workaround for no MKLDNN kernel

Update dist lookup table design doc

bugfix/libpaddle_inference_api.so double free

release branch in the futrue

Adding corresponding hl_fini() for each hl_init() used

Remove Op::Clone method

Fix data balance on single GPU

check return value of patchelf in setup.py.in

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into update-dist-lookup-table-design-doc

add inference_transpiler doc to develop branch

* use -rpath to fix libmklml_intel.so not found

* only remain $ORIGIN/../libs in rpath of core.so

* test

* test

* add rpath of libmkldnn.so.0

* check return value of os.system

* remove check return value of patchelf

Change default APISpec URL

Do NOT clone input vars if op NOT in pserver_program

Add template specialization of Sprintf

move SetNumThreads to platform

fix conflicts

* move to platform

* "move init from framework to platform"

* "remove used init"

* "fix ci"

* "fix ci"

* "fix generic"

* "fix ci"

* "fix ci"

* "fix ci"

* "disable fragile test"

Add design doc for lookup remote table in Fluid

* add_nccl2_dist_design

* update

* update by comments

Check API changes

This reverts commit 36d9e4be.