Optimize implementation of torch.pow #46830

Kiyosora · 2020-10-26T02:34:21Z

Related with Implement NumPy-like function torch.float_power() #44937
Use resize_output instead of resize_as
Tuning the native_functions.yaml, move the inplace variant pow_ next to the other pow entries

dr-ci · 2020-10-26T03:26:39Z

💊 CI failures summary and remediations

As of commit dfe8670175 (more details on the Dr. CI page):

💚 💚 Looks good so far! There are no failures yet. 💚 💚

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.

mrshenli · 2020-10-27T15:01:02Z

adding @zou3519, as this PR is adding named tensor support to the operator

ezyang

okey dokey

facebook-github-bot

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

ezyang · 2020-10-30T15:52:07Z

@anjali411 I just realized that this PR intersects with your fix for the accidental complex promotion. Do you think you could work these changes into your patchset in some way? Otherwise I'll wait for your fix to land and rebase this on top of yours.

Kiyosora · 2020-11-11T08:06:46Z

I rebased this PR since it has been a while. BTW, if there is anything else I can do for this PR, I‘d like to be the help.

anjali411 · 2020-11-11T18:49:28Z

@ezyang sorry I completely missed this PR. let me merge my pow PR as is and then we can merge this as a follow-up change.

anjali411 · 2021-01-13T00:30:16Z

Hi @Kiyosora can you rebase this PR on the latest master?

Kiyosora · 2021-01-13T14:24:22Z

Hi @Kiyosora can you rebase this PR on the latest master?

Sure, I've rebase this PR on the latest master. @anjali411

anjali411 · 2021-01-13T15:24:40Z

aten/src/ATen/native/Pow.cpp

-    result.resize_as_(base).copy_(base);
+  auto exponent = (exp.isComplex()) ? exp.toComplexDouble() : exp.toDouble();
+
+  if (exponent == 0.0) {


@Kiyosora can you please retain this check to exp.equal(0.0) (as in master)?

anjali411 · 2021-01-13T15:24:55Z

aten/src/ATen/native/Pow.cpp

-    result.resize_as_(base).fill_(1);
-  } else if (exp.equal(1.0)) {
-    result.resize_as_(base).copy_(base);
+  auto exponent = (exp.isComplex()) ? exp.toComplexDouble() : exp.toDouble();


this is no longer needed. let's remove this!

anjali411 · 2021-01-13T15:25:04Z

aten/src/ATen/native/Pow.cpp

+    resize_output(result, base.sizes());
+    result.fill_(1);
+    namedinference::propagate_names(result, base);
+  } else if (exponent == 1.0) {


same comment as above

anjali411 · 2021-01-13T15:25:19Z

aten/src/ATen/native/Pow.cpp

-  } else if (!base.isComplex() && base.toDouble() == 1.0) {
-    result.resize_as_(exp).fill_(1);
+
+  auto exponent = (base.isComplex()) ? base.toComplexDouble() : base.toDouble();


same comment as above

anjali411 · 2021-01-13T15:25:26Z

aten/src/ATen/native/Pow.cpp

+
+  auto exponent = (base.isComplex()) ? base.toComplexDouble() : base.toDouble();
+
+  if (exponent == 1.0) {


same as above

anjali411

LGTM thanks @Kiyosora

facebook-github-bot

@anjali411 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot · 2021-01-18T22:24:52Z

@anjali411 merged this pull request in d140ca8.

pytorchbot added the open source label Oct 26, 2020

Kiyosora force-pushed the pow_implementation_optimize branch from b58e34a to de7da46 Compare October 26, 2020 03:17

Kiyosora force-pushed the pow_implementation_optimize branch 4 times, most recently from e0f9115 to 0b9deb5 Compare October 27, 2020 06:17

Kiyosora changed the title ~~[WIP] Optimize implementation of torch.pow~~ Optimize implementation of torch.pow Oct 27, 2020

