Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Sum kernel for CPU supporting BF16 and SelectedRows #32631

Merged

Conversation

arogowie-intel
Copy link
Contributor

PR types

New features

PR changes

OPs

Describe

Introduce CPU native kernel supporting SUM operator with BF16 data type and SelectedRows input tensors.

@paddle-bot-old
Copy link

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

paddle/fluid/platform/mkldnn_reuse.h Outdated Show resolved Hide resolved
paddle/fluid/operators/math/blas_impl.h Outdated Show resolved Hide resolved
jczaja
jczaja previously approved these changes Apr 29, 2021
Copy link
Contributor

@jczaja jczaja left a comment

Choose a reason for hiding this comment

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

LGTM

* reported by coverage CI.
arlesniak
arlesniak previously approved these changes May 5, 2021
Copy link
Contributor

@arlesniak arlesniak left a comment

Choose a reason for hiding this comment

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

LGTM

@arogowie-intel
Copy link
Contributor Author

arogowie-intel commented May 5, 2021

@luotao1 The CI failures are as following:

  1. PR-CI-APPROVAL is because I'm skipping UT on platforms not supporting BF16 data type.
  2. PR-CI-Coverage is IMHO a false positive, since I can set a breakpoint and stop in gdb in lines which it doesn't see. Attached file with backtrace: sum_op_py_bt.log command: gdb --args python python/paddle/fluid/tests/unittests/test_sum_op.py TestSelectedRowsSumOpInt
  3. PR-CI-OP-benchmark is running GPU benchmarks and this PR's changes are not related to GPU.

Copy link
Contributor

@lidanqing-intel lidanqing-intel left a comment

Choose a reason for hiding this comment

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

I have some doubts. Thanks

paddle/fluid/platform/mkldnn_reuse.h Outdated Show resolved Hide resolved
paddle/fluid/platform/mkldnn_reuse.h Outdated Show resolved Hide resolved
@lidanqing-intel
Copy link
Contributor

LGTM.

@arogowie-intel
Copy link
Contributor Author

@luotao1 Could you start your review please? I've explained in earlier comment why some CI runs failed.

@luotao1 luotao1 merged commit 9599c3b into PaddlePaddle:develop May 6, 2021
@luotao1
Copy link
Contributor

luotao1 commented May 6, 2021

Please cherry-pick into release/2.1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants