr(name, val)
Y
yuyang18 已提交
909

F
fengjiayi 已提交
910 911 912 913 914
    @property
    def attr_names(self):
        return self.desc.attr_names()

    def attr(self, name):
915
        """
916 917
        Get the attribute by name.

918
        Args:
919
            name(str): the attribute name.
920

921 922
        Returns:
            bool|int|str|float|list: The attribute value. The return value
923 924
            can be any valid attribute type.
        """
F
fengjiayi 已提交
925
        return self.desc.attr(name)
Y
Yu Yang 已提交
926

W
Wu Yi 已提交
927
    def _block_attr_id(self, name):
928
        """
G
gongweibao 已提交
929
        Get the block attribute's id by name.
930

931 932
        Args:
            name(str): the attribute name.
933

934 935
        Returns:
            int: the block index.
936
        """
W
Wu Yi 已提交
937
        return self.desc._block_attr_id(name)
G
gongweibao 已提交
938

W
Wu Yi 已提交
939
    def _block_attr(self, name):
G
gongweibao 已提交
940 941 942 943 944 945 946 947 948 949
        """
        Get the block attribute  by name.

        Args:
            name(str): the attribute name.

        Returns:
            block: the block attribute.
        """

W
Wu Yi 已提交
950
        id = self._block_attr_id(name)
G
gongweibao 已提交
951 952 953
        assert (id >= 0 and id < len(self.block.program.blocks))
        return self.block.program.blocks[id]

W
Wu Yi 已提交
954
    def _blocks_attr(self, name):
G
gongweibao 已提交
955 956 957 958 959 960 961 962 963 964
        """
        Get the blocks attribute  by name.

        Args:
            name(str): the attribute name.

        Returns:
            list: list of the blocks attribute.
        """
        attrs = []
W
Wu Yi 已提交
965
        for i in self._blocks_attr_ids(name):
G
gongweibao 已提交
966 967 968 969 970
            assert (i >= 0 and i < len(self.block.program.blocks))
            attrs.append(self.block.program.blocks[i])

        return attrs

W
Wu Yi 已提交
971
    def _blocks_attr_ids(self, name):
G
gongweibao 已提交
972 973 974 975 976 977 978 979 980 981
        """
        Get the blocks attribute's ids by name.

        Args:
            name(str): the attribute name.

        Returns:
            list: list of the blocks ids.
        """

W
Wu Yi 已提交
982
        return self.desc._blocks_attr_ids(name)
Y
Yu Yang 已提交
983

J
JiayiFeng 已提交
984
    def all_attrs(self):
F
fengjiayi 已提交
985
        """
986 987 988
        Get the attribute dict.

        Returns:
G
gongweibao 已提交
989
            dict: The Operator's attribute dict, name->attr.
F
fengjiayi 已提交
990 991 992 993
        """
        attr_names = self.attr_names
        attr_map = {}
        for n in attr_names:
G
gongweibao 已提交
994 995
            attr_type = self.desc.attr_type(n)
            if attr_type == core.AttrType.BLOCK:
W
Wu Yi 已提交
996
                attr_map[n] = self._block_attr(n)
G
gongweibao 已提交
997 998 999
                continue

            if attr_type == core.AttrType.BLOCKS:
W
Wu Yi 已提交
1000
                attr_map[n] = self._blocks_attr(n)
G
gongweibao 已提交
1001 1002 1003 1004
                continue

            attr_map[n] = self.attr(n)

F
fengjiayi 已提交
1005 1006
        return attr_map

Y
Yu Yang 已提交
1007

Y
Yu Yang 已提交
1008
class Block(object):
1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022
    """
    In Fluid, a Program is consistence of multi-Block, and Block stores
    VarDesc and OpDesc. In a specific Block, a VarDesc have a unique name.
    One block could have some child blocks, and child block's name scopes
    should inherit the parent's so that OpDesc in child block can reference
    a VarDesc that is stored in the parent block.
    Please reference the framework.proto for details.

    Args:
        program(Program): The Program that the Block belongs to.
        idx(int): The block's id in the Program.

    Notes:
        The constructor of Block should not be invoked directly. Please
W
Wu Yi 已提交
1023
        use `Program._create_block()` to create a block.
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037

    Examples:
        .. code-block:: python

            cur_program = Program()
            cur_block = cur_program.current_block()
            var = cur_block.create_var(name="X",
                                       shape=[-1, 23, 48],
                                       dtype='float32')
            cur_block.append_op(type="abs",
                                inputs={"X": [var]},
                                outputs={"Out": [var]})
    """

Y
Yu Yang 已提交
1038
    def __init__(self, program, idx):
Y
Yu Yang 已提交
1039
        self.desc = program.desc.block(idx)
1040
        self.vars = collections.OrderedDict()  # var_name --> var
Q
qiaolongfei 已提交
1041
        self.ops = list()  # operator list
Y
Yu Yang 已提交
1042
        self.program = program
1043
        self.removed_vars = collections.OrderedDict()
Y
Yu Yang 已提交
1044

1045
    def __str__(self):
Y
Yang Yang(Tony) 已提交
1046 1047
        return self.to_string(True)

F
fengjiayi 已提交
1048 1049
    def to_string(self, throw_on_error, with_details=False):
        """
1050 1051
        Get debug string.

F
fengjiayi 已提交
1052 1053
        Args:
            throw_on_error(bool): raise exception when self is not initialized
1054
                when throw_on_error is True.
F
update  
fengjiayi 已提交
1055
            with_details(bool): more details about variables and parameters
1056 1057
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False.
F
fengjiayi 已提交
1058

1059 1060
        Returns:
            str: The debug string.
F
fengjiayi 已提交
1061 1062 1063 1064
        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
F
fengjiayi 已提交
1065
            re_add_indent = re.compile(r"\n(.)")
F
fengjiayi 已提交
1066 1067
            res_str = "blocks {\n  idx: %d\n  parent_idx: %d" % (
                self.idx, self.parent_idx)
1068
            for var in list(self.vars.values()):
F
fengjiayi 已提交
1069
                res_str += "\n  vars {\n    %s  }" % re_add_indent.sub(
F
update  
fengjiayi 已提交
1070
                    r"\n    \1", var.to_string(throw_on_error, with_details))
F
fengjiayi 已提交
1071
            for op in self.ops:
F
fengjiayi 已提交
1072 1073
                res_str += "\n  ops {\n    %s  }" % re_add_indent.sub(
                    r"\n    \1", op.to_string(throw_on_error))
F
fengjiayi 已提交
1074 1075 1076
            res_str += "\n}"
        else:
            protostr = self.desc.serialize_to_string()
1077 1078
            proto = framework_pb2.BlockDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
1079 1080
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1081 1082 1083

    __repr__ = __str__

Y
Yu Yang 已提交
1084 1085
    @property
    def parent_idx(self):
Y
Yu Yang 已提交
1086
        return self.desc.parent
Y
Yu Yang 已提交
1087

Y
Yu Yang 已提交
1088 1089 1090 1091
    @property
    def forward_block_idx(self):
        return self.desc.get_forward_block_idx()

W
Wu Yi 已提交
1092
    def _set_forward_block_idx(self, idx):
1093 1094 1095 1096 1097 1098 1099 1100 1101
        """
        Set the forward block Idx.

        Args:
            idx(int): the block index.

        Returns:
            None
        """
W
Wu Yi 已提交
1102
        self.desc._set_forward_block_idx(idx)
Y
Yu Yang 已提交
1103

Y
Yu Yang 已提交
1104 1105
    @property
    def idx(self):
Y
Yu Yang 已提交
1106
        return self.desc.id
Y
Yu Yang 已提交
1107

Q
Qiao Longfei 已提交
1108
    def var(self, name):
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
        """
        Get a Variable by name from this block.

        Args:
            name(str): the Variable's name.

        Raises:
            ValueError: The If input's type is not str, or this block
                doesn't have a Variable with the giving name.

