merge develop

CMakeLists.txt

@@ -213,9 +213,11 @@ include(configure)          # add paddle env configuration
 if(WITH_GPU)
     include(cuda)
     include(tensorrt)
+endif()
+if(WITH_MKL OR WITH_MKLML)
     include(external/anakin)
 elseif()
-    set(WITH_ANAKIN OFF CACHE STRING "Anakin is used in GPU only now." FORCE)
+    set(WITH_ANAKIN OFF CACHE STRING "Anakin is used in MKL only now." FORCE)
 endif()
 
 include(generic)            # simplify cmake module

Dockerfile

@@ -53,7 +53,7 @@ RUN curl -s -q https://glide.sh/get | sh
 #    and its size is only one-third of the official one.
 # 2. Manually add ~IPluginFactory() in IPluginFactory class of NvInfer.h, otherwise, it couldn't work in paddle.
 #    See https://github.com/PaddlePaddle/Paddle/issues/10129 for details.
-RUN wget -qO- http://paddlepaddledeps.bj.bcebos.com/TensorRT-4.0.0.3.Ubuntu-16.04.4.x86_64-gnu.cuda-8.0.cudnn7.0.tar.gz | \
+RUN wget -qO- http://paddlepaddledeps.cdn.bcebos.com/TensorRT-4.0.0.3.Ubuntu-16.04.4.x86_64-gnu.cuda-8.0.cudnn7.0.tar.gz | \
     tar -xz -C /usr/local && \
     cp -rf /usr/local/TensorRT/include /usr && \
     cp -rf /usr/local/TensorRT/lib /usr

README.md

@@ -76,33 +76,26 @@ pip install paddlepaddle-gpu==0.14.0.post85
 
 ## Installation
 
-It is recommended to check out the
-[Docker installation guide](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/build_and_install/docker_install_en.html)
-before looking into the
-[build from source guide](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/build_and_install/build_from_source_en.html).
+It is recommended to read [this doc](http://paddlepaddle.org/documentation/docs/zh/0.14.0/new_docs/beginners_guide/install/install_doc.html) on our website.
 
 ## Documentation
 
-We provide [English](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/index_en.html) and
-[Chinese](http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/index_cn.html) documentation.
+We provide [English](http://paddlepaddle.org/documentation/docs/en/0.14.0/getstarted/index_en.html) and
+[Chinese](http://paddlepaddle.org/documentation/docs/zh/0.14.0/new_docs/beginners_guide/index.html) documentation.
 
-- [Deep Learning 101](http://www.paddlepaddle.org/docs/develop/book/01.fit_a_line/index.html)
+- [Deep Learning 101](https://github.com/PaddlePaddle/book)
 
   You might want to start from this online interactive book that can run in a Jupyter Notebook.
 
-- [Distributed Training](http://www.paddlepaddle.org/docs/develop/documentation/en/howto/cluster/index_en.html)
+- [Distributed Training](http://paddlepaddle.org/documentation/docs/zh/0.14.0/new_docs/user_guides/howto/training/cluster_howto.html)
 
   You can run distributed training jobs on MPI clusters.
 
-- [Distributed Training on Kubernetes](http://www.paddlepaddle.org/docs/develop/documentation/en/howto/cluster/multi_cluster/k8s_en.html)
-
-   You can also run distributed training jobs on Kubernetes clusters.
-
-- [Python API](http://www.paddlepaddle.org/docs/develop/api/en/overview.html)
+- [Python API](http://paddlepaddle.org/documentation/api/zh/0.14.0/fluid.html)
 
    Our new API enables much shorter programs.
 
-- [How to Contribute](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/dev/contribute_to_paddle_en.html)
+- [How to Contribute](http://paddlepaddle.org/documentation/docs/zh/0.14.0/new_docs/advanced_usage/development/contribute_to_paddle.html)
 
    We appreciate your contributions!
 

benchmark/fluid/Dockerfile

@@ -11,6 +11,7 @@ RUN ln -s /usr/lib/x86_64-linux-gnu/libcudnn.so.7 /usr/lib/libcudnn.so && ln -s
 # Add "ENV http_proxy=http://ip:port" if your download is slow, and don't forget to unset it at runtime.
 # exmaple: unset http_proxy && unset https_proxy && python fluid_benchmark.py ...
 
+
 RUN pip install -U pip
 RUN pip install -U kubernetes paddlepaddle
 
@@ -27,5 +28,6 @@ ADD *.whl /
 RUN pip install /*.whl && rm -f /*.whl 
 
 ENV LD_LIBRARY_PATH=/usr/local/lib
-ADD fluid_benchmark.py recordio_converter.py args.py recordio_converter.py run.sh run_fluid_benchmark.sh /workspace/
+ADD fluid_benchmark.py recordio_converter.py args.py recordio_converter.py run.sh run_fluid_benchmark.sh imagenet_reader.py /workspace/
 ADD models/ /workspace/models/
+

benchmark/fluid/args.py

@@ -17,7 +17,8 @@ import argparse
 __all__ = ['parse_args', ]
 
 BENCHMARK_MODELS = [
-    "machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm"
+    "machine_translation", "resnet", "se_resnext", "vgg", "mnist",
+    "stacked_dynamic_lstm", "resnet_with_preprocess"
 ]
 
 
@@ -67,12 +68,12 @@ def parse_args():
         '--cpus',
         type=int,
         default=1,
-        help='If cpus > 1, will use ParallelDo to run, else use Executor.')
+        help='If cpus > 1, will set ParallelExecutor to use multiple threads.')
     parser.add_argument(
         '--data_set',
         type=str,
         default='flowers',
-        choices=['cifar10', 'flowers'],
+        choices=['cifar10', 'flowers', 'imagenet'],
         help='Optional dataset for benchmark.')
     parser.add_argument(
         '--infer_only', action='store_true', help='If set, run forward only.')
@@ -122,6 +123,11 @@ def parse_args():
         type=str,
         default="",
         help='Directory that contains all the training recordio files.')
+    parser.add_argument(
+        '--test_data_path',
+        type=str,
+        default="",
+        help='Directory that contains all the test data (NOT recordio).')
     parser.add_argument(
         '--use_inference_transpiler',
         action='store_true',
@@ -130,5 +136,9 @@ def parse_args():
         '--no_random',
         action='store_true',
         help='If set, keep the random seed and do not shuffle the data.')
+    parser.add_argument(
+        '--use_lars',
+        action='store_true',
+        help='If set, use lars for optimizers, ONLY support resnet module.')
     args = parser.parse_args()
     return args

benchmark/fluid/kube_gen_job.py

@@ -163,6 +163,19 @@ def gen_job():
         volumes.append({"name": "dshm", "emptyDir": {"medium": "Memory"}})
         volumeMounts.append({"mountPath": "/dev/shm", "name": "dshm"})
 
+    # add ceph volumes
+    volumes.append({
+        "name": "ceph-data",
+        "cephfs": {
+            "monitors": ["192.168.16.23:6789"],
+            "secretRef": {
+                "name": "ceph-secret"
+            },
+            "user": "admin",
+        }
+    })
+    volumeMounts.append({"mountPath": "/mnt/data", "name": "ceph-data"})
+
     tn["spec"]["template"]["spec"]["volumes"] = volumes
     tn_container["volumeMounts"] = volumeMounts
 

benchmark/fluid/models/__init__.py

@@ -13,5 +13,6 @@
 # limitations under the License.
 
 __all__ = [
-    "machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm"
+    "machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm",
+    "resnet_with_preprocess"
 ]

benchmark/fluid/models/machine_translation.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """seq2seq model for fluid."""
+
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
@@ -181,7 +182,7 @@ def lodtensor_to_ndarray(lod_tensor):
     return ndarray
 
 
-def get_model(args):
+def get_model(args, is_train, main_prog, startup_prog):
     if args.use_reader_op:
         raise Exception("machine_translation do not support reader op for now.")
     embedding_dim = 512
@@ -190,30 +191,27 @@ def get_model(args):
     dict_size = 30000
     beam_size = 3
     max_length = 250
-    avg_cost, feeding_list = seq_to_seq_net(
-        embedding_dim,
-        encoder_size,
-        decoder_size,
-        dict_size,
-        dict_size,
-        False,
-        beam_size=beam_size,
-        max_length=max_length)
-
-    # clone from default main program
-    inference_program = fluid.default_main_program().clone()
-
-    optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
-
-    train_batch_generator = paddle.batch(
-        paddle.reader.shuffle(
-            paddle.dataset.wmt14.train(dict_size), buf_size=1000),
-        batch_size=args.batch_size * args.gpus)
 
-    test_batch_generator = paddle.batch(
+    with fluid.program_guard(main_prog, startup_prog):
+        with fluid.unique_name.guard():
+            avg_cost, feeding_list = seq_to_seq_net(
+                embedding_dim,
+                encoder_size,
+                decoder_size,
+                dict_size,
+                dict_size,
+                False,
+                beam_size=beam_size,
+                max_length=max_length)
+    if is_train:
+        optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
+        optimizer.minimize(avg_cost)
+
+    batch_generator = paddle.batch(
         paddle.reader.shuffle(
-            paddle.dataset.wmt14.test(dict_size), buf_size=1000),
-        batch_size=args.batch_size)
+            paddle.dataset.wmt14.train(dict_size)
+            if is_train else paddle.dataset.wmt14.test(dict_size),
+            buf_size=1000),
+        batch_size=args.batch_size * args.gpus)
 
-    return avg_cost, inference_program, optimizer, train_batch_generator, \
-           test_batch_generator, None
+    return avg_cost, optimizer, [], batch_generator, None

benchmark/fluid/models/mnist.py

@@ -65,61 +65,50 @@ def cnn_model(data):
     return predict
 
 
-def get_model(args):
-    if args.use_reader_op:
-        filelist = [
-            os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
-        ]
-        data_file = fluid.layers.open_files(
-            filenames=filelist,
-            shapes=[[-1, 1, 28, 28], (-1, 1)],
-            lod_levels=[0, 0],
-            dtypes=["float32", "int64"],
-            thread_num=args.gpus,
-            pass_num=args.pass_num)
-        data_file = fluid.layers.double_buffer(
-            fluid.layers.batch(
-                data_file, batch_size=args.batch_size))
-        images, label = fluid.layers.read_file(data_file)
-    else:
-        images = fluid.layers.data(name='pixel', shape=[1, 28, 28], dtype=DTYPE)
-        label = fluid.layers.data(name='label', shape=[1], dtype='int64')
-
-    if args.device == 'CPU' and args.cpus > 1:
-        places = fluid.layers.get_places(args.cpus)
-        pd = fluid.layers.ParallelDo(places)
-        with pd.do():
-            predict = cnn_model(pd.read_input(images))
-            label = pd.read_input(label)
+def get_model(args, is_train, main_prog, startup_prog):
+    # NOTE: mnist is small, we don't implement data sharding yet.
+    filelist = [
+        os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
+    ]
+    with fluid.program_guard(main_prog, startup_prog):
+        if args.use_reader_op:
+            data_file_handle = fluid.layers.open_files(
+                filenames=filelist,
+                shapes=[[-1, 1, 28, 28], (-1, 1)],
+                lod_levels=[0, 0],
+                dtypes=["float32", "int64"],
+                thread_num=1,
+                pass_num=1)
+            data_file = fluid.layers.double_buffer(
+                fluid.layers.batch(
+                    data_file_handle, batch_size=args.batch_size))
+        with fluid.unique_name.guard():
+            if args.use_reader_op:
+                input, label = fluid.layers.read_file(data_file)
+            else:
+                images = fluid.layers.data(
+                    name='pixel', shape=[1, 28, 28], dtype='float32')
+                label = fluid.layers.data(
+                    name='label', shape=[1], dtype='int64')
+
+            predict = cnn_model(images)
             cost = fluid.layers.cross_entropy(input=predict, label=label)
             avg_cost = fluid.layers.mean(x=cost)
+            # Evaluator
             batch_acc = fluid.layers.accuracy(input=predict, label=label)
-
-            pd.write_output(avg_cost)
-            pd.write_output(batch_acc)
-
-        avg_cost, batch_acc = pd()
-        avg_cost = fluid.layers.mean(avg_cost)
-        batch_acc = fluid.layers.mean(batch_acc)
-    else:
-        # Train program
-        predict = cnn_model(images)
-        cost = fluid.layers.cross_entropy(input=predict, label=label)
-        avg_cost = fluid.layers.mean(x=cost)
-
-        # Evaluator
-        batch_acc = fluid.layers.accuracy(input=predict, label=label)
-
-    # inference program
-    inference_program = fluid.default_main_program().clone()
-
-    # Optimization
-    opt = fluid.optimizer.AdamOptimizer(
-        learning_rate=0.001, beta1=0.9, beta2=0.999)
+            # Optimization
+            if is_train:
+                opt = fluid.optimizer.AdamOptimizer(
+                    learning_rate=0.001, beta1=0.9, beta2=0.999)
+                opt.minimize()
+                if args.memory_optimize:
+                    fluid.memory_optimize(main_prog)
 
     # Reader
-    train_reader = paddle.batch(
-        paddle.dataset.mnist.train(), batch_size=args.batch_size * args.gpus)
-    test_reader = paddle.batch(
-        paddle.dataset.mnist.test(), batch_size=args.batch_size)
-    return avg_cost, inference_program, opt, train_reader, test_reader, batch_acc
+    if is_train:
+        reader = paddle.dataset.mnist.train()
+    else:
+        reader = paddle.dataset.mnist.test()
+    batched_reader = paddle.batch(
+        reader, batch_size=args.batch_size * args.gpus)
+    return avg_cost, opt, [batch_acc], batched_reader, data_file_handle

benchmark/fluid/models/stacked_dynamic_lstm.py

@@ -26,7 +26,6 @@ import numpy
 import paddle
 import paddle.dataset.imdb as imdb
 import paddle.fluid as fluid
-import paddle.batch as batch
 import paddle.fluid.profiler as profiler
 
 word_dict = imdb.word_dict()
@@ -43,19 +42,7 @@ def crop_sentence(reader, crop_size):
     return __impl__
 
 
-def get_model(args):
-    if args.use_reader_op:
-        raise Exception(
-            "stacked_dynamic_lstm do not support reader op for now.")
-    lstm_size = 512
-    emb_dim = 512
-    crop_size = 1500
-
-    data = fluid.layers.data(
-        name="words", shape=[1], lod_level=1, dtype='int64')
-    sentence = fluid.layers.embedding(
-        input=data, size=[len(word_dict), emb_dim])
-
+def lstm_net(sentence, lstm_size):
     sentence = fluid.layers.fc(input=sentence, size=lstm_size, act='tanh')
 
     rnn = fluid.layers.DynamicRNN()
@@ -97,31 +84,47 @@ def get_model(args):
 
     last = fluid.layers.sequence_pool(rnn(), 'last')
     logit = fluid.layers.fc(input=last, size=2, act='softmax')
-    loss = fluid.layers.cross_entropy(
-        input=logit,
-        label=fluid.layers.data(
-            name='label', shape=[1], dtype='int64'))
-    loss = fluid.layers.mean(x=loss)
+    return logit
 
-    # add acc
-    batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
-    batch_acc = fluid.layers.accuracy(input=logit, label=fluid.layers.data(name='label', \
-                shape=[1], dtype='int64'), total=batch_size_tensor)
 
-    inference_program = fluid.default_main_program().clone()
-    with fluid.program_guard(inference_program):
-        inference_program = fluid.io.get_inference_program(
-            target_vars=[batch_acc, batch_size_tensor])
-
-    adam = fluid.optimizer.Adam()
+def get_model(args, is_train, main_prog, startup_prog):
+    if args.use_reader_op:
+        raise Exception(
+            "stacked_dynamic_lstm do not support reader op for now.")
+    lstm_size = 512
+    emb_dim = 512
+    crop_size = 1500
 
-    train_reader = batch(
+    with fluid.program_guard(main_prog, startup_prog):
+        with fluid.unique_name.guard():
+            data = fluid.layers.data(
+                name="words", shape=[1], lod_level=1, dtype='int64')
+            sentence = fluid.layers.embedding(
+                input=data, size=[len(word_dict), emb_dim])
+            logit = lstm_net(sentence, lstm_size)
+            loss = fluid.layers.cross_entropy(
+                input=logit,
+                label=fluid.layers.data(
+                    name='label', shape=[1], dtype='int64'))
+            loss = fluid.layers.mean(x=loss)
+
+            # add acc
+            batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
+            batch_acc = fluid.layers.accuracy(input=logit, label=fluid.layers.data(name='label', \
+                        shape=[1], dtype='int64'), total=batch_size_tensor)
+
+            if is_train:
+                adam = fluid.optimizer.Adam()
+                adam.minimize(loss)
+
+    if is_train:
+        reader = crop_sentence(imdb.train(word_dict), crop_size)
+    else:
+        reader = crop_sentence(imdb.test(word_dict), crop_size)
+
+    batched_reader = paddle.batch(
         paddle.reader.shuffle(
-            crop_sentence(imdb.train(word_dict), crop_size), buf_size=25000),
+            reader, buf_size=25000),
         batch_size=args.batch_size * args.gpus)
-    test_reader = batch(
-        paddle.reader.shuffle(
-            crop_sentence(imdb.test(word_dict), crop_size), buf_size=25000),
-        batch_size=args.batch_size)
 
-    return loss, inference_program, adam, train_reader, test_reader, batch_acc
+    return loss, adam, [batch_acc], batched_reader, None

benchmark/fluid/models/vgg.py

@@ -25,7 +25,7 @@ import functools
 import os
 
 
-def vgg16_bn_drop(input):
+def vgg16_bn_drop(input, is_train=True):
     def conv_block(input, num_filter, groups, dropouts):
         return fluid.nets.img_conv_group(
             input=input,
@@ -46,13 +46,13 @@ def vgg16_bn_drop(input):
 
     drop = fluid.layers.dropout(x=conv5, dropout_prob=0.5)
     fc1 = fluid.layers.fc(input=drop, size=512, act=None)
-    bn = fluid.layers.batch_norm(input=fc1, act='relu')
+    bn = fluid.layers.batch_norm(input=fc1, act='relu', is_test=not is_train)
     drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5)
     fc2 = fluid.layers.fc(input=drop2, size=512, act=None)
     return fc2
 
 
-def get_model(args):
+def get_model(args, is_train, main_prog, startup_prog):
     if args.data_set == "cifar10":
         classdim = 10
         if args.data_format == 'NCHW':
@@ -65,57 +65,56 @@ def get_model(args):
             data_shape = [3, 224, 224]
         else:
             data_shape = [224, 224, 3]
+    filelist = [
+        os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
+    ]
+    with fluid.program_guard(main_prog, startup_prog):
+        if args.use_reader_op:
+            data_file_handle = fluid.layers.open_files(
+                filenames=filelist,
+                shapes=[[-1] + data_shape, (-1, 1)],
+                lod_levels=[0, 0],
+                dtypes=["float32", "int64"],
+                thread_num=1,
+                pass_num=1)
+            data_file = fluid.layers.double_buffer(
+                fluid.layers.batch(
+                    data_file_handle, batch_size=args.batch_size))
+        with fluid.unique_name.guard():
+            if args.use_reader_op:
+                images, label = fluid.layers.read_file(data_file)
+            else:
+                images = fluid.layers.data(
+                    name='data', shape=data_shape, dtype='float32')
+                label = fluid.layers.data(
+                    name='label', shape=[1], dtype='int64')
+            # Train program
+            net = vgg16_bn_drop(images, is_train=is_train)
+            predict = fluid.layers.fc(input=net, size=classdim, act='softmax')
+            cost = fluid.layers.cross_entropy(input=predict, label=label)
+            avg_cost = fluid.layers.mean(x=cost)
 
-    if args.use_reader_op:
-        filelist = [
-            os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
-        ]
-        data_file = fluid.layers.open_files(
-            filenames=filelist,
-            shapes=[[-1] + data_shape, (-1, 1)],
-            lod_levels=[0, 0],
-            dtypes=["float32", "int64"],
-            thread_num=args.gpus,
-            pass_num=args.pass_num)
-        data_file = fluid.layers.double_buffer(
-            fluid.layers.batch(
-                data_file, batch_size=args.batch_size))
-        images, label = fluid.layers.read_file(data_file)
-    else:
-        images = fluid.layers.data(
-            name='data', shape=data_shape, dtype='float32')
-        label = fluid.layers.data(name='label', shape=[1], dtype='int64')
-
-    # Train program
-    net = vgg16_bn_drop(images)
-    predict = fluid.layers.fc(input=net, size=classdim, act='softmax')
-    cost = fluid.layers.cross_entropy(input=predict, label=label)
-    avg_cost = fluid.layers.mean(x=cost)
-
-    # Evaluator
-    batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
-    batch_acc = fluid.layers.accuracy(
-        input=predict, label=label, total=batch_size_tensor)
-
-    # inference program
-    inference_program = fluid.default_main_program().clone()
-    with fluid.program_guard(inference_program):
-        inference_program = fluid.io.get_inference_program(
-            target_vars=[batch_acc, batch_size_tensor])
-
-    # Optimization
-    optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
+            # Evaluator
+            batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
+            batch_acc = fluid.layers.accuracy(
+                input=predict, label=label, total=batch_size_tensor)
+            # Optimization
+            if is_train:
+                optimizer = fluid.optimizer.Adam(
+                    learning_rate=args.learning_rate)
+                optimizer.minimize(avg_cost)
 
     # data reader
-    train_reader = paddle.batch(
+    if is_train:
+        reader = paddle.dataset.cifar.train10() \
+            if args.data_set == 'cifar10' else paddle.dataset.flowers.train()
+    else:
+        reader = paddle.dataset.cifar.test10() \
+            if args.data_set == 'cifar10' else paddle.dataset.flowers.test()
+
+    batched_reader = paddle.batch(
         paddle.reader.shuffle(
-            paddle.dataset.cifar.train10()
-            if args.data_set == 'cifar10' else paddle.dataset.flowers.train(),
-            buf_size=5120),
+            reader, buf_size=5120),
         batch_size=args.batch_size * args.gpus)
-    test_reader = paddle.batch(
-        paddle.dataset.cifar.test10()
-        if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
-        batch_size=args.batch_size)
 
-    return avg_cost, inference_program, optimizer, train_reader, test_reader, batch_acc
+    return avg_cost, optimizer, [batch_acc], batched_reader, data_file_handle

cmake/cuda.cmake

@@ -169,14 +169,19 @@ set(CUDA_PROPAGATE_HOST_FLAGS OFF)
 
 # Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
 # So, don't set these flags here.
+if (NOT WIN32) # windows msvc2015 support c++11 natively. 
+# -std=c++11 -fPIC not recoginize by msvc, -Xcompiler will be added by cmake.
 list(APPEND CUDA_NVCC_FLAGS "-std=c++11")
-list(APPEND CUDA_NVCC_FLAGS "--use_fast_math")
 list(APPEND CUDA_NVCC_FLAGS "-Xcompiler -fPIC")
+endif(NOT WIN32)
+
+list(APPEND CUDA_NVCC_FLAGS "--use_fast_math")
 # in cuda9, suppress cuda warning on eigen 
 list(APPEND CUDA_NVCC_FLAGS "-w")
 # Set :expt-relaxed-constexpr to suppress Eigen warnings
 list(APPEND CUDA_NVCC_FLAGS "--expt-relaxed-constexpr")
 
+if (NOT WIN32)
 if(CMAKE_BUILD_TYPE  STREQUAL "Debug")
     list(APPEND CUDA_NVCC_FLAGS  ${CMAKE_CXX_FLAGS_DEBUG})
 elseif(CMAKE_BUILD_TYPE  STREQUAL "Release")
@@ -187,6 +192,13 @@ elseif(CMAKE_BUILD_TYPE  STREQUAL "MinSizeRel")
     # nvcc 9 does not support -Os. Use Release flags instead
     list(APPEND CUDA_NVCC_FLAGS  ${CMAKE_CXX_FLAGS_RELEASE})
 endif()
+else(NOT WIN32)
+if(CMAKE_BUILD_TYPE STREQUAL "Release")
+  list(APPEND CUDA_NVCC_FLAGS "-O3 -DNDEBUG")
+else()
+  message(FATAL "Windows only support Release build now. Please set visual studio build type to Release, x64 build.")
+endif()
+endif(NOT WIN32)
 
 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD)
 mark_as_advanced(CUDA_SDK_ROOT_DIR CUDA_SEPARABLE_COMPILATION)

cmake/external/anakin.cmake

@@ -16,16 +16,6 @@ set(ANAKIN_LIBRARY     ${ANAKIN_INSTALL_DIR})
 set(ANAKIN_SHARED_LIB  ${ANAKIN_LIBRARY}/libanakin.so)
 set(ANAKIN_SABER_LIB   ${ANAKIN_LIBRARY}/libanakin_saber_common.so)
 
-# TODO(luotao): ANAKIN_MODLE_URL etc will move to demo ci later.
-set(INFERENCE_URL "http://paddle-inference-dist.bj.bcebos.com")
-set(ANAKIN_MODLE_URL "${INFERENCE_URL}/mobilenet_v2.anakin.bin")
-set(ANAKIN_RNN_MODLE_URL "${INFERENCE_URL}/anakin_test%2Fditu_rnn.anakin2.model.bin")
-set(ANAKIN_RNN_DATA_URL "${INFERENCE_URL}/anakin_test%2Fditu_rnn_data.txt")
-execute_process(COMMAND bash -c "mkdir -p ${ANAKIN_SOURCE_DIR}")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_MODLE_URL} -N")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_MODLE_URL} -N")
-execute_process(COMMAND bash -c "cd ${ANAKIN_SOURCE_DIR}; wget -q --no-check-certificate ${ANAKIN_RNN_DATA_URL} -N")
-
 include_directories(${ANAKIN_INCLUDE})
 include_directories(${ANAKIN_INCLUDE}/saber/)
 include_directories(${ANAKIN_INCLUDE}/saber/core/)
@@ -48,21 +38,24 @@ set(ANAKIN_COMPILE_EXTRA_FLAGS
     -Wno-reorder
     -Wno-error=cpp)
 
+if(WITH_GPU)
+    set(CMAKE_ARGS_PREFIX -DUSE_GPU_PLACE=YES -DCUDNN_ROOT=${CUDNN_ROOT} -DCUDNN_INCLUDE_DIR=${CUDNN_INCLUDE_DIR})
+else()
+    set(CMAKE_ARGS_PREFIX -DUSE_GPU_PLACE=NO)
+endif()
 ExternalProject_Add(
     extern_anakin
     ${EXTERNAL_PROJECT_LOG_ARGS}
     DEPENDS             ${MKLML_PROJECT}
     GIT_REPOSITORY      "https://github.com/PaddlePaddle/Anakin"
-    GIT_TAG             "9424277cf9ae180a14aff09560d3cd60a49c76d2"
+    GIT_TAG             "3c8554f4978628183566ab7dd6c1e7e66493c7cd"
     PREFIX              ${ANAKIN_SOURCE_DIR}
     UPDATE_COMMAND      ""
-    CMAKE_ARGS          -DUSE_GPU_PLACE=YES
+    CMAKE_ARGS          ${CMAKE_ARGS_PREFIX}
                         -DUSE_X86_PLACE=YES
                         -DBUILD_WITH_UNIT_TEST=NO
                         -DPROTOBUF_ROOT=${THIRD_PARTY_PATH}/install/protobuf
                         -DMKLML_ROOT=${THIRD_PARTY_PATH}/install/mklml
-                        -DCUDNN_ROOT=${CUDNN_ROOT}
-                        -DCUDNN_INCLUDE_DIR=${CUDNN_INCLUDE_DIR}
                         -DENABLE_OP_TIMER=${ANAKIN_ENABLE_OP_TIMER}
                         ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS    -DCMAKE_INSTALL_PREFIX:PATH=${ANAKIN_INSTALL_DIR}

cmake/external/grpc.cmake

@@ -44,7 +44,7 @@ ExternalProject_Add(
     # 3. keep only zlib, cares, protobuf, boringssl under "third_party",
     #    checkout and clean other dirs under third_party
     # 4. remove .git, and package the directory.
-    URL "http://paddlepaddledeps.bj.bcebos.com/grpc-v1.10.x.tar.gz"
+    URL "http://paddlepaddledeps.cdn.bcebos.com/grpc-v1.10.x.tar.gz"
     URL_MD5  "1f268a2aff6759839dccd256adcc91cf"
     PREFIX          ${GRPC_SOURCES_DIR}
     UPDATE_COMMAND  ""

cmake/inference_lib.cmake

@@ -128,16 +128,13 @@ set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
 set(dst_dir "${FLUID_INSTALL_DIR}/paddle/fluid")
 set(module "framework")
 if (NOT WIN32)
-copy(framework_lib DEPS framework_py_proto 
-  SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/details/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/framework/framework.pb.h
-  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/details ${dst_dir}/${module}
-)
-else()
-copy(framework_lib
+set(framework_lib_deps framework_py_proto)
+endif(NOT WIN32)
+copy(framework_lib DEPS ${framework_lib_deps}
   SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/details/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/framework/framework.pb.h
-  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/details ${dst_dir}/${module}
+       ${src_dir}/${module}/ir/*.h
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/details ${dst_dir}/${module} ${dst_dir}/${module}/ir
 )
-endif(NOT WIN32)
 
 set(module "memory")
 copy(memory_lib
@@ -148,12 +145,12 @@ copy(memory_lib
 set(inference_deps paddle_fluid_shared paddle_fluid)
 
 set(module "inference/api")
-if (WITH_ANAKIN AND WITH_GPU)
+if (WITH_ANAKIN AND WITH_MKL)
     copy(anakin_inference_lib DEPS paddle_inference_api inference_anakin_api
         SRCS
         ${PADDLE_BINARY_DIR}/paddle/fluid/inference/api/libinference_anakin_api* # compiled anakin api
         ${ANAKIN_INSTALL_DIR} # anakin release
-        DSTS ${dst_dir}/inference/anakin ${dst_dir}/inference/anakin)
+        DSTS ${dst_dir}/inference/anakin ${FLUID_INSTALL_DIR}/third_party/install/anakin)
      list(APPEND inference_deps anakin_inference_lib)
 endif()
 
@@ -161,7 +158,8 @@ set(module "inference")
 copy(inference_lib DEPS ${inference_deps}
   SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/fluid/inference/libpaddle_fluid.*
        ${src_dir}/${module}/api/paddle_inference_api.h ${src_dir}/${module}/api/demo_ci
-  DSTS ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module}
+       ${PADDLE_BINARY_DIR}/paddle/fluid/inference/api/paddle_inference_pass.h
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module} ${dst_dir}/${module}
 )
 
 set(module "platform")

doc/fluid/api/layers.rst

@@ -822,6 +822,14 @@ pad
 ..  autofunction:: paddle.fluid.layers.pad
     :noindex:
 
+.. _api_fluid_layers_pad_constant_like:
+
+pad_constant_like
+---
+
+..  autofunction:: paddle.fluid.layers.pad_constant_like
+    :noindex:
+
 .. _api_fluid_layers_label_smooth:
 
 label_smooth
@@ -1145,6 +1153,14 @@ sigmoid
 ..  autofunction:: paddle.fluid.layers.sigmoid
     :noindex:
 
+.. _api_fluid_layers_hsigmoid:
+
+hsigmoid
+-------
+
+..  autofunction:: paddle.fluid.layers.hsigmoid
+    :noindex:
+
 .. _api_fluid_layers_logsigmoid:
 
 logsigmoid

doc/fluid/dev/releasing_process_cn.md

@@ -1,24 +1,23 @@
 # PaddlePaddle发行规范
 
-PaddlePaddle使用git-flow branching model做分支管理，使用[Semantic Versioning](http://semver.org/)标准表示PaddlePaddle版本号。
+PaddlePaddle使用Trunk Based Development，使用[Semantic Versioning](http://semver.org/)标准表示PaddlePaddle版本号。
 
 PaddlePaddle每次发新的版本，遵循以下流程:
 
 1. 从`develop`分支派生出新的分支，分支名为`release/版本号`。例如，`release/0.10.0`
-1. 将新分支的版本打上tag，tag为`版本号rc.Patch号`。第一个tag为`0.10.0rc1`，第二个为`0.10.0rc2`，依次类推。
-1. 对这个版本的提交，做如下几个操作:
-  * 使用Regression Test List作为检查列表，测试本次release的正确性。
-	  * 如果失败，记录下所有失败的例子，在这个`release/版本号`分支中，修复所有bug后，Patch号加一，到第二步
-	* 修改`python/setup.py.in`中的版本信息,并将`istaged`字段设为`True`。
-	* 将这个版本的python wheel包发布到pypi。
-	* 更新Docker镜像（参考后面的操作细节）。
-1. 第三步完成后，将`release/版本号`分支合入master分支，将master分支的合入commit打上tag，tag为`版本号`。同时再将`master`分支合入`develop`分支。
-1. 协同完成Release Note的书写。
+2. 将新分支的版本打上tag，tag为`版本号rc-Patch号`。例如，第一个tag为`0.10.0-rc0`。
+3. 新分支一般不接受新的feature和优化。QA在release分支上进行测试。研发基于最新的develop开发。
+4. QA和研发发现的bug，在develop上修复验证后，cherry-pick修复到release分支。直到release分支相对稳定。
+5. 如果有需要，在release分支最新代码上打上新的tag，比如`0.10.0-rc1`，让更多的用户加入测试。重复3-4步。
+6. release分支稳定后，打上正式的release tag，比如`0.10.0`。
+7. 将这个版本的python wheel包发布到pypi。
+8. 更新Docker镜像（参考后面的操作细节）。
 
 需要注意的是:
 
-* `release/版本号`分支一旦建立，一般不允许再从`develop`分支合入`release/版本号`。这样保证`release/版本号`分支功能的封闭，方便测试人员测试PaddlePaddle的行为。
-* 在`release/版本号`分支存在的时候，如果有bugfix的行为，需要将bugfix的分支同时merge到`master`, `develop`和`release/版本号`这三个分支。
+* bug修复需要先在develop上进行，然后进入release分支。而不是直接在release分支上开发。
+
+* release分支原则上只接受修复类的修改，不接受新feature。
 
 ## 发布wheel包到pypi
 
@@ -61,24 +60,21 @@ docker push [镜像]:[version]
 
 ## PaddlePaddle 分支规范
 
-PaddlePaddle开发过程使用[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范，并适应github的特性做了一些区别。
-
-* PaddlePaddle的主版本库遵循[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范。其中:
-	* `master`分支为稳定(stable branch)版本分支。每一个`master`分支的版本都是经过单元测试和回归测试的版本。
-	* `develop`分支为开发(develop branch)版本分支。每一个`develop`分支的版本都经过单元测试，但并没有经过回归测试。
-	* `release/版本号`分支为每一次Release时建立的临时分支。在这个阶段的代码正在经历回归测试。
+PaddlePaddle开发过程使用[Trunk Based Development](https://trunkbaseddevelopment.com/) 开发规范。
 
-* 其他用户的fork版本库并不需要严格遵守[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范，但所有fork的版本库的所有分支都相当于特性分支。
-	* 建议，开发者fork的版本库使用`develop`分支同步主版本库的`develop`分支
-	* 建议，开发者fork的版本库中，再基于`develop`版本fork出自己的功能分支。
-	* 当功能分支开发完毕后，向PaddlePaddle的主版本库提交`Pull Reuqest`，进而进行代码评审。
-		* 在评审过程中，开发者修改自己的代码，可以继续在自己的功能分支提交代码。
+* `develop`分支为开发(develop branch)版本分支。每一个`develop`分支的版本都经过单元测试。并且会经过模型回归测试。
+* `release/版本号`分支为每一次Release时建立的临时分支。release分支主要用于测试，bug修复和最终发版。
+* `master`分支因为历史原因，已经废弃。
 
-* BugFix分支也是在开发者自己的fork版本库维护，与功能分支不同的是，BugFix分支需要分别给主版本库的`master`、`develop`与可能有的`release/版本号`分支，同时提起`Pull Request`。
+* 其他开发者fork的feature branch。
+	* 建议，开发者的feature branch需要同步主版本库的`develop`分支。
+	* 建议，开发者的feature branch需要基于主版本库中的`develop`分支。
+	* 当feature branch开发完毕后，向PaddlePaddle的主版本库提交`Pull Reuqest`，进而进行代码评审。
+		* 在评审过程中，开发者修改自己的代码，可以继续在自己的feature branch提交代码。
 
 ## PaddlePaddle回归测试列表
 
-本列表说明PaddlePaddle发版之前需要测试的功能点。
+TODO
 
 ### PaddlePaddle Book中所有章节
 

doc/fluid/dev/releasing_process_en.md

@@ -4,26 +4,21 @@ PaddlePaddle manages its branches using "git-flow branching model", and [Semanti
 
 Each time we release a new PaddlePaddle version, we should follow the below steps:
 
-1. Fork a new branch from `develop` named `release/[version]`, e.g. `release/0.10.0`.
-1. Push a new tag on the release branch, the tag name should be like `[version]rc.patch`. The
-   first tag should be `0.10.0rc1`, and the second should be `0.10.0.rc2` and so on.
-1. After that, we should do:
-  * Run all regression test on the Regression Test List (see PaddlePaddle TeamCity CI), to confirm
-      that this release has no major bugs.
-        * If regression test fails, we must fix those bugs and create a new `release/[version]`
-          branch from previous release branch.
-    * Modify `python/setup.py.in`, change the version number and change `ISTAGED` to `True`.
-    * Publish PaddlePaddle release wheel packages to pypi (see below instructions for detail).
-    * Update the Docker images (see below instructions for detail).
-1. After above step, merge `release/[version]` branch to master and push a tag on the master commit,
-   then merge `master` to `develop`.
-1. Update the Release Note.          
-
-***NOTE:***
-
-* Do ***NOT*** merge commits from develop branch to release branches to keep the release branch contain
-  features only for current release, so that we can test on that version.
-* If we want to fix bugs on release branches, we must merge the fix to master, develop and release branch.
+1. Create a new release branch from `develop`，named `release/[version]`. E.g.，`release/0.10.0`
+2. Create a new tag for the release branch, tag format: `version-rc.Patch`. E.g. the first tag is `0.10.0-rc0`。
+3. New release branch normally doesn't accept new features or optimizations. QA will test on the release branch. Developer should develop based on `develop` branch.
+4. If QA or Developer find bugs. They should first fix and verify on `develop` branch. Then cherry-pick the fix to the release branch. Wait until the release branch is stable.
+5. If necessary, create a new tag on the relese branch, e.g. `0.10.0-rc1`. Involve more users to try it and repeat step 3-4.
+6. After release branch is stable，Create the official release tag，such as `0.10.0`.
+7. Release the python wheel package to pypi.
+8. Update the docker image (More details below).
+
+NOTE:
+
+* bug fix should happen on `develop` branch, then cherry-pick to relese branch. Avoid developing directly on release branch.
+
+* release normally only accept bug fixes. Don't add new features.
+
 
 ## Publish Wheel Packages to pypi
 
@@ -97,26 +92,22 @@ You can then checkout the latest pushed tags at https://hub.docker.com/r/paddlep
 
 ## Branching Model
 
-We use [git-flow](http://nvie.com/posts/a-successful-git-branching-model/) as our branching model,
-with some modifications:
-
-* `master` branch is the stable branch. Each version on the master branch is tested and guaranteed.
-* `develop` branch is for development. Each commit on develop branch has passed CI unit test, but no
-  regression tests are run.
-* `release/[version]` branch is used to publish each release. Latest release version branches have
-  bugfix only for that version, but no feature updates.
-* Developer forks are not required to follow
-  [git-flow](http://nvie.com/posts/a-successful-git-branching-model/)
-  branching model, all forks is like a feature branch.
-    * Advise: developer fork's develop branch is used to sync up with main repo's develop branch.
-    * Advise: developer use it's fork's develop branch to for new branch to start developing.
-  * Use that branch on developer's fork to create pull requests and start reviews.
-      * developer can push new commits to that branch when the pull request is open.
-* Bug fixes are also started from developers forked repo. And, bug fixes branch can merge to
-  `master`, `develop` and `releases`.
+PaddlePaddle uses [Trunk Based Development](https://trunkbaseddevelopment.com/) as our branching model.
+
+* `develop` branch is used for development. Each comment to `develop` branc goes through unit tests and model regression tests.
+* `release/[version]` branch is used for each release. Release branch is used for tests, bug fix and evetual release.
+* `master` branch as been deprecated for historical reasons
+
+* Developer's feature branch。
+	* Developer's feature branch should sync with upstream `develop` branch.
+	* Developer's feature branch should be forked from upstream `develop` branch.
+	* After feature branch is ready, create a `Pull Request` against the Paddle repo and go through code review.
+	   * In the review process, develop modify codes and push to their own feature branch.
 
 ## PaddlePaddle Regression Test List
 
+TODO
+
 ### All Chapters of PaddlePaddle Book
 
 We need to guarantee that all the chapters of PaddlePaddle Book can run correctly. Including

doc/fluid/new_docs/advanced_usage/deploy/index_anakin.rst

@@ -1,5 +1,5 @@
-服务器端部署 - Anakin
-#####################
+Anakin - 服务器端加速引擎
+#######################
 
 
 使用文档

doc/fluid/new_docs/advanced_usage/deploy/index_native.rst

@@ -1,8 +0,0 @@
-服务器端部署 - 原生引擎
-#######################
-
-..  toctree::
-    :maxdepth: 2
-
-    build_and_install_lib_cn.rst
-    native_infer.rst