Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into add-merge-splited-ids

benchmark/fluid/fluid_benchmark.py

@@ -180,7 +180,7 @@ def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
         print_train_time(start_time, time.time(), num_samples)
         print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses))),
         # evaluation
-        if not args.no_test and batch_acc:
+        if not args.no_test and batch_acc and not args.use_reader_op:
             pass_test_acc = test(exe, infer_prog, test_reader, feeder,
                                  batch_acc)
             print(", Test Accuracy: %f" % pass_test_acc)
@@ -277,11 +277,12 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
             batch_id += 1
 
         print_train_time(start_time, time.time(), num_samples)
-        if not args.no_test and batch_acc:
+        if not args.no_test and batch_acc and not args.use_reader_op:
+            # we have not implement record io for test
+            # skip test when use args.use_reader_op
             test_acc = test(startup_exe, infer_prog, test_reader, feeder,
                             batch_acc)
             print("Pass: %d, Test Accuracy: %f\n" % (pass_id, test_acc))
-        exit(0)
 
 
 def print_arguments(args):

benchmark/fluid/models/resnet.py

@@ -199,7 +199,10 @@ def get_model(args):
     batched_train_reader = paddle.batch(
         paddle.reader.shuffle(
             train_reader, buf_size=5120),
-        batch_size=args.batch_size * args.gpus)
-    batched_test_reader = paddle.batch(train_reader, batch_size=args.batch_size)
+        batch_size=args.batch_size * args.gpus,
+        drop_last=True)
+    batched_test_reader = paddle.batch(
+        train_reader, batch_size=args.batch_size, drop_last=True)
 
-    return avg_cost, inference_program, optimizer, batched_train_reader, batched_test_reader, batch_acc
+    return avg_cost, inference_program, optimizer, batched_train_reader,\
+                   batched_test_reader, batch_acc

doc/fluid/api/gen_doc.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-python gen_doc.py layers --submodules control_flow device io nn ops tensor > layers.rst
+python gen_doc.py layers --submodules control_flow device io nn ops tensor detection > layers.rst
 
 for module in data_feeder clip metrics executor initializer io nets optimizer param_attr profiler regularizer
 do

doc/fluid/api/io.rst

@@ -59,21 +59,3 @@ get_inference_program
 ..  autofunction:: paddle.fluid.io.get_inference_program
     :noindex:
 
-save_checkpoint
----------------
-
-..  autofunction:: paddle.fluid.io.save_checkpoint
-    :noindex:
-
-load_checkpoint
----------------
-
-..  autofunction:: paddle.fluid.io.load_checkpoint
-    :noindex:
-
-clean_checkpoint
-----------------
-
-..  autofunction:: paddle.fluid.io.clean_checkpoint
-    :noindex:
-

doc/fluid/api/layers.rst

@@ -181,12 +181,6 @@ Print
 ..  autofunction:: paddle.fluid.layers.Print
     :noindex:
 
-is_empty
---------
-
-..  autofunction:: paddle.fluid.layers.is_empty
-    :noindex:
-
 device
 ======
 
@@ -261,19 +255,6 @@ double_buffer
 ..  autofunction:: paddle.fluid.layers.double_buffer
     :noindex:
 
-random_data_generator
----------------------
-
-..  autofunction:: paddle.fluid.layers.random_data_generator
-    :noindex:
-
-Preprocessor
-------------
-
-..  autoclass:: paddle.fluid.layers.Preprocessor
-    :members:
-    :noindex:
-
 nn
 ==
 
@@ -613,30 +594,6 @@ roi_pool
 ..  autofunction:: paddle.fluid.layers.roi_pool
     :noindex:
 
-dice_loss
----------
-
-..  autofunction:: paddle.fluid.layers.dice_loss
-    :noindex:
-
-resize_bilinear
----------------
-
-..  autofunction:: paddle.fluid.layers.resize_bilinear
-    :noindex:
-
-gather
-------
-
-..  autofunction:: paddle.fluid.layers.gather
-    :noindex:
-
-random_crop
------------
-
-..  autofunction:: paddle.fluid.layers.random_crop
-    :noindex:
-
 ops
 ===
 
@@ -784,12 +741,6 @@ sum
 ..  autofunction:: paddle.fluid.layers.sum
     :noindex:
 
-shape
------
-
-..  autofunction:: paddle.fluid.layers.shape
-    :noindex:
-
 sigmoid
 -------
 
@@ -1039,3 +990,54 @@ zeros
 ..  autofunction:: paddle.fluid.layers.zeros
     :noindex:
 
