remove conflict

.copyright.hook

+from __future__ import absolute_import
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import argparse
+import io, re
+import sys, os
+import subprocess
+import platform
+
+COPYRIGHT = '''
+  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+'''
+
+LANG_COMMENT_MARK = None
+
+NEW_LINE_MARK = None
+
+COPYRIGHT_HEADER = None
+
+if platform.system() == "Windows":
+    NEW_LINE_MARK = "\r\n"
+else:
+    NEW_LINE_MARK = '\n'
+    COPYRIGHT_HEADER = COPYRIGHT.split(NEW_LINE_MARK)[1]
+    p = re.search('(\d{4})', COPYRIGHT_HEADER).group(0)
+    process = subprocess.Popen(["date", "+%Y"], stdout=subprocess.PIPE)
+    date, err = process.communicate()
+    date = date.decode("utf-8").rstrip("\n")
+    COPYRIGHT_HEADER = COPYRIGHT_HEADER.replace(p, date)
+
+
+def generate_copyright(template, lang='C'):
+    if lang == 'Python':
+        LANG_COMMENT_MARK = '#'
+    else:
+        LANG_COMMENT_MARK = "//"
+
+    lines = template.split(NEW_LINE_MARK)
+    ans = LANG_COMMENT_MARK + " " + COPYRIGHT_HEADER + NEW_LINE_MARK
+    for lino, line in enumerate(lines):
+        if lino == 0 or lino == 1 or lino == len(lines) - 1: continue
+        ans += LANG_COMMENT_MARK + " " + line + NEW_LINE_MARK
+
+    return ans + "\n"
+
+
+def lang_type(filename):
+    if filename.endswith(".py"):
+        return "Python"
+    elif filename.endswith(".h"):
+        return "C"
+    elif filename.endswith(".hpp"):
+        return "C"
+    elif filename.endswith(".cc"):
+        return "C"
+    elif filename.endswith(".cpp"):
+        return "C"
+    elif filename.endswith(".cu"):
+        return "C"
+    elif filename.endswith(".cuh"):
+        return "C"
+    elif filename.endswith(".go"):
+        return "C"
+    elif filename.endswith(".proto"):
+        return "C"
+    else:
+        print("Unsupported filetype")
+        exit(0)
+
+
+def main(argv=None):
+    parser = argparse.ArgumentParser(
+        description='Checker for copyright declaration.')
+    parser.add_argument('filenames', nargs='*', help='Filenames to check')
+    args = parser.parse_args(argv)
+
+    retv = 0
+    for filename in args.filenames:
+        first_line = io.open(filename).readline()
+        if "COPYRIGHT" in first_line.upper() : continue
+        original_contents = io.open(filename).read()
+        new_contents = generate_copyright(
+            COPYRIGHT, lang_type(filename)) + original_contents
+        print('Auto Insert Copyright Header {}'.format(filename))
+        retv = 1
+        with io.open(filename, 'w') as output_file:
+            output_file.write(new_contents)
+
+    return retv
+
+
+if __name__ == '__main__':
+    exit(main())

.pre-commit-config.yaml

@@ -31,3 +31,11 @@
     -   id: go-fmt
         types:
         - go
+-   repo: local
+    hooks:
+    -   id: copyright_checker
+        name: copyright_checker
+        entry: python ./.copyright.hook
+        language: system
+        files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto|py)$
+        exclude: (?!.*third_party)^.*$ | (?!.*book)^.*$

CODE_OF_CONDUCT.md

+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as contributors and maintainers pledge to making participation in our project and our community a harassment-free experience for everyone, regardless of age, body size, disability, ethnicity, gender identity and expression, level of experience, nationality, personal appearance, race, religion, or sexual identity and orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable behavior and are expected to take appropriate and fair corrective action in response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or reject comments, commits, code, wiki edits, issues, and other contributions that are not aligned to this Code of Conduct, or to ban temporarily or permanently any contributor for other behaviors that they deem inappropriate, threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies both within project spaces and in public spaces when an individual is representing the project or its community. Examples of representing a project or community include using an official project e-mail address, posting via an official social media account, or acting as an appointed representative at an online or offline event. Representation of a project may be further defined and clarified by project maintainers.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be reported by contacting the project team at paddle-dev@baidu.com. The project team will review and investigate all complaints, and will respond in a way that it deems appropriate to the circumstances. The project team is obligated to maintain confidentiality with regard to the reporter of an incident. Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good faith may face temporary or permanent repercussions as determined by other members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4, available at [http://contributor-covenant.org/version/1/4][version]
+
+[homepage]: http://contributor-covenant.org
+[version]: http://contributor-covenant.org/version/1/4/

CODE_OF_CONDUCT_cn.md

+# 貢獻者公約
+
+## 我們的承諾
+
+為了促進一個開放透明且受歡迎的環境，我們作為貢獻者和維護者保證，無論年齡、種族、民族、性別認同和表達、體型、殘疾、經驗水平、國籍、個人表現、宗教或性別取向，在我們的專案以及社群的參與者都有不被騷擾的體驗。
+
+## 我們的準則
+
+舉例來說有助於創造正面環境的行為包括：
+* 使用歡迎和包容性語言
+* 尊重不同的觀點和經驗
+* 優雅地接受建設性批評
+* 關注在對於社群最好的事情上
+* 對其他社群成員的表現友善
+
+舉例來說身為參與者不能接受的行為包括：
+* 使用與性有關的言語或是圖像，以及不受歡迎的性騷擾
+* 酸民/反串/釣魚行為或進行侮辱/貶損的評論，人身攻擊及政治攻擊
+* 公開或私下的騷擾
+* 未經許可地發布他人的個人資料，例如住址或是電子地址
+* 其他可以被合理地認定為不恰當或者違反職業操守的行為
+
+## 我們的責任
+
+專案維護者有責任為"可接受的行為"準則做出詮釋，以及對已發生的不被接受的行為採取恰當且公平的糾正措施。
+
+專案維護者有權力及責任去刪除、編輯、拒絕與本行為準則有所違背的評論(comments)、提交(commits)、程式碼、wiki 編輯、問題(issues)和其他貢獻，以及專案維護者可暫時或永久性的禁止任何他們認為有不適當、威脅、冒犯、有害行為的貢獻者。
+
+## 使用範圍
+
+當一個人代表該專案或是其社群時，本行為準則適用於其專案平台和公共平台。
+
+代表專案或是社群的情況，舉例來說包括使用官方專案的電子郵件地址、通過官方的社群媒體帳號發布或線上或線下事件中擔任指定代表。
+
+該專案的呈現方式可由其專案維護者進行進一步的定義及解釋。
+
+## 強制執行
+
+可以透過paddle-dev@baidu.com，來聯繫專案團隊來報告濫用、騷擾或其他不被接受的行為。
+
+任何維護團隊認為有必要且適合的所有投訴都將進行審查及調查，並做出相對應的回應。專案小組有對事件回報者有保密的義務。具體執行的方針近一步細節可能會單獨公佈。
+
+沒有真誠的遵守或是執行本行為準則的專案維護人員，可能會因專案領導人或是其他成員的決定，暫時或是永久的取消其身份。
+
+## 來源
+
+本行為準則改編自[貢獻者公約][首頁]，版本 1.4
+可在此觀看https://www.contributor-covenant.org/zh-tw/version/1/4/code-of-conduct.html
+
+[首頁]: https://www.contributor-covenant.org

README.md

@@ -37,6 +37,7 @@ Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddl
 
       - Optimized math operations through SSE/AVX intrinsics, BLAS libraries
       (e.g. MKL, OpenBLAS, cuBLAS) or customized CPU/GPU kernels.
+      - Optimized CNN networks through MKL-DNN library.
       - Highly optimized recurrent networks which can handle **variable-length**
       sequence without padding.
       - Optimized local and distributed training for models with high dimensional

adversarial/advbox/attacks/base.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 """
 The base model of the model.
 """

adversarial/advbox/attacks/gradientsign.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 """
 This module provide the attack method for FGSM's implement.
 """
@@ -36,3 +49,39 @@ class GradientSignAttack(Attack):
 
 
 FGSM = GradientSignAttack
+
+
+class IteratorGradientSignAttack(Attack):
+    """
+    This attack was originally implemented by Alexey Kurakin(Google Brain).
+    Paper link: https://arxiv.org/pdf/1607.02533.pdf
+    """
+
+    def _apply(self, image_label, epsilons=100, steps=10):
+        """
+        Apply the iterative gradient sign attack.
+        Args:
+            image_label(list): The image and label tuple list of one element.
+            epsilons(list|tuple|int): The epsilon (input variation parameter).
+            steps(int): The number of iterator steps.
+        Return:
+            numpy.ndarray: The adversarail sample generated by the algorithm.
+        """
+        assert len(image_label) == 1
+        pre_label = np.argmax(self.model.predict(image_label))
+        gradient = self.model.gradient(image_label)
+        min_, max_ = self.model.bounds()
+
+        if not isinstance(epsilons, Iterable):
+            epsilons = np.linspace(0, 1, num=epsilons + 1)
+
+        for epsilon in epsilons:
+            adv_img = image_label[0][0].reshape(gradient.shape)
+            for _ in range(steps):
+                gradient = self.model.gradient([(adv_img, image_label[0][1])])
+                gradient_sign = np.sign(gradient) * (max_ - min_)
+                adv_img = adv_img + epsilon * gradient_sign
+                adv_img = np.clip(adv_img, min_, max_)
+                adv_label = np.argmax(self.model.predict([(adv_img, 0)]))
+                if pre_label != adv_label:
+                    return adv_img

