* add fleet api for program pass

* turn on apply pass for CI test

* fix disable fuse_all_optimizer bug

* try to test ci

* fix CI

* fill unspecified op role

* fix fuse_allreduce

* add ut to improve coverage

* remove useless change

* improve c++ coverage

* follow some comments

* test ir pass pipeline

* update doc

* reduce ut time again

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.

* update, test=develop

* update, test=develop

* merge CMakeList.txt manual

* add platform for changethreadnum

* repair some bugs according to make error

* do nothing just flush CI

* forget change thread num

* add inplace_atol param for check_output_with_place

* Windows

* std:min and std::max should be change because of windows

* support weight sharing

* release gil before op run

* support npu grad test

* fix op_test

* add the matmul v2 grad kernel

* relief the test case time

* update the test case for the matmul double grad

* remove the unsed code for the matmul double grad

* update the test case for the double grad matmul

* remove the unused code in dot

* fix scatter_nd_add doc, test=document_fix

* update
test=document_fix

* add conv op check for illegal input or attributes

* [NPU] update batch norm op, test=develop

* add NHWC support for bn, test=develop

Co-authored-by: Nroot <root@bjyz-sys-gpu-kongming9.bjyz.baidu.com>

* Add DPADDLE_WITH_CUDA for GCC

* polish code

* [NPU] fix for test_norm_op_npu

* [NPU] add norm_grad

* [NPU] add CheckAxis for axis

* [NPU] delete debug codes

* norm can not use L2Normalize, norm_grad can use L2NormalizeGrad

* [NPU] delete useless codes

* [NPU] optimize norm_grad OpMaker

* Update python import path

* [NPU] log_softmax_grad, test=develop

* remove debug files, test=develop

* update lookup_table_v2 for CANN 5.0.x, test=develop

* fix for reshape2

* added reviewers sugestions

* add AsExtra in data_norm op

* pass data_layout from python to data_norm op

* fix data_layout in data_norm op
Co-authored-by: Nroot <root@bjyz-sys-gpu-kongming9.bjyz.baidu.com>

* fix commit

* Open unittest

* fix unittest on Windows

* fix constructor

* open test_resnet_amp on Windows

* disable on Windows CPU CI for timeout

* disable on Windows CPU CI for timeout

* fix code style

* transfer the static.accurcay to v2 api

* remove the unused code

* [HIP] fix op not support AMD GPU bug, the flag PADDLE_WITH_ROCM is invalid

* [HIP] fix op not support AMD GPU bug, the flag PADDLE_WITH_ROCM is invalid

* [HIP] fix op not support AMD GPU bug

* support double in deformable conv

* add double for dcn v2

* Add fusion_lstm INT8 PTQ

* Correct mkldnn_cache_capacity and enable fc_lstm_fuse_pass only for this test

* Change mkldnn_cache_capacity

* add pool2d grad grad

* dbg

* add unittest

* update format

* add more unittests

* dbg

* add kernel, stride check

* add unitest for param out of range

* delete max limit check

* add depthwise_conv_npu_grad op

* add depthwise_conv_npu_grad op

* add depthwise_conv_npu_grad op

* add NHWC test case

* support numpy dtype and polish code of list index.

* polish code.

This PR adds error exception in fused transformer python interface.
The function body are not implemented (will be implemented later).
Following zhiqiu's comment in previous PR-35206 (merged already), it is better to raise an exception instead of using "pass".

* where op add broadcast, when x_shape == y_shape, and x_shape != cond_shape

* add static api tests, and delete debug codes

* support lodtensorarray

* [NPU] Add huber_loss op

* [NPU] Add huber_loss op

* [NPU] Add huber_loss p[

* [NPU] Add huber_loss

* [NPU] add int64 kernels, test=develop

* update ci scripts to be able to trun WITH_ASCEND_INT64 on, test=develop