* update prior profile, test=develop

* fix review. test=develop

* test=develop

* fix opencl backend. test=develop

[ARM] [NPU] Fix the overflow of layer_norm op bridge, fix the registration of lstm op kernel (#4007) (#4017)

* [NPU] Fix the overflow of layer_norm op bridge
test=develop

* [ARM] Fix the registration of lstm op kernel
test=develop

* fix arm winograd compute segment. test=develop

 [Framework] Fix the issue that `CreatePaddlePredictor` method is not safe for multi-thread condition #4014

* add elu act

* fix elu act not find error, test=develop

* fix format.test=develop

* fix int8 model opt error in conv+conv fusion, test=develop

* fix format. test=develop

* test=develop

* add op affine_grid. test=develop

* fix format. test=develop

* change arm  transpose kernel. develop=test

* [LITE][BM] modify interpolate if,test=develop

* [LITE][BM] compiler opt=1 to opt=2, test=develop

* [ASCEND] update based on latest engine.h, test=develop

* fix opencl ci random failure, test=develop

* [ASCEND] Add Huawei Ascend310 support, test=develop

* [ASCEND] fix some typos, test=develop

* [ASCEND] address comments and fix opt ci python file, test=develop

* [ASCEND] update based on new ascend env, test=develop

* [ASCEND] update after develop merge, test=develop

* [Core][ARM] Fix beam_search, eltwise_mul supports broadcast and int64_t data type, add print op and kernel, add exeception
test=develop

* Fix the dims of parent idx of the arm kernel of beam_search op

* elementwise_mul supports int64_t data type with broadcasting

* Add print op and kernel for debugging

* Support throwing the exception when the internal error occurs

* Refine while and conditional_block op kernel

* Support the graph optimization on subblocks

* Pass program_desc and block_idx into the kernel of the control flow ops(while/conditional_block/subgraph), and create the RuntimeProgram online, it make it possiable to call the control flow ops recursively

*Add unit test for masked transformer model