        Returns:
            Variable: the Variable with the giving name.
        """
1122
        if not isinstance(name, six.string_types):
M
minqiyang 已提交
1123 1124 1125
            raise TypeError(
                "var require string as parameter, but get %s instead." %
                (type(name)))
Y
Yu Yang 已提交
1126 1127
        v = self.vars.get(name, None)
        if v is None:
Q
Qiao Longfei 已提交
1128
            raise ValueError("var %s not in this block" % name)
Y
Yu Yang 已提交
1129
        return v
Q
Qiao Longfei 已提交
1130

X
Xin Pan 已提交
1131
    def _find_var_recursive(self, name):
1132 1133 1134 1135 1136 1137 1138
        """
        Get a Variable by name from this block recursively.

        Args:
            name(str): the Variable's name.

        Returns:
X
Xin Pan 已提交
1139
            Variable: the Variable with the giving name. Or None if not found.
1140
        """
Y
Yu Yang 已提交
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164
        frontier = list()
        visited = set()

        frontier.append(self)

        prog = self.program

        while len(frontier) != 0:  # BFS
            cur = frontier[0]
            frontier = frontier[1:]

            if id(cur) in visited:
                continue

            if cur.has_var(name):
                return cur.var(name)

            if cur.parent_idx != -1:
                frontier.append(prog.block(cur.parent_idx))

            if cur.forward_block_idx != -1:
                frontier.append(prog.block(cur.forward_block_idx))

            visited.add(id(cur))
X
Xin Pan 已提交
1165
        return None
Y
Yu Yang 已提交
1166

X
Xin Pan 已提交
1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185
    def _var_recursive(self, name):
        """
        Get a Variable by name from this block recursively.

        Args:
            name(str): the Variable's name.

        Raises:
            ValueError: this block and this parent block doesn't
                have a Variable with the giving name.

        Returns:
            Variable: the Variable with the giving name.
        """
        var = self._find_var_recursive(name)
        if var:
            return var
        else:
            raise ValueError("Var {0} is not found recursively".format(name))
F
fengjiayi 已提交
1186

Q
Qiao Longfei 已提交
1187
    def all_parameters(self):
1188
        return list(self.iter_parameters())
1189

1190
    def iter_parameters(self):
M
minqiyang 已提交
1191
        return (item[1] for item in six.iteritems(self.vars)
1192
                if isinstance(item[1], Parameter))
Q
Qiao Longfei 已提交
1193

Y
Yu Yang 已提交
1194
    def create_var(self, *args, **kwargs):
1195
        var = Variable(block=self, *args, **kwargs)
1196 1197
        if 'initializer' in kwargs:
            kwargs['initializer'](var, self)
Q
Qiao Longfei 已提交
1198
        return var
Y
Yu Yang 已提交
1199

Q
Qiao Longfei 已提交
1200 1201 1202
    def has_var(self, name):
        return name in self.vars

W
Wu Yi 已提交
1203
    def _rename_var(self, name, new_name):
T
typhoonzero 已提交
1204 1205
        """
        Rename variable in vars and ops' inputs and outputs
1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217

        Args:
            name(str): the name that need to be renamed.
            new_name(str): the name that need to rename to.

        Raises:
            ValueError: If this block doesn't have this the giving name,
                or the type of the var with the giving name is not Parameter
                or Variable.

        Returns:
            Variable: the Variable with the giving name.
T
typhoonzero 已提交
1218
        """
M
minqiyang 已提交
1219 1220
        name = cpt.to_text(name)
        new_name = cpt.to_text(new_name)
M
minqiyang 已提交
1221

T
typhoonzero 已提交
1222
        if not self.has_var(name):
1223
            raise ValueError("var %s is not in current block" % name)
T
wip  
typhoonzero 已提交
1224 1225
        v = self.var(name)
        if type(v) == Parameter:
T
typhoonzero 已提交
1226
            var_type = "Parameter"
T
wip  
typhoonzero 已提交
1227 1228 1229 1230 1231 1232 1233
            stop_gradient = v.stop_gradient
            trainable = v.trainable
            optimize_attr = v.optimize_attr
            regularizer = v.regularizer
            gradient_clip_attr = v.gradient_clip_attr
            error_clip = v.error_clip
        elif type(v) == Variable:
T
typhoonzero 已提交
1234
            var_type = "Variable"
T
wip  
typhoonzero 已提交
1235 1236 1237 1238
            error_clip = v.error_clip
            stop_gradient = v.stop_gradient
        else:
            raise ValueError("unsupported var type: %s", type(v))
T
typhoonzero 已提交
1239
        orig_var_type = v.type
M
minqiyang 已提交
1240
        self.desc._rename_var(cpt.to_bytes(name), cpt.to_bytes(new_name))
W
Wu Yi 已提交
1241
        # NOTE: v is destroyed by C++ after calling _rename_var.
M
minqiyang 已提交
1242
        d = self.desc.find_var(cpt.to_bytes(new_name))
T
typhoonzero 已提交
1243
        if var_type == "Parameter":
T
wip  
typhoonzero 已提交
1244 1245 1246 1247
            var = Parameter(
                self,
                d.shape(),
                d.dtype(),
T
typhoonzero 已提交
1248
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1249 1250 1251 1252 1253 1254 1255
                name=new_name,
                stop_gradient=stop_gradient,
                trainable=trainable,
                optimize_attr=optimize_attr,
                regularizer=regularizer,
                gradient_clip_attr=gradient_clip_attr,
                error_clip=error_clip)
T
typhoonzero 已提交
1256
        elif var_type == "Variable":
T
wip  
typhoonzero 已提交
1257 1258
            var = Variable(
                self,
T
typhoonzero 已提交
1259
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1260 1261 1262 1263
                name=new_name,
                error_clip=error_clip,
                stop_gradient=stop_gradient)

W
Wu Yi 已提交
1264
        # rename the python side, _sync_with_cpp will only add
T
wip  
typhoonzero 已提交
1265 1266 1267
        # new vars/ops to python side.
        self.vars[new_name] = var
        del self.vars[name]
W
Wu Yi 已提交
1268
        self._sync_with_cpp()
1269
        return var
T
typhoonzero 已提交
1270

W
Wu Yi 已提交
1271 1272
    def _remove_var(self, name):
        self._sync_with_cpp()
M
minqiyang 已提交
1273
        self.desc._remove_var(cpt.to_bytes(name))
1274 1275
        del self.vars[name]

Y
Yu Yang 已提交
1276 1277
    def create_parameter(self, *args, **kwargs):
        global_block = self.program.global_block()
Q
Qiao Longfei 已提交
1278
        param = Parameter(global_block, *args, **kwargs)
1279
        if 'initializer' in kwargs:
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299

            def _is_inited_by(block, var):
                init_ops = []
                for op in block.ops:
                    if var.name in op.output_arg_names:
                        init_ops.append(op)
                return init_ops

            initializer = kwargs['initializer']
            init_ops = _is_inited_by(global_block, param)
            init_ops_len = len(init_ops)
            if init_ops_len > 1:
                raise RuntimeError("param " + param.name +
                                   " is inited by multiple init ops " + str(
                                       init_ops))
            elif init_ops_len == 1:
                #TODO already inited, do nothing, should log a warning
                pass
            else:
                initializer(param, self)
Q
Qiao Longfei 已提交
1300
        return param
Y
Yu Yang 已提交
1301

Y
Yu Yang 已提交
1302
    def append_op(self, *args, **kwargs):
1303 1304 1305 1306 1307 1308
        """
        Appends a new Operator according to the giving arguments.

