* Fix gather_op by adding OurOfRangeCheck for param[Index]

* Code Optimization

* add npu operator for gaussian random.

* bugfix: add wait after memory copy.

* update gaussian random op: use TensorCopy.

* add dropout npu op

* fix bugs

* add unittest

* fix bugs

* support 1-D input

* change ScatterAdd to EmbeddingDenseGrad in lookup_table NPU op

* EmbeddingDenseGrad only supports dim 32

* fix shape error

* softmax mask fuse upper triangle

* cover not implemented cpu code

* 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>

* fix vector overflow

* refine code

* refine ut

* 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 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

* fix interpolate for shape[i]=0, test=develop

* fix test_trilinear_interp_v2 random failure, test=develop

* change Add to AddN in sum npu op

* add AddInputNames

* change fp16 to fp32 because numpy has accuracy loss in fp16 adding

* delete check

* fix runner error

* enhance error message when x or y is empty in elementwise_op

* format code

* format code

* delete useless GELU in gelu npu op

* add description

* fix format

* add check_grad in gelu unittest