* add scalar by tensor op

* scalar xxx by tensor

* rm useless

* fix

* rm useless code

* support int64

* fix

* add test ofrecord decoder

* int32, int64, float, double OK, bytes not ready

* add bytes_list codec

* char to int8 work

* bytes_list test OK

* check in updates for ubuntu buid

* 3rd praty ok

* fix py cmd

* make py3 default

* fix opencv flag

* dont build test

* rm debug lines

* fix minor vsn

* use 10.0

* fix vsn

* rm proxy

* fix libcublas_static

* fix cublas

* use 10.1

* compability

* fix docker file and script

* force 10.0

* rm build arg

* revert

* fix cublas link

* rm proxy arg

* copy_comm_net pass when regst dynamic data is empty

* refine name and remove useless back insert in blob; and unittest scripts

* dynamic binary split/concat op

* dynamic binary split kernel implement

* dynamic binary concat kernel implment

* more strict check shape and add log

* Regster::GetSoleBlob()

* half implement of dynamic binary python wrapper

* fix bug and pass test

* add test scripts and refine python wrapper

* rm log

* mut contiguous header ptr and fix format

* constant like op

* sqrt grad

* square grad

* fix

* fix

* add test case

* format

* fix rsqrt op, rm epsilon

* fix

* test case

* rm empty

* unsorted_segment_sum_kernel_util

* unsorted_segment_sum test_case

* unsorted_segment_sum_kernel_util

* UnsortedSegmentSumKernel use UnsortedSegmentSumKernelUtil

* gather_ms0_grad_kernel

* gather data_type_seq

* checkout nn_ops.py from dev_sparse_cross_entropy

* rm old sigmoid cross entropy with logits

* add test_sparse_softmax_cross_entropy.py

* add broadcast logical binary op

* add broadcast_logical_kernel.cpp

* add broadcast op conf

* ndarray patch

* add test

* copy_comm_net pass when regst dynamic data is empty

* refine name and remove useless back insert in blob; and unittest scripts

* more strict check shape and add log

* Regster::GetSoleBlob()

* Support Gelu half

* Remove gelu half kernel of old complicated version

* Refactor the impl of gelu float/half kernel

* indexed_slices_reduce_sum_op

* indexed_slices_reduce_sum_kernel_util

* indexed_slices_reduce_sum_op_util

* indexed_slices_reduce_sum_kernel

* python op and test_case

* op_conf

* op

* of_format

* kernel util

* empty kernel

* unique_with_counts

* py

* gpu pass

* refine

* iter

* fix type seq

* fix

* cpu pass

* remove check kernel

* op_conf

* op

* of_format

* kernel util

* empty kernel

* unique_with_counts

* py

* gpu pass

* refine

* iter

* fix type seq

* fix

* cpu pass

* add add_n in math_ops.py

* add_n bug fix

* add test_add_n.py

* rm usless debug codes

* check inputs lenght > 1 for add_n

* rm uselss lines

* rm r = None