        Returns:
            Operator: the append Operator.
        """
Y
Yu Yang 已提交
1309
        op_desc = self.desc.append_op()
1310 1311 1312 1313 1314 1315 1316 1317
        op = Operator(
            block=self,
            desc=op_desc,
            type=kwargs.get("type", None),
            inputs=kwargs.get("inputs", None),
            outputs=kwargs.get("outputs", None),
            attrs=kwargs.get("attrs", None))
        self.ops.append(op)
M
minqiyang 已提交
1318

M
minqiyang 已提交
1319 1320
        # TODO(minqiyang): add stop_gradient support in static mode too.
        # currently, we only support stop_gradient in imperative mode.
1321 1322 1323 1324
        self._trace_op(op, kwargs.get("stop_gradient", False))
        return op

    def _trace_op(self, op, stop_gradient=False):
1325
        if _in_imperative_mode():
1326
            _imperative_tracer().trace(op.iop, op.inputs, op.outputs, self.desc,
M
minqiyang 已提交
1327 1328
                                       _imperative_current_expected_place_,
                                       stop_gradient)
Y
Yu Yang 已提交
1329

W
Wu Yi 已提交
1330
    def _insert_op(self, index, *args, **kwargs):
1331 1332 1333 1334 1335 1336 1337 1338 1339
        """
        Insert a Operator according to the giving arguments.

        Args:
            index(int): the place that the operator to insert.

        Returns:
            Operator: the insert Operator.
        """
W
Wu Yi 已提交
1340 1341
        self._sync_with_cpp()
        op_desc = self.desc._insert_op(index)
Q
qiaolongfei 已提交
1342 1343 1344 1345
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
        self.ops.insert(index, op)
        return op

W
Wu Yi 已提交
1346
    def _remove_op(self, index):
1347 1348 1349 1350 1351 1352 1353 1354 1355
        """
        Remove the specific position operator.

        Args:
            index(int): the position that the operator to insert.

        Returns:
            None
        """
W
Wu Yi 已提交
1356 1357
        self._sync_with_cpp()
        self.desc._remove_op(index, index + 1)
1358 1359
        del self.ops[index]

W
Wu Yi 已提交
1360
    def _slice_ops(self, start, end):
1361 1362 1363 1364 1365 1366 1367 1368 1369 1370
        """
        Return the Operator between start and end.

        Args:
            start(int): the start position.
            end(int): the end position.

        Returns:
            list: the Operators between start and end.
        """
Q
qiaolongfei 已提交
1371
        return self.ops[start:end]
Y
Yancey1989 已提交
1372

W
Wu Yi 已提交
1373 1374
    def _prepend_op(self, *args, **kwargs):
        op_desc = self.desc._prepend_op()
1375 1376 1377 1378 1379 1380 1381
        op = Operator(
            self,
            op_desc,
            type=kwargs.get("type", None),
            inputs=kwargs.get("inputs", None),
            outputs=kwargs.get("outputs", None),
            attrs=kwargs.get("attrs", None))
Q
qiaolongfei 已提交
1382
        self.ops.insert(0, op)
1383
        self._trace_op(op, kwargs.get("stop_gradient", False))
Y
Yu Yang 已提交
1384 1385
        return op

W
Wu Yi 已提交
1386
    def _sync_with_cpp(self):
1387
        """
1388 1389
        Sync from the desc on the c++ end. This method is used to synchronize
        the c++ desc instance generated by backward.
1390
        """
Q
Qiao Longfei 已提交
1391 1392 1393 1394 1395
        # sync variables from cpp
        for var in self.desc.all_vars():
            if not self.has_var(var.name()):
                self.create_var(name=var.name(), desc=var, type=var.type())

1396
        # sync variables removed from c++ end
1397
        for var in list(self.vars.keys()):
M
minqiyang 已提交
1398
            if not self.desc.find_var(cpt.to_bytes(var)):
1399 1400
                self.vars.pop(var)

Q
Qiao Longfei 已提交
1401
        # sync operators from cpp
1402 1403 1404 1405
        ops_in_cpp = []
        for op_idx in range(0, self.desc.op_size()):
            ops_in_cpp.append(self.desc.op(op_idx))

Y
Yu Yang 已提交
1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421
        if len(self.ops) != 0:
            first_op_in_python = self.ops[0].desc
            last_op_in_python = self.ops[len(self.ops) - 1].desc
            start_index = None
            end_index = None
            for index in range(len(ops_in_cpp)):
                if first_op_in_python == ops_in_cpp[index]:
                    start_index = index
                if last_op_in_python == ops_in_cpp[index]:
                    end_index = index
            assert start_index is not None
            assert end_index is not None
            assert start_index <= end_index
        else:
            start_index = 0
            end_index = -1
Q
Qiao Longfei 已提交
1422 1423 1424 1425 1426

        # sync ops append to the head of cpp_ops
        for index in range((start_index - 1 - 1), -1, -1):
            op_desc = ops_in_cpp[index]
            op = Operator(self, op_desc)
Q
qiaolongfei 已提交
1427
            self.ops.insert(0, op)
Q
Qiao Longfei 已提交
1428 1429 1430 1431 1432 1433 1434

        # sync ops append to the end of cpp_ops
        for index in range((end_index + 1), len(ops_in_cpp)):
            op_desc = ops_in_cpp[index]
            op = Operator(self, op_desc)
            self.ops.append(op)

1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447
        # sync ops removed from c++ end
        if end_index != -1 and end_index < len(self.ops):
            ops_in_cpp_index = 0
            ops_in_python_index = 0
            while ops_in_python_index < len(
                    self.ops) and ops_in_cpp_index < len(ops_in_cpp):
                if self.ops[ops_in_python_index].desc != ops_in_cpp[
                        ops_in_cpp_index]:
                    del self.ops[ops_in_python_index]
                else:
                    ops_in_cpp_index += 1
                    ops_in_python_index += 1

Q
Qiao Longfei 已提交
1448 1449 1450 1451
        assert len(self.ops) == len(ops_in_cpp)
        for index in range(len(self.ops)):
            assert self.ops[index].desc == ops_in_cpp[index]

W
Wu Yi 已提交
1452
    def _copy_param_info_from(self, other):
1453
        """
1454 1455
        Copy the information of parameters from the other block.

1456
        Args:
1457 1458 1459 1460 1461
            other(Block): the other block.

        Raises:
            ValueError: If type of input is not Block, or the `other` and this
                block is not in the same topology.
1462 1463 1464 1465 1466

        Returns:
            None
        """
        if not isinstance(other, Block):
W
Wu Yi 已提交
1467 1468
            raise TypeError(
                "_copy_param_info_from should be invoked with Block")
1469
        for p in other.iter_parameters():
1470 1471 1472
            assert isinstance(p, Parameter)
            v = self.vars.get(p.name, None)
            if v is None:
W
Wu Yi 已提交
1473
                raise ValueError("_copy_param_info_from should be invoked with "
1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485
                                 "same topology")
            assert isinstance(v, Variable)
            new_p = Parameter(
                block=self,
                shape=v.shape,
                dtype=v.dtype,
                type=v.type,
                lod_level=v.lod_level,
                stop_gradient=p.stop_gradient,
                trainable=p.trainable,
                optimize_attr=p.optimize_attr,
                regularizer=p.regularizer,
F
fengjiayi 已提交
1486
                gradient_clip_attr=p.gradient_clip_attr,
F
fengjiayi 已提交
1487
                error_clip=p.error_clip,
1488 1489 1490
                name=v.name)
            self.vars[new_p.name] = new_p

W
Wu Yi 已提交
1491
    def _clone_variable(self, var):
1492 1493
        """
        Clone a variable into current block.
1494

1495 1496 1497 1498
        Args:
            var: the variable to be cloned.

