* Optimize softmax. When the input tensor is 2-D and axis is 1, there is no need to resize.

* Optimize the gru, avoid calling Tensor::Slice.
test=develop

* Remove a std::vector in softmax.
test=develop

* Define CalculateSeqWidth to get the width of a sequence.
test=develop

* replace gru RowWiseAdd Eigen with c implementation
test=develop

* Optimize the transform from Paddle' Tensor to EigenVector, avoiding defining multiple DDim.

* Optimize the compute implementation of several operators.
test=develop

-fix fluid-lite-subgraph x86 compile error
    - Replace FLAGS with environment variables

* Remove GEMM padding in fc_compute.
test=develop

* Write a common ParallelFor function to run the for loop in parallel.

* Add the codes of padding GEMM back in fc.

* Refine the code of fc when padding_weight is false to avoid the definition of temporary Tensor.

* Refine the unit test of fc and add testing case of padding and parallel.
test=develop

* Enable more test cases in common fc unittest, including padding and parallel for x86 target.

* Remove the fc test under kernels/x86.
test=develop

* Disable relu in test of fc for non-x86 target.
test=develop

* Change the eps of arm.
test=develop

* fix pooling bug and speed

* fix build error

* delete VLOGin pool, test=develop

* add openmp, test=develop

* fix lite/kernels/arm/pool_compute_test basic_pooling compute error bug, test=develop

* update pooling 2-pad to 4-pad, test=develop

* fix 2-pad to 4-pad in operators/pool_op.h, AttachKernel will set param, so 2-pad to 4-pad funcs should put in AttachKernel. test=ddevellop

* put 2-pad to 4-pad in AttachImpl, test=develop

* according to reviews, fix some format error. test=develop

* fix format errorr, add (). test=develop

* change paddings type to support dynamically modify, test=develop

* update padding type int other devices, test=develop

* fix x8d build error on shared_ptr, test=ddevelop

* fix formmat in operators pool_op.cc, test=develop

* add x86 op and kernel : search_fc and sequence_topk_avg_pooling   for content-dnn model test=develop

fix jit::matmul bug. Input x shape is (m, k), weight shape is (k, n). When k < 512, m==1, and n is a multiple of 16, the weight pointer is not correctly updated in the group calculation in the implementation of jit::matmul, resulting in the result diff 

* add tanh op in x86 platform and its unittest, test=develop

* add gelu op on x86 platform and add its unittests, test=develop

* update depends for math_function for activation for gelu, test=develop

* enable pool2d op on x86 and add its unit tests, test=develop

* enable dropout op and add its unit tests, test=develop

* add tranpose, transpose2 op and add their unit tests, test=develop

* fix asr modle related kernel bugs test=develop

* add full_api_static target and fix building errors, test=develop

* fix build errors, test=develop

* fix code style, test=develop

* fix lite/model_parser/pb/var_desc.cc, test=develop

* fix building errors, test=develop

* modify lite/tools/debug/CMakeLists.txt, test=develop

* fix building model_optimize_tool error on mac, test=develop

* fix model_optimize_tool build error, test=develop

* add x86 gru&&relu&&sequence_expand_as op test=develop

add math function: lstm  and selected_rows into lite/x86/math
add selected_rows and rw_lock into lite/fluid
add lstm_cpu_kernel and  lstm_kernel into lite/x86/detail

* add lite x86 ops for ASR test=develop

* add lite x86 ops for ASR test=develop

* fix x86 ci run test problems test=develop

* fix mkl path for CI test=develop