adversarial/advbox/models/base.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 """
 The base model of the model.
 """

adversarial/advbox/models/paddle.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from __future__ import absolute_import
 
 import numpy as np

adversarial/fluid_mnist.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 """
 CNN on mnist data using fluid api of paddlepaddle
 """

adversarial/mnist_tutorial_fgsm.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 """
 FGSM demos on mnist using advbox tool.
 """

benchmark/paddle/image/alexnet.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 
 from paddle.trainer_config_helpers import *

benchmark/paddle/image/googlenet.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 from paddle.trainer_config_helpers import *
 

benchmark/paddle/image/provider.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import io, os
 import random
 import numpy as np

benchmark/paddle/image/resnet.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 from paddle.trainer_config_helpers import *
 

benchmark/paddle/image/smallnet_mnist_cifar.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 
 from paddle.trainer_config_helpers import *

benchmark/paddle/image/vgg.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 from paddle.trainer_config_helpers import *
 

benchmark/paddle/rnn/provider.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import io, os
 import random
 import numpy as np

benchmark/paddle/rnn/rnn.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 
 from paddle.trainer_config_helpers import *

benchmark/tensorflow/image/alexnet.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from six.moves import xrange  # pylint: disable=redefined-builtin
 from datetime import datetime
 import math

benchmark/tensorflow/image/alexnet_multi_gpu.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from six.moves import xrange  # pylint: disable=redefined-builtin
 from datetime import datetime
 import math

benchmark/tensorflow/image/googlenet.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from six.moves import xrange
 from datetime import datetime
 import math

benchmark/tensorflow/image/googlenet_multi_gpu.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from six.moves import xrange  # pylint: disable=redefined-builtin
 from datetime import datetime
 import math

benchmark/tensorflow/image/smallnet_mnist_cifar.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from six.moves import xrange  # pylint: disable=redefined-builtin
 from datetime import datetime
 import math

benchmark/tensorflow/rnn/reader.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import os.path
 import io
 import numpy as np

benchmark/tensorflow/rnn/rnn.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 from six.moves import xrange  # pylint: disable=redefined-builtin
 import math

benchmark/tensorflow/rnn/rnn_multi_gpu.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/env python
 from six.moves import xrange  # pylint: disable=redefined-builtin
 import re

cmake/external/grpc.cmake

@@ -33,7 +33,7 @@ ExternalProject_Add(
     extern_grpc
     DEPENDS protobuf zlib
     GIT_REPOSITORY "https://github.com/grpc/grpc.git"
-    GIT_TAG "v1.7.x"
+    GIT_TAG "v1.8.x"
     PREFIX          ${GRPC_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CONFIGURE_COMMAND ""

cmake/make_resource.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import os
 import re
 import sys

doc/api/v1/data_provider/src/mnist_config.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 define_py_data_sources2(

doc/api/v1/data_provider/src/mnist_provider.dict.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer.PyDataProvider2 import *
 
 

doc/api/v1/data_provider/src/sentimental_config.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 dictionary = dict()

doc/api/v1/data_provider/src/sentimental_provider.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer.PyDataProvider2 import *
 
 

doc/api/v2/fluid/layers.rst

@@ -358,6 +358,18 @@ reduce_min
 ..  autofunction:: paddle.v2.fluid.layers.reduce_min
     :noindex:
 
+
+split
+-----
+..  autofunction:: paddle.v2.fluid.layers.split
+    :noindex:
+
+
+matmul
+------
+..  autofunction:: paddle.v2.fluid.layers.matmul
+    :noindex:
+
 logsigmoid
 ----------
 ..  autofunction:: paddle.v2.fluid.layers.logsigmoid
@@ -487,3 +499,8 @@ swish
 ------
 ..  autofunction:: paddle.v2.fluid.layers.swish
     :noindex:
+
+l2_normalize
+------------
+..  autofunction:: paddle.v2.fluid.layers.l2_normalize
+    :noindex:

doc/api/v2/fluid/nets.rst

@@ -20,3 +20,14 @@ sequence_conv_pool
     :noindex:
 
 
+glu
+---
+..  autofunction:: paddle.v2.fluid.nets.glu
+    :noindex:
+
+
+dot_product_attention
+---------------------
+..  autofunction:: paddle.v2.fluid.nets.dot_product_attention
+    :noindex:
+

doc/faq/local/src/reduce_min_pool_size.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 @provider(min_pool_size=0, ...)
 def process(settings, filename):
     os.system('shuf %s > %s.shuf' % (filename, filename))  # shuffle before.

doc/faq/local/src/word2vec_config.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 ...  # the settings and define data provider is omitted.
 DICT_DIM = 3000  # dictionary dimension.
 word_ids = data_layer('word_ids', size=DICT_DIM)

doc/faq/local/src/word2vec_dataprovider.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 DICT_DIM = 3000
 
 

doc/getstarted/concepts/src/infer.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import paddle.v2 as paddle
 import numpy as np
 

doc/getstarted/concepts/src/train.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import paddle.v2 as paddle
 import numpy as np
 

doc/howto/usage/capi/organization_of_the_inputs_cn.md

@@ -19,7 +19,7 @@
 
 ### 基本使用概念
 
-- 在PaddlePaddle内部，神经网络中一个计算层的输入/输出被组织为一个 `Argument` 结构体，如果神经网络有多个输入或者多个输入，每一个输入/输入都会对应有自己的`Argument`。
+- 在PaddlePaddle内部，神经网络中一个计算层的输入/输出被组织为一个 `Argument` 结构体，如果神经网络有多个输入或者多个输出，每一个输入/输出都会对应有自己的`Argument`。
 - `Argument` 并不真正“存储”数据，而是将输入/输出信息有机地组织在一起。
 - 在`Argument`内部由`IVector`（对应着上文提到的一维整型数组）和`Matrix`（对应着上文提到的二维浮点型矩阵）来实际存储数据；由 `Sequence Start Positions` (下文详细解释) 来描述输入/输出的序列信息。
 

doc/howto/usage/cluster/fluid_cluster_train_en.md

+# Fluid Distributed Training
+
+## Introduction
+
+In this article, we'll explain how to config and run distributed training jobs with PaddlePaddle Fluid in a bare metal cluster.
+
+## Preparations
+
+### Get your cluster ready
+
+Prepare your computer nodes in the cluster. Nodes in this cluster can be of any specification that runs PaddlePaddle, and with a unique IP address assigned to it. Make sure they can communicate with each other.
+
+### Have PaddlePaddle installed
+
+PaddlePaddle must be installed on all nodes. If you have GPU cards on your nodes, be sure to properly install drivers and CUDA libraries.
+
+PaddlePaddle build and installation guide can be found from [here](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/index_en.html).
+
+### Update training script
+
+#### Non-cluster training script
+
+Let's take [Deep Learning 101](http://www.paddlepaddle.org/docs/develop/book/01.fit_a_line/index.html)'s first chapter: "fit a line" as an example.
+
+This demo's non-cluster version with fluid API is as follows:
+
+``` python
+import paddle.v2 as paddle
+import paddle.v2.fluid as fluid
+
+x = fluid.layers.data(name='x', shape=[13], dtype='float32')
+y_predict = fluid.layers.fc(input=x, size=1, act=None)
+y = fluid.layers.data(name='y', shape=[1], dtype='float32')
+
+cost = fluid.layers.square_error_cost(input=y_predict, label=y)
+avg_cost = fluid.layers.mean(x=cost)
+
+sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001)
+sgd_optimizer.minimize(avg_cost)
+
+BATCH_SIZE = 20
+
+train_reader = paddle.batch(
+    paddle.reader.shuffle(
+        paddle.dataset.uci_housing.train(), buf_size=500),
+    batch_size=BATCH_SIZE)
+
+place = fluid.CPUPlace()
+feeder = fluid.DataFeeder(place=place, feed_list=[x, y])
+exe = fluid.Executor(place)
+
+exe.run(fluid.default_startup_program())
+
+PASS_NUM = 100
+for pass_id in range(PASS_NUM):
+    fluid.io.save_persistables(exe, "./fit_a_line.model/")
+    fluid.io.load_persistables(exe, "./fit_a_line.model/")
+    for data in train_reader():
+        avg_loss_value, = exe.run(fluid.default_main_program(),
+                                  feed=feeder.feed(data),
+                                  fetch_list=[avg_cost])
+
+        if avg_loss_value[0] < 10.0:
+            exit(0)  # if avg cost less than 10.0, we think our code is good.
+exit(1)
+```
+
+We created a simple fully connected neural networks training program and handed it to the fluid executor to run for 100 passes.
+
+Now let's try to convert it to a distributed version to run in a cluster.
+
+#### Introducing parameter server
+
+As you see from the non-cluster version of training script, there is only one role in it: the trainer, who does the computing as well as holding parameters. In cluster training, since multi-trainers are working on the same task, they need one centralized place to hold and distribute parameters. This centralized place is called the Parameter Server in PaddlePaddle.
+
+![parameter server architect](src/trainer.png)
+
+Parameter Server in fluid does not only hold parameters but is also assigned with a part of the program. Trainers communicate with parameter servers via send/receive OPs. For more tech detail, please refer to this [document](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/dist_refactor/distributed_architecture.md).
+
+Now we need to create program for both trainers and parameter servers, the question is how?
+
+#### Slice the program
+
+Fluid provides a tool called "Distribute Transpiler" to automatically convert the non-cluster program into cluster program.
+
+The idea behind this tool is to find optimize OPs and gradient parameters, slice the program into 2 pieces and connect them with send/receive OP.
+
+Optimize OPs and gradient parameters can be found from the return values of optimizer's minimize function.
+
+To put them together:
+
+``` python
+... #define the program, cost, and create sgd optimizer
+
+optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost) #get optimize OPs and gradient parameters
+
+t = fluid.DistributeTranspiler() # create transpiler instance
+# slice the program into 2 pieces with optimizer_ops and gradient parameters list, as well as pserver_endpoints, which is a comma separated list of [IP:PORT] and number of trainers
+t.transpile(optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) 
+
+... #create executor
+
+# in pserver, run this
+exe.run(fluid.default_startup_program())
+#current_endpoint here means current pserver IP:PORT you wish to run on
+exe.run(t.get_pserver_program(current_endpoint, optimize_ops)) 
+
+# in trainer, run this
+... # define data reader
+exe.run(fluid.default_startup_program())
+for pass_id in range(100):
+    for data in train_reader():
+        exe.run(t.get_trainer_program())
+
+
+```
+
+### E2E demo
+
+Please find the complete demo from [here](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/fluid/tests/book_distribute/notest_dist_fit_a_line.py). In parameter server node run this in the command line:
+
+``` bash
+PSERVERS=192.168.1.2:6174 SERVER_ENDPOINT=192.168.1.2:6174 TRAINING_ROLE=PSERVER python notest_dist_fit_a_line.py
+```
+
+*please note we assume that your parameter server runs at 192.168.1.2:6174*
+
+Wait until the prompt `Server listening on 192.168.1.2:6174`
+
+Then in 2 of your trainer node run this:
+
+``` bash
+PSERVERS=192.168.1.2:6174 SERVER_ENDPOINT=192.168.1.2:6174 TRAINING_ROLE=TRAINER python notest_dist_fit_a_line.py
+```
+
+*the reason you need to run this command twice in 2 nodes is: in the script we set the trainer count to be 2. You can change this setting on line 50*
+
+Now you have 2 trainers and 1 parameter server up and running.

