optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid_backward, tanh_backward, addcmul, addcdiv by mingfeima · Pull Request #55221 · pytorch/pytorch

mingfeima · 2021-04-02T07:39:24Z

Stack from ghstack:

add BFloat16 support for BatchNorm2d on CPU #61015 add BFloat16 support for BatchNorm2d on CPU
add BFloat16 support for UpSample on CPU #58297 add BFloat16 support for UpSample on CPU
optimize transpose copy for float32 and bfloat16 on CPU #56904 optimize transpose copy for float32 and bfloat16 on CPU
add BFloat16 support for MaxPool2d on CPU #56903 add BFloat16 support for MaxPool2d on CPU
add BFloat16 support for AdaptiveAvgPool2d on CPU #56902 add BFloat16 support for AdaptiveAvgPool2d on CPU
add BFloat16 support for bernoulli and Dropout on CPU #56372 add BFloat16 support for bernoulli and Dropout on CPU
add bf16 support for bucketize #55588 add bf16 support for bucketize
optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid_backward, tanh_backward, addcmul, addcdiv #55221 optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid_backward, tanh_backward, addcmul, addcdiv

Differential Revision: D28836797

…l, addcdiv [ghstack-poisoned]

facebook-github-bot · 2021-04-02T07:39:32Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/55221
📄 Preview docs built from this PR

💊 CI failures summary and remediations

As of commit 6cc108f (more details on the Dr. CI page):

2/2 failures possibly* introduced in this PR
- 1/2 non-scanned failure(s)

🕵️ 1 new failure recognized by patterns

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

linux-bionic-py3.8-gcc9-coverage / test (default, 1, 2, linux.2xlarge) (1/1)

Step: "Test PyTorch" (full log | diagnosis details | 🔁 rerun)

2021-08-18T03:00:36.3795161Z Build left local git repository checkout dirty

2021-08-18T03:00:28.3723603Z real	84m57.530s
2021-08-18T03:00:28.3724134Z user	191m29.737s
2021-08-18T03:00:28.3724487Z sys	25m30.980s
2021-08-18T03:00:28.3725112Z + assert_git_not_dirty
2021-08-18T03:00:28.3726217Z + [[ linux-bionic-py3.8-gcc9-coverage-default != *rocm* ]]
2021-08-18T03:00:28.3727798Z + [[ linux-bionic-py3.8-gcc9-coverage-default != *xla* ]]
2021-08-18T03:00:28.3728540Z ++ git status --porcelain
2021-08-18T03:00:36.3792906Z + git_status='?? third_party/pocketfft/'
2021-08-18T03:00:36.3793672Z + [[ -n ?? third_party/pocketfft/ ]]
2021-08-18T03:00:36.3794662Z + echo 'Build left local git repository checkout dirty'
2021-08-18T03:00:36.3795161Z Build left local git repository checkout dirty
2021-08-18T03:00:36.3795697Z + echo 'git status --porcelain:'
2021-08-18T03:00:36.3796141Z git status --porcelain:
2021-08-18T03:00:36.3796604Z + echo '?? third_party/pocketfft/'
2021-08-18T03:00:36.3796977Z ?? third_party/pocketfft/
2021-08-18T03:00:36.3797294Z + exit 1
2021-08-18T03:00:36.3797552Z + cleanup
2021-08-18T03:00:36.3797836Z + retcode=1
2021-08-18T03:00:36.3798095Z + set +x
2021-08-18T03:00:36.3798442Z =================== sccache compilation log ===================
2021-08-18T03:00:36.3974513Z =========== If your build fails, please take a look at the log above for possible reasons ===========

ci.pytorch.org: 1 failed

Failed: pr/pytorch-linux-bionic-rocm4.2-py3.6

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 to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

…l, addcdiv ghstack-source-id: 1c4aa33 Pull Request resolved: #55221

mingfeima · 2021-04-02T07:48:58Z

Since this PR is not related to parallelization feature, only single core perf is tested:
NB: tanh_backward doesn't have BFloat16 support previously, the before perf refers to simple impl as following:

