[new-exec] fit mkldnn #41058
[new-exec] fit mkldnn #41058
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
| // NOTE(zhiqiu): hot fix, follow the same logic in DataCopy() in fetch_op.cc | ||
| if (in_layout == framework::DataLayout::kMKLDNN && | ||
| var_name == framework::GradVarName("Filter") && is_fetch_v2) { | ||
| out_layout = framework::DataLayout::kNCHW; |
There was a problem hiding this comment.
Why out_layout is set to kNCHW ?
There was a problem hiding this comment.
I don't really understand the tricky, just follow DataCopy() in fetch_op.cc
There was a problem hiding this comment.
Ok. Thanks I understand now.
| VLOG(4) << in_layout << "->" << out_layout << " " << in_tensor.layout(); | ||
| if (!in_tensor.IsInitialized() && in_layout == DataLayout::kMKLDNN && | ||
| out_layout == DataLayout::kNHWC) { |
There was a problem hiding this comment.
Recently DataLayout was extended with 5D NHWC layout e.g. kNDHWC for which also MatchShapeToLayout() should be called
There was a problem hiding this comment.
Ok, I think this can be refined in the future, here just follow the same logic in executor.cc
|
@zhiqiu We just started our own tests on this PR just to see functional and performance impact of changes . We will come back to you when we have results (this week) so please wait with merging |
@jczaja Thanks. Since paddle is approaching release-2.3, and there are some works depending on this PR, can we merge it first? If any problems are found, we can fix them in the next PR. |
|
@zhiqiu Yes. we just finished light testing and this PR passed our C++ workloads . So should be fine. It can be merged. |
Thanks. |
PR types
New features
PR changes
Others
Describe
fit mkldnn for new executor