+detection
+=========
+
+multi_box_head
+--------------
+
+..  autofunction:: paddle.fluid.layers.multi_box_head
+    :noindex:
+
+bipartite_match
+---------------
+
+..  autofunction:: paddle.fluid.layers.bipartite_match
+    :noindex:
+
+target_assign
+-------------
+
+..  autofunction:: paddle.fluid.layers.target_assign
+    :noindex:
+
+detection_output
+----------------
+
+..  autofunction:: paddle.fluid.layers.detection_output
+    :noindex:
+
+ssd_loss
+--------
+
+..  autofunction:: paddle.fluid.layers.ssd_loss
+    :noindex:
+
+detection_map
+-------------
+
+..  autofunction:: paddle.fluid.layers.detection_map
+    :noindex:
+
+iou_similarity
+--------------
+
+..  autofunction:: paddle.fluid.layers.iou_similarity
+    :noindex:
+
+box_coder
+---------
+
+..  autofunction:: paddle.fluid.layers.box_coder
+    :noindex:
+

doc/fluid/api/optimizer.rst

@@ -89,13 +89,6 @@ DecayedAdagradOptimizer
     :members:
     :noindex:
 
-RMSPropOptimizer
-----------------
-
-..  autoclass:: paddle.fluid.optimizer.RMSPropOptimizer
-    :members:
-    :noindex:
-
 Adadelta
 --------
 

doc/fluid/api/profiler.rst

@@ -23,15 +23,3 @@ profiler
 ..  autofunction:: paddle.fluid.profiler.profiler
     :noindex:
 
-start_profiler
---------------
-
-..  autofunction:: paddle.fluid.profiler.start_profiler
-    :noindex:
-
-stop_profiler
--------------
-
-..  autofunction:: paddle.fluid.profiler.stop_profiler
-    :noindex:
-

paddle/fluid/framework/CMakeLists.txt

@@ -84,7 +84,7 @@ cc_library(lod_rank_table SRCS lod_rank_table.cc DEPS lod_tensor)
 cc_library(feed_fetch_method SRCS feed_fetch_method.cc DEPS lod_tensor scope glog)
 
 if(WITH_DISTRIBUTE)
-  cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method sendrecvop_grpc grpc++_unsecure grpc_unsecure gpr)
+  cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto glog lod_rank_table feed_fetch_method sendrecvop_grpc cares grpc++_unsecure grpc_unsecure gpr)
   set(DISTRIBUTE_COMPILE_FLAGS "-Wno-non-virtual-dtor -Wno-error=non-virtual-dtor -Wno-error=delete-non-virtual-dtor")
   set_source_files_properties(executor.cc PROPERTIES COMPILE_FLAGS ${DISTRIBUTE_COMPILE_FLAGS})
 else()

paddle/fluid/framework/reader.h

@@ -35,14 +35,15 @@ class ReaderBase {
 
 class DecoratedReader : public ReaderBase {
  public:
-  explicit DecoratedReader(ReaderBase* reader) : ReaderBase(), reader_(reader) {
+  explicit DecoratedReader(const std::shared_ptr<ReaderBase>& reader)
+      : ReaderBase(), reader_(reader) {
     PADDLE_ENFORCE_NOT_NULL(reader_);
   }
 
   void ReInit() override { reader_->ReInit(); }
 
  protected:
-  ReaderBase* reader_;
+  std::shared_ptr<ReaderBase> reader_;
 };
 
 class FileReader : public ReaderBase {
@@ -64,7 +65,7 @@ class ReaderHolder {
  public:
   void Reset(ReaderBase* reader) { reader_.reset(reader); }
 
-  ReaderBase* Get() const { return reader_.get(); }
+  std::shared_ptr<ReaderBase> Get() const { return reader_; }
 
   void ReadNext(std::vector<LoDTensor>* out) {
     PADDLE_ENFORCE_NOT_NULL(reader_);
@@ -76,7 +77,7 @@ class ReaderHolder {
   }
 
  private:
-  std::unique_ptr<ReaderBase> reader_;
+  std::shared_ptr<ReaderBase> reader_;
 };
 
 }  // namespace framework

paddle/fluid/operators/batch_norm_op.cc

@@ -110,19 +110,19 @@ class BatchNormOp : public framework::OperatorWithKernel {
                                          ctx.Input<Tensor>("Variance")->type()),
                       "Variance input should be of float type");
 
-    framework::LibraryType library_{framework::LibraryType::kPlain};
     // TODO(pzelazko-intel): enable MKLDNN layout when it's ready
+    framework::LibraryType library = framework::LibraryType::kPlain;
     framework::DataLayout layout = framework::DataLayout::kAnyLayout;
-
 #ifdef PADDLE_WITH_MKLDNN
-    if (library_ == framework::LibraryType::kPlain &&
+    if (library == framework::LibraryType::kPlain &&
         platform::CanMKLDNNBeUsed(ctx)) {
-      library_ = framework::LibraryType::kMKLDNN;
+      library = framework::LibraryType::kMKLDNN;
       layout = framework::DataLayout::kMKLDNN;
     }
 #endif
+
     return framework::OpKernelType(input_data_type, ctx.GetPlace(), layout,
-                                   library_);
+                                   library);
   }
 };
 
