backward.py 69.2 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

17
from paddle.fluid import framework as framework
F
update  
fengjiayi 已提交
18
from . import core
F
update  
fengjiayi 已提交
19
import collections
20
import copy
21
import six
22
import logging
M
minqiyang 已提交
23
from .. import compat as cpt
24
from . import unique_name
25
from . import log_helper
L
liym27 已提交
26
import paddle.fluid
M
mapingshuo 已提交
27 28 29 30 31
__all__ = [
    'append_backward',
    'gradients',
]

32 33 34
_logger = log_helper.get_logger(
    __name__, logging.INFO, fmt='%(asctime)s-%(levelname)s: %(message)s')

M
mapingshuo 已提交
35 36 37 38 39 40 41 42 43 44 45

class ProgramStats(object):
    def __init__(self, block, ops):
        self.block = block
        self.ops = ops
        self.op_deps = {}  # op-> in_ops, out_ops
        self.var_op_deps = {}  # var as input op, var as output op

    def get_input_nodes(self):
        input_names = []
        for name in self.var_op_deps:
46
            if len(self.var_op_deps[name]["var_as_output_ops"]) == 0 and \
M
mapingshuo 已提交
47 48 49 50 51 52 53 54 55 56 57 58
               len(self.var_op_deps[name]["var_as_input_ops"]) > 0:
                if self.block.var(name).persistable:
                    continue
                input_names.append(name)
        for op in self.ops:
            if op.desc.type() == "read":
                input_names.extend(op.desc.output_arg_names())
        return input_names

    def get_reserved_vars(self):
        var_name = []
        for op in self.ops:
M
mapingshuo 已提交
59
            if op.desc.type() == "seed":
M
mapingshuo 已提交
60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122
                var_name.extend(op.desc.output_arg_names())
        return var_name

    def get_out_of_subgraph_vars(self, begin_op_idx, end_op_idx):
        var_name = []
        for i in range(begin_op_idx, end_op_idx, 1):
            for name in self.ops[i].desc.output_arg_names():
                if name in self.var_op_deps:
                    for idx in self.var_op_deps[name]["var_as_input_ops"]:
                        if idx >= end_op_idx:
                            var_name.append(name)
        return var_name

    def is_subgraph(self, var_group1, var_group2):
        # should traverse from var_group1 to var_group2
        # max op idx in var_group2
        # min op idx in var_group1
        min_op_idx = len(self.ops)
        max_op_idx = -1
        for name in var_group1:
            if name not in self.var_op_deps:
                return False, min_op_idx, max_op_idx
        for name in var_group2:
            if name not in self.var_op_deps:
                return False, min_op_idx, max_op_idx
        for name in var_group1:
            op_idx = self.var_op_deps[name]["var_as_input_ops"]
            for idx in op_idx:
                min_op_idx = min(min_op_idx, idx)
        for name in var_group2:
            op_idx = self.var_op_deps[name]["var_as_output_ops"]
            for idx in op_idx:
                max_op_idx = max(max_op_idx, idx)
        if min_op_idx >= max_op_idx:
            return False, min_op_idx, max_op_idx
        return True, min_op_idx, max_op_idx

    def build_stats(self):
        for i, op in enumerate(self.ops):
            self.op_deps[i] = {"in_ops": [], "out_ops": []}
            for j, name in enumerate(op.desc.input_arg_names()):
                if name in self.var_op_deps:
                    self.op_deps[i]["in_ops"].extend(self.var_op_deps[name][
                        "var_as_output_ops"])
            for j, name in enumerate(op.desc.input_arg_names()):
                if name in self.var_op_deps:
                    self.var_op_deps[name]["var_as_input_ops"].extend([i])
                else:
                    self.var_op_deps[name] = {}
                    self.var_op_deps[name]["var_as_input_ops"] = [i]
                    self.var_op_deps[name]["var_as_output_ops"] = []

            for j, name in enumerate(op.desc.output_arg_names()):
                if name in self.var_op_deps:
                    self.var_op_deps[name]["var_as_output_ops"].extend([i])
                else:
                    self.var_op_deps[name] = {}
                    self.var_op_deps[name]["var_as_input_ops"] = []
                    self.var_op_deps[name]["var_as_output_ops"] = [i]

            for op_idx in self.op_deps[i]["in_ops"]:
                self.op_deps[op_idx]["out_ops"].extend([i])

123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
    def sort_checkpoints(self, checkpoints_name):
        sorted_checkpoints = []
        for name in checkpoints_name:
            if name not in self.var_op_deps:
                _logger.debug(
                    "Recompute Optimizer: deleted %s from checkpoints, because it is not used in paddle program."
                    % name)
            elif self.var_op_deps[name]["var_as_output_ops"] == []:
                # input nodes
                sorted_checkpoints.append((name, -1))
            else:
                sorted_checkpoints.append(
                    (name, max(self.var_op_deps[name]["var_as_output_ops"])))
        sorted_checkpoints = sorted(sorted_checkpoints, key=lambda x: x[1])
        return [x[0] for x in sorted_checkpoints]

M
mapingshuo 已提交
139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174
    def modify_forward_desc_for_recompute(self):
        op_types = [op.desc.type() for op in self.ops]
        if "dropout" not in op_types:
            return

        op_idx = 0
        while (op_idx < len(self.ops)):
            op = self.ops[op_idx]
            if op.desc.type() != "dropout":
                op_idx += 1
                continue
            # add a seed op so that the two dropout op can generate same output
            op_unique_name = unique_name.generate("seed")
            var_unique_name = unique_name.generate_with_ignorable_key(".".join(
                [op_unique_name, 'tmp']))
            added_var = self.block.create_var(
                name=var_unique_name,
                dtype='int32',
                type=core.VarDesc.VarType.LOD_TENSOR,
                persistable=False,
                stop_gradient=False)
            seed = 0 if op.attr("fix_seed") is False else int(op.attr("seed"))
            added_op = self.block._insert_op(
                index=op.idx,
                type='seed',
                inputs={},
                outputs={'Out': [added_var]},
                attrs={'seed': seed})
            self.ops.insert(op_idx, added_op)
            # modify dropout op desc so that it accept a seed var as input
            op.desc.set_input("Seed", [var_unique_name])
            op.desc.remove_attr("fix_seed")
            op.desc.remove_attr("seed")
            self.block._sync_with_cpp()
            op_idx += 2

M
mapingshuo 已提交
175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236

def _pretty_op_desc_(op_desc, prefix):
    out_s = "%s\tname:[%s]\n%s    \tinputs:[%s]\n%s    \toutputs:[%s]" % \
            (prefix + "_op", str(op_desc.type()), prefix + "_input", " ".join(op_desc.input_arg_names()),
             prefix + "_output", " ".join(op_desc.output_arg_names()))
    return out_s


def _add_needed_descs_to_block(descs, block, main_block, in_memory_vars):
    if len(descs) == 0:
        return []
    result_descs = []
    op_role_attr_name = \
            core.op_proto_and_checker_maker.kOpRoleAttrName()
    backward = core.op_proto_and_checker_maker.OpRole.Backward
    for desc in descs:
        if isinstance(desc, framework.Operator):
            desc = desc.desc
        if isinstance(desc, tuple):
            desc = desc[0]
        is_needed = False
        for name in desc.output_arg_names():
            if main_block.has_var(name) and main_block.var(name).persistable:
                continue
            if name not in in_memory_vars:
                is_needed = True
        if is_needed:
            new_op_desc = block.desc.append_op()
            new_op_desc.copy_from(desc)
            new_op_desc._set_attr(op_role_attr_name, backward)
            result_descs.append(new_op_desc)
    return result_descs


def _add_descs_to_block(descs, block):
    if len(descs) == 0:
        return []
    result_descs = []
    op_role_attr_name = \
        core.op_proto_and_checker_maker.kOpRoleAttrName()
    backward = core.op_proto_and_checker_maker.OpRole.Backward
    for desc in descs:
        if isinstance(desc, framework.Operator):
            desc = desc.desc
        if isinstance(desc, tuple):
            desc = desc[0]
        new_op_desc = block.desc.append_op()
        new_op_desc.copy_from(desc)
        new_op_desc._set_attr(op_role_attr_name, backward)
        result_descs.append(new_op_desc)
    return result_descs


def _find_loss_op_(loss):
    for op in reversed(loss.block.ops):
        assert isinstance(op, framework.Operator)
        if len(op.output_arg_names) == 1 and op.output_arg_names[
                0] == loss.name:
            loss.op = op
            break
    if loss.op is None:
        raise ValueError("loss.op is None. Should not happend")
237 238


239 240
def _rename_arg_(op_descs, old_name, new_name, begin_idx=None, end_idx=None):
    """
241
    Traverse all ops in op_descs[begin_idx : end_idx],
242 243
    if any op has inputs/outputs named "old_name", rename it as 'new_name'
    """
F
update  
fengjiayi 已提交
244 245 246
    if begin_idx is None:
        begin_idx = 0
    if end_idx is None:
247
        end_idx = len(op_descs)
F
update  
fengjiayi 已提交
248
    for i in range(begin_idx, end_idx):
249
        op_desc = op_descs[i]
F
fengjiayi 已提交
250 251
        if isinstance(op_desc, tuple):
            op_desc = op_desc[0]
W
Wu Yi 已提交
252 253
        op_desc._rename_input(old_name, new_name)
        op_desc._rename_output(old_name, new_name)
F
update  
fengjiayi 已提交
254 255