Kiyosora marked this pull request as ready for review October 27, 2020 10:02

mrshenli requested a review from ezyang October 27, 2020 14:57

mrshenli added the module: custom-operators custom operators, custom ops, custom-operators, custom-ops label Oct 27, 2020

mrshenli requested review from gchanan and zou3519 October 27, 2020 14:58

mrshenli added module: named tensor Named tensor support triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module labels Oct 27, 2020

ezyang approved these changes Oct 27, 2020

View reviewed changes

facebook-github-bot reviewed Oct 27, 2020

View reviewed changes

ezyang requested a review from anjali411 October 30, 2020 15:51

facebook-github-bot added the cla signed label Nov 2, 2020

Kiyosora mentioned this pull request Nov 2, 2020

Implement NumPy-like function torch.float_power() #44937

Closed

Kiyosora force-pushed the pow_implementation_optimize branch from 0b9deb5 to 37c25a3 Compare November 11, 2020 06:23

Kiyosora force-pushed the pow_implementation_optimize branch from 37c25a3 to 929ad85 Compare November 30, 2020 05:33

Kiyosora force-pushed the pow_implementation_optimize branch from 929ad85 to d159ac7 Compare December 16, 2020 07:44

Kiyosora force-pushed the pow_implementation_optimize branch from d159ac7 to 4da9ade Compare January 13, 2021 02:22

anjali411 reviewed Jan 13, 2021

View reviewed changes

Kiyosora force-pushed the pow_implementation_optimize branch 2 times, most recently from dfe8670 to 779026a Compare January 14, 2021 14:47

Kiyosora added 2 commits January 15, 2021 05:15

optimize implementation of torch.pow

6283e92

feedback

d014a9f

Kiyosora force-pushed the pow_implementation_optimize branch from 779026a to d014a9f Compare January 14, 2021 21:16

Kiyosora requested a review from anjali411 January 15, 2021 07:18

anjali411 approved these changes Jan 18, 2021

View reviewed changes

facebook-github-bot reviewed Jan 18, 2021

View reviewed changes

facebook-github-bot closed this in d140ca8 Jan 18, 2021

facebook-github-bot added the Merged label Jan 18, 2021

Kiyosora deleted the pow_implementation_optimize branch January 25, 2021 02:02

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimize implementation of torch.pow #46830

Optimize implementation of torch.pow #46830

Kiyosora commented Oct 26, 2020 •

edited

Loading

dr-ci bot commented Oct 26, 2020 •

edited by facebook-github-bot

Loading

mrshenli commented Oct 27, 2020

ezyang left a comment

facebook-github-bot left a comment

ezyang commented Oct 30, 2020

Kiyosora commented Nov 11, 2020

anjali411 commented Nov 11, 2020

anjali411 commented Jan 13, 2021

Kiyosora commented Jan 13, 2021

anjali411 Jan 13, 2021

Kiyosora Jan 15, 2021

anjali411 Jan 13, 2021

Kiyosora Jan 15, 2021

anjali411 Jan 13, 2021

Kiyosora Jan 15, 2021

anjali411 Jan 13, 2021

Kiyosora Jan 15, 2021

anjali411 Jan 13, 2021

Kiyosora Jan 15, 2021

anjali411 left a comment

facebook-github-bot left a comment

facebook-github-bot commented Jan 18, 2021


		auto exponent = (base.isComplex()) ? base.toComplexDouble() : base.toDouble();

		if (exponent == 1.0) {

Optimize implementation of torch.pow #46830

Optimize implementation of torch.pow #46830

Conversation

Kiyosora commented Oct 26, 2020 • edited Loading

dr-ci bot commented Oct 26, 2020 • edited by facebook-github-bot Loading

💊 CI failures summary and remediations

mrshenli commented Oct 27, 2020

ezyang left a comment

Choose a reason for hiding this comment

facebook-github-bot left a comment

Choose a reason for hiding this comment

ezyang commented Oct 30, 2020

Kiyosora commented Nov 11, 2020

anjali411 commented Nov 11, 2020

anjali411 commented Jan 13, 2021

Kiyosora commented Jan 13, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

anjali411 left a comment

Choose a reason for hiding this comment

facebook-github-bot left a comment

Choose a reason for hiding this comment

facebook-github-bot commented Jan 18, 2021

Kiyosora commented Oct 26, 2020 •

edited

Loading

dr-ci bot commented Oct 26, 2020 •

edited by facebook-github-bot

Loading