Make fuse_optimizer_op_pass also work when the model contains sparse gradients.#18664

Merged

chengduoZH merged 11 commits intoPaddlePaddle:developfrom

chengduoZH:support_sparse_gradient

Jul 23, 2019

Commits on Jul 17, 2019

support sparse gradients
chengduozh
committed
polish code
chengduozh
committed
update alloc_continuous_space_for_grad_pass
chengduozh
committed
use coalesce_tensor
chengduozh
committed
polish unit test
chengduozh
committed

Commits on Jul 18, 2019

open fuse_all_optimizer_ops
chengduozh
committed
follow comments
chengduozh
committed

Commits on Jul 21, 2019

polish code
chengduozh
committed
disable fuse optimization op in test_buffer_shared_inplace_pass
chengduozh
committed
disable fuse_optimization
chengduozh
committed

Commits on Jul 22, 2019

fix fuse all reduce
chengduozh
committed