F
fengjiayi 已提交
256
def _create_op_desc_(op_type, inputs, outputs, attrs):
257 258 259
    """
    Create a C++ OpDesc object with specified inputs, outputs and attributes.
    """
F
fengjiayi 已提交
260 261
    op_desc = core.OpDesc()
    op_desc.set_type(op_type)
M
minqiyang 已提交
262
    for para, args in six.iteritems(inputs):
263 264 265 266 267
        op_desc.set_input(
            para,
            list(
                map(lambda arg: arg.decode() if isinstance(arg, six.binary_type) else arg,
                    args)))
M
minqiyang 已提交
268
    for para, args in six.iteritems(outputs):
269 270 271 272 273
        op_desc.set_output(
            para,
            list(
                map(lambda arg: arg.decode() if isinstance(arg, six.binary_type) else arg,
                    args)))
Y
yuyang18 已提交
274 275

    op_role_attr_name = core.op_proto_and_checker_maker.kOpRoleAttrName()
276
    op_device_attr_name = core.op_proto_and_checker_maker.kOpDeviceAttrName()
Y
yuyang18 已提交
277 278 279 280

    if op_role_attr_name not in attrs:
        attrs[
            op_role_attr_name] = core.op_proto_and_checker_maker.OpRole.Backward
281 282
    if op_device_attr_name not in attrs:
        attrs[op_device_attr_name] = ""
M
minqiyang 已提交
283
    for name, val in six.iteritems(attrs):
F
fengjiayi 已提交
284 285 286
        if isinstance(val, framework.Block):
            op_desc.set_block_attr(name, val.desc)
        else:
W
Wu Yi 已提交
287
            op_desc._set_attr(name, val)
F
fengjiayi 已提交
288 289 290
    return op_desc


M
mapingshuo 已提交
291 292 293 294 295 296 297 298 299 300
def _create_loss_op_desc_(loss):
    op_desc = _create_op_desc_(
        "fill_constant", {}, {"Out": [_append_grad_suffix_(loss.name)]}, {
            "shape": [1],
            "value": 1.0,
            "dtype": loss.dtype,
            "force_cpu": False,
            core.op_proto_and_checker_maker.kOpRoleAttrName():
            int(core.op_proto_and_checker_maker.OpRole.Backward) |
            int(core.op_proto_and_checker_maker.OpRole.Loss),
301 302
            core.op_proto_and_checker_maker.kOpDeviceAttrName():
            loss.op.attr(core.op_proto_and_checker_maker.kOpDeviceAttrName())
M
mapingshuo 已提交
303 304 305 306
        })
    return op_desc


307
def _infer_var_data_type_shape_(grad_var_name, block):
308
    """
309
    Infer the data type and shape of given grad variable
310
    """
M
minqiyang 已提交
311 312 313 314
    grad_var = block.desc.find_var(cpt.to_bytes(grad_var_name))
    fwd_name = _strip_grad_suffix_(grad_var_name)
    if block.desc.has_var_recursive(cpt.to_bytes(fwd_name)):
        fwd_var = block.desc.find_var_recursive(cpt.to_bytes(fwd_name))
F
fengjiayi 已提交
315
        grad_var.set_dtype(fwd_var.dtype())
316
        grad_var.set_shape(fwd_var.shape())
F
fengjiayi 已提交
317
    else:
318
        grad_var.set_dtype(core.VarDesc.VarType.FP32)
F
fengjiayi 已提交
319 320


F
fengjiayi 已提交
321
def _all_in_set_(cands, s):
322 323 324
    """
    Test if all elements of 'cands' are in set 's'
    """
F
fengjiayi 已提交
325 326
    if len(cands) == 0:
        return False
F
fengjiayi 已提交
327 328 329 330 331 332
    for c in cands:
        if not c in s:
            return False
    return True


333 334 335 336 337 338
def _some_in_set_(cands, s):
    """
    Test if some elements of 'cands' are in set 's'
    """
    if len(cands) == 0:
        return False
M
minqiyang 已提交
339 340
    literal_set = cpt.to_text(s)
    literal_cands = cpt.to_text(cands)
M
minqiyang 已提交
341 342
    for c in literal_cands:
        if c in literal_set:
343 344 345 346
            return True
    return False


F
fengjiayi 已提交
347
def _strip_grad_suffix_(name):
348
    """
M
mapingshuo 已提交
349
    Strip the grad suffix from the given variable name
350 351 352
    e.g. x@GRAD ==> x
         y@GRAD@RENAME@1 ==> y
    """
M
minqiyang 已提交
353
    name = cpt.to_text(name)
M
minqiyang 已提交
354
    pos = name.find(core.grad_var_suffix())
F
fengjiayi 已提交
355
    return name[:pos] if pos != -1 else name
F
fengjiayi 已提交
356 357 358


def _append_grad_suffix_(name):
359 360 361 362
    """
    Append grad suffix to the given variable name
    e.g. x ==> x@GRAD
    """
M
minqiyang 已提交
363
    return cpt.to_text(name) + core.grad_var_suffix()
F
fengjiayi 已提交
364 365


366
def _addup_repetitive_outputs_(op_descs, block_idx):
367 368
    """
    In backward part, an variable may be the output of more than one ops.
F
fengjiayi 已提交
369 370
    And one op may yield its multiple outputs to the same variable.
    In these cases, the variable should be the accumulation of all the outputs.
371 372
    `sum_op`s are added to implement the accumulate.
    """
F
update  
fengjiayi 已提交
373 374
    pending_sum_ops = []
    var_rename_count = collections.defaultdict(int)
F
fengjiayi 已提交
375
    renamed_vars = collections.defaultdict(list)
376
    renamed_var_start_idx = collections.defaultdict(list)
F
fengjiayi 已提交
377
    for idx, op_desc in enumerate(op_descs):
F
update  
fengjiayi 已提交
378
        for var_name in op_desc.input_arg_names():
M
mapingshuo 已提交
379 380
            if "@GRAD" not in var_name:
                continue
F
fengjiayi 已提交
381
            if len(renamed_vars[var_name]) > 1:
382 383 384
                pending_sum_ops.append((_create_op_desc_(
                    "sum", {"X": renamed_vars[var_name]}, {"Out": [var_name]},
                    {"use_mkldnn": False}), idx))
F
fengjiayi 已提交
385
                renamed_vars[var_name] = [var_name]
F
update  
fengjiayi 已提交
386
        for param_idx, param_name in enumerate(op_desc.output_names()):
F
fengjiayi 已提交
387 388
            arg_names = op_desc.output(param_name)
            for arg_idx, var_name in enumerate(arg_names):
M
mapingshuo 已提交
389 390 391 392
                if "@GRAD" not in var_name:
                    continue
                #if "@RENAME@" in var_name:
                #    continue
F
fengjiayi 已提交
393 394 395 396 397 398 399
                if var_name == core.empty_var_name(
                ) or var_name in op_desc.input_arg_names():
                    # empty variable or inplace op
                    continue
                if len(renamed_vars[var_name]) == 0:
                    # it's the first time we get the variable
                    renamed_vars[var_name] = [var_name]
400
                    renamed_var_start_idx[var_name] = idx
F
fengjiayi 已提交
401 402
                else:
                    if len(renamed_vars[var_name]) == 1:
403
                        new_name = var_name + "@RENAME@block" + str(block_idx) + "@" + \
F
fengjiayi 已提交
404 405 406 407
                            str(var_rename_count[var_name])
                        var_rename_count[var_name] += 1
                        # rename original var_name
                        renamed_vars[var_name][0] = new_name
408 409 410 411 412 413
                        # before change: _rename_arg_(op_descs, var_name,
                        #                             new_name, 0, idx)
                        # rename arg from idx of the first appearance
                        # in backward, not always from 0
                        _rename_arg_(op_descs, var_name, new_name,
                                     renamed_var_start_idx[var_name], idx)
F
fengjiayi 已提交
414 415
                        _rename_arg_(pending_sum_ops, var_name, new_name)

F
update  
fengjiayi 已提交
416 417 418 419 420 421 422 423 424 425 426 427 428
                        for p in op_desc.output_names()[:param_idx]:
                            p_arg_names = op_desc.output(p)
                            if var_name in p_arg_names:
                                op_desc.set_output(p, [
                                    new_name if x == var_name else x
                                    for x in p_arg_names
                                ])

                        arg_names = [
                            new_name if x == var_name else x
                            for x in arg_names[:arg_idx]
                        ] + arg_names[arg_idx:]

429
                    new_name = var_name + "@RENAME@block" + str(block_idx) + "@" + \
F
fengjiayi 已提交
430
                        str(var_rename_count[var_name])
F
fengjiayi 已提交
431
                    var_rename_count[var_name] += 1
F
fengjiayi 已提交
432 433 434
                    arg_names[arg_idx] = new_name
                    op_desc.set_output(param_name, arg_names)
                    renamed_vars[var_name].append(new_name)
F
update  
fengjiayi 已提交
435

