Skip to content

Cherry-pick from PaddlePaddle:feature/anakin-engine: Anakin subgraph support.#16330

Merged
NHZlX merged 18 commits intoPaddlePaddle:developfrom
NHZlX:merge_anakin_branch_to_dev
Mar 27, 2019
Merged

Cherry-pick from PaddlePaddle:feature/anakin-engine: Anakin subgraph support.#16330
NHZlX merged 18 commits intoPaddlePaddle:developfrom
NHZlX:merge_anakin_branch_to_dev

Conversation

@NHZlX
Copy link
Contributor

@NHZlX NHZlX commented Mar 20, 2019

This branch Align with branch PaddlePaddle:feature/anakin-engine except for:

1. conv_fusion cudnn workspace problem,  feature/anakin-engine branch takes a hard code way. A better way should be considered.
2. add int32 data type for zerocopy.
3. should add ut for paddle-anakin.
cmake .. -DANAKIN_ROOT=/anakin_root/

The 3 points above will be added in this pr later.

Superjomn and others added 13 commits March 20, 2019 19:50
* add activation op

* test conv2d relu sigmoid tanh
…_norm/flatten/reshape op (PaddlePaddle#16020)

* add anakin softmax/ flatten/reshape/transpose/batch_norm op converter
* use anakin batch norm and scale implement fluid batch norm
…addlePaddle#16313

1. refine anakin engine
2. add data type for zero copy

align dev branch and PaddlePaddle:feature/anakin-engine brach
the cudnn workspace modify was not included for now, because we use a hard code way
in feature/anakin-engine branch. There should be a better way to implement it,
and subsequent submissions will be made.

test=develop
@NHZlX NHZlX requested review from Superjomn and fc500110 March 20, 2019 13:54
Shixiaowei02
Shixiaowei02 previously approved these changes Mar 22, 2019
Copy link
Contributor

@Superjomn Superjomn left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@NHZlX NHZlX merged commit fa1796a into PaddlePaddle:develop Mar 27, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants