功能：本PR的目标是提高attention模块的计算性能。
为了减少框架层对op的调度开销，本PR通过在C++层手动实现attention模块，对外提供attention 大op；
为了减少防存开销，本PR采取了两种优化方法：
（1）在q,k,v计算时通过共享输入X，将该处的gemm，transpose和bias add从三次调用减少为一次；
（2）使用kernel融合优化技术，在不同cuda kernel之间通过寄存器传输数据；

The bug is that access to mean and var is incorrect, and the array will be out of bounds: the shape of mean and var is [batch_size], and the range of thread idx is 0~feature_size, so mean[idx] and var[idx] is incorrect.

When batch_size=1, the correct access is mean[0] and var[0], and a unit test with batch_size=1 is added.

Refactor the organization of layer_norm cuda impl so that it can be reused in fused attention op.

    Extract the layer_norm cuda impl form layer_norm_op.cu to layer_norm_kernel.cu.h.
    Define fused/attention_layer_norm.h, which can be used in fused attention op in next PR.

* 1, remove layernorm dynamic fp16; 2, let reshape out in dynamic shape

* remove useless code

* Eliminate numerical differences of LayerNorm; fix LayerNorm Nan Bug while large data input

* fix bug while large shape of data input

* add dynamic layer_norm plugin

* fix bug

* fix numpy.allclose

* fix format

* fix code style

* remove shepe in dynamic shape

* code format

* remove layer norm fp16

* fix format

built-in `rsqrt` is shadowed

* fix compile problem when cuda_arch < 6000

* refine code

* refine code

* layernorm fw opt

* layernorm bw opt

* fix typo, test=develop

* remove const dim3 for windows CI compatibility

* merge develop
Co-authored-by: Nzlsh80826 <zlsh80826@gmail.com>

* add fp16 for layer_norm op

* revert layernorm api

* fix forward

* fix forward

* fix backward for layernorm with fp16

* fix unit test for layernorm with fp16

* fix with_mkldnn compile error for layernorm with fp16

* 1. revert to PADDLE_ENFORCE_NOT_NULL, 2. change static_cast<float> to static_cast<U>

* fix with_mkldnn compile error for layernorm with fp16

* fix with_mkldnn compile error for layernorm with fp16
Co-authored-by: Nzhiqiu <chenqiuliang@baidu.com>

* API/OP (group_norm, layer_norm, unpool) error message enhancement test=develop

restrict block num of layer_norm_grad cuda kernel to 128, test=develop

* add gelu plugin

* align trt bert with gpu

* add support for fused fc with relu,

* add unittest for bert trt