Skip to content

Support fusing 3D Conv with Add/Mul.#23

Merged
weixingzhang merged 5 commits intomasterfrom
fusion-resnet3d
Dec 3, 2018
Merged

Support fusing 3D Conv with Add/Mul.#23
weixingzhang merged 5 commits intomasterfrom
fusion-resnet3d

Conversation

@weixingzhang
Copy link
Contributor

@weixingzhang weixingzhang commented Nov 25, 2018

With this PR, the subgraph 3D Conve->Add->Mul in Resnet3D can be fused into one 3D Conv.

ke1337
ke1337 reviewed Nov 27, 2018
View reviewed changes
ke1337
ke1337 previously approved these changes Nov 28, 2018
View reviewed changes
Copy link
Contributor

@ke1337 ke1337 left a comment

Choose a reason for hiding this comment

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

:shipit:

ke1337
ke1337 previously approved these changes Nov 28, 2018
View reviewed changes
Copy link
Contributor

@ke1337 ke1337 left a comment

Choose a reason for hiding this comment

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

:shipit:

@weixingzhang weixingzhang requested a review from a team as a code owner November 29, 2018 23:23
tracysh
tracysh reviewed Nov 29, 2018
View reviewed changes
onnxruntime/core/graph/conv_mul_fusion.cc Outdated
Copy link
Contributor

@tracysh tracysh Nov 29, 2018

Choose a reason for hiding this comment

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

Caculate->Calculate

Copy link
Contributor Author

@weixingzhang weixingzhang Nov 29, 2018

Choose a reason for hiding this comment

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

Good catch. Thanks!

tracysh
tracysh reviewed Nov 29, 2018
View reviewed changes
onnxruntime/core/graph/conv_mul_fusion.cc Outdated
Copy link
Contributor

@tracysh tracysh Nov 29, 2018

Choose a reason for hiding this comment

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

Does this have the same problem as the conv_add_fusion where the other dims of the multiplier should be 1?

Copy link
Contributor Author

@weixingzhang weixingzhang Nov 29, 2018

Choose a reason for hiding this comment

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

Yeah, it is better to check it.

tracysh
tracysh reviewed Nov 29, 2018
View reviewed changes
onnxruntime/core/graph/conv_add_fusion.cc Outdated
Copy link
Contributor

@tracysh tracysh Nov 29, 2018

Choose a reason for hiding this comment

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

Caculate->Calculate

tracysh
tracysh reviewed Nov 29, 2018
View reviewed changes
onnxruntime/core/graph/initializer.h Outdated
Copy link
Contributor

@tracysh tracysh Nov 29, 2018

Choose a reason for hiding this comment

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

These helpers are a little weird with the broadcasting rules. mul now behaves different than add/sub/etc. Maybe just have a scale_by_scalar to handle the case for now?

Copy link
Contributor Author

@weixingzhang weixingzhang Nov 30, 2018

Choose a reason for hiding this comment

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

Agree. The PR is updated. Please review it! Thanks!

@weixingzhang weixingzhang requested a review from tracysh November 30, 2018 02:55
@weixingzhang
Copy link
Contributor Author

weixingzhang commented Nov 30, 2018

It would be better to hoist the other.size() work outside the for(j) loop.

I would also encourage you to look at the disassembly for these routines: the calls to data() inside the loop end up being calls to go fetch the data pointer on every cycle through the loop for something that should be loaded once. Also if size_ were cached in a local, then the compiler would be able to vectorize the loop. As it stands, the compiler doesn't have enough aliasing info to know that writing to data[] may not trash size_. This cleanup can happen in a different PR though.

@weixingzhang weixingzhang reopened this Nov 30, 2018
@raymondxyang
Copy link

raymondxyang commented Dec 3, 2018

@weixingzhang its good for merge now

@weixingzhang weixingzhang merged commit aa549cd into master Dec 3, 2018
@raymondxyang raymondxyang deleted the fusion-resnet3d branch December 3, 2018 21:11
tmccrmck pushed a commit to tmccrmck/onnxruntime that referenced this pull request Aug 28, 2019
@NonStatic2014
Merge pull request microsoft#23 from NonStatic2014/bohu/generate_prot…
05847b0 
…obuf_status

Create a mapping from ONNX Runtime Status to Protobuf Status
@yueyihua yueyihua mentioned this pull request Jan 9, 2020
Closed
@Liujingxiu23 Liujingxiu23 mentioned this pull request Jun 12, 2020
Closed
@matthill matthill mentioned this pull request Apr 15, 2021
Closed
natke referenced this pull request in natke/onnxruntime Feb 15, 2022
Add Linux CI (#23)
66417fb
@ns-wxin ns-wxin mentioned this pull request Jan 31, 2023
Closed
@lfch lfch mentioned this pull request Oct 13, 2023
Closed
@horror-proton horror-proton mentioned this pull request Feb 21, 2024
Closed
@tianleiwu tianleiwu mentioned this pull request Mar 20, 2024
Merged
@frenetj frenetj mentioned this pull request Oct 15, 2024
Closed
@FaithyQi FaithyQi mentioned this pull request Oct 22, 2024
Closed
@Consulting4J Consulting4J mentioned this pull request Feb 20, 2025
Closed
@hui-li-xf hui-li-xf mentioned this pull request Jun 10, 2025
Closed
@laurenspriem laurenspriem mentioned this pull request Nov 28, 2025
Closed
@qq296216078 qq296216078 mentioned this pull request Dec 11, 2025
Open
@pawanake2333 pawanake2333 mentioned this pull request Jan 6, 2026
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@pranavsharma pranavsharma Awaiting requested review from pranavsharma

@linkerzhang linkerzhang Awaiting requested review from linkerzhang

2 more reviewers

@tracysh tracysh tracysh approved these changes

@ke1337 ke1337 ke1337 left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@weixingzhang @raymondxyang @ke1337 @tracysh