        Returns:
1499
            Variable: the new  variable cloned from 'var' in current block.
1500 1501
        """
        assert isinstance(var, Variable)
T
update  
typhoonzero 已提交
1502 1503 1504 1505 1506
        ret_var = None
        # make STEP_SCOPES var can be safely cloned.
        if var.type == core.VarDesc.VarType.STEP_SCOPES:
            ret_var = self.create_var(
                name=var.name, persistable=var.persistable, type=var.type)
T
tangwei12 已提交
1507 1508
        elif var.type == core.VarDesc.VarType.RAW:
            ret_var = self.create_var(
T
tangwei12 已提交
1509
                name=var.name, persistable=var.persistable, type=var.type)
T
typhoonzero 已提交
1510 1511 1512 1513 1514 1515
        elif var.type == core.VarDesc.VarType.SELECTED_ROWS:
            ret_var = self.create_var(
                name=var.name,
                shape=var.shape,
                dtype=var.dtype,
                type=var.type,
F
fengjiayi 已提交
1516 1517
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1518 1519 1520 1521 1522 1523 1524
        else:
            ret_var = self.create_var(
                name=var.name,
                shape=var.shape,
                dtype=var.dtype,
                type=var.type,
                lod_level=var.lod_level,
F
fengjiayi 已提交
1525 1526
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1527
        return ret_var
1528

Y
Yu Yang 已提交
1529

1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677
class IrGraph(object):
    """
    IrGraph uses core.Graph as the delegation to accomplish the manipulation.
    """

    def __init__(self, graph, for_test=False):
        """
        Construct the IrGraph using core.Graph.
        Args:
            graph(core.Graph): C++ Graph.
            for_test(bool): True for the test graph and false for the train graph.
        """
        assert isinstance(
            graph, core.Graph), 'graph must be the instance of core.Graph.'
        self.graph = graph
        self._for_test = for_test

    def is_test(self):
        return self._for_test

    def all_parameters(self):
        param_nodes = set()
        for node in self.graph.nodes():
            if node.is_var() and node.var() is not None and node.var(
            ).persistable():
                param_nodes.add(node)
        return param_nodes

    def all_vars(self):
        return {node for node in self.graph.nodes() if node.is_var()}

    def all_ops(self):
        return {node for node in self.graph.nodes() if node.is_op()}

    def create_param_node(self, name, var_type, shape, var_dtype):
        var_desc = core.VarDesc(name)
        var_desc.set_type(var_type)
        var_desc.set_shape(shape)
        var_desc.set_dtype(var_dtype)
        var_desc.set_persistable(True)
        return self.graph.create_var_node(var_desc)

    def create_var_node(self, name, var_type, shape, var_dtype):
        var_desc = core.VarDesc(name)
        var_desc.set_type(var_type)
        var_desc.set_shape(shape)
        var_desc.set_dtype(var_dtype)
        return self.graph.create_var_node(var_desc)

    def create_var_node_from_desc(self, var_desc):
        return self.graph.create_var_node(var_desc)

    def create_op_node(self, op_type, attrs, inputs, outputs):
        op_desc = core.OpDesc()
        op_desc.set_type(op_type)
        for attr, value in attrs.iteritems():
            self._update_desc_attr(op_desc, attr, value)
        for input_name, var_nodes in inputs.iteritems():
            if not isinstance(var_nodes, list):
                var_nodes = [var_nodes]
            op_desc.set_input(input_name,
                              [var_node.name() for var_node in var_nodes])
        for output_name, var_nodes in outputs.iteritems():
            if not isinstance(var_nodes, list):
                var_nodes = [var_nodes]
            op_desc.set_output(output_name,
                               [var_node.name() for var_node in var_nodes])
        return self.graph.create_op_node(op_desc)

    def create_op_node_from_desc(self, op_desc):
        return self.graph.create_op_node(op_desc)

    def update_input_link(self, old_input_node, new_input_node, op_node):
        assert old_input_node in self.graph.nodes() and new_input_node in self.graph.nodes() and \
            op_node in self.graph.nodes(), 'Th three arguments must be in the graph nodes.'
        old_input_node.outputs_remove(op_node)
        op_node.inputs_remove(old_input_node)
        new_input_node.outputs_append(op_node)
        op_node.inputs_append(new_input_node)
        op_node.op()._rename_input(old_input_node.name(), new_input_node.name())

    def link_to(self, node_in, node_out):
        assert node_in in self.graph.nodes() and node_out in self.graph.nodes(), \
            'Th two arguments must be in the graph nodes.'
        node_in.outputs_append(node_out)
        node_out.inputs_append(node_in)

    def safe_remove_nodes(self, remove_nodes):
        if not isinstance(remove_nodes, set):
            remove_nodes = set(remove_nodes)
        core.graph_safe_remove_nodes(self.graph, remove_nodes)

    def draw(self, save_path, name, marked_nodes=None):
        def _convert_to_pdf(dot_file_path):
            pdf_save_path = os.path.splitext(dot_file_path)[0] + '.pdf'
            exited_code = subprocess.call('dot -Tpdf ' + dot_file_path \
                            + ' -o ' + pdf_save_path, shell=True)
            if exited_code != 0:
                print('The dot command is needed for creating pdf files.')
                print('The {} is saved as the dot filetype.'.format(
                    dot_file_path))

        remove_ctr_vars = set()
        ops_num = 0
        for node in self.graph.nodes():
            if node.is_ctrl_var():
                remove_ctr_vars.add(node)
            elif node.is_op():
                ops_num += 1
        print('Total ops num = {}.'.format(ops_num))
        self.safe_remove_nodes(remove_ctr_vars)
        if marked_nodes is not None:
            if not isinstance(marked_nodes, set):
                marked_nodes = set(marked_nodes)
            marked_nodes = marked_nodes - remove_ctr_vars
            if self.graph.has('__graphviz__marked_node__'):
                self.graph.erase('__graphviz__marked_node__')
            self.graph.set('__graphviz__marked_node__', marked_nodes)
        viz_dot_path = os.path.join(save_path, name) + '.dot'
        viz_pass = core.get_pass('graph_viz_pass')
        viz_pass.set('graph_viz_path', viz_dot_path)
        viz_pass.apply(self.graph)
        _convert_to_pdf(viz_dot_path)

    def to_program(self):
        convert_pass = core.get_pass('graph_to_program_pass')
        convert_pass.set('program', Program().desc)
        convert_pass.apply(self.graph)
        desc = convert_pass.get_program('program')
        program = Program._construct_from_desc(desc)
        return program

    def _update_desc_attr(self, desc, name, val):
        """
        Update the value of desc's attribute by attribute's name.
        """
        if isinstance(val, Block):
            desc.set_block_attr(name, val.desc)
        elif isinstance(val, list) and val and all(
                isinstance(v, Block) for v in val):
            desc.set_blocks_attr(name, [v.desc for v in val])
        elif isinstance(val, core.BlockDesc) or \
                isinstance(val, core.ProgramDesc):
            desc.set_serialized_attr(name, val.serialize_to_string())
        else:
            desc._set_attr(name, val)


Y
Yu Yang 已提交
1678
class Program(object):
D
dzhwinter 已提交
1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689
    """
    Python Program. Beneath it is a ProgramDesc, which is used for
    create c++ Program. A program is a self-contained programing
    language like container. It has at least one Block, when the
    control flow op like conditional_block, while_op is included,
    it will contains nested block.
    Please reference the framework.proto for details.

    Notes: we have default_startup_program and default_main_program
    by default, a pair of them will shared the parameters.
    The default_startup_program only run once to initialize parameters,
Y
yuyang18 已提交
1690
    default_main_program run in every mini batch and adjust the weights.
D
dzhwinter 已提交
1691 1692

    Returns:
Y
yuyang18 已提交
1693
        A empty program.
D
dzhwinter 已提交
1694 1695

    Examples:
Y
yuyang18 已提交
1696 1697 1698 1699 1700 1701
        >>> main_program = fluid.Program()
        >>> startup_program = fluid.Program()
        >>> with fluid.program_guard(main_program=main_program, startup_program=startup_program):
        >>>     fluid.layers.data(name="x", shape=[-1, 784], dtype='float32')
        >>>     fluid.layers.data(name="y", shape=[-1, 1], dtype='int32')
        >>>     fluid.layers.fc(name="fc", shape=[10], dtype='float32', act="relu")
D
dzhwinter 已提交
1702 1703 1704

    """

1705 1706
    def __init__(self):
        self.desc = core.ProgramDesc()
Y
Yu Yang 已提交
1707 1708
        self.blocks = [Block(self, 0)]
        self.current_block_idx = 0
D
dzhwinter 已提交
1709
        self._seed = 0
Y
yuyang18 已提交
1710
        self._current_role = core.op_proto_and_checker_maker.OpRole.Forward
Y
yuyang18 已提交
1711
        self._op_role_var = []
T
tangwei12 已提交
1712

1713 1714
        # for distribute training
        # _is_distributed = True if under distributed training
T
tangwei12 已提交
1715
        self._is_distributed = False
1716
        # _is_chief = True if the trainer is the first one, usually No.0
T
tangwei12 已提交
1717
        self._is_chief = False
1718 1719 1720
        # _parameters_on_pservers records all the parameters distributed on parameter servers.
        self._parameters_on_pservers = None
        # _endpoints is a list about parameter servers ip:port, such as ["ip:port","ip:port"]
T
tangwei12 已提交
1721
        self._endpoints = []
1722 1723 1724
        # if current role is parameter server, the _ps_endpoint is its "ip:port"
        self._ps_endpoint = None
        # trainers_endpoints, it is used for distribution.
1725
        self._trainers_endpoints = []
1726
        # the distributed lookup table names
T
tangwei12 已提交
1727
        self._distributed_lookup_table = None
Y
yuyang18 已提交
1728 1729 1730

    @property
    def op_role(self):
Y
yuyang18 已提交
1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743
        """
        The operator role. In a enum {Forward, Backward, Optimize}.

