* add elementwise max grad op for npu

* add elementwise max grad op for npu

* add elementwise max grad op for npu

* add elementwise max grad op for npu

* add elementwise max grad op for npu

* polish the error message of paddle.slice.

* polish code.

[oneDNN] disable caching oneDNN primitives in  matmul v2, Reduce grad and elementwise_add grad, expand_v2 (#35132)

* - grad caching disabled of matmul_v1

- compilation fix

- compilation fix

* - reduction removed

* - Matmul v2 disabled caching

* Draft of further changes

* - workaround for reducegrad

* - fixes to UT

* - fix to compilation

* - another fix

* - fix

* fix assign bug support fp16 uint8

* fix dygragh assign bool bug

* modify code style

* revoke bool modification

* Support dropout backward in eval mode

* add downscale case

* minor fix

* minor fix

* Modify into QueueSync QueueAsync

* fix complie on MacOS

* fix pointer

* fix conflict

* polish unittest

* fix windows fetch error

* polish code according reviewer

* fix device_guard on CPU place

Describe

Add feed_forward for fused attention op.
(1) Encapsulate matmul impl (forward and backward) used in attention op.
(2) Implement bias_add (forward and backward) used in attention op.

* add_roi_align_npu

* update

* update

* update

* temporary change

* fix for expand_v2

* changes after review, activated ppyolov inference test

* add fetch, test=develop

* fix fetch2op, test=develop

* fix fetch2op, test=develop

* refine, test=develop

* fix fetch ctx, test=develop

* add wait, test=develop

* rename fetch2 to fetch_v2, test=develop

* merge, test=develop

* fix bmm bug

* bmm style

* fix bmm

* - concat refactoring draft

* - cmpilation fixes

* - yet another compilation fix

* - fix

* - compilation fix

* - fixes to compilation

* - another compilation fix

* - fix

* - Added overloaded AcquirePrimitiveDesc for concat

* - fix

* - reserve introduced

* - UT fixes

* - test concat int8 improved

* - fixes

* - fix to crash

* - lint fixes

* - fixes after review

* - some other fixes from review

* add conv_op_npu and test

* add more tests

* clean headers & support fp16

* update

* add pool2d_op_npu and test

* update

* update pool2d_backward_navie

* clean headers

* - disabled interpolate onednn

* - compilation fix

* - draft of batch_norm cache disabling

* - fixes to UT

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.

* Support getitem by Bool index

* delete some debug info of bool index

* support the case that the shape of bool index is different from indexed tensor

* adamw support cuda

* adamw support cuda

* add rmsprop npu

* add argsort npu

* add argsort npu

* modify according to review

* modify sharedatawith according to review

* modify reshape according to review

* rm dygraph=false

* [NPU] Support npu kernel for pad3d op

* fix for comment of zhouwei25

* fix some bugs according to qili93's comments

* add support and test for paddings in input

* delete VLOG used for debug

* add depthwise_conv2d npu

* add some tests

* Delete test_unique_op_npu.py

* delete trans input

* [NPU] Support npu op where and where grad

* fix use const_cast

* delete a test

* add (N,C,*) input support for GroupNorm

* --amend