Optimize thread pool

* add fp16 backward support
test=develop

* add sum_op fp16 test

* disable test_dist_save_load
test=develop

* add check_grad for sum

* add unit test for softmax_grad fp16
test=develop

* add scale_op unit test

* add mul_grad_op unit test for fp16

* add cross_entropy_grad and eman_grad unit test for fp16
test=develop

* fix cross_entropy unit test

* add pool2d fp16 unit test

* refine conv2d fp16 unit test
test=develop

* refine activation unit test
test=develop

* fix ci
test=develop

* follow zhihong's comment, copy from https://github.com/PaddlePaddle/Paddle/pull/12796
test=develop

test=develop

Revert " Exhaustive search for cuDNN conv."

This reverts commit ce7d9b07.

* exhaustive search for cuDNN conv.
* Refine code and add unit testing.
* Clean code
* Fix model load in fluid/inference and unit testing in conv2d
* Follow comments.

Fix rmsprop_op enforce bug

test=develop

fix jit on mac

Remove some locks in ParallelExecutor

Stream Callback Support in CUDA 10

run dist tests in serial

test=develop

test=develop

test=develop

* add dam test

* update fuse_statis

* use separated dam model.

* Revert "use separated dam model."

This reverts commit 13e775c8.

* test=develop

* modify the cmake file about infer test, test=develop.

* remove one comment, test=develop.

test=develop

Warning only at first when CUDA CC not matched.

test=develop

test=develop

* Fix build error of affine grid op in mac os.
test=develop

* Make function return reference.
test=develop

Refine/jit/vmulcode

Fix lod_level share bug in read_op

* wip

* add ref_by_trainer_id op

* ready to test

* fix ref inputs

* refine rpc_op_handle

* fix merge bug

Residual data reorder in MKLDNN convolution

Fix avx illegal instuctions

throw error when mismatch cpu avx version

test=develop

test=develop

Fea/jit/gen

fix to only check block 0