* nvcc_lazy

* nvcc_lazy

* conv_fusion

Modify the unittests of the assign_value, iou_similarity, one_hot_v2, reduce_mean, roi_align op. test=kunlun (#44061)

* make only win32 and 11.6 use external/cub

* unify cub version in linux and windows when cuda >= 11.6

* Revert "Revert "make inference_api_test compile with dynamic linking library (#41944)" (#43882)"

This reverts commit e6d81ddf.

* modify third_party cmake

* move SKIP_CPP_TEST return in the begining

* fix or backend many inputs bug

* fix or backend many inputs bug

* fix or backend many inputs bug

* fix or backend many inputs bug

* code format

* code format

* cmake-lint

* update

* Fix cmakelint errors for generic.cmake

* Fix cmake errors

* make all phi kernels to 2(host/device) static libraries directly

* fix calling kernel_declare

* fix compile

* fix cpu compile

* fix rocm compile

* fix xpu compile

* fix xpu kp compile

* fix inference compile

* modify paddle2onnx cmake

* modify paddle2onnx cmake

* modify export interface

* modify paddle2onnx export

* paddle2onnx add mac windows

* modify paddle2onnx mac windows cmake

* modify paddle2onnx mac windows cmake

* modify paddle2onnx cmake support windows

* modify paddle2onnx cmake support windows

* modify paddle2onnx cmake support windows
Co-authored-by: Nxiegegege <xiege01@baidu.com>

* back fl

* delete ssl cert

* .

* make warning

* .

* unittest paral degree

* solve unittest

* heter & multi cloud commm ready

* .

* .

* fl-ps v1.0

* .

* support N + N mode

* .

* .

* .

* .

* delete print

* .

* .

* .

* .

* update xpu cmake: xdnn 0527. test=kunlun

* update to xdnn 0531.

* update to xdnn 0531. test=kunlun

* update to xdnn 0601. test=kunlun

* experimental nvcc-lazy-module-loading

* remove two empty last line from two files

* fix maybe-uninitialized warning

* fix compile

* fix xpu compile

* fix npu compile

* fix infer compile

* fix compile

* fix compile

* [NPU] add take_along_axis and take_along_axis_grad ops

* [NPU] add take_along_axis and take_along_axis_grad ops

* fix ut because cpu kernel can not be fallbacked

* [NPU] add multinomial op

* fix place

* deal with cann version

* fix for old operator

* change another way

* Add the new XDNN implementation. test=kunlun

* Add the new XDNN implementation. test=kunlun

* Modify the code based on review, test=kunlun

* update xpu depends
*test=kunlun

* minor
*test=kunlun
Co-authored-by: Nroot <root@yq01-sys-hic-p40-0091.yq01.baidu.com>

* support nhwc format for kunlun conv/batch_norm
*test=kunlun

* minor
*test=kunlun