* softmax mask fuse upper triangle

* cover not implemented cpu code

* remove check on kwargs

* refine code, reuse commom function

* Use CBLAS for SelectedRows elementwise add operation. It's faster.

* template compilation fix

* reverted template compilation fix

* slimmed template compilation fix
Co-authored-by: NAdam Osewski <adam.osewski@intel.com>

* opt dygraph python code, test=develop

* refine, test=develop

* fix vector overflow

* refine code

* refine ut

* PTQ save quantized model

* Wrap simulated layer

* post process the inference model

* correct conditions of gather in opteller

* test=develop

* test=allcase

* fix error for no python coverage data

* delete the function of saving layer object.

* edit doc of paddle.save/load and polish error message

* opt dygraph python API, test=develop

* Fix unbind bug in manipulation.py

* add num_iters in fit/evalate, test=develop

* Fix test_jit_save_load random failure.

* Since CI is not activated, recommit the code.

* delete temp file.

* NpuOpRunner supports host tensor as input

* fix compile issue

* added prelu bf16/fp32 fwd/bwd kernel

* added checking if md uses blocking format

* minor change

* removed unnecessary line

* add delta score, scale show

* so parser

* windows

* windows

* modify logic

* test=allcase

* test=document_fix

* add gpu implementation of shuffle batch
test=develop

* add thrust cuda patches
test=develop

* fix macro guard

* fix shuffle batch compile on windows/hip

* fix hip compilation error

* refine CMakeLists.txt

* fix windows compile error

* try to fix windows CI compilation error

* fix windows compilation again

* fix shuffle_batch op test on Windows