@@ -370,19 +370,21 @@ class BatchNormGradOp : public framework::OperatorWithKernel {
       PADDLE_THROW("can't find Y@GRAD");
     }
 
-    framework::LibraryType library_{framework::LibraryType::kPlain};
     // TODO(pzelazko-intel): enable MKLDNN layout when it's ready
-    framework::DataLayout layout_ = framework::DataLayout::kAnyLayout;
+    framework::LibraryType library = framework::LibraryType::kPlain;
+    framework::DataLayout layout = framework::DataLayout::kAnyLayout;
+
 #ifdef PADDLE_WITH_MKLDNN
-    if (library_ == framework::LibraryType::kPlain &&
+    if (library == framework::LibraryType::kPlain &&
         platform::CanMKLDNNBeUsed(ctx)) {
-      library_ = framework::LibraryType::kMKLDNN;
-      layout_ = framework::DataLayout::kMKLDNN;
+      library = framework::LibraryType::kMKLDNN;
+      layout = framework::DataLayout::kMKLDNN;
     }
 #endif
+
     return framework::OpKernelType(
         framework::ToDataType(ctx.Input<Tensor>("X")->type()), ctx.GetPlace(),
-        layout_, library_);
+        layout, library);
   }
 };
 

paddle/fluid/operators/reader/create_batch_reader_op.cc

@@ -20,7 +20,7 @@ namespace reader {
 
 class BatchReader : public framework::DecoratedReader {
  public:
-  BatchReader(ReaderBase* reader, int batch_size)
+  BatchReader(const std::shared_ptr<ReaderBase>& reader, int batch_size)
       : DecoratedReader(reader), batch_size_(batch_size) {
     buffer_.reserve(batch_size_);
   }

paddle/fluid/operators/reader/create_custom_reader_op.cc

@@ -22,7 +22,8 @@ namespace reader {
 
 class CustomReader : public framework::DecoratedReader {
  public:
-  CustomReader(ReaderBase* reader, const framework::BlockDesc& sub_block,
+  CustomReader(const std::shared_ptr<ReaderBase>& reader,
+               const framework::BlockDesc& sub_block,
                const std::vector<std::string>& source_var_names,
                const std::vector<std::string>& sink_var_names)
       : DecoratedReader(reader),

paddle/fluid/operators/reader/create_double_buffer_reader_op.cc

@@ -34,7 +34,8 @@ static constexpr size_t kChannelSize = 1;  // kCacheSize - 2
 class DoubleBufferReader : public framework::DecoratedReader {
  public:
   explicit DoubleBufferReader(
-      ReaderBase* reader, platform::Place target_place = platform::CPUPlace())
+      const std::shared_ptr<ReaderBase>& reader,
+      platform::Place target_place = platform::CPUPlace())
       : DecoratedReader(reader), place_(target_place) {
     cpu_tensor_cache_.resize(kCacheSize);
     gpu_tensor_cache_.resize(kCacheSize);

paddle/fluid/operators/reader/create_multi_pass_reader_op.cc

@@ -21,7 +21,7 @@ namespace reader {
 
 class MultiPassReader : public framework::DecoratedReader {
  public:
-  MultiPassReader(ReaderBase* reader, int pass_num)
+  MultiPassReader(const std::shared_ptr<ReaderBase>& reader, int pass_num)
       : DecoratedReader(reader), pass_num_(pass_num), pass_count_(0) {}
 
   void ReadNext(std::vector<framework::LoDTensor>* out) override {

paddle/fluid/operators/reader/create_shuffle_reader_op.cc

@@ -23,7 +23,8 @@ namespace reader {
 
 class ShuffleReader : public framework::DecoratedReader {
  public:
-  ShuffleReader(ReaderBase* reader, size_t buffer_size, size_t seed = 0)
+  ShuffleReader(const std::shared_ptr<ReaderBase>& reader, size_t buffer_size,
+                size_t seed = 0)
       : DecoratedReader(reader), buffer_size_(buffer_size), seed_(seed) {
     VLOG(10) << "Create shuffle reader of " << reader_;
     if (seed_ == 0) {

paddle/fluid/operators/reader/create_threaded_reader_op.cc

@@ -21,7 +21,8 @@ namespace reader {
 
 class ThreadedReader : public framework::DecoratedReader {
  public:
-  explicit ThreadedReader(ReaderBase* reader) : DecoratedReader(reader) {}
+  explicit ThreadedReader(const std::shared_ptr<ReaderBase>& reader)
+      : DecoratedReader(reader) {}
 
   void ReadNext(std::vector<framework::LoDTensor>* out) override {
     std::lock_guard<std::mutex> lock(mutex_);

paddle/fluid/platform/cpu_info.cc

@@ -21,12 +21,17 @@ limitations under the License. */
 #include <unistd.h>
 #endif
 
+#include <algorithm>
 #include "gflags/gflags.h"
 
 DEFINE_double(fraction_of_cpu_memory_to_use, 1,
               "Default use 100% of CPU memory for PaddlePaddle,"
               "reserve the rest for page tables, etc");
 
+DEFINE_uint64(
+    initial_cpu_memory_in_mb, 500,
+    "Default initial 500MB of CPU memory for PaddlePaddle, in MD unit.");
+
 DEFINE_double(
     fraction_of_cuda_pinned_memory_to_use, 0.5,
     "Default use 50% of CPU memory as the pinned_memory for PaddlePaddle,"
@@ -54,7 +59,10 @@ inline size_t CpuTotalPhysicalMemory() {
 size_t CpuMaxAllocSize() {
   // For distributed systems, it requires configuring and limiting
   // the fraction of memory to use.
-  return FLAGS_fraction_of_cpu_memory_to_use * CpuTotalPhysicalMemory();
+  return std::min(
+      static_cast<size_t>(FLAGS_fraction_of_cpu_memory_to_use *
+                          CpuTotalPhysicalMemory()),
+      static_cast<size_t>(FLAGS_initial_cpu_memory_in_mb * 1 << 20));
 }
 
 size_t CpuMinChunkSize() {

python/paddle/fluid/framework.py

@@ -382,7 +382,7 @@ class Operator(object):
         'rnn_memory_helper_grad', 'conditional_block', 'while', 'send', 'recv',
         'listen_and_serv', 'parallel_do', 'save_combine', 'load_combine',
         'ncclInit', 'channel_create', 'channel_close', 'channel_send',
-        'channel_recv', 'select'
+        'channel_recv', 'select', 'gen_nccl_id'
     }
 
     def __init__(self,

python/paddle/fluid/layers/nn.py

@@ -261,9 +261,10 @@ def embedding(input,
     return tmp
 
 
-# TODO(qijun): expose H0 and C0
 def dynamic_lstm(input,
                  size,
+                 h_0=None,
+                 c_0=None,
                  param_attr=None,
                  bias_attr=None,
                  use_peepholes=True,
@@ -324,6 +325,13 @@ def dynamic_lstm(input,
                          (T X 4D), where T is the total time steps in this
                          mini-batch, D is the hidden size.
         size(int): 4 * hidden size.
+        h_0(Variable): The initial hidden state is an optional input, default is zero.
+                       This is a tensor with shape (N x D), where N is the
+                       batch size and D is the hidden size.
+        c_0(Variable): The initial cell state is an optional input, default is zero.
+                       This is a tensor with shape (N x D), where N is the
+                       batch size. `h_0` and `c_0` can be NULL but only at the same time.
+
         param_attr(ParamAttr|None): The parameter attribute for the learnable
                                hidden-hidden weights.
 
@@ -387,12 +395,20 @@ def dynamic_lstm(input,
     cell = helper.create_tmp_variable(dtype)
     batch_gate = helper.create_tmp_variable(dtype)
     batch_cell_pre_act = helper.create_tmp_variable(dtype)
+    inputs = {'Input': input, 'Weight': weight, 'Bias': bias}
+    batch_size = input.shape[0]
+    if h_0:
+        assert h_0.shape == (batch_size, size), \
+            'The shape of h0 should be (batch_size, %d)' % size
+        inputs['H0'] = h_0
+    if c_0:
+        assert c_0.shape == (batch_size, size), \
+            'The shape of c0 should be (batch_size, %d)' % size
+        inputs['C0'] = c_0
 
     helper.append_op(
         type='lstm',
-        inputs={'Input': input,
-                'Weight': weight,
-                'Bias': bias},
+        inputs=inputs,
         outputs={
             'Hidden': hidden,
             'Cell': cell,
@@ -677,11 +693,13 @@ def dynamic_gru(input,
         attr=helper.param_attr, shape=[size, 3 * size], dtype=dtype)
     bias = helper.create_parameter(
         attr=helper.bias_attr, shape=[1, 3 * size], dtype=dtype, is_bias=True)
+    batch_size = input.shape[0]
     inputs = {'Input': input, 'Weight': weight, 'Bias': bias}
     if h_0 != None:
         assert h_0.shape == (
-            size, size), 'The shape of h0 should be(%d, %d)' % (size, size)
-        inputs['h0'] = h_0
+            batch_size, size
+        ), 'The shape of h0 should be(batch_size, %d)' % size
+        inputs['H0'] = h_0
 
     hidden = helper.create_tmp_variable(dtype)
     batch_gate = helper.create_tmp_variable(dtype)