doc/howto/usage/cluster/src/k8s_train/start_paddle.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 #!/usr/bin/python
 # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
 #

doc/howto/usage/cluster/src/word2vec/api_train_v2.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import gzip
 import math
 

doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import math
 import os
 import paddle.v2 as paddle

go/pserver/client/c/test/test_mnist.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import paddle.v2 as paddle
 import gzip
 

go/pserver/client/c/test/test_train.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import paddle.v2 as paddle
 import paddle.v2.dataset.uci_housing as uci_housing
 import paddle.v2.master as master

paddle/api/test/testTrainConfig.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 settings(batch_size=100, learning_method=AdamOptimizer())

paddle/capi/examples/model_inference/common/common.h

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 #ifndef __CAPI_EXAMPLE_COMMON_H__
 #define __CAPI_EXAMPLE_COMMON_H__
 #include <stdio.h>

paddle/capi/examples/model_inference/dense/merge_v2_model.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.utils.merge_model import merge_v2_model
 
 from mnist_v2 import network

paddle/capi/examples/model_inference/dense/mnist_v2.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import os
 import sys
 import gzip

paddle/capi/examples/model_inference/dense/trainer_config.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 img = data_layer(name='pixel', size=784)

paddle/capi/examples/model_inference/sequence/trainer_config.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 WORD_DIM = 3000

paddle/capi/tests/test_predict_network.py

+#  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+#
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 settings(batch_size=100)

paddle/cuda/src/avx_mathfun.h

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 /*
    AVX implementation of sin, cos, sincos, exp and log
 

paddle/framework/backward_test.cc

-/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 
 #include "paddle/framework/backward.h"
 

paddle/framework/dim.h

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 #pragma once
 
 #include <iostream>

paddle/framework/dim_test.cu

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 #include <thrust/device_vector.h>
 #include <sstream>
 

paddle/framework/eigen_test.cc

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 /*
   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
   Licensed under the Apache License, Version 2.0 (the "License");

paddle/framework/lod_tensor.cc

@@ -135,6 +135,65 @@ bool operator==(const LoD &a, const LoD &b) {
   return true;
 }
 
+bool CheckLoD(const LoD &in, int tensor_height) {
+  if (in.empty()) return true;
+  for (const auto &level : in) {
+    // check: there should be more than 2 offsets existing in each level.
+    if (level.size() < 2) return false;
+    // check: the first offset(the begin offset) of each level should be 0.
+    if (level.front() != 0) return false;
+    // check: all the offsets in a level should be ascending(no same items
+    // allows).
+    if (!std::is_sorted(level.begin(), level.begin(), [](size_t a, size_t b) {
+          if (a < b) return true;
+          return false;
+        })) {
+      LOG(INFO) << "ascending error";
+      return false;
+    }
+  }
+  // check: the lowest level's last offset should equals `tensor_height` if
+  //        tensor_height>0.
+  if (tensor_height > 0 && (size_t)tensor_height != in.back().back())
+    return false;
+
+  // check: the higher level's last offset should equals the lower level's
+  // size-1.
+  // NOTE LoD store the levels from top to bottom, so the higher level goes
+  // first.
+  for (size_t level = 0; level < in.size() - 1; level++) {
+    if (in[level].back() != in[level + 1].size() - 1) return false;
+  }
+  return true;
+}
+
+bool CheckAbsLoD(const LoD &in, int tensor_height) {
+  if (in.empty()) return true;
+  for (const auto &level : in) {
+    // check: all the offsets in a level should be ascending(no same items
+    // allows).
+    if (!std::is_sorted(level.begin(), level.begin(), [](size_t a, size_t b) {
+          if (a < b) return true;
+          return false;
+        })) {
+      return false;
+    }
+
+    // check: there should be more than 2 offsets existing in each level.
+    if (level.size() < 2) return false;
+
+    // check: the first offset of each level should be 0, and the last should be
+    // the same(the height of underlying tensor).
+    if (level.front() != 0) return false;
+    if (tensor_height < 0) {
+      tensor_height = level.back();
+    } else if ((size_t)tensor_height != level.back()) {
+      return false;
+    }
+  }
+  return true;
+}
+
 using LoDAndOffset = std::pair<LoD, std::pair<size_t, size_t>>;
 LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx,
                                         size_t end_idx, size_t start_level) {
@@ -232,23 +291,32 @@ std::vector<LoDTensor> LoDTensor::SplitLoDTensor(
     const std::vector<platform::Place> places) const {
   check_memory_size();
   PADDLE_ENFORCE(lod().empty(), "Disable parallel lod for now");
-  PADDLE_ENFORCE(dims()[0] % places.size() == 0,
-                 "Batch size should be divided by places size");
-
-  std::vector<LoDTensor> lods;
-  for (size_t place_idx = 0; place_idx < places.size(); ++place_idx) {
-    int begin = place_idx * dims()[0] / places.size();
-    int end = (place_idx + 1) * dims()[0] / places.size();
+  size_t result_size = std::min(static_cast<size_t>(dims()[0]), places.size());
+  size_t remainder = dims()[0] % places.size();
+
+  std::vector<LoDTensor> results;
+  results.reserve(result_size);
+
+  int step_width = static_cast<int>(dims()[0] / result_size);
+  for (size_t i = 0; i < result_size; ++i) {
+    int begin = static_cast<int>(i * step_width);
+    int end = static_cast<int>((i + 1) * step_width);
+    if (i + 1 == places.size()) {  // last
+      end += remainder;
+    }
 
     auto src = Slice(begin, end);
-    auto &dst_place = places[place_idx];
+    auto &dst_place = places[i];
     LoDTensor dst;
-    framework::Copy(src, dst_place, &dst);
-
-    lods.emplace_back(dst);
+    if (!(dst_place == place())) {
+      framework::Copy(src, dst_place, &dst);
+    } else {  // It is no need to copy if src_place and dst_place are same.
+      dst.ShareDataWith(src);
+    }
+    results.emplace_back(dst);
   }
 
-  return lods;
+  return results;
 }
 
 // TODO(tonyyang-svail): make this function support LoD
@@ -259,12 +327,17 @@ void LoDTensor::MergeLoDTensor(
   framework::DDim new_dim = lod_tensors[0]->dims();
   std::type_index new_type = lod_tensors[0]->type();
   auto new_layout = lod_tensors[0]->layout();
+  int64_t new_height = 0;
   for (auto *lod : lod_tensors) {
-    PADDLE_ENFORCE(new_dim == lod->dims());
-    PADDLE_ENFORCE(new_type == lod->type());
-    PADDLE_ENFORCE(new_layout == lod->layout());
+    new_height += lod->dims()[0];
+    for (int i = 1; i < new_dim.size(); ++i) {
+      PADDLE_ENFORCE_EQ(new_dim[i], lod->dims()[i]);
+    }
+
+    PADDLE_ENFORCE_EQ(new_type, lod->type());
+    PADDLE_ENFORCE_EQ(new_layout, lod->layout());
   }
-  new_dim[0] *= lod_tensors.size();
+  new_dim[0] = new_height;
   Resize(new_dim);
   set_layout(new_layout);
 

paddle/framework/lod_tensor.h

@@ -71,6 +71,38 @@ LoD ToAbsOffset(const LoD& in);
 
 bool operator==(const LoD& a, const LoD& b);
 
