framework.py 104.4 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
W
WangZhen 已提交
19
from collections import Iterable
Q
qiaolongfei 已提交
20
import contextlib
S
rename  
sneaxiy 已提交
21
from .wrapped_decorator import signature_safe_contextmanager
P
peizhilin 已提交
22
import os
F
fengjiayi 已提交
23
import re
24
import traceback
25
import six
26

Y
Yu Yang 已提交
27
import numpy as np
28
import subprocess
Q
qiaolongfei 已提交
29

M
minqiyang 已提交
30
from .. import compat as cpt
31
from .proto import framework_pb2
32
try:
P
peizhilin 已提交
33
    if os.name == 'nt':
P
peizhilin 已提交
34
        import sys
P
peizhilin 已提交
35 36 37 38 39
        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)

40
    from . import core
41
except ImportError as e:
P
peizhilin 已提交
42
    if os.name == 'nt':
43
        executable_path = os.path.abspath(os.path.dirname(sys.executable))
P
peizhilin 已提交
44
        raise ImportError(
45 46 47 48 49
            """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
50 51 52 53 54 55
    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))
56
except Exception as e:
57
    raise e
58
from . import unique_name
Y
Yu Yang 已提交
59

60
__all__ = [
61 62 63 64
    'Program',
    'default_startup_program',
    'default_main_program',
    'program_guard',
65
    'name_scope',
66
]
Y
Yu Yang 已提交
67

Q
qiaolongfei 已提交
68 69 70 71
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
72 73
CONTROL_DEP_VAR_PREFIX = core.kControlDepVarName()

74
_imperative_tracer_ = None
M
minqiyang 已提交
75
_imperative_current_expected_place_ = None
76 77 78 79 80 81 82 83 84


def _in_imperative_mode():
    return _imperative_tracer_ is not None


def _imperative_tracer():
    return _imperative_tracer_

W
Wu Yi 已提交
85

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


S
rename  
sneaxiy 已提交
116
@signature_safe_contextmanager
117 118 119 120 121 122 123 124 125 126 127 128
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
129

130 131 132 133
          with name_scope("encoder"):
             ...
          with name_scope("decoder"):
             ...
T
Tink_Y 已提交
134 135
          with name_scope("attention"):
             ...
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
    """
    # 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
155 156 157
def generate_control_dev_var_name():
    import random
    return CONTROL_DEP_VAR_PREFIX + "@" + str(random.random())
Q
qiaolongfei 已提交
158 159 160 161


def grad_var_name(var_name):
    """
162 163
    Returns:
        str: gradient name for a certain var name
Q
qiaolongfei 已提交
164 165 166
    """
    return var_name + GRAD_VAR_SUFFIX

Y
Yu Yang 已提交
167

168
def convert_np_dtype_to_dtype_(np_dtype):
169 170
    """
    Convert the data type in numpy to the data type in Paddle
171

172
    Args:
173
        np_dtype(np.dtype): the data type in numpy.
174

175 176
    Returns:
        core.VarDesc.VarType: the data type in Paddle.
177 178

    """
179 180
    dtype = np.dtype(np_dtype)
    if dtype == np.float32:
181
        return core.VarDesc.VarType.FP32
182
    elif dtype == np.float64:
183
        return core.VarDesc.VarType.FP64
184
    elif dtype == np.float16:
185
        return core.VarDesc.VarType.FP16
186
    elif dtype == np.int32:
187
        return core.VarDesc.VarType.INT32
188
    elif dtype == np.int16:
189
        return core.VarDesc.VarType.INT16
190
    elif dtype == np.int64:
191
        return core.VarDesc.VarType.INT64
192
    elif dtype == np.bool:
193
        return core.VarDesc.VarType.BOOL
194 195
    elif dtype == np.uint16:
        return core.VarDesc.VarType.INT16
196 197
    elif dtype == np.uint8:
        return core.VarDesc.VarType.UINT8
Q
qingqing01 已提交
198 199
    elif dtype == np.int8:
        return core.VarDesc.VarType.INT8
200
    else:
M
minqiyang 已提交
201
        raise ValueError("Not supported numpy dtype %s" % dtype)
202 203 204


def dtype_is_floating(dtype):
205 206 207
    """
    Check the data type is floating or not.
    Args:
208
        dtype(np.dtype|core.VarDesc.VarType): data type.
209 210 211 212 213
            Could be numpy format or Paddle format

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

    """
214
    if not isinstance(dtype, core.VarDesc.VarType):
215 216
        dtype = convert_np_dtype_to_dtype_(dtype)

217 218 219 220
    return dtype in [
        core.VarDesc.VarType.FP16, core.VarDesc.VarType.FP32,
        core.VarDesc.VarType.FP64
    ]
221 222


Y
Yang Yang(Tony) 已提交
223
def _debug_string_(proto, throw_on_error=True):
224 225 226 227 228 229 230 231 232 233 234
    """
    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
235
    error_fields = list()
Y
Yang Yang(Tony) 已提交
236
    if not proto.IsInitialized(error_fields) and throw_on_error:
C
caoying03 已提交
237 238
        raise ValueError("{0} are not initialized.\nThe message is {1}:\n".
                         format(error_fields, proto))
Y
Yu Yang 已提交
239 240 241
    return proto.__str__()


X
Xin Pan 已提交
242
class Variable(object):
243
    """
244 245 246
    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
247
    two variables in different blocks could have the same name.
248

249 250
    There are many kinds of variables. Each kind of them has its own attributes
    and usages. Please reference the framework.proto for details.
251

252
    Most of a Variable's member variables can be setted to be None. It mean
253
    it is not available or will be specified later.
254 255

    Args:
256
        block(Block): The block that the variable belongs to.
257 258
        type(core.VarDesc.VarType): Variable type. Please reference the
            framework.proto for details.
259 260
        name(str|None): The name of the variable. If setted None, it will be
            generated automatically. Default: None
261
        shape(tuple|list|None): The shape of the variable. -1 means the batch size.
262
            Some kinds of variable do not contain shape, just set it to None.
263 264 265
            Default: None
        dtype(np.dtype|core.VarDesc.VarType|str|None): The data type of variable.
            Default: None
266
        lod_level (int|None): The level of lod tensor. 0 means it is not a time
267
            series data.
268
            Default: None
269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290
        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')
291 292
    """

Y
Yu Yang 已提交
293 294
    def __init__(self,
                 block,
Y
Yu Yang 已提交
295
                 type=core.VarDesc.VarType.LOD_TENSOR,
Y
Yu Yang 已提交
296 297 298 299
                 name=None,
                 shape=None,
                 dtype=None,
                 lod_level=None,
300
                 capacity=None,
Q
QI JUN 已提交
301
                 persistable=None,
F
fengjiayi 已提交
302
                 error_clip=None,
Y
Yu Yang 已提交
303
                 stop_gradient=False,
F
fengjiayi 已提交
304
                 is_data=False,
Y
Yu Yang 已提交
305
                 **kwargs):
Y
Yu Yang 已提交
306 307
        self.block = block
        if name is None:
Y
Yu Yang 已提交
308
            name = unique_name.generate('_generated_var')
D
Dong Zhihong 已提交
309

Y
Yu Yang 已提交
310
        if dtype is not None:
311
            if not isinstance(dtype, core.VarDesc.VarType):
312
                dtype = convert_np_dtype_to_dtype_(dtype)
313

X
Xin Pan 已提交
314
        if _in_imperative_mode():
M
minqiyang 已提交
315
            # record vars in tracer rather than blocks
M
minqiyang 已提交
316 317
            self._ivar = kwargs.get("ivar", None)
            if not self._ivar:
318 319 320 321 322
                self._ivar = core.VarBase(
                    name, dtype if dtype else core.VarDesc.VarType.FP32,
                    list(shape) if shape else [],
                    _current_expected_place(), True
                    if persistable else False, stop_gradient)
M
minqiyang 已提交
323
            if persistable:
324
                _imperative_tracer().trace_var(name, self)
M
minqiyang 已提交
325
        else:
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
            self.error_clip = error_clip

            is_new_var = False
            name = cpt.to_text(name)
            self.desc = self.block.desc.find_var(cpt.to_bytes(name))

            if self.desc is None:
                self.desc = self.block.desc.var(cpt.to_bytes(name))
                is_new_var = True

            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))

            if shape is not None:
                if is_new_var:
                    self.desc.set_shape(shape)
                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))
            if dtype is not None:
                if is_new_var:
                    self.desc.set_dtype(dtype)
                else:
                    old_dtype = self.dtype
                    if dtype != old_dtype:
                        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))

            if lod_level is not None:
                if is_new_var:
                    self.desc.set_lod_level(lod_level)
                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))
            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))

            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

M
minqiyang 已提交
398
            self.block.vars[name] = self
399 400 401
            self.op = None
            self.stop_gradient = stop_gradient
            self.is_data = is_data
Y
Yu Yang 已提交
402

403
    def _numpy(self):
M
minqiyang 已提交
404
        new_ivar = self._ivar._copy_to(core.CPUPlace(), True)
P
Paddle CI 已提交
405
        return np.array(new_ivar.value().get_tensor())
406 407

    def _backward(self):
X
Xin Pan 已提交
408
        self._ivar._run_backward()
409 410

    def _gradient(self):
M
minqiyang 已提交
411
        return np.array(self._ivar._grad_value())
412

X
Xin Pan 已提交
413 414
    def _clear_gradient(self):
        self._ivar._clear_gradient()
X
Xin Pan 已提交
415

416
    def __str__(self):
Y
Yang Yang(Tony) 已提交
417 418
        return self.to_string(True)

F
update  
fengjiayi 已提交
419
    def to_string(self, throw_on_error, with_details=False):
420 421 422 423
        """
        Get debug string.

        Args:
424 425
            throw_on_error(bool): True if raise an exception when self is
                not initialized.
F
update  
fengjiayi 已提交
426
            with_details(bool): more details about variables and parameters
427 428
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False;
429

430 431
        Returns:
            str: The debug string.
432
        """
433 434 435 436 437
        if _in_imperative_mode():
            # TODO(panyx0718): add imperative debug info.
            return 'name %s, dtype: %s shape: %s' % (self.name, self.dtype,
                                                     self.shape)

F
update  
fengjiayi 已提交
438 439
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
440
        protostr = self.desc.serialize_to_string()
441
        proto = framework_pb2.VarDesc.FromString(six.binary_type(protostr))
F
update  
fengjiayi 已提交
442 443 444 445
        res_str = _debug_string_(proto, throw_on_error)
        if with_details:
            additional_attr = ("error_clip", "stop_gradient")
            for attr_name in additional_attr:
446 447
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
448
        return res_str
449 450 451

    __repr__ = __str__

W
Wu Yi 已提交
452
    def _set_desc(self, input):
453 454 455 456 457 458 459 460 461
        """
        Set the variable description.

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

        Returns:
            None
        """
462 463
        self.desc = input

464 465
    @property
    def _stop_gradient(self):
M
minqiyang 已提交
466 467 468 469
        if _in_imperative_mode():
            return self._ivar.stop_gradient
        else:
            return self.stop_gradient
470 471 472

    @_stop_gradient.setter
    def _stop_gradient(self, s):
M
minqiyang 已提交
473 474
        if _in_imperative_mode():
            self._ivar.stop_gradient = s
475 476
        else:
            self.stop_gradient = s
477

478 479
    @property
    def persistable(self):
480 481 482 483
        if _in_imperative_mode():
            return self._ivar.persistable
        else:
            return self.desc.persistable()
484

Y
Yu Yang 已提交
485 486
    @persistable.setter
    def persistable(self, p):
487 488 489 490
        if _in_imperative_mode():
            return self._ivar.persistable
        else:
            self.desc.set_persistable(p)
Y
Yu Yang 已提交
491

Y
Yu Yang 已提交
492 493
    @property
    def name(self):
494 495 496 497
        if _in_imperative_mode():
            return self._ivar.name
        else:
            return cpt.to_text(self.desc.name())
Y
Yu Yang 已提交
498

T
typhoonzero 已提交
499 500
    @name.setter
    def name(self, new_name):
501 502 503 504
        if _in_imperative_mode():
            self._ivar.name = new_name
        else:
            self.desc.set_name(new_name)
T
typhoonzero 已提交
505

Y
Yu Yang 已提交
506 507 508
    @property
    def shape(self):
        # convert to tuple, make it as same as numpy API.
509 510 511 512
        if _in_imperative_mode():
            return self._ivar.shape
        else:
            return tuple(self.desc.shape())
Y
Yu Yang 已提交
513 514

    @property
F
fengjiayi 已提交
515
    def dtype(self):
516 517 518 519
        if _in_imperative_mode():
            return self._ivar.dtype
        else:
            return self.desc.dtype()
Y
Yu Yang 已提交
520 521 522

    @property
    def lod_level(self):
523
        # TODO(minqiyang): Support lod_level in imperative mode
524
        return self.desc.lod_level()
Y
Yu Yang 已提交
525

Y
Yu Yang 已提交
526 527
    @property
    def type(self):
528 529 530 531
        if _in_imperative_mode():
            return self._ivar.dtype
        else:
            return self.desc.type()
Y
Yu Yang 已提交
532

W
Wu Yi 已提交
533
    def _set_error_clip(self, error_clip):
534 535 536 537 538 539 540 541 542
        """
        Set the error_clip.

        Args:
            error_clip(BaseErrorClipAttr) : The new error_clip.

        Returns:
            None
        """
543 544
        self.error_clip = error_clip

Y
Yu Yang 已提交
545

F
fengjiayi 已提交
546 547 548
def get_all_op_protos():
    """
    Get all registered op proto from PaddlePaddle C++ end.
549

550 551
    Returns:
       list: list of OpProto.
F
fengjiayi 已提交
552 553 554 555
    """
    protostrs = core.get_all_op_protos()
    ret_values = []
    for pbstr in protostrs:
556
        op_proto = framework_pb2.OpProto.FromString(six.binary_type(pbstr))
F
fengjiayi 已提交
557 558 559 560 561
        ret_values.append(op_proto)
    return ret_values


class OpProtoHolder(object):
562 563 564 565
    """
    A global variable to hold all OpProtos from C++ as a map
    """

F
fengjiayi 已提交
566 567 568 569 570 571 572 573 574
    @classmethod
    def instance(cls):
        if not hasattr(cls, '_instance'):
            cls._instance = cls()
        return cls._instance

    def __init__(self):
        assert not hasattr(
            self.__class__,
575
            '_instance'), 'Please use `instance()` to get OpProtoHolder object!'
F
fengjiayi 已提交
576 577 578 579 580 581
        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):
582 583 584 585 586 587 588 589
        """
        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
590 591
        if type not in self.op_proto_map:
            raise ValueError("Operator \"%s\" has not been registered." % type)
F
fengjiayi 已提交
592 593
        return self.op_proto_map[type]

594 595 596 597
    @staticmethod
    def generated_op_attr_names():
        return {
            core.op_proto_and_checker_maker.kOpRoleAttrName(),
S
sneaxiy 已提交
598
            core.op_proto_and_checker_maker.kOpRoleVarAttrName(),
599 600
            core.op_proto_and_checker_maker.kOpNameScopeAttrName(),
            core.op_proto_and_checker_maker.kOpCreationCallstackAttrName()
601 602
        }

F
fengjiayi 已提交
603

X
Xin Pan 已提交
604
class Operator(object):
605
    """
606 607 608 609 610 611 612
    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
613
        type(str): The type of operator. Default None.
614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633
        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
634
        Block.append_op or Block._prepend_op instead.
635 636 637 638 639 640 641 642 643 644

    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]})
645
    """
646 647 648
    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
649 650
        'listen_and_serv', 'save_combine', 'load_combine', 'ncclInit', 'select',
        'checkpoint_notify', 'gen_nccl_id'
651
    }
652

Y
Yu Yang 已提交
653 654
    def __init__(self,
                 block,
Y
Yu Yang 已提交
655
                 desc,
Y
Yu Yang 已提交
656 657 658
                 type=None,
                 inputs=None,
                 outputs=None,
M
minqiyang 已提交
659
                 attrs=None):
X
Xin Pan 已提交
660
        if _in_imperative_mode():
661 662 663 664
            if type is None:
                raise ValueError(
                    "`type` to initilized an Operator can not be None.")
            self.iop = core.OpBase(type)
M
minqiyang 已提交
665

666 667
            # TODO(minqiyang): remove these lines after we take apart all
            # backward grads and forward variables
X
Xin Pan 已提交
668
            self.inputs = defaultdict(list)
X
Xin Pan 已提交
669
            if inputs is not None:
X
Xin Pan 已提交
670 671 672 673 674
                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])
M
minqiyang 已提交
675

X
Xin Pan 已提交
676
            self.outputs = defaultdict(list)
X
Xin Pan 已提交
677
            if outputs is not None:
X
Xin Pan 已提交
678 679 680 681 682
                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
683

684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798
            self.attrs = attrs if attrs else {}
        else:
            self.block = block
            self.desc = desc
            # 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()
            del attrs

            op_maker = core.op_proto_and_checker_maker

            if op_maker.kOpRoleAttrName() not in op_attrs:
                op_attrs[op_maker.kOpRoleAttrName(
                )] = self.block.program.op_role

            role_var_name = op_maker.kOpRoleVarAttrName()
            if len(self.block.program.
                   op_role_var) != 0 and role_var_name not in op_attrs:
                op_attrs[role_var_name] = self.block.program.op_role_var

            if role_var_name in op_attrs and len(op_attrs[role_var_name]) == 0:
                del op_attrs[role_var_name]

            if len(self.desc.type()) != 0:
                return
            if type is None:
                raise ValueError(
                    "`type` to initilized an Operator can not be None.")
            else:
                callstack_var_name = op_maker.kOpCreationCallstackAttrName()
                op_attrs[callstack_var_name] = list(
                    reversed(traceback.format_stack()))[1:]

            self.desc.set_type(type)
            proto = OpProtoHolder.instance().get_op_proto(type)

            namescope_var_name = op_maker.kOpNameScopeAttrName()
            op_attrs[namescope_var_name] = _full_name_scope()

            def find_name(var_list, name):
                for var_name in var_list:
                    if var_list[var_name] is not None and var_name == name:
                        return True
                return False

            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:
                        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:
                            raise ValueError(
                                "Input %s expects only one input, but %d are given."
                                % (in_proto.name, len(in_args)))
                        in_arg_names = []
                        for arg in in_args:
                            if isinstance(arg, six.string_types):
                                in_arg_names.append(arg)
                            elif isinstance(arg, six.binary_type):
                                in_arg_names.append(arg.decode())
                            else:
                                in_arg_names.append(cpt.to_text(arg.name))
                        self.desc.set_input(in_proto.name, in_arg_names)
                    else:
                        self.desc.set_input(in_proto.name, [])

            if outputs is not None:
                for m in proto.outputs:
                    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))
                for out_proto in proto.outputs:
                    if out_proto.name not in outputs:
                        continue
                    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:
                        raise ValueError(
                            "Output %s expects only one output, but %d are given."
                            % (out_proto.name, len(out_args)))
                    out_arg_names = []
                    for arg in out_args:
                        out_arg_names.append(cpt.to_text(arg.name))
                        # TODO(minqiyang): could we remove variable's op in static mode?
                        if not _in_imperative_mode():
                            arg.op = self
                    self.desc.set_output(out_proto.name, out_arg_names)

            if op_attrs is not None:
                if not isinstance(op_attrs, dict):
                    raise TypeError("'attrs' should be a dict.")
                for attr in proto.attrs:
                    attr_name = attr.name
                    if (attr_name not in op_attrs) or (
                            op_attrs[attr_name] is None):
                        continue
                    attr_val = op_attrs[attr_name]
                    self._update_desc_attr(attr_name, attr_val)

            self.desc.check_attrs()
            if self._has_kernel(type):
                self.desc.infer_var_type(self.block.desc)
                self.desc.infer_shape(self.block.desc)

W
Wu Yi 已提交
799
    def _has_kernel(self, op_type):
800 801
        return op_type not in self.OP_WITHOUT_KERNEL_SET

Y
Yang Yang(Tony) 已提交
802
    def to_string(self, throw_on_error):
803
        """
804 805
        Get debug string.

806
        Args:
807 808
            throw_on_error(bool): Whether to raise exception if self is not
                initialized.
809

810 811
        Returns:
            str: The debug string.
812 813

        """
814
        protostr = self.desc.serialize_to_string()
815
        proto = framework_pb2.OpDesc.FromString(six.binary_type(protostr))
Y
Yang Yang(Tony) 已提交
816 817 818 819
        return _debug_string_(proto, throw_on_error)

    def __str__(self):
        return self.to_string(True)
820 821 822

    __repr__ = __str__

F
fengjiayi 已提交
823 824 825 826 827
    @property
    def type(self):
        return self.desc.type()

    def input(self, name):
828
        """
829
        Get the input arguments according to the input parameter name.
830

831 832
        Args:
            name(str): The input parameter name.
833

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

W
Wu Yi 已提交
840
    def _rename_input(self, old_name, new_name):
841 842 843 844 845 846 847 848 849 850
        """
        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
851
        self.desc._rename_input(old_name, new_name)
T
typhoonzero 已提交
852

W
Wu Yi 已提交
853
    def _rename_output(self, old_name, new_name):
854 855 856 857 858 859 860 861 862 863
        """
        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
864
        self.desc._rename_output(old_name, new_name)
T
typhoonzero 已提交
865

F
fengjiayi 已提交
866 867 868 869
    @property
    def input_names(self):
        return self.desc.input_names()

T
typhoonzero 已提交
870 871 872 873 874 875 876 877
    @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
878
    def output(self, name):
879
        """
880
        Get output arguments by the output parameter name.
881

882 883
        Args:
            name(str): The output parameter name.
884

885 886 887
        Returns:
            list: return the list of argument names associated with \
                the specific parameter name.
888
        """
F
fengjiayi 已提交
889 890 891 892 893 894
        return self.desc.output(name)

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

895 896 897 898 899 900 901 902
    @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
903
    def has_attr(self, name):
904
        """
905 906
        Whether this Operator has the attribute with name or not.

907
        Args:
908
            name(str): the attribute name.
909

910 911
        Returns:
            bool: True if has this attribute.
912 913

        """
F
fengjiayi 已提交
914 915 916
        return self.desc.has_attr(name)

    def attr_type(self, name):
917
        """
918
        Get the type of attribute by attribute's name.
919

920 921
        Args:
            name(str): the attribute name.
922

923 924
        Returns:
            core.AttrType: the attribute type.
925
        """
F
fengjiayi 已提交
926 927
        return self.desc.attr_type(name)

W
Wu Yi 已提交
928
    def _set_attr(self, name, val):
929 930 931 932 933 934 935 936 937 938
        """
        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
939 940 941 942 943 944 945 946 947 948 949 950 951
        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
952 953
        if isinstance(val, Block):
            self.desc.set_block_attr(name, val.desc)
Y
Yancey1989 已提交
954 955
        elif isinstance(val, list) and val and all(
                isinstance(v, Block) for v in val):
956
            self.desc.set_blocks_attr(name, [v.desc for v in val])
Q
Qiyang Min 已提交
957 958 959 960
        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
961
            self.desc._set_attr(name, val)
Y
yuyang18 已提交
962

F
fengjiayi 已提交
963 964 965 966 967
    @property
    def attr_names(self):
        return self.desc.attr_names()

    def attr(self, name):
968
        """
969 970
        Get the attribute by name.

971
        Args:
972
            name(str): the attribute name.
973

974 975
        Returns:
            bool|int|str|float|list: The attribute value. The return value
976 977
            can be any valid attribute type.
        """
F
fengjiayi 已提交
978
        return self.desc.attr(name)
Y
Yu Yang 已提交
979

W
Wu Yi 已提交
980
    def _block_attr_id(self, name):
981
        """
G
gongweibao 已提交
982
        Get the block attribute's id by name.
983

984 985
        Args:
            name(str): the attribute name.
986

987 988
        Returns:
            int: the block index.
989
        """
W
Wu Yi 已提交
990
        return self.desc._block_attr_id(name)
G
gongweibao 已提交
991

W
Wu Yi 已提交
992
    def _block_attr(self, name):
G
gongweibao 已提交
993 994 995 996 997 998 999 1000 1001 1002
        """
        Get the block attribute  by name.

        Args:
            name(str): the attribute name.

        Returns:
            block: the block attribute.
        """

W
Wu Yi 已提交
1003
        id = self._block_attr_id(name)
G
gongweibao 已提交
1004 1005 1006
        assert (id >= 0 and id < len(self.block.program.blocks))
        return self.block.program.blocks[id]

W
Wu Yi 已提交
1007
    def _blocks_attr(self, name):
G
gongweibao 已提交
1008 1009 1010 1011 1012 1013 1014 1015 1016 1017
        """
        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
1018
        for i in self._blocks_attr_ids(name):
G
gongweibao 已提交
1019 1020 1021 1022 1023
            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
1024
    def _blocks_attr_ids(self, name):
G
gongweibao 已提交
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
        """
        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
1035
        return self.desc._blocks_attr_ids(name)
Y
Yu Yang 已提交
1036

J
JiayiFeng 已提交
1037
    def all_attrs(self):
F
fengjiayi 已提交
1038
        """
1039 1040 1041
        Get the attribute dict.

        Returns:
G
gongweibao 已提交
1042
            dict: The Operator's attribute dict, name->attr.
F
fengjiayi 已提交
1043 1044 1045 1046
        """
        attr_names = self.attr_names
        attr_map = {}
        for n in attr_names:
G
gongweibao 已提交
1047 1048
            attr_type = self.desc.attr_type(n)
            if attr_type == core.AttrType.BLOCK:
W
Wu Yi 已提交
1049
                attr_map[n] = self._block_attr(n)
G
gongweibao 已提交
1050 1051 1052
                continue

            if attr_type == core.AttrType.BLOCKS:
W
Wu Yi 已提交
1053
                attr_map[n] = self._blocks_attr(n)
G
gongweibao 已提交
1054 1055 1056 1057
                continue

            attr_map[n] = self.attr(n)

F
fengjiayi 已提交
1058 1059
        return attr_map

Y
Yu Yang 已提交
1060

Y
Yu Yang 已提交
1061
class Block(object):
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
    """
    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
1076
        use `Program._create_block()` to create a block.
1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090

    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
1091
    def __init__(self, program, idx):
Y
Yu Yang 已提交
1092
        self.desc = program.desc.block(idx)
1093
        self.vars = collections.OrderedDict()  # var_name --> var
Q
qiaolongfei 已提交
1094
        self.ops = list()  # operator list
Y
Yu Yang 已提交
1095
        self.program = program
1096
        self.removed_vars = collections.OrderedDict()
Y
Yu Yang 已提交
1097

1098
    def __str__(self):
Y
Yang Yang(Tony) 已提交
1099 1100
        return self.to_string(True)

F
fengjiayi 已提交
1101 1102
    def to_string(self, throw_on_error, with_details=False):
        """
1103 1104
        Get debug string.

F
fengjiayi 已提交
1105 1106
        Args:
            throw_on_error(bool): raise exception when self is not initialized
1107
                when throw_on_error is True.
F
update  
fengjiayi 已提交
1108
            with_details(bool): more details about variables and parameters
1109 1110
                (e.g. trainable, optimize_attr, ...) will be printed when
                with_details is True. Default False.
F
fengjiayi 已提交
1111

1112 1113
        Returns:
            str: The debug string.
F
fengjiayi 已提交
1114 1115 1116 1117
        """
        assert isinstance(throw_on_error, bool) and isinstance(with_details,
                                                               bool)
        if with_details:
F
fengjiayi 已提交
1118
            re_add_indent = re.compile(r"\n(.)")
F
fengjiayi 已提交
1119 1120
            res_str = "blocks {\n  idx: %d\n  parent_idx: %d" % (
                self.idx, self.parent_idx)
1121
            for var in list(self.vars.values()):
F
fengjiayi 已提交
1122
                res_str += "\n  vars {\n    %s  }" % re_add_indent.sub(
F
update  
fengjiayi 已提交
1123
                    r"\n    \1", var.to_string(throw_on_error, with_details))
F
fengjiayi 已提交
1124
            for op in self.ops:
F
fengjiayi 已提交
1125 1126
                res_str += "\n  ops {\n    %s  }" % re_add_indent.sub(
                    r"\n    \1", op.to_string(throw_on_error))
F
fengjiayi 已提交
1127 1128 1129
            res_str += "\n}"
        else:
            protostr = self.desc.serialize_to_string()
1130 1131
            proto = framework_pb2.BlockDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
1132 1133
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
1134 1135 1136

    __repr__ = __str__

Y
Yu Yang 已提交
1137 1138
    @property
    def parent_idx(self):
Y
Yu Yang 已提交
1139
        return self.desc.parent
Y
Yu Yang 已提交
1140

Y
Yu Yang 已提交
1141 1142 1143 1144
    @property
    def forward_block_idx(self):
        return self.desc.get_forward_block_idx()

W
Wu Yi 已提交
1145
    def _set_forward_block_idx(self, idx):
1146 1147 1148 1149 1150 1151 1152 1153 1154
        """
        Set the forward block Idx.

        Args:
            idx(int): the block index.

        Returns:
            None
        """
W
Wu Yi 已提交
1155
        self.desc._set_forward_block_idx(idx)
Y
Yu Yang 已提交
1156

Y
Yu Yang 已提交
1157 1158
    @property
    def idx(self):
Y
Yu Yang 已提交
1159
        return self.desc.id
Y
Yu Yang 已提交
1160

Q
Qiao Longfei 已提交
1161
    def var(self, name):
1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174
        """
        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.
        """
1175
        if not isinstance(name, six.string_types):
M
minqiyang 已提交
1176 1177 1178
            raise TypeError(
                "var require string as parameter, but get %s instead." %
                (type(name)))
Y
Yu Yang 已提交
1179 1180
        v = self.vars.get(name, None)
        if v is None:
Q
Qiao Longfei 已提交
1181
            raise ValueError("var %s not in this block" % name)
Y
Yu Yang 已提交
1182
        return v
Q
Qiao Longfei 已提交
1183

X
Xin Pan 已提交
1184
    def _find_var_recursive(self, name):
1185 1186 1187 1188 1189 1190 1191
        """
        Get a Variable by name from this block recursively.

