Move resize logic for bmm from codegen to native code. by gchanan · Pull Request #37955 · pytorch/pytorch

gchanan · 2020-05-06T19:59:40Z

Stack from ghstack:

Kill resize-ing and zero-ing from codegen. #37958 Kill resize-ing and zero-ing from codegen.
Move resize / zero logic for _thnn_conv_depthwise2d from codegen to native code. #37957 Move resize / zero logic for _thnn_conv_depthwise2d from codegen to native code.
Move _thnn_conv2d resize and zero code from codegen to native code. #37956 Move _thnn_conv2d resize and zero code from codegen to native code.
Move resize logic for bmm from codegen to native code. #37955 Move resize logic for bmm from codegen to native code.
[RESUBMIT] Kill broadcasting from the codegen layer. #37907 [RESUBMIT] Kill broadcasting from the codegen layer.

Differential Revision: D21433213

[ghstack-poisoned]

dr-ci · 2020-05-06T20:34:25Z

💊 Build failures summary and remediations

As of commit ad5df71 (more details on the Dr. CI page):

1/2 failures introduced in this PR
1/2 broken upstream at merge base 91c1505 on May 05 from 2:00pm to 6:02pm PDT (21 commits; 06e1b68 - e3d1c4e)

🕵️ 1 new failure recognized by patterns

The following build failures do not appear to be due to upstream breakages:

pytorch_linux_backward_compatibility_check_test (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

May 06 20:30:39 The PR is introducing backward incompatible changes to the operator library. Please contact PyTorch team to confirm whether this change is wanted or not.

May 06 20:30:39 processing existing schema:  aten::var.out(Tensor self, int[1] dim, bool unbiased=True, bool keepdim=False, *, Tensor(a!) out) -> (Tensor(a!)) 
May 06 20:30:39 processing existing schema:  aten::var.names_dim(Tensor self, str[1] dim, bool unbiased=True, bool keepdim=False) -> (Tensor) 
May 06 20:30:39 processing existing schema:  aten::var.names_out(Tensor self, str[1] dim, bool unbiased=True, bool keepdim=False, *, Tensor(a!) out) -> (Tensor(a!)) 
May 06 20:30:39 processing existing schema:  aten::var_mean(Tensor self, bool unbiased=True) -> (Tensor, Tensor) 
May 06 20:30:39 processing existing schema:  aten::var_mean.dim(Tensor self, int[1] dim, bool unbiased=True, bool keepdim=False) -> (Tensor, Tensor) 
May 06 20:30:39 processing existing schema:  aten::var_mean.names_dim(Tensor self, str[1] dim, bool unbiased=True, bool keepdim=False) -> (Tensor, Tensor) 
May 06 20:30:39 processing existing schema:  aten::view_as(Tensor self, Tensor other) -> (Tensor) 
May 06 20:30:39 processing existing schema:  aten::where.self(Tensor condition, Tensor self, Tensor other) -> (Tensor) 
May 06 20:30:39 processing existing schema:  aten::where(Tensor condition) -> (Tensor[]) 
May 06 20:30:39 processing existing schema:  aten::_s_where(Tensor condition, Tensor self, Tensor other) -> (Tensor) 
May 06 20:30:39 The PR is introducing backward incompatible changes to the operator library. Please contact PyTorch team to confirm whether this change is wanted or not.  
May 06 20:30:39  
May 06 20:30:39 Broken ops: [ 
May 06 20:30:39 	quantized::conv_unpack(__torch__.torch.classes.quantized.Conv2dPackedParamsBase packed_weights) -> (Tensor unpacked_weights, Tensor? B_origin) 
May 06 20:30:39 ] 
May 06 20:30:39 =================== sccache compilation log =================== 
May 06 20:30:39 + cleanup 
May 06 20:30:39 + retcode=1 
May 06 20:30:39 + set +x 
May 06 20:30:39 =========== If your build fails, please take a look at the log above for possible reasons =========== 
May 06 20:30:39 Compile requests                 0

🚧 1 fixed upstream failure:

These were probably caused by upstream breakages that were already fixed.