+/*
+ * Check whether this lod's format is valid.
+ *
+ * ATTENTION:
+ *   - Empty lod is treated as valid.
+ *
+ * It will check two things:
+ *
+ *  1. all the offsets in a level should be ascending(no same items allows).
+ *  2. there should be more than 2 offsets existing in each level.
+ *  3. the higher level's last offset should equals the lower level's size-1.
+ *  4. the first offset(the begin offset) of each level should be 0.
+ *  5. the lowest level's last offset should equals `tensor_height` if
+ * tensor_height>0.
+ */
+
+bool CheckLoD(const LoD& in, int tensor_height = -1);
+/*
+ * Check whether this absolute lod's format is valid.
+ *
+ * ATTENTION:
+ *   - Empty lod is treated as valid.
+ *
+ * It will check two things:
+ *  1. all the offsets in a level should be ascending(no same items allows)
+ *  2. there should be more than 2 offsets existing in each level.
+ *  3. the first offset of each level should be 0, and the last should be the
+ *     same(the height of underlying tensor) or `tensor_height` if
+ *     tensor_height>0.
+ */
+bool CheckAbsLoD(const LoD& in, int tensor_height = -1);
+
 /*
  * LoDTensor (Level of details Tensor)
  * see https://en.wikipedia.org/wiki/Level_of_details for reference.

paddle/framework/lod_tensor_test.cc

-/*
-  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
-  Licensed under the Apache License, Version 2.0 (the "License");
-  you may not use this file except in compliance with the License.
-  You may obtain a copy of the License at
-  http://www.apache.org/licenses/LICENSE-2.0
-  Unless required by applicable law or agreed to in writing, software
-  distributed under the License is distributed on an "AS IS" BASIS,
-  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-  See the License for the specific language governing permissions and
-  limitations under the License.
-*/
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
 
 #include "paddle/framework/lod_tensor.h"
 
@@ -22,38 +23,6 @@
 namespace paddle {
 namespace framework {
 
-const int kLodTensorSize = 20 * 128;
-
-class LoDTensorTester : public ::testing::Test {
- public:
-  virtual void SetUp() override {
-    // tensor's batch_size: 30
-    // 3 levels
-    // 0 10 20
-    // 0 5 10 15 20
-    // 0 2 5 7 10 12 15 20
-    LoD lod;
-    lod.push_back(std::vector<size_t>{0, 2, 3});
-    lod.push_back(std::vector<size_t>{0, 2, 5, 8});
-    lod.push_back(std::vector<size_t>{0, 2, 5, 7, 10, 12, 15, 17, 20});
-
-    ASSERT_EQ(lod.size(), 3UL);
-
-    lod_tensor_.Resize({20 /*batch size*/, 128 /*dim*/});
-    // malloc memory
-    float* dst_ptr = lod_tensor_.mutable_data<float>(place);
-    for (int i = 0; i < kLodTensorSize; ++i) {
-      dst_ptr[i] = i;
-    }
-
-    lod_tensor_.set_lod(lod);
-  }
-
- protected:
-  platform::CPUPlace place;
-  LoDTensor lod_tensor_;
-};
-
 TEST(LodExpand, test) {
   LoD lod{{0, 2}};
   LoDTensor tensor;
@@ -131,5 +100,53 @@ TEST(LoD, ToAbsOffset) {
   EXPECT_EQ(abs_lod, expected);
 }
 
+TEST(LoD, CheckLoD) {
+  LoD relative_lod;
+  relative_lod.push_back(std::vector<size_t>({0, 2}));
+  relative_lod.push_back(std::vector<size_t>({0, 1, 3}));
+  relative_lod.push_back(std::vector<size_t>({0, 2, 4, 5}));
+
+  // check compatible
+  ASSERT_TRUE(CheckLoD(relative_lod));
+  relative_lod[1].back()++;
+  ASSERT_FALSE(CheckLoD(relative_lod));
+  relative_lod[1].back()--;  // recover it
+
+  // check empty
+  LoD empty_lod;
+  ASSERT_TRUE(CheckLoD(empty_lod));
+
+  // check less than 2 offsets in a level
+  LoD some_lod0;
+  some_lod0.push_back(std::vector<size_t>({0}));
+  ASSERT_FALSE(CheckLoD(some_lod0));
+
+  // check with underlying tensor storage.
+  ASSERT_TRUE(CheckLoD(relative_lod, 5));
+  ASSERT_FALSE(CheckLoD(relative_lod, 9));
+}
+
+TEST(LoD, CheckAbsLoD) {
+  LoD relative_lod;
+  relative_lod.push_back(std::vector<size_t>({0, 2}));
+  relative_lod.push_back(std::vector<size_t>({0, 1, 3}));
+  relative_lod.push_back(std::vector<size_t>({0, 2, 4, 5}));
+
+  auto abs_lod = ToAbsOffset(relative_lod);
+
+  ASSERT_TRUE(CheckAbsLoD(abs_lod));
+
+  // check less than 2 offsets in a level.
+
+  // check the last item should be compatible with tensor height.
+  abs_lod.back().back()++;
+  ASSERT_FALSE(CheckAbsLoD(abs_lod));
+  abs_lod.back().back()--;  // restore
+
+  // check less than 2 offsets in a lod.
+  LoD abs_lod0;
+  abs_lod0.push_back(std::vector<size_t>({0}));
+  ASSERT_FALSE(CheckAbsLoD(abs_lod0));
+}
 }  // namespace framework
 }  // namespace paddle

paddle/framework/lod_tensor_test.cu

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 /*
   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
   Licensed under the Apache License, Version 2.0 (the "License");

paddle/framework/op_registry.h

@@ -177,16 +177,16 @@ class OpKernelRegistrar : public Registrar {
 /**
  * Macro to register OperatorKernel.
  */
-#define REGISTER_OP_KERNEL(op_type, DEVICE_TYPE, place_class, ...)        \
-  STATIC_ASSERT_GLOBAL_NAMESPACE(                                         \
-      __reg_op_kernel_##op_type##_##DEVICE_TYPE##__,                      \
-      "REGISTER_OP_KERNEL must be called in global namespace");           \
-  static ::paddle::framework::OpKernelRegistrar<place_class, __VA_ARGS__> \
-      __op_kernel_registrar_##op_type##_##DEVICE_TYPE##__(#op_type,       \
-                                                          #DEVICE_TYPE);  \
-  int TouchOpKernelRegistrar_##op_type##_##DEVICE_TYPE() {                \
-    __op_kernel_registrar_##op_type##_##DEVICE_TYPE##__.Touch();          \
-    return 0;                                                             \
+#define REGISTER_OP_KERNEL(op_type, LIBRARY_TYPE, place_class, ...)        \
+  STATIC_ASSERT_GLOBAL_NAMESPACE(                                          \
+      __reg_op_kernel_##op_type##_##LIBRARY_TYPE##__,                      \
+      "REGISTER_OP_KERNEL must be called in global namespace");            \
+  static ::paddle::framework::OpKernelRegistrar<place_class, __VA_ARGS__>  \
+      __op_kernel_registrar_##op_type##_##LIBRARY_TYPE##__(#op_type,       \
+                                                           #LIBRARY_TYPE); \
+  int TouchOpKernelRegistrar_##op_type##_##LIBRARY_TYPE() {                \
+    __op_kernel_registrar_##op_type##_##LIBRARY_TYPE##__.Touch();          \
+    return 0;                                                              \
   }
 
 #define REGISTER_OP_CUDA_KERNEL(op_type, ...) \
@@ -208,14 +208,14 @@ class OpKernelRegistrar : public Registrar {
   static int use_op_itself_##op_type##_ __attribute__((unused)) = \
       TouchOpRegistrar_##op_type()
 
-#define USE_OP_DEVICE_KERNEL(op_type, DEVICE_TYPE)               \
-  STATIC_ASSERT_GLOBAL_NAMESPACE(                                \
-      __use_op_kernel_##op_type##_##DEVICE_TYPE##__,             \
-      "USE_OP_DEVICE_KERNEL must be in global namespace");       \
-  extern int TouchOpKernelRegistrar_##op_type##_##DEVICE_TYPE(); \
-  static int use_op_kernel_##op_type##_##DEVICE_TYPE##_          \
-      __attribute__((unused)) =                                  \
-          TouchOpKernelRegistrar_##op_type##_##DEVICE_TYPE()
+#define USE_OP_DEVICE_KERNEL(op_type, LIBRARY_TYPE)               \
+  STATIC_ASSERT_GLOBAL_NAMESPACE(                                 \
+      __use_op_kernel_##op_type##_##LIBRARY_TYPE##__,             \
+      "USE_OP_DEVICE_KERNEL must be in global namespace");        \
+  extern int TouchOpKernelRegistrar_##op_type##_##LIBRARY_TYPE(); \
+  static int use_op_kernel_##op_type##_##LIBRARY_TYPE##_          \
+      __attribute__((unused)) =                                   \
+          TouchOpKernelRegistrar_##op_type##_##LIBRARY_TYPE()
 
 // TODO(fengjiayi): The following macros
 // seems ugly, do we have better method?