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

        Returns:
X
Xin Pan 已提交
1192
            Variable: the Variable with the giving name. Or None if not found.
1193
        """
Y
Yu Yang 已提交
1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217
        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
1218
        return None
Y
Yu Yang 已提交
1219

X
Xin Pan 已提交
1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
    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
1239

Q
Qiao Longfei 已提交
1240
    def all_parameters(self):
1241
        return list(self.iter_parameters())
1242

1243
    def iter_parameters(self):
M
minqiyang 已提交
1244
        return (item[1] for item in six.iteritems(self.vars)
1245
                if isinstance(item[1], Parameter))
Q
Qiao Longfei 已提交
1246

Y
Yu Yang 已提交
1247
    def create_var(self, *args, **kwargs):
1248
        var = Variable(block=self, *args, **kwargs)
1249 1250
        if 'initializer' in kwargs:
            kwargs['initializer'](var, self)
Q
Qiao Longfei 已提交
1251
        return var
Y
Yu Yang 已提交
1252

Q
Qiao Longfei 已提交
1253 1254 1255
    def has_var(self, name):
        return name in self.vars

W
Wu Yi 已提交
1256
    def _rename_var(self, name, new_name):
T
typhoonzero 已提交
1257 1258
        """
        Rename variable in vars and ops' inputs and outputs
1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270

        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
1271
        """
M
minqiyang 已提交
1272 1273
        name = cpt.to_text(name)
        new_name = cpt.to_text(new_name)
M
minqiyang 已提交
1274

T
typhoonzero 已提交
1275
        if not self.has_var(name):
1276
            raise ValueError("var %s is not in current block" % name)
T
wip  
typhoonzero 已提交
1277 1278
        v = self.var(name)
        if type(v) == Parameter:
T
typhoonzero 已提交
1279
            var_type = "Parameter"
T
wip  
typhoonzero 已提交
1280 1281 1282 1283 1284 1285 1286
            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
1287
            var_type = "Variable"
T
wip  
typhoonzero 已提交
1288 1289 1290 1291
            error_clip = v.error_clip
            stop_gradient = v.stop_gradient
        else:
            raise ValueError("unsupported var type: %s", type(v))
T
typhoonzero 已提交
1292
        orig_var_type = v.type
M
minqiyang 已提交
1293
        self.desc._rename_var(cpt.to_bytes(name), cpt.to_bytes(new_name))
W
Wu Yi 已提交
1294
        # NOTE: v is destroyed by C++ after calling _rename_var.
M
minqiyang 已提交
1295
        d = self.desc.find_var(cpt.to_bytes(new_name))
T
typhoonzero 已提交
1296
        if var_type == "Parameter":
T
wip  
typhoonzero 已提交
1297 1298 1299 1300
            var = Parameter(
                self,
                d.shape(),
                d.dtype(),
T
typhoonzero 已提交
1301
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1302 1303 1304 1305 1306 1307 1308
                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
1309
        elif var_type == "Variable":
T
wip  
typhoonzero 已提交
1310 1311
            var = Variable(
                self,
T
typhoonzero 已提交
1312
                type=orig_var_type,
T
wip  
typhoonzero 已提交
1313 1314 1315 1316
                name=new_name,
                error_clip=error_clip,
                stop_gradient=stop_gradient)

W
Wu Yi 已提交
1317
        # rename the python side, _sync_with_cpp will only add
T
wip  
typhoonzero 已提交
1318 1319 1320
        # new vars/ops to python side.
        self.vars[new_name] = var
        del self.vars[name]
W
Wu Yi 已提交
1321
        self._sync_with_cpp()
1322
        return var
T
typhoonzero 已提交
1323

W
Wu Yi 已提交
1324 1325
    def _remove_var(self, name):
        self._sync_with_cpp()
M
minqiyang 已提交
1326
        self.desc._remove_var(cpt.to_bytes(name))
1327 1328
        del self.vars[name]

Y
Yu Yang 已提交
1329 1330
    def create_parameter(self, *args, **kwargs):
        global_block = self.program.global_block()
Q
Qiao Longfei 已提交
1331
        param = Parameter(global_block, *args, **kwargs)
1332
        if 'initializer' in kwargs:
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352

            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
1353
        return param
Y
Yu Yang 已提交
1354

Y
Yu Yang 已提交
1355
    def append_op(self, *args, **kwargs):
1356 1357 1358 1359 1360 1361
        """
        Appends a new Operator according to the giving arguments.

        Returns:
            Operator: the append Operator.
        """
M
minqiyang 已提交
1362
        if _in_imperative_mode():
1363 1364 1365 1366 1367 1368 1369 1370
            op = Operator(
                block=self,
                desc=None,
                type=kwargs.get("type", None),
                inputs=kwargs.get("inputs", None),
                outputs=kwargs.get("outputs", None),
                attrs=kwargs.get("attrs", None))

M
minqiyang 已提交
1371 1372 1373 1374
            # record ops in tracer rather than blocks
            #
            # TODO(minqiyang): add op stop_gradient support in static mode too.
            # currently, we only support stop_gradient in imperative mode.
M
minqiyang 已提交
1375 1376 1377
            _imperative_tracer().trace_op(op,
                                          kwargs.get("stop_gradient", False))
        else:
1378 1379 1380 1381 1382 1383 1384 1385 1386
            op_desc = self.desc.append_op()
            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))

M
minqiyang 已提交
1387
            self.ops.append(op)
M
minqiyang 已提交
1388

1389 1390
        return op

W
Wu Yi 已提交
1391
    def _insert_op(self, index, *args, **kwargs):
1392 1393 1394 1395 1396 1397 1398 1399 1400
        """
        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
1401 1402
        self._sync_with_cpp()
        op_desc = self.desc._insert_op(index)
Q
qiaolongfei 已提交
1403 1404 1405 1406
        op = Operator(block=self, desc=op_desc, *args, **kwargs)
        self.ops.insert(index, op)
        return op

W
Wu Yi 已提交
1407
    def _remove_op(self, index):
1408 1409 1410 1411 1412 1413 1414 1415 1416
        """
        Remove the specific position operator.

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

        Returns:
            None
        """
W
Wu Yi 已提交
1417 1418
        self._sync_with_cpp()
        self.desc._remove_op(index, index + 1)
1419 1420
        del self.ops[index]

W
Wu Yi 已提交
1421
    def _slice_ops(self, start, end):
1422 1423 1424 1425 1426 1427 1428 1429 1430 1431
        """
        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
1432
        return self.ops[start:end]
Y
Yancey1989 已提交
1433

W
Wu Yi 已提交
1434
    def _prepend_op(self, *args, **kwargs):
M
minqiyang 已提交
1435
        if _in_imperative_mode():
1436 1437 1438 1439 1440 1441 1442
            op = Operator(
                self,
                None,
                type=kwargs.get("type", None),
                inputs=kwargs.get("inputs", None),
                outputs=kwargs.get("outputs", None),
                attrs=kwargs.get("attrs", None))
M
minqiyang 已提交
1443 1444 1445
            _imperative_tracer().trace_op(op,
                                          kwargs.get("stop_gradient", False))
        else:
1446 1447 1448 1449 1450 1451 1452 1453
            op_desc = self.desc._prepend_op()
            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))
M
minqiyang 已提交
1454
            self.ops.insert(0, op)
1455

Y
Yu Yang 已提交
1456 1457
        return op

W
Wu Yi 已提交
1458
    def _sync_with_cpp(self):
1459
        """
1460 1461
        Sync from the desc on the c++ end. This method is used to synchronize
        the c++ desc instance generated by backward.
1462
        """
Q
Qiao Longfei 已提交
1463 1464 1465 1466 1467
        # 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())

1468
        # sync variables removed from c++ end
1469
        for var in list(self.vars.keys()):
M
minqiyang 已提交
1470
            if not self.desc.find_var(cpt.to_bytes(var)):
1471 1472
                self.vars.pop(var)

Q
Qiao Longfei 已提交
1473
        # sync operators from cpp
1474 1475 1476 1477
        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
1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493
        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
1494 1495 1496 1497 1498

        # 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
1499
            self.ops.insert(0, op)
Q
Qiao Longfei 已提交
1500 1501 1502 1503 1504 1505 1506

        # 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)

1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519
        # 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
1520 1521 1522 1523
        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
1524
    def _copy_param_info_from(self, other):
1525
        """
1526 1527
        Copy the information of parameters from the other block.

1528
        Args:
1529 1530 1531 1532 1533
            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.
1534 1535 1536 1537 1538

        Returns:
            None
        """
        if not isinstance(other, Block):
W
Wu Yi 已提交
1539 1540
            raise TypeError(
                "_copy_param_info_from should be invoked with Block")
1541
        for p in other.iter_parameters():
1542 1543 1544
            assert isinstance(p, Parameter)
            v = self.vars.get(p.name, None)
            if v is None:
W
Wu Yi 已提交
1545
                raise ValueError("_copy_param_info_from should be invoked with "
1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
                                 "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
1558
                gradient_clip_attr=p.gradient_clip_attr,
F
fengjiayi 已提交
1559
                error_clip=p.error_clip,
1560 1561 1562
                name=v.name)
            self.vars[new_p.name] = new_p

W
Wu Yi 已提交
1563
    def _clone_variable(self, var):
1564 1565
        """
        Clone a variable into current block.
1566

1567 1568 1569 1570
        Args:
            var: the variable to be cloned.

        Returns:
1571
            Variable: the new  variable cloned from 'var' in current block.
1572 1573
        """
        assert isinstance(var, Variable)
T
update  
typhoonzero 已提交
1574 1575 1576 1577 1578
        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
1579 1580
        elif var.type == core.VarDesc.VarType.RAW:
            ret_var = self.create_var(
T
tangwei12 已提交
1581
                name=var.name, persistable=var.persistable, type=var.type)
T
typhoonzero 已提交
1582 1583 1584 1585 1586 1587
        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
1588 1589
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1590 1591 1592 1593 1594 1595 1596
        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
1597 1598
                persistable=True,
                is_data=var.is_data)
T
update  
typhoonzero 已提交
1599
        return ret_var
1600

Y
Yu Yang 已提交
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 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696
class IrNode(object):
    """
    Python IrNode. Beneath it is a core.Node, which is used for Ir Pass.
    """

    def __init__(self, node):
        """
        Construct an IrNode using core.Node.

        Args:
            node(core.Node): C++ Node.
        """
        assert isinstance(node,
                          core.Node), 'node must be the instance of core.Node.'
        self.node = node

    def name(self):
        """
        Return the node name.

        Returns:
            str: node name.
        """
        return self.node.name()

    def node_type(self):
        """
        Return the node type.

        Returns:
            core.Node.Type: node type(core.Node.Type.Operation or core.Node.Type.Variable).
        """
        return self.node.node_type()

    def var(self):
        """
        Return the node variable description.

        Returns:
            core.VarDesc: node variable description.
        """
        return self.node.var()

    def op(self):
        """
        Return the node operator description.

        Returns:
            core.OpDesc: node operator description.
        """
        return self.node.op()

    def id(self):
        """
        Return the node id.

        Returns:
            int: node id.
        """
        return self.node.id()

    def is_op(self):
        """
        If the node is an operator, then return true.

