This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
fallback to dense version for grad(reshape), grad(expand_dims) #13599
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
zheng-da
reviewed
Dec 12, 2018
| .set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& n) { | ||
| return std::vector<ResourceRequest>{ResourceRequest::kTempSpace}; | ||
| }) | ||
| #endif |
There was a problem hiding this comment.
does mkldnn support reshape?
if you want to optimize for mkldnn, you should add FComputeEx
There was a problem hiding this comment.
FYI, #12980 is enabled the FW of MKL-DNN supported reshape but BW is still WIP @huangzhiyuan
There was a problem hiding this comment.
Then I guess we need to remove this for now.
|
Otherwise, it looks good to me. BTW, I don't think it has anything to do with sparse. |
yzhliu
force-pushed
the
reshape_backward
branch
from
ccc66e1 to
986803c
Compare
| [](const NodeAttrs& attrs){ | ||
| return std::vector<std::pair<int, int> >{{0, 0}}; | ||
| }) | ||
| .set_attr<FInferStorageType>("FInferStorageType", ElemwiseStorageType<1, 1, false, false, false>) |
There was a problem hiding this comment.
If an op only supports dense tensors and FCompute, FInferStorageType is not needed
yzhliu
force-pushed
the
reshape_backward
branch
from
986803c to
04587b5
Compare
eric-haibin-lin
approved these changes
Dec 20, 2018
rondogency
pushed a commit
to rondogency/incubator-mxnet
that referenced
this pull request
Jan 9, 2019
…e#13599) * fallback to dense version for grad(reshape), grad(expand_dims) * add _backward_reshape gpu version * reshape test case comments * fix gpu test * remove mkldnn support for _backward_reshape
haohuanw
pushed a commit
to haohuanw/incubator-mxnet
that referenced
this pull request
Jun 23, 2019
…e#13599) * fallback to dense version for grad(reshape), grad(expand_dims) * add _backward_reshape gpu version * reshape test case comments * fix gpu test * remove mkldnn support for _backward_reshape
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Current
reshapeuses_backward_copyto do gradient calculation.But
_backward_copysparse calculation requires same shapes of input and output - which is fair since sparsity is bind to # of rows and cols.And though
reshapeitself does not have sparse version, we can easily construct a network, in which its backward input and output are sparse - see test case.Thus we need to provide a fallback version for
reshapebackward computation, so asexpand_dims.@eric-haibin-lin @zheng-da
Checklist
Essentials
Please feel free to remove inapplicable items for your PR.