Skip to content

Add flag use_mkl_packed#7183

Merged
luotao1 merged 3 commits intoPaddlePaddle:developfrom
tensor-tang:use_mkl_packed
Jan 4, 2018
Merged

Add flag use_mkl_packed#7183
luotao1 merged 3 commits intoPaddlePaddle:developfrom
tensor-tang:use_mkl_packed

Conversation

@tensor-tang
Copy link
Contributor

follow #6612 about how to add MKL Packed interface

@tensor-tang tensor-tang requested a review from luotao1 January 4, 2018 03:18
DEFINE_bool(use_mkl_packed, true, "Default use MKL Packed Optimization");
#else
DEFINE_bool(use_mkl_packed, false, "Whether to use MKL Packed Optimization");
#endif
Copy link
Contributor

Choose a reason for hiding this comment

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

  • 现在直接替换过早了,虽然单测过了,但是整体收敛效果如何还是要跑几个完整的例子来测下,类似v2和fluid的对比实验。how to add MKL Packed interface #6612 (comment) 这里说的test不仅仅是单测。
  • false, "Whether to use MKL Packed Optimization"这里的注释对么?paddle_with_mklml没打开的时候,use_mkl_packed也不能用了吧。
Copy link
Contributor

Choose a reason for hiding this comment

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

最终是变成:

#ifdef PADDLE_WITH_MKLML
DEFINE_bool(use_mkl_packed, true, "whether to use MKL Packed Optimization, default true");
#else
DEFINE_bool(use_mkl_packed, false, "Not to use MKL Packed Optimization");
#endif

但由于现在default还不能开,可以改成:

DEFINE_bool(use_mkl_packed, false, "Whether to use MKL Packed Optimization");

这里可以加��个TODO

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Thx, It's done.

Copy link
Contributor

@luotao1 luotao1 left a comment

Choose a reason for hiding this comment

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

LGTM

@luotao1 luotao1 merged commit ee341ef into PaddlePaddle:develop Jan 4, 2018
@tensor-tang tensor-tang deleted the use_mkl_packed branch January 4, 2018 13:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

2 participants