M
minqiyang 已提交
436
    for var_name, inputs in six.iteritems(renamed_vars):
F
update  
fengjiayi 已提交
437
        if len(inputs) > 1:
438 439 440
            pending_sum_ops.append(
                (_create_op_desc_("sum", {"X": inputs}, {"Out": [var_name]},
                                  {"use_mkldnn": False}), len(op_descs)))
F
fengjiayi 已提交
441
    # sum_op descs are sorted according to their insert position
F
update  
fengjiayi 已提交
442
    for p in reversed(pending_sum_ops):
F
fengjiayi 已提交
443 444 445 446 447 448
        op_descs.insert(p[1], p[0])

    return op_descs


def _remove_no_grad_branch_(op_descs, no_grad_set):
449 450 451 452
    """
    Remove unnecessary grad ops
    A grad op can be removed in two cases:
        1. all outputs of the grad op are in 'no_grad_set'
F
fengjiayi 已提交
453
        2. all grad inputs of the grad op are in 'no_grad_set'
454
    """
F
fengjiayi 已提交
455 456

    def _op_can_be_removed_(op_desc, no_grad_set):
F
fengjiayi 已提交
457 458
        out_arg_names = op_desc.output_arg_names()
        if len(out_arg_names) == 0 or _all_in_set_(out_arg_names, no_grad_set):
F
fengjiayi 已提交
459
            return True
460 461 462 463
        if _all_in_set_([
                name for name in op_desc.input_arg_names()
                if name.find(core.grad_var_suffix()) != -1
        ], no_grad_set):
F
fengjiayi 已提交
464
            no_grad_set.update(out_arg_names)
F
fengjiayi 已提交
465 466 467
            return True
        return False

F
fengjiayi 已提交
468
    # Remove ops whose outputs are all in no_grad_dict
469 470 471 472
    op_descs = [
        op_desc for op_desc in op_descs
        if not _op_can_be_removed_(op_desc, no_grad_set)
    ]
F
fengjiayi 已提交
473 474
    # Insert fill_zeros_like_op
    to_insert = []
F
fengjiayi 已提交
475
    for idx, op_desc in enumerate(op_descs):
F
fengjiayi 已提交
476
        for arg in op_desc.input_arg_names():
M
mapingshuo 已提交
477
            # arg is a gradient var name and arg should not have gradient
F
fengjiayi 已提交
478
            if core.grad_var_suffix() in arg and arg in no_grad_set:
479
                x_in = _strip_grad_suffix_(arg)
M
mapingshuo 已提交
480 481
                # the reason should be: arg can be input of another grad op
                # and the op is a not-to-remove op
482 483
                to_insert.append((_create_op_desc_(
                    "fill_zeros_like", {"X": [x_in]}, {"Out": [arg]}, {}), idx))
F
fengjiayi 已提交
484

485
    list([op_descs.insert(p[1], p[0]) for p in reversed(to_insert)])
F
fengjiayi 已提交
486 487 488 489

    return op_descs


C
chengduo 已提交
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
def _find_not_need_ops(grad_op_descs, forward_ops, input_grad_names_set):
    """
    Pruning Program with Structural Analysis Method of Computational Graph.
    The nodes of the computational graph composed of backward OPS should be
    interconnected. If there are unconnected sub-graphs in the computational graph,
    these sub-graphs should be cut off.

    Args:
        grad_op_descs(list[core.OpDesc]): The candidate backward OpDescs.
        forward_ops(list[Operator]): The forward ops.
        input_grad_names_set(set): this set is used to store the gradients' name
            which is generated by backward ops, and input_grad_names_set can help
            to prune the unnecessary backward ops.

    Return:
505
        (set[core.OpDesc]): A set of OpDescs which should be pruned.
C
chengduo 已提交
506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
    """

    class Var(object):
        def __init__(self, var_name):
            self.var_name = var_name
            self.gen_op = None
            self.pendding_ops = []

        def set_gen_op(self, gen_op):
            assert isinstance(gen_op, Op)
            assert self.gen_op is None
            self.gen_op = gen_op

        def add_pending_op(self, op):
            assert isinstance(op, Op)
            self.pendding_ops.append(op)

    class Op(object):
        def __init__(self, op_desc):
            self.op_desc = op_desc
            self.inputs = []
            self.outputs = []

        def insert_input(self, var):
            assert isinstance(var, Var)
            self.inputs.append(var)

        def insert_output(self, var):
            assert isinstance(var, Var)
            self.outputs.append(var)

    var_versions = dict()

    def _create_node(name):
        if name not in var_versions.keys():
            var_versions[name] = [Var(name)]
        else:
            var_versions[name].append(Var(name))
        return var_versions[name][-1]

    def _create_or_get_last_version_node(name):
        if name not in var_versions.keys():
            var_versions[name] = [Var(name)]
        return var_versions[name][-1]

    def _create_op_node(op_desc):
        op_node = Op(op_desc)
        for input in op_desc.input_arg_names():
            var = _create_or_get_last_version_node(name=input)
            var.add_pending_op(op_node)
            op_node.insert_input(var)
        for output in op_desc.output_arg_names():
            var = _create_node(name=output)
            var.set_gen_op(op_node)
            op_node.insert_output(var)
        return op_node

    # Record the forward vars
    forward_vars_set = set() if input_grad_names_set is None else set(
        input_grad_names_set)
    for op in forward_ops:
        forward_vars_set.update(op.desc.input_arg_names())
        forward_vars_set.update(op.desc.output_arg_names())

    # Record the vars which are created during backward and is not generated by op.
    backward_vars_set = set()
    # special_op_nodes is the candidate sub-graph head node.
    special_op_nodes = set()
    for op_desc in grad_op_descs:
        input_set = set(op_desc.input_arg_names())
        # The new_vars are created during backward and is not generated by op.
        new_vars = input_set - forward_vars_set - backward_vars_set
        backward_vars_set.update(op_desc.output_arg_names())

        op_node = _create_op_node(op_desc)
        if len(new_vars) == len(input_set):
            special_op_nodes.add(op_node)

    not_need_op_descs = []
    # Start traversing all candidate sub-graph headers to check whether
    # they are connected to backward computational graphs, and if they are
    # not, list them in not_need_op_descs
    for special_op_node in special_op_nodes:
        op_list = [special_op_node]
        ready_vars = set(special_op_node.inputs)
        remove_ops = True
        candidate_ops = [special_op_node]
        while len(candidate_ops) > 0:
            op_node = candidate_ops.pop(0)
            if _all_in_set_(op_node.inputs, ready_vars):
                for out_var in op_node.outputs:
                    candidate_ops.extend(out_var.pendding_ops)
                    op_list.extend(out_var.pendding_ops)
                ready_vars.update(op_node.outputs)
            else:
                remove_ops = False
                break
        if remove_ops:
            not_need_op_descs.extend([node.op_desc for node in op_list])
605 606 607 608 609 610 611
    not_need_op_descs_set = set(not_need_op_descs)
    grad_op_descs_set = set(grad_op_descs)
    # If a backward computational graph is simply one sub-graph header, the
    # not_need_op_descs will be whole graph, this IF clause avoids it. 
    if grad_op_descs_set == not_need_op_descs_set:
        return set()
    return not_need_op_descs_set
C
chengduo 已提交
612 613


614
from .proto import framework_pb2
Y
Yang Yang 已提交
615 616 617 618


def serialize_op_decs(op_desc):
    protostr = op_desc.serialize_to_string()
M
minqiyang 已提交
619
    proto = framework_pb2.OpDesc.FromString(six.binary_type(protostr))
Y
Yang Yang 已提交
620 621 622
    return proto.__str__()


M
mapingshuo 已提交
623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
def _append_backward_ops_with_checkpoints_(
        block, ops, target_block, no_grad_dict, grad_to_var, checkpoints):
    """
    Create grad ops with forward ops, and insert them into given block

    Args:
        block(Block): the block where forward ops are
        ops(Op): the forward operators whose forward recomputation backward ops need to be added
        target_block(Block): the block which is going to hold new generated grad ops
        no_grad_dict(dict):
            key(int) block index
            val(str): corresponding forward variable name
        checkpoints: variables that a user defined as checkpoint for forward recomputation

    Algorithms:
M
mapingshuo 已提交
638
        0) deal with forward recomputing program descs
M
mapingshuo 已提交
639 640 641 642 643
        1) find ops between checkpoints, i.e. recompute_segments
        2) go through all forward ops and induct all variables that will be hold in memory
            a. variables that are used across segments will be held in memory
            b. output of dropout op will be held in memory
            c. input variables will be held in memory
M
mapingshuo 已提交
644 645 646
        3) go through each recompute_segments, add backward ops with forward recomputation
            a. add ops in current recompute_segment as forward recomputation ops
            b. rename all non-checkpoint variables in recomputation ops
M
mapingshuo 已提交
647 648
            c. add backward ops of current recomputation ops
            d. add sum op for repetitive_outputs
M
mapingshuo 已提交
649 650
        4) remove no grad branch as it is in _remove_no_grad_branch_
        5) Note1: all appended ops' OpRole are Backward
M
mapingshuo 已提交
651 652
        6) Note2: all variables with new name should be returned so that _append_backward_vars_ can be called
        7) Note3: current forward recomputation backpropagation does not handle programs with subblock
M
mapingshuo 已提交
653
    """
M
mapingshuo 已提交
654 655

    checkpoints_name = [x.name for x in checkpoints]
656
    checkpoints_name = list(set(checkpoints_name))
M
mapingshuo 已提交
657 658
    local_block = block.program._create_block()
    buffer_block = block.program._create_block()
M
mapingshuo 已提交
659
    # 0) deal with forward recomputing program descs  
M
mapingshuo 已提交
660
    program_stat = ProgramStats(block, ops)
M
mapingshuo 已提交
661
    program_stat.modify_forward_desc_for_recompute()
M
mapingshuo 已提交
662
    program_stat.build_stats()
M
mapingshuo 已提交
663 664

    # 1) find ops between checkpoints, i.e. recompute_segments
665
    checkpoints_name = program_stat.sort_checkpoints(checkpoints_name)
M
mapingshuo 已提交
666 667
    segments = []

668
    if len(checkpoints_name) == 1:
M
mapingshuo 已提交
669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695
        # only one checkpoint
        max_op_idx = -1
        var_group = [checkpoints_name[0]]
        for name in var_group:
            if name not in program_stat.var_op_deps:
                break
            op_idx = program_stat.var_op_deps[name]["var_as_output_ops"]
            for idx in op_idx:
                max_op_idx = max(max_op_idx, idx)
        if max_op_idx > 0:
            segments.append([0, max_op_idx + 1])
    else:
        start_idx = 0
        while True:
            if start_idx >= len(checkpoints_name) - 1:
                break
            flag, min_idx, max_idx = program_stat.is_subgraph(
                [checkpoints_name[start_idx]],
                [checkpoints_name[start_idx + 1]])
            if flag:
                segments.append([min_idx, max_idx + 1])
            start_idx += 1

    if segments != [] and segments[0][0] != 0:
        recompute_segments = [[0, segments[0][0]]] + segments
    else:
        recompute_segments = segments
M
mapingshuo 已提交
696 697

    # 2) go through all forward ops and induct all variables that will be hold in memory
M
mapingshuo 已提交
698
    vars_should_be_hold = []
699
    # a. variables that are used across segments will be held in memory
M
mapingshuo 已提交
700 701 702
    for segment in recompute_segments:
        vars_should_be_hold.extend(
            program_stat.get_out_of_subgraph_vars(segment[0], segment[1]))
M
mapingshuo 已提交
703
    # b. output of dropout op will be held in memory
M
mapingshuo 已提交
704
    vars_should_be_hold.extend(program_stat.get_reserved_vars())
M
mapingshuo 已提交
705
    # c. input variables are checkpoints
M
mapingshuo 已提交
706 707 708
    vars_should_be_hold.extend(program_stat.get_input_nodes())
    vars_should_be_hold = list(set(vars_should_be_hold))

M
mapingshuo 已提交
709
    # 3) go through each recompute_segments, add backward ops with forward recomputation
M
mapingshuo 已提交
710 711 712 713 714 715 716
    grad_op_descs = []
    var_name_dict = {}

    vars_in_memory = vars_should_be_hold + checkpoints_name

    max_calculated_op_position = len(ops)
    if recompute_segments == []:
717
        # if there is no recompute segment, add backward ops like
M
mapingshuo 已提交
718
        # _append_backward_ops_ function
M
mapingshuo 已提交
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
        gap_ops = ops[0:max_calculated_op_position]
        for op in reversed(gap_ops):
            if op.has_attr("sub_block"):
                raise Exception("Recompute don't support ops with sub_block"
                                "invoke op: %s" %
                                _pretty_op_desc_(op.desc, "with_sub_block"))
            grad_op_desc, op_grad_to_var = core.get_grad_op_desc(
                op.desc, cpt.to_text(no_grad_dict[block.idx]), [])
            added_descs = _add_descs_to_block(grad_op_desc, local_block)
            grad_op_descs.extend(added_descs)
            grad_to_var.update(op_grad_to_var)

    for i, segment in enumerate(recompute_segments[::-1]):
        # add grad op for ops not in any segments
        gap_ops = ops[segment[1]:max_calculated_op_position]
        max_calculated_op_position = segment[0]
        for op in reversed(gap_ops):
            if op.has_attr("sub_block"):
                raise Exception("Recompute don't support ops with sub_block"
                                "invoke op: %s" %
                                _pretty_op_desc_(op.desc, "with_sub_block"))
            grad_op_desc, op_grad_to_var = core.get_grad_op_desc(
                op.desc, cpt.to_text(no_grad_dict[block.idx]), [])
            added_descs = _add_descs_to_block(grad_op_desc, local_block)
            grad_op_descs.extend(added_descs)
            grad_to_var.update(op_grad_to_var)

        ff_ops = ops[segment[0]:segment[1]]
        var_suffix = ".subprog_%d" % i

        for op in ff_ops:
            if op.has_attr("sub_block"):
                raise Exception("Recompute don't support ops with sub_block"
                                "invoke op: %s" %
                                _pretty_op_desc_(op.desc, "with_sub_block"))
            input_and_output_names = []
            input_and_output_names.extend(op.desc.input_arg_names())
            input_and_output_names.extend(op.desc.output_arg_names())
            for name in input_and_output_names:
                if block.var(name).persistable or name in checkpoints_name:
                    continue
                if name in vars_should_be_hold:
                    continue
                if name not in var_name_dict:
                    var_name_dict[name] = name + var_suffix
M
mapingshuo 已提交
764
        # 3.a. add ops in current recompute_segment as forward recomputation ops
M
mapingshuo 已提交
765 766 767 768
        buffer_descs = _add_needed_descs_to_block(ff_ops, buffer_block, block,
                                                  vars_in_memory)
        added_descs = _add_descs_to_block(ff_ops, local_block)

M
mapingshuo 已提交
769
        # 3.b. rename all non-checkpoint variables in recomputation ops
M
mapingshuo 已提交
770 771 772 773 774 775
        for key in var_name_dict:
            _rename_arg_(buffer_descs, key, var_name_dict[key])

        # added_descs should be in grad_op_descs because it is backward op desc
        grad_op_descs.extend(buffer_descs)

M
mapingshuo 已提交
776
        # 3.c. add backward ops of current recomputation ops
M
mapingshuo 已提交
777 778 779 780 781 782 783 784
        for op_desc in reversed(added_descs):
            grad_op_desc, op_grad_to_var = core.get_grad_op_desc(
                op_desc, cpt.to_text(no_grad_dict[block.idx]), [])
            for key in var_name_dict:
                _rename_arg_(grad_op_desc, key, var_name_dict[key])
            grad_op_descs.extend(grad_op_desc)
            grad_to_var.update(op_grad_to_var)

M
mapingshuo 已提交
785
    # 3.d. add sum op for repetitive_outputs
786
    grad_op_descs = _addup_repetitive_outputs_(grad_op_descs, block.idx)
M
mapingshuo 已提交
787
    # 4) remove no grad branch as it is in _remove_no_grad_branch_
M
mapingshuo 已提交
788 789 790 791 792 793
    grad_op_descs = _remove_no_grad_branch_(grad_op_descs,
                                            no_grad_dict[block.idx])
    added_descs = _add_descs_to_block(grad_op_descs, target_block)
    return program_stat, checkpoints_name, vars_should_be_hold, recompute_segments


794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809
def _get_sub_block_path(sub_block, sub_block_op_desc, no_grad_set):
    """
    Get output vars in subblock which will be assigned to parent block.
    It is used to find the grad path in subblock
    """
    assert sub_block_op_desc.has_attr(
        "sub_block") and sub_block.idx == sub_block_op_desc._block_attr_id(
            "sub_block")
    # TODO(huihuangzheng): add support for recurrent op and while op
    if sub_block_op_desc.type == "conditional_block":
        sub_outputs = []
        sub_assign_to_out_ops = []
        for var in sub_block_op_desc.output_arg_names:
            for op_desc in sub_block.ops:
                if op_desc.type == "assign" and var in op_desc.output_arg_names:
                    sub_assign_to_out_ops.append(op_desc)
810 811 812 813
                    for name in op_desc.input_arg_names:
                        if sub_block.has_var(name):
                            sub_outputs.append(sub_block.var(name))

814 815 816 817 818 819 820 821 822 823
        sub_block_op_path = _find_op_path_(sub_block, sub_outputs, [],
                                           no_grad_set)
        # TODO better way than finding in list
        for op_desc in sub_assign_to_out_ops:
            if op_desc not in sub_block_op_path:
                sub_block_op_path.append(op_desc)
        return sub_block_op_path
    return sub_block.ops


824 825
def _append_backward_ops_(block,
                          ops,
F
fengjiayi 已提交
826 827 828
                          target_block,
                          no_grad_dict,
                          grad_to_var,
829 830
                          callbacks=None,
                          input_grad_names_set=None):
831 832 833 834 835
    """
    Create all grad ops, and insert them into given block

    Args:
        block(Block): the block where forward ops are
836
        ops(Op): the forward operators whose backward ops need to be added
837
        target_block(Block): the block which is going to hold new generated grad ops
838
        no_grad_dict(dict):
839
            key(int)  block index
T
tianshuo78520a 已提交
840
            val(set) a set of variable names. These variables have no gradient
841 842 843
        grad_to_var(dict)(output argument):
            key(str): grad variable name
            val(str): corresponding forward variable name
C
chengduo 已提交
844 845 846 847
        callbacks(callable object): a callable object used to decorate new generated grad ops
        input_grad_names_set(set): this set is used to store the gradients' name which is
            generated by backward ops, and input_grad_names_set can help to prune the unnecessary
            backward ops.
848
    """
Y
Yang Yang 已提交
849
    if callbacks is not None:
Y
Yang Yang 已提交
850 851 852 853
        assert (isinstance(callbacks, list))
        for cb in callbacks:
            if not hasattr(cb, '__call__'):
                raise ValueError("'callback' must be a callable object.")
F
fengjiayi 已提交
854

F
fengjiayi 已提交
855
    # grad_op_descs holds created grad_op, and will be appended to target_block
F
fengjiayi 已提交
856 857
    grad_op_descs = []
    program = block.program
858 859

    # add grad_op_desc by reversed ops
860
    for op in reversed(ops):
F
fengjiayi 已提交
861 862 863
        grad_sub_block_list = []
        # If the op has its own sub-block, deal with the sub-block first
        if op.has_attr("sub_block"):
W
Wu Yi 已提交
864
            sub_block = program.block(op._block_attr_id("sub_block"))
W
Wu Yi 已提交
865
            grad_sub_block = program._create_block()
W
Wu Yi 已提交
866
            grad_sub_block._set_forward_block_idx(sub_block.idx)
867 868 869
            # see follwing comments for why set None here.
            pre_input_grad_names_set = copy.copy(input_grad_names_set)
            input_grad_names_set = None
870 871 872
            sub_block_path = _get_sub_block_path(sub_block, op,
                                                 no_grad_dict[sub_block.idx])
            _append_backward_ops_(sub_block, sub_block_path, grad_sub_block,
873 874 875
                                  no_grad_dict, grad_to_var, callbacks,
                                  input_grad_names_set)
            input_grad_names_set = pre_input_grad_names_set
Y
Yu Yang 已提交
876

W
Wu Yi 已提交
877
            program._rollback()
F
fengjiayi 已提交
878 879
            grad_sub_block_list.append(grad_sub_block.desc)

F
fengjiayi 已提交
880
        # Getting op's corresponding grad_op
F
fengjiayi 已提交
881
        grad_op_desc, op_grad_to_var = core.get_grad_op_desc(
M
minqiyang 已提交
882
            op.desc, cpt.to_text(no_grad_dict[block.idx]), grad_sub_block_list)
Y
Yang Yu 已提交
883

884 885
        # Set device for grad_op according to forward Op
        device_attr_name = core.op_proto_and_checker_maker.kOpDeviceAttrName()
886 887 888 889
        if op.desc.has_attr(device_attr_name):
            op_device = op.desc.attr(device_attr_name)
            for op_desc in grad_op_desc:
                op_desc._set_attr(device_attr_name, op_device)
890

891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917
        # If input_grad_names_set is not None, extend grad_op_descs only when
        # any input grad in outputs of previous grad ops.
        # But this strategy is not suited for while op for some control flow,
        # for example, for while op, the grads maybe generated in next loop.
        if input_grad_names_set is not None:
            is_append_grad = False
            for op_desc in grad_op_desc:
                input_grad_names = [
                    name for name in op_desc.input_arg_names()
                    if name.find(core.grad_var_suffix()) != -1
                ]
                # some code of gradient ops, like increment, are not very
                # standard, there is no @GRAD in these ops' inputs.
                if len(input_grad_names) == 0:
                    is_append_grad = True
                    break

                if _some_in_set_(input_grad_names, input_grad_names_set):
                    grad_op_descs.append(op_desc)
                    is_append_grad = True
                    for name in op_desc.output_arg_names():
                        input_grad_names_set.add(name)
            if is_append_grad:
                grad_to_var.update(op_grad_to_var)
        else:
            grad_op_descs.extend(grad_op_desc)
            grad_to_var.update(op_grad_to_var)
F
fengjiayi 已提交
918

M
mapingshuo 已提交
919
    # sum parameter's gradients' var given multiple var gradient
920
    grad_op_descs = _addup_repetitive_outputs_(grad_op_descs, block.idx)
F
fengjiayi 已提交
921

M
mapingshuo 已提交
922 923
    # if all outputs of the grad op are in no_grad_set, then just remove and fill zero
    # if all inputs of the grad op are in no_grad_set, just remove this op
F
fengjiayi 已提交
924 925
    grad_op_descs = _remove_no_grad_branch_(grad_op_descs,
                                            no_grad_dict[block.idx])
F
fengjiayi 已提交
926

M
mapingshuo 已提交
927
    # remove some backward ops
C
chengduo 已提交
928
    not_need_ops = _find_not_need_ops(grad_op_descs, ops, input_grad_names_set)
M
mapingshuo 已提交
929

C
chengduo 已提交
930 931 932
    grad_op_descs = [
        op_desc for op_desc in grad_op_descs if op_desc not in not_need_ops
    ]
933

F
fengjiayi 已提交
934
    # append op_desc in grad_op_descs to target_block
Y
yuyang18 已提交
935 936
    op_role_attr_name = core.op_proto_and_checker_maker.kOpRoleAttrName()
    backward = core.op_proto_and_checker_maker.OpRole.Backward
F
update  
fengjiayi 已提交
937
    for op_desc in grad_op_descs:
F
fengjiayi 已提交
938 939
        new_op_desc = target_block.desc.append_op()
        new_op_desc.copy_from(op_desc)
W
Wu Yi 已提交
940
        new_op_desc._set_attr(op_role_attr_name, backward)
Y
Yang Yang 已提交
941
        grad_to_var["__current_op_desc__"] = new_op_desc
Y
Yang Yang 已提交
942 943 944 945
        if callbacks is not None:
            assert (isinstance(callbacks, list))
            for cb in callbacks:
                cb(block=target_block, context=grad_to_var)
F
update  
fengjiayi 已提交
946

F
fengjiayi 已提交
947

948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
def _is_grad_var_(var_name):
    return core.grad_var_suffix() in var_name


# Find the op who holds the sub_block as its "sub_block" attr
def _find_parent_op_(sub_block):
    sub_block_id = sub_block.idx

    if sub_block_id == 0:
        return None

    program = sub_block.program
    for block_id in six.moves.range(program.num_blocks):
        block_desc = program.block(block_id).desc
        for op_idx in six.moves.range(block_desc.op_size()):
            op = block_desc.op(op_idx)
            if op.has_attr("sub_block") and op._block_attr_id(
                    "sub_block") == sub_block_id:
                return op

    # NOTE(paddle-dev): When optimizer is added in conditional block, 
    # sub_block may not be found.
    return None


F
fengjiayi 已提交
973
def _append_backward_vars_(block, start_op_idx, grad_to_var, grad_info_map):
974 975 976 977 978 979 980 981 982 983 984 985
    """
    Create new variables required by backward pass.

    Args:
        block(Block): the block where new variables will be created
        start_op_idx(int): Only variables required by ops in block.ops[start_op_idx : ] will be created
        grad_to_var(dict):
            key(str): grad variable name
            val(str): corresponding forward variable name
            In most cases, this dict is generated by _append_backward_ops_()
        grad_info_map(dict)(output argument):
            key(str): forward variable name
986
            val(tuple): a tuple of (str, Block), str is the corresponding grad name, Block is the block containing grad variable
987
    """
988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004
    ops_to_remove = []
    '''
    NOTE(paddle-dev): while_grad op may hold some inputs which are not found 
    in the parent/forward block, and they are also the outputs of while_grad 
    op. These kinds of inputs are the recursive outputs inside while_grad op. 
    They should be considered as "already created" when scanning the inner 
    ops of while_grad ops.  
    '''
    parent_op = _find_parent_op_(block)
    parent_op_vars = []
    if parent_op is not None:
        input_args = parent_op.input_arg_names()
        output_args = parent_op.output_arg_names()
        for in_arg in input_args:
            if in_arg in output_args:
                parent_op_vars.append(in_arg)

F
fengjiayi 已提交
1005 1006 1007
    for op_idx in range(start_op_idx, block.desc.op_size()):
        op_desc = block.desc.op(op_idx)
        if op_desc.has_attr("sub_block"):
W
Wu Yi 已提交
1008
            sub_block = block.program.block(op_desc._block_attr_id("sub_block"))
F
fengjiayi 已提交
1009
            _append_backward_vars_(sub_block, 0, grad_to_var, grad_info_map)
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052

        grad_var_ins = [
            var for var in op_desc.input_arg_names() if _is_grad_var_(var)
        ]
        grad_var_outs = [
            var for var in op_desc.output_arg_names() if _is_grad_var_(var)
        ]

        inputs = [
            var for var in op_desc.input_arg_names()
            if var != core.empty_var_name()
        ]
        outputs = [
            var for var in op_desc.output_arg_names()
            if var != core.empty_var_name()
        ]

        # If the outputs of grad op is empty, just remove it 
        if not outputs:
            ops_to_remove.append(op_idx)
            continue
        else:
            '''
            If the output is not empty and there is any grad input, find 
            whether there is any existing input. If not, just remove it.
            '''
            if grad_var_ins:
                existing_grad_var_ins = [
                    var for var in grad_var_ins
                    if block.desc.has_var_recursive(cpt.to_bytes(var)) or var in
                    parent_op_vars
                ]
                if not existing_grad_var_ins:
                    '''
                    FIXME(paddle-dev, zengjinle): rnn_memory_helper_grad is used
                    in recurrent op. The input of this op does not even exist in 
                    the program! Therefore, any dependency analysis would not 
                    work to this op! If I do not add the following code, this op
                    would be pruned, and the calculation result would be wrong. 
                    Maybe we should re-design this op later...  
                    '''
                    if op_desc.type() not in ['rnn_memory_helper_grad']:
                        ops_to_remove.append(op_idx)
1053
                        continue
1054

F
fengjiayi 已提交
1055 1056 1057
        new_vars = set()
        # create new gradient variables
        for grad_var_name in op_desc.output_arg_names():
M
minqiyang 已提交
1058 1059
            if block.desc.has_var_recursive(cpt.to_bytes(
                    grad_var_name)) or grad_var_name == core.empty_var_name():
F
fengjiayi 已提交
1060
                continue
M
minqiyang 已提交
1061
            block.desc.var(cpt.to_bytes(grad_var_name))
F
fengjiayi 已提交
1062
            new_vars.add(grad_var_name)
1063
            if grad_var_name not in grad_to_var:
F
fengjiayi 已提交
1064 1065 1066 1067 1068
                continue
            grad_info_map[grad_to_var[grad_var_name]] = (grad_var_name, block)
        # infer_shape and infer_type
        op_desc.infer_var_type(block.desc)
        op_desc.infer_shape(block.desc)
1069