-    AT_DISPATCH_FLOATING_TYPES(iter.dtype(), "tanh_backward_cpu", [&]() {
+    AT_DISPATCH_FLOATING_TYPES_AND(kBFloat16, iter.dtype(), "tanh_backward_cpu", [&]() {

performance update on avx512 machine: Xeon(R) Gold 6248 CPU @ 2.50GHz

before: sigmoid: 1x128x1024: fp32: 0.101 ms; bf16: 0.272 ms
after:  sigmoid: 1x128x1024: fp32: 0.109 ms; bf16: 0.144 ms

before: sigmoid backward: 1x128x1024: fp32: 0.114 ms; bf16: 0.154 ms
after:  sigmoid backward: 1x128x1024: fp32: 0.111 ms; bf16: 0.091 ms

before: tanh backward: 1x128x1024: fp32: 0.110 ms; bf16: 0.185 ms
after:  tanh backward: 1x128x1024: fp32: 0.111 ms; bf16: 0.085 ms

performance update on avx2 machine: Xeon(R) CPU E5-2680 v3 @ 2.50GHz

before: sigmoid: 1x128x1024: fp32: 0.241 ms; bf16: 0.480 ms
after:  sigmoid: 1x128x1024: fp32: 0.237 ms; bf16: 0.238 ms

before: sigmoid backward: 1x128x1024: fp32: 0.128 ms; bf16: 0.299 ms
after:  sigmoid backward: 1x128x1024: fp32: 0.131 ms; bf16: 0.149 ms

before: tanh backward: 1x128x1024: fp32: 0.134 ms; bf16: 0.299 ms
after:  tanh backward: 1x128x1024: fp32: 0.136 ms; bf16: 0.131 ms

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

…l, addcdiv ghstack-source-id: 6de211e Pull Request resolved: pytorch#55221

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

…l, addcdiv ghstack-source-id: 0705d16 Pull Request resolved: pytorch#55221

VitalyFedyunin · 2021-06-02T16:12:48Z

@VitalyFedyunin has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

…_backward, tanh_backward, addcmul, addcdiv" Differential Revision: [D28836797](https://our.internmc.facebook.com/intern/diff/D28836797) [ghstack-poisoned]

VitalyFedyunin · 2021-06-23T16:41:29Z

@VitalyFedyunin has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

VitalyFedyunin

Please change code pattern from dispatching on all types and specializing on bfloat16 to check type first and dispatch only on right types.

…_backward, tanh_backward, addcmul, addcdiv" Differential Revision: [D28836797](https://our.internmc.facebook.com/intern/diff/D28836797) [ghstack-poisoned]

VitalyFedyunin · 2021-06-29T18:12:18Z

@VitalyFedyunin has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

…_backward, tanh_backward, addcmul, addcdiv" Differential Revision: [D28836797](https://our.internmc.facebook.com/intern/diff/D28836797) [ghstack-poisoned]

mingfeima · 2021-07-09T01:04:13Z

rebased!

…_backward, tanh_backward, addcmul, addcdiv" Differential Revision: [D28836797](https://our.internmc.facebook.com/intern/diff/D28836797) [ghstack-poisoned]

mingfeima · 2021-07-26T01:53:11Z

rebased!

…_backward, tanh_backward, addcmul, addcdiv" Differential Revision: [D28836797](https://our.internmc.facebook.com/intern/diff/D28836797) [ghstack-poisoned]

mingfeima · 2021-08-04T06:13:35Z

rebased!

…_backward, tanh_backward, addcmul, addcdiv" Differential Revision: [D28836797](https://our.internmc.facebook.com/intern/diff/D28836797) [ghstack-poisoned]

mingfeima · 2021-08-18T01:12:00Z

rebased!

VitalyFedyunin · 2021-08-23T00:04:21Z

@VitalyFedyunin has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot · 2021-08-24T15:58:32Z

@VitalyFedyunin merged this pull request in 94d6215.

elewise op bf16 cpu: sigmoid, sigmoid_backward, tanh_backward, addcmu…

fcf751e

…l, addcdiv [ghstack-poisoned]

mingfeima mentioned this pull request Apr 2, 2021

Optimize some reduction operators on CPU BFloat16 #55202

Closed

facebook-github-bot added the cla signed label Apr 2, 2021

mingfeima mentioned this pull request Apr 2, 2021

add BFloat16 support for LayerNorm CPU #55210

Closed

mingfeima mentioned this pull request Apr 2, 2021

SumKernel (BFloat16): use float as accumulation type #55217

Closed

mingfeima added a commit that referenced this pull request Apr 2, 2021

elewise op bf16 cpu: sigmoid, sigmoid_backward, tanh_backward, addcmu…

64fc765

…l, addcdiv ghstack-source-id: 1c4aa33 Pull Request resolved: #55221

pytorchbot added the open source label Apr 2, 2021

mingfeima changed the title ~~elewise op bf16 cpu: sigmoid, sigmoid_backward, tanh_backward, addcmul, addcdiv~~ optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid_backward, tanh_backward, addcmul, addcdiv Apr 2, 2021