        Notes: this is a low level API. It is used only for ParallelExecutor to
        duplicate or schedule operator to devices.

        For example, the forward operator should be executed on every device.
        The backward operator should be executed on every device and the
        parameter gradient of backward (use :code:`op_role_var` to get this
        variable) operator should be merged to one device. The optimization
        operators should be executed on only one device and broadcast the
        optimization result, i.e., the new parameter, to every other device.
        """
Y
yuyang18 已提交
1744 1745 1746 1747 1748 1749 1750 1751
        return self._current_role

    @op_role.setter
    def set_op_role(self, role):
        self._current_role = role

    @property
    def op_role_var(self):
Y
yuyang18 已提交
1752 1753 1754 1755 1756 1757 1758
        """
        The auxiliary variables for :code:`op_role` property.

        See Also: :code:`Program.op_role`'s documentation for details.

        Notes: This is a very low-level API. Users should not use it directly.
        """
Y
yuyang18 已提交
1759 1760 1761 1762
        return self._op_role_var

    @op_role_var.setter
    def set_op_role_var(self, var_name):
Y
yuyang18 已提交
1763
        self._op_role_var = [var_name]
Y
yuyang18 已提交
1764 1765

    @contextlib.contextmanager
W
Wu Yi 已提交
1766
    def _optimized_guard(self, param_and_grads):
Y
yuyang18 已提交
1767 1768 1769 1770 1771 1772 1773
        """
        A with guard to set :code:`Optimization` :code:`OpRole` and
        :code:`OpRoleVar` automatically.

        Notes: This is a very low level API. Users should not use it directly.

        Args:
1774
            param_and_grads(list): The variables (names) to be optimized.
Y
yuyang18 已提交
1775 1776 1777 1778

        Examples:

            >>> p, g = backward(...)
W
Wu Yi 已提交
1779
            >>> with program._optimized_guard([p,g]):
Y
yuyang18 已提交
1780 1781
            >>>     p = p - 0.001 * g
        """
X
Xin Pan 已提交
1782 1783 1784
        tmp_role = self._current_role
        tmp_var = self._op_role_var

Y
yuyang18 已提交
1785 1786
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.Optimize
1787 1788 1789 1790
        self._op_role_var = [
            var.name if isinstance(var, Variable) else var
            for var in param_and_grads
        ]
Y
yuyang18 已提交
1791
        yield
X
Xin Pan 已提交
1792 1793
        self._op_role_var = tmp_var
        self._current_role = tmp_role
Y
Yu Yang 已提交
1794

1795
    @contextlib.contextmanager
X
Xin Pan 已提交
1796
    def _lr_schedule_guard(self, is_with_opt=False):
1797 1798 1799 1800 1801 1802 1803
        """
        A with guard to set :code:`LRSched` :code:`OpRole` and
        :code:`OpRoleVar` automatically. The :code:`OpRoleVar` is
        set to the target learning rate.

        Notes: This is a very low level API. Users should not use it directly.

X
Xin Pan 已提交
1804 1805 1806 1807
        Args:
            is_with_opt: Only set to true if these ops a in the middle
                 of a bunch of optimize ops so that it can be treated
                 correctly. For example, sgd->lr_op->sgd->lr_op->sgd.
1808 1809 1810 1811 1812 1813 1814

        Examples:

            >>> p, g = backward(...)
            >>> with program.lr_schedule_guard():
            >>>     lr = lr * decay
        """
1815 1816 1817 1818

        tmp_role = self._current_role
        tmp_var = self._op_role_var

1819 1820
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.LRSched
X
Xin Pan 已提交
1821 1822
        if is_with_opt:
            self._current_role = int(OpRole.LRSched) | int(OpRole.Optimize)
1823 1824 1825
        # TODO(typhoonzero): how to set target learning rate var
        self._op_role_var = []
        yield
1826 1827
        self._op_role_var = tmp_var
        self._current_role = tmp_role
1828

1829
    def __str__(self):
Y
yuyang18 已提交
1830 1831 1832 1833 1834 1835 1836 1837 1838
        """
        Get the protobuf debug string of this Program.

        Returns:
            (str): The protobuf debug string.

        Raises:
            ValueError: If any of required fields is not set.
        """
Y
Yang Yang(Tony) 已提交
1839 1840
        return self.to_string(True)

F
fengjiayi 已提交
1841 1842 1843
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
Y
yuyang18 已提交
1844

F
fengjiayi 已提交
1845
        Args:
Y
yuyang18 已提交
1846 1847
            throw_on_error(bool): raise Value error when any of required fields
                is not set.
F
fengjiayi 已提交
1848

Y
yuyang18 已提交
1849 1850 1851 1852
            with_details(bool): True if more details about variables and
                parameters, e.g., :code:`trainable`, :code:`optimize_attr`, need
                to print.

H
haowang101779990 已提交
1853 1854
        Returns:
            str : The debug string.
Y
yuyang18 已提交
1855 1856 1857 1858

        Raises:
            ValueError: If any of required fields is not set and throw_on_error is
                True.
F
fengjiayi 已提交
1859 1860 1861 1862 1863 1864 1865 1866 1867 1868

        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
            res_str = ""
            for block in self.blocks:
                res_str += block.to_string(throw_on_error, with_details)
        else:
            protostr = self.desc.serialize_to_string()
1869 1870
            proto = framework_pb2.ProgramDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
1871 1872
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1873

W
Wu Yi 已提交
1874
    def _get_desc(self):
Y
yuyang18 已提交
1875 1876 1877 1878 1879 1880 1881
        """
        Get the C++ side of `ProgramDesc` object pointer. The C++ object is
        exposed by :code:`pybind`.

        Notes: This is a very low level API. Users should not use this API
        directly.
        """
1882 1883
        return self.desc

X
version  
Xin Pan 已提交
1884 1885 1886
    def _version(self):
        return self.desc._version()

1887
    def clone(self, for_test=False):
Y
yuyang18 已提交
1888 1889 1890
        """
        Create a new, duplicated program.

1891

Y
yuyang18 已提交
1892 1893 1894 1895
        Some operators, e.g., :code:`batch_norm`, behave differently between
        training and testing. They have an attribute, :code:`is_test`, to
        control this behaviour. This method will change the :code:`is_test`
        attribute of them to :code:`True` when :code:`for_test=True`.
1896

Y
yuyang18 已提交
1897 1898 1899 1900
        * Set for_test to False when we want to clone the program for training.
        * Set for_test to True when we want to clone the program for testing.

        Notes: This API DOES NOT prune any operator. Use
L
Luo Tao 已提交
1901 1902 1903 1904 1905
        :code:`clone(for_test=True)` before backward and optimization please. e.g.

