Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into no_counter_on_pserver

CMakeLists.txt

@@ -137,7 +137,7 @@ include(external/openblas)  # download, build, install openblas
 include(external/mkldnn)    # download, build, install mkldnn
 include(external/swig)      # download, build, install swig
 include(external/warpctc)   # download, build, install warpctc
-include(external/boost)     # download, build, install boost
+include(external/boost)     # download boost
 include(external/any)       # download libn::any
 include(external/eigen)     # download eigen3
 include(external/pybind11)  # download pybind11

benchmark/cluster/vgg16/Dockerfile

+#FROM python:2.7.14
+FROM nvidia/cuda:8.0-cudnn5-runtime-ubuntu16.04
+RUN apt-get update && apt-get install -y python
+RUN pip install -U kubernetes opencv-python &&   apt-get update -y &&   apt-get install -y iputils-ping libgtk2.0-dev
+# NOTE: By default CI built wheel packages turn WITH_DISTRIBUTE=OFF,
+#       so we must build one with distribute support to install in this image.
+RUN pip install paddlepaddle
+RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.cifar.train10()" | python'
+RUN pip uninstall -y paddlepaddle
+
+# below lines may change a lot for debugging
+ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/paddle_k8s /usr/bin
+ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/k8s_tools.py /root
+ADD *.whl /
+RUN pip install /*.whl && rm -f /*.whl && \
+chmod +x /usr/bin/paddle_k8s
+ENV LD_LIBRARY_PATH=/usr/local/lib
+ADD vgg16_fluid.py vgg16_v2.py /workspace/

benchmark/cluster/vgg16/README.md

+# Performance for Distributed vgg16
+
+## Test Result
+
+### Hardware Infomation
+
+- CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
+- cpu MHz		: 2101.000
+- cache size	: 20480 KB
+
+### Single Node Single Thread
+
+- PServer Count: 10
+- Trainer Count: 20
+- Metrics: samples / sec
+
+| Batch Size | 32 | 64 | 128 | 256 |
+| -- | -- | -- | -- | -- |
+| PaddlePaddle Fluid | 15.44 | 16.32 | 16.74 | 16.79 |
+| PaddlePaddle v2 | 15.97 | 17.04 | 17.60 | 17.83 |
+| TensorFlow | - | - | - | - |
+
+### Different Batch Size
+
+- PServer Count: 10
+- Trainer Count: 20
+- Per trainer CPU Core: 1
+- Metrics: samples / sec
+
+| Batch Size | 32 | 64 | 128 | 256 |
+| -- | -- | -- | -- | -- |
+| PaddlePaddle Fluid | 190.20 | 222.15 | 247.40 | 258.18 |
+| PaddlePaddle v2 | 170.96 | 233.71 | 256.14 | 329.23 |
+| TensorFlow | - | - | - | - |
+
+
+### Accelerate Rate
+
+- Pserver Count: 20
+- Batch Size: 128
+- Metrics: samples / sec
+
+| Trainer Count | 20 | 40 | 80 | 100 |
+| -- | -- | -- | -- | -- |
+| PaddlePaddle Fluid | 263.29 (78.64%) | 518.80 (77.47%) | 836.26 (62.44%) | 1019.29 (60.89%) |
+| PaddlePaddle v2 (need more tests) | 326.85 (92.85%) | 534.58 (75.93%) | 853.30 (60.60%) | 1041.99 (59.20%) |
+| TensorFlow | - | - | - | - |
+
+### Different Pserver Count
+
+- Trainer Count: 60
+- Batch Size: 128
+- Metrics: samples/ sec
+
+| PServer Count | 3 | 6 |10 | 20 |
+| -- | -- | -- | -- | -- |
+| PaddlePaddle Fluid(should fix in next PR) | 589.1 | 592.6 | 656.4 | 655.8 |
+| PaddlePaddle v2 | 593.4 | 791.3 | 729.7 | 821.7 |
+| TensorFlow | - | - | - | - |
+
+*The performance gap between Fuild and v2 comes from the network interference.*
+
+
+## Steps to Run the Performance Test
+
+1. You must re-compile PaddlePaddle and enable `-DWITH_DISTRIBUTE` to build PaddlePaddle with distributed support.
+1. When the build finishes, copy the output `whl` package located under `build/python/dist` to current directory.
+1. Run `docker build -t [image:tag] .` to build the docker image and run `docker push [image:tag]` to push the image to reponsitory so kubernetes can find it.
+1. Run `kubectl create -f pserver.yaml && kubectl create -f trainer.yaml` to start the job on your kubernetes cluster (you must configure the `kubectl` client before this step).
+1. Run `kubectl get po` to get running pods, and run `kubectl logs [podID]` to fetch the pod log of pservers and trainers.
+
+Check the logs for the distributed training progress and analyze the performance.
+
+## Enable Verbos Logs
+
+Edit `pserver.yaml` and `trainer.yaml` and add an environment variable `GLOG_v=3` and `GLOG_logtostderr=1` to see what happend in detail.

benchmark/cluster/vgg16/fluid_pserver.yaml

+apiVersion: extensions/v1beta1
+kind: ReplicaSet
+metadata:
+  name: vgg16job-pserver
+spec:
+  replicas: 10
+  template:
+    metadata:
+      labels:
+        paddle-job-pserver: vgg16job
+    spec:
+      hostNetwork: true
+      imagePullSecrets:
+      - name: job-registry-secret
+      containers:
+      - name: pserver
+        image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
+        imagePullPolicy: Always
+        ports:
+        - name: jobport-30236
+          containerPort: 30236
+        env:
+        - name: PADDLE_JOB_NAME
+          value: vgg16job
+        - name: MKL_NUM_THREADS
+          value: "1"
+        - name: TRAINING_ROLE
+          value: "PSERVER"
+        - name: TRAINERS
+          value: "20"
+        - name: PSERVERS
+          value: "10"
+        - name: TOPOLOGY
+          value: ""
+        - name: ENTRY
+          value: "MKL_NUM_THREADS=1 python /workspace/vgg16_fluid.py --local 0"
+        - name: TRAINER_PACKAGE
+          value: "/workspace"
+        - name: PADDLE_INIT_PORT
+          value: "30236"
+        - name: PADDLE_INIT_NICS
+          value: "xgbe0"
+        - name: PADDLE_INIT_TRAINER_COUNT
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
+          value: "1"
+        - name: PADDLE_INIT_NUM_GRADIENT_SERVERS
+          value: "20"
+        - name: PADDLE_INIT_NUM_PASSES
+          value: "1"
+        - name: PADDLE_INIT_USE_GPU
+          value: "0"
+        - name: LD_LIBRARY_PATH
+          value: "/usr/local/lib:/usr/local/nvidia/lib64"
+        - name: NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: "metadata.namespace"
+        - name: POD_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: "status.podIP"
+        command: ["paddle_k8s", "start_fluid"]
+        resources:
+          requests:
+            memory: 10Gi
+            cpu: 4
+          limits:
+            memory: 10Gi
+            cpu: 4

benchmark/cluster/vgg16/fluid_trainer.yaml

+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: vgg16job-trainer
+spec:
+  parallelism: 20
+  completions: 20
+  template:
+    metadata:
+      labels:
+        paddle-job: vgg16job
+    spec:
+      imagePullSecrets:
+        - name: job-registry-secret
+      hostNetwork: true
+      containers:
+      - name: trainer
+        image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
+        imagePullPolicy: Always
+        command: ["paddle_k8s", "start_fluid"]
+        env:
+        - name: PADDLE_JOB_NAME
+          value: vgg16job
+        - name: TRAINING_ROLE
+          value: "TRAINER"
+        - name: TRAINERS
+          value: "20"
+        - name: PSERVERS
+          value: "10"
+        - name: TOPOLOGY
+          value: ""
+        - name: ENTRY
+          value: "MKL_NUM_THREADS=1 python /workspace/vgg16_fluid.py --local 0 --batch_size 128"
+        - name: TRAINER_PACKAGE
+          value: "/workspace"
+        - name: PADDLE_INIT_PORT
+          value: "30236"
+        - name: PADDLE_INIT_NICS
+          value: "xgbe0"
+        - name: PADDLE_INIT_TRAINER_COUNT
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
+          value: "1"
+        - name: PADDLE_INIT_NUM_GRADIENT_SERVERS
+          value: "20"
+        - name: PADDLE_INIT_NUM_PASSES
+          value: "1"
+        - name: PADDLE_INIT_USE_GPU
+          value: "0"
+        - name: LD_LIBRARY_PATH
+          value: "/usr/local/lib:/usr/local/nvidia/lib64"
+        - name: NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: "metadata.namespace"
+        - name: POD_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: "status.podIP"
+        resources:
+          requests:
+            memory: 40Gi
+            cpu: 2
+          limits:
+            memory: 40Gi
+            cpu: 2
+      restartPolicy: Never

benchmark/cluster/vgg16/v2_pserver.yaml

+apiVersion: extensions/v1beta1
+kind: ReplicaSet
+metadata:
+  name: vgg16v2job-pserver
+spec:
+  replicas: 10
+  template:
+    metadata:
+      labels:
+        paddle-job-pserver: vgg16v2job
+    spec:
+      hostNetwork: true
+      imagePullSecrets:
+      - name: job-registry-secret
+      containers:
+      - name: pserver
+        image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
+        imagePullPolicy: Always
+        ports:
+        - name: jobport-30236
+          containerPort: 30236
+        env:
+        - name: PADDLE_JOB_NAME
+          value: vgg16v2job
+        - name: TRAINERS
+          value: "20"
+        - name: PSERVERS
+          value: "10"
+        - name: TOPOLOGY
+          value: ""
+        - name: ENTRY
+          value: "python train.py"
+        - name: TRAINER_PACKAGE
+          value: "/workspace"
+        - name: PADDLE_INIT_PORT
+          value: "30236"
+        - name: PADDLE_INIT_NICS
+          value: "xgbe0"
+        - name: PADDLE_INIT_TRAINER_COUNT
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
+          value: "1"
+        - name: PADDLE_INIT_NUM_GRADIENT_SERVERS
+          value: "20"
+        - name: PADDLE_INIT_NUM_PASSES
+          value: "1"
+        - name: PADDLE_INIT_USE_GPU
+          value: "0"
+        - name: LD_LIBRARY_PATH
+          value: "/usr/local/lib:/usr/local/nvidia/lib64"
+        - name: NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: "metadata.namespace"
+        command: ["paddle_k8s", "start_pserver"]
+        resources:
+          requests:
+            memory: 10Gi
+            cpu: 4
+          limits:
+            memory: 10Gi
+            cpu: 4

benchmark/cluster/vgg16/v2_trainer.yaml

+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: vgg16v2job-trainer
+spec:
+  parallelism: 20
+  completions: 20
+  template:
+    metadata:
+      labels:
+        paddle-job: vgg16v2job
+    spec:
+      imagePullSecrets:
+        - name: job-registry-secret
+      hostNetwork: true
+      containers:
+      - name: trainer
+        image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
+        imagePullPolicy: Always
+        command: ["paddle_k8s", "start_trainer", "v2"]
+        env:
+        - name: PADDLE_JOB_NAME
+          value: vgg16v2job
+        - name: BATCH_SIZE
+          value: "256"
+        - name: TRAINERS
+          value: "20"
+        - name: PSERVERS
+          value: "10"
+        - name: TOPOLOGY
+          value: ""
+        - name: ENTRY
+          value: "cd /workspace && MKL_NUM_THREADS=1 python /workspace/vgg16_v2.py"
+        - name: TRAINER_PACKAGE
+          value: "/workspace"
+        - name: PADDLE_INIT_PORT
+          value: "30236"
+        - name: PADDLE_INIT_NICS
+          value: "xgbe0"
+        - name: PADDLE_INIT_TRAINER_COUNT
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM
+          value: "1"
+        - name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
+          value: "1"
+        - name: PADDLE_INIT_NUM_GRADIENT_SERVERS
+          value: "20"
+        - name: PADDLE_INIT_NUM_PASSES
+          value: "2"
+        - name: PADDLE_INIT_USE_GPU
+          value: "0"
+        - name: LD_LIBRARY_PATH
+          value: "/usr/local/lib:/usr/local/nvidia/lib64"
+        - name: NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: "metadata.namespace"
+        resources:
+          requests:
+            memory: 40Gi
+            cpu: 2
+          limits:
+            memory: 40Gi
+            cpu: 2
+      restartPolicy: Never

benchmark/cluster/vgg16/vgg16_fluid.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.
+"""VGG16 benchmark in Fluid"""
+from __future__ import print_function
+
+import sys
+import time
+import numpy as np
+import paddle.v2 as paddle
+import paddle.v2.fluid as fluid
+import paddle.v2.fluid.core as core
+import paddle.v2.fluid.profiler as profiler
+import argparse
+import functools
+import os
+
+
+def str2bool(v):
+    if v.lower() in ('yes', 'true', 't', 'y', '1'):
+        return True
+    elif v.lower() in ('no', 'false', 'f', 'n', '0'):
+        return False
+    else:
+        raise argparse.ArgumentTypeError('Boolean value expected.')
+
+
+parser = argparse.ArgumentParser(description=__doc__)
+parser.add_argument(
+    '--batch_size', type=int, default=128, help="Batch size for training.")
+parser.add_argument(
+    '--learning_rate',
+    type=float,
+    default=1e-3,
+    help="Learning rate for training.")
+parser.add_argument('--num_passes', type=int, default=50, help="No. of passes.")
+parser.add_argument(
+    '--device',
+    type=str,
+    default='CPU',
+    choices=['CPU', 'GPU'],
+    help="The device type.")
+parser.add_argument('--device_id', type=int, default=0, help="The device id.")
+parser.add_argument(
+    '--data_format',
+    type=str,
+    default='NCHW',
+    choices=['NCHW', 'NHWC'],
+    help='The data order, now only support NCHW.')
+parser.add_argument(
+    '--data_set',
+    type=str,
+    default='cifar10',
+    choices=['cifar10', 'flowers'],
+    help='Optional dataset for benchmark.')
+parser.add_argument(
+    '--local',
+    type=str2bool,
+    default=True,
+    help='Whether to run as local mode.')
+args = parser.parse_args()
+
+
+def vgg16_bn_drop(input):
+    def conv_block(input, num_filter, groups, dropouts):
+        return fluid.nets.img_conv_group(
+            input=input,
+            pool_size=2,
+            pool_stride=2,
+            conv_num_filter=[num_filter] * groups,
+            conv_filter_size=3,
+            conv_act='relu',
+            conv_with_batchnorm=True,
+            conv_batchnorm_drop_rate=dropouts,
+            pool_type='max')
+
+    conv1 = conv_block(input, 64, 2, [0.3, 0])
+    conv2 = conv_block(conv1, 128, 2, [0.4, 0])
+    conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0])
+    conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0])
+    conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0])
+
+    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')
+    drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5)
+    fc2 = fluid.layers.fc(input=drop2, size=512, act=None)
+    return fc2
+
+
+def main():
+    if args.data_set == "cifar10":
+        classdim = 10
+        if args.data_format == 'NCHW':
+            data_shape = [3, 32, 32]
+        else:
+            data_shape = [32, 32, 3]
+    else:
+        classdim = 102
+        if args.data_format == 'NCHW':
+            data_shape = [3, 224, 224]
+        else:
+            data_shape = [224, 224, 3]
+
+    # Input data
+    images = fluid.layers.data(name='pixel', 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
+    accuracy = fluid.evaluator.Accuracy(input=predict, label=label)
+
+    # inference program
+    inference_program = fluid.default_main_program().clone()
+    with fluid.program_guard(inference_program):
+        test_target = accuracy.metrics + accuracy.states
+        inference_program = fluid.io.get_inference_program(test_target)
+
+    # Optimization
+    optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
+    optimize_ops, params_grads = optimizer.minimize(avg_cost)
+
+    # Initialize executor
+    place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(
+        args.device_id)
+    exe = fluid.Executor(place)
+
+    # test
+    def test(exe):
+        accuracy.reset(exe)
+        for batch_id, data in enumerate(test_reader()):
+            img_data = np.array(map(lambda x: x[0].reshape(data_shape),
+                                    data)).astype("float32")
+            y_data = np.array(map(lambda x: x[1], data)).astype("int64")
+            y_data = y_data.reshape([-1, 1])
+
+            exe.run(inference_program,
+                    feed={"pixel": img_data,
+                          "label": y_data})
+
+        return accuracy.eval(exe)
+
+    def train_loop(exe, trainer_prog):
+        iters = 0
+        ts = time.time()
+        for pass_id in range(args.num_passes):
+            # train
+            start_time = time.time()
+            num_samples = 0
+            accuracy.reset(exe)
+            with profiler.profiler("CPU", 'total') as prof:
+                for batch_id, data in enumerate(train_reader()):
+                    ts = time.time()
+                    img_data = np.array(
+                        map(lambda x: x[0].reshape(data_shape), data)).astype(
+                            "float32")
+                    y_data = np.array(map(lambda x: x[1], data)).astype("int64")
+                    y_data = y_data.reshape([-1, 1])
+
+                    loss, acc = exe.run(
+                        trainer_prog,
+                        feed={"pixel": img_data,
+                              "label": y_data},
+                        fetch_list=[avg_cost] + accuracy.metrics)
+                    iters += 1
+                    num_samples += len(data)
+                    print(
+                        "Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, spent %f"
+                        % (pass_id, iters, loss, acc, time.time() - ts)
+                    )  # The accuracy is the accumulation of batches, but not the current batch.
+
+            pass_elapsed = time.time() - start_time
+            pass_train_acc = accuracy.eval(exe)
+            pass_test_acc = test(exe)
+            print(
+                "Pass = %d, Training performance = %f imgs/s, Train accuracy = %f, Test accuracy = %f\n"
+                % (pass_id, num_samples / pass_elapsed, pass_train_acc,
+                   pass_test_acc))
+
+    if args.local:
+        # Parameter initialization
+        exe.run(fluid.default_startup_program())
+
+        # data reader
+        train_reader = paddle.batch(
+            paddle.reader.shuffle(
+                paddle.dataset.cifar.train10() if args.data_set == 'cifar10'
+                else paddle.dataset.flowers.train(),
+                buf_size=5120),
+            batch_size=args.batch_size)
+        test_reader = paddle.batch(
+            paddle.dataset.cifar.test10()
+            if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
+            batch_size=args.batch_size)
+        train_loop(exe, fluid.default_main_program())
+    else:
+        pserver_ips = os.getenv("PADDLE_INIT_PSERVERS")  # all pserver endpoints
+        eplist = []
+        for ip in pserver_ips.split(","):
+            eplist.append(':'.join([ip, "6174"]))
+        pserver_endpoints = ",".join(eplist)
+        print("pserver endpoints: ", pserver_endpoints)
+        trainers = int(os.getenv("TRAINERS"))  # total trainer count
+        print("trainers total: ", trainers)
+        current_endpoint = os.getenv(
+            "POD_IP") + ":6174"  # current pserver endpoint
+        training_role = os.getenv(
+            "TRAINING_ROLE",
+            "TRAINER")  # get the training role: trainer/pserver
+        t = fluid.DistributeTranspiler()
+        t.transpile(
+            optimize_ops,
+            params_grads,
+            pservers=pserver_endpoints,
+            trainers=trainers)
+
+        if training_role == "PSERVER":
+            if not current_endpoint:
+                print("need env SERVER_ENDPOINT")
+                exit(1)
+            pserver_prog = t.get_pserver_program(current_endpoint)
+            pserver_startup = t.get_startup_program(current_endpoint,
+                                                    pserver_prog)
+            print("starting server side startup")
+            exe.run(pserver_startup)
+            print("starting parameter server...")
+            exe.run(pserver_prog)
+        elif training_role == "TRAINER":
+            # Parameter initialization
+            exe.run(fluid.default_startup_program())
+
+            # data reader
+            train_reader = paddle.batch(
+                paddle.reader.shuffle(
+                    paddle.dataset.cifar.train10() if args.data_set == 'cifar10'
+                    else paddle.dataset.flowers.train(),
+                    buf_size=5120),
+                batch_size=args.batch_size)
+            test_reader = paddle.batch(
+                paddle.dataset.cifar.test10() if args.data_set == 'cifar10' else
+                paddle.dataset.flowers.test(),
+                batch_size=args.batch_size)
+
+            trainer_prog = t.get_trainer_program()
+            feeder = fluid.DataFeeder(feed_list=[images, label], place=place)
+            # TODO(typhoonzero): change trainer startup program to fetch parameters from pserver
+            exe.run(fluid.default_startup_program())
+            train_loop(exe, trainer_prog)
+        else:
+            print("environment var TRAINER_ROLE should be TRAINER os PSERVER")
+
+
+def print_arguments():
+    print('-----------  Configuration Arguments -----------')
+    for arg, value in sorted(vars(args).iteritems()):
+        print('%s: %s' % (arg, value))
+    print('------------------------------------------------')
+
+
+if __name__ == "__main__":
+    print_arguments()
+    main()

benchmark/cluster/vgg16/vgg16_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 paddle.v2.dataset.cifar as cifar
+import paddle.v2 as paddle
+import time
+import os
+
+DATA_DIM = 3 * 32 * 32
+CLASS_DIM = 10
+BATCH_SIZE = os.getenv("BATCH_SIZE")
+if BATCH_SIZE:
+    BATCH_SIZE = int(BATCH_SIZE)
+else:
+    BATCH_SIZE = 128
+print "batch_size", BATCH_SIZE
+NODE_COUNT = int(os.getenv("TRAINERS"))
+ts = 0
+
+
+def vgg(input, nums, class_dim):
+    def conv_block(input, num_filter, groups, num_channels=None):
+        return paddle.networks.img_conv_group(
+            input=input,
+            num_channels=num_channels,
+            pool_size=2,
+            pool_stride=2,
+            conv_num_filter=[num_filter] * groups,
+            conv_filter_size=3,
+            conv_act=paddle.activation.Relu(),
+            pool_type=paddle.pooling.Max())
+
+    assert len(nums) == 5
+    # the channel of input feature is 3
+    conv1 = conv_block(input, 64, nums[0], 3)
+    conv2 = conv_block(conv1, 128, nums[1])
+    conv3 = conv_block(conv2, 256, nums[2])
+    conv4 = conv_block(conv3, 512, nums[3])
+    conv5 = conv_block(conv4, 512, nums[4])
+
+    fc_dim = 512
+    fc1 = paddle.layer.fc(input=conv5,
+                          size=fc_dim,
+                          act=paddle.activation.Relu(),
+                          layer_attr=paddle.attr.Extra(drop_rate=0.5))
+    fc2 = paddle.layer.fc(input=fc1,
+                          size=fc_dim,
+                          act=paddle.activation.Relu(),
+                          layer_attr=paddle.attr.Extra(drop_rate=0.5))
+    out = paddle.layer.fc(input=fc2,
+                          size=class_dim,
+                          act=paddle.activation.Softmax())
+    return out
+
+
+def vgg13(input, class_dim):
+    nums = [2, 2, 2, 2, 2]
+    return vgg(input, nums, class_dim)
+
+
+def vgg16(input, class_dim):
+    nums = [2, 2, 3, 3, 3]
+    return vgg(input, nums, class_dim)
+
+
+def vgg19(input, class_dim):
+    nums = [2, 2, 4, 4, 4]
+    return vgg(input, nums, class_dim)
+
+
+def main():
+    global ts
+    paddle.init(use_gpu=False)
+    image = paddle.layer.data(
+        name="image", type=paddle.data_type.dense_vector(DATA_DIM))
+    lbl = paddle.layer.data(
+        name="label", type=paddle.data_type.integer_value(CLASS_DIM))
+
+    extra_layers = None
+    # NOTE: for v2 distributed training need averaging updates.
+    learning_rate = 1e-3 / NODE_COUNT
+    out = vgg16(image, class_dim=CLASS_DIM)
+    cost = paddle.layer.classification_cost(input=out, label=lbl)
+
+    # Create parameters
+    parameters = paddle.parameters.create(cost)
+
+    # Create optimizer
+    optimizer = paddle.optimizer.Momentum(
+        momentum=0.9,
+        regularization=paddle.optimizer.L2Regularization(rate=0.0005 *
+                                                         BATCH_SIZE),
+        learning_rate=learning_rate / BATCH_SIZE,
+        learning_rate_decay_a=0.1,
+        learning_rate_decay_b=128000 * 35,
+        learning_rate_schedule="discexp", )
+
+    train_reader = paddle.batch(
+        paddle.reader.shuffle(
+            cifar.train10(),
+            # To use other data, replace the above line with:
+            # reader.train_reader('train.list'),
+            buf_size=1000),
+        batch_size=BATCH_SIZE)
+    test_reader = paddle.batch(
+        cifar.test10(),
+        # To use other data, replace the above line with:
+        # reader.test_reader('val.list'),
+        batch_size=BATCH_SIZE)
+
+    # Create trainer
+    trainer = paddle.trainer.SGD(cost=cost,
+                                 parameters=parameters,
+                                 update_equation=optimizer,
+                                 extra_layers=extra_layers,
+                                 is_local=False)
+
+    # End batch and end pass event handler
+    def event_handler(event):
+        global ts, ts_pass
+        if isinstance(event, paddle.event.BeginPass):
+            ts_pass = time.time()
+        if isinstance(event, paddle.event.BeginIteration):
+            ts = time.time()
+        if isinstance(event, paddle.event.EndIteration):
+            if event.batch_id % 1 == 0:
+                print "\nPass %d, Batch %d, Cost %f, %s, spent: %f" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics,
+                    time.time() - ts)
+        if isinstance(event, paddle.event.EndPass):
+            print "Pass %d end, spent: %f" % (event.pass_id,
+                                              time.time() - ts_pass)
+            result = trainer.test(reader=test_reader)
+            print "\nTest with Pass %d, %s" % (event.pass_id, result.metrics)
+
+    trainer.train(
+        reader=train_reader, num_passes=200, event_handler=event_handler)
+
+
+if __name__ == '__main__':
+    main()

cmake/external/boost.cmake

@@ -21,6 +21,7 @@ set(BOOST_URL           "http://sourceforge.net/projects/boost/files/boost/${BOO
 set(BOOST_SOURCES_DIR ${THIRD_PARTY_PATH}/boost)
 set(BOOST_DOWNLOAD_DIR  "${BOOST_SOURCES_DIR}/src/${BOOST_PROJECT}")
 set(BOOST_INCLUDE_DIR "${BOOST_DOWNLOAD_DIR}/${BOOST_TAR}" CACHE PATH "boost include directory." FORCE)
+set_directory_properties(PROPERTIES CLEAN_NO_CUSTOM 1)
 
 include_directories(${BOOST_INCLUDE_DIR})
 

cmake/generic.cmake

@@ -186,6 +186,11 @@ function(cc_library TARGET_NAME)
       add_library(${TARGET_NAME} STATIC ${cc_library_SRCS})
     endif()
     if (cc_library_DEPS)
+      # Don't need link libwarpctc.so
+      if ("${cc_library_DEPS};" MATCHES "warpctc;")
+        list(REMOVE_ITEM cc_library_DEPS warpctc)
+        add_dependencies(${TARGET_NAME} warpctc)
+      endif()
       add_dependencies(${TARGET_NAME} ${cc_library_DEPS})
       target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
     endif()
@@ -465,10 +470,10 @@ function(py_test TARGET_NAME)
   if(WITH_TESTING)
     set(options "")
     set(oneValueArgs "")
-    set(multiValueArgs SRCS DEPS ARGS)
+    set(multiValueArgs SRCS DEPS ARGS ENVS)
     cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
     add_test(NAME ${TARGET_NAME}
-             COMMAND env PYTHONPATH=${PADDLE_PYTHON_BUILD_DIR}/lib-python
+             COMMAND env PYTHONPATH=${PADDLE_PYTHON_BUILD_DIR}/lib-python ${py_test_ENVS}
              ${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}
              WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
   endif()

doc/CMakeLists.txt

@@ -47,3 +47,5 @@ sphinx_add_target(paddle_docs_cn
                   ${SPHINX_CACHE_DIR_CN}
                   ${CMAKE_CURRENT_SOURCE_DIR}
                   ${SPHINX_HTML_DIR_CN})
+
+add_subdirectory(api)

doc/api/CMakeLists.txt

+# configured documentation tools and intermediate build results
+set(BINARY_BUILD_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_build")
+
+# Sphinx cache with pickled ReST documents
+set(SPHINX_CACHE_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_doctrees")
+
+# HTML output director
+set(SPHINX_HTML_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/html")
+
+configure_file(
+    "${CMAKE_CURRENT_SOURCE_DIR}/../templates/conf.py.en.in"
+    "${BINARY_BUILD_DIR_EN}/conf.py"
+    @ONLY)
+
+sphinx_add_target(paddle_api_docs
+                  html
+                  ${BINARY_BUILD_DIR_EN}
+                  ${SPHINX_CACHE_DIR_EN}
+                  ${CMAKE_CURRENT_SOURCE_DIR}
+                  ${SPHINX_HTML_DIR_EN})

doc/api/v2/config/layer.rst

@@ -87,6 +87,11 @@ roi_pool
 ..  autoclass:: paddle.v2.layer.roi_pool
     :noindex:
 
+pad
+----
+..  autoclass:: paddle.v2.layer.pad
+    :noindex:
+
 Norm Layer
 ==========
 
@@ -133,6 +138,11 @@ grumemory
 ..  autoclass:: paddle.v2.layer.grumemory
     :noindex:
 
+gated_unit
+-----------
+..  autoclass:: paddle.v2.layer.gated_unit
+    :noindex:
+    
 Recurrent Layer Group
 =====================
 
@@ -340,6 +350,11 @@ bilinear_interp
 ..  autoclass:: paddle.v2.layer.bilinear_interp
     :noindex:
 
+dropout
+--------
+..  autoclass:: paddle.v2.layer.dropout
+    :noindex:
+    
 dot_prod
 ---------
 .. autoclass:: paddle.v2.layer.dot_prod
@@ -402,6 +417,11 @@ scale_shift
 ..  autoclass:: paddle.v2.layer.scale_shift
     :noindex:
 
+factorization_machine
+---------------------
+..  autoclass:: paddle.v2.layer.factorization_machine
+    :noindex:
+
 Sampling Layers
 ===============
 
@@ -420,22 +440,6 @@ multiplex
 ..  autoclass:: paddle.v2.layer.multiplex
     :noindex:
 
-Factorization Machine Layer
-============================
-
-factorization_machine
----------------------
-..  autoclass:: paddle.v2.layer.factorization_machine
-    :noindex:
-
-Slicing and Joining Layers
-==========================
-
-pad
-----
-..  autoclass:: paddle.v2.layer.pad
-    :noindex:
-
 ..  _api_v2.layer_costs:
 
 Cost Layers
@@ -526,6 +530,11 @@ multibox_loss
 ..  autoclass:: paddle.v2.layer.multibox_loss
     :noindex:
 
+detection_output
+----------------
+..  autoclass:: paddle.v2.layer.detection_output
+    :noindex:
+    
 Check Layer
 ============
 
@@ -534,31 +543,10 @@ eos
 ..  autoclass:: paddle.v2.layer.eos
     :noindex:
 
-Miscs
-=====
-
-dropout
---------
-..  autoclass:: paddle.v2.layer.dropout
-    :noindex:
-
-Activation with learnable parameter
-===================================
+Activation
+==========
 
 prelu
 --------
 ..  autoclass:: paddle.v2.layer.prelu
     :noindex:
-
-gated_unit
------------
-..  autoclass:: paddle.v2.layer.gated_unit
-    :noindex:
-
-Detection output Layer
-======================
-
-detection_output
-----------------
-..  autoclass:: paddle.v2.layer.detection_output
-    :noindex:

doc/api/v2/data/dataset.rst

@@ -73,3 +73,10 @@ wmt14
 ..  automodule:: paddle.v2.dataset.wmt14
     :members:
     :noindex:
+
+wmt16
++++++
+
+..  automodule:: paddle.v2.dataset.wmt16
+    :members:
+    :noindex:

doc/design/csp.md

@@ -42,7 +42,7 @@ The type *channel* is conceptually the blocking queue.  In Go, its implemented i
 
 The `select` operation has been in OS kernels long before Go language.  All Unix kernels implement system calls *poll* and *select*.  They monitor multiple file descriptors to see if I/O is possible on any of them.  This takes O(N) time.  Since Linux 2.6, a new system call, *epoll*, can do the same in O(1) time.  In BSD systems, there is a similar system call *kqueue*.  Go's Linux implementation uses epoll.
 
-It might be a good idea to implement Fluid's select using epoll too.  In this design doc, we start from the O(N) way, so we could focus on Python binding and the syntax.
+It might be a good idea to implement Fluid's select using epoll too.  In this design doc, we start from the O(N) way so that we could focus on Python binding and the syntax.
 
 ### Type Channel
 
@@ -71,14 +71,14 @@ ch1 := make(chan int, 100)  // a channel that can buffer 100 ints.
 In Fluid, we should be able to do the same:
 
 ```python
-ch  = fluid.make_chan(dtype=INT)
-ch1 = fluid.make_chan(dtype=INT, 100)
+ch  = fluid.make_channel(dtype=INT)
+ch1 = fluid.make_channel(dtype=INT, 100)
 ```
 
 In addition to that, we want channels that can hold more complex element types, e.g., Tensors of float16:
 
 ```python
-ch = fluid.make_chan(dtype=Tensor, etype=float16)
+ch = fluid.make_channel(dtype=Tensor, etype=float16)
 ```
 
 or Tensors of Tensors of float16 etc.
@@ -87,8 +87,135 @@ The point here is that we need a consistent way to compose types, like in C++ we
 
 ### Send and Recv
 
+Go's CSP implementation depends on data type *channel*. There are two types of channels:
+
+1. The unblocked channel, or buffered channel, is a blocking queue with a non-zero sized buffer. The sending to buffered channel blocks if the buffer is full, and the receive operation blocks if the buffer is empty.
+1. blocked channel, or unbuffered channel, is a blocking queue with no buffer.  Both sending and receiving block with unbuffered channels.
+
+There are four types of actions with a channel:
+
+1. Create a channel
+
+   ```go
+   ch := make(chan int) // this is an unbuffered channel
+   ch := make(chan int, 100) // this is a buffered channel of 100 ints.
+   ```
+
+1. Send
+
+   ```go
+   ch <- 111
+   ```
+
+1. Recv
+
+   ```go
+   y, ok <- ch
+   ```
+
+1. Close
+
+   ```go
+   close(ch)
+   ```
+   
+   Please be aware that a closed channel is not a nil channel, which is `var ch chan int`.
+   
+There are some [axioms with channels](https://dave.cheney.net/2014/03/19/channel-axioms):
+
+1. A send to a nil channel blocks forever
+
+1. A receive from a nil channel blocks forever
+
+1. A send to a closed channel panics
+
+1. A receive from a closed channel returns the residual values and then zeros.
+
+In Fluid, we have [buffered channels](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/details/buffered_channel.h) and [unbuffered channels](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/details/unbuffered_channel.h)
+
+The following program illustrates the Python syntax for accessing Fluid buffers.
+
+```python
+import fluid
+
+buffer_size = 10
+ch = fluid.make_channel(dtype=INT, buffer_size)
+
+# Now write three elements to the channel
+with fluid.while(steps=buffer_size):
+  fluid.send(ch, step)
+  fluid.close_channel(ch)
+  
+with fluid.while(steps=buffer_size):
+  fluid.print(fluid.recv(ch))
+```
+
+The following example shows that to avoid the always-blocking behavior of unbuffered channels, we need to use Fluid's goroutines.
+
+```python
+import fluid
+
+ch = fluid.make_channel(dtype=INT)
+
+with fluid.go():
+  fluid.send(ch)
+
+y = fluid.recv(ch)
+
+fluid.close_channel(ch)
+```
+
 ### Select
 
+In Go, the `select` statement lets a goroutine wait on multiple communication operations. A `select` blocks until one of its cases can run, then it executes that case. It chooses one at random if multiple are ready.
+
+```go
+
+ch1  := make(chan int)       
+ch2  := make(chan int, 100)
+
+x := 0
+
+for {
+    select {
+    case ch1 <- x:
+      x := x + 1
+    case y <- ch2:
+      fmt.Println("Received on channel")
+    default:
+      fmt.Println("Default")
+    }
+  }
+
+```
+
+In Fluid, we should be able to do the same:
+
+```python
+ch1  = fluid.make_chan(dtype=INT)
+ch2 = fluid.make_chan(dtype=INT, 100)
+
+sel = fluid.select()
+
+with sel.case(ch1, 'w', X):
+    fluid.layers.increment(X)
+
+with sel.case(ch2, 'r', Y):
+    fluid.print("Received on Channel")
+
+with sel.default():
+    fluid.print("Default")
+
+```
+
+In the above code snippet, `X` and `Y` are variables. Now let us look at each of these statements one by one.
+
+- `sel.case(ch1, 'w', X)` : This specifies that we are writing to `ch1` and we want to write the integer in variable `X` to the channel. The character `w` is used here to make the syntax familiar to write syntax in Python I/O.
+
+- `sel.case(ch2, 'r', Y)` : This specifies that we would like to read the result from `ch2` into variable `Y`. The character `r` is used here to make the syntax familiar to read syntax in Python I/O.
+
+- `sel.default()` : This is equivalent to the default in Go `select`. If none of the channels are ready for read or write, then the fluid code in the default block will be executed.
+
 ## Example Programs
 
 ### 1. RPC between Trainers and Parameter Servers

doc/design/switch.md

+### Design Doc: Switch
+
+### Background
+
+Many programming languages provide `switch` as a generalization of `if-elif-else`.  We want to add it to Fluid.
+
+The following example shows the usage of `fluid.switch`.
+
+```python
+a = fluid.Var(10)
+b = fluid.Var(0)
+
+switch = fluid.switch()
+with switch.block():
+    with switch.case(fluid.less_equal(a, 10)):
+        fluid.print("Case 1")
+    with switch.case(fluid.larger(a, 0)):
+        fluid.print("Case 2")
+    with switch.default():
+        fluid.print("Case 3")
+```
+
+### The Semantics
+
+1. A `switch` control-flow checks cases one-by-one.
+1. The condition of each case is a boolean value, which is a scalar, and differs from the `fluid.if_else` control-flow, which condition could be a vector of boolean values.
+1. It runs the first matched case, or the default case if there is one.
+1. Once it matches a case, it runs the corresponding branch and only that branch.  It's like there is a C's `break` keyword at the end of each case.
+
+The above program should print and print only "Case 1".
+
+The implementation of the backward pass of the `switch` control-flow is easier than the backward of the `if_else`, because `switch` runs at most one branch, whereas `if-else` could run more than one branches.

doc/howto/usage/cluster/cluster_train_cn.md

@@ -92,11 +92,11 @@ paddle.init(
 参数说明
 
 - use_gpu： **可选，默认False**，是否启用GPU训练
-- trainer_count：**必选，默认1**，当前训练任务trainer总个数
+- trainer_count：**必选，默认1**，当前trainer的线程数目
 - port：**必选，默认7164**，连接到pserver的端口
 - ports_num：**必选，默认1**，连接到pserver的端口个数
 - ports_num_for_sparse：**必选，默认0**，和pserver之间用于稀疏类型参数通信的端口个数
-- num_gradient_servers：**必选，默认1**，当前训练任务pserver总数
+- num_gradient_servers：**必选，默认1**，当前训练任务trainer总数
 - trainer_id：**必选，默认0**，每个trainer的唯一ID，从0开始的整数
 - pservers：**必选，默认127.0.0.1**，当前训练任务启动的pserver的IP列表，多个IP使用“,”隔开
 

doc/howto/usage/cluster/cluster_train_en.md

@@ -95,11 +95,11 @@ paddle.init(
 Parameter Description
 
 - use_gpu: **optional, default False**, set to "True" to enable GPU training.
-- trainer_count: **required, default 1**, total count of trainers in the training job.
+- trainer_count: **required, default 1**, number of threads in current trainer.
 - port: **required, default 7164**, port to connect to parameter server.
 - ports_num: **required, default 1**, number of ports for communication.
 - ports_num_for_sparse: **required, default 0**, number of ports for sparse type caculation.
-- num_gradient_servers: **required, default 1**, total number of gradient server.
+- num_gradient_servers: **required, default 1**, number of trainers in current job.
 - trainer_id: **required, default 0**, ID for every trainer, start from 0.
 - pservers: **required, default 127.0.0.1**, list of IPs of parameter servers, separated by ",".
 

doc/index_cn.rst

@@ -8,4 +8,3 @@ PaddlePaddle 文档
   howto/index_cn.rst
   api/index_cn.rst
   faq/index_cn.rst
-  mobile/index_cn.rst

doc/index_en.rst

@@ -7,4 +7,3 @@ PaddlePaddle Documentation
   getstarted/index_en.rst
   howto/index_en.rst
   api/index_en.rst
-  mobile/index_en.rst

doc/mobile/index_cn.rst

-MOBILE
-======
-
-..  toctree::
-  :maxdepth: 1
-
-  cross_compiling_for_android_cn.md
-  cross_compiling_for_ios_cn.md
-  cross_compiling_for_raspberry_cn.md

doc/mobile/index_en.rst

-MOBILE
-======
-
-..  toctree::
-  :maxdepth: 1
-
-  cross_compiling_for_android_en.md
-  cross_compiling_for_ios_en.md
-  cross_compiling_for_raspberry_en.md

paddle/framework/backward.cc

@@ -534,7 +534,7 @@ ParamGradInfoMap AppendBackward(
   auto root_block = program_desc.MutableBlock(root_block_idx);
 
   std::string fill_one_op_out = GradVarName(target.Name());
-  bool is_scalar = target.Shape() == std::vector<int64_t>{1};
+  bool is_scalar = target.GetShape() == std::vector<int64_t>{1};
   PADDLE_ENFORCE(is_scalar, "target should be scalar");
   VLOG(3) << "backward from loss=" << target.Name()
           << " data_type=" << target.GetDataType();
@@ -565,7 +565,7 @@ ParamGradInfoMap AppendBackward(
 
   auto var = root_block->Var(fill_one_op_out);
   var->SetDataType(target.GetDataType());
-  var->SetShape(target.Shape());
+  var->SetShape(target.GetShape());
   auto& target_grad = retv[target.Name()];
   target_grad.name_ = fill_one_op_out;
   target_grad.block_idx_ = root_block_idx;

paddle/framework/channel.h

@@ -23,8 +23,8 @@ namespace framework {
 template <typename T>
 class Channel {
  public:
-  virtual void Send(T*) = 0;
-  virtual void Receive(T*) = 0;
+  virtual bool Send(T*) = 0;
+  virtual bool Receive(T*) = 0;
   virtual size_t Cap() = 0;
   virtual void Close() = 0;
   virtual ~Channel() {}

paddle/framework/channel_test.cc

@@ -29,16 +29,16 @@ TEST(Channel, MakeAndClose) {
   {
     // MakeChannel should return a buffered channel is buffer_size > 0.
     auto ch = MakeChannel<int>(10);
-    EXPECT_NE(dynamic_cast<Buffered<int>*>(ch), nullptr);
-    EXPECT_EQ(dynamic_cast<UnBuffered<int>*>(ch), nullptr);
+    EXPECT_NE(dynamic_cast<Buffered<int> *>(ch), nullptr);
+    EXPECT_EQ(dynamic_cast<UnBuffered<int> *>(ch), nullptr);
     CloseChannel(ch);
     delete ch;
   }
   {
     // MakeChannel should return an un-buffered channel is buffer_size = 0.
     auto ch = MakeChannel<int>(0);
-    EXPECT_EQ(dynamic_cast<Buffered<int>*>(ch), nullptr);
-    EXPECT_NE(dynamic_cast<UnBuffered<int>*>(ch), nullptr);
+    EXPECT_EQ(dynamic_cast<Buffered<int> *>(ch), nullptr);
+    EXPECT_NE(dynamic_cast<UnBuffered<int> *>(ch), nullptr);
     CloseChannel(ch);
     delete ch;
   }
@@ -48,12 +48,12 @@ TEST(Channel, SufficientBufferSizeDoesntBlock) {
   const size_t buffer_size = 10;
   auto ch = MakeChannel<size_t>(buffer_size);
   for (size_t i = 0; i < buffer_size; ++i) {
-    ch->Send(&i);  // should not block
+    EXPECT_EQ(ch->Send(&i), true);  // should not block
   }
 