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

7c24bf2

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

mingfeima mentioned this pull request Apr 8, 2021

add bf16 support for bucketize #55588

Closed

mingfeima added 2 commits April 12, 2021 19:27

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

c48236c

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

85764ae

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

mingfeima mentioned this pull request Apr 19, 2021

add BFloat16 support for bernoulli and Dropout on CPU #56372

Closed

mingfeima added 2 commits April 20, 2021 12:44

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

92b9059

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

5d5effc

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

This was referenced Apr 26, 2021

add BFloat16 support for AdaptiveAvgPool2d on CPU #56902

Closed

add BFloat16 support for MaxPool2d on CPU #56903

Closed

optimize transpose copy for float32 and bfloat16 on CPU #56904

Closed

mingfeima added 2 commits April 27, 2021 09:55

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

6d11bd7

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

7d7d892

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

mingfeima added a commit to mingfeima/pytorch that referenced this pull request Apr 28, 2021

elewise op bf16 cpu: sigmoid, sigmoid_backward, tanh_backward, addcmu…

756761f

…l, addcdiv ghstack-source-id: 6de211e Pull Request resolved: pytorch#55221

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

721de7f

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

mingfeima mentioned this pull request May 13, 2021

use mkldnn for Linear on CPU BFloat16 dtype #58210

Closed

mingfeima added 2 commits May 13, 2021 10:52

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

3de5317

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

Update on "optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid…

504963e

…_backward, tanh_backward, addcmul, addcdiv" [ghstack-poisoned]

mingfeima mentioned this pull request May 14, 2021

add BFloat16 support for UpSample on CPU #58297

Closed

dgl-intel pushed a commit to dgl-intel/pytorch that referenced this pull request May 14, 2021

elewise op bf16 cpu: sigmoid, sigmoid_backward, tanh_backward, addcmu…

8b6ab28

…l, addcdiv ghstack-source-id: 0705d16 Pull Request resolved: pytorch#55221

mingfeima added 4 commits June 4, 2021 14:26

VitalyFedyunin suggested changes Jun 25, 2021

View reviewed changes

Comment thread aten/src/ATen/native/cpu/BinaryOpsKernel.cpp Outdated

Comment thread aten/src/ATen/native/cpu/BinaryOpsKernel.cpp Outdated

VitalyFedyunin reviewed Jun 25, 2021

View reviewed changes

Comment thread aten/src/ATen/native/cpu/BinaryOpsKernel.cpp Outdated

mingfeima mentioned this pull request Jun 30, 2021

add BFloat16 support for BatchNorm2d on CPU #61015

Closed

mingfeima added 2 commits June 30, 2021 13:31

mingfeima added 2 commits August 4, 2021 16:06

VitalyFedyunin approved these changes Aug 11, 2021

View reviewed changes

facebook-github-bot closed this in 94d6215 Aug 24, 2021

facebook-github-bot added the Merged label Aug 24, 2021

facebook-github-bot deleted the gh/mingfeima/19/head branch August 28, 2021 14:16

CaoE mentioned this pull request Dec 7, 2022

torch.addcdiv_ != torch.addcdiv (in place addcdiv_ ensures higher precision vs out of place addcdiv) #83231

Closed

Conversation

mingfeima commented Apr 2, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Apr 2, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

💊 CI failures summary and remediations

🕵️ 1 new failure recognized by patterns

linux-bionic-py3.8-gcc9-coverage / test (default, 1, 2, linux.2xlarge) (1/1)

ci.pytorch.org: 1 failed

Uh oh!

mingfeima commented Apr 2, 2021

Uh oh!

VitalyFedyunin commented Jun 2, 2021

Uh oh!

VitalyFedyunin commented Jun 23, 2021

Uh oh!

VitalyFedyunin left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

VitalyFedyunin commented Jun 29, 2021

Uh oh!

mingfeima commented Jul 9, 2021

Uh oh!

mingfeima commented Jul 26, 2021

Uh oh!

mingfeima commented Aug 4, 2021

Uh oh!

mingfeima commented Aug 18, 2021

Uh oh!

VitalyFedyunin commented Aug 23, 2021

Uh oh!

facebook-github-bot commented Aug 24, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

mingfeima commented Apr 2, 2021 •

edited

Loading

facebook-github-bot commented Apr 2, 2021 •

edited

Loading