matmul and matmul_v2 refactor #42732
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
| @@ -519,43 +519,6 @@ def init_data_type(self): | |||
| self.data_type_ = np.int8 | |||
|
|
|||
|
|
|||
| class TestMatMulOpTransposeReshapeTransposeAxisNotSupportedException( | |||
There was a problem hiding this comment.
Why these tests are deleted?
There was a problem hiding this comment.
These tests only checked if after setting incorrect values for fuse_reshape_out and fuse_transpose_out test would fail. I deleted it, because the only way of setting these parameters is via fuse_pass, which has all PADDLE_ENFORCES to prevent such case. Also, we have separate unit test dedicated for matmul+transpose+reshape fuse pass.
There was a problem hiding this comment.
Really nice that you have spotted that! Thanks!
Maybe you need to reassign license ? |
I checked the CLA assistant manually and everything looks fine. Unfortunately, despite the renewal of the license, the CI is still not updated. Can this be merged as it is, or should I trigger CI with empty commit? |
|
@baoachun please review |
|
很抱歉，经过我们的反复讨论，你的PR暂未达到合入标准，请阅读飞桨原生算子开发规范，你可以重新提交新的PR，我们先将此PR关闭，感谢你的贡献。 |
paddle/phi/core/ddim.cc
Outdated
| @@ -171,11 +171,25 @@ DDim stride_numel(const DDim& ddim) { | |||
| return strides; | |||
| } | |||
|
|
|||
| DDim DDim::reshape(const std::vector<int>& shape) const { | |||
| DDim DDim::reshape(const std::vector<int>& new_shape) const { | |||
| std::vector<int> shape = new_shape; | |||
There was a problem hiding this comment.
Creating a new vector here and do ifelse branch in core/ddim.cc will influence each cpu/gpu op I think ?
Silv3S
dismissed stale reviews from lidanqing-intel, jczaja, and jakpiase
via
9dbb43c
|
你的PR已合入Paddle库，请关注后续测试结果。 |
PR types
Others
PR changes
OPs
Describe
reshapefunction, so it can be reused,PADDLE_ENFORCEwhich are checked both in fuse pass and operator, with same conditions,fuse_transpose_outandfuse_reshape_outare intentionally set incorrectly,