F
fengjiayi 已提交
1070 1071
        for arg in op_desc.output_arg_names():
            if arg in new_vars:
1072
                _infer_var_data_type_shape_(arg, block)
F
update  
fengjiayi 已提交
1073

1074 1075 1076
    for op_idx in reversed(ops_to_remove):
        block.desc._remove_op(op_idx, op_idx + 1)

F
update  
fengjiayi 已提交
1077

1078 1079 1080 1081 1082 1083
def _rename_grad_(block, start_op_idx, grad_to_var, target_grad_map):
    var_map = copy.copy(target_grad_map)
    for op_idx in range(start_op_idx, block.desc.op_size()):
        op_desc = block.desc.op(op_idx)
        for name in op_desc.input_arg_names():
            if name in var_map:
W
Wu Yi 已提交
1084
                op_desc._rename_input(name, var_map[name])
1085 1086

        for name in op_desc.output_arg_names():
M
mapingshuo 已提交
1087 1088
            if "@GRAD" not in name:
                continue
1089
            if block.desc.find_var(name.encode("ascii")):
Y
Yu Yang 已提交
1090
                new_name = unique_name.generate(name)
W
Wu Yi 已提交
1091
                op_desc._rename_output(name, new_name)
1092 1093
                var_map[name] = new_name

M
minqiyang 已提交
1094
    for g, ng in six.iteritems(var_map):
1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
        if g in grad_to_var:
            grad_to_var[ng] = grad_to_var[g]
            grad_to_var.pop(g)


def _get_stop_gradients_(program):
    no_grad_dict = dict()
    assert isinstance(program, framework.Program)
    for block in program.blocks:
        assert isinstance(block, framework.Block)
        block_no_grad_set = set()
1106
        for var in list(block.vars.values()):
1107 1108 1109 1110 1111 1112 1113
            assert isinstance(var, framework.Variable)
            if var.stop_gradient:
                block_no_grad_set.add(_append_grad_suffix_(var.name))
        no_grad_dict[block.idx] = block_no_grad_set
    return no_grad_dict


1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124
def _get_son_parent_block_idx_dict(program, current_block_idx):

    son_parent_block_idx_dict = collections.OrderedDict()
    while current_block_idx >= 0:
        parent_block_idx = program.block(current_block_idx).parent_idx
        son_parent_block_idx_dict[current_block_idx] = parent_block_idx
        current_block_idx = parent_block_idx

    return son_parent_block_idx_dict


1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
def _get_no_grad_set_name(no_grad_set):
    no_grad_set_name = set()
    if no_grad_set is not None:
        if isinstance(no_grad_set, (set, list, tuple)):
            for i, no_grad_var in enumerate(no_grad_set):
                if isinstance(no_grad_var, framework.Variable):
                    no_grad_set_name.add(no_grad_var.name)
                elif isinstance(no_grad_var, six.string_types):
                    no_grad_set_name.add(no_grad_var)
                else:
                    raise TypeError(
                        "The type of no_grad_set's member must be paddle.fluid.Variable or str, but received %s."
                        % (type(no_grad_var)))
        else:
            raise TypeError(
                "The type of no_grad_set should be set or list or tuple, but received {}".
                format(type(no_grad_set)))
    return no_grad_set_name


M
mapingshuo 已提交
1145 1146 1147 1148 1149
def append_backward(loss,
                    parameter_list=None,
                    no_grad_set=None,
                    callbacks=None,
                    checkpoints=None):
1150
    """
1151
    This function appends backward part to main_program.
F
fengjiayi 已提交
1152

1153 1154
    A complete neural network training is made up of forward and backward
    propagation. However, when we configure a network, we only need to
1155 1156
    specify its forward part. This function uses the chain rule to automatically
    generate the backward part according to the forward part.
F
fengjiayi 已提交
1157

1158 1159
    In most cases, users do not need to invoke this function manually.
    It will be automatically invoked by the optimizer's `minimize` function.
F
fengjiayi 已提交
1160

1161 1162
    Parameters:
        loss( :ref:`api_guide_Variable_en` ): The loss variable of the network.
1163 1164
        parameter_list(list[Variable|str], optional): List of Parameters or Parameter.names
                                           that need to be updated by optimizers.
1165
                                           If it is None, all parameters
F
fengjiayi 已提交
1166
                                           will be updated.
1167
                                           Default: None.
1168
        no_grad_set(set[Variable|str], optional): Set of Variables or Variable.names in the :ref:`api_guide_Block_en` 0 whose gradients
1169
                               should be ignored. All variables with
1170
                               `stop_gradient=True` from all blocks will
F
fengjiayi 已提交
1171
                               be automatically added into this set.
1172
                               If this parameter is not None, the Variables or Variable.names in this set will be added to the default set.
1173
                               Default: None.
1174
        callbacks(list[callable object], optional): List of callback functions.
1175
                                               The callbacks are used for
1176 1177 1178 1179 1180 1181
                                               doing some custom jobs during
                                               backward part building. All
                                               callable objects in it will
                                               be invoked once each time a
                                               new gradient operator is added
                                               into the program. The callable
Z
zhangchunle 已提交
1182
                                               object must have two input
1183
                                               parameters: 'block' and 'context'.
1184
                                               The 'block' is the :ref:`api_guide_Block_en` which
1185 1186 1187 1188
                                               the new gradient operator will
                                               be added to. The 'context' is a
                                               map, whose keys are gradient
                                               variable names and values are
1189
                                               corresponding original :ref:`api_guide_Variable_en` .
1190 1191 1192 1193 1194 1195
                                               In addition to this, the 'context'
                                               has another special key-value pair:
                                               the key is string '__current_op_desc__'
                                               and the value is the op_desc of the
                                               gradient operator who has just
                                               triggered the callable object.
1196
                                               Default: None.
F
fengjiayi 已提交
1197 1198

    Returns:
1199 1200
        list of tuple ( :ref:`api_guide_Variable_en` , :ref:`api_guide_Variable_en` ): Pairs of parameter and its corresponding gradients.
        The key is the parameter and the value is gradient variable.
F
fengjiayi 已提交
1201 1202 1203 1204 1205 1206 1207

    Raises:
        AssertionError: If `loss` is not an instance of Variable.

    Examples:
        .. code-block:: python

1208
            import paddle.fluid as fluid
L
lujun 已提交
1209

1210 1211 1212 1213
            x = fluid.data(name='x', shape=[None, 13], dtype='int64')
            y = fluid.data(name='y', shape=[None, 1], dtype='float32')
            x_emb = fluid.embedding(x, size=[100, 256])
            y_predict = fluid.layers.fc(input=x_emb, size=1, act=None, name='my_fc')
L
lujun 已提交
1214
            loss = fluid.layers.square_error_cost(input=y_predict, label=y)
F
fengjiayi 已提交
1215
            avg_loss = fluid.layers.mean(loss)
1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242

            # Get all weights in main_program, not include bias.
            all_weights = [param for param in fluid.default_main_program().block(0).all_parameters() if 'w_' in param.name]
            all_weights_name = [w.name for w in all_weights]

            # return all param_grads needed to be updated if parameter_list set default None.
            p_g_list1 = fluid.backward.append_backward(loss=avg_loss)
            # output: [(embedding_0.w_0, embedding_0.w_0@GRAD), (my_fc.w_0, my_fc.w_0@GRAD), (my_fc.b_0, my_fc.b_0@GRAD)]

            # return the param_grads corresponding to parameter_list that can be list of param (Variable).
            p_g_list2 = fluid.backward.append_backward(loss=avg_loss, parameter_list=all_weights)
            # output: [(embedding_0.w_0, embedding_0.w_0@GRAD), (my_fc.w_0, my_fc.w_0@GRAD)]

            # parameter_list can be list of param.name (str).
            p_g_list3 = fluid.backward.append_backward(loss=avg_loss, parameter_list=all_weights_name)
            # output: [(embedding_0.w_0, embedding_0.w_0@GRAD), (my_fc.w_0, my_fc.w_0@GRAD)]

            # no_grad_set can be set of Variables that means grad will be cut off from these Variables.
            p_g_list4 = fluid.backward.append_backward(loss=avg_loss, no_grad_set=set([x_emb]))
            # output: [(my_fc.w_0, my_fc.w_0@GRAD), (my_fc.b_0, my_fc.b_0@GRAD)]

            # no_grad_set can be set of Variable.name when the Variable is created inside layers and can't be specified explicitly.
            p_g_list5 = fluid.backward.append_backward(loss=avg_loss, no_grad_set=set(['my_fc.b_0']))
            # output: [(embedding_0.w_0, embedding_0.w_0@GRAD), (my_fc.w_0, my_fc.w_0@GRAD)]

            # return [] because all param_grads are filtered by no_grad_set.
            p_g_list6 = fluid.backward.append_backward(loss=avg_loss, parameter_list=all_weights, no_grad_set=set(all_weights))
1243

1244 1245
    """
    assert isinstance(loss, framework.Variable)
Y
yuyang18 已提交
1246

Y
Fix bug  
yuyang18 已提交
1247 1248
    if loss.op is None:
        # the loss is from a cloned program. Find loss op manually.
M
mapingshuo 已提交
1249
        _find_loss_op_(loss)
Y
Fix bug  
yuyang18 已提交
1250

W
Wu Yi 已提交
1251 1252 1253
    loss.op._set_attr(core.op_proto_and_checker_maker.kOpRoleAttrName(),
                      int(core.op_proto_and_checker_maker.OpRole.Forward) |
                      int(core.op_proto_and_checker_maker.OpRole.Loss))
Y
yuyang18 已提交
1254

Y
Yang Yang 已提交
1255 1256
    if callbacks is not None:
        isinstance(callbacks, list)
Y
Yu Yang 已提交
1257

F
fengjiayi 已提交
1258
    program = loss.block.program
1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
    root_block = program.block(0)
    current_block_idx = program.current_block_idx
    current_block = program.block(current_block_idx)

    is_in_control_flow = current_block_idx != 0

    # Double grad is not supported in sub-block (control flow)
    if not is_in_control_flow:
        # _appending_grad_times used for double grad
        program._appending_grad_times += 1
1269

F
fengjiayi 已提交
1270
    if no_grad_set is None:
1271
        no_grad_set = set()
1272 1273
    else:
        no_grad_set = _get_no_grad_set_name(copy.copy(no_grad_set))
1274
    no_grad_dict = _get_stop_gradients_(program)
1275 1276
    # no_grad_set only contains vars in block 0
    # Todo(liym27): support vars in sub block
1277
    no_grad_dict[0].update(list(map(_append_grad_suffix_, no_grad_set)))
Y
Yu Yang 已提交
1278

1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
    # Currently it is only to support the optimizer.minimize
    # in a switch branch, which can append_backward in a sub_block.
    # Note: while_loop is in control flow, but it makes no sense to call optimizer in while.
    # Todo: report error when it is in while_loop
    if is_in_control_flow:
        # create grad block if in switch control flow.
        target_grad_block = program._create_block(
            parent_idx=current_block.parent_idx)
        target_grad_block._set_forward_block_idx(current_block_idx)
        # after _create_block, program.current_block changes
    else:
        target_grad_block = root_block

    son_parent_block_idx_dict = _get_son_parent_block_idx_dict(
        program, current_block_idx)

    block_fwd_op_num_dict = {}  # block_id: fwd_op_num
    for idx in son_parent_block_idx_dict:
        block_fwd_op_num_dict[idx] = program.block(idx).desc.op_size()
F
fengjiayi 已提交
1298

F
fengjiayi 已提交
1299 1300
    grad_to_var = dict()

M
mapingshuo 已提交
1301
    op_desc = _create_loss_op_desc_(loss)
1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360
    target_grad_block.desc.append_op().copy_from(op_desc)

    for block_idx in son_parent_block_idx_dict:
        block = program.block(block_idx)

        block_no_grad_set = set(
            map(_strip_grad_suffix_, no_grad_dict[block_idx]))
        op_path = _find_op_path_(block, [loss], [], block_no_grad_set)

        no_grad_vars = _find_no_grad_vars(block, op_path, [loss],
                                          block_no_grad_set)

        block_no_grad_set.update(no_grad_vars)
        no_grad_dict[block_idx].update(
            list(map(_append_grad_suffix_, block_no_grad_set)))

        input_grad_names_set = None
        # For double backward, input_grad_names is used for filtering
        # some non-used gradients op(s).

        # Todo(liym27): need a better design.
        # not support double grad in control flow sub-block now.
        if not is_in_control_flow:
            if program._appending_grad_times > 1:
                input_grad_names_set = set([_append_grad_suffix_(loss.name)])

        # Todo: support _append_backward_ops_with_checkpoints_ in
        #  sub-block (control flow)
        if checkpoints != None and \
                isinstance(checkpoints, list) and \
                len(checkpoints) > 0:
            program_stat, checkpoint_names, \
            vars_should_be_hold, \
            recompute_segments = \
                _append_backward_ops_with_checkpoints_(
                    root_block,
                    op_path,
                    root_block,
                    no_grad_dict,
                    grad_to_var,
                    checkpoints)
        else:
            _append_backward_ops_(
                block,  # the block where forward ops are in
                op_path,
                target_grad_block,
                no_grad_dict,
                grad_to_var,
                callbacks,
                input_grad_names_set=input_grad_names_set)

    grad_info_map = dict()

    # if in control flow, target_grad_block is a created new block which only contains grad ops,
    # so fwd_op_num is set to 0.
    fwd_op_num = block_fwd_op_num_dict[
        current_block_idx] if not is_in_control_flow else 0

    # Because append_backward may be called multiple times,
1361 1362
    # we need rename the internal gradient variables so that they have
    # different names.
1363
    _rename_grad_(target_grad_block, fwd_op_num, grad_to_var, {})
1364

1365 1366
    _append_backward_vars_(target_grad_block, fwd_op_num, grad_to_var,
                           grad_info_map)
F
fengjiayi 已提交
1367

F
fengjiayi 已提交
1368
    program.current_block_idx = current_block_idx
W
Wu Yi 已提交
1369
    program._sync_with_cpp()
F
fengjiayi 已提交
1370

1371
    if parameter_list is not None:
1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385
        if not isinstance(parameter_list, (list, tuple, set)):
            raise TypeError(
                "The type of parameter_list argument must be list or tuple or set, but received %s."
                % (type(parameter_list)))
        parameters = []
        for i, param in enumerate(parameter_list):
            if isinstance(param, framework.Variable):
                parameters.append(param.name)
            elif isinstance(param, six.string_types):
                parameters.append(param)
            else:
                raise TypeError(
                    "The type of parameter_list's member must be paddle.fluid.Variable or str, but received %s."
                    % (type(param)))
1386
    else:
F
fengjiayi 已提交
1387
        params = program.global_block().all_parameters()
C
chengduo 已提交
1388
        parameters = [param.name for param in params if param.trainable]
1389

1390
    params_and_grads = []
1391
    op_role_var_attr_name = core.op_proto_and_checker_maker.kOpRoleVarAttrName()
1392
    for param in parameters:
M
minqiyang 已提交
1393
        if cpt.to_text(param) not in grad_info_map:
F
fengjiayi 已提交
1394
            continue
F
update  
fengjiayi 已提交
1395
        grad_info = grad_info_map[param]
F
fengjiayi 已提交
1396
        grad_block = grad_info[1]
1397 1398 1399 1400
        if not grad_block.has_var(grad_info[0]):
            raise ValueError("grad block[{0}] did not have grad var {1}".format(
                grad_info[1], grad_info[0]))
        # Get the param var from the global block
F
fengjiayi 已提交
1401
        param_var = program.global_block().var(param)
1402
        grad_var = grad_block.var(grad_info[0])
1403 1404 1405 1406 1407
        if not is_in_control_flow:
            if loss.block.has_var(grad_info[0]):
                params_and_grads.append((param_var, grad_var))
            else:
                params_and_grads.append((param_var, None))
1408
        else:
1409
            params_and_grads.append((param_var, grad_var))
Y
yuyang18 已提交
1410 1411 1412 1413

    for p, g in params_and_grads:
        if g is None:
            continue
1414 1415 1416
        ops = grad_block.ops if is_in_control_flow else program.global_block(
        ).ops
        for op in reversed(ops):
Y
yuyang18 已提交
1417 1418 1419 1420 1421 1422 1423
            assert isinstance(op, framework.Operator)
            if g.name in op.output_arg_names:
                g.op = op
                break

        if g.op is None:
            raise ValueError("Unexpected branch")
Y
yuyang18 已提交
1424
        attr_val = [p.name, g.name]
Y
yuyang18 已提交
1425 1426
        if g.op.has_attr(op_role_var_attr_name):
            attr_val.extend(g.op.attr(op_role_var_attr_name))
W
Wu Yi 已提交
1427
        g.op._set_attr(op_role_var_attr_name, attr_val)
Y
yuyang18 已提交
1428

1429
    return params_and_grads
1430 1431 1432 1433 1434 1435 1436 1437


def _as_list(x):
    if x is None:
        return []
    return list(x) if isinstance(x, collections.Sequence) else [x]


1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487
def _is_ancestor_block(ancestor_block, block):
    prog = block.program
    ancestor_idx = ancestor_block.idx
    parent_idx = block.parent_idx

    while parent_idx != -1:
        if parent_idx == ancestor_idx:
            return True
        parent_idx = prog.block(parent_idx).parent_idx

    return False


def _get_output_names(cur_block, targets):
    """
    In `cur_block`, get output names those linked to targets.
    NOTE:
    1. `targets` can be in `cur_block`;
    Usually, `targets` is in `cur_block`. However, considering control flow,
    2. `targets` may be in sub-block but `cur_block` is an ancestor of `targets[0].block`;
    3. `targets` may be in the block which is ancestor of `cur_block`.
    """

    block = targets[0].block if targets else cur_block
    prog = cur_block.program
    if _is_ancestor_block(block, cur_block):
        return set()

    current_output_names = set([out.name for out in targets])

    # if `cur_block` is an ancestor of `targets[0].block`, run while loop
    while block.idx != cur_block.idx:
        assert block.parent_idx != -1
        parent_block = prog.block(block.parent_idx)

        parent_block_output_names = set()
        for op in reversed(block.ops):
            if _some_in_set_(op.desc.output_arg_names(), current_output_names):
                for name in op.desc.input_arg_names():
                    current_output_names.add(name)
                    if not block.desc.find_var(cpt.to_bytes(name)) \
                            and parent_block.desc.find_var(cpt.to_bytes(name)):
                        parent_block_output_names.add(name)

        block = parent_block
        current_output_names = parent_block_output_names

    return current_output_names


1488 1489 1490
def _find_no_grad_vars(block, op_path, targets, no_grad_set):
    """
    Find the vars which is not used in the program, and
1491
    those vars belong to no_grad_var.
1492
    """
1493
    output_names = _get_output_names(block, targets)
1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507
    no_grad_var = []
    for i, op in reversed(list(enumerate(op_path))):
        # If the op has sub_block, it is too complicated to find the correct no_grad_var.
        if not op.has_attr("sub_block"):
            for out_var in op.desc.output_arg_names():
                if out_var not in output_names and out_var not in op.desc.input_arg_names(
                ) and not block.vars[out_var].stop_gradient:
                    no_grad_var.append(out_var)
        for name in op.desc.input_arg_names():
            if name not in no_grad_set:
                output_names.add(name)
    return set(no_grad_var)


1508 1509 1510 1511 1512
def _find_op_path_(block, outputs, inputs, no_grad_set):
    """
    no_grad_set will also be changed
    """
    input_names = set([inp.name for inp in inputs])
1513
    output_names = _get_output_names(block, outputs)
1514 1515 1516 1517 1518 1519

    relevant_op_flags = [True] * len(block.ops)

    # All the inputs of the block are used if inputs is empty,
    if inputs:
        for i, op in enumerate(block.ops):
1520 1521 1522
            if _some_in_set_(
                    op.desc.input_arg_names(),
                    input_names) and core.has_non_empty_grad_op_maker(op.type):
1523 1524 1525 1526 1527 1528 1529
                for name in op.desc.output_arg_names():
                    if name not in no_grad_set:
                        input_names.add(name)
            else:
                relevant_op_flags[i] = False

    for i, op in reversed(list(enumerate(block.ops))):
1530 1531 1532
        if _some_in_set_(
                op.desc.output_arg_names(),
                output_names) and core.has_non_empty_grad_op_maker(op.type):
1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
            for name in op.desc.input_arg_names():
                if name not in no_grad_set:
                    output_names.add(name)
        else:
            relevant_op_flags[i] = False

    op_path = [
        block.ops[i] for i in range(len(block.ops)) if relevant_op_flags[i]
    ]

    if inputs:
        for op in op_path:
            for name in op.desc.input_arg_names():
1546
                if name not in input_names and block.vars[name].stop_gradient:
1547 1548 1549 1550 1551 1552 1553
                    no_grad_set.add(name)

    return op_path


def calc_gradient(targets, inputs, target_gradients=None, no_grad_set=None):
    """
1554
    Backpropagate the gradients of targets to inputs.
1555 1556 1557 1558

    Args:
        targets(Variable|list[Variable]): The target variables
        inputs(Variable|list[Variable]): The input variables
1559
        target_gradients (Variable|list[Variable], optional): The gradient variables
1560 1561
            of targets which has the same shape with targets, If None, ones will
            be created for them.
1562 1563 1564 1565 1566 1567
        no_grad_set(set[Variable|str], optional): Set of Variables or Variable.names in the :ref:`api_guide_Block_en` 0 whose gradients
                               should be ignored. All variables with
                               `stop_gradient=True` from all blocks will
                               be automatically added into this set.
                               If this parameter is not None, the Variables or Variable.names in this set will be added to the default set.
                               Default: None.
1568 1569

    Return:
1570
        (list[Variable]): A list of gradients for inputs
1571 1572 1573 1574 1575 1576 1577 1578 1579
        If an input does not affect targets, the corresponding gradient variable
        will be None
    """
    targets = _as_list(targets)
    inputs = _as_list(inputs)
    target_gradients = _as_list(target_gradients)

    block = targets[0].block
    prog = block.program
1580 1581
    # increase appending gradients times
    prog._appending_grad_times += 1
1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592
    block_idx = block.idx

    if not target_gradients:
        target_gradients = [None] * len(targets)

    if len(targets) != len(target_gradients):
        raise ValueError(
            "Should have the same number of target_gradients as targets")

    if no_grad_set is None:
        no_grad_set = set()
1593 1594
    else:
        no_grad_set = _get_no_grad_set_name(copy.copy(no_grad_set))
1595
    no_grad_dict = _get_stop_gradients_(prog)
1596
    no_grad_dict[0].update(list(map(_append_grad_suffix_, no_grad_set)))
1597 1598 1599

    fwd_op_num = block.desc.op_size()

1600 1601
    input_grad_names_set = set()

1602 1603 1604 1605 1606
    target_grad_map = {}
    for i, grad in enumerate(target_gradients):
        target = targets[i]
        if grad is None:
            grad_name = _append_grad_suffix_(target.name)
L
lvmengsi 已提交
1607 1608 1609 1610 1611
            target_shape = target.name + '_shape'
            block.desc.append_op().copy_from(
                _create_op_desc_("shape", {'Input': [target.name]},
                                 {"Out": [target_shape]}, {}))
            input_grad_names_set.add(target_shape)
L
liym27 已提交
1612
            op_desc = _create_op_desc_("fill_constant",
L
lvmengsi 已提交
1613
                                       {"ShapeTensor": [target_shape]},
1614
                                       {"Out": [grad_name]}, {
1615
                                           "shape": target.shape,
1616 1617 1618
                                           "value": 1.0,
                                           "dtype": target.dtype,
                                       })
L
liym27 已提交
1619

1620
            block.desc.append_op().copy_from(op_desc)
1621
            input_grad_names_set.add(grad_name)
1622 1623 1624 1625 1626 1627 1628 1629
        else:
            if target.block.idx != block_idx or target.block.program != prog:
                raise ValueError("all targets must be in the same block")
            if target.shape != grad.shape:
                raise ValueError(
                    "The shapes of target and grad are different: %s %s" % (
                        target.name, grad.name))
            target_grad_map[_append_grad_suffix_(target.name)] = grad.name
1630 1631 1632 1633 1634 1635
            input_grad_names_set.add(grad.name)

    # For double backward, input_grad_names is used for filter
    # some non-used gradients op.
    if prog._appending_grad_times == 1:
        input_grad_names_set = None
1636 1637 1638 1639 1640 1641 1642

    for input in inputs:
        if input.block.program != prog:
            raise "input must be in the same program as targets"

    block_no_grad_set = set(map(_strip_grad_suffix_, no_grad_dict[0]))
    op_path = _find_op_path_(block, targets, inputs, block_no_grad_set)
1643
    no_grad_dict[0].update(list(map(_append_grad_suffix_, block_no_grad_set)))
1644 1645
    grad_to_var = dict()
    grad_info_map = dict()
1646 1647 1648 1649 1650 1651 1652
    _append_backward_ops_(
        block,
        op_path,
        block,
        no_grad_dict,
        grad_to_var,
        input_grad_names_set=input_grad_names_set)
1653 1654 1655 1656 1657 1658 1659

    # Because calc_gradient may be called multiple times,
    # we need rename the internal gradient variables so that they have
    # different names.
    _rename_grad_(block, fwd_op_num, grad_to_var, target_grad_map)

    _append_backward_vars_(block, fwd_op_num, grad_to_var, grad_info_map)
W
Wu Yi 已提交
1660
    prog._sync_with_cpp()
1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675

    grad_vars = []
    for input_var in inputs:
        if input_var.name not in grad_info_map:
            grad_vars.append(None)
        else:
            grad_info = grad_info_map[input_var.name]
            grad_block = grad_info[1]
            grad_var = grad_block.var(grad_info[0])
            grad_vars.append(grad_var)

    if len(grad_vars) == 1:
        return grad_vars[0]
    else:
        return grad_vars
1676 1677 1678 1679 1680 1681 1682 1683 1684


def gradients(targets, inputs, target_gradients=None, no_grad_set=None):
    """
    Backpropagate the gradients of targets to inputs.

    Args:
        targets (Variable|list[Variable]): The target variables.
        inputs (Variable|list[Variable]): The input variables.
1685
        target_gradients (Variable|list[Variable], optional): The gradient variables
1686 1687
            of targets which has the same shape with targets, If None, ones will
            be created for them.
1688 1689 1690 1691
        no_grad_set (set[Variable|str], optional): Set of Variables or Variable.names in the :ref:`api_guide_Block_en` 0 whose gradients
            should be ignored. All variables with `stop_gradient=True` from all blocks will
            be automatically added into this set. If this parameter is not None, the Variables or Variable.names
            in this set will be added to the default set. Default: None.
1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702

    Return:
        (list[Variable]): A list of gradients for inputs
        If an input does not affect targets, the corresponding gradient variable
        will be None.

    Examples:
        .. code-block:: python

            import paddle.fluid as fluid

1703
            x = fluid.data(name='x', shape=[None,2,8,8], dtype='float32')
1704 1705 1706 1707 1708 1709 1710 1711 1712 1713
            x.stop_gradient=False
            y = fluid.layers.conv2d(x, 4, 1, bias_attr=False)
            y = fluid.layers.relu(y)
            y = fluid.layers.conv2d(y, 4, 1, bias_attr=False)
            y = fluid.layers.relu(y)
            z = fluid.gradients([y], x)
            print(z)
    """
    outs = calc_gradient(targets, inputs, target_gradients, no_grad_set)
    return _as_list(outs)