MKLDNN layout: Support for activation operator #11124
Conversation
mozga-intel force-pushed the mozga-intel/Activation_mkldnn_layout branch from 8fac6b8 to 2ba42fd Compare mozga-intel force-pushed the mozga-intel/Activation_mkldnn_layout branch 3 times, most recently from 5150a23 to 5f8cb3d Compare | @luotao1, @tensor-tang The code is prepared to the code-review process. |
| "For some kinds of element wise operations (namely ReLU with alpha parameter equals 0) dst and src can be interchangeable for the backward pass, which allows performing in-place forward even for training." |
| We have the "out-of-place" implementation of code. |
There was a problem hiding this comment.
Please try to use unique_ptr next time.
There was a problem hiding this comment.
Ok. I will change it, next time.
There was a problem hiding this comment.
Everyplace, where we have the GetExpectedKernelType function we use the protected keyword. This the only one reason, why I used it.
There was a problem hiding this comment.
why remove the typename?
There was a problem hiding this comment.
Because, I did't need to have it - redundant code.
mozga-intel force-pushed the mozga-intel/Activation_mkldnn_layout branch from 5f8cb3d to 792d3b2 Compare | @luotao1 Could you continue the checking of this code, please. |
3 participants
Waiting for supports of the mkldnn’s layout.
Please have a look at this activation operator supported by the MKLDNN’s layout and assess if we can keep this design of the code.
This code uses the implementation of layout which is available in the last pull-request. Therefore some of the function in this code are not available in this pull-request, but are available here
This version of code can be merged into the main branch when the pull-request with layout is accepted.
The concept of splits of the long code was proposed by @luotao1
Pull-request is related to #11040