paddle/framework/tensor_test.cc

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 /*
   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
   Licensed under the Apache License, Version 2.0 (the "License");

paddle/framework/tensor_util.h

@@ -315,9 +315,8 @@ inline void DeserializeFromStream(std::istream& is, Tensor* tensor,
           desc.data_type(),
           DeserializedDataFunctor(&buf, &cpu_tensor, ctx.GetPlace()));
       is.read(static_cast<char*>(buf), cpu_tensor.memory_size());
-      auto cpu_place = new platform::CPUPlace();
-      framework::Copy(cpu_tensor, *cpu_place, dev_ctx, tensor);
-      delete cpu_place;
+      auto dst_place = dev_ctx.GetPlace();
+      framework::Copy(cpu_tensor, dst_place, dev_ctx, tensor);
 #else
       PADDLE_THROW("Unexpected branch");
 #endif

paddle/framework/tensor_util_test.cc

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 /*
   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
   Licensed under the Apache License, Version 2.0 (the "License");

paddle/framework/variable.h

+//  Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 /*
   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
   Licensed under the Apache License, Version 2.0 (the "License");

paddle/gserver/layers/MKLDNNConcatLayer.cpp

@@ -43,7 +43,7 @@ void MKLDNNConcatLayer::reshape(
   channels_[0] = ic;
   oc = ic;
   for (size_t i = 1; i < inputLayers_.size(); i++) {
-    int batchsize, height, witdh;
+    int batchsize = 0, height = 0, witdh = 0;
     reshapeInput(batchsize, height, witdh, i);
     CHECK_EQ(bs, batchsize);
     CHECK_EQ(ih, height);
@@ -84,6 +84,7 @@ void MKLDNNConcatLayer::resetFwdBuffers(std::vector<MKLDNNMatrixPtr>& inputs,
   bool has8c = false, has16c = false, hasnc = false;
   for (size_t i = 0; i < inputs.size(); i++) {
     resetInValue(inputs[i], nullptr, i, channels_[i]);
+    inputs[i]->downSpatial();
     CHECK(inputs[i]);
     auto dm = inputs[i]->getDims();
     // inputs format can be different, but ndims must equal

paddle/gserver/tests/img_conv_cudnn.py

-#edit-mode: -*- python -*-
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 
 from paddle.trainer_config_helpers import *
 

paddle/gserver/tests/img_conv_exconv.py

-#edit-mode: -*- python -*-
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 
 from paddle.trainer_config_helpers import *
 

paddle/gserver/tests/pyDataProvider.py

-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import numpy
 import struct
 import traceback

paddle/gserver/tests/rnn_data_provider.py

-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer.PyDataProvider2 import *
 
 # Note that each config should has an independent provider

paddle/gserver/tests/sequenceGen.py

-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import os
 import sys
 

paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py

-# edit-mode: -*- python -*-
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 from paddle.trainer_config_helpers import *
 
 ######################## data source ################################

paddle/gserver/tests/sequence_recurrent.py

-#!/usr/bin/env python
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 
 from paddle.trainer_config_helpers import *
 

paddle/gserver/tests/sequence_recurrent_group.py

-#!/usr/bin/env python
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 from paddle.trainer_config_helpers import *
 
 ######################## data source ################################

paddle/gserver/tests/sequence_rnn_matched_inputs.py

-# edit-mode: -*- python -*-
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 
 from paddle.trainer_config_helpers import *
 

paddle/gserver/tests/sequence_rnn_mixed_inputs.py

-# edit-mode: -*- python -*-
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 
 from paddle.trainer_config_helpers import *
 

paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py

-#edit-mode: -*- python -*-
-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 
 from paddle.trainer_config_helpers import *
 

paddle/gserver/tests/test_PyDataProvider2.py

-# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
+#  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
 #
-# 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
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
 #
-#     http://www.apache.org/licenses/LICENSE-2.0
+#    http://www.apache.org/licenses/LICENSE-2.0
 #
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
 import random
 
 from paddle.trainer.PyDataProvider2 import *

paddle/operators/CMakeLists.txt

@@ -149,7 +149,7 @@ op_library(sequence_pool_op DEPS sequence_pooling)
 op_library(lstm_op DEPS sequence2batch lstm_compute)
 op_library(gru_op DEPS sequence2batch gru_compute)
 op_library(recurrent_op DEPS executor)
-op_library(warpctc_op DEPS dynload_warpctc sequence_padding math_function)
+op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale math_function)
 op_library(cos_sim_op DEPS cos_sim_functor)
 op_library(parallel_do_op DEPS executor)
 

paddle/operators/clip_op.cc

@@ -51,8 +51,8 @@ class ClipOpMaker : public framework::OpProtoAndCheckerMaker {
     AddComment(R"DOC(
 Clip Operator.
 
-The clip operator limits the value of given input within an interval. The interval is
-specified with arguments 'min' and 'max':
+The clip operator limits the value of given input within an interval. The
+interval is specified with arguments 'min' and 'max':
 
 $$
 Out = \min(\max(X, min), max)

paddle/operators/detail/grpc_client.cc

@@ -87,7 +87,7 @@ bool RPCClient::AsyncGetVariable(const std::string& ep,
   return true;
 }
 
-bool RPCClient::wait() {
+bool RPCClient::Wait() {
   bool ok = true;
 
   while (true) {
@@ -96,7 +96,6 @@ bool RPCClient::wait() {
     }
 
     if (!Proceed()) {
-      LOG(ERROR) << "Get meets CompletionQueue error";
       return false;
     }
   }
@@ -110,9 +109,9 @@ bool RPCClient::Proceed() {
 
   // request counts.
   if (!cq_.Next(&tag, &ok)) {
+    LOG(ERROR) << "Get meets CompletionQueue error";
     return false;
   }
-  req_count_--;
 
   GPR_ASSERT(ok);
   PADDLE_ENFORCE(tag);
@@ -120,12 +119,15 @@ bool RPCClient::Proceed() {
   // TODO(gongwb): add more retries.
   ClientBase* c = static_cast<ClientBase*>(tag);
   if (!c->status_.ok()) {
+    LOG(ERROR) << "proc param error:" << c->var_h_.String()
+               << " grpc error:" << c->status_.error_message();
     delete c;
-    return true;
+    return false;
   }
 
   c->Process();
   delete c;
+  req_count_--;
   return true;
 }
 
@@ -135,8 +137,12 @@ std::shared_ptr<grpc::Channel> RPCClient::GetChannel(const std::string& ep) {
     return it->second;
   }
 
+  grpc::ChannelArguments args;
+  args.SetMaxSendMessageSize(std::numeric_limits<int>::max());
+  args.SetMaxReceiveMessageSize(std::numeric_limits<int>::max());
+
   auto ch = std::shared_ptr<grpc::Channel>(
-      grpc::CreateChannel(ep, grpc::InsecureChannelCredentials()));
+      grpc::CreateCustomChannel(ep, grpc::InsecureChannelCredentials(), args));
 
   channels_[ep] = ch;
   return ch;

paddle/operators/detail/grpc_client.h

@@ -130,7 +130,7 @@ class RPCClient {
                         const framework::Scope& scope,
                         const std::string& var_name,
                         int64_t time_out = 600 * 1000);
-  bool wait();
+  bool Wait();
 
  private:
   bool Proceed();

paddle/operators/detail/grpc_server.cc

@@ -28,12 +28,15 @@ class RequestBase {
  public:
   explicit RequestBase(sendrecv::SendRecvService::AsyncService* service,
                        grpc::ServerCompletionQueue* cq)
-      : service_(service), cq_(cq), status_(PROCESS) {}
+      : service_(service), cq_(cq), status_(PROCESS) {
+    PADDLE_ENFORCE(cq_);
+  }
   virtual ~RequestBase() {}
   virtual void Process() { assert(false); }
 
   CallStatus Status() { return status_; }
   void SetStatus(CallStatus status) { status_ = status; }
+  virtual std::string GetReqName() { assert(false); }
 
  protected:
   grpc::ServerContext ctx_;
@@ -56,12 +59,14 @@ class RequestSend final : public RequestBase {
 
   virtual ~RequestSend() {}
 
+  virtual std::string GetReqName() { return request_.varname(); }
+
   virtual void Process() {
     MessageWithName msg_with_name =
         std::make_pair(request_.varname(), std::move(request_));
     queue_->Push(std::move(msg_with_name));
-    // TODO(gongwb): check var's info.
     responder_.Finish(reply_, grpc::Status::OK, this);
+    status_ = FINISH;
   }
 
  protected:
@@ -74,20 +79,27 @@ class RequestSend final : public RequestBase {
 class RequestGet final : public RequestBase {
  public:
   explicit RequestGet(sendrecv::SendRecvService::AsyncService* service,
-                      grpc::ServerCompletionQueue* cq, framework::Scope* scope)
-      : RequestBase(service, cq), responder_(&ctx_), scope_(scope) {
+                      grpc::ServerCompletionQueue* cq, framework::Scope* scope,
+                      const platform::DeviceContext* dev_ctx)
+      : RequestBase(service, cq),
+        responder_(&ctx_),
+        scope_(scope),
+        dev_ctx_(dev_ctx) {
     service_->RequestGetVariable(&ctx_, &request_, &responder_, cq_, cq_, this);
   }
 
   virtual ~RequestGet() {}
 
+  virtual std::string GetReqName() { return request_.varname(); }
+
   virtual void Process() {
     // proc request.
     std::string var_name = request_.varname();
     auto* var = scope_->FindVar(var_name);
-    SerializeToMessage(var_name, var, platform::CPUDeviceContext(), &reply_);
+    SerializeToMessage(var_name, var, *dev_ctx_, &reply_);
     // TODO(gongwb): check var's info.
     responder_.Finish(reply_, grpc::Status::OK, this);
+    status_ = FINISH;
   }
 
  protected:
@@ -95,11 +107,14 @@ class RequestGet final : public RequestBase {
   sendrecv::VariableMessage reply_;
   ServerAsyncResponseWriter<sendrecv::VariableMessage> responder_;
   framework::Scope* scope_;
+  const platform::DeviceContext* dev_ctx_;
 };
 
 void AsyncGRPCServer::RunSyncUpdate() {
   grpc::ServerBuilder builder;
   builder.AddListeningPort(address_, grpc::InsecureServerCredentials());
+  builder.SetMaxSendMessageSize(std::numeric_limits<int>::max());
+  builder.SetMaxReceiveMessageSize(std::numeric_limits<int>::max());
   builder.RegisterService(&service_);
 
   cq_send_ = builder.AddCompletionQueue();
@@ -155,22 +170,10 @@ void AsyncGRPCServer::TryToRegisterNewGetOne() {
   if (is_shut_down_) {
     return;
   }
-  RequestGet* get = new RequestGet(&service_, cq_get_.get(), scope_);
+  RequestGet* get = new RequestGet(&service_, cq_get_.get(), scope_, dev_ctx_);
   VLOG(4) << "create Requestget status:" << get->Status();
 }
 