            >>> test_program = fluid.default_main_program().clone(for_test=True)
            >>> optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
            >>> optimizer.minimize()
1906 1907

        Args:
Y
yuyang18 已提交
1908 1909
            for_test(bool): True if change the :code:`is_test` attribute of
                operators to :code:`True`.
1910

D
dzhwinter 已提交
1911
        Returns:
Y
yuyang18 已提交
1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964
            Program: The new, duplicated Program object.

        Examples:

            1. To clone a test program, the sample code is:

            >>> import paddle.fluid as fluid
            >>> train_program = fluid.Program()
            >>> startup_program = fluid.Program()
            >>> with fluid.program_guard(train_program, startup_program):
            >>>     img = fluid.layers.data(name='image', shape=[784])
            >>>     hidden = fluid.layers.fc(input=img, size=200, act='relu')
            >>>     hidden = fluid.layers.dropout(hidden, dropout_prob=0.5)
            >>>     loss = fluid.layers.cross_entropy(
            >>>                 input=fluid.layers.fc(hidden, size=10, act='softmax'),
            >>>                 label=fluid.layers.data(name='label', shape=[1], dtype='int64'))
            >>>
            >>> test_program = train_program.clone(for_test=True)
            >>>
            >>> sgd = fluid.optimizer.SGD(learning_rate=1e-3)
            >>> with fluid.program_guard(train_program, startup_program):
            >>>     sgd.minimize(loss)

            2. The :code:`clone` method can be avoid if you create program for
            training and program for testing individually.

            >>> import paddle.fluid as fluid
            >>>
            >>> def network(is_test):
            >>>     img = fluid.layers.data(name='image', shape=[784])
            >>>     hidden = fluid.layers.fc(input=img, size=200, act='relu')
            >>>     hidden = fluid.layers.dropout(hidden, dropout_prob=0.5, is_test=is_test)
            >>>     loss = fluid.layers.cross_entropy(
            >>>                 input=fluid.layers.fc(hidden, size=10, act='softmax'),
            >>>                 label=fluid.layers.data(name='label', shape=[1], dtype='int64'))
            >>>     return loss
            >>>
            >>> train_program = fluid.Program()
            >>> startup_program = fluid.Program()
            >>> test_program = fluid.Program()
            >>>
            >>> with fluid.program_guard(train_program, startup_program):
            >>>     with fluid.unique_name.guard():
            >>>         loss = network(is_test=False)
            >>>         sgd = fluid.optimizer.SGD(learning_rate=1e-3)
            >>>         sgd.minimize(loss)
            >>>
            >>> # the test startup program is not used.
            >>> with fluid.program_guard(test_program, fluid.Program()):
            >>>     with fluid.unique_name.guard():
            >>>         loss = network(is_test=True)

            The two code snippets above will generate same programs.
1965 1966
        """
        if for_test:
X
Xin Pan 已提交
1967
            p = self._inference_optimize(prune_read_op=False)
1968
        else:
1969
            p = Program()
G
gongweibao 已提交
1970 1971
            p.current_block_idx = self.current_block_idx
            p._seed = self._seed
1972
            p.desc = core.ProgramDesc(self.desc)
M
minqiyang 已提交
1973 1974 1975
            p.blocks = [
                Block(p, i) for i in six.moves.range(self.desc.num_blocks())
            ]
G
gongweibao 已提交
1976 1977 1978 1979

            p._current_role = self._current_role
            p._op_role_var = self._op_role_var

W
Wu Yi 已提交
1980
            p._sync_with_cpp()
1981

W
Wu Yi 已提交
1982
        p._copy_param_info_from(self)
W
Wu Yi 已提交
1983
        p._copy_data_info_from(self)
1984
        p._copy_dist_param_info_from(self)
Y
Yu Yang 已提交
1985
        return p
1986

W
Wu Yi 已提交
1987
    def _prune(self, targets):
Y
yuyang18 已提交
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
        """
        Prune operators and variables which are not needed to generate
        :code:`targets`.

        Notes: This is a very low level API. Users should not use this API
        directly. This API is in flux and not stable.

        Args:
            targets(list|Variable|Operator): A list of variables or operators
                need to be pruned

        Returns:
            Program:  A new, pruned program.

        """
2003 2004 2005 2006 2007 2008
        if not isinstance(targets, list):
            targets = [targets]
        targets_idx = []
        for t in targets:
            if not isinstance(t, Operator):
                if isinstance(t, Variable):
2009 2010
                    # After transpiler processing, the op that output this
                    # variable maybe has been changed, so t.op is not reliable
2011
                    # and we need to find the current op that generate this
2012 2013 2014 2015 2016 2017 2018 2019
                    # variable here.
                    t.op = None
                    global_block = self.global_block()
                    for idx, op in enumerate(global_block.ops):
                        if t.name in op.output_arg_names:
                            t.op = op
                            break

2020
                    t = t.op
2021 2022 2023 2024
                    if t is None:
                        raise ValueError(
                            "The target variable must have an "
                            "associated operator that generates it.")
2025
                else:
2026 2027
                    raise ValueError("All targets of prune() can only be "
                                     "Variable or Operator.")
2028 2029 2030 2031

            targets_idx.append([t.block.idx, t.idx])
        res = Program()
        res.desc = core.prune(self.desc, targets_idx)
M
minqiyang 已提交
2032 2033 2034
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
2035
        res._sync_with_cpp()
2036 2037
        return res

X
Xin Pan 已提交
2038
    def _inference_optimize(self, prune_read_op=True):
Y
yuyang18 已提交
2039
        """
F
fengjiayi 已提交
2040 2041 2042 2043 2044
        This method will create a new program and do following adjustments on it:
        1. Remove all reader variables and their creator ops if exist.

        2. Remove the :code:`read_op` if exists.

2045
        3. change the :code:`is_test`
Y
yuyang18 已提交
2046 2047 2048
        attribute of operators to :code:`True`. All the :code:`Parameter`
        information will be lost.

2049
        Args:
X
Xin Pan 已提交
2050 2051
            prune_read_op(bool): remove the read ops that are added by py_reader
                                 for cpp inference library
2052

Y
yuyang18 已提交
2053 2054 2055 2056 2057 2058
        Notes: This API is a very low level API. Use
        :code:`Program.clone(for_test=True)` instead.

        Returns:
            Program: The new program.
        """
2059
        res = Program()
2060
        res.desc = core.ProgramDesc(self.desc)
F
fengjiayi 已提交
2061 2062 2063 2064

        # remove all readers and the read_op if exist
        read_op_idx = 0
        root_block = res.desc.block(0)
X
Xin Pan 已提交
2065
        if prune_read_op:
2066 2067 2068 2069 2070 2071 2072 2073 2074
            while True:
                if read_op_idx >= root_block.op_size() or root_block.op(
                        read_op_idx).type() == 'read':
                    break
                read_op_idx += 1
            if read_op_idx < root_block.op_size():
                root_block._remove_op(0, read_op_idx + 1)
            for var in root_block.all_vars():
                if var.type() == core.VarDesc.VarType.READER:
M
minqiyang 已提交
2075
                    root_block._remove_var(cpt.to_bytes(var.name()))
F
fengjiayi 已提交
2076 2077

        # change all `is_test` attributes to True
M
minqiyang 已提交
2078
        for i in six.moves.range(res.desc.num_blocks()):
2079
            block = res.desc.block(i)
M
minqiyang 已提交
2080
            for j in six.moves.range(block.op_size()):
2081 2082
                op = block.op(j)
                if op.has_attr('is_test'):
W
Wu Yi 已提交
2083
                    op._set_attr('is_test', True)
M
minqiyang 已提交
2084 2085 2086
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
2087
        res._sync_with_cpp()
2088 2089
        return res

2090 2091
    @staticmethod
    def parse_from_string(binary_str):
Y
yuyang18 已提交
2092 2093 2094 2095 2096 2097 2098
        """
        Deserialize a program desc from protobuf binary string.

        Notes: All information about parameters will be lost after serialization
        and deserialization.