   size_t out;
   for (size_t i = 0; i < buffer_size; ++i) {
-    ch->Receive(&out);  // should not block
+    EXPECT_EQ(ch->Receive(&out), true);  // should not block
     EXPECT_EQ(out, i);
   }
   CloseChannel(ch);
@@ -67,7 +67,10 @@ TEST(Channel, ConcurrentSendNonConcurrentReceiveWithSufficientBufferSize) {
   std::thread t([&]() {
     // Try to write more than buffer size.
     for (size_t i = 0; i < 2 * buffer_size; ++i) {
-      ch->Send(&i);  // should not block
+      if (i < buffer_size)
+        EXPECT_EQ(ch->Send(&i), true);  // should block after 10 iterations
+      else
+        EXPECT_EQ(ch->Send(&i), false);
       sum += i;
     }
   });
@@ -78,3 +81,262 @@ TEST(Channel, ConcurrentSendNonConcurrentReceiveWithSufficientBufferSize) {
   t.join();
   delete ch;
 }
+
+TEST(Channel, SimpleUnbufferedChannelTest) {
+  auto ch = MakeChannel<int>(0);
+  unsigned sum_send = 0;
+  std::thread t([&]() {
+    for (int i = 0; i < 5; i++) {
+      EXPECT_EQ(ch->Send(&i), true);
+      sum_send += i;
+    }
+  });
+  for (int i = 0; i < 5; i++) {
+    int recv;
+    EXPECT_EQ(ch->Receive(&recv), true);
+    EXPECT_EQ(recv, i);
+  }
+
+  CloseChannel(ch);
+  t.join();
+  EXPECT_EQ(sum_send, 10U);
+  delete ch;
+}
+
+// This tests that closing a buffered channel also unblocks
+//  any receivers waiting on the channel
+TEST(Channel, BufferedChannelCloseUnblocksReceiversTest) {
+  auto ch = MakeChannel<int>(1);
+  size_t num_threads = 5;
+  std::thread t[num_threads];
+  bool thread_ended[num_threads];
+
+  // Launches threads that try to read and are blocked because of no writers
+  for (size_t i = 0; i < num_threads; i++) {
+    thread_ended[i] = false;
+    t[i] = std::thread(
+        [&](bool *p) {
+          int data;
+          // All reads should return false
+          EXPECT_EQ(ch->Receive(&data), false);
+          *p = true;
+        },
+        &thread_ended[i]);
+  }
+  std::this_thread::sleep_for(std::chrono::milliseconds(100));  // wait
+
+  // Verify that all threads are blocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], false);
+  }
+
+  // Explicitly close the channel
+  // This should unblock all receivers
+  CloseChannel(ch);
+
+  std::this_thread::sleep_for(std::chrono::milliseconds(200));  // wait
+
+  // Verify that all threads got unblocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], true);
+  }
+
+  for (size_t i = 0; i < num_threads; i++) t[i].join();
+  delete ch;
+}
+
+// This tests that closing a buffered channel also unblocks
+//  any senders waiting for channel to have write space
+TEST(Channel, BufferedChannelCloseUnblocksSendersTest) {
+  auto ch = MakeChannel<int>(1);
+  size_t num_threads = 5;
+  std::thread t[num_threads];
+  bool thread_ended[num_threads];
+  bool send_success[num_threads];
+
+  // Launches threads that try to write and are blocked because of no readers
+  for (size_t i = 0; i < num_threads; i++) {
+    thread_ended[i] = false;
+    send_success[i] = false;
+    t[i] = std::thread(
+        [&](bool *ended, bool *success) {
+          int data = 10;
+          *success = ch->Send(&data);
+          *ended = true;
+        },
+        &thread_ended[i], &send_success[i]);
+  }
+  std::this_thread::sleep_for(std::chrono::milliseconds(100));  // wait
+
+  // Verify that atleast 4 threads are blocked
+  int ct = 0;
+  for (size_t i = 0; i < num_threads; i++) {
+    if (thread_ended[i] == false) ct++;
+  }
+  // Atleast 4 threads must be blocked
+  EXPECT_GE(ct, 4);
+
+  // Explicitly close the thread
+  // This should unblock all senders
+  CloseChannel(ch);
+
+  std::this_thread::sleep_for(std::chrono::milliseconds(200));  // wait
+
+  // Verify that all threads got unblocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], true);
+  }
+
+  // Verify that only 1 send was successful
+  ct = 0;
+  for (size_t i = 0; i < num_threads; i++) {
+    if (send_success[i]) ct++;
+  }
+  // Only 1 send must be successful
+  EXPECT_EQ(ct, 1);
+
+  for (size_t i = 0; i < num_threads; i++) t[i].join();
+  delete ch;
+}
+
+// This tests that closing an unbuffered channel also unblocks
+//  unblocks any receivers waiting for senders
+TEST(Channel, UnbufferedChannelCloseUnblocksReceiversTest) {
+  auto ch = MakeChannel<int>(0);
+  size_t num_threads = 5;
+  std::thread t[num_threads];
+  bool thread_ended[num_threads];
+
+  // Launches threads that try to read and are blocked becausew of no writers
+  for (size_t i = 0; i < num_threads; i++) {
+    thread_ended[i] = false;
+    t[i] = std::thread(
+        [&](bool *p) {
+          int data;
+          EXPECT_EQ(ch->Receive(&data), false);
+          *p = true;
+        },
+        &thread_ended[i]);
+  }
+  std::this_thread::sleep_for(std::chrono::milliseconds(500));  // wait 0.5 sec
+
+  // Verify that all the threads are blocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], false);
+  }
+
+  // Explicitly close the thread
+  // This should unblock all receivers
+  CloseChannel(ch);
+
+  std::this_thread::sleep_for(std::chrono::milliseconds(500));  // wait 0.5 sec
+
+  // Verify that all threads got unblocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], true);
+  }
+
+  for (size_t i = 0; i < num_threads; i++) t[i].join();
+  delete ch;
+}
+
+// This tests that closing an unbuffered channel also unblocks
+//  unblocks any senders waiting for senders
+TEST(Channel, UnbufferedChannelCloseUnblocksSendersTest) {
+  auto ch = MakeChannel<int>(0);
+  size_t num_threads = 5;
+  std::thread t[num_threads];
+  bool thread_ended[num_threads];
+
+  // Launches threads that try to read and are blocked becausew of no writers
+  for (size_t i = 0; i < num_threads; i++) {
+    thread_ended[i] = false;
+    t[i] = std::thread(
+        [&](bool *p) {
+          int data = 10;
+          EXPECT_EQ(ch->Send(&data), false);
+          *p = true;
+        },
+        &thread_ended[i]);
+  }
+  std::this_thread::sleep_for(std::chrono::milliseconds(500));  // wait 0.5 sec
+
+  // Verify that all the threads are blocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], false);
+  }
+
+  // Explicitly close the thread
+  // This should unblock all receivers
+  CloseChannel(ch);
+
+  std::this_thread::sleep_for(std::chrono::milliseconds(500));  // wait 0.5 sec
+
+  // Verify that all threads got unblocked
+  for (size_t i = 0; i < num_threads; i++) {
+    EXPECT_EQ(thread_ended[i], true);
+  }
+
+  for (size_t i = 0; i < num_threads; i++) t[i].join();
+  delete ch;
+}
+
+TEST(Channel, UnbufferedLessReceiveMoreSendTest) {
+  auto ch = MakeChannel<int>(0);
+  unsigned sum_send = 0;
+  // Send should block after three iterations
+  // since we only have three receivers.
+  std::thread t([&]() {
+    // Try to send more number of times
+    // than receivers
+    for (int i = 0; i < 4; i++) {
+      ch->Send(&i);
+      sum_send += i;
+    }
+  });
+  for (int i = 0; i < 3; i++) {
+    int recv;
+    ch->Receive(&recv);
+    EXPECT_EQ(recv, i);
+  }
+  std::this_thread::sleep_for(std::chrono::milliseconds(100));  // wait 0.5 sec
+  EXPECT_EQ(sum_send, 3U);
+
+  CloseChannel(ch);
+  t.join();
+  delete ch;
+}
+
+TEST(Channel, UnbufferedMoreReceiveLessSendTest) {
+  auto ch = MakeChannel<int>(0);
+  unsigned sum_send = 0;
+  unsigned sum_receive = 0;
+  // The receiver should block after 5
+  // iterations, since there are only 5 senders.
+  std::thread t([&]() {
+    for (int i = 0; i < 8; i++) {
+      int recv;
+      ch->Receive(&recv);  // should block after the fifth iteration.
+      EXPECT_EQ(recv, i);
+      sum_receive += i;
+    }
+  });
+  for (int i = 0; i < 5; i++) {
+    ch->Send(&i);
+    sum_send += i;
+  }
+  std::this_thread::sleep_for(std::chrono::milliseconds(500));  // wait 0.5 sec
+  EXPECT_EQ(sum_send, 10U);
+  EXPECT_EQ(sum_receive, 10U);
+  // send three more elements
+  for (int i = 5; i < 8; i++) {
+    ch->Send(&i);
+    sum_send += i;
+  }
+
+  CloseChannel(ch);
+  t.join();
+  EXPECT_EQ(sum_send, 28U);
+  EXPECT_EQ(sum_receive, 28U);
+  delete ch;
+}

paddle/framework/details/buffered_channel.h

@@ -30,8 +30,8 @@ class Buffered : public paddle::framework::Channel<T> {
   friend void paddle::framework::CloseChannel<T>(Channel<T>*);
 
  public:
-  virtual void Send(T*);
-  virtual void Receive(T*);
+  virtual bool Send(T*);
+  virtual bool Receive(T*);
   virtual size_t Cap() { return cap_; }
   virtual void Close();
   virtual ~Buffered();
@@ -48,39 +48,43 @@ class Buffered : public paddle::framework::Channel<T> {
     PADDLE_ENFORCE_GT(cap, 0);
   }
 
-  void NotifyAllSenders(std::unique_lock<std::mutex>*);
+  void NotifyAllParticipants(std::unique_lock<std::mutex>*);
 };
 
 template <typename T>
-void Buffered<T>::Send(T* item) {
+bool Buffered<T>::Send(T* item) {
   std::unique_lock<std::mutex> lock(mu_);
   full_cond_var_.wait(lock,
                       [this]() { return channel_.size() < cap_ || closed_; });
+  bool ret = false;
   if (!closed_) {
     channel_.push_back(std::move(*item));
     lock.unlock();
     empty_cond_var_.notify_one();
+    ret = true;
   }
+  return ret;
 }
 
 template <typename T>
-void Buffered<T>::Receive(T* item) {
+bool Buffered<T>::Receive(T* item) {
   std::unique_lock<std::mutex> lock(mu_);
   empty_cond_var_.wait(lock, [this]() { return !channel_.empty() || closed_; });
-  if (!closed_) {
+  bool ret = false;
+  if (!channel_.empty()) {
     *item = std::move(channel_.front());
     channel_.pop_front();
-    NotifyAllSenders(&lock);
-  } else {
-    item = nullptr;
+    full_cond_var_.notify_one();
+    ret = true;
   }
+  return ret;
 }
 
 template <typename T>
 void Buffered<T>::Close() {
   std::unique_lock<std::mutex> lock(mu_);
   closed_ = true;
-  NotifyAllSenders(&lock);
+  NotifyAllParticipants(&lock);
 }
 
 template <typename T>
@@ -88,13 +92,14 @@ Buffered<T>::~Buffered() {
   std::unique_lock<std::mutex> lock(mu_);
   closed_ = true;
   channel_.clear();
-  NotifyAllSenders(&lock);
+  NotifyAllParticipants(&lock);
 }
 
 template <typename T>
-void Buffered<T>::NotifyAllSenders(std::unique_lock<std::mutex>* lock) {
+void Buffered<T>::NotifyAllParticipants(std::unique_lock<std::mutex>* lock) {
   lock->unlock();
   full_cond_var_.notify_all();
+  empty_cond_var_.notify_all();
 }
 
 }  // namespace details

paddle/framework/details/unbuffered_channel.h

-/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -13,8 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #pragma once
+#include <atomic>
 #include <condition_variable>
-#include <deque>
 #include <mutex>
 
 #include "paddle/framework/channel.h"
@@ -29,27 +29,117 @@ class UnBuffered : public paddle::framework::Channel<T> {
   friend void paddle::framework::CloseChannel<T>(Channel<T>*);
 
  public:
-  virtual void Send(T*);
-  virtual void Receive(T*);
+  virtual bool Send(T*);
+  virtual bool Receive(T*);
   virtual size_t Cap() { return 0; }
   virtual void Close();
   virtual ~UnBuffered();
 
  private:
-  UnBuffered() {}
+  std::mutex mu_ch_;
+  // Mutex for readers and writers who are waiting for other reader
+  // and writer to complete execution
+  std::recursive_mutex mu_read_, mu_write_;
+  // reader_found_ is set true when a reader is ready to accept data
+  // writer_found_ is set true when a writer is ready to send data
+  // A transaction occurs only when both are true
+  std::atomic<bool> reader_found_{false}, writer_found_{false};
+  std::condition_variable cv_channel_;
+  std::condition_variable_any cv_reader_, cv_writer_;
+  T* item{nullptr};
+  std::atomic<bool> closed_{false};
+
+  UnBuffered() : closed_(false) {}
+
+  void NotifyAllParticipants(std::unique_lock<std::mutex>*);
 };
 
+// This function implements the concept of how data should
+// be sent from a writer to a reader.
+template <typename T>
+bool UnBuffered<T>::Send(T* data) {
+  // Prevent other writers from entering
+  std::unique_lock<std::recursive_mutex> writer_lock(mu_write_);
+  writer_found_ = true;
+  std::unique_lock<std::recursive_mutex> cv_lock(mu_write_);
+  // If writer comes first, it should wait till a reader arrives
+  cv_writer_.wait(cv_lock,
+                  [this]() { return reader_found_ == true || closed_; });
+  cv_reader_.notify_one();
+  bool ret = false;
+  if (!closed_) {
+    std::unique_lock<std::mutex> channel_lock(mu_ch_);
+    item = data;
+    channel_lock.unlock();
+    cv_channel_.notify_one();
+    channel_lock.lock();
+    cv_channel_.wait(channel_lock,
+                     [this]() { return item == nullptr || closed_; });
+    ret = true;
+  }
+  writer_found_ = false;
+  return ret;
+}
+
+// This function implements the concept of how
+// data that was sent by a writer is read from a reader.
 template <typename T>
-void UnBuffered<T>::Send(T* channel_element) {}
+bool UnBuffered<T>::Receive(T* data) {
+  // Prevent other readers from entering
+  std::unique_lock<std::recursive_mutex> read_lock{mu_read_};
+  reader_found_ = true;
+  std::unique_lock<std::recursive_mutex> cv_lock{mu_read_};
+  // If reader comes first, it should wait till a writer arrives
+  cv_reader_.wait(cv_lock,
+                  [this]() { return writer_found_ == true || closed_; });
+  cv_writer_.notify_one();
+  bool ret = false;
+  if (!closed_) {
+    std::unique_lock<std::mutex> lock_ch{mu_ch_};
+    // Reader should wait for the writer to first write its data
+    cv_channel_.wait(lock_ch, [this]() { return item != nullptr || closed_; });
+    if (!closed_) {
+      *data = std::move(*item);
+      item = nullptr;
+      lock_ch.unlock();
+      ret = true;
+    }
+    cv_channel_.notify_one();
+  }
+  reader_found_ = false;
+  return ret;
+}
 
+// This function implements the sequence of events
+// that take place once the channel is closed.
 template <typename T>
-void UnBuffered<T>::Receive(T*) {}
+void UnBuffered<T>::Close() {
+  std::unique_lock<std::mutex> lock(mu_ch_);
+  item = nullptr;
+  closed_ = true;
+  NotifyAllParticipants(&lock);
+}
 
+// This function implements the sequence of events
+// that are executed once the object of an UnBuffered
+// channel is destroyed.
 template <typename T>
-void UnBuffered<T>::Close() {}
+UnBuffered<T>::~UnBuffered() {
+  std::unique_lock<std::mutex> lock(mu_ch_);
+  item = nullptr;
+  closed_ = true;
+  NotifyAllParticipants(&lock);
+}
 
+// This function notifies all the readers, writers and
+// the channel condition variables.
 template <typename T>
-UnBuffered<T>::~UnBuffered() {}
+void UnBuffered<T>::NotifyAllParticipants(std::unique_lock<std::mutex>* lock) {
+  lock->unlock();
+  cv_writer_.notify_all();
+  cv_channel_.notify_all();
+  cv_reader_.notify_all();
+}
 
 }  // namespace details
 }  // namespace framework

paddle/framework/framework.proto

@@ -116,6 +116,8 @@ message LoDTensorArrayDesc {
   optional int32 lod_level = 2 [ default = 0 ];
 }
 
+message Reader { repeated LoDTensorDesc lod_tensor = 1; }
+
 message VarDesc {
   enum VarType {
     LOD_TENSOR = 1;
@@ -126,13 +128,15 @@ message VarDesc {
     LOD_RANK_TABLE = 6;
     LOD_TENSOR_ARRAY = 7;
     PLACE_LIST = 8;
+    READER = 9;
   }
   required string name = 1;
   required VarType type = 2;
-  optional LoDTensorDesc lod_tensor = 3;
-  optional TensorDesc selected_rows = 4;
+  optional bool persistable = 3 [ default = false ];
+  optional LoDTensorDesc lod_tensor = 4;
+  optional TensorDesc selected_rows = 5;
   optional LoDTensorArrayDesc tensor_array = 6;
-  optional bool persistable = 5 [ default = false ];
+  optional Reader reader = 7;
 }
 
 message BlockDesc {

paddle/framework/mixed_vector.h

@@ -34,18 +34,6 @@ namespace framework {
 
 template <typename T>
 class Vector : public std::vector<T> {
- public:
-  /* NOTE(dzhwinter):
-   * Data always store and modified on Host.
-   * If the data is modified when use cuda_data interface,
-   * You need to call the CopyFromCUDA explicitly to synchronize data.
-   *
-   */
-  enum class kDataPosition {
-    kDataOnHost = 0,
-    kDataOnDevice = 1,
-  };
-
  public:
   using std::vector<T>::vector;
 
@@ -55,11 +43,12 @@ class Vector : public std::vector<T> {
   virtual ~Vector() {
 #ifdef PADDLE_WITH_CUDA
     if (cuda_ptr_ != nullptr) {
-      memory::Free<platform::CUDAPlace>(place_, static_cast<void *>(cuda_ptr_));
+      memory::Free<platform::CUDAPlace>(place_, cuda_ptr_);
     }
 #endif
   }
 
+  /* Get device vector */
   T *cuda_data() {
     CopyToCUDA();
     PADDLE_ENFORCE_NOT_NULL(
@@ -67,81 +56,73 @@ class Vector : public std::vector<T> {
     return static_cast<T *>(cuda_ptr_);
   }
 
+  /* Get host vector */
   T *data() { return std::vector<T>::data(); }
-
   const T *data() const { return std::vector<T>::data(); }
 
+  /* Synchronize host vector to device vector */
   void CopyToCUDA();
-
+  /* Synchronize device vector to host vector */
   void CopyFromCUDA();
-
+  /* Switch device vector location */
   void CopyToPeer(platform::Place);
 
  private:
   void *cuda_ptr_ = nullptr;
-  size_t cuda_size_ = 0;
-  /*The DataPosition is unused now,
-    if we want support random access from cpu and cuda,
-    we need to overload all the vector method */
-
-  kDataPosition position_ = kDataPosition::kDataOnHost;
+  size_t cuda_size_ = 0;  // device vector numel
   platform::CUDAPlace place_;
 };
 
 template <typename T>
 void Vector<T>::CopyToCUDA() {
 #ifdef PADDLE_WITH_CUDA
-  if (cuda_ptr_ == nullptr) {
+  if (cuda_size_ < this->size()) {
+    if (cuda_ptr_ != nullptr) {
+      memory::Free<platform::CUDAPlace>(place_, cuda_ptr_);
+    }
     cuda_ptr_ =
         memory::Alloc<platform::CUDAPlace>(place_, this->size() * sizeof(T));
   }
+  cuda_size_ = this->size();
   platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
-  auto *cuda_ctx = pool.GetByPlace(place_);
-
-  memory::Copy(place_, static_cast<void *>(cuda_ptr_), platform::CPUPlace(),
+  auto *ctx = pool.GetByPlace(place_);
+  memory::Copy(place_, cuda_ptr_, platform::CPUPlace(),
                static_cast<const void *>(this->data()),
-               this->size() * sizeof(T), cuda_ctx->stream());
-  cuda_ctx->Wait();
-
-  cuda_size_ = this->size();
+               this->size() * sizeof(T), ctx->stream());
+  ctx->Wait();
 #endif
 }
 
 template <typename T>
 void Vector<T>::CopyFromCUDA() {
 #ifdef PADDLE_WITH_CUDA
-  platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
-  auto *cuda_ctx = pool.GetByPlace(place_);
   if (cuda_ptr_ == nullptr) {
-    LOG(WARNING) << "No uncommited cuda data.";
+    LOG(WARNING) << "No uncommitted cuda data.";
     return;
   }
   this->resize(cuda_size_);
+  platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
+  auto *ctx = pool.GetByPlace(place_);
   memory::Copy(platform::CPUPlace(), static_cast<void *>(this->data()), place_,
                static_cast<const void *>(cuda_ptr_), this->size() * sizeof(T),
-               cuda_ctx->stream());
-  cuda_ctx->Wait();
-
+               ctx->stream());
+  ctx->Wait();
 #endif
 }
 
 template <typename T>
 void Vector<T>::CopyToPeer(platform::Place peer_place) {
-  if (platform::is_cpu_place(peer_place)) {
-    return;
-  }
 #ifdef PADDLE_WITH_CUDA
-  auto *cuda_ctx = platform::DeviceContextPool::Instance().GetByPlace(place_);
-  void *peer_cuda_ptr_ = memory::Alloc<platform::CUDAPlace>(
+  auto *ctx = platform::DeviceContextPool::Instance().GetByPlace(place_);
+  void *peer_cuda_ptr = memory::Alloc<platform::CUDAPlace>(
       boost::get<platform::CUDAPlace>(peer_place), this->size() * sizeof(T));
-  memory::Copy(boost::get<platform::CUDAPlace>(peer_place),
-               static_cast<void *>(peer_cuda_ptr_), place_,
-               static_cast<const void *>(cuda_ptr_), this->size() * sizeof(T),
-               cuda_ctx->stream());
-  cuda_ctx->Wait();
-  memory::Free<platform::CUDAPlace>(place_, static_cast<void *>(cuda_ptr_));
+  memory::Copy(boost::get<platform::CUDAPlace>(peer_place), peer_cuda_ptr,
+               place_, cuda_ptr_, this->size() * sizeof(T), ctx->stream());
+  ctx->Wait();
+
+  memory::Free<platform::CUDAPlace>(place_, cuda_ptr_);
   place_ = boost::get<platform::CUDAPlace>(peer_place);
-  cuda_ptr_ = peer_cuda_ptr_;
+  cuda_ptr_ = peer_cuda_ptr;
 #endif
 }
 

paddle/framework/op_desc.cc

@@ -39,10 +39,6 @@ class CompileTimeInferShapeContext : public InferShapeContext {
 
   bool HasOutputs(const std::string &name) const override;
 
-  DDim GetInputDim(const std::string &name) const override;
-
-  void SetOutputDim(const std::string &name, const DDim &dim) override;
-
   AttrReader Attrs() const override;
 
   const std::vector<std::string> &Inputs(
@@ -444,21 +440,6 @@ bool CompileTimeInferShapeContext::HasOutputs(const std::string &name) const {
   return true;
 }
 
-DDim CompileTimeInferShapeContext::GetInputDim(const std::string &name) const {
-  std::vector<DDim> ddims = GetInputsDim(name);
-  auto length = ddims.size();
-  PADDLE_ENFORCE_EQ(length, 1UL,
-                    "Input(%s) should have 1 value, "
-                    "but it has %d now",
-                    name, length);
-  return ddims[0];
-}
-
-void CompileTimeInferShapeContext::SetOutputDim(const std::string &name,
-                                                const DDim &dim) {
-  SetOutputsDim(name, {dim});
-}
-
 AttrReader CompileTimeInferShapeContext::Attrs() const {
   return AttrReader(op_.GetAttrMap());
 }
@@ -477,11 +458,11 @@ DDim CompileTimeInferShapeContext::GetDim(const std::string &name) const {
   auto var = block_.FindVarRecursive(name);
   PADDLE_ENFORCE(var != nullptr, "Cannot find variable %s", name);
   try {
-    auto shape = var->Shape();
+    auto shape = var->GetShape();
     if (shape.empty()) {
       return framework::make_ddim({0UL});
     } else {
-      return framework::make_ddim(var->Shape());
+      return framework::make_ddim(var->GetShape());
     }
   } catch (...) {
     VLOG(5) << "GetDim of variable " << name << " error";

paddle/framework/operator.cc

@@ -366,14 +366,6 @@ class RuntimeInferShapeContext : public InferShapeContext {
     return true;
   }
 
-  DDim GetInputDim(const std::string& name) const override {
-    return GetDim(op_.Input(name));
-  }
-
-  void SetOutputDim(const std::string& name, const DDim& dim) override {
-    SetDim(op_.Output(name), dim);
-  }
-
   AttrReader Attrs() const override { return AttrReader(op_.Attrs()); }
 
   const std::vector<std::string>& Inputs(

paddle/framework/program_desc_test.cc

@@ -53,7 +53,7 @@ TEST(ProgramDesc, copy_ctor) {
     ASSERT_NE(copy, var_before);
     ASSERT_EQ(copy->Name(), var_before->Name());
     ASSERT_EQ(copy->GetType(), var_before->GetType());
-    ASSERT_EQ(copy->Shape(), var_before->Shape());
+    ASSERT_EQ(copy->GetShape(), var_before->GetShape());
     ASSERT_EQ(copy->Proto()->SerializeAsString(),
               var_before->Proto()->SerializeAsString());
   };
@@ -117,7 +117,7 @@ TEST(ProgramDescBind, serialize_and_deserialize) {
     ASSERT_NE(restored, var_before);
     ASSERT_EQ(restored->Name(), var_before->Name());
     ASSERT_EQ(restored->GetType(), var_before->GetType());
-    ASSERT_EQ(restored->Shape(), var_before->Shape());
+    ASSERT_EQ(restored->GetShape(), var_before->GetShape());
     ASSERT_EQ(restored->Proto()->SerializeAsString(),
               var_before->Proto()->SerializeAsString());
   };

paddle/framework/shape_inference.cc

@@ -18,10 +18,18 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
-std::vector<framework::DDim> InferShapeContext::GetInputsDim(
+DDim InferShapeContext::GetInputDim(const std::string &name) const {
+  const std::vector<std::string> &arg_names = Inputs(name);
+  PADDLE_ENFORCE_EQ(arg_names.size(), 1UL,
+                    "Input(%s) should hold one element, but now it holds %d",
+                    name, arg_names.size());
+  return this->GetDim(arg_names[0]);
+}
+
+std::vector<DDim> InferShapeContext::GetInputsDim(
     const std::string &name) const {
-  const std::vector<std::string> &names = Inputs(name);
-  return GetDims(names);
+  const std::vector<std::string> &arg_names = Inputs(name);
+  return GetDims(arg_names);
 }
 
 DDim InferShapeContext::GetInputsElementDim(const std::string &name,
@@ -30,24 +38,31 @@ DDim InferShapeContext::GetInputsElementDim(const std::string &name,
   return this->GetDim(names[idx]);
 }
 
-void InferShapeContext::SetOutputsDim(
-    const std::string &name, const std::vector<framework::DDim> &dims) {
+void InferShapeContext::SetOutputDim(const std::string &name, const DDim &dim) {
+  auto &arg_names = Outputs(name);
+  PADDLE_ENFORCE_EQ(arg_names.size(), 1UL,
+                    "Output(%s) should hold one element, but now it holds %d",
+                    name, arg_names.size());
+  SetDim(arg_names[0], dim);
+}
+
+void InferShapeContext::SetOutputsDim(const std::string &name,
+                                      const std::vector<DDim> &dims) {
   auto &names = Outputs(name);
   SetDims(names, dims);
 }
 