-void AsyncGRPCServer::SetFinishOrDelete(RequestBase*& last) {
-  std::unique_lock<std::mutex> lock(cq_mutex_);
-  if (is_shut_down_) {
-    delete last;
-    last = NULL;
-    return;
-  }
-
-  last->SetStatus(FINISH);
-  return;
-}
-
 void AsyncGRPCServer::HandleRequest(bool wait, grpc::ServerCompletionQueue* cq,
                                     std::string cq_name,
                                     std::function<void()> TryToRegisterNewOne) {
@@ -184,13 +187,19 @@ void AsyncGRPCServer::HandleRequest(bool wait, grpc::ServerCompletionQueue* cq,
       break;
     }
 
+    PADDLE_ENFORCE(tag);
     if (wait && !done_) {
       Wait();
     }
 
     RequestBase* base = (RequestBase*)tag;
+    // reference:
+    // https://github.com/tensorflow/tensorflow/issues/5596
+    // https://groups.google.com/forum/#!topic/grpc-io/xftlRy-IQwM
+    // https://groups.google.com/forum/#!topic/grpc-io/ywATt88Ef_I
     if (!ok) {
-      VLOG(4) << cq_name << " recv no regular event";
+      LOG(WARNING) << cq_name << " recv no regular event:argument name"
+                   << base->GetReqName();
       TryToRegisterNewOne();
       delete base;
       continue;
@@ -201,7 +210,6 @@ void AsyncGRPCServer::HandleRequest(bool wait, grpc::ServerCompletionQueue* cq,
         VLOG(4) << cq_name << " status:" << base->Status();
         TryToRegisterNewOne();
         base->Process();
-        SetFinishOrDelete(base);
         break;
       }
       case FINISH: {

paddle/operators/detail/grpc_server.h

@@ -37,7 +37,7 @@ class RequestBase;
 
 class AsyncGRPCServer final : public sendrecv::SendRecvService::Service {
  public:
-  explicit AsyncGRPCServer(std::string address) { address_ = address; }
+  explicit AsyncGRPCServer(const std::string &address) : address_(address) {}
 
   void RunSyncUpdate();
 
@@ -47,6 +47,8 @@ class AsyncGRPCServer final : public sendrecv::SendRecvService::Service {
 
   void SetScope(framework::Scope *scope) { scope_ = scope; }
 
+  void SetDevCtx(const platform::DeviceContext *dev_ctx) { dev_ctx_ = dev_ctx; }
+
   const MessageWithName Get() { return this->var_recv_queue_.Pop(); }
 
   void Push(const MessageWithName &msg) { this->var_recv_queue_.Push(msg); }
@@ -60,7 +62,6 @@ class AsyncGRPCServer final : public sendrecv::SendRecvService::Service {
                      std::function<void()> TryToRegisterNewOne);
   void TryToRegisterNewSendOne();
   void TryToRegisterNewGetOne();
-  void SetFinishOrDelete(RequestBase *&last);
   void ShutdownQueue();
 
  private:
@@ -74,6 +75,7 @@ class AsyncGRPCServer final : public sendrecv::SendRecvService::Service {
 
   std::string address_;
   framework::Scope *scope_;
+  const platform::DeviceContext *dev_ctx_;
   // received variable from RPC, operators fetch variable from this queue.
   SimpleBlockQueue<MessageWithName> var_recv_queue_;
 

paddle/operators/elementwise_add_op.h

@@ -28,39 +28,7 @@ template <typename DeviceContext, typename T>
 class ElementwiseAddKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    using Tensor = framework::Tensor;
-
-    auto* x = ctx.Input<Tensor>("X");
-    auto* y = ctx.Input<Tensor>("Y");
-    auto* z = ctx.Output<Tensor>("Out");
-    z->mutable_data<T>(ctx.GetPlace());
-    TransformFunctor<AddFunctor<T>, T, DeviceContext> functor(
-        x, y, z, ctx.template device_context<DeviceContext>(), AddFunctor<T>());
-
-    auto x_dims = x->dims();
-    auto y_dims = y->dims();
-    PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(),
-                      "Rank of first input must >= rank of second input.");
-
-    if (x_dims == y_dims) {
-      functor.Run();
-      return;
-    }
-
-    int axis = ctx.Attr<int>("axis");
-    axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
-    PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(),
-                   "Axis should be in range [0, x_dims)");
-
-    int pre, n, post;
-    get_mid_dims(x_dims, y_dims, axis, pre, n, post);
-    if (post == 1) {
-      functor.RunRowWise(n, pre);
-      return;
-    } else {
-      functor.RunMidWise(n, pre, post);
-      return;
-    }
+    ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(ctx);
   }
 };
 
@@ -81,23 +49,6 @@ struct ElementwiseAddGradFunctor {
   }
 };
 
-template <typename T>
-struct ElementwiseAddOneGradFunctor {
-  template <typename Device, typename X, typename Y, typename Z, typename dX,
-            typename dY, typename dZ>
-  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz) {
-    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
-    if (dx) {
-      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
-      dx_e.device(d) = dz_e;
-    }
-    if (dy) {
-      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
-      dy_e.device(d) = dz_e.sum();
-    }
-  }
-};
-
 template <typename T>
 struct ElementwiseAddBroadCastGradFunctor {
   template <typename Device, typename X, typename Y, typename Z, typename dX,
@@ -142,7 +93,6 @@ class ElementwiseAddGradKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
     ElementwiseGradCompute<DeviceContext, T, ElementwiseAddGradFunctor<T>,
-                           ElementwiseAddOneGradFunctor<T>,
                            ElementwiseAddBroadCastGradFunctor<T>,
                            ElementwiseAddBroadCast2GradFunctor<T>>(ctx);
   }

paddle/operators/elementwise_div_op.h

@@ -19,11 +19,16 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
+template <typename T>
+struct DivFunctor {
+  inline HOSTDEVICE T operator()(T a, T b) const { return a / b; }
+};
+
 template <typename DeviceContext, typename T>
 class ElementwiseDivKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseCompute<EigenDivFunctor, DeviceContext, T>(ctx);
+    ElementwiseComputeEx<DivFunctor<T>, DeviceContext, T>(ctx);
   }
 };
 
@@ -107,7 +112,6 @@ class ElementwiseDivGradKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
     ElementwiseGradCompute<DeviceContext, T, ElementwiseDivGradFunctor<T>,
-                           ElementwiseDivGradFunctor<T>,
                            ElementwiseDivBroadCastGradFunctor<T>,
                            ElementwiseDivBroadCast2GradFunctor<T>>(ctx);
   }

paddle/operators/elementwise_max_op.cc

+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#include "paddle/operators/elementwise_max_op.h"
+#include "paddle/operators/elementwise_op.h"
+
+namespace paddle {
+namespace operators {
+class ElementwiseMaxOpMaker : public ElementwiseOpMaker {
+ public:
+  ElementwiseMaxOpMaker(OpProto* proto, OpAttrChecker* op_checker)
+      : ElementwiseOpMaker(proto, op_checker) {
+    SetComment("Max", "Out = max(X, Y)");
+    AddComment(comment_);
+  }
+};
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP(elementwise_max, ops::ElementwiseOp, ops::ElementwiseMaxOpMaker,
+            elementwise_max_grad, ops::ElementwiseOpGrad);
+REGISTER_OP_CPU_KERNEL(
+    elementwise_max,
+    ops::ElementwiseMaxKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::ElementwiseMaxKernel<paddle::platform::CPUDeviceContext, double>,
+    ops::ElementwiseMaxKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwiseMaxKernel<paddle::platform::CPUDeviceContext, int64_t>);
+REGISTER_OP_CPU_KERNEL(
+    elementwise_max_grad,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CPUDeviceContext, double>,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CPUDeviceContext, int64_t>);

paddle/operators/elementwise_max_op.cu

+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#define EIGEN_USE_GPU
+#include "paddle/operators/elementwise_max_op.h"
+
+namespace ops = paddle::operators;
+
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_max,
+    ops::ElementwiseMaxKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::ElementwiseMaxKernel<paddle::platform::CUDADeviceContext, double>,
+    ops::ElementwiseMaxKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ElementwiseMaxKernel<paddle::platform::CUDADeviceContext, int64_t>);
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_max_grad,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CUDADeviceContext, double>,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ElementwiseMaxGradKernel<paddle::platform::CUDADeviceContext,
+                                  int64_t>);

paddle/operators/elementwise_max_op.h

+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#pragma once
+
+#include "paddle/operators/elementwise_op_function.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename T>
+struct MaxFunctor {
+  inline HOSTDEVICE T operator()(T a, T b) const { return a > b ? a : b; }
+};
+
+template <typename DeviceContext, typename T>
+class ElementwiseMaxKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    ElementwiseComputeEx<MaxFunctor<T>, DeviceContext, T>(ctx);
+  }
+};
+
+template <typename T>
+struct ElementwiseMaxGradFunctor {
+  template <typename Device, typename X, typename Y, typename Z, typename dX,
+            typename dY, typename dZ>
+  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz) {
+    auto x_e = framework::EigenVector<T>::Flatten(*x);
+    auto y_e = framework::EigenVector<T>::Flatten(*y);
+    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
+
+    if (dx) {
+      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
+      dx_e.device(d) = (x_e > y_e).template cast<T>() * dz_e;
+    }
+    if (dy) {
+      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
+      dy_e.device(d) = (x_e <= y_e).template cast<T>() * dz_e;
+    }
+  }
+};
+
+template <typename T>
+struct ElementwiseMaxBroadCastGradFunctor {
+  template <typename Device, typename X, typename Y, typename Z, typename dX,
+            typename dY, typename dZ, typename Pre, typename N>
+  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n) {
+    auto x_e = framework::EigenVector<T>::Flatten(*x);
+    auto y_e = framework::EigenVector<T>::Flatten(*y);
+    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
+
+    auto y_e_bcast = y_e.reshape(Eigen::DSizes<int, 2>(1, n))
+                         .broadcast(Eigen::DSizes<int, 2>(pre, 1))
+                         .reshape(Eigen::DSizes<int, 1>(x_e.size()));
+
+    if (dx) {
+      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
+      dx_e.device(d) = (x_e > y_e_bcast).template cast<T>() * dz_e;
+    }
+
+    if (dy) {
+      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
+      dy_e.device(d) = ((x_e <= y_e_bcast).template cast<T>() * dz_e)
+                           .reshape(Eigen::DSizes<int, 2>(pre, n))
+                           .sum(Eigen::array<int, 1>{{0}});
+    }
+  }
+};
+
+template <typename T>
+struct ElementwiseMaxBroadCast2GradFunctor {
+  template <typename Device, typename X, typename Y, typename Z, typename dX,
+            typename dY, typename dZ, typename Pre, typename N, typename Post>
+  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n,
+                  Post post) {
+    auto x_e = framework::EigenVector<T>::Flatten(*x);
+    auto y_e = framework::EigenVector<T>::Flatten(*y);
+    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
+
+    auto y_e_bcast = y_e.reshape(Eigen::DSizes<int, 3>(1, n, 1))
+                         .broadcast(Eigen::DSizes<int, 3>(pre, 1, post))
+                         .reshape(Eigen::DSizes<int, 1>(x_e.size()));
+    if (dx) {
+      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
+      dx_e.device(d) = (x_e > y_e_bcast).template cast<T>() * dz_e;
+    }
+
+    if (dy) {
+      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
+      dy_e.device(d) = ((x_e <= y_e_bcast).template cast<T>() * dz_e)
+                           .reshape(Eigen::DSizes<int, 3>(pre, n, post))
+                           .sum(Eigen::array<int, 2>{{0, 2}});
+    }
+  }
+};
+
+template <typename DeviceContext, typename T>
+class ElementwiseMaxGradKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    ElementwiseGradCompute<DeviceContext, T, ElementwiseMaxGradFunctor<T>,
+                           ElementwiseMaxBroadCastGradFunctor<T>,
+                           ElementwiseMaxBroadCast2GradFunctor<T>>(ctx);
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle

paddle/operators/elementwise_min_op.cc

+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#include "paddle/operators/elementwise_min_op.h"
+#include "paddle/operators/elementwise_op.h"
+
+namespace paddle {
+namespace operators {
+class ElementwiseMinOpMaker : public ElementwiseOpMaker {
+ public:
+  ElementwiseMinOpMaker(OpProto* proto, OpAttrChecker* op_checker)
+      : ElementwiseOpMaker(proto, op_checker) {
+    SetComment("Max", "Out = min(X, Y)");
+    AddComment(comment_);
+  }
+};
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP(elementwise_min, ops::ElementwiseOp, ops::ElementwiseMinOpMaker,
+            elementwise_min_grad, ops::ElementwiseOpGrad);
+REGISTER_OP_CPU_KERNEL(
+    elementwise_min,
+    ops::ElementwiseMinKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::ElementwiseMinKernel<paddle::platform::CPUDeviceContext, double>,
+    ops::ElementwiseMinKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwiseMinKernel<paddle::platform::CPUDeviceContext, int64_t>);
+REGISTER_OP_CPU_KERNEL(
+    elementwise_min_grad,
+    ops::ElementwiseMinGradKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::ElementwiseMinGradKernel<paddle::platform::CPUDeviceContext, double>,
+    ops::ElementwiseMinGradKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwiseMinGradKernel<paddle::platform::CPUDeviceContext, int64_t>);

paddle/operators/elementwise_min_op.cu

+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#define EIGEN_USE_GPU
+#include "paddle/operators/elementwise_min_op.h"
+
+namespace ops = paddle::operators;
+
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_min,
+    ops::ElementwiseMinKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::ElementwiseMinKernel<paddle::platform::CUDADeviceContext, double>,
+    ops::ElementwiseMinKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ElementwiseMinKernel<paddle::platform::CUDADeviceContext, int64_t>);
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_min_grad,
+    ops::ElementwiseMinGradKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::ElementwiseMinGradKernel<paddle::platform::CUDADeviceContext, double>,
+    ops::ElementwiseMinGradKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ElementwiseMinGradKernel<paddle::platform::CUDADeviceContext,
+                                  int64_t>);

paddle/operators/elementwise_min_op.h

+/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#pragma once
+
+#include "paddle/operators/elementwise_op_function.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename T>
+struct MinFunctor {
+  inline HOSTDEVICE T operator()(T a, T b) const { return a < b ? a : b; }
+};
+
+template <typename DeviceContext, typename T>
+class ElementwiseMinKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    ElementwiseComputeEx<MinFunctor<T>, DeviceContext, T>(ctx);
+  }
+};
+
+template <typename T>
+struct ElementwiseMinGradFunctor {
+  template <typename Device, typename X, typename Y, typename Z, typename dX,
+            typename dY, typename dZ>
+  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz) {
+    auto x_e = framework::EigenVector<T>::Flatten(*x);
+    auto y_e = framework::EigenVector<T>::Flatten(*y);
+    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
+
+    if (dx) {
+      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
+      dx_e.device(d) = (x_e < y_e).template cast<T>() * dz_e;
+    }
+    if (dy) {
+      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
+      dy_e.device(d) = (x_e >= y_e).template cast<T>() * dz_e;
+    }
+  }
+};
+
+template <typename T>
+struct ElementwiseMinBroadCastGradFunctor {
+  template <typename Device, typename X, typename Y, typename Z, typename dX,
+            typename dY, typename dZ, typename Pre, typename N>
+  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n) {
+    auto x_e = framework::EigenVector<T>::Flatten(*x);
+    auto y_e = framework::EigenVector<T>::Flatten(*y);
+    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
+
+    auto y_e_bcast = y_e.reshape(Eigen::DSizes<int, 2>(1, n))
+                         .broadcast(Eigen::DSizes<int, 2>(pre, 1))
+                         .reshape(Eigen::DSizes<int, 1>(x_e.size()));
+
+    if (dx) {
+      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
+      dx_e.device(d) = (x_e < y_e_bcast).template cast<T>() * dz_e;
+    }
+
+    if (dy) {
+      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
+      dy_e.device(d) = ((x_e >= y_e_bcast).template cast<T>() * dz_e)
+                           .reshape(Eigen::DSizes<int, 2>(pre, n))
+                           .sum(Eigen::array<int, 1>{{0}});
+    }
+  }
+};
+
+template <typename T>
+struct ElementwiseMinBroadCast2GradFunctor {
+  template <typename Device, typename X, typename Y, typename Z, typename dX,
+            typename dY, typename dZ, typename Pre, typename N, typename Post>
+  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n,
+                  Post post) {
+    auto x_e = framework::EigenVector<T>::Flatten(*x);
+    auto y_e = framework::EigenVector<T>::Flatten(*y);
+    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
+
+    auto y_e_bcast = y_e.reshape(Eigen::DSizes<int, 3>(1, n, 1))
+                         .broadcast(Eigen::DSizes<int, 3>(pre, 1, post))
+                         .reshape(Eigen::DSizes<int, 1>(x_e.size()));
+    if (dx) {
+      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
+      dx_e.device(d) = (x_e < y_e_bcast).template cast<T>() * dz_e;
+    }
+
+    if (dy) {
+      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
+      dy_e.device(d) = ((x_e >= y_e_bcast).template cast<T>() * dz_e)
+                           .reshape(Eigen::DSizes<int, 3>(pre, n, post))
+                           .sum(Eigen::array<int, 2>{{0, 2}});
+    }
+  }
+};
+
+template <typename DeviceContext, typename T>
+class ElementwiseMinGradKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    ElementwiseGradCompute<DeviceContext, T, ElementwiseMinGradFunctor<T>,
+                           ElementwiseMinBroadCastGradFunctor<T>,
+                           ElementwiseMinBroadCast2GradFunctor<T>>(ctx);
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle

paddle/operators/elementwise_mul_op.h

@@ -18,11 +18,16 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
+template <typename T>
+struct MulFunctor {
+  inline HOSTDEVICE T operator()(T a, T b) const { return a * b; }
+};
+
 template <typename DeviceContext, typename T>
 class ElementwiseMulKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseCompute<EigenMulFunctor, DeviceContext, T>(ctx);
+    ElementwiseComputeEx<MulFunctor<T>, DeviceContext, T>(ctx);
   }
 };
 