Please rebase on the viable/strict branch (expand for instructions)

Since your merge base is older than viable/strict, run these commands:

git fetch https://github.com/pytorch/pytorch viable/strict
git rebase FETCH_HEAD

Check out the recency history of this "viable master" tracking branch.

pytorch_macos_10_13_py3_test on May 05 from 2:00pm to 6:02pm PDT (21 commits; 06e1b68 - e3d1c4e)
- 🔁 rerun

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker.

See how this bot performed.

This comment has been revised 2 times.

bhosmer · 2020-05-06T21:16:30Z

 }

-Tensor bmm_cuda(const Tensor& self, const Tensor& mat2) {
-  return legacy::cuda::_th_bmm(self, mat2);


curious, does this removal mean there's now a generated legacy::cuda::_th_bmm that nobody calls?

yes, there's no way to control that separately. I've considered adding another annotation in order to reduce the amount of generated code, but haven't done it yet.

facebook-github-bot · 2020-05-07T16:12:01Z

@gchanan merged this pull request in b2cc992.

ghstack-source-id: e37d407 Pull Request resolved: pytorch/pytorch#37955

Summary: Pull Request resolved: pytorch#37955 Test Plan: Imported from OSS Differential Revision: D21433213 Pulled By: gchanan fbshipit-source-id: 421c566471279b53348bc77e738af13a1f3e1f9e

Move resize logic for bmm from codegen to native code.

ad5df71

[ghstack-poisoned]

This was referenced May 6, 2020

Move _thnn_conv2d resize and zero code from codegen to native code. #37956

Closed

Move resize / zero logic for _thnn_conv_depthwise2d from codegen to native code. #37957

Closed

Kill resize-ing and zero-ing from codegen. #37958

Closed

gchanan requested a review from bhosmer May 6, 2020 20:12

bhosmer approved these changes May 6, 2020

View reviewed changes

facebook-github-bot closed this in b2cc992 May 7, 2020

facebook-github-bot added the merged label May 7, 2020

This was referenced May 7, 2020

Macro generate ScalarTypeToCPPType, including all ScalarTypes. #38068

Closed

Remove codegen for IntArrayRefStride, which isn't used. #38069

Closed

Move (most) generated return statements for TH functions out of the switch. #38070

Closed

xuezhou1998 pushed a commit to xuezhou1998/new_pytorch that referenced this pull request May 9, 2020

Move resize logic for bmm from codegen to native code.

87a76b7

ghstack-source-id: e37d407 Pull Request resolved: pytorch/pytorch#37955

facebook-github-bot deleted the gh/gchanan/267/head branch May 11, 2020 14:19

mruberry added the Merged label Oct 28, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Move resize logic for bmm from codegen to native code.#37955

Move resize logic for bmm from codegen to native code.#37955
gchanan wants to merge 1 commit intogh/gchanan/267/basefrom
gh/gchanan/267/head

gchanan commented May 6, 2020 •

edited

Loading

Uh oh!

dr-ci Bot commented May 6, 2020 •

edited

Loading

Uh oh!

bhosmer May 6, 2020

Uh oh!

gchanan May 7, 2020

Uh oh!

facebook-github-bot commented May 7, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

gchanan commented May 6, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dr-ci Bot commented May 6, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

💊 Build failures summary and remediations

🕵️ 1 new failure recognized by patterns

pytorch_linux_backward_compatibility_check_test (1/1)

🚧 1 fixed upstream failure:

Uh oh!

bhosmer May 6, 2020

Choose a reason for hiding this comment

Uh oh!

gchanan May 7, 2020

Choose a reason for hiding this comment

Uh oh!

facebook-github-bot commented May 7, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

gchanan commented May 6, 2020 •

edited

Loading

dr-ci Bot commented May 6, 2020 •

edited

Loading