        Returns:
            bool: indicate whether the node is an operator.
        """
        return self.node.is_op()

    def is_var(self):
        """
        If the node is a variable, then return true.

        Returns:
            bool: indicate whether the node is a variable.
        """
        return self.node.is_var()

    def is_ctrl_var(self):
        """
        If the node is a control dependence variable, then return true.

        Returns:
            bool: indicate whether the node is a control dependence variable.
        """
        return self.node.is_ctrl_var()

    def clear_inputs(self):
        """
        Clear the node inputs. After executing the `clear_inputs` function,
        the node inputs will be empty.
        """
        self.node.clear_inputs()

1697
    def remove_input_by_id(self, node_id):
1698 1699 1700 1701 1702 1703
        """
        Remove a node from inputs by the given node id.

        Args:
            node_id(int): the given node id.
        """
1704
        self.node.remove_input(node_id)
1705

1706
    def remove_input(self, node):
1707 1708 1709 1710
        """
        Remove a node from inputs.

        Args:
1711
            node(IrNode): the node being removed.
1712
        """
1713
        self.node.remove_input(node.node)
1714

1715
    def append_input(self, node):
1716 1717 1718 1719
        """
        Append a node in inputs.

        Args:
1720
            node(IrNode): the node being appended.
1721
        """
1722
        self.node.append_input(node.node)
1723 1724 1725 1726 1727 1728 1729 1730

    def clear_outputs(self):
        """
        Clear the node outputs. After executing the `clear_outputs` function,
        the node outputs will be empty.
        """
        self.node.clear_outputs()

1731
    def remove_output_by_id(self, node_id):
1732 1733 1734 1735 1736 1737
        """
        Remove a node from outputs by the given node id.

        Args:
            node_id(int): the given node id.
        """
1738
        self.node.remove_output(node_id)
1739

1740
    def remove_output(self, node):
1741 1742 1743 1744
        """
        Remove a node from outputs.

        Args:
1745
            node(IrNode): the node being removed.
1746
        """
1747
        self.node.remove_output(node.node)
1748

1749
    def append_output(self, node):
1750 1751 1752 1753
        """
        Append a node in outputs.

        Args:
1754
            node(IrNode): the node being appended.
1755
        """
1756
        self.node.append_output(node.node)
1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817

    @property
    def inputs(self):
        """
        Return the node inputs.

        Returns:
            list(IrNode): node inputs wrapped by IrNode.
        """
        return [IrNode(n) for n in self.node.inputs]

    @property
    def outputs(self):
        """
        Return the node outputs.

        Returns:
            list(IrNode): node outputs wrapped by IrNode.
        """
        return [IrNode(n) for n in self.node.outputs]


class IrVarNode(IrNode):
    """
    Python IrVarNode. Beneath it is a core.Node, it inherits from IrNode.
    """

    def __init__(self, node):
        """
        Construct an IrVarNode using core.Node.

        Args:
            node(core.Node): C++ Node.
        """
        assert isinstance(node, core.Node) and node.is_var(), \
            'node must be the instance of core.Node and it must be a variable node.'
        super(IrVarNode, self).__init__(node)
        self.node = node

    def set_shape(self, shape):
        """
        Set the node variable shape.

        Args:
            shape(list): shape to be set.
        """
        assert self.node.var() is not None, \
            "The node variable description cannot be None."
        self.node.var().set_shape(shape)

    def persistable(self):
        """
        If the variable node is a persistable variable, then return true.

        Returns:
            bool: indicate whether the variable is persistable.
        """
        assert self.node.var() is not None, \
            "The node variable description cannot be None."
        return self.node.var().persistable()

1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850
    def type(self):
        """
        Return the variable type.

        Returns:
            core.VarDesc.VarType: the variable type.
        """
        assert self.node.var() is not None, \
            "The node variable description cannot be None."
        return self.node.var().type()

    def dtype(self):
        """
        Return the variable data type.

        Returns:
            core.VarDesc.VarType: the variable data type.
        """
        assert self.node.var() is not None, \
            "The node variable description cannot be None."
        return self.node.var().dtype()

    def shape(self):
        """
        Return the variable shape.

        Returns:
            list: the variable shape.
        """
        assert self.node.var() is not None, \
            "The node variable description cannot be None."
        return self.node.var().shape()

1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900
    @property
    def inputs(self):
        """
        Return the node inputs.

        Returns:
            list(IrOpNode): node inputs wrapped by IrOpNode.
        """
        return [IrOpNode(n) for n in self.node.inputs]

    @property
    def outputs(self):
        """
        Return the node outputs.

        Returns:
            list(IrOpNode): node outputs wrapped by IrOpNode.
        """
        return [IrOpNode(n) for n in self.node.outputs]


class IrOpNode(IrNode):
    """
    Python IrOpNode. Beneath it is a core.Node, it inherits from IrNode.
    """

    def __init__(self, node):
        """
        Construct an IrOpNode using core.Node.

        Args:
            node(core.Node): C++ Node.
        """
        assert isinstance(node, core.Node) and node.is_op(), \
            'node must be the instance of core.Node and it must be a operator node.'
        super(IrOpNode, self).__init__(node)
        self.node = node

    def rename_input(self, old_input_name, new_input_name):
        """
        Rename the input of this node.

        Args:
            old_input_name(str): the old input name.
            new_input_name(str): the new input name.
        """
        assert self.node.op() is not None, \
            "The node operator description cannot be None."
        self.node.op()._rename_input(old_input_name, new_input_name)

1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 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
    def input(self, name):
        """
        Get the argument name list by the parameter name for input.

        Args:
            name(str): the parameter name.

        Returns:
            list(str): the argument name list.
        """
        assert self.node.op() is not None, \
            "The node operator description cannot be None."
        return self.node.op().input(name)

    def output(self, name):
        """
        Get the argument name list by the parameter name for output.

        Args:
            name(str): the parameter name.

        Returns:
            list(str): the argument name list.
        """
        assert self.node.op() is not None, \
            "The node operator description cannot be None."
        return self.node.op().output(name)

    def set_type(self, new_type):
        """
        Change the operator type into new type.

        Args:
            new_type(str): new operator type to be set.
        """
        assert self.node.op() is not None, \
            "The node operator description cannot be None."
        return self.node.op().set_type(new_type)

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 1965 1966 1967
    def set_attr(self, name, val):
        """
        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.
        """
        self._update_desc_attr(name, val)

    def _update_desc_attr(self, name, val):
        """
        Update the value of the op desc's attribute by attribute's name.
        """
        assert self.node.op() is not None, \
            "The node operator description cannot be None."
        desc = self.node.op()
        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)

1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
    @property
    def inputs(self):
        """
        Return the node inputs.

        Returns:
            list(IrVarNode): node inputs wrapped by IrVarNode.
        """
        return [IrVarNode(n) for n in self.node.inputs]

    @property
    def outputs(self):
        """
        Return the node outputs.

        Returns:
            list(IrVarNode): node outputs wrapped by IrVarNode.
        """
        return [IrVarNode(n) for n in self.node.outputs]


1989 1990
class IrGraph(object):
    """
1991
    Python IrGraph. Beneath it is a core.Graph, which is used for
1992
    creating a c++ Ir Pass Graph. An IrGraph is just a graph view of
1993 1994
    a Program. In an IrGraph, both Variables and Operators are graph
    nodes.
1995 1996 1997 1998
    """

    def __init__(self, graph, for_test=False):
        """
1999 2000
        Construct an IrGraph using core.Graph.

2001 2002 2003 2004 2005 2006 2007 2008 2009
        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

2010 2011 2012 2013
    def clone(self):
        """
        Create a new and duplicated IrGraph.

2014 2015 2016
        Warns:
            The method only clones the graph structure, not its attributes.

2017 2018 2019
        Returns:
            IrGraph: A new and duplicated graph.
        """
2020
        g = self.graph.clone()
2021 2022
        return IrGraph(g, self._for_test)

2023
    def is_test(self):
2024 2025 2026
        """
        If the graph is used for testing, the function returns true. Otherwise, returns false.
        """
2027 2028
        return self._for_test

W
WangZhen 已提交
2029
    def all_nodes(self):
2030 2031 2032
        """
        Return all nodes included in the graph as a set.
        """
2033
        return {IrNode(node) for node in self.graph.nodes()}
2034

2035
    def all_var_nodes(self):
2036 2037 2038
        """
        Return all variable nodes included in the graph as a set.
        """
2039
        return {IrVarNode(node) for node in self.graph.nodes() if node.is_var()}
2040

2041
    def all_persistable_nodes(self):
2042 2043 2044
        """
        Return all persistable variable nodes included in the graph as a set.
        """
W
WangZhen 已提交
2045 2046 2047 2048 2049
        persistable_nodes = set()
        for node in self.graph.nodes():
            if node.is_var() and node.var() is not None and node.var(
            ).persistable():
                persistable_nodes.add(node)
2050
        return {IrVarNode(p) for p in persistable_nodes}
W
WangZhen 已提交
2051

2052
    def all_op_nodes(self):
2053 2054 2055
        """
        Return all operator nodes included in the graph as a set.
        """
2056
        return {IrOpNode(node) for node in self.graph.nodes() if node.is_op()}
2057

W
WangZhen 已提交
2058 2059
    def var_node(self, name):
        """
2060 2061
        Get a variable node by name from the graph.

W
WangZhen 已提交
2062 2063
        Args:
            name(str): the name of the variable node.
2064

W
WangZhen 已提交
2065 2066 2067
        Raises:
            ValueError: The If input's type is not str, or this graph
            doesn't have a variable with the giving name.
2068

W
WangZhen 已提交
2069
        Returns:
2070
            IrVarNode: the variable node with the giving name.
W
WangZhen 已提交
2071 2072 2073 2074 2075 2076
        """
        if not isinstance(name, six.string_types):
            raise TypeError(
                "var require string as parameter, but get %s instead." %
                (type(name)))
        target_var_node = None
2077
        var_nodes = self.all_var_nodes()
W
WangZhen 已提交
2078 2079 2080 2081 2082 2083 2084
        for var_node in var_nodes:
            if var_node.name() == name:
                target_var_node = var_node
        if target_var_node is None:
            raise ValueError("var_node %s not in this graph" % name)
        return target_var_node

2085
    def create_persistable_node(self, name, var_type, shape, var_dtype):
2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096
        """
        Create a persistable variable node in the graph. In IrGraph,
        it can not distinguish between persistable variables and parameters.

        Args:
            name(str): the name of the persistable variable node.
            vart_type(core.VarDesc.VarType): the type of the persistable variable node.
            shape(list): the shape of the persistable variable node.
            var_dtype(core.VarDesc.VarType): the data type of the persistable variable node.

        Returns:
2097
            IrVarNode: the created persistable variable node.
2098
        """
2099 2100 2101 2102 2103
        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)
2104
        return IrVarNode(self.graph.create_var_node(var_desc))
2105 2106

    def create_var_node(self, name, var_type, shape, var_dtype):
2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117
        """
        Create a variable node in the graph. The created variable node is
        not persistable.

        Args:
            name(str): the name of the variable node.
            vart_type(core.VarDesc.VarType): the type of the variable node.
            shape(list): the shape of the variable node.
            var_dtype(core.VarDesc.VarType): the data type of the variable node.

        Returns:
2118
            IrVarNode: the created variable node.
2119 2120
        """

2121 2122 2123 2124
        var_desc = core.VarDesc(name)
        var_desc.set_type(var_type)
        var_desc.set_shape(shape)
        var_desc.set_dtype(var_dtype)
2125
        return IrVarNode(self.graph.create_var_node(var_desc))
2126 2127

    def create_var_node_from_desc(self, var_desc):
2128 2129 2130 2131 2132 2133 2134 2135
        """
        Create a variable node by using an existing VarDesc in the graph.
        Depend on the giving VarDesc, the created variable node may be persistable.

        Args:
            var_desc(core.VarDesc): the giving variable description.

        Returns:
2136
            IrVarNode: the created variable node.
2137
        """
2138
        return IrVarNode(self.graph.create_var_node(var_desc))
2139 2140

    def create_op_node(self, op_type, attrs, inputs, outputs):
2141 2142 2143 2144 2145 2146 2147 2148 2149 2150
        """
        Create a operator node in the graph.

        Args:
            op_type(str): the type of the operator node.
            attrs(dict): the attributes of the operator node.
            inputs(dict): the inputs of the operator node.
            outputs(dict): the outpus of the operator node.

        Returns:
2151
            IrOpNode: the created operator node.
2152
        """
2153 2154
        op_desc = core.OpDesc()
        op_desc.set_type(op_type)
2155
        for attr, value in six.iteritems(attrs):
2156
            self._update_desc_attr(op_desc, attr, value)
2157
        for input_name, var_nodes in six.iteritems(inputs):
2158 2159 2160 2161
            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])
2162
        for output_name, var_nodes in six.iteritems(outputs):
2163 2164 2165 2166
            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])
2167
        return IrOpNode(self.graph.create_op_node(op_desc))
2168 2169

    def create_op_node_from_desc(self, op_desc):
2170 2171 2172 2173 2174 2175 2176
        """
        Create a operator node by using an existing OpDesc in the graph.

        Args:
            op_desc(core.VarDesc): the giving operator description.

        Returns:
2177
            IrOpNode: the created operator node.
2178
        """
2179
        return IrOpNode(self.graph.create_op_node(op_desc))
2180 2181

    def update_input_link(self, old_input_node, new_input_node, op_node):
2182 2183 2184 2185
        """
        Update the input's link of a operator node.

        Args:
2186 2187 2188
            old_input_node(IrNode): the old input node of the giving op_node.
            new_input_node(IrNode): the new input node of the giving op_node.
            op_node(IrOpNode): the operator node that is needed to update input's link.
2189
        """
2190 2191
        assert old_input_node.node in self.graph.nodes() and new_input_node.node in \
        self.graph.nodes() and op_node.node in self.graph.nodes(), \
W
WangZhen 已提交
2192
        'The three arguments(old_input_node&new_input_node&op_node) must be in the graph nodes.'
2193 2194 2195 2196
        old_input_node.remove_output(op_node)
        op_node.remove_input(old_input_node)
        new_input_node.append_output(op_node)
        op_node.append_input(new_input_node)
2197
        op_node.rename_input(old_input_node.name(), new_input_node.name())
2198 2199

    def link_to(self, node_in, node_out):
2200 2201 2202 2203
        """
        Connect two nodes.

        Args:
2204 2205
            node_in(IrNode): the input node.
            node_out(IrNode): the output node.
2206
        """
2207
        assert node_in.node in self.graph.nodes() and node_out.node in self.graph.nodes(), \
W
WangZhen 已提交
2208
            'The two arguments(node_in&node_out) must be in the graph nodes.'
2209 2210
        node_in.append_output(node_out)
        node_out.append_input(node_in)
2211 2212

    def safe_remove_nodes(self, remove_nodes):
2213 2214 2215 2216 2217 2218 2219
        """
        Remove nodes safely since links connected to these removed nodes are
        also removed.

        Args:
            remove_nodes(set): the nodes prepared to be removed.
        """
2220
        if not isinstance(remove_nodes, set):
W
WangZhen 已提交
2221 2222 2223 2224
            if isinstance(remove_nodes, Iterable):
                remove_nodes = set(remove_nodes)
            else:
                remove_nodes = {remove_nodes}
2225 2226
        original_nodes = {n.node for n in remove_nodes}
        core.graph_safe_remove_nodes(self.graph, original_nodes)
2227

W
WangZhen 已提交
2228
    def has_circle(self):
2229 2230 2231 2232 2233 2234
        """
        Check if the graph has a circle.

        Returns:
            bool: True if the graph has a circle else False.
        """
W
WangZhen 已提交
2235 2236 2237
        return core.has_circle(self.graph)

    def graph_num(self):
2238 2239 2240 2241 2242 2243
        """
        Count the number of unconnected graphs in this graph.

        Returns:
            int: the number of unconnected graphs.
        """
W
WangZhen 已提交
2244 2245 2246
        return core.graph_num(self.graph)

    def topology_sort(self):
2247 2248 2249 2250 2251 2252
        """
        Perform the topology sort operation on the graph.

        Notes: the `graph` cannot contain a circle.

        Returns:
Z
Zhen Wang 已提交
2253
            list(IrNode): nodes in topology order.
2254
        """
2255
        ordered_nodes = core.topology_sort(self.graph)
Z
Zhen Wang 已提交
2256
        return [IrNode(n) for n in ordered_nodes]
W
WangZhen 已提交
2257 2258

    def build_adjacency_list(self):
2259 2260 2261 2262
        """
        Build an adjacency list of operations for the `graph`.

        Returns:
2263
            dict{IrNode: set(IrNode)}: the adjacency list.
2264
        """
2265 2266 2267 2268 2269
        adj_list = core.build_adjacency_list(self.graph)
        wrapped_adj_list = dict()
        for k, v in six.iteritems(adj_list):
            wrapped_adj_list[IrNode(k)] = {IrNode(n) for n in v}
        return wrapped_adj_list
W
WangZhen 已提交
2270

2271 2272 2273 2274 2275 2276 2277 2278
    def draw(self, save_path, name, marked_nodes=None, remove_ctr_var=True):
        """
        Draw the graph. If `dot` command is installed, the drawn graph
        will be saved as pdf file type, otherwise dot file type is used.

        Args:
            save_path(str): the save path of drawn graph.
            name(str): the name of drawn graph.
2279
            marked_nodes(set(IrNode)): nodes that are needed to be marked.
2280 2281 2282 2283 2284
            Default value is None.
            remove_ctr_var(bool): If it is set True, all control variable nodes
            in the graph will be removed. Default value is True.
        """

2285 2286 2287 2288 2289 2290 2291 2292 2293
        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))

2294
        remove_ctr_vars = set()
2295
        if remove_ctr_var:
2296
            for node in self.all_var_nodes():
2297 2298 2299
                if node.is_ctrl_var():
                    remove_ctr_vars.add(node)
            self.safe_remove_nodes(remove_ctr_vars)
2300 2301
        print('Total ops num = {}.'.format(len(self.all_op_nodes())))

2302 2303
        if marked_nodes is not None:
            if not isinstance(marked_nodes, set):
2304 2305 2306 2307 2308 2309
                if isinstance(marked_nodes, Iterable):
                    marked_nodes = set(marked_nodes)
                else:
                    marked_nodes = {marked_nodes}
            marked_nodes = {n.node for n in marked_nodes}
            remove_ctr_vars = {n.node for n in remove_ctr_vars}
2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320
            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):
2321 2322 2323
        """
        Convert the graph into a Program.

Z
Zhen Wang 已提交
2324
        WARN: When the graph includes backward operator nodes, the
2325 2326 2327 2328 2329 2330
        conversion process may be failed. Usually, this function is
        only used to convert a test graph.

        Returns:
            Program: a program converted from the graph.
        """
2331
        convert_pass = core.get_pass('graph_to_program_pass')
2332 2333
        desc = core.ProgramDesc()
        convert_pass.set_not_owned('program', desc)
2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353
        convert_pass.apply(self.graph)
        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
2354
class Program(object):
D
dzhwinter 已提交
2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365
    """
    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
2366
    default_main_program run in every mini batch and adjust the weights.
D
dzhwinter 已提交
2367 2368

    Returns:
Y
yuyang18 已提交
2369
        A empty program.
D
dzhwinter 已提交
2370 2371

    Examples:
Y
yuyang18 已提交
2372 2373 2374 2375 2376 2377
        >>> 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
2378 2379 2380

    """

2381 2382
    def __init__(self):
        self.desc = core.ProgramDesc()
Y
Yu Yang 已提交
2383 2384
        self.blocks = [Block(self, 0)]
        self.current_block_idx = 0
D
dzhwinter 已提交
2385
        self._seed = 0
Y
yuyang18 已提交
2386
        self._current_role = core.op_proto_and_checker_maker.OpRole.Forward
Y
yuyang18 已提交
2387
        self._op_role_var = []
T
tangwei12 已提交
2388

2389 2390
        # for distribute training
        # _is_distributed = True if under distributed training
T
tangwei12 已提交
2391
        self._is_distributed = False
2392
        # _is_chief = True if the trainer is the first one, usually No.0
T
tangwei12 已提交
2393
        self._is_chief = False
2394 2395 2396
        # _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
2397
        self._endpoints = []
2398 2399 2400
        # if current role is parameter server, the _ps_endpoint is its "ip:port"
        self._ps_endpoint = None
        # trainers_endpoints, it is used for distribution.
2401
        self._trainers_endpoints = []
2402
        # the distributed lookup table names
T
tangwei12 已提交
2403
        self._distributed_lookup_table = None
D
dzhwinter 已提交
2404
        # @deprecated(the python memory optimize transpiler is deprecated)
D
dzhwinter 已提交
2405
        # whether the program is optimized by memory_optimize_transpiler
D
dzhwinter 已提交
2406
        self.__is_mem_optimized = False
D
dzhwinter 已提交
2407 2408

    @property
D
dzhwinter 已提交
2409
    def _is_mem_optimized(self):
D
dzhwinter 已提交
2410 2411
        # if the program is optimized, operator input/outputs
        # maybe same, which conflict with save_inference_model.
D
dzhwinter 已提交
2412
        return self.__is_mem_optimized
D
dzhwinter 已提交
2413

D
dzhwinter 已提交
2414 2415 2416
    @_is_mem_optimized.setter
    def _is_mem_optimized(self, target):
        self.__is_mem_optimized = target
Y
yuyang18 已提交
2417 2418 2419

    @property
    def op_role(self):
Y
yuyang18 已提交
2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432
        """
        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
2433 2434 2435
        return self._current_role

    @op_role.setter
D
dzhwinter 已提交
2436
    def op_role(self, role):
Y
yuyang18 已提交
2437 2438 2439 2440
        self._current_role = role

    @property
    def op_role_var(self):
Y
yuyang18 已提交
2441 2442 2443 2444 2445 2446 2447
        """
        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
2448 2449 2450 2451
        return self._op_role_var

    @op_role_var.setter
    def set_op_role_var(self, var_name):
Y
yuyang18 已提交
2452
        self._op_role_var = [var_name]
Y
yuyang18 已提交
2453

S
rename  
sneaxiy 已提交
2454
    @signature_safe_contextmanager
W
Wu Yi 已提交
2455
    def _optimized_guard(self, param_and_grads):
Y
yuyang18 已提交
2456 2457 2458 2459 2460 2461 2462
        """
        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:
2463
            param_and_grads(list): The variables (names) to be optimized.
Y
yuyang18 已提交
2464 2465 2466 2467

        Examples:

            >>> p, g = backward(...)
W
Wu Yi 已提交
2468
            >>> with program._optimized_guard([p,g]):
Y
yuyang18 已提交
2469 2470
            >>>     p = p - 0.001 * g
        """
X
Xin Pan 已提交
2471 2472 2473
        tmp_role = self._current_role
        tmp_var = self._op_role_var

Y
yuyang18 已提交
2474 2475
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.Optimize
2476 2477 2478 2479
        self._op_role_var = [
            var.name if isinstance(var, Variable) else var
            for var in param_and_grads
        ]
Y
yuyang18 已提交
2480
        yield
X
Xin Pan 已提交
2481 2482
        self._op_role_var = tmp_var
        self._current_role = tmp_role
Y
Yu Yang 已提交
2483

S
rename  
sneaxiy 已提交
2484
    @signature_safe_contextmanager
X
Xin Pan 已提交
2485
    def _lr_schedule_guard(self, is_with_opt=False):
2486 2487 2488 2489 2490 2491 2492
        """
        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
2493 2494 2495 2496
        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.
2497 2498 2499 2500 2501 2502 2503

        Examples:

            >>> p, g = backward(...)
            >>> with program.lr_schedule_guard():
            >>>     lr = lr * decay
        """
2504 2505 2506 2507

        tmp_role = self._current_role
        tmp_var = self._op_role_var

2508 2509
        OpRole = core.op_proto_and_checker_maker.OpRole
        self._current_role = OpRole.LRSched
X
Xin Pan 已提交
2510 2511
        if is_with_opt:
            self._current_role = int(OpRole.LRSched) | int(OpRole.Optimize)
2512 2513 2514
        # TODO(typhoonzero): how to set target learning rate var
        self._op_role_var = []
        yield
2515 2516
        self._op_role_var = tmp_var
        self._current_role = tmp_role
2517

2518
    def __str__(self):
Y
yuyang18 已提交
2519 2520 2521 2522 2523 2524 2525 2526 2527
        """
        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
2528 2529
        return self.to_string(True)

F
fengjiayi 已提交
2530 2531 2532
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
Y
yuyang18 已提交
2533

F
fengjiayi 已提交
2534
        Args:
Y
yuyang18 已提交
2535 2536
            throw_on_error(bool): raise Value error when any of required fields
                is not set.
F
fengjiayi 已提交
2537

Y
yuyang18 已提交
2538 2539 2540 2541
            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
2542 2543
        Returns:
            str : The debug string.
Y
yuyang18 已提交
2544 2545 2546 2547

        Raises:
            ValueError: If any of required fields is not set and throw_on_error is
                True.
F
fengjiayi 已提交
2548 2549 2550 2551 2552 2553 2554 2555 2556 2557

        """
        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()
2558 2559
            proto = framework_pb2.ProgramDesc.FromString(
                six.binary_type(protostr))
F
fengjiayi 已提交
2560 2561
            res_str = _debug_string_(proto, throw_on_error)
        return res_str
2562

W
Wu Yi 已提交
2563
    def _get_desc(self):
Y
yuyang18 已提交
2564 2565 2566 2567 2568 2569 2570
        """
        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.
        """
2571 2572
        return self.desc

X
version  
Xin Pan 已提交
2573 2574 2575
    def _version(self):
        return self.desc._version()

2576
    def clone(self, for_test=False):
Y
yuyang18 已提交
2577 2578 2579
        """
        Create a new, duplicated program.

2580

Y
yuyang18 已提交
2581 2582 2583 2584
        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`.
2585

Y
yuyang18 已提交
2586 2587 2588 2589
        * 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
2590 2591 2592 2593 2594
        :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()
2595 2596

        Args:
Y
yuyang18 已提交
2597 2598
            for_test(bool): True if change the :code:`is_test` attribute of
                operators to :code:`True`.
2599

D
dzhwinter 已提交
2600
        Returns:
Y
yuyang18 已提交
2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653
            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.
2654 2655
        """
        if for_test:
X
Xin Pan 已提交
2656
            p = self._inference_optimize(prune_read_op=False)
2657
        else:
2658
            p = Program()
G
gongweibao 已提交
2659 2660
            p.current_block_idx = self.current_block_idx
            p._seed = self._seed
2661
            p.desc = core.ProgramDesc(self.desc)
M
minqiyang 已提交
2662 2663 2664
            p.blocks = [
                Block(p, i) for i in six.moves.range(self.desc.num_blocks())
            ]
G
gongweibao 已提交
2665 2666 2667 2668

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

W
Wu Yi 已提交
2669
            p._sync_with_cpp()
2670

W
Wu Yi 已提交
2671
        p._copy_param_info_from(self)
W
Wu Yi 已提交
2672
        p._copy_data_info_from(self)
2673
        p._copy_dist_param_info_from(self)
Y
Yu Yang 已提交
2674
        return p
2675

W
Wu Yi 已提交
2676
    def _prune(self, targets):
Y
yuyang18 已提交
2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691
        """
        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.

        """
2692 2693 2694 2695 2696 2697
        if not isinstance(targets, list):
            targets = [targets]
        targets_idx = []
        for t in targets:
            if not isinstance(t, Operator):
                if isinstance(t, Variable):
2698 2699
                    # After transpiler processing, the op that output this
                    # variable maybe has been changed, so t.op is not reliable
2700
                    # and we need to find the current op that generate this
2701 2702 2703 2704 2705 2706 2707 2708
                    # 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

2709
                    t = t.op
2710 2711 2712 2713
                    if t is None:
                        raise ValueError(
                            "The target variable must have an "
                            "associated operator that generates it.")
2714
                else:
2715 2716
                    raise ValueError("All targets of prune() can only be "
                                     "Variable or Operator.")
2717 2718 2719 2720

            targets_idx.append([t.block.idx, t.idx])
        res = Program()
        res.desc = core.prune(self.desc, targets_idx)
M
minqiyang 已提交
2721 2722 2723
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
2724
        res._sync_with_cpp()
2725 2726
        return res

X
Xin Pan 已提交
2727
    def _inference_optimize(self, prune_read_op=True):
Y
yuyang18 已提交
2728
        """
F
fengjiayi 已提交
2729 2730 2731 2732 2733
        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.

2734
        3. change the :code:`is_test`
Y
yuyang18 已提交
2735 2736 2737
        attribute of operators to :code:`True`. All the :code:`Parameter`
        information will be lost.

2738
        Args:
X
Xin Pan 已提交
2739 2740
            prune_read_op(bool): remove the read ops that are added by py_reader
                                 for cpp inference library
2741

Y
yuyang18 已提交
2742 2743 2744 2745 2746 2747
        Notes: This API is a very low level API. Use
        :code:`Program.clone(for_test=True)` instead.

        Returns:
            Program: The new program.
        """
2748
        res = Program()
2749
        res.desc = core.ProgramDesc(self.desc)
F
fengjiayi 已提交
2750 2751 2752 2753

        # 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
2754
        if prune_read_op:
2755 2756 2757 2758 2759 2760 2761 2762 2763
            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
2764
                    root_block._remove_var(cpt.to_bytes(var.name()))
F
fengjiayi 已提交
2765 2766

        # change all `is_test` attributes to True
M
minqiyang 已提交
2767
        for i in six.moves.range(res.desc.num_blocks()):
2768
            block = res.desc.block(i)
M
minqiyang 已提交
2769
            for j in six.moves.range(block.op_size()):
2770 2771
                op = block.op(j)
                if op.has_attr('is_test'):
W
Wu Yi 已提交
2772
                    op._set_attr('is_test', True)
M
minqiyang 已提交
2773 2774 2775
        res.blocks = [
            Block(res, i) for i in six.moves.range(res.desc.num_blocks())
        ]
W
Wu Yi 已提交
2776
        res._sync_with_cpp()
2777 2778
        return res

2779 2780
    @staticmethod
    def parse_from_string(binary_str):
Y
yuyang18 已提交
2781 2782 2783 2784 2785 2786 2787
        """
        Deserialize a program desc from protobuf binary string.

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

        Args:
2788
            binary_str_type(str): The binary prootbuf string.
Y
yuyang18 已提交
2789 2790 2791 2792

        Returns:
            Program: A deserialized program desc.
        """
2793 2794
        p = Program()
        p.desc = core.ProgramDesc(binary_str)
M
minqiyang 已提交
2795
        p.blocks = [Block(p, i) for i in six.moves.range(p.desc.num_blocks())]
W
Wu Yi 已提交
2796
        p._sync_with_cpp()
2797
        return p
Y
Yu Yang 已提交
2798

2799
    @staticmethod
2800
    def _construct_from_desc(desc):
2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815
        """
        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
2816 2817
    @property
    def random_seed(self):
Y
yuyang18 已提交
2818 2819 2820 2821 2822 2823
        """
        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
2824 2825
        return self._seed

Q
qiaolongfei 已提交
2826 2827
    @property
    def num_blocks(self):
Y
yuyang18 已提交
2828 2829 2830
        """
        The number of blocks in this program.
        """
Q
qiaolongfei 已提交
2831 2832
        return self.desc.num_blocks()

D
dzhwinter 已提交
2833 2834 2835 2836 2837 2838
    @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
2839
    def __repr__(self):
2840
        return self.__str__()
2841

Y
Yu Yang 已提交
2842
    def global_block(self):
Y
yuyang18 已提交
2843 2844 2845
        """
        Get the first block of this program.
        """
Y
Yu Yang 已提交
2846 2847
        return self.blocks[0]

Q
Qiao Longfei 已提交
2848
    def block(self, index):
Y
yuyang18 已提交
2849 2850 2851 2852 2853 2854 2855 2856
        """
        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
2857 2858
        return self.blocks[index]

Y
Yu Yang 已提交
2859
    def current_block(self):
Y
yuyang18 已提交
2860 2861 2862 2863
        """
        Get the current block. The :code:`current` block is the block to append
        operators.
        """
Y
Yu Yang 已提交
2864 2865
        return self.blocks[self.current_block_idx]

W
Wu Yi 已提交
2866
    def _create_block(self, parent_idx=None):
Y
yuyang18 已提交
2867 2868 2869 2870 2871 2872 2873 2874 2875 2876
        """
        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
2877
        new_block_idx = len(self.blocks)
F
update  
fengjiayi 已提交
2878 2879 2880
        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
2881 2882 2883 2884
        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
2885
    def _rollback(self):
Y
yuyang18 已提交
2886 2887 2888 2889 2890
        """
        Exit a code block, i.e., roll back to the parent block.
        Returns:
            None
        """
Y
Yu Yang 已提交
2891 2892
        self.current_block_idx = self.current_block().parent_idx

W
Wu Yi 已提交
2893
    def _sync_with_cpp(self):
Y
yuyang18 已提交
2894 2895 2896 2897 2898 2899 2900 2901 2902 2903
        """
        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
2904 2905 2906
        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
2907
            block._sync_with_cpp()
Q
Qiao Longfei 已提交
2908

W
Wu Yi 已提交
2909
    def _copy_param_info_from(self, other):
2910
        """
2911
        Copy the information of parameters from other program.
D
dzhwinter 已提交
2912

Y
yuyang18 已提交
2913 2914 2915
        Notes: This is a very low level API. Users should not invoke it
        directly.

2916 2917 2918 2919 2920 2921 2922
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
2923
            raise TypeError("_copy_param_info_from should be invoked with "
2924 2925 2926
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
2927
            raise ValueError("_copy_param_info_from should be invoked with two "
2928
                             "program, with represent the same topology")
W
Wu Yi 已提交
2929
        self.global_block()._copy_param_info_from(other.global_block())
2930

2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945
    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
2946
        self._parameters_on_pservers = other._parameters_on_pservers
2947
        self._endpoints = other._endpoints
2948
        self._ps_endpoint = other._ps_endpoint
2949 2950
        self._distributed_lookup_table = other._distributed_lookup_table

W
Wu Yi 已提交
2951
    def _copy_data_info_from(self, other):
F
fengjiayi 已提交
2952 2953
        """
        Copy the information of data variables from other program.
D
dzhwinter 已提交
2954

Y
yuyang18 已提交
2955 2956 2957
        Notes: This is a very low level API. Users should not invoke it
        directly.

F
fengjiayi 已提交
2958 2959 2960 2961 2962 2963 2964
        Args:
            other(Program): Other program

        Returns:
            None
        """
        if not isinstance(other, Program):
W
Wu Yi 已提交
2965
            raise TypeError("_copy_param_info_from should be invoked with "
F
fengjiayi 已提交
2966 2967 2968
                            "Program")

        if len(self.blocks) != len(other.blocks):
W
Wu Yi 已提交
2969
            raise ValueError("_copy_param_info_from should be invoked with two "
F
fengjiayi 已提交
2970
                             "program, with represent the same topology")
2971
        for var in list(other.global_block().vars.values()):
F
fengjiayi 已提交
2972 2973 2974
            if var.is_data:
                self.global_block().var(var.name).is_data = True

2975
    def list_vars(self):
Y
yuyang18 已提交
2976 2977 2978 2979 2980 2981
        """
        Get all variables from this Program. A iterable object is returned.

        Returns:
            iterable: The generator will yield every variable in this program.
        """
2982
        for each_block in self.blocks:
2983
            for each_var in list(each_block.vars.values()):
2984 2985
                yield each_var

Y
Yu Yang 已提交
2986

Y
Yu Yang 已提交
2987
class Parameter(Variable):
2988
    """
2989
    Parameter is derived from Variable. A parameter is a persistable
2990
    Variable, and will be updated by optimizers after each iteration.
2991
    The training of a neural network is essentially the updating of
2992 2993
    its parameters.

2994
    Relative to a general Variable, a Parameter has several its own
2995 2996
    member variables:

2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008
    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.
3009 3010
    """

Y
Yu Yang 已提交
3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
    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")
3021 3022 3023

        Variable.__init__(
            self, block, persistable=True, shape=shape, dtype=dtype, **kwargs)
Y
Yu Yang 已提交
3024 3025 3026 3027
        self.trainable = kwargs.get('trainable', True)

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

3028 3029
        self.regularizer = kwargs.get('regularizer', None)

F
fengjiayi 已提交
3030
        self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None)
Y
Yu Yang 已提交
3031

W
wanghaoshuang 已提交
3032
        self.do_model_average = kwargs.get('do_model_average', None)
W
wanghaoshuang 已提交
3033

F
fengjiayi 已提交
3034 3035 3036
    def __str__(self):
        return self.to_string(True)

F
update  
fengjiayi 已提交
3037 3038 3039
    def to_string(self, throw_on_error, with_details=False):
        """
        To debug string.
D
dzhwinter 已提交
3040

F
update  
fengjiayi 已提交
3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054
        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
3055
                               "gradient_clip_attr", "do_model_average")
F
update  
fengjiayi 已提交
3056
            for attr_name in additional_attr:
3057 3058
                res_str += "%s: %s\n" % (
                    attr_name, six.binary_type(getattr(self, attr_name)))
F
update  
fengjiayi 已提交
3059 3060
        else:
            res_str = Variable.to_string(self, throw_on_error, False)
F
fengjiayi 已提交
3061 3062 3063 3064
        return res_str

    __repr__ = __str__

Y
Yu Yang 已提交
3065

Y
Yu Yang 已提交
3066
# program is a global instance.
Y
Yu Yang 已提交
3067 3068
_main_program_ = Program()
_startup_program_ = Program()
3069

3070

3071
def default_startup_program():
Y
Yu Yang 已提交
3072
    """
Y
yuyang18 已提交
3073 3074 3075 3076 3077 3078 3079 3080 3081
    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.
3082

Y
Yu Yang 已提交
3083 3084 3085
    Returns:
        Program: startup program
    """
Y
Yu Yang 已提交
3086
    return _startup_program_
3087

3088

3089
def default_main_program():
Y
Yu Yang 已提交
3090
    """
Y
yuyang18 已提交
3091 3092 3093 3094 3095 3096 3097 3098 3099
    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.
3100

Y
Yu Yang 已提交
3101 3102 3103
    Returns:
        Program: main program
    """
Y
Yu Yang 已提交
3104
    return _main_program_
Y
Yu Yang 已提交
3105 3106 3107 3108 3109


def switch_main_program(program):
    """
    Switch the main program to a new program.
3110

Y
Yu Yang 已提交
3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124
    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):
    """
3125
    Switch the startup program to a new program
Y
Yu Yang 已提交
3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137
    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


S
rename  
sneaxiy 已提交
3138
@signature_safe_contextmanager
Y
Yu Yang 已提交
3139 3140
def program_guard(main_program, startup_program=None):
    """
Y
yuyang18 已提交
3141 3142 3143
    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.
3144

Y
Yu Yang 已提交
3145
    Examples:
Y
yuyang18 已提交
3146 3147 3148 3149 3150 3151 3152 3153 3154 3155

        >>> 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.
3156

Y
Yu Yang 已提交
3157
    Examples:
Y
yuyang18 已提交
3158 3159 3160 3161 3162 3163

        >>> 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 = ...
3164

Y
Yu Yang 已提交
3165
    Args:
Y
yuyang18 已提交
3166
        main_program(Program): New main program inside `with` statement.
3167
        startup_program(Program): New startup program inside `with` statement.
Y
Yu Yang 已提交
3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180
            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
3181 3182


W
Wu Yi 已提交
3183
def _get_var(name, program=None):
X
xuwei06 已提交
3184
    """
Y
yuyang18 已提交
3185
    Get a variable by name from the global block of a program.
F
fengjiayi 已提交
3186

X
xuwei06 已提交
3187 3188 3189
    Args:
        name(str): name of the variable
        program(Program|None): program object.
T
tangwei12 已提交
3190
        If None, default_global_program() will be used.
X
xuwei06 已提交
3191 3192 3193 3194 3195 3196 3197

    Returns:
        Variable
    """
    if program is None:
        program = default_main_program()
    assert isinstance(name, str)
3198
    assert isinstance(program, Program)
X
xuwei06 已提交
3199 3200

    return program.global_block().var(name)
3201 3202


S
rename  
sneaxiy 已提交
3203
@signature_safe_contextmanager
3204 3205 3206 3207
def _imperative_guard(tracer):
    global _imperative_tracer_
    tmp_trace = _imperative_tracer_
    _imperative_tracer_ = tracer
M
minqiyang 已提交
3208

3209
    yield
P
Paddle CI 已提交
3210

3211
    _imperative_tracer_ = tmp_trace
P
Paddle CI 已提交
3212 3213


S
rename  
sneaxiy 已提交
3214
@signature_safe_contextmanager
P
Paddle CI 已提交
3215
def _imperative_place_guard(place):
M
minqiyang 已提交
3216 3217 3218
    global _imperative_current_expected_place_
    tmp_place = _imperative_current_expected_place_
    _imperative_current_expected_place_ = place
M
minqiyang 已提交
3219

3220
    yield
M
minqiyang 已提交
3221

M
minqiyang 已提交
3222
    _imperative_current_expected_place_ = tmp_place