@@ -106,7 +111,6 @@ class ElementwiseMulGradKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
     ElementwiseGradCompute<DeviceContext, T, ElementwiseMulGradFunctor<T>,
-                           ElementwiseMulGradFunctor<T>,
                            ElementwiseMulBroadCastGradFunctor<T>,
                            ElementwiseMulBroadCast2GradFunctor<T>>(ctx);
   }

paddle/operators/elementwise_op.h

@@ -26,9 +26,9 @@ class ElementwiseOp : public framework::OperatorWithKernel {
   using Tensor = framework::Tensor;
   void InferShape(framework::InferShapeContext* ctx) const override {
     PADDLE_ENFORCE(ctx->HasInput("X"),
-                   "Input(X) of elementwise op should not be null");
+                   "Input(X) of elementwise op should not be null.");
     PADDLE_ENFORCE(ctx->HasInput("Y"),
-                   "Input(Y) of elementwise op should not be null");
+                   "Input(Y) of elementwise op should not be null.");
     PADDLE_ENFORCE(ctx->HasOutput("Out"),
                    "Output(Out) of elementwise op should not be null.");
 
@@ -45,12 +45,12 @@ class ElementwiseOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
   ElementwiseOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
-    AddInput("X", "(Tensor) The first input tensor of elementwise op");
-    AddInput("Y", "(Tensor) The second input tensor of elementwise op");
-    AddOutput("Out", "The output of elementwise op");
+    AddInput("X", "(Tensor), The first input tensor of elementwise op.");
+    AddInput("Y", "(Tensor), The second input tensor of elementwise op.");
+    AddOutput("Out", "The output of elementwise op.");
     AddAttr<int>("axis",
-                 "(int, default -1) The starting dimension index "
-                 "for broadcasting Y onto X")
+                 "(int, default -1). The start dimension index "
+                 "for broadcasting Y onto X.")
         .SetDefault(-1)
         .EqualGreaterThan(-1);
     comment_ = R"DOC(
@@ -58,19 +58,18 @@ Limited Elementwise {name} Operator.
 
 The equation is:
 
-.. math::
-  {equation}
+$${equation}$$
 
-X is a tensor of any dimension and the dimensions of tensor Y must be smaller than
-or equal to the dimensions of X. 
+$X$ is a tensor of any dimension and the dimensions of tensor $Y$ must be
+smaller than or equal to the dimensions of $X$.
 
 There are two cases for this operator:
-1. The shape of Y is same with X;
-2. The shape of Y is a subset of X.
+1. The shape of $Y$ is same with $X$;
+2. The shape of $Y$ is a subset of $X$.
 
 For case 2:
-Y will be broadcasted to match the shape of X and axis should be 
-the starting dimension index for broadcasting Y onto X.
+$Y$ will be broadcasted to match the shape of $X$ and axis should be
+set to index of the start dimension to broadcast $Y$ onto $X$.
 
 For example
   .. code-block:: python
@@ -81,7 +80,8 @@ For example
     shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1
     shape(X) = (2, 3, 4, 5), shape(Y) = (2), with axis=0
 
-Either of the inputs X and Y or none can carry the LoD (Level of Details) information. However, the output only shares the LoD information with input X.
+Either of the inputs $X$ and $Y$ or none can carry the LoD (Level of Details)
+information. However, the output only shares the LoD information with input $X$.
 
 )DOC";
     AddComment(comment_);

paddle/operators/elementwise_op_function.h

@@ -311,8 +311,7 @@ EIGEN_FUNCTOR(Mul, EIGEN_MUL);
 EIGEN_FUNCTOR(Div, EIGEN_DIV);
 
 template <typename DeviceContext, typename T, typename functor,
-          typename functor1, typename broadcastfunctor,
-          typename broadcast2functor>
+          typename broadcastfunctor, typename broadcast2functor>
 void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
   using Tensor = framework::Tensor;
 
@@ -341,6 +340,13 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
     return;
   }
 
+  if (y_dims.size() == 1 && y_dims[0] == 1) {
+    // y is a scalar
+    auto extended_dims = framework::vectorize(x_dims);
+    extended_dims.push_back(1);
+    x_dims = framework::make_ddim(extended_dims);
+  }
+
   int axis = ctx.Attr<int>("axis");
   axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
 
@@ -357,5 +363,50 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
     return;
   }
 }
+
+template <typename Functor, typename DeviceContext, typename T>
+void ElementwiseComputeEx(const framework::ExecutionContext& ctx) {
+  using Tensor = framework::Tensor;
+
+  auto* x = ctx.Input<Tensor>("X");
+  auto* y = ctx.Input<Tensor>("Y");
+  auto* z = ctx.Output<Tensor>("Out");
+  z->mutable_data<T>(ctx.GetPlace());
+  TransformFunctor<Functor, T, DeviceContext> functor(
+      x, y, z, ctx.template device_context<DeviceContext>(), Functor());
+
+  auto x_dims = x->dims();
+  auto y_dims = y->dims();
+  PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(),
+                    "Rank of first input must >= rank of second input.");
+
+  if (x_dims == y_dims) {
+    functor.Run();
+    return;
+  }
+
+  if (y_dims.size() == 1 && y_dims[0] == 1) {
+    // y is a scalar
+    auto extended_dims = framework::vectorize(x_dims);
+    extended_dims.push_back(1);
+    x_dims = framework::make_ddim(extended_dims);
+  }
+
+  int axis = ctx.Attr<int>("axis");
+  axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
+  PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(),
+                 "Axis should be in range [0, x_dims)");
+
+  int pre, n, post;
+  get_mid_dims(x_dims, y_dims, axis, pre, n, post);
+  if (post == 1) {
+    functor.RunRowWise(n, pre);
+    return;
+  } else {
+    functor.RunMidWise(n, pre, post);
+    return;
+  }
+}
+
 }  // namespace operators
 }  // namespace paddle

paddle/operators/elementwise_sub_op.h

@@ -18,11 +18,16 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
+template <typename T>
+struct SubFunctor {
+  inline HOSTDEVICE T operator()(T a, T b) const { return a - b; }
+};
+
 template <typename DeviceContext, typename T>
 class ElementwiseSubKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseCompute<EigenSubFunctor, DeviceContext, T>(ctx);
+    ElementwiseComputeEx<SubFunctor<T>, DeviceContext, T>(ctx);
   }
 };
 
@@ -43,23 +48,6 @@ struct ElementwiseSubGradFunctor {
   }
 };
 
-template <typename T>
-struct ElementwiseSubOneGradFunctor {
-  template <typename Device, typename X, typename Y, typename Z, typename dX,
-            typename dY, typename dZ>
-  void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz) {
-    auto dz_e = framework::EigenVector<T>::Flatten(*dz);
-    if (dx) {
-      auto dx_e = framework::EigenVector<T>::Flatten(*dx);
-      dx_e.device(d) = dz_e;
-    }
-    if (dy) {
-      auto dy_e = framework::EigenVector<T>::Flatten(*dy);
-      dy_e.device(d) = (-1.0) * dz_e.sum();
-    }
-  }
-};
-
 template <typename T>
 struct ElementwiseSubBroadCastGradFunctor {
   template <typename Device, typename X, typename Y, typename Z, typename dX,
@@ -106,7 +94,6 @@ class ElementwiseSubGradKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
     ElementwiseGradCompute<DeviceContext, T, ElementwiseSubGradFunctor<T>,
-                           ElementwiseSubOneGradFunctor<T>,
                            ElementwiseSubBroadCastGradFunctor<T>,
                            ElementwiseSubBroadCast2GradFunctor<T>>(ctx);
   }

paddle/operators/expand_op.cc

@@ -58,21 +58,21 @@ class ExpandOpMaker : public framework::OpProtoAndCheckerMaker {
   ExpandOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
-             "(Tensor, default Tensor<float>) A tensor with rank in [1, 6]."
-             "X is the input tensor to be expanded.");
+             "(Tensor, default Tensor<float>). A tensor with rank in [1, 6]."
+             "X is the input to be expanded.");
     AddOutput("Out",
-              "(Tensor, default Tensor<float>) A tensor with rank in [1, 6]."
-              "The rank of Output(Out) is same as Input(X) except that each "
-              "dimension size of Output(Out) is equal to corresponding "
-              "dimension size of Input(X) multiplying corresponding value of "
-              "Attr(expand_times).");
+              "(Tensor, default Tensor<float>). A tensor with rank in [1, 6]."
+              "The rank of Output(Out) have the same with Input(X). "
+              "After expanding, size of each dimension of Output(Out) is equal "
+              "to size of the corresponding dimension of Input(X) multiplying "
+              "the corresponding value given by Attr(expand_times).");
     AddAttr<std::vector<int>>("expand_times",
                               "Expand times number for each dimension.");
     AddComment(R"DOC(
 Expand operator tiles the input by given times number. You should set times
 number for each dimension by providing attribute 'expand_times'. The rank of X
-should be in [1, 6]. Please notice that size of 'expand_times' must be same with
-X's rank. Following is a using case:
+should be in [1, 6]. Please note that size of 'expand_times' must be the same
+with X's rank. Following is a using case:
 
 Input(X) is a 3-D tensor with shape [2, 3, 1]: