* 1. Create buddy allocator in each places before NcclBcast the variables
2. Check the memory usage of ALL gpus rather than the first one

* 1. Make NCCLGroupGuard guards only the ncclBcast part, which avoid ncclGroupEnd blocking the exception throwing
2. NOTE the usage of NCCLGroupGuard

* Remove the memory usage check of gpus

* Fix code style

* Add mean_iou op.

* Add unitest for mean iou op.

* Add optional collections of confusion matrix and mean_iou.

* Fix cuda kernel.

* Refine code.
1. Merge computing in GPU to two kernel.
2. Use wrong array and correct array instead of confusion matrix.

* Add python api and fix cuda kernel.

* Fix comments.

* Small fix.

* Small fix.

In cupti samples, only cuptiFlush is used.
I can't find any places calling cuptiFinalize and
this API can error out as not_implemented in some
cuda installation.

* fix #11416

* make sgd check tape has been backwarded_

* add error message

* change dynamic to tape

* update readme link

* wip

* use executor.complete to end trainer

* fix build

* fix build with distribute off

* fix typo

* fix cmake typo

* fix build

* Add normalization operator.
1. Refine the raw norm_op and let it more general to support to normalize Tensor along any axis.
2. There is a bug in l2_normalize API, which lacks sqrt after `reduce_sum`.
3. Use norm_op to refine the l2_normalize API.
4. Fix bug in test_normalization_wrapper.py.

* Add slice op.

* Remove using from header file and fix doc.

* Fix doc

* Small fix.

The status is updated in the Process() thread
and can be checked in another HandleRequest() thread.

* "add inplace attribute"

* "register inplace attribute"

* "change se-next model for memory-reuse"

* "fix typo"

* repick

* fix merge conflict

* "fix stupid error"