Update from the original

* fix CUDNN_VERSION for backward of CudnnBatchNormLayer

* fix cudnn version number for BatchNorm

Update from the original

Custom stack trace refines

Lazy install Paddle wheels

Fix a PyDataProvider2 bug when use calc_batch_size

* Need PyGuard when set args for calc_batch_size

Change-Id: I2a72eb0a296a8cf297e2a53029d2602d85f40296

Merge branch 'custom_stack_trace_refines' of https://github.com/reyoung/Paddle into custom_stack_trace_refines

* Make layer stack trace shows ThreadId, Forward or Backward.

Change-Id: Iba1477adb8c9115c3a67ff2959bb5c878ca706c7

* Replace DataProviderWrapperConverter with DataProviderConverter in all
demo prediction scripts.

* Minor refinement of swig_py_paddle.rst to keep consistency with the
English version.

Fix bug in NO_AVX when using lstm

* libpaddle_cuda add wrong -mavx to some cpp files.

* Make layer stack trace shows ThreadId, Forward or Backward.

Change-Id: Iba1477adb8c9115c3a67ff2959bb5c878ca706c7

* install wheels when invoke paddle script if current python don't
  have paddle packages, or installed a previous version.
* Also add `make install` to travis

Merge Baidu Changes into github

Change-Id: I028e402c964ca4f4431cbf8153bea4379dd4df70

Change-Id: Ia84860cdc25945ececba84fc9807495c9e5f047b

DataProviderWrapperConverter with DataProviderConverter.

Change-Id: Ibc39ca1d1a27d0d28569e29f41a5647659f8c764

The reference return type causes ThreadParameterUpdater.cpp:123 seg fault under gcc5.4.

Change-Id: I7a1c155892722076a7cb48793b83d5ee525747d1

Change-Id: I0b8608feab8f6be5094e8981fc5f65cb401ed415

Change-Id: I636cf13af5becb1168bc9749266b55580c46f6c9

* the last layer in the stack already has all the suffixes

Change-Id: Ia60c008a8c922f66e6b5e2ca3e488fc4625d6506