-std::vector<framework::DDim> InferShapeContext::GetDims(
+std::vector<DDim> InferShapeContext::GetDims(
     const std::vector<std::string> &names) const {
-  std::vector<framework::DDim> ret;
+  std::vector<DDim> ret;
   ret.reserve(names.size());
   std::transform(
       names.begin(), names.end(), std::back_inserter(ret),
       [this](const std::string &name) { return this->GetDim(name); });
   return ret;
 }
-
 void InferShapeContext::SetDims(const std::vector<std::string> &names,
-                                const std::vector<framework::DDim> &dims) {
+                                const std::vector<DDim> &dims) {
   size_t length = names.size();
   PADDLE_ENFORCE_EQ(length, dims.size());
   for (size_t i = 0; i < length; ++i) {

paddle/framework/shape_inference.h

@@ -35,14 +35,13 @@ class InferShapeContext {
   virtual bool HasInputs(const std::string &name) const = 0;
   virtual bool HasOutputs(const std::string &name) const = 0;
 
-  virtual framework::DDim GetInputDim(const std::string &name) const = 0;
+  DDim GetInputDim(const std::string &name) const;
 
-  std::vector<framework::DDim> GetInputsDim(const std::string &name) const;
+  std::vector<DDim> GetInputsDim(const std::string &name) const;
   DDim GetInputsElementDim(const std::string &name, int idx) const;
 
-  virtual void SetOutputDim(const std::string &name, const DDim &dim) = 0;
-  void SetOutputsDim(const std::string &name,
-                     const std::vector<framework::DDim> &dims);
+  void SetOutputDim(const std::string &name, const DDim &dim);
+  void SetOutputsDim(const std::string &name, const std::vector<DDim> &dims);
 
   virtual AttrReader Attrs() const = 0;
   virtual const std::vector<std::string> &Inputs(
@@ -57,15 +56,13 @@ class InferShapeContext {
 
   // Note: In while op, we need this to be public
   void SetDims(const std::vector<std::string> &names,
-               const std::vector<framework::DDim> &dims);
+               const std::vector<DDim> &dims);
 
  protected:
-  virtual framework::DDim GetDim(const std::string &name) const = 0;
-  virtual void SetDim(const std::string &name, const framework::DDim &dim) = 0;
-
-  std::vector<framework::DDim> GetDims(
-      const std::vector<std::string> &names) const;
+  virtual DDim GetDim(const std::string &name) const = 0;
+  virtual void SetDim(const std::string &name, const DDim &dim) = 0;
 
+  std::vector<DDim> GetDims(const std::vector<std::string> &names) const;
   std::vector<proto::VarDesc::VarType> GetVarTypes(
       const std::vector<std::string> &names) const;
 

paddle/framework/var_desc.cc

@@ -26,18 +26,91 @@ void VarDesc::SetShape(const std::vector<int64_t> &dims) {
   VectorToRepeated(dims, mutable_tensor_desc()->mutable_dims());
 }
 
+void VarDesc::SetTensorDescNum(size_t num) {
+  switch (desc_.type()) {
+    case proto::VarDesc::READER: {
+      auto *lod_tensors_ptr = desc_.mutable_reader()->mutable_lod_tensor();
+      lod_tensors_ptr->Clear();
+      for (size_t i = 0; i < num; ++i) {
+        lod_tensors_ptr->Add();
+      }
+      return;
+    } break;
+    default:
+      PADDLE_THROW(
+          "Setting 'sub_tensor_number' is not supported by the type of var %s.",
+          this->Name());
+  }
+}
+
+size_t VarDesc::GetTensorDescNum() const {
+  switch (desc_.type()) {
+    case proto::VarDesc::READER:
+      return desc_.reader().lod_tensor_size();
+      break;
+    default:
+      PADDLE_THROW(
+          "Getting 'sub_tensor_number' is not supported by the type of var %s.",
+          this->Name());
+  }
+}
+
+void VarDesc::SetShapes(
+    const std::vector<std::vector<int64_t>> &multiple_dims) {
+  PADDLE_ENFORCE_EQ(multiple_dims.size(), GetTensorDescNum(),
+                    "The number of given shapes(%d) doesn't equal to the "
+                    "number of sub tensor.",
+                    multiple_dims.size(), GetTensorDescNum());
+  std::vector<proto::TensorDesc *> tensors = mutable_tensor_descs();
+  for (size_t i = 0; i < multiple_dims.size(); ++i) {
+    VectorToRepeated(multiple_dims[i], tensors[i]->mutable_dims());
+  }
+}
+
+std::vector<int64_t> VarDesc::GetShape() const {
+  return RepeatedToVector(tensor_desc().dims());
+}
+
+std::vector<std::vector<int64_t>> VarDesc::GetShapes() const {
+  std::vector<proto::TensorDesc> descs = tensor_descs();
+  std::vector<std::vector<int64_t>> res;
+  res.reserve(descs.size());
+  for (const auto &tensor_desc : descs) {
+    res.push_back(RepeatedToVector(tensor_desc.dims()));
+  }
+  return res;
+}
+
 void VarDesc::SetDataType(proto::DataType data_type) {
   mutable_tensor_desc()->set_data_type(data_type);
 }
 
-std::vector<int64_t> VarDesc::Shape() const {
-  return RepeatedToVector(tensor_desc().dims());
+void VarDesc::SetDataTypes(
+    const std::vector<proto::DataType> &multiple_data_type) {
+  PADDLE_ENFORCE_EQ(multiple_data_type.size(), GetTensorDescNum(),
+                    "The number of given data types(%d) doesn't equal to the "
+                    "number of sub tensor.",
+                    multiple_data_type.size(), GetTensorDescNum());
+  std::vector<proto::TensorDesc *> tensor_descs = mutable_tensor_descs();
+  for (size_t i = 0; i < multiple_data_type.size(); ++i) {
+    tensor_descs[i]->set_data_type(multiple_data_type[i]);
+  }
 }
 
 proto::DataType VarDesc::GetDataType() const {
   return tensor_desc().data_type();
 }
 
+std::vector<proto::DataType> VarDesc::GetDataTypes() const {
+  std::vector<proto::TensorDesc> descs = tensor_descs();
+  std::vector<proto::DataType> res;
+  res.reserve(descs.size());
+  for (const auto &tensor_desc : descs) {
+    res.push_back(tensor_desc.data_type());
+  }
+  return res;
+}
+
 void VarDesc::SetLoDLevel(int32_t lod_level) {
   switch (desc_.type()) {
     case proto::VarDesc::LOD_TENSOR:
@@ -47,8 +120,28 @@ void VarDesc::SetLoDLevel(int32_t lod_level) {
       desc_.mutable_tensor_array()->set_lod_level(lod_level);
       break;
     default:
-      PADDLE_THROW("Tensor type=%d does not support LoDLevel",
-                   desc_.tensor_array().lod_level());
+      PADDLE_THROW(
+          "Setting 'lod_level' is not supported by the type of var %s.",
+          this->Name());
+  }
+}
+
+void VarDesc::SetLoDLevels(const std::vector<int32_t> &multiple_lod_level) {
+  PADDLE_ENFORCE_EQ(multiple_lod_level.size(), GetTensorDescNum(),
+                    "The number of given data types(%d) doesn't equal to the "
+                    "number of sub tensor.",
+                    multiple_lod_level.size(), GetTensorDescNum());
+  switch (desc_.type()) {
+    case proto::VarDesc::READER: {
+      size_t i = 0;
+      for (auto &lod_tensor : *desc_.mutable_reader()->mutable_lod_tensor()) {
+        lod_tensor.set_lod_level(multiple_lod_level[i++]);
+      }
+    } break;
+    default:
+      PADDLE_THROW(
+          "Setting 'lod_levels' is not supported by the type of var %s.",
+          this->Name());
   }
 }
 
@@ -59,13 +152,31 @@ int32_t VarDesc::GetLoDLevel() const {
     case proto::VarDesc::LOD_TENSOR_ARRAY:
       return desc_.tensor_array().lod_level();
     default:
-      PADDLE_THROW("Tensor type=%d does not support LoDLevel",
-                   desc_.tensor_array().lod_level());
+      PADDLE_THROW(
+          "Getting 'lod_level' is not supported by the type of var %s.",
+          this->Name());
+  }
+}
+
+std::vector<int32_t> VarDesc::GetLoDLevels() const {
+  std::vector<int32_t> res;
+  switch (desc_.type()) {
+    case proto::VarDesc::READER:
+      res.reserve(desc_.reader().lod_tensor_size());
+      for (auto &lod_tensor : desc_.reader().lod_tensor()) {
+        res.push_back(lod_tensor.lod_level());
+      }
+      return res;
+      break;
+    default:
+      PADDLE_THROW(
+          "Getting 'lod_levels' is not supported by the type of var %s.",
+          this->Name());
   }
 }
 
 const proto::TensorDesc &VarDesc::tensor_desc() const {
-  PADDLE_ENFORCE(desc_.has_type(), "invoke TensorDesc must after set type");
+  PADDLE_ENFORCE(desc_.has_type(), "The var's type hasn't been set.");
   switch (desc_.type()) {
     case proto::VarDesc::SELECTED_ROWS:
       return desc_.selected_rows();
@@ -74,13 +185,32 @@ const proto::TensorDesc &VarDesc::tensor_desc() const {
     case proto::VarDesc::LOD_TENSOR_ARRAY:
       return desc_.tensor_array().tensor();
     default:
-      PADDLE_THROW("The type of var %s is unsupported.", this->Name());
+      PADDLE_THROW(
+          "Getting 'tensor_desc' is not supported by the type of var %s.",
+          this->Name());
+  }
+}
+
+std::vector<proto::TensorDesc> VarDesc::tensor_descs() const {
+  PADDLE_ENFORCE(desc_.has_type(), "The var type hasn't been set.");
+  std::vector<proto::TensorDesc> res;
+  res.reserve(GetTensorDescNum());
+  switch (desc_.type()) {
+    case proto::VarDesc::READER:
+      for (const auto &lod_tensor : desc_.reader().lod_tensor()) {
+        res.push_back(lod_tensor.tensor());
+      }
+      return res;
+    default:
+      PADDLE_THROW(
+          "Getting 'tensor_descs' is not supported by the type of var "
+          "%s.",
+          this->Name());
   }
 }
 
 proto::TensorDesc *VarDesc::mutable_tensor_desc() {
-  PADDLE_ENFORCE(desc_.has_type(),
-                 "invoke MutableTensorDesc must after set type");
+  PADDLE_ENFORCE(desc_.has_type(), "The var type hasn't been set.");
   switch (desc_.type()) {
     case proto::VarDesc::SELECTED_ROWS:
       return desc_.mutable_selected_rows();
@@ -89,8 +219,30 @@ proto::TensorDesc *VarDesc::mutable_tensor_desc() {
     case proto::VarDesc::LOD_TENSOR_ARRAY:
       return desc_.mutable_tensor_array()->mutable_tensor();
     default:
-      PADDLE_THROW("Unexpected branch.");
+      PADDLE_THROW(
+          "Getting 'mutable_tensor_desc' is not supported by the type of var "
+          "%s.",
+          this->Name());
   }
 }
+
+std::vector<proto::TensorDesc *> VarDesc::mutable_tensor_descs() {
+  PADDLE_ENFORCE(desc_.has_type(), "The var type hasn't been set.");
+  std::vector<proto::TensorDesc *> res;
+  res.reserve(GetTensorDescNum());
+  switch (desc_.type()) {
+    case proto::VarDesc::READER:
+      for (auto &lod_tensor : *desc_.mutable_reader()->mutable_lod_tensor()) {
+        res.push_back(lod_tensor.mutable_tensor());
+      }
+      return res;
+    default:
+      PADDLE_THROW(
+          "Getting 'tensor_descs' is not supported by the type of var "
+          "%s.",
+          this->Name());
+  }
+}
+
 }  // namespace framework
 }  // namespace paddle

paddle/framework/var_desc.h

@@ -68,18 +68,34 @@ class VarDesc {
 
   void SetName(std::string name) { desc_.set_name(name); }
 
+  void SetTensorDescNum(size_t num);
+
+  size_t GetTensorDescNum() const;
+
   void SetShape(const std::vector<int64_t> &dims);
 
+  void SetShapes(const std::vector<std::vector<int64_t>> &multiple_dims);
+
+  std::vector<int64_t> GetShape() const;
+
+  std::vector<std::vector<int64_t>> GetShapes() const;
+
   void SetDataType(proto::DataType data_type);
 
-  std::vector<int64_t> Shape() const;
+  void SetDataTypes(const std::vector<proto::DataType> &multiple_data_type);
 
   proto::DataType GetDataType() const;
 
+  std::vector<proto::DataType> GetDataTypes() const;
+
   void SetLoDLevel(int32_t lod_level);
 
+  void SetLoDLevels(const std::vector<int32_t> &multiple_lod_level);
+
   int32_t GetLoDLevel() const;
 
+  std::vector<int32_t> GetLoDLevels() const;
+
   proto::VarDesc::VarType GetType() const;
 
   void SetType(proto::VarDesc::VarType type);
@@ -90,7 +106,9 @@ class VarDesc {
 
  private:
   const proto::TensorDesc &tensor_desc() const;
+  std::vector<proto::TensorDesc> tensor_descs() const;
   proto::TensorDesc *mutable_tensor_desc();
+  std::vector<proto::TensorDesc *> mutable_tensor_descs();
 
   proto::VarDesc desc_;
 };

paddle/function/GemmConvOp.cpp

@@ -178,19 +178,22 @@ public:
     real* inputData = inputs[0].data<real>();
     real* filterData = inputs[1].data<real>();
     real* outputData = outputs[0].data<real>();
+    real* colData = NULL;
     bool needIm2col = isNeedIm2col(filter);
 
     TensorShape imShape =
         TensorShape({inputChannels / groups_, inputHeight, inputWidth});
-
     TensorShape colShape;
-    real* colData = NULL;
 
-    size_t colHeight = inputChannels / groups_ * filterHeight * filterWidth;
-    size_t colWidth = outputHeight * outputWidth;
-    // Max col matrix height 256, Max col matrix width 1024
-    size_t stepColHeight = std::min(colHeight, static_cast<size_t>(256));
-    size_t stepColWidth = std::min(colWidth, static_cast<size_t>(2048));
+    // Max col matrix width 4096, Max col matrix size 4M.
+    size_t outputHeightSteps =
+        std::min(std::max(4096 / outputWidth, (size_t)1), outputHeight);
+    size_t maxColWidth = outputHeightSteps * outputWidth;
+    size_t channelSteps =
+        std::min(std::max((1048576 / maxColWidth) / filterHeight * filterWidth,
+                          (size_t)1),
+                 inputChannels / groups_);
+    size_t maxColHeight = channelSteps * filterHeight * filterWidth;
 
     if (needIm2col) {
       colShape = TensorShape({inputChannels / groups_,
@@ -199,7 +202,7 @@ public:
                               outputHeight,
                               outputWidth});
 
-      resizeBuffer<Device>(stepColHeight * stepColWidth * sizeof(real));
+      resizeBuffer<Device>(maxColHeight * maxColWidth * sizeof(real));
       colData = reinterpret_cast<real*>(memory_->getBuf());
     }
 
@@ -209,20 +212,24 @@ public:
         (outputChannels / groups_) * outputHeight * outputWidth;
     size_t filterOffset = filter.getElements() / groups_;
 
-    int nStride = colWidth;
-    int kStride = colHeight;
+    int nStride = outputHeight * outputWidth;
+    int kStride = inputChannels / groups_ * filterHeight * filterWidth;
     for (size_t i = 0; i < batchSize; i++) {
+      filterData = inputs[1].data<real>();
       for (size_t g = 0; g < groups_; g++) {
         if (needIm2col) {
           real beta_ = beta;
-          for (size_t colHeightStart = 0; colHeightStart < colHeight;
-               colHeightStart += stepColHeight) {
-            for (size_t colWidthStart = 0; colWidthStart < colWidth;
-                 colWidthStart += stepColWidth) {
-              int N = std::min(colWidth - colWidthStart, stepColWidth);
-              int K = std::min(colHeight - colHeightStart, stepColHeight);
+          for (size_t ic = 0; ic < inputChannels / groups_;
+               ic += channelSteps) {
+            int channels = std::min(inputChannels / groups_ - ic, channelSteps);
+            for (size_t oh = 0; oh < outputHeight; oh += outputHeightSteps) {
+              int height = std::min(outputHeight - oh, outputHeightSteps);
+
+              int M = outputChannels / groups_;
+              int N = height * outputWidth;
+              int K = channels * filterHeight * filterWidth;
               // im2col
-              im2col(inputData + g * inputOffset,
+              im2col(inputData,
                      imShape,
                      colData,
                      colShape,
@@ -232,13 +239,12 @@ public:
                      paddingW(),
                      dilationH(),
                      dilationW(),
-                     colHeightStart,
-                     K,
-                     colWidthStart,
+                     channels,
+                     oh,
+                     height,
                      N);
 
               // gemm
-              int M = outputChannels / groups_;
               BlasGemm<Device, real>::compute(
                   false,
                   false,
@@ -246,12 +252,12 @@ public:
                   N,
                   K,
                   1.0f,
-                  filterData + g * filterOffset + colHeightStart,
+                  filterData + ic * filterHeight * filterWidth,
                   kStride,
                   colData,
                   N,
                   beta_,
-                  outputData + g * outputOffset + colWidthStart,
+                  outputData + oh * outputWidth,
                   nStride);
             }
             beta_ = 1.0;
@@ -266,17 +272,18 @@ public:
                                           N,
                                           K,
                                           1.0f,
-                                          filterData + g * filterOffset,
+                                          filterData,
                                           K,
-                                          inputData + g * inputOffset,
+                                          inputData,
                                           N,
                                           beta,
-                                          outputData + g * outputOffset,
+                                          outputData,
                                           N);
         }
+        inputData += inputOffset;
+        outputData += outputOffset;
+        filterData += filterOffset;
       }
-      inputData += inputChannels * inputHeight * inputWidth;
-      outputData += outputChannels * outputHeight * outputWidth;
     }
 
     memory_.reset();

paddle/function/Im2Col.h

@@ -111,39 +111,42 @@ public:
                   int paddingWidth,
                   int dilationHeight,
                   int dilationWidth,
-                  int colHeightStart,
-                  int colHeightSize,
-                  int colWidthStart,
-                  int colWidthSize) {
+                  int inputChannels,
+                  int colOffset,
+                  int colOutputHeight,
+                  int colWidth) {
     int inputHeight = imShape[1];
     int inputWidth = imShape[2];
     int filterHeight = colShape[1];
     int filterWidth = colShape[2];
     int outputWidth = colShape[4];
 
-    for (int colh = 0; colh < colHeightSize; colh++) {
-      int wOffset = (colHeightStart + colh) % filterWidth;
-      int hOffset = ((colHeightStart + colh) / filterWidth) % filterHeight;
-      int c_im = (colHeightStart + colh) / filterWidth / filterHeight;
-
-      for (int colw = 0; colw < colWidthSize; colw++) {
-        int h = (colWidthStart + colw) / outputWidth;
-        int w = (colWidthStart + colw) % outputWidth;
-
-        int imRowIdx = h * strideHeight + hOffset * dilationHeight;
-        int imColIdx = w * strideWidth + wOffset * dilationWidth;
-        if ((imRowIdx - paddingHeight) < 0 ||
-            (imRowIdx - paddingHeight) >= inputHeight ||
-            (imColIdx - paddingWidth) < 0 ||
-            (imColIdx - paddingWidth) >= inputWidth) {
-          colData[colh * colWidthSize + colw] = static_cast<T>(0);
-        } else {
-          imRowIdx += c_im * inputHeight - paddingHeight;
-          imColIdx -= paddingWidth;
-          colData[colh * colWidthSize + colw] =
-              imData[imRowIdx * inputWidth + imColIdx];
+    for (int ic = 0; ic < inputChannels; ic++) {
+      for (int oh = 0; oh < colOutputHeight; oh++) {
+        T* dstData = colData + oh * outputWidth;
+        for (int fh = 0; fh < filterHeight; fh++) {
+          for (int fw = 0; fw < filterWidth; fw++) {
+            int imRowIdx = (oh + colOffset) * strideHeight +
+                           fh * dilationHeight - paddingHeight;
+            if (imRowIdx < 0 || imRowIdx >= inputHeight) {
+              memset(dstData, 0, outputWidth * sizeof(T));
+            } else {
+              for (int ow = 0; ow < outputWidth; ow++) {
+                int imColIdx =
+                    ow * strideWidth + fw * dilationWidth - paddingWidth;
+                if (imColIdx < 0 || imColIdx >= inputWidth) {
+                  dstData[ow] = T(0);
+                } else {
+                  dstData[ow] = imData[imRowIdx * inputWidth + imColIdx];
+                }
+              }
+            }
+            dstData += colWidth;
+          }
         }
       }
+      colData += filterHeight * filterWidth * colWidth;
+      imData += inputHeight * inputWidth;
     }
   }
 };

paddle/function/Im2ColTest.cpp

@@ -202,10 +202,10 @@ void TestIm2ColMobileFunctor() {
                           padding,
                           dilation,
                           dilation,
+                          channels,
                           0,
-                          height,
-                          0,
-                          width);
+                          outputHeight,
+                          outputHeight * outputWidth);
 
                   autotest::TensorCheckEqual(*output1, *output2);
                 }

paddle/inference/io.cc

@@ -55,7 +55,7 @@ void LoadPersistables(framework::Executor& executor,
       VLOG(3) << "parameter's name: " << var->Name();
 
       framework::VarDesc* new_var = load_block->Var(var->Name());
-      new_var->SetShape(var->Shape());
+      new_var->SetShape(var->GetShape());
       new_var->SetDataType(var->GetDataType());
       new_var->SetType(var->GetType());
       new_var->SetLoDLevel(var->GetLoDLevel());

paddle/inference/tests/book/CMakeLists.txt

@@ -4,4 +4,4 @@ cc_test(test_inference_recognize_digits_mlp
     DEPS ARCHIVE_START paddle_fluid ARCHIVE_END
     ARGS --dirname=${PYTHON_TESTS_DIR}/book/recognize_digits_mlp.inference.model)
 set_tests_properties(test_inference_recognize_digits_mlp
-    PROPERTIES DEPENDS test_recognize_digits_mlp_cpu)
+    PROPERTIES DEPENDS test_recognize_digits)

paddle/inference/tests/book/test_inference_recognize_digits.cc

@@ -58,6 +58,47 @@ void TestInference(const std::string& dirname,
   delete scope;
 }
 
+template <typename T>
+void SetupTensor(paddle::framework::LoDTensor& input,
+                 paddle::framework::DDim dims,
+                 T lower,
+                 T upper) {
+  srand(time(0));
+  float* input_ptr = input.mutable_data<T>(dims, paddle::platform::CPUPlace());
+  for (int i = 0; i < input.numel(); ++i) {
+    input_ptr[i] =
+        (static_cast<T>(rand()) / static_cast<T>(RAND_MAX)) * (upper - lower) +
+        lower;
+  }
+}
+
+template <typename T>
+void CheckError(paddle::framework::LoDTensor& output1,
+                paddle::framework::LoDTensor& output2) {
+  // Check lod information
+  EXPECT_EQ(output1.lod(), output2.lod());
+
+  EXPECT_EQ(output1.dims(), output2.dims());
+  EXPECT_EQ(output1.numel(), output2.numel());
+
+  T err = static_cast<T>(0);
+  if (typeid(T) == typeid(float)) {
+    err = 1E-3;
+  } else if (typeid(T) == typeid(double)) {
+    err = 1E-6;
+  } else {
+    err = 0;
+  }
+
+  size_t count = 0;
+  for (int64_t i = 0; i < output1.numel(); ++i) {
+    if (fabs(output1.data<T>()[i] - output2.data<T>()[i]) > err) {
+      count++;
+    }
+  }
+  EXPECT_EQ(count, 0) << "There are " << count << " different elements.";
+}
+
 TEST(inference, recognize_digits) {
   if (FLAGS_dirname.empty()) {
     LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model";
@@ -70,12 +111,10 @@ TEST(inference, recognize_digits) {
   // In unittests, this is done in paddle/testing/paddle_gtest_main.cc
 
   paddle::framework::LoDTensor input;
-  srand(time(0));
-  float* input_ptr =
-      input.mutable_data<float>({1, 28, 28}, paddle::platform::CPUPlace());
-  for (int i = 0; i < 784; ++i) {
-    input_ptr[i] = rand() / (static_cast<float>(RAND_MAX));
-  }
+  // Use normilized image pixels as input data,
+  // which should be in the range [-1.0, 1.0].
+  SetupTensor<float>(
+      input, {1, 28, 28}, static_cast<float>(-1), static_cast<float>(1));
   std::vector<paddle::framework::LoDTensor*> cpu_feeds;
   cpu_feeds.push_back(&input);
 
@@ -98,16 +137,6 @@ TEST(inference, recognize_digits) {
       dirname, cpu_feeds, cpu_fetchs2);
   LOG(INFO) << output2.dims();
 
-  EXPECT_EQ(output1.dims(), output2.dims());
-  EXPECT_EQ(output1.numel(), output2.numel());
-
-  float err = 1E-3;
-  int count = 0;
-  for (int64_t i = 0; i < output1.numel(); ++i) {
-    if (fabs(output1.data<float>()[i] - output2.data<float>()[i]) > err) {
-      count++;
-    }
-  }
-  EXPECT_EQ(count, 0) << "There are " << count << " different elements.";
+  CheckError<float>(output1, output2);
 #endif
 }

paddle/math/Matrix.cpp

@@ -2015,13 +2015,6 @@ void CpuMatrix::maxPoolForward(Matrix& inputMat,
     CHECK_EQ(channels * outLength, maskMatP->getWidth());
   }
 
-  /* initialize the data_ */
-  for (size_t i = 0; i < height_; i++) {
-    for (size_t j = 0; j < width_; j++) {
-      outData[i * outStride + j] = -(real)FLT_MAX;
-    }
-  }
-
   /* pool max one by one */
   for (size_t n = 0; n < num; ++n) {  // frame by frame
     if (!isContiguous()) {
@@ -2030,19 +2023,24 @@ void CpuMatrix::maxPoolForward(Matrix& inputMat,
     for (size_t c = 0; c < channels; ++c) {  // channel by channel
       for (size_t ph = 0; ph < outputH; ++ph) {
         int hstart = ph * strideH - paddingH;
-        int hend = std::min(hstart + sizeY, imgSizeH);
-        hstart = std::max(hstart, 0);
+        int hend = hstart + sizeY;
+        hstart = hstart < 0 ? 0 : hstart;
+        hend = hend < (int)imgSizeH ? hend : (int)imgSizeH;
         for (size_t pw = 0; pw < outputW; ++pw) {
           int wstart = pw * strideW - paddingW;
-          int wend = std::min(wstart + sizeX, imgSizeW);
-          wstart = std::max(wstart, 0);
+          int wend = wstart + sizeX;
+          wstart = wstart < 0 ? 0 : wstart;
+          wend = wend < (int)imgSizeW ? wend : (int)imgSizeW;
           if (maskData == NULL) {
+            real tmp = -(real)FLT_MAX;
             for (int h = hstart; h < hend; ++h) {
               for (int w = wstart; w < wend; ++w) {
-                outData[ph * outputW + pw] = std::max(
-                    outData[ph * outputW + pw], inputData[h * imgSizeW + w]);
+                tmp = tmp < inputData[h * imgSizeW + w]
+                          ? inputData[h * imgSizeW + w]
+                          : tmp;
               }
             }
+            outData[ph * outputW + pw] = tmp;
           } else {
             for (int h = hstart; h < hend; ++h) {
               for (int w = wstart; w < wend; ++w) {

paddle/operators/CMakeLists.txt

@@ -158,7 +158,10 @@ op_library(parallel_do_op DEPS executor)
 
 # Regist multiple Kernel to pybind
 if (WITH_GPU)
-op_library(conv_op SRCS conv_op.cc conv_op.cu.cc conv_cudnn_op.cu.cc DEPS vol2col)
+
+op_library(conv_op SRCS conv_op.cc conv_op.cu.cc conv_cudnn_op.cu.cc DEPS
+    vol2col depthwise_conv)
+
 op_library(edit_distance_op SRCS edit_distance_op.cc edit_distance_op.cu DEPS math_function)
 op_library(pool_op SRCS pool_op.cc pool_op.cu.cc pool_cudnn_op.cu.cc DEPS pooling)
 op_library(conv_transpose_op SRCS conv_transpose_op.cc conv_transpose_op.cu.cc

paddle/operators/bipartite_match_op.cc

-/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
+/* 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.
@@ -28,12 +28,18 @@ class BipartiteMatchOp : public framework::OperatorWithKernel {
   void InferShape(framework::InferShapeContext* ctx) const override {
     PADDLE_ENFORCE(ctx->HasInput("DistMat"),
                    "Input(DistMat) of BipartiteMatch should not be null.");
+    PADDLE_ENFORCE(
+        ctx->HasOutput("ColToRowMatchIndices"),
+        "Output(ColToRowMatchIndices) of BipartiteMatch should not be null.");
+    PADDLE_ENFORCE(
+        ctx->HasOutput("ColToRowMatchDist"),
+        "Output(ColToRowMatchDist) of BipartiteMatch should not be null.");
 
     auto dims = ctx->GetInputDim("DistMat");
     PADDLE_ENFORCE_EQ(dims.size(), 2, "The rank of Input(DistMat) must be 2.");
 
     ctx->SetOutputDim("ColToRowMatchIndices", dims);
-    ctx->SetOutputDim("ColToRowMatchDis", dims);
+    ctx->SetOutputDim("ColToRowMatchDist", dims);
   }
 };
 
@@ -91,7 +97,7 @@ class BipartiteMatchKernel : public framework::OpKernel<T> {
   void Compute(const framework::ExecutionContext& context) const override {
     auto* dist_mat = context.Input<LoDTensor>("DistMat");
     auto* match_indices = context.Output<Tensor>("ColToRowMatchIndices");
-    auto* match_dist = context.Output<Tensor>("ColToRowMatchDis");
+    auto* match_dist = context.Output<Tensor>("ColToRowMatchDist");
 
     auto& dev_ctx = context.device_context<platform::CPUDeviceContext>();
 
@@ -148,13 +154,13 @@ class BipartiteMatchOpMaker : public framework::OpProtoAndCheckerMaker {
               "Otherwise, it means B[j] is matched to row "
               "ColToRowMatchIndices[i][j] in i-th instance. The row number of "
               "i-th instance is saved in ColToRowMatchIndices[i][j].");
-    AddOutput("ColToRowMatchDis",
+    AddOutput("ColToRowMatchDist",
               "(Tensor) A 2-D Tensor with shape [N, M] in float type. "
               "N is batch size. If ColToRowMatchIndices[i][j] is -1, "
-              "ColToRowMatchDis[i][j] is also -1.0. Otherwise, assumed "
+              "ColToRowMatchDist[i][j] is also -1.0. Otherwise, assumed "
               "ColToRowMatchIndices[i][j] = d, and the row offsets of each "
               "instance are called LoD. Then "
-              "ColToRowMatchDis[i][j] = DistMat[d+LoD[i]][j]");
+              "ColToRowMatchDist[i][j] = DistMat[d+LoD[i]][j]");
     AddComment(R"DOC(
 This operator is a greedy bipartite matching algorithm, which is used to
 obtain the matching with the maximum distance based on the input

paddle/operators/box_coder_op.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. */
+
+#include "paddle/operators/box_coder_op.h"
+
+namespace paddle {
+namespace operators {
+
+class BoxCoderOp : public framework::OperatorWithKernel {
+ public:
+  using framework::OperatorWithKernel::OperatorWithKernel;
+
+ protected:
+  void InferShape(framework::InferShapeContext *ctx) const override {
+    PADDLE_ENFORCE(ctx->HasInput("PriorBox"),
+                   "Input(PriorBox) of BoxCoderOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasInput("PriorBoxVar"),
+                   "Input(PriorBoxVar) of BoxCoderOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasInput("TargetBox"),
+                   "Input(TargetBox) of BoxCoderOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasOutput("OutputBox"),
+                   "Output(OutputBox) of BoxCoderOp should not be null.");
+
+    auto prior_box_dims = ctx->GetInputDim("PriorBox");
+    auto prior_box_var_dims = ctx->GetInputDim("PriorBoxVar");
+    auto target_box_dims = ctx->GetInputDim("TargetBox");
+
+    PADDLE_ENFORCE_EQ(prior_box_dims.size(), 2,
+                      "The rank of Input of PriorBoxVar must be 2");
+    PADDLE_ENFORCE_EQ(prior_box_dims[1], 4, "The shape of PriorBox is [N, 4]");
+    PADDLE_ENFORCE_EQ(prior_box_dims, prior_box_var_dims);
+    PADDLE_ENFORCE_EQ(target_box_dims.size(), 2,
+                      "The rank of Input of TargetBox must be 2");
+    PADDLE_ENFORCE_EQ(target_box_dims[1], 4,
+                      "The shape of TargetBox is [M, 4]");
+
+    GetBoxCodeType(ctx->Attrs().Get<std::string>("code_type"));
+
+    ctx->SetOutputDim(
+        "OutputBox",
+        framework::make_ddim({target_box_dims[0], prior_box_dims[0], 4}));
+    ctx->ShareLoD("TargetBox", /*->*/ "OutputBox");
+  }
+};
+
+class BoxCoderOpMaker : public framework::OpProtoAndCheckerMaker {
+ public:
+  BoxCoderOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : OpProtoAndCheckerMaker(proto, op_checker) {
+    AddInput(
+        "PriorBox",
+        "(Tensor, default Tensor<float>) "
+        "Box list PriorBox is a 2-D Tensor with shape [M, 4] holds M boxes, "
+        "each box is represented as [xmin, ymin, xmax, ymax], "
+        "[xmin, ymin] is the left top coordinate of the anchor box, "
+        "if the input is image feature map, they are close to the origin "
+        "of the coordinate system. [xmax, ymax] is the right bottom "
+        "coordinate of the anchor box.");
+    AddInput("PriorBoxVar",
+             "(Tensor, default Tensor<float>) "
+             "PriorBoxVar is a 2-D Tensor with shape [M, 4] holds M group "
+             "of variance.");
+    AddInput(
+        "TargetBox",
+        "(LoDTensor or Tensor) this input is a 2-D LoDTensor with shape "
+        "[N, 4], each box is represented as [xmin, ymin, xmax, ymax], "
+        "[xmin, ymin] is the left top coordinate of the box if the input "
+        "is image feature map, they are close to the origin of the coordinate "
+        "system. [xmax, ymax] is the right bottom coordinate of the box. "
+        "This tensor can contain LoD information to represent a batch "
+        "of inputs. One instance of this batch can contain different "
+        "numbers of entities.");
+    AddAttr<std::string>("code_type",
+                         "(string, default encode_center_size) "
+                         "the code type used with the target box")
+        .SetDefault("encode_center_size")
+        .InEnum({"encode_center_size", "decode_center_size"});
+    AddOutput(
+        "OutputBox",
+        "(LoDTensor or Tensor) "
+        "(Tensor) The output of box_coder_op, a tensor with shape [N, M, 4] "
+        "representing the result of N target boxes encoded/decoded with "
+        "M Prior boxes and variances.");
+
+    AddComment(R"DOC(
+Bounding Box Coder Operator.
+Encode/Decode the target bounding box with the priorbox information.
+The Encoding schema described below:
+ox = (tx - px) / pw / pxv
+oy = (ty - py) / ph / pyv
+ow = log(abs(tw / pw)) / pwv 
+oh = log(abs(th / ph)) / phv 
+The Decoding schema described below:
+ox = (pw * pxv * tx * + px) - tw / 2
+oy = (ph * pyv * ty * + py) - th / 2
+ow = exp(pwv * tw) * pw + tw / 2
+oh = exp(phv * th) * ph + th / 2
+where tx, ty, tw, th denote the target box's center coordinates, width and
+height respectively. Similarly, px, py, pw, ph denote the priorbox's(anchor)
+center coordinates, width and height. pxv, pyv, pwv, phv denote the variance
+of the priorbox and ox, oy, ow, oh denote the encoded/decoded coordinates,
+width and height.
+)DOC");
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP_WITHOUT_GRADIENT(box_coder, ops::BoxCoderOp, ops::BoxCoderOpMaker);
+REGISTER_OP_CPU_KERNEL(box_coder, ops::BoxCoderKernel<float>,
+                       ops::BoxCoderKernel<double>);

paddle/operators/box_coder_op.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 "paddle/operators/box_coder_op.h"
+#include "paddle/platform/cuda_helper.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename T>
+__global__ void EncodeCenterSizeKernel(const T* prior_box_data,
+                                       const T* prior_box_var_data,
+                                       const T* target_box_data, const int row,
+                                       const int col, const int len,
+                                       T* output) {
+  const int idx = threadIdx.x + blockIdx.x * blockDim.x;
+  if (idx < row * col) {
+    const int row_idx = idx / col;
+    const int col_idx = idx % col;
+    T prior_box_width =
+        prior_box_data[col_idx * len + 2] - prior_box_data[col_idx * len];
+    T prior_box_height =
+        prior_box_data[col_idx * len + 3] - prior_box_data[col_idx * len + 1];
+    T prior_box_center_x =
+        (prior_box_data[col_idx * len + 2] + prior_box_data[col_idx * len]) / 2;
+    T prior_box_center_y = (prior_box_data[col_idx * len + 3] +
+                            prior_box_data[col_idx * len + 1]) /
+                           2;
+
+    T target_box_center_x =
+        (target_box_data[row_idx * len + 2] + target_box_data[row_idx * len]) /
+        2;
+    T target_box_center_y = (target_box_data[row_idx * len + 3] +
+                             target_box_data[row_idx * len + 1]) /
+                            2;
+    T target_box_width =
+        target_box_data[row_idx * len + 2] - target_box_data[row_idx * len];
+    T target_box_height =
+        target_box_data[row_idx * len + 3] - target_box_data[row_idx * len + 1];
+
+    output[idx * len] = (target_box_center_x - prior_box_center_x) /
+                        prior_box_width / prior_box_var_data[col_idx * len];
+    output[idx * len + 1] = (target_box_center_y - prior_box_center_y) /
+                            prior_box_height /
+                            prior_box_var_data[col_idx * len + 1];
+    output[idx * len + 2] = log(fabs(target_box_width / prior_box_width)) /
+                            prior_box_var_data[col_idx * len + 2];
+    output[idx * len + 3] = log(fabs(target_box_height / prior_box_height)) /
+                            prior_box_var_data[col_idx * len + 3];
+  }
+}
+
+template <typename T>
+__global__ void DecodeCenterSizeKernel(const T* prior_box_data,
+                                       const T* prior_box_var_data,
+                                       const T* target_box_data, const int row,
+                                       const int col, const int len,
+                                       T* output) {
+  const int idx = threadIdx.x + blockIdx.x * blockDim.x;
+  if (idx < row * col) {
+    const int row_idx = idx / col;
+    const int col_idx = idx % col;
+    T prior_box_width =
+        prior_box_data[col_idx * len + 2] - prior_box_data[col_idx * len];
+    T prior_box_height =
+        prior_box_data[col_idx * len + 3] - prior_box_data[col_idx * len + 1];
+    T prior_box_center_x =
+        (prior_box_data[col_idx * len + 2] + prior_box_data[col_idx * len]) / 2;
+    T prior_box_center_y = (prior_box_data[col_idx * len + 3] +
+                            prior_box_data[col_idx * len + 1]) /
+                           2;
+
+    T target_box_width = exp(prior_box_var_data[col_idx * len + 2] *
+                             target_box_data[row_idx * len + 2]) *
+                         prior_box_width;
+    T target_box_height = exp(prior_box_var_data[col_idx * len + 3] *
+                              target_box_data[row_idx * len + 3]) *
+                          prior_box_height;
+    T target_box_center_x = prior_box_var_data[col_idx * len] *
+                                target_box_data[row_idx * len] *
+                                prior_box_width +
+                            prior_box_center_x;
+    T target_box_center_y = prior_box_var_data[col_idx * len + 1] *
+                                target_box_data[row_idx * len + 1] *
+                                prior_box_height +
+                            prior_box_center_y;
+
+    output[idx * len] = target_box_center_x - target_box_width / 2;
+    output[idx * len + 1] = target_box_center_y - target_box_height / 2;
+    output[idx * len + 2] = target_box_center_x + target_box_width / 2;
+    output[idx * len + 3] = target_box_center_y + target_box_height / 2;
+  }
+}
+
+template <typename T>
+class BoxCoderCUDAKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& context) const override {
+    PADDLE_ENFORCE(platform::is_gpu_place(context.GetPlace()),
+                   "This kernel only runs on GPU device.");
+    auto* prior_box = context.Input<framework::Tensor>("PriorBox");
+    auto* prior_box_var = context.Input<framework::Tensor>("PriorBoxVar");
+    auto* target_box = context.Input<framework::LoDTensor>("TargetBox");
+    auto* output_box = context.Output<framework::Tensor>("OutputBox");
+
+    if (target_box->lod().size()) {
+      PADDLE_ENFORCE_EQ(target_box->lod().size(), 1,
+                        "Only support 1 level of LoD.");
+    }
+    auto row = target_box->dims()[0];
+    auto col = prior_box->dims()[0];
+    auto len = prior_box->dims()[1];
+    int block = 512;
+    int grid = (row * col + block - 1) / block;
+    auto& device_ctx = context.cuda_device_context();
+
+    const T* prior_box_data = prior_box->data<T>();
+    const T* prior_box_var_data = prior_box_var->data<T>();
+    const T* target_box_data = target_box->data<T>();
+
+    output_box->mutable_data<T>({row, col, len}, context.GetPlace());
+    T* output = output_box->data<T>();
+
+    auto code_type = GetBoxCodeType(context.Attr<std::string>("code_type"));
+    if (code_type == BoxCodeType::kEncodeCenterSize) {
+      EncodeCenterSizeKernel<T><<<grid, block, 0, device_ctx.stream()>>>(
+          prior_box_data, prior_box_var_data, target_box_data, row, col, len,
+          output);
+    } else if (code_type == BoxCodeType::kDecodeCenterSize) {
+      DecodeCenterSizeKernel<T><<<grid, block, 0, device_ctx.stream()>>>(
+          prior_box_data, prior_box_var_data, target_box_data, row, col, len,
+          output);
+    }
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP_CUDA_KERNEL(box_coder, ops::BoxCoderCUDAKernel<float>,
+                        ops::BoxCoderCUDAKernel<double>);

paddle/operators/box_coder_op.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 "paddle/framework/op_registry.h"
+#include "paddle/operators/math/math_function.h"
+
+namespace paddle {
+namespace operators {
+
+enum class BoxCodeType { kEncodeCenterSize = 0, kDecodeCenterSize = 1 };
+
+inline BoxCodeType GetBoxCodeType(const std::string& type) {
+  if (type == "encode_center_size") {
+    return BoxCodeType::kEncodeCenterSize;
+  } else if (type == "decode_center_size") {
+    return BoxCodeType::kDecodeCenterSize;
+  }
+  PADDLE_THROW("Not support type %s.", type);
+}
+
+template <typename T>
+class BoxCoderKernel : public framework::OpKernel<T> {
+ public:
+  void EncodeCenterSize(const framework::Tensor& target_box,
+                        const framework::Tensor& prior_box,
+                        const framework::Tensor& prior_box_var,
+                        T* output) const {
+    int64_t row = target_box.dims()[0];
+    int64_t col = prior_box.dims()[0];
+    int64_t len = prior_box.dims()[1];
+    auto* target_box_data = target_box.data<T>();
+    auto* prior_box_data = prior_box.data<T>();
+    auto* prior_box_var_data = prior_box_var.data<T>();
+
+    for (int64_t i = 0; i < row; ++i) {
+      for (int64_t j = 0; j < col; ++j) {
+        T prior_box_width =
+            prior_box_data[j * len + 2] - prior_box_data[j * len];
+        T prior_box_height =
+            prior_box_data[j * len + 3] - prior_box_data[j * len + 1];
+        T prior_box_center_x =
+            (prior_box_data[j * len + 2] + prior_box_data[j * len]) / 2;
+        T prior_box_center_y =
+            (prior_box_data[j * len + 3] + prior_box_data[j * len + 1]) / 2;
+
+        T target_box_center_x =
+            (target_box_data[i * len + 2] + target_box_data[i * len]) / 2;
+        T target_box_center_y =
+            (target_box_data[i * len + 3] + target_box_data[i * len + 1]) / 2;
+        T target_box_width =
+            target_box_data[i * len + 2] - target_box_data[i * len];
+        T target_box_height =
+            target_box_data[i * len + 3] - target_box_data[i * len + 1];
+
+        size_t offset = i * col * len + j * len;
+        output[offset] = (target_box_center_x - prior_box_center_x) /
+                         prior_box_width / prior_box_var_data[j * len];
+        output[offset + 1] = (target_box_center_y - prior_box_center_y) /
+                             prior_box_height / prior_box_var_data[j * len + 1];
+        output[offset + 2] =
+            std::log(std::fabs(target_box_width / prior_box_width)) /
+            prior_box_var_data[j * len + 2];
+        output[offset + 3] =
+            std::log(std::fabs(target_box_height / prior_box_height)) /
+            prior_box_var_data[j * len + 3];
+      }
+    }
+  }
+  void DecodeCenterSize(const framework::Tensor& target_box,
+                        const framework::Tensor& prior_box,
+                        const framework::Tensor& prior_box_var,
+                        T* output) const {
+    int64_t row = target_box.dims()[0];
+    int64_t col = prior_box.dims()[0];
+    int64_t len = prior_box.dims()[1];
+
+    auto* target_box_data = target_box.data<T>();
+    auto* prior_box_data = prior_box.data<T>();
+    auto* prior_box_var_data = prior_box_var.data<T>();
+
+    for (int64_t i = 0; i < row; ++i) {
+      for (int64_t j = 0; j < col; ++j) {
+        T prior_box_width =
+            prior_box_data[j * len + 2] - prior_box_data[j * len];
+        T prior_box_height =
+            prior_box_data[j * len + 3] - prior_box_data[j * len + 1];
+        T prior_box_center_x =
+            (prior_box_data[j * len + 2] + prior_box_data[j * len]) / 2;
+        T prior_box_center_y =
+            (prior_box_data[j * len + 3] + prior_box_data[j * len + 1]) / 2;
+
+        T target_box_center_x = prior_box_var_data[j * len] *
+                                    target_box_data[i * len] * prior_box_width +
+                                prior_box_center_x;
+        T target_box_center_y = prior_box_var_data[j * len + 1] *
+                                    target_box_data[i * len + 1] *
+                                    prior_box_height +
+                                prior_box_center_y;
+        T target_box_width = std::exp(prior_box_var_data[j * len + 2] *
+                                      target_box_data[i * len + 2]) *
+                             prior_box_width;
+        T target_box_height = std::exp(prior_box_var_data[j * len + 3] *
+                                       target_box_data[i * len + 3]) *
+                              prior_box_height;
+
+        size_t offset = i * col * len + j * len;
+        output[offset] = target_box_center_x - target_box_width / 2;
+        output[offset + 1] = target_box_center_y - target_box_height / 2;
+        output[offset + 2] = target_box_center_x + target_box_width / 2;
+        output[offset + 3] = target_box_center_y + target_box_height / 2;
+      }
+    }
+  }
+
+  void Compute(const framework::ExecutionContext& context) const override {
+    auto* prior_box = context.Input<framework::Tensor>("PriorBox");
+    auto* prior_box_var = context.Input<framework::Tensor>("PriorBoxVar");
+    auto* target_box = context.Input<framework::LoDTensor>("TargetBox");
+    auto* output_box = context.Output<framework::Tensor>("OutputBox");
+
+    if (target_box->lod().size()) {
+      PADDLE_ENFORCE_EQ(target_box->lod().size(), 1UL,
+                        "Only support 1 level of LoD.");
+    }
+    auto row = target_box->dims()[0];
+    auto col = prior_box->dims()[0];
+    auto len = prior_box->dims()[1];
+
+    output_box->mutable_data<T>({row, col, len}, context.GetPlace());
+
+    auto code_type = GetBoxCodeType(context.Attr<std::string>("code_type"));
+    T* output = output_box->data<T>();
+    if (code_type == BoxCodeType::kEncodeCenterSize) {
+      EncodeCenterSize(*target_box, *prior_box, *prior_box_var, output);
+    } else if (code_type == BoxCodeType::kDecodeCenterSize) {
+      DecodeCenterSize(*target_box, *prior_box, *prior_box_var, output);
+    }
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle

paddle/operators/compare_op.h

@@ -54,7 +54,15 @@ class CompareOpKernel
  public:
   void Compute(const framework::ExecutionContext& context) const override {
     using T = typename Functor::ELEM_TYPE;
-    ElementwiseComputeEx<Functor, DeviceContext, T, bool>(context);
+    using Tensor = framework::Tensor;
+
+    auto* x = context.Input<Tensor>("X");
+    auto* y = context.Input<Tensor>("Y");
+    auto* z = context.Output<Tensor>("Out");
+    z->mutable_data<T>(context.GetPlace());
+    int axis = context.Attr<int>("axis");
+    ElementwiseComputeEx<Functor, DeviceContext, T, bool>(context, x, y, axis,
+                                                          z);
   }
 };
 

paddle/operators/conv_op.cc

@@ -318,9 +318,25 @@ framework::OpKernelType ConvOpGrad::GetExpectedKernelType(
 namespace ops = paddle::operators;
 REGISTER_OP(conv2d, ops::ConvOp, ops::Conv2DOpMaker, conv2d_grad,
             ops::ConvOpGrad);
+
+// depthwise convolution op
+REGISTER_OP(depthwise_conv2d, ops::ConvOp, ops::Conv2DOpMaker,
+            depthwise_conv2d_grad, ops::ConvOpGrad);
 REGISTER_OP(conv3d, ops::ConvOp, ops::Conv3DOpMaker, conv3d_grad,
             ops::ConvOpGrad);
 
+// depthwise conv kernel
+// TODO(xingzhaolong): neon kernel for mobile
+REGISTER_OP_CPU_KERNEL(
+    depthwise_conv2d,
+    ops::GemmConvKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::GemmConvKernel<paddle::platform::CPUDeviceContext, double>);
+
+REGISTER_OP_CPU_KERNEL(
+    depthwise_conv2d_grad,
+    ops::GemmConvGradKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::GemmConvGradKernel<paddle::platform::CPUDeviceContext, double>);
+
 REGISTER_OP_CPU_KERNEL(
     conv2d, ops::GemmConvKernel<paddle::platform::CPUDeviceContext, float>,
     ops::GemmConvKernel<paddle::platform::CPUDeviceContext, double>);

paddle/operators/conv_op.cu.cc

@@ -16,6 +16,16 @@ limitations under the License. */
 
 namespace ops = paddle::operators;
 
+REGISTER_OP_CUDA_KERNEL(
+    depthwise_conv2d,
+    ops::DepthwiseConvKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::DepthwiseConvKernel<paddle::platform::CUDADeviceContext, double>);
+
+REGISTER_OP_CUDA_KERNEL(
+    depthwise_conv2d_grad,
+    ops::DepthwiseConvGradKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::DepthwiseConvGradKernel<paddle::platform::CUDADeviceContext, double>);
+
 REGISTER_OP_CUDA_KERNEL(
     conv2d, ops::GemmConvKernel<paddle::platform::CUDADeviceContext, float>,
     ops::GemmConvKernel<paddle::platform::CUDADeviceContext, double>);

paddle/operators/conv_op.h

@@ -16,6 +16,7 @@ limitations under the License. */
 
 #include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
+#include "paddle/operators/math/depthwise_conv.h"
 #include "paddle/operators/math/im2col.h"
 #include "paddle/operators/math/math_function.h"
 #include "paddle/operators/math/vol2col.h"
@@ -350,5 +351,72 @@ class GemmConvGradKernel : public framework::OpKernel<T> {
     }
   }
 };
+
+template <typename DeviceContext, typename T>
+class DepthwiseConvKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& context) const override {
+    const Tensor* input = context.Input<Tensor>("Input");
+    Tensor filter = *context.Input<Tensor>("Filter");
+    Tensor* output = context.Output<Tensor>("Output");
+    output->mutable_data<T>(context.GetPlace());
+
+    PADDLE_ENFORCE_EQ(
+        output->dims()[1] % input->dims()[1], 0,
+        "The output channels must be a multiple of the input channels");
+    std::vector<int> strides = context.Attr<std::vector<int>>("strides");
+    std::vector<int> paddings = context.Attr<std::vector<int>>("paddings");
+    std::vector<int> dilations = context.Attr<std::vector<int>>("dilations");
+
+    math::DepthwiseConvFunctor<DeviceContext, T> depthwiseConv;
+
+    auto& dev_ctx = context.template device_context<DeviceContext>();
+    depthwiseConv(dev_ctx, *input, filter, strides, paddings, output);
+  }
+};
+
+template <typename DeviceContext, typename T>
+class DepthwiseConvGradKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& context) const override {
+    const Tensor* input = context.Input<Tensor>("Input");
+    const Tensor* output_grad =
+        context.Input<Tensor>(framework::GradVarName("Output"));
+    Tensor* input_grad =
+        context.Output<Tensor>(framework::GradVarName("Input"));
+    Tensor* filter_grad =
+        context.Output<Tensor>(framework::GradVarName("Filter"));
+    Tensor filter = *context.Input<Tensor>("Filter");
+
+    if (!input_grad && !filter_grad) return;
+
+    std::vector<int> strides = context.Attr<std::vector<int>>("strides");
+    std::vector<int> paddings = context.Attr<std::vector<int>>("paddings");
+    std::vector<int> dilations = context.Attr<std::vector<int>>("dilations");
+
+    math::SetConstant<DeviceContext, T> set_zero;
+    auto& dev_ctx = context.template device_context<DeviceContext>();
+
+    math::DepthwiseConvInputGradFunctor<DeviceContext, T>
+        depthwiseConvInputGrad;
+    math::DepthwiseConvFilterGradFunctor<DeviceContext, T>
+        depthwiseConvFilterGrad;
+
+    if (input_grad) {
+      input_grad->mutable_data<T>(context.GetPlace());
+      set_zero(dev_ctx, input_grad, static_cast<T>(0));
+      depthwiseConvInputGrad(dev_ctx, *input, filter, *output_grad, strides,
+                             paddings, input_grad);
+    }
+
+    if (filter_grad) {
+      filter_grad->mutable_data<T>(context.GetPlace());
+      set_zero(dev_ctx, filter_grad, static_cast<T>(0));
+      depthwiseConvFilterGrad(dev_ctx, *input, *output_grad, strides, paddings,
+                              filter_grad);
+    }
+  }
+};
+
 }  // namespace operators
 }  // namespace paddle

paddle/operators/elementwise_add_op.h

@@ -28,7 +28,14 @@ template <typename DeviceContext, typename T>
 class ElementwiseAddKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 
@@ -92,9 +99,19 @@ template <typename DeviceContext, typename T>
 class ElementwiseAddGradKernel : 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* out = ctx.Input<Tensor>("Out");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
     ElementwiseGradCompute<DeviceContext, T, ElementwiseAddGradFunctor<T>,
                            ElementwiseAddBroadCastGradFunctor<T>,
-                           ElementwiseAddBroadCast2GradFunctor<T>>(ctx);
+                           ElementwiseAddBroadCast2GradFunctor<T>>(
+        ctx, x, y, out, dout, axis, dx, dy);
   }
 };
 

paddle/operators/elementwise_div_op.h

@@ -28,7 +28,14 @@ template <typename DeviceContext, typename T>
 class ElementwiseDivKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseComputeEx<DivFunctor<T>, DeviceContext, T>(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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<DivFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 
@@ -111,9 +118,19 @@ template <typename DeviceContext, typename T>
 class ElementwiseDivGradKernel : 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* out = ctx.Input<Tensor>("Out");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
     ElementwiseGradCompute<DeviceContext, T, ElementwiseDivGradFunctor<T>,
                            ElementwiseDivBroadCastGradFunctor<T>,
-                           ElementwiseDivBroadCast2GradFunctor<T>>(ctx);
+                           ElementwiseDivBroadCast2GradFunctor<T>>(
+        ctx, x, y, out, dout, axis, dx, dy);
   }
 };
 

paddle/operators/elementwise_max_op.h

@@ -28,7 +28,14 @@ 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);
+    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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<MaxFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 
@@ -110,9 +117,19 @@ template <typename DeviceContext, typename T>
 class ElementwiseMaxGradKernel : 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* out = ctx.Input<Tensor>("Out");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
     ElementwiseGradCompute<DeviceContext, T, ElementwiseMaxGradFunctor<T>,
                            ElementwiseMaxBroadCastGradFunctor<T>,
-                           ElementwiseMaxBroadCast2GradFunctor<T>>(ctx);
+                           ElementwiseMaxBroadCast2GradFunctor<T>>(
+        ctx, x, y, out, dout, axis, dx, dy);
   }
 };
 

paddle/operators/elementwise_min_op.h

@@ -28,7 +28,14 @@ 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);
+    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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<MinFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 
@@ -110,9 +117,19 @@ template <typename DeviceContext, typename T>
 class ElementwiseMinGradKernel : 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* out = ctx.Input<Tensor>("Out");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
     ElementwiseGradCompute<DeviceContext, T, ElementwiseMinGradFunctor<T>,
                            ElementwiseMinBroadCastGradFunctor<T>,
-                           ElementwiseMinBroadCast2GradFunctor<T>>(ctx);
+                           ElementwiseMinBroadCast2GradFunctor<T>>(
+        ctx, x, y, out, dout, axis, dx, dy);
   }
 };
 

paddle/operators/elementwise_mul_op.h

@@ -27,7 +27,14 @@ template <typename DeviceContext, typename T>
 class ElementwiseMulKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseComputeEx<MulFunctor<T>, DeviceContext, T>(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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<MulFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 
@@ -110,9 +117,19 @@ template <typename DeviceContext, typename T>
 class ElementwiseMulGradKernel : 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* out = ctx.Input<Tensor>("Out");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
     ElementwiseGradCompute<DeviceContext, T, ElementwiseMulGradFunctor<T>,
                            ElementwiseMulBroadCastGradFunctor<T>,
-                           ElementwiseMulBroadCast2GradFunctor<T>>(ctx);
+                           ElementwiseMulBroadCast2GradFunctor<T>>(
+        ctx, x, y, out, dout, axis, dx, dy);
   }
 };
 

paddle/operators/elementwise_op_function.h

@@ -313,21 +313,18 @@ EIGEN_FUNCTOR(Div, EIGEN_DIV);
 
 template <typename DeviceContext, typename T, typename functor,
           typename broadcastfunctor, typename broadcast2functor>
-void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
-  using Tensor = framework::Tensor;
-
-  auto* x = ctx.Input<Tensor>("X");
-  auto* y = ctx.Input<Tensor>("Y");
-  auto* out = ctx.Input<Tensor>("Out");
-  auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+void ElementwiseGradCompute(const framework::ExecutionContext& ctx,
 
+                            const framework::Tensor* x,
+                            const framework::Tensor* y,
+                            const framework::Tensor* out,
+                            const framework::Tensor* dout, int axis,
+                            framework::Tensor* dx, framework::Tensor* dy) {
   auto& place = *ctx.template device_context<DeviceContext>().eigen_device();
 
   auto x_dims = x->dims();
   auto y_dims = y->dims();
 
-  auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
-  auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
   if (dx) {
     dx->mutable_data<T>(ctx.GetPlace());
   }
@@ -348,7 +345,6 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
     x_dims = framework::make_ddim(extended_dims);
   }
 
-  int axis = ctx.Attr<int>("axis");
   axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis);
 
   int pre, n, post;
@@ -367,13 +363,10 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) {
 
 template <typename Functor, typename DeviceContext, typename T,
           typename OutType = 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<OutType>(ctx.GetPlace());
+void ElementwiseComputeEx(const framework::ExecutionContext& ctx,
+                          const framework::Tensor* x,
+                          const framework::Tensor* y, int axis,
+                          framework::Tensor* z) {
   TransformFunctor<Functor, T, DeviceContext, OutType> functor(
       x, y, z, ctx.template device_context<DeviceContext>(), Functor());
 
@@ -394,7 +387,6 @@ void ElementwiseComputeEx(const framework::ExecutionContext& ctx) {
     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)");

paddle/operators/elementwise_pow_op.h

@@ -29,7 +29,14 @@ template <typename DeviceContext, typename T>
 class ElementwisePowKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseComputeEx<PowFunctor<T>, DeviceContext, T>(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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<PowFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 

paddle/operators/elementwise_sub_op.h

@@ -27,7 +27,14 @@ template <typename DeviceContext, typename T>
 class ElementwiseSubKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    ElementwiseComputeEx<SubFunctor<T>, DeviceContext, T>(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());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<SubFunctor<T>, DeviceContext, T>(ctx, x, y, axis, z);
   }
 };
 
@@ -93,9 +100,19 @@ template <typename DeviceContext, typename T>
 class ElementwiseSubGradKernel : 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* out = ctx.Input<Tensor>("Out");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
     ElementwiseGradCompute<DeviceContext, T, ElementwiseSubGradFunctor<T>,
                            ElementwiseSubBroadCastGradFunctor<T>,
-                           ElementwiseSubBroadCast2GradFunctor<T>>(ctx);
+                           ElementwiseSubBroadCast2GradFunctor<T>>(
+        ctx, x, y, out, dout, axis, dx, dy);
   }
 };
 

paddle/operators/math/CMakeLists.txt

@@ -8,6 +8,7 @@ if(WITH_GPU)
     nv_library(softmax SRCS softmax.cc softmax.cu DEPS device_context)
     nv_library(cross_entropy SRCS cross_entropy.cc cross_entropy.cu DEPS device_context)
     nv_library(pooling SRCS pooling.cc pooling.cu DEPS device_context)
+    nv_library(depthwise_conv SRCS depthwise_conv.cu DEPS device_context)
     nv_library(sequence_pooling SRCS sequence_pooling.cc sequence_pooling.cu DEPS device_context math_function)
     nv_library(vol2col SRCS vol2col.cc vol2col.cu DEPS device_context tensor)
     nv_library(context_project SRCS context_project.cc context_project.cu DEPS device_context math_function)

paddle/operators/math/depthwise_conv.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. */
+
+#include "paddle/operators/math/depthwise_conv.h"
+#include "paddle/platform/cuda_helper.h"
+
+namespace paddle {
+namespace operators {
+namespace math {
+
+// A Cuda kernel to compute the depthwise convolution forward pass
+// in NCHW format.
+template <typename T>
+__global__ void KernelDepthwiseConv(
+    const int nthreads, const T* const input_data, const T* const filter_data,
+    const int batch_size, const int output_channels, const int output_height,
+    const int output_width, const int input_channels, const int input_height,
+    const int input_width, const int filter_multiplier, const int filter_height,
+    const int filter_width, const int stride_height, const int stride_width,
+    const int padding_height, const int padding_width, T* const output_data) {
+  int index = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
+
+  if (index < nthreads) {
+    const int batch = index / output_channels / output_height / output_width;
+    const int c_out = (index / output_height / output_width) % output_channels;
+    const int h_out = (index / output_width) % output_height;
+    const int w_out = index % output_width;
+
+    const int c_in = c_out / filter_multiplier;
+    const T* weight = filter_data + c_out * filter_height * filter_width;
+    T value = 0;
+    const int h_in_start = -padding_height + h_out * stride_height;
+    const int w_in_start = -padding_width + w_out * stride_width;
+    const int h_in_end = h_in_start + filter_height;
+    const int w_in_end = w_in_start + filter_width;
+
+    const int in_offset =
+        ((batch * input_channels + c_in) * input_height) * input_width;
+
+    const int h_end = h_in_end < input_height ? h_in_end : input_height;
+    const int w_end = w_in_end < input_width ? w_in_end : input_width;
+    const int h_start = h_in_start > 0 ? h_in_start : 0;
+    const int w_start = w_in_start > 0 ? w_in_start : 0;
+
+    for (int h_in = h_start; h_in < h_end; h_in++) {
+      for (int w_in = w_start; w_in < w_end; w_in++) {
+        const int offset = in_offset + h_in * input_width + w_in;
+        value +=
+            weight[(h_in - h_in_start) * filter_width + (w_in - w_in_start)] *
+            input_data[offset];
+      }
+    }
+    output_data[index] = value;
+  }
+}
+
+// CUDA kernel to compute the depthwise convolution backprop w.r.t input.
+template <typename T>
+__global__ void KernelDepthwiseConvInputGrad(
+    const int nthreads, const T* const output_grad_data,
+    const T* const filter_data, const int batch_size, const int output_channels,
+    const int output_height, const int output_width, const int input_channels,
+    const int input_height, const int input_width, const int filter_multiplier,
+    const int filter_height, const int filter_width, const int stride_height,
+    const int stride_width, const int padding_height, const int padding_width,
+    T* const input_grad_data) {
+  int index = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
+  if (index < nthreads) {
+    const int batch = index / input_channels / input_height / input_width;
+    const int c_in = (index / input_height / input_width) % input_channels;
+    const int h_in = (index / input_width) % input_height;
+    const int w_in = index % input_width;
+
+    const int c_out_start = c_in * filter_multiplier;
+
+    int h_out_start =
+        (h_in - filter_height + padding_height + stride_height) / stride_height;
+    h_out_start = 0 > h_out_start ? 0 : h_out_start;
+
+    int h_out_end = (h_in + padding_height) / stride_height;
+    h_out_end = output_height - 1 < h_out_end ? output_height - 1 : h_out_end;
+
+    int w_out_start =
+        (w_in - filter_width + padding_width + stride_width) / stride_width;
+    w_out_start = 0 > w_out_start ? 0 : w_out_start;
+
+    int w_out_end = (w_in + padding_width) / stride_width;
+    w_out_end = output_width - 1 < w_out_end ? output_width - 1 : w_out_end;
+
+    T value = 0;
+
+    for (int c_out = c_out_start; c_out < c_out_start + filter_multiplier;
+         c_out++) {
+      for (int h_out = h_out_start; h_out <= h_out_end; ++h_out) {
+        const int filter_h = h_in + padding_height - h_out * stride_height;
+        for (int w_out = w_out_start; w_out <= w_out_end; ++w_out) {
+          const int filter_w = w_in + padding_width - w_out * stride_width;
+          const int filter_offset = c_out * filter_height * filter_width +
+                                    filter_h * filter_width + filter_w;
+          const int output_grad_offset =
+              ((batch * output_channels + c_out) * output_height + h_out) *
+                  output_width +
+              w_out;
+          value +=
+              output_grad_data[output_grad_offset] * filter_data[filter_offset];
+        }
+      }
+    }
+    input_grad_data[index] += value;
+  }
+}
+
+// Cuda kernel to compute the depthwise convolution backprop w.r.t. filter.
+template <typename T>
+__global__ void KernelDepthwiseConvFilterGrad(
+    const int nthreads, const T* const output_grad_data,
+    const T* const input_data, const int num, const int output_channels,
+    const int output_height, const int output_width, const int input_channels,
+    const int input_height, const int input_width, const int filter_multiplier,
+    const int filter_height, const int filter_width, const int stride_height,
+    const int stride_width, const int padding_height, const int padding_width,
+    T* const filter_grad_data) {
+  int index = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
+  if (index < nthreads) {
+    const int w_out = index % output_width;
+    const int h_out = (index / output_width) % output_height;
+    const int c_out = (index / output_width / output_height) % output_channels;
+    const int batch = (index / output_width / output_height / output_channels);
+    const int c_in = c_out / filter_multiplier;
+    const int h_in_start = -padding_height + h_out * stride_height;
+    const int w_in_start = -padding_width + w_out * stride_width;
+    const int h_in_end =
+        -padding_height + h_out * stride_height + filter_height;
+    const int w_in_end = -padding_width + w_out * stride_width + filter_width;
+    const int in_offset =
+        (batch * input_channels + c_in) * input_height * input_width;
+
+    T* addr_offset = filter_grad_data + c_out * filter_height * filter_width;
+    const int h_end = h_in_end < input_height ? h_in_end : input_height;
+    const int w_end = w_in_end < input_width ? w_in_end : input_width;
+    const int h_start = h_in_start > 0 ? h_in_start : 0;
+    const int w_start = w_in_start > 0 ? w_in_start : 0;
+
+    for (int h_in = h_start; h_in < h_end; h_in++) {
+      for (int w_in = w_start; w_in < w_end; w_in++) {
+        const int offset = in_offset + h_in * input_width + w_in;
+        const T diff_temp = output_grad_data[index] * input_data[offset];
+        T* addr = addr_offset + (h_in - h_in_start) * filter_width +
+                  (w_in - w_in_start);
+        paddle::platform::CudaAtomicAdd(addr, diff_temp);
+      }
+    }
+  }
+}
+
+/*
+ * All tensors are in NCHW format.
+ * Ksize, strides, paddings are two elements. These two elements represent
+ * height and width, respectively.
+ */
+template <class T>
+class DepthwiseConvFunctor<platform::CUDADeviceContext, T> {
+ public:
+  void operator()(const platform::CUDADeviceContext& context,
+                  const framework::Tensor& input,
+                  const framework::Tensor& filter,
+                  const std::vector<int>& strides,
+                  const std::vector<int>& paddings, framework::Tensor* output) {
+    const int batch_size = input.dims()[0];
+    const int input_channels = input.dims()[1];
+    const int input_height = input.dims()[2];
+    const int input_width = input.dims()[3];
+    const int output_channels = output->dims()[1];
+    const int output_height = output->dims()[2];
+    const int output_width = output->dims()[3];
+    const int ksize_height = filter.dims()[2];
+    const int ksize_width = filter.dims()[3];
+    const int stride_height = strides[0];
+    const int stride_width = strides[1];
+    const int padding_height = paddings[0];
+    const int padding_width = paddings[1];
+
+    const T* input_data = input.data<T>();
+    const T* filter_data = filter.data<T>();
+    T* output_data = output->mutable_data<T>(context.GetPlace());
+
+    int nthreads = batch_size * output_channels * output_height * output_width;
+    int blocks = (nthreads + 1024 - 1) / 1024;
+    dim3 threads(1024, 1);
+    dim3 grid(blocks, 1);
+
+    KernelDepthwiseConv<T><<<grid, threads, 0, context.stream()>>>(
+        nthreads, input_data, filter_data, batch_size, output_channels,
+        output_height, output_width, input_channels, input_height, input_width,
+        output_channels / input_channels, ksize_height, ksize_width,
+        stride_height, stride_width, padding_height, padding_width,
+        output_data);
+  }
+};
+
+template <typename T>
+class DepthwiseConvInputGradFunctor<platform::CUDADeviceContext, T> {
+ public:
+  void operator()(const platform::CUDADeviceContext& context,
+                  const framework::Tensor& input,
+                  const framework::Tensor& filter,
+                  const framework::Tensor& output_grad,
+                  const std::vector<int>& strides,
+                  const std::vector<int>& paddings,
+                  framework::Tensor* input_grad) {
+    const int batch_size = input.dims()[0];
+    const int input_channels = input.dims()[1];
+    const int input_height = input.dims()[2];
+    const int input_width = input.dims()[3];
+    const int output_channels = output_grad.dims()[1];
+    const int output_height = output_grad.dims()[2];
+    const int output_width = output_grad.dims()[3];
+    const int ksize_height = filter.dims()[2];
+    const int ksize_width = filter.dims()[3];
+    const int stride_height = strides[0];
+    const int stride_width = strides[1];
+    const int padding_height = paddings[0];
+    const int padding_width = paddings[1];
+
+    const T* filter_data = filter.data<T>();
+    const T* output_grad_data = output_grad.data<T>();
+    T* input_grad_data = input_grad->mutable_data<T>(context.GetPlace());
+
+    int nthreads = batch_size * input_channels * input_height * input_width;
+    int blocks = (nthreads + 1024 - 1) / 1024;
+    dim3 threads(1024, 1);
+    dim3 grid(blocks, 1);
+
+    KernelDepthwiseConvInputGrad<T><<<grid, threads, 0, context.stream()>>>(
+        nthreads, output_grad_data, filter_data, batch_size, output_channels,
+        output_height, output_width, input_channels, input_height, input_width,
+        output_channels / input_channels, ksize_height, ksize_width,
+        stride_height, stride_width, padding_height, padding_width,
+        input_grad_data);
+  }
+};
+
+template <typename T>
+class DepthwiseConvFilterGradFunctor<platform::CUDADeviceContext, T> {
+ public:
+  void operator()(const platform::CUDADeviceContext& context,
+                  const framework::Tensor& input,
+                  const framework::Tensor& output_grad,
+                  const std::vector<int>& strides,
+                  const std::vector<int>& paddings,
+                  framework::Tensor* filter_grad) {
+    const int batch_size = input.dims()[0];
+    const int input_channels = input.dims()[1];
+    const int input_height = input.dims()[2];
+    const int input_width = input.dims()[3];
+    const int output_channels = output_grad.dims()[1];
+    const int output_height = output_grad.dims()[2];
+    const int output_width = output_grad.dims()[3];
+    const int ksize_height = filter_grad->dims()[2];
+    const int ksize_width = filter_grad->dims()[3];
+    const int stride_height = strides[0];
+    const int stride_width = strides[1];
+    const int padding_height = paddings[0];
+    const int padding_width = paddings[1];
+
+    const T* input_data = input.data<T>();
+    const T* output_grad_data = output_grad.data<T>();
+    T* filter_grad_data = filter_grad->mutable_data<T>(context.GetPlace());
+
+    int nthreads = batch_size * output_channels * output_height * output_width;
+
+    int blocks = (nthreads + 1024 - 1) / 1024;
+    dim3 threads(1024, 1);
+    dim3 grid(blocks, 1);
+
+    KernelDepthwiseConvFilterGrad<T><<<grid, threads, 0, context.stream()>>>(
+        nthreads, output_grad_data, input_data, batch_size, output_channels,
+        output_height, output_width, input_channels, input_height, input_width,
+        output_channels / input_channels, ksize_height, ksize_width,
+        stride_height, stride_width, padding_height, padding_width,
+        filter_grad_data);
+  }
+};
+
+template class DepthwiseConvFunctor<platform::CUDADeviceContext, float>;
+template class DepthwiseConvFunctor<platform::CUDADeviceContext, double>;
+
+template class DepthwiseConvInputGradFunctor<platform::CUDADeviceContext,
+                                             float>;
+template class DepthwiseConvInputGradFunctor<platform::CUDADeviceContext,
+                                             double>;
+
+template class DepthwiseConvFilterGradFunctor<platform::CUDADeviceContext,
+                                              float>;
+template class DepthwiseConvFilterGradFunctor<platform::CUDADeviceContext,
+                                              double>;
+
+}  // namespace math
+}  // namespace operators
+}  // namespace paddle

paddle/operators/math/depthwise_conv.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/framework/tensor.h"
+#include "paddle/platform/device_context.h"
+#include "paddle/platform/hostdevice.h"
+
+namespace paddle {
+namespace operators {
+namespace math {
+
+/*
+ * \brief Compute the depthwise convolution which include
+ * forward process and backpropagation process
+ */
+template <typename DeviceContext, typename T>
+class DepthwiseConvFunctor {
+ public:
+  void operator()(const DeviceContext& context, const framework::Tensor& input,
+                  const framework::Tensor& filter,
+                  const std::vector<int>& strides,
+                  const std::vector<int>& paddings, framework::Tensor* output);
+};
+
+template <typename DeviceContext, typename T>
+class DepthwiseConvInputGradFunctor {
+ public:
+  void operator()(const DeviceContext& context, const framework::Tensor& input,
+                  const framework::Tensor& filter,
+                  const framework::Tensor& output_grad,
+                  const std::vector<int>& strides,
+                  const std::vector<int>& paddings,
+                  framework::Tensor* input_grad);
+};
+
+template <typename DeviceContext, typename T>
+class DepthwiseConvFilterGradFunctor {
+ public:
+  void operator()(const DeviceContext& context, const framework::Tensor& input,
+                  const framework::Tensor& output_grad,
+                  const std::vector<int>& strides,
+                  const std::vector<int>& paddings,
+                  framework::Tensor* filter_grad);
+};
+
+}  // namespace math
+}  // namespace operators
+}  // namespace paddle

paddle/operators/mine_hard_examples_op.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. */
+
+#include "paddle/framework/eigen.h"
+#include "paddle/framework/op_registry.h"
+
+namespace paddle {
+namespace operators {
+
+enum MiningType { kNone = 0, kMaxNegative, kHardExample };
+
+template <typename T>
+bool SortScoreDescend(const std::pair<float, T>& pair1,
+                      const std::pair<float, T>& pair2) {
+  return pair1.first > pair2.first;
+}
+
+inline bool IsEligibleMining(const MiningType mining_type, const int match_idx,
+                             const float match_dist,
+                             const float neg_dist_threshold) {
+  if (mining_type == MiningType::kMaxNegative) {
+    return match_idx == -1 && match_dist < neg_dist_threshold;
+  } else if (mining_type == MiningType::kHardExample) {
+    return true;
+  } else {
+    return false;
+  }
+}
+
+inline MiningType GetMiningType(std::string str) {
+  if (str == "max_negative") {
+    return MiningType::kMaxNegative;
+  } else if (str == "hard_example") {
+    return MiningType::kHardExample;
+  } else {
+    return MiningType::kNone;
+  }
+}
+
+template <typename DeviceContext, typename T>
+class MineHardExamplesKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    auto* in_cls_loss = ctx.Input<framework::Tensor>("ClsLoss");
+    auto* in_loc_loss = ctx.Input<framework::Tensor>("LocLoss");
+    auto* in_matched_indices = ctx.Input<framework::Tensor>("MatchIndices");
+    auto* in_match_dist = ctx.Input<framework::Tensor>("MatchDist");
+    float neg_pos_ratio = ctx.Attr<float>("neg_pos_ratio");
+    T neg_dist_threshold =
+        static_cast<T>(ctx.Attr<float>("neg_dist_threshold"));
+    int sample_size = ctx.Attr<int>("sample_size");
+    MiningType mining_type =
+        GetMiningType(ctx.Attr<std::string>("mining_type"));
+
+    auto out_neg_indices = ctx.Output<framework::LoDTensor>("NegIndices");
+    auto out_match_indices =
+        ctx.Output<framework::Tensor>("UpdatedMatchIndices");
+
+    framework::Copy(*in_matched_indices, ctx.GetPlace(), out_match_indices);
+
+    int batch_size = in_matched_indices->dims()[0];
+    int prior_num = in_matched_indices->dims()[1];
+
+    auto match_indices = framework::EigenMatrix<int>::From(*in_matched_indices);
+
+    auto match_indices_et =
+        framework::EigenMatrix<int>::From(*out_match_indices);
+
+    auto match_dist = framework::EigenMatrix<T>::From(*in_match_dist);
+
+    const T* cls_loss = in_cls_loss->data<T>();
+    const T* loc_loss = nullptr;
+    if (in_loc_loss) {
+      loc_loss = in_loc_loss->data<T>();
+    }
+
+    std::vector<std::vector<int>> all_neg_indices;
+    std::vector<size_t> batch_starts = {0};
+    for (int n = 0; n < batch_size; ++n) {
+      std::vector<std::pair<T, size_t>> loss_idx;
+      int neg_sel = 0;
+      for (int m = 0; m < prior_num; ++m) {
+        if (IsEligibleMining(mining_type, match_indices(n, m), match_dist(n, m),
+                             neg_dist_threshold)) {
+          T loss = cls_loss[n * prior_num + m];
+          if (mining_type == MiningType::kHardExample && loc_loss != nullptr) {
+            loss = cls_loss[n * prior_num + m] + loc_loss[n * prior_num + m];
+          }
+          loss_idx.push_back(std::make_pair(loss, m));
+          ++neg_sel;
+        }
+      }
+
+      if (mining_type == MiningType::kMaxNegative) {
+        int num_pos = 0;
+        for (int m = 0; m < prior_num; ++m) {
+          if (match_indices(n, m) != -1) ++num_pos;
+        }
+        neg_sel = std::min(static_cast<int>(num_pos * neg_pos_ratio), neg_sel);
+      } else if (mining_type == MiningType::kHardExample) {
+        neg_sel = std::min(sample_size, neg_sel);
+      }
+
+      std::sort(loss_idx.begin(), loss_idx.end(), SortScoreDescend<size_t>);
+      std::set<int> sel_indices;
+      std::vector<int> neg_indices;
+      std::transform(loss_idx.begin(), loss_idx.begin() + neg_sel,
+                     std::inserter(sel_indices, sel_indices.begin()),
+                     [](std::pair<T, size_t>& l) -> int {
+                       return static_cast<int>(l.second);
+                     });
+
+      if (mining_type == MiningType::kHardExample) {
+        for (int m = 0; m < prior_num; ++m) {
+          if (match_indices(n, m) > -1) {
+            if (sel_indices.find(m) == sel_indices.end()) {
+              match_indices_et(n, m) = -1;
+            }
+          } else {
+            if (sel_indices.find(m) != sel_indices.end()) {
+              neg_indices.push_back(m);
+            }
+          }
+        }
+      } else {
+        neg_indices.resize(sel_indices.size());
+        std::copy(sel_indices.begin(), sel_indices.end(), neg_indices.begin());
+      }
+
+      all_neg_indices.push_back(neg_indices);
+      batch_starts.push_back(batch_starts.back() + neg_indices.size());
+    }
+
+    framework::LoD out_neg_indices_lod;
+    out_neg_indices_lod.emplace_back(batch_starts);
+    int neg_offset = 0;
+    auto neg_data = out_neg_indices->mutable_data<int>(
+        framework::make_ddim({static_cast<int>(batch_starts.back()), 1}),
+        ctx.GetPlace());
+
+    for (auto neg_indices : all_neg_indices) {
+      std::copy(neg_indices.begin(), neg_indices.end(), neg_data + neg_offset);
+      neg_offset += neg_indices.size();
+    }
+    out_neg_indices->set_lod(out_neg_indices_lod);
+    return;
+  }
+};
+
+class MineHardExamplesOp : public framework::OperatorWithKernel {
+ public:
+  using framework::OperatorWithKernel::OperatorWithKernel;
+
+ protected:
+  void InferShape(framework::InferShapeContext* ctx) const override {
+    PADDLE_ENFORCE(ctx->HasInput("ClsLoss"),
+                   "Input(ClsLoss) of MineHardExamplesOp should not be null.");
+    PADDLE_ENFORCE(
+        ctx->HasInput("MatchIndices"),
+        "Input(MatchIndices) of MineHardExamplesOp should not be null.");
+    PADDLE_ENFORCE(
+        ctx->HasInput("MatchDist"),
+        "Input(MatchDist) of MineHardExamplesOp should not be null.");
+    PADDLE_ENFORCE(
+        ctx->HasOutput("NegIndices"),
+        "Output(NegIndices) of MineHardExamplesOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasOutput("UpdatedMatchIndices"),
+                   "Output(UpdatedMatchIndices) of MineHardExamplesOp should "
+                   "not be null.");
+
+    auto cls_loss_dims = ctx->GetInputDim("ClsLoss");
+    auto idx_dims = ctx->GetInputDim("MatchIndices");
+    auto dis_dims = ctx->GetInputDim("MatchDist");
+
+    PADDLE_ENFORCE_EQ(cls_loss_dims.size(), 2UL,
+                      "The shape of ClsLoss is [N, Np].");
+    PADDLE_ENFORCE_EQ(idx_dims.size(), 2UL,
+                      "The shape of MatchIndices is [N, Np].");
+    PADDLE_ENFORCE_EQ(dis_dims.size(), 2UL,
+                      "The shape of MatchDist is [N, Np].");
+
+    if (ctx->HasInput("LocLoss")) {
+      auto loc_loss_dims = ctx->GetInputDim("LocLoss");
+      PADDLE_ENFORCE_EQ(loc_loss_dims.size(), 2UL,
+                        "The shape of LocLoss is [N, Np].");
+      PADDLE_ENFORCE_EQ(cls_loss_dims[0], loc_loss_dims[0],
+                        "Batch size of ClsLoss and LocLoss must be the same.");
+      PADDLE_ENFORCE_EQ(
+          cls_loss_dims[1], loc_loss_dims[1],
+          "Prior box number of ClsLoss and LocLoss must be the same.");
+    }
+
+    PADDLE_ENFORCE_EQ(
+        cls_loss_dims[0], idx_dims[0],
+        "Batch size of ClsLoss and MatchIndices must be the same.");
+    PADDLE_ENFORCE_EQ(
+        cls_loss_dims[1], idx_dims[1],
+        "Prior box number of ClsLoss and MatchIndices must be the same.");
+
+    PADDLE_ENFORCE_EQ(cls_loss_dims[0], dis_dims[0],
+                      "Batch size of ClsLoss and MatchDist must be the same.");
+    PADDLE_ENFORCE_EQ(
+        cls_loss_dims[1], idx_dims[1],
+        "Prior box number of ClsLoss and MatchDist must be the same.");
+
+    auto mining_type =
+        GetMiningType(ctx->Attrs().Get<std::string>("mining_type"));
+
+    PADDLE_ENFORCE_NE(mining_type, MiningType::kNone,
+                      "mining_type must be hard_example or max_negative");
+
+    if (mining_type == MiningType::kMaxNegative) {
+      auto neg_pos_ratio = ctx->Attrs().Get<float>("neg_pos_ratio");
+      auto neg_dist_threshold = ctx->Attrs().Get<float>("neg_dist_threshold");
+      PADDLE_ENFORCE_GT(
+          neg_pos_ratio, 0.0f,
+          "neg_pos_ratio must greater than zero in max_negative mode");
+      PADDLE_ENFORCE_GT(
+          neg_dist_threshold, 0.0f,
+          "neg_dist_threshold must greater than zero in max_negative mode");
+    } else if (mining_type == MiningType::kHardExample) {
+      auto sample_size = ctx->Attrs().Get<int>("sample_size");
+      PADDLE_ENFORCE_GT(
+          sample_size, 0,
+          "sample_size must greater than zero in hard_example mode");
+    }
+
+    ctx->SetOutputDim("UpdatedMatchIndices", idx_dims);
+  }
+
+ protected:
+  framework::OpKernelType GetExpectedKernelType(
+      const framework::ExecutionContext& ctx) const override {
+    return framework::OpKernelType(
+        framework::ToDataType(ctx.Input<framework::Tensor>("ClsLoss")->type()),
+        ctx.device_context());
+  }
+};
+
+class MineHardExamplesOpMaker : public framework::OpProtoAndCheckerMaker {
+ public:
+  MineHardExamplesOpMaker(OpProto* proto, OpAttrChecker* op_checker)
+      : OpProtoAndCheckerMaker(proto, op_checker) {
+    AddInput(
+        "ClsLoss",
+        "(Tensor, default Tensor<float>), The classification loss with shape "
+        "[N, Np], N is the batch size and Np is the number of prior box.");
+    AddInput("LocLoss",
+             "(Tensor, optional, default Tensor<float>), The localization loss "
+             "with shape [N, Np], N is the batch size and Np is the number of "
+             "prior box.")
+        .AsDispensable();
+    AddInput("MatchIndices",
+             "(Tensor, Tensor<int>), Matched indices with shape [N, Np], N is "
+             "the batch size and Np is the number of prior box. "
+             "MatchIndices[i][j] equal -1 means the j-th prior box in i-th "
+             "instance does not match any entity, otherwise means it is "
+             "matched to row.");
+    AddInput("MatchDist",
+             "(Tensor, default Tensor<float>) Matched indices with shape [N, "
+             "Np], N is the batch size and Np is the number of prior box.");
+    AddAttr<float>("neg_pos_ratio",
+                   "(float) The ratio of the negative box to the positive "
+                   "box. Use only when mining_type is max_negative.")
+        .SetDefault(1.0);
+    AddAttr<float>("neg_dist_threshold",
+                   "(float) The negative overlap upper bound for the unmatched "
+                   "predictions. Use only when mining_type is max_negative.")
+        .SetDefault(0.5);
+    AddAttr<int>("sample_size",
+                 "(float) The max sample size of negative box. Use only when "
+                 "mining_type is hard_example.")
+        .SetDefault(0);
+    AddAttr<std::string>("mining_type",
+                         "(float) The mining algorithm name, the value is "
+                         "hard_example or max_negative.")
+        .SetDefault("max_negative")
+        .InEnum({"hard_example", "max_negative"});
+
+    AddOutput(
+        "NegIndices",
+        "(LoDTensor<int>) The output of negative example indices. a LoDTensor "
+        "with shape [Neg, 1]. The size of lod[0] minus 1 is batch size, "
+        "and each element is the prior box index. "
+        "For example, the batch size is 2, the lod is [[0, 1, 2]], "
+        "the sample 0's box 1(MatchIndices[0][1]) is selected, "
+        "and sample 1's box 0 is selected. The output NegIndices is "
+        "[[1], [0]].");
+
+    AddOutput("UpdatedMatchIndices",
+              "(Tensor<int>) The output of updated MatchIndices, a tensor with "
+              "shape [N, Np]. Only update when mining_type is "
+              "hard_example. The input MatchIndices elements will be update to "
+              "-1 when it is not in the candidate high loss list of negative "
+              "examples.");
+
+    AddComment(R"DOC(
+Mine hard examples Operator.
+This operator implements hard example mining to select a subset of negative box indices.
+For each image, selects the box with highest losses. subject to the condition that the 
+box cannot have an Matcht > neg_dist_threshold when mining_type is max_negative. 
+The selected number is min(sample_size, max_negative_box_number) when mining_type is 
+hard_example, or min(neg_pos_ratio * positive_box_number, max_negative_box_number) 
+when mining_type is max_negative, where the max_negative_box_number is the count of 
+MatchIndices elements with value -1.
+)DOC");
+  }
+};
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OP_WITHOUT_GRADIENT(mine_hard_examples, ops::MineHardExamplesOp,
+                             ops::MineHardExamplesOpMaker);
+
+REGISTER_OP_CPU_KERNEL(
+    mine_hard_examples,
+    ops::MineHardExamplesKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::MineHardExamplesKernel<paddle::platform::CPUDeviceContext, double>);

paddle/operators/multiclass_nms_op.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. */
+
+#include "paddle/framework/op_registry.h"
+
+namespace paddle {
+namespace operators {
+
+using Tensor = framework::Tensor;
+using LoDTensor = framework::LoDTensor;
+
+constexpr int64_t kOutputDim = 6;
+constexpr int64_t kBBoxSize = 4;
+
+class MultiClassNMSOp : public framework::OperatorWithKernel {
+ public:
+  using framework::OperatorWithKernel::OperatorWithKernel;
+
+  void InferShape(framework::InferShapeContext* ctx) const override {
+    PADDLE_ENFORCE(ctx->HasInput("BBoxes"),
+                   "Input(BBoxes) of MultiClassNMS should not be null.");
+    PADDLE_ENFORCE(ctx->HasInput("Scores"),
+                   "Input(Scores) of MultiClassNMS should not be null.");
+    PADDLE_ENFORCE(ctx->HasOutput("Out"),
+                   "Output(Out) of MultiClassNMS should not be null.");
+
+    auto box_dims = ctx->GetInputDim("BBoxes");
+    auto score_dims = ctx->GetInputDim("Scores");
+
+    PADDLE_ENFORCE_EQ(box_dims.size(), 2,
+                      "The rank of Input(BBoxes) must be 2.");
+    PADDLE_ENFORCE_EQ(score_dims.size(), 3,
+                      "The rank of Input(Scores) must be 3.");
+    PADDLE_ENFORCE_EQ(box_dims[1], 4,
+                      "The 2nd dimension of Input(BBoxes) must be 4, "
+                      "represents the layout of coordinate "
+                      "[xmin, ymin, xmax, ymax]");
+    PADDLE_ENFORCE_EQ(box_dims[0], score_dims[2],
+                      "The 1st dimensiong of Input(BBoxes) must be equal to "
+                      "3rd dimension of Input(Scores), which represents the "
+                      "predicted bboxes.");
+
+    // Here the box_dims[0] is not the real dimension of output.
+    // It will be rewritten in the computing kernel.
+    ctx->SetOutputDim("Out", {box_dims[0], 6});
+  }
+
+ protected:
+  framework::OpKernelType GetExpectedKernelType(
+      const framework::ExecutionContext& ctx) const override {
+    return framework::OpKernelType(
+        framework::ToDataType(
+            ctx.Input<framework::LoDTensor>("Scores")->type()),
+        ctx.device_context());
+  }
+};
+
+template <class T>
+bool SortScorePairDescend(const std::pair<float, T>& pair1,
+                          const std::pair<float, T>& pair2) {
+  return pair1.first > pair2.first;
+}
+
+template <class T>
+static inline void GetMaxScoreIndex(
+    const std::vector<T>& scores, const T threshold, int top_k,
+    std::vector<std::pair<T, int>>* sorted_indices) {
+  for (size_t i = 0; i < scores.size(); ++i) {
+    if (scores[i] > threshold) {
+      sorted_indices->push_back(std::make_pair(scores[i], i));
+    }
+  }
+  // Sort the score pair according to the scores in descending order
+  std::stable_sort(sorted_indices->begin(), sorted_indices->end(),
+                   SortScorePairDescend<int>);
+  // Keep top_k scores if needed.
+  if (top_k > -1 && top_k < static_cast<int>(sorted_indices->size())) {
+    sorted_indices->resize(top_k);
+  }
+}
+
+template <class T>
+static inline T BBoxArea(const T* box, const bool normalized) {
+  if (box[2] < box[0] || box[3] < box[1]) {
+    // If coordinate values are is invalid
+    // (e.g. xmax < xmin or ymax < ymin), return 0.
+    return static_cast<T>(0.);
+  } else {
+    const T w = box[2] - box[0];
+    const T h = box[3] - box[1];
+    if (normalized) {
+      return w * h;
+    } else {
+      // If coordinate values are not within range [0, 1].
+      return (w + 1) * (h + 1);
+    }
+  }
+}
+
+template <class T>
+static inline T JaccardOverlap(const T* box1, const T* box2,
+                               const bool normalized) {
+  if (box2[0] > box1[2] || box2[2] < box1[0] || box2[1] > box1[3] ||
+      box2[3] < box1[1]) {
+    return static_cast<T>(0.);
+  } else {
+    const T inter_xmin = std::max(box1[0], box2[0]);
+    const T inter_ymin = std::max(box1[1], box2[1]);
+    const T inter_xmax = std::min(box1[2], box2[2]);
+    const T inter_ymax = std::min(box1[3], box2[3]);
+    const T inter_w = inter_xmax - inter_xmin;
+    const T inter_h = inter_ymax - inter_ymin;
+    const T inter_area = inter_w * inter_h;
+    const T bbox1_area = BBoxArea<T>(box1, normalized);
+    const T bbox2_area = BBoxArea<T>(box2, normalized);
+    return inter_area / (bbox1_area + bbox2_area - inter_area);
+  }
+}
+
+template <typename T>
+class MultiClassNMSKernel : public framework::OpKernel<T> {
+ public:
+  void NMSFast(const Tensor& bbox, const Tensor& scores,
+               const T score_threshold, const T nms_threshold, const T eta,
+               const int64_t top_k, std::vector<int>* selected_indices) const {
+    // The total boxes for each instance.
+    int64_t num_boxes = bbox.dims()[0];
+    // 4: [xmin ymin xmax ymax]
+    int64_t box_size = bbox.dims()[1];
+
+    std::vector<T> scores_data(num_boxes);
+    std::copy_n(scores.data<T>(), num_boxes, scores_data.begin());
+    std::vector<std::pair<T, int>> sorted_indices;
+    GetMaxScoreIndex(scores_data, score_threshold, top_k, &sorted_indices);
+
+    selected_indices->clear();
+    T adaptive_threshold = nms_threshold;
+    const T* bbox_data = bbox.data<T>();
+
+    while (sorted_indices.size() != 0) {
+      const int idx = sorted_indices.front().second;
+      bool keep = true;
+      for (size_t k = 0; k < selected_indices->size(); ++k) {
+        if (keep) {
+          const int kept_idx = (*selected_indices)[k];
+          T overlap = JaccardOverlap<T>(bbox_data + idx * box_size,
+                                        bbox_data + kept_idx * box_size, true);
+          keep = overlap <= adaptive_threshold;
+        } else {
+          break;
+        }
+      }
+      if (keep) {
+        selected_indices->push_back(idx);
+      }
+      sorted_indices.erase(sorted_indices.begin());
+      if (keep && eta < 1 && adaptive_threshold > 0.5) {
+        adaptive_threshold *= eta;
+      }
+    }
+  }
+
+  void MultiClassNMS(const framework::ExecutionContext& ctx,
+                     const Tensor& scores, const Tensor& bboxes,
+                     std::map<int, std::vector<int>>& indices,
+                     int& num_nmsed_out) const {
+    int64_t background_label = ctx.Attr<int>("background_label");
+    int64_t nms_top_k = ctx.Attr<int>("nms_top_k");
+    int64_t keep_top_k = ctx.Attr<int>("keep_top_k");
+    T nms_threshold = static_cast<T>(ctx.Attr<float>("nms_threshold"));
+    T nms_eta = static_cast<T>(ctx.Attr<float>("nms_eta"));
+    T score_threshold = static_cast<T>(ctx.Attr<float>("score_threshold"));
+
+    int64_t class_num = scores.dims()[0];
+    int64_t predict_dim = scores.dims()[1];
+    int num_det = 0;
+    for (int64_t c = 0; c < class_num; ++c) {
+      if (c == background_label) continue;
+      Tensor score = scores.Slice(c, c + 1);
+      NMSFast(bboxes, score, score_threshold, nms_threshold, nms_eta, nms_top_k,
+              &(indices[c]));
+      num_det += indices[c].size();
+    }
+
+    num_nmsed_out = num_det;
+    const T* scores_data = scores.data<T>();
+    if (keep_top_k > -1 && num_det > keep_top_k) {
+      std::vector<std::pair<float, std::pair<int, int>>> score_index_pairs;
+      for (const auto& it : indices) {
+        int label = it.first;
+        const T* sdata = scores_data + label * predict_dim;
+        const std::vector<int>& label_indices = it.second;
+        for (size_t j = 0; j < label_indices.size(); ++j) {
+          int idx = label_indices[j];
+          PADDLE_ENFORCE_LT(idx, predict_dim);
+          score_index_pairs.push_back(
+              std::make_pair(sdata[idx], std::make_pair(label, idx)));
+        }
+      }
+      // Keep top k results per image.
+      std::stable_sort(score_index_pairs.begin(), score_index_pairs.end(),
+                       SortScorePairDescend<std::pair<int, int>>);
+      score_index_pairs.resize(keep_top_k);
+
+      // Store the new indices.
+      std::map<int, std::vector<int>> new_indices;
+      for (size_t j = 0; j < score_index_pairs.size(); ++j) {
+        int label = score_index_pairs[j].second.first;
+        int idx = score_index_pairs[j].second.second;
+        new_indices[label].push_back(idx);
+      }
+      new_indices.swap(indices);
+      num_nmsed_out = keep_top_k;
+    }
+  }
+
+  void MultiClassOutput(const Tensor& scores, const Tensor& bboxes,
+                        std::map<int, std::vector<int>>& selected_indices,
+                        Tensor* outs) const {
+    int predict_dim = scores.dims()[1];
+    auto* scores_data = scores.data<T>();
+    auto* bboxes_data = bboxes.data<T>();
+    auto* odata = outs->data<T>();
+
+    int count = 0;
+    for (const auto& it : selected_indices) {
+      int label = it.first;
+      const T* sdata = scores_data + label * predict_dim;
+      const std::vector<int>& indices = it.second;
+      for (size_t j = 0; j < indices.size(); ++j) {
+        int idx = indices[j];
+        const T* bdata = bboxes_data + idx * kBBoxSize;
+        odata[count * kOutputDim] = label;           // label
+        odata[count * kOutputDim + 1] = sdata[idx];  // score
+        // xmin, ymin, xmax, ymax
+        std::memcpy(odata + count * kOutputDim + 2, bdata, 4 * sizeof(T));
+        count++;
+      }
+    }
+  }
+
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    auto* boxes = ctx.Input<Tensor>("BBoxes");
+    auto* scores = ctx.Input<Tensor>("Scores");
+    auto* outs = ctx.Output<LoDTensor>("Out");
+
+    auto score_dims = scores->dims();
+
+    int64_t batch_size = score_dims[0];
+    int64_t class_num = score_dims[1];
+    int64_t predict_dim = score_dims[2];
+
+    std::vector<std::map<int, std::vector<int>>> all_indices;
+    std::vector<size_t> batch_starts = {0};
+    for (int64_t i = 0; i < batch_size; ++i) {
+      Tensor ins_score = scores->Slice(i, i + 1);
+      ins_score.Resize({class_num, predict_dim});
+      std::map<int, std::vector<int>> indices;
+      int num_nmsed_out = 0;
+      MultiClassNMS(ctx, ins_score, *boxes, indices, num_nmsed_out);
+      all_indices.push_back(indices);
+      batch_starts.push_back(batch_starts.back() + num_nmsed_out);
+    }
+
+    int num_kept = batch_starts.back();
+    if (num_kept == 0) {
+      T* od = outs->mutable_data<T>({1}, ctx.GetPlace());
+      od[0] = -1;
+    } else {
+      outs->mutable_data<T>({num_kept, kOutputDim}, ctx.GetPlace());
+      for (int64_t i = 0; i < batch_size; ++i) {
+        Tensor ins_score = scores->Slice(i, i + 1);
+        ins_score.Resize({class_num, predict_dim});
+        int64_t s = batch_starts[i];
+        int64_t e = batch_starts[i + 1];
+        if (e > s) {
+          Tensor out = outs->Slice(s, e);
+          MultiClassOutput(ins_score, *boxes, all_indices[i], &out);
+        }
+      }
+    }
+
+    framework::LoD lod;
+    lod.emplace_back(batch_starts);
+
+    outs->set_lod(lod);
+  }
+};
+
+class MultiClassNMSOpMaker : public framework::OpProtoAndCheckerMaker {
+ public:
+  MultiClassNMSOpMaker(OpProto* proto, OpAttrChecker* op_checker)
+      : OpProtoAndCheckerMaker(proto, op_checker) {
+    AddInput("BBoxes",
+             "(Tensor) A 2-D Tensor with shape [M, 4] represents the "
+             "predicted locations of M bounding bboxes. Each bounding box "
+             "has four coordinate values and the layout is "
+             "[xmin, ymin, xmax, ymax].");
+    AddInput("Scores",
+             "(Tensor) A 3-D Tensor with shape [N, C, M] represents the "
+             "predicted confidence predictions. N is the batch size, C is the "
+             "class number, M is number of bounding boxes. For each category "
+             "there are total M scores which corresponding M bounding boxes. "
+             " Please note, M is equal to the 1st dimension of BBoxes. ");
+    AddAttr<int>(
+        "background_label",
+        "(int64_t, defalut: 0) "
+        "The index of background label, the background label will be ignored. "
+        "If set to -1, then all categories will be considered.")
+        .SetDefault(0);
+    AddAttr<float>("score_threshold",
+                   "(float) "
+                   "Threshold to filter out bounding boxes with low "
+                   "confidence score. If not provided, consider all boxes.");
+    AddAttr<int>("nms_top_k",
+                 "(int64_t) "
+                 "Maximum number of detections to be kept according to the "
+                 "confidences aftern the filtering detections based on "
+                 "score_threshold");
+    AddAttr<float>("nms_threshold",
+                   "(float, defalut: 0.3) "
+                   "The threshold to be used in NMS.")
+        .SetDefault(0.3);
+    AddAttr<float>("nms_eta",
+                   "(float) "
+                   "The parameter for adaptive NMS.")
+        .SetDefault(1.0);
+    AddAttr<int>("keep_top_k",
+                 "(int64_t) "
+                 "Number of total bboxes to be kept per image after NMS "
+                 "step. -1 means keeping all bboxes after NMS step.");
+    AddOutput("Out",
+              "(LoDTensor) A 2-D LoDTensor with shape [No, 6] represents the "
+              "detections. Each row has 6 values: "
+              "[label, confidence, xmin, ymin, xmax, ymax], No is the total "
+              "number of detections in this mini-batch. For each instance, "
+              "the offsets in first dimension are called LoD, the number of "
+              "offset is N + 1, if LoD[i + 1] - LoD[i] == 0, means there is "
+              "no detected bbox.");
+    AddComment(R"DOC(
+This operator is to do multi-class non maximum suppression (NMS) on a batched
+of boxes and scores.
+
+In the NMS step, this operator greedily selects a subset of detection bounding
+boxes that have high scores larger than score_threshold, if providing this
+threshold, then selects the largest nms_top_k confidences scores if nms_top_k
+is larger than -1. Then this operator pruns away boxes that have high IOU
+(intersection over union) overlap with already selected boxes by adaptive
+threshold NMS based on parameters of nms_threshold and nms_eta.
+
+Aftern NMS step, at most keep_top_k number of total bboxes are to be kept
+per image if keep_top_k is larger than -1.
+
+This operator support multi-class and batched inputs. It applying NMS
+independently for each class. The outputs is a 2-D LoDTenosr, for each
+image, the offsets in first dimension of LoDTensor are called LoD, the number
+of offset is N + 1, where N is the batch size. If LoD[i + 1] - LoD[i] == 0,
+means there is no detected bbox for this image. If there is no detected boxes
+for all images, all the elements in LoD are 0, and the Out only contains one
+value which is -1.
+)DOC");
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+REGISTER_OPERATOR(multiclass_nms, ops::MultiClassNMSOp,
+                  ops::MultiClassNMSOpMaker,
+                  paddle::framework::EmptyGradOpMaker);
+REGISTER_OP_CPU_KERNEL(multiclass_nms, ops::MultiClassNMSKernel<float>,
+                       ops::MultiClassNMSKernel<double>);

paddle/operators/prior_box_op.cc

@@ -44,12 +44,6 @@ class PriorBoxOp : public framework::OperatorWithKernel {
     auto aspect_ratios = ctx->Attrs().Get<std::vector<float>>("aspect_ratios");
     bool flip = ctx->Attrs().Get<bool>("flip");
 
-    PADDLE_ENFORCE_GT(min_sizes.size(), 0,
-                      "Size of min_sizes must be at least 1.");
-    for (size_t i = 0; i < min_sizes.size(); ++i) {
-      PADDLE_ENFORCE_GT(min_sizes[i], 0, "min_sizes[%d] must be positive.", i);
-    }
-
     std::vector<float> aspect_ratios_vec;
     ExpandAspectRatios(aspect_ratios, flip, aspect_ratios_vec);
 
@@ -65,17 +59,6 @@ class PriorBoxOp : public framework::OperatorWithKernel {
       }
     }
 
-    PADDLE_ENFORCE_EQ(variances.size(), 4, "Must and only provide 4 variance.");
-    for (size_t i = 0; i < variances.size(); ++i) {
-      PADDLE_ENFORCE_GT(variances[i], 0.0,
-                        "variance[%d] must be greater than 0.", i);
-    }
-
-    const float step_h = ctx->Attrs().Get<float>("step_h");
-    PADDLE_ENFORCE_GT(step_h, 0.0, "step_h should be larger than 0.");
-    const float step_w = ctx->Attrs().Get<float>("step_w");
-    PADDLE_ENFORCE_GT(step_w, 0.0, "step_w should be larger than 0.");
-
     std::vector<int64_t> dim_vec(4);
     dim_vec[0] = input_dims[2];
     dim_vec[1] = input_dims[3];
@@ -106,26 +89,54 @@ class PriorBoxOpMaker : public framework::OpProtoAndCheckerMaker {
               "PriorBoxOp. The layout is [H, W, num_priors, 4]. "
               "H is the height of input, W is the width of input, num_priors "
               "is the box count of each position.");
-    AddAttr<std::vector<int>>("min_sizes", "(vector<int>) ",
-                              "List of min sizes of generated prior boxes.");
-    AddAttr<std::vector<int>>("max_sizes", "(vector<int>) ",
-                              "List of max sizes of generated prior boxes.");
+
+    AddAttr<std::vector<int>>("min_sizes",
+                              "(vector<int>) List of min sizes "
+                              "of generated prior boxes.")
+        .AddCustomChecker([](const std::vector<int>& min_sizes) {
+          PADDLE_ENFORCE_GT(min_sizes.size(), 0,
+                            "Size of min_sizes must be at least 1.");
+          for (size_t i = 0; i < min_sizes.size(); ++i) {
+            PADDLE_ENFORCE_GT(min_sizes[i], 0,
+                              "min_sizes[%d] must be positive.", i);
+          }
+        });
+    AddAttr<std::vector<int>>(
+        "max_sizes",
+        "(vector<int>) List of max sizes of generated prior boxes.");
     AddAttr<std::vector<float>>(
-        "aspect_ratios", "(vector<float>) ",
-        "List of aspect ratios of generated prior boxes.");
+        "aspect_ratios",
+        "(vector<float>) List of aspect ratios of generated prior boxes.");
+
     AddAttr<std::vector<float>>(
-        "variances", "(vector<float>) ",
-        "List of variances to be encoded in prior boxes.");
-    AddAttr<bool>("flip", "(bool) ", "Whether to flip aspect ratios.")
+        "variances",
+        "(vector<float>) List of variances to be encoded in prior boxes.")
+        .AddCustomChecker([](const std::vector<float>& variances) {
+          PADDLE_ENFORCE_EQ(variances.size(), 4,
+                            "Must and only provide 4 variance.");
+          for (size_t i = 0; i < variances.size(); ++i) {
+            PADDLE_ENFORCE_GT(variances[i], 0.0,
+                              "variance[%d] must be greater than 0.", i);
+          }
+        });
+    AddAttr<bool>("flip", "(bool) Whether to flip aspect ratios.")
         .SetDefault(true);
-    AddAttr<bool>("clip", "(bool) ", "Whether to clip out-of-boundary boxes.")
+    AddAttr<bool>("clip", "(bool) Whether to clip out-of-boundary boxes.")
         .SetDefault(true);
+
     AddAttr<float>("step_w",
                    "Prior boxes step across width, 0 for auto calculation.")
-        .SetDefault(0.0);
+        .SetDefault(0.0)
+        .AddCustomChecker([](const float& step_w) {
+          PADDLE_ENFORCE_GT(step_w, 0.0, "step_w should be larger than 0.");
+        });
     AddAttr<float>("step_h",
                    "Prior boxes step across height, 0 for auto calculation.")
-        .SetDefault(0.0);
+        .SetDefault(0.0)
+        .AddCustomChecker([](const float& step_h) {
+          PADDLE_ENFORCE_GT(step_h, 0.0, "step_h should be larger than 0.");
+        });
+
     AddAttr<float>("offset",
                    "(float) "
                    "Prior boxes center offset.")

paddle/operators/prior_box_op.h

@@ -25,7 +25,7 @@ inline void ExpandAspectRatios(const std::vector<float>& input_aspect_ratior,
                                std::vector<float>& output_aspect_ratior) {
   constexpr float epsilon = 1e-6;
   output_aspect_ratior.clear();
-  output_aspect_ratior.push_back(1.);
+  output_aspect_ratior.push_back(1.0f);
   for (size_t i = 0; i < input_aspect_ratior.size(); ++i) {
     float ar = input_aspect_ratior[i];
     bool already_exist = false;
@@ -38,7 +38,7 @@ inline void ExpandAspectRatios(const std::vector<float>& input_aspect_ratior,
     if (!already_exist) {
       output_aspect_ratior.push_back(ar);
       if (flip) {
-        output_aspect_ratior.push_back(1. / ar);
+        output_aspect_ratior.push_back(1.0f / ar);
       }
     }
   }
@@ -46,7 +46,7 @@ inline void ExpandAspectRatios(const std::vector<float>& input_aspect_ratior,
 
 template <typename T>
 struct ClipFunctor {
-  HOSTDEVICE T operator()(T in) const {
+  HOSTDEVICE inline T operator()(T in) const {
     return std::min<T>(std::max<T>(in, 0.), 1.);
   }
 };
@@ -97,6 +97,9 @@ class PriorBoxOpKernel : public framework::OpKernel<T> {
     boxes->mutable_data<T>(ctx.GetPlace());
     vars->mutable_data<T>(ctx.GetPlace());
 
+    T inv_img_width = 1.0 / img_width;
+    T inv_img_height = 1.0 / img_height;
+
     auto e_boxes = framework::EigenTensor<T, 4>::From(*boxes);
     for (int h = 0; h < feature_height; ++h) {
       for (int w = 0; w < feature_width; ++w) {
@@ -109,13 +112,15 @@ class PriorBoxOpKernel : public framework::OpKernel<T> {
           // first prior: aspect_ratio = 1, size = min_size
           box_width = box_height = min_size;
           // xmin
-          e_boxes(h, w, idx, 0) = (center_x - box_width / 2.) / img_width;
+          e_boxes(h, w, idx, 0) = (center_x - box_width * 0.5) * inv_img_width;
           // ymin
-          e_boxes(h, w, idx, 1) = (center_y - box_height / 2.) / img_height;
+          e_boxes(h, w, idx, 1) =
+              (center_y - box_height * 0.5) * inv_img_height;
           // xmax
-          e_boxes(h, w, idx, 2) = (center_x + box_width / 2.) / img_width;
+          e_boxes(h, w, idx, 2) = (center_x + box_width * 0.5) * inv_img_width;
           // ymax
-          e_boxes(h, w, idx, 3) = (center_y + box_height / 2.) / img_height;
+          e_boxes(h, w, idx, 3) =
+              (center_y + box_height * 0.5) * inv_img_height;
 
           idx++;
           if (max_sizes.size() > 0) {
@@ -124,13 +129,17 @@ class PriorBoxOpKernel : public framework::OpKernel<T> {
             // size = sqrt(min_size * max_size)
             box_width = box_height = sqrt(min_size * max_size);
             // xmin
-            e_boxes(h, w, idx, 0) = (center_x - box_width / 2.) / img_width;
+            e_boxes(h, w, idx, 0) =
+                (center_x - box_width * 0.5) * inv_img_width;
             // ymin
-            e_boxes(h, w, idx, 1) = (center_y - box_height / 2.) / img_height;
+            e_boxes(h, w, idx, 1) =
+                (center_y - box_height * 0.5) * inv_img_height;
             // xmax
-            e_boxes(h, w, idx, 2) = (center_x + box_width / 2.) / img_width;
+            e_boxes(h, w, idx, 2) =
+                (center_x + box_width * 0.5) * inv_img_width;
             // ymax
-            e_boxes(h, w, idx, 3) = (center_y + box_height / 2.) / img_height;
+            e_boxes(h, w, idx, 3) =
+                (center_y + box_height * 0.5) * inv_img_height;
             idx++;
           }
 
@@ -143,13 +152,17 @@ class PriorBoxOpKernel : public framework::OpKernel<T> {
             box_width = min_size * sqrt(ar);
             box_height = min_size / sqrt(ar);
             // xmin
-            e_boxes(h, w, idx, 0) = (center_x - box_width / 2.) / img_width;
+            e_boxes(h, w, idx, 0) =
+                (center_x - box_width * 0.5) * inv_img_width;
             // ymin
-            e_boxes(h, w, idx, 1) = (center_y - box_height / 2.) / img_height;
+            e_boxes(h, w, idx, 1) =
+                (center_y - box_height * 0.5) * inv_img_height;
             // xmax
-            e_boxes(h, w, idx, 2) = (center_x + box_width / 2.) / img_width;
+            e_boxes(h, w, idx, 2) =
+                (center_x + box_width * 0.5) * inv_img_width;
             // ymax
-            e_boxes(h, w, idx, 3) = (center_y + box_height / 2.) / img_height;
+            e_boxes(h, w, idx, 3) =
+                (center_y + box_height * 0.5) * inv_img_height;
             idx++;
           }
         }

paddle/operators/while_op.cc

@@ -53,6 +53,8 @@ class WhileOp : public framework::OperatorBase {
     auto step_scopes =
         scope.FindVar(Output(kStepScopes))->GetMutable<StepScopeVar>();
 
+    PADDLE_ENFORCE(platform::is_cpu_place(cond.place()),
+                   "Condition of while op must in CPU memory.");
     while (cond.data<bool>()[0]) {
       auto &current_scope = scope.NewScope();
       step_scopes->push_back(&current_scope);
@@ -99,6 +101,9 @@ class WhileGradOp : public framework::OperatorBase {
 
   void Run(const framework::Scope &scope,
            const platform::Place &dev_place) const override {
+    // get device context from pool
+    platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
+    auto &dev_ctx = *pool.Get(dev_place);
     framework::Executor executor(dev_place);
     auto *block = Attr<framework::BlockDesc *>(kStepBlock);
     auto *program = block->Program();
@@ -205,6 +210,8 @@ class WhileGradOp : public framework::OperatorBase {
         sum_op->Run(cur_scope, dev_place);
         cur_scope.Rename(new_inside_name, inside_grad_name);
       }
+      dev_ctx.Wait();
+      const_cast<framework::Scope &>(scope).DeleteScope(&cur_scope);
     }
   }
 };

paddle/platform/profiler.cc

@@ -233,7 +233,7 @@ void ParseEvents(std::vector<std::vector<Event>>& events,
       };
       break;
     default:
-      sorted_domain = "event end time";
+      sorted_domain = "event first end time";
   }
 
   std::vector<std::vector<EventItem>> events_table;

paddle/pybind/protobuf.cc

@@ -220,11 +220,20 @@ void BindVarDsec(py::module &m) {
            py::return_value_policy::reference)
       .def("set_name", &VarDesc::SetName)
       .def("set_shape", &VarDesc::SetShape)
+      .def("set_shapes", &VarDesc::SetShapes)
       .def("set_dtype", &VarDesc::SetDataType)
-      .def("shape", &VarDesc::Shape, py::return_value_policy::reference)
+      .def("set_dtypes", &VarDesc::SetDataTypes)
+      .def("set_tensor_num", &VarDesc::SetTensorDescNum)
+      .def("tensor_num", &VarDesc::GetTensorDescNum)
+      .def("shape", &VarDesc::GetShape, py::return_value_policy::reference)
+      .def("shapes", &VarDesc::GetShapes, py::return_value_policy::reference)
       .def("dtype", &VarDesc::GetDataType, py::return_value_policy::reference)
+      .def("dtypes", &VarDesc::GetDataTypes, py::return_value_policy::reference)
       .def("lod_level", &VarDesc::GetLoDLevel)
+      .def("lod_levels", &VarDesc::GetLoDLevels,
+           py::return_value_policy::reference)
       .def("set_lod_level", &VarDesc::SetLoDLevel)
+      .def("set_lod_levels", &VarDesc::SetLoDLevels)
       .def("type", &VarDesc::GetType)
       .def("set_type", &VarDesc::SetType)
       .def("serialize_to_string", SerializeMessage<VarDesc>)
@@ -239,7 +248,8 @@ void BindVarDsec(py::module &m) {
       .value("STEP_SCOPES", proto::VarDesc::STEP_SCOPES)
       .value("LOD_RANK_TABLE", proto::VarDesc::LOD_RANK_TABLE)
       .value("LOD_TENSOR_ARRAY", proto::VarDesc::LOD_TENSOR_ARRAY)
-      .value("PLACE_LIST", proto::VarDesc::PLACE_LIST);
+      .value("PLACE_LIST", proto::VarDesc::PLACE_LIST)
+      .value("READER", proto::VarDesc::READER);
 }
 
 void BindOpDesc(py::module &m) {

paddle/scripts/docker/build.sh

@@ -79,6 +79,7 @@ function run_build() {
     Building in /paddle/build ...
     ============================================
 EOF
+    make clean
     make -j `nproc`
 }
 
@@ -116,7 +117,7 @@ EOF
             -DWITH_STYLE_CHECK=OFF
         make -j `nproc` gen_proto_py
         make -j `nproc` paddle_python
-        make -j `nproc` paddle_docs paddle_docs_cn
+        make -j `nproc` paddle_docs paddle_docs_cn paddle_api_docs
         make -j `nproc` print_operators_doc
         paddle/pybind/print_operators_doc > doc/en/html/operators.json
         popd

paddle/scripts/travis/build_doc.sh

@@ -9,13 +9,14 @@ cd $TRAVIS_BUILD_DIR/build
 cmake .. -DCMAKE_BUILD_TYPE=Debug -DWITH_GPU=OFF -DWITH_MKL=OFF -DWITH_DOC=ON
 make -j `nproc` gen_proto_py
 make -j `nproc` paddle_python
-make -j `nproc` paddle_docs paddle_docs_cn
+make -j `nproc` paddle_docs paddle_docs_cn paddle_api_docs
 make -j `nproc` print_operators_doc
 paddle/pybind/print_operators_doc > doc/en/html/operators.json
 
 # check websites for broken links
 linkchecker doc/en/html/index.html
 linkchecker doc/cn/html/index.html
+linkchecker doc/api/en/html/index.html
 
 # Parse Github URL
 REPO=`git config remote.origin.url`
@@ -54,10 +55,11 @@ function deploy_docs() {
   mkdir -p ${DIR}
   # remove old docs. mv new docs.
   set +e
-  rm -rf ${DIR}/doc ${DIR}/doc_cn
+  rm -rf ${DIR}/doc ${DIR}/doc_cn ${DIR}/api_doc
   set -e
   cp -r ../doc/cn/html ${DIR}/doc_cn
   cp -r ../doc/en/html ${DIR}/doc
+  cp -r ../doc/api/en/html ${DIR}/api_doc
   git add .
 }
 

python/paddle/v2/fluid/tests/book/.gitignore

+recognize_digits_*.inference.model