        Args:
2099
            binary_str_type(str): The binary prootbuf string.
Y
yuyang18 已提交
2100 2101 2102 2103

        Returns:
            Program: A deserialized program desc.
        """
2104 2105
        p = Program()
        p.desc = core.ProgramDesc(binary_str)
M
minqiyang 已提交
2106
        p.blocks = [Block(p, i) for i in six.moves.range(p.desc.num_blocks())]
W
Wu Yi 已提交
2107
        p._sync_with_cpp()
2108
        return p
Y
Yu Yang 已提交
2109

2110
    @staticmethod
2111
    def _construct_from_desc(desc):
2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126
        """
        Construct a program from program desc.

        Args:
            desc(core.ProgramDesc): The program desc for constructing.

        Returns:
            Program: A program.
        """
        p = Program()
        p.desc = desc
        p.blocks = [Block(p, i) for i in six.moves.range(p.desc.num_blocks())]
        p._sync_with_cpp()
        return p

D
dzhwinter 已提交
2127 2128
    @property
    def random_seed(self):
Y
yuyang18 已提交
2129 2130 2131 2132 2133 2134
        """
        The default random seed for random operators in Program. Zero means get
        the random seed from random device.

        Notes: It must be set before the operators have been added.
        """
D
dzhwinter 已提交
2135 2136
        return self._seed

Q
qiaolongfei 已提交
2137 2138
    @property
    def num_blocks(self):
Y
yuyang18 已提交
2139 2140 2141
        """
        The number of blocks in this program.
        """
Q
qiaolongfei 已提交
2142 2143
        return self.desc.num_blocks()

D
dzhwinter 已提交
2144 2145 2146 2147 2148 2149
    @random_seed.setter
    def random_seed(self, seed):
        if not isinstance(seed, int):
            raise ValueError("Seed must be a integer.")
        self._seed = seed

Y
Yu Yang 已提交
2150
    def __repr__(self):
2151
        return self.__str__()
2152

Y
Yu Yang 已提交
2153
    def global_block(self):
Y
yuyang18 已提交
2154 2155 2156
        """
        Get the first block of this program.
        """
Y
Yu Yang 已提交
2157 2158
        return self.blocks[0]

Q
Qiao Longfei 已提交
2159
    def block(self, index):
Y
yuyang18 已提交
2160 2161 2162 2163 2164 2165 2166 2167
        """
        Get the :code:`index` block of this program
        Args:
            index(int): The index of block to get

        Returns:
            Block: The :code:`index` block
        """
Q
Qiao Longfei 已提交
2168 2169
        return self.blocks[index]

Y
Yu Yang 已提交
2170
    def current_block(self):
Y
yuyang18 已提交
2171 2172 2173 2174
        """
        Get the current block. The :code:`current` block is the block to append
        operators.
        """
Y
Yu Yang 已提交
2175 2176
        return self.blocks[self.current_block_idx]

W
Wu Yi 已提交
2177
    def _create_block(self, parent_idx=None):
Y
yuyang18 已提交
2178 2179 2180 2181 2182 2183 2184 2185 2186 2187
        """
        Create a new block with the :code:`parent_idx` and change the current block
        to new block.

        Args:
            parent_idx(int): The parent block index.

        Returns:
            Block: The new block.
        """
Y
Yu Yang 已提交
2188
        new_block_idx = len(self.blocks)
F
update  
fengjiayi 已提交
2189 2190 2191
        parent = self.current_block() if parent_idx is None else self.block(
            parent_idx)
        self.desc.append_block(parent.desc)
Y
Yu Yang 已提交
2192 2193 2194 2195
        self.current_block_idx = new_block_idx
        self.blocks.append(Block(self, self.current_block_idx))
        return self.current_block()

W
Wu Yi 已提交
2196
    def _rollback(self):
Y
yuyang18 已提交
2197 2198 2199 2200 2201
        """
        Exit a code block, i.e., roll back to the parent block.
        Returns:
            None
        """
Y
Yu Yang 已提交
2202 2203
        self.current_block_idx = self.current_block().parent_idx

W
Wu Yi 已提交
2204
    def _sync_with_cpp(self):
Y
yuyang18 已提交
2205 2206 2207 2208 2209 2210 2211 2212 2213 2214
        """
        Synchronize Python instance to its binding C++ object instance.
        If the program is modified in C++ space, this method should be invoked.

        Notes: This is a very low level API. Users should not invoke it
        directly.

        Returns:
            None
        """
Q
Qiao Longfei 已提交
2215 2216 2217
        for block_idx in range(len(self.blocks), self.desc.num_blocks()):
            self.blocks.append(Block(self, block_idx))
        for block in self.blocks:
W
Wu Yi 已提交
2218
            block._sync_with_cpp()
Q
Qiao Longfei 已提交
2219

W
Wu Yi 已提交
2220
    def _copy_param_info_from(self, other):
2221
        """
2222
        Copy the information of parameters from other program.
D
dzhwinter 已提交
2223

Y
yuyang18 已提交
2224 2225 2226
        Notes: This is a very low level API. Users should not invoke it
        directly.

2227 2228 2229 2230 2231 2232 2233
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
2234
            raise TypeError("_copy_param_info_from should be invoked with "
2235 2236 2237
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
2238
            raise ValueError("_copy_param_info_from should be invoked with two "
2239
                             "program, with represent the same topology")
W
Wu Yi 已提交
2240
        self.global_block()._copy_param_info_from(other.global_block())
2241

2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256
    def _copy_dist_param_info_from(self, other):
        """
        Copy the information of distributed information from other program.

        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
            raise TypeError("_copy_dist_param_info_from should be invoked with "
                            "Program")
        self._is_distributed = other._is_distributed
        self._is_chief = other._is_chief
2257
        self._parameters_on_pservers = other._parameters_on_pservers
2258
        self._endpoints = other._endpoints
2259
        self._ps_endpoint = other._ps_endpoint
2260 2261
        self._distributed_lookup_table = other._distributed_lookup_table

W
Wu Yi 已提交
2262
    def _copy_data_info_from(self, other):
F
fengjiayi 已提交
2263 2264
        """
        Copy the information of data variables from other program.
D
dzhwinter 已提交
2265

Y
yuyang18 已提交
2266 2267 2268
        Notes: This is a very low level API. Users should not invoke it
        directly.

F
fengjiayi 已提交
2269 2270 2271 2272 2273 2274 2275
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
2276
            raise TypeError("_copy_param_info_from should be invoked with "
F
fengjiayi 已提交
2277 2278 2279
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
2280
            raise ValueError("_copy_param_info_from should be invoked with two "
F
fengjiayi 已提交
2281
                             "program, with represent the same topology")
2282
        for var in list(other.global_block().vars.values()):
F
fengjiayi 已提交
2283 2284 2285
            if var.is_data:
                self.global_block().var(var.name).is_data = True

2286
    def list_vars(self):
Y
yuyang18 已提交
2287 2288 2289 2290 2291 2292
        """
        Get all variables from this Program. A iterable object is returned.

        Returns:
            iterable: The generator will yield every variable in this program.
        """
2293
        for each_block in self.blocks:
2294
            for each_var in list(each_block.vars.values()):
2295 2296
                yield each_var

Y
Yu Yang 已提交
2297

Y
Yu Yang 已提交
2298
class Parameter(Variable):
2299
    """
2300
    Parameter is derived from Variable. A parameter is a persistable
2301
    Variable, and will be updated by optimizers after each iteration.
2302
    The training of a neural network is essentially the updating of
2303 2304
    its parameters.

2305
    Relative to a general Variable, a Parameter has several its own
2306 2307
    member variables:

2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319
    Args:
        trainable(bool): True if the parameter need to be updated after
            iterations.
        optimize_attr(map): Parameter attributes related with optimizing.
            Currently, it only contains 'learning_rate'.
            Default: {'learning_rate': 1.0}
        regularizer(WeightDecayRegularizer): The Regularizer which will
            be applied on the parameter. Default: None
        gradient_clip_attr(BaseGradientClipAttr): The gradint clip strategy
            which will be applied on the parameter. Default: None
        do_model_average(bool): True if the model average strategy will
            be applied on this parameter.
2320 2321
    """

Y
Yu Yang 已提交
2322 2323 2324 2325 2326 2327 2328 2329 2330 2331
    def __init__(self, block, shape, dtype, **kwargs):
        if shape is None or dtype is None:
            raise ValueError("Parameter must set shape and dtype")
        if len(shape) == 0:
            raise ValueError("Parameter shape cannot be empty")

        for each in shape:
            if each < 0:
                raise ValueError("Parameter shape should not be related with "
                                 "batch-size")
2332 2333 2334

        Variable.__init__(
            self, block, persistable=True, shape=shape, dtype=dtype, **kwargs)
Y
Yu Yang 已提交
2335 2336 2337 2338
        self.trainable = kwargs.get('trainable', True)

        self.optimize_attr = kwargs.get('optimize_attr', {'learning_rate': 1.0})

2339 2340
        self.regularizer = kwargs.get('regularizer', None)

F
fengjiayi 已提交
2341
        self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None)
Y
Yu Yang 已提交
2342

W
wanghaoshuang 已提交
2343
        self.do_model_average = kwargs.get('do_model_average', None)
W
wanghaoshuang 已提交
2344

F
fengjiayi 已提交
2345 2346 2347
    def __str__(self):
        return self.to_string(True)

F
update  
fengjiayi 已提交
2348 2349 2350
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
D
dzhwinter 已提交
2351

F
update  
fengjiayi 已提交
2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365
        Args:
            throw_on_error(bool): raise exception when self is not initialized
                when throw_on_error is True
            with_details(bool): more details about variables and parameters
                (e.g. trainable, optimize_attr, ...) will be printed when with_details is True

        Returns(str): The debug string.

        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
            res_str = Variable.to_string(self, throw_on_error, True)
            additional_attr = ("trainable", "optimize_attr", "regularizer",
W
wanghaoshuang 已提交
2366
                               "gradient_clip_attr", "do_model_average")
F
update  
fengjiayi 已提交
2367
            for attr_name in additional_attr:
2368 2369
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
2370 2371
        else:
            res_str = Variable.to_string(self, throw_on_error, False)
F
fengjiayi 已提交
2372 2373 2374 2375
        return res_str

    __repr__ = __str__

Y
Yu Yang 已提交
2376

Y
Yu Yang 已提交
2377
# program is a global instance.
Y
Yu Yang 已提交
2378 2379
_main_program_ = Program()
_startup_program_ = Program()
2380

2381

2382
def default_startup_program():
Y
Yu Yang 已提交
2383
    """
Y
yuyang18 已提交
2384 2385 2386 2387 2388 2389 2390 2391 2392
    Get default/global startup program.

    The layer function in :code:`fluid.layers` will create parameters, readers,
    NCCL handles as global variables. The :code:`startup_program` will
    initialize them by the operators in startup program. The layer function will
    append these initialization operators into startup program.

    This method will return the :code:`default` or the :code:`current` startup
    program. Users can use :code:`fluid.program_guard` to switch program.
2393

Y
Yu Yang 已提交
2394 2395 2396
    Returns:
        Program: startup program
    """
Y
Yu Yang 已提交
2397
    return _startup_program_
2398

2399

2400
def default_main_program():
Y
Yu Yang 已提交
2401
    """
Y
yuyang18 已提交
2402 2403 2404 2405 2406 2407 2408 2409 2410
    Get default/global main program. The main program is used for training or
    testing.

    All layer function in :code:`fluid.layers` will append operators and
    variables to the :code:`default_main_program`.

    The :code:`default_main_program` is the default program in a lot of APIs.
    For example, the :code:`Executor.run()` will execute the
    :code:`default_main_program` when the program is not specified.
2411

Y
Yu Yang 已提交
2412 2413 2414
    Returns:
        Program: main program
    """
Y
Yu Yang 已提交
2415
    return _main_program_
Y
Yu Yang 已提交
2416 2417 2418 2419 2420


def switch_main_program(program):
    """
    Switch the main program to a new program.
2421

Y
Yu Yang 已提交
2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435
    Args:
        program(Program): The new main program

    Returns:
        Program: The previous main program
    """
    global _main_program_
    prev_program = _main_program_
    _main_program_ = program
    return prev_program


def switch_startup_program(program):
    """
2436
    Switch the startup program to a new program
Y
Yu Yang 已提交
2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451
    Args:
        program(Program): The new startup program

    Returns:
        Program: The previous startup program
    """
    global _startup_program_
    prev_program = _startup_program_
    _startup_program_ = program
    return prev_program


@contextlib.contextmanager
def program_guard(main_program, startup_program=None):
    """
Y
yuyang18 已提交
2452 2453 2454
    Change the global main program and startup program with `with` statement.
    Layer functions in the Python `with` block will append operators and
    variables to the new main programs.
2455

Y
Yu Yang 已提交
2456
    Examples:
Y
yuyang18 已提交
2457 2458 2459 2460 2461 2462 2463 2464 2465 2466

        >>> import paddle.fluid as fluid
        >>> main_program = fluid.Program()
        >>> startup_program = fluid.Program()
        >>> with fluid.program_guard(main_program, startup_program):
        >>>     data = fluid.layers.data(...)
        >>>     hidden = fluid.layers.fc(...)

    Notes: The temporary :code:`Program` can be used if the user does not need
    to construct either of startup program or main program.
2467

Y
Yu Yang 已提交
2468
    Examples:
Y
yuyang18 已提交
2469 2470 2471 2472 2473 2474

        >>> import paddle.fluid as fluid
        >>> main_program = fluid.Program()
        >>> # does not care about startup program. Just pass a temporary value.
        >>> with fluid.program_guard(main_program, fluid.Program()):
        >>>     data = ...
2475

Y
Yu Yang 已提交
2476
    Args:
Y
yuyang18 已提交
2477
        main_program(Program): New main program inside `with` statement.
2478
        startup_program(Program): New startup program inside `with` statement.
Y
Yu Yang 已提交
2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
            None means do not change startup program.
    """
    if not isinstance(main_program, Program):
        raise TypeError("main_program should be Program")
    main_program = switch_main_program(main_program)
    if startup_program is not None:
        if not isinstance(startup_program, Program):
            raise TypeError("startup_program should be Program")
        startup_program = switch_startup_program(startup_program)
    yield
    switch_main_program(main_program)
    if startup_program is not None:
        switch_startup_program(startup_program)
X
xuwei06 已提交
2492 2493


W
Wu Yi 已提交
2494
def _get_var(name, program=None):
X
xuwei06 已提交
2495
    """
Y
yuyang18 已提交
2496
    Get a variable by name from the global block of a program.
F
fengjiayi 已提交
2497

X
xuwei06 已提交
2498 2499 2500
    Args:
        name(str): name of the variable
        program(Program|None): program object.
T
tangwei12 已提交
2501
        If None, default_global_program() will be used.
X
xuwei06 已提交
2502 2503 2504 2505 2506 2507 2508

    Returns:
        Variable
    """
    if program is None:
        program = default_main_program()
    assert isinstance(name, str)
2509
    assert isinstance(program, Program)
X
xuwei06 已提交
2510 2511

    return program.global_block().var(name)
2512 2513 2514


@contextlib.contextmanager
P
Paddle CI 已提交
2515
def _imperative_guard(tracer):
2516 2517 2518
    global _imperative_tracer_
    tmp_trace = _imperative_tracer_
    _imperative_tracer_ = tracer
M
minqiyang 已提交
2519

P
Paddle CI 已提交
2520 2521 2522 2523 2524 2525 2526
    yield

    _imperative_tracer_ = tmp_trace


@contextlib.contextmanager
def _imperative_place_guard(place):
M
minqiyang 已提交
2527 2528 2529
    global _imperative_current_expected_place_
    tmp_place = _imperative_current_expected_place_
    _imperative_current_expected_place_ = place
M
minqiyang 已提交
2530

2531
    yield
M
minqiyang 已提交
2532

M
minqiyang 已提交
2533
    _imperative_current_expected_place_ = tmp_place