* remove "gpu_primitives.h" in fluid namespace

* fix PR-CI-GpuPS fail

* fix PR-CI-GpuPS fail

* add vectorized bfloat16 atomicAdd

* fix compile error

* fix compile error again

* fix V100 compile error

* fix V100 compile again

* support pure bfloat16

* support bf16 linear

* update PR to pass CI

* tiny fix where_grad_kernel.cu

* add bfloat16 to selu_grad to pass CI

* fix selu grad compilation error

* add init file

* add op definition and infermeta

* add kernel definition funcs

* add broadcast infer shape

* add gpu forward kernel

* delete SUB and DIV

* add x_grad

* add template

* add e_grad for min and max

* fix small bug

* temp commit

* temp commit

* add e_grad for sum and mean

* fix some compile bug

* fix compile bugs

* fix compile problem

* add sum forward unittest

* fix broadcast error, add kernel sig, register e_grad, change unit test

* fix grad

* add temp grad fix

* temp commit

* add min max unittest

* add max, min unittest, fix mul bug

* add cpu forward sum and mean

* add forward min max, fix mean unittest

* add cpu backward min max

* fix code-style

* add backward sum mean

* fix rocm ci

* set uniitest timeout

* fix bug of x broadcast to e, gpu grad

* fix bug of x broadcast to e, cpu grad

* rename BOOST_GET_CONST macro

* fix rocm ci

* mv graph_send_e_recv to graph_send_ue_recv

* move out_size to IntArray

* add eager op test

* fix max pool type bug, add unittest for api

* revise api doc

* add fp16 for atomic min and max, add unittest

* add unittest

* add fp16 support for graph_send_recv

* fix unittest fp16 bug

* change OutSizeTensor to Out_size

* move E to Y

* add copyright, fix comment

* review code

* fix thread block size

* fix thread block size

* change api attribute name: pool_type to reduce_op, compute_type to message_op

* change api attribute name, move pool_type to reduce_op, move compute_type to message_op

* add scale gather sum

* refine CUDA_ATOMIC_WRAPPER ADD for bf16

* add gather unittest

* solve conflict

* add scale uinttest

* add sum unittest

* solve conflict

* refine gather unittest

* refine unittest

* Fix compile error on cuda_arch less than 700.

* optimize block config  and fp16 atomicAdd perf for lookup_table_v2_grad.

* fix cuda atomicAdd for FP16

* try to fix ci

* refine structure for cuda and rocm

* update

* update

* update

* update

* add cpu version, using set: sum, min, max

* add cpu version: mean

* improve cpu code and fix dynamic memory allcation problem

* fix arg error, add index judge, delete fp16

* fix bug in CudaAtomicMax and CudaAtomicMin

* add CUDA version

* fix grad_op bug for index

* add op test, add correct cpu grad op

* Add correct CUDA Mean grad

* [Add] Successful MEAN and SUM

* [Add] Successful MIN and MAX in CPU

* [Add] Successful MIN and MAX in CUDA

* fix windows dtype ci

* fix ROCM ci by adding HIP flag

* rename fused_gather_scatter to send_recv

* unify name as send and recv

* change zero index return time

* add send_recv incubate api

* fix index data type, add unittest case for API

* delete redundant input tensor

* fix en example and docs, add default value in pool_type

* add shape judge and max grid judge

* fix comment

* fix index type bug

* add const &

* fix en docs

* delete numpy in examples

* add unittest for int input

* fix send_recv comment

* change send_recv to graph_send_recv

* replace and remove complex64/128 types in custom OP and other files

* fix custom_tensor_test fail bug

* fix custom_conj_test fail bug

* fix dispatch_test_op build fail bug

* add support for complex grad accumulated

* add unittest for coverage

* update test dtype

* remove useless blank line

fix cpplint error for the autmic max/min

fix cuda atomic for ARCH<350 for the automic_max

Add GPU Kernels of Segment Ops,  support, sum, max, min, mean

* "rewrite the test case"

* "follow comment"

* cherry picked

* "cherry picked platform"

* "add comment"

* "fix ci"

* fix __shfl_down_sync_ of cross_entropy

* use reduceSum

* "fix ci"

* "re-commit "

* "picked up"

* "fix ci"

* "fix pdb hang up issue in cuda 9"

* add cuda_device_functions.h

* move reduceSum to elementwise_op_function.h

* Support int64 for sum op

* Refine code

* refind accuracy_op

* follow comments

* follow comments

1. finish lookup table CPU and GPU kernel
2. Add some cuda helper
3. Add some math funtor