Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Port addcmul operator from the TH code to Aten #22797

Closed
VitalyFedyunin opened this issue Jul 12, 2019 · 2 comments
Closed

Port addcmul operator from the TH code to Aten #22797

VitalyFedyunin opened this issue Jul 12, 2019 · 2 comments
Assignees
@VitalyFedyunin
Labels
better-engineering Relatively self-contained tasks for better engineering contributors module: cpu CPU specific problem (e.g., perf, algorithm) module: cuda Related to torch.cuda, and CUDA support in general module: porting Issues related to porting TH/THNN legacy to ATen native triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Comments

@VitalyFedyunin
Copy link
Contributor

VitalyFedyunin commented Jul 12, 2019 •
edited

addcmul is the point-wise math operator so porting if from the TH code to Aten (and TensorIterator) expected to be easy. Such migration will help to clean up the code, simplify dispatch as well as provide immediate 2-3x operator performance gain.

Porting guide: https://github.com/pytorch/pytorch/wiki/TH-to-ATen-porting-guide

Example PR with porting of the adaptive_avg_pool2d: #14714

How to use TensorIterator: https://github.com/pytorch/pytorch/wiki/How-to-use-TensorIterator
@VitalyFedyunin VitalyFedyunin added module: cpu CPU specific problem (e.g., perf, algorithm) module: cuda Related to torch.cuda, and CUDA support in general module: operators module: porting Issues related to porting TH/THNN legacy to ATen native triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module labels Jul 12, 2019
@VitalyFedyunin VitalyFedyunin mentioned this issue Jul 12, 2019
Closed
@VitalyFedyunin VitalyFedyunin changed the title Port addcdiv operator from the TH code to Aten Port addcmul operator from the TH code to Aten Jul 12, 2019
@ezyang ezyang added the better-engineering Relatively self-contained tasks for better engineering contributors label Jul 12, 2019
@VitalyFedyunin VitalyFedyunin self-assigned this Jul 15, 2019
@VitalyFedyunin
Copy link
Contributor Author

I will make the example out of it.

@VitalyFedyunin VitalyFedyunin mentioned this issue Jul 15, 2019
Closed
petrex pushed a commit to petrex/pytorch that referenced this issue Aug 1, 2019
@VitalyFedyunin @facebook-github-bot
Move addcmul to Aten (pytorch#22874)
3b5daef 
Summary:
Move CPU implementation of the `addcmul` operator to Aten ( pytorch#22797 )

### before

```python
In [11]: timeit x.addcmul(a, b)
1.31 ms ± 18.3 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
```

### after

```python
In [9]: timeit x.addcmul(a, b)
588 µs ± 22.8 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
```

Adding custom code for the case when `value == 1`, doesn't provide significant performance gain.
Pull Request resolved: pytorch#22874

Differential Revision: D16359348

Pulled By: VitalyFedyunin

fbshipit-source-id: 941ead835672fca78a1fcc762da052e64308b111
zdevito pushed a commit to zdevito/ATen that referenced this issue Aug 1, 2019
@VitalyFedyunin @facebook-github-bot
Move addcmul to Aten (#22874)
70c439c 
Summary:
Move CPU implementation of the `addcmul` operator to Aten ( pytorch/pytorch#22797 )

### before

```python
In [11]: timeit x.addcmul(a, b)
1.31 ms ± 18.3 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
```

### after

```python
In [9]: timeit x.addcmul(a, b)
588 µs ± 22.8 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
```

Adding custom code for the case when `value == 1`, doesn't provide significant performance gain.
Pull Request resolved: pytorch/pytorch#22874

Differential Revision: D16359348

Pulled By: VitalyFedyunin

fbshipit-source-id: 941ead835672fca78a1fcc762da052e64308b111
@ifedan ifedan mentioned this issue Aug 5, 2019
Closed
salexspb pushed a commit to salexspb/pytorch that referenced this issue Aug 8, 2019
@ifedan @facebook-github-bot
Move addcmul to Aten(CUDA) (pytorch#23814)
341d593 
Summary:
pytorch#22797
Pull Request resolved: pytorch#23814

Differential Revision: D16712381

Pulled By: ifedan

fbshipit-source-id: aeca4fdb9b10143932f195900b1f424ef6d26c89
zdevito pushed a commit to zdevito/ATen that referenced this issue Aug 8, 2019
@ifedan @facebook-github-bot
Move addcmul to Aten(CUDA) (#23814)
0d3d17e 
Summary:
pytorch/pytorch#22797
Pull Request resolved: pytorch/pytorch#23814

Differential Revision: D16712381

Pulled By: ifedan

fbshipit-source-id: aeca4fdb9b10143932f195900b1f424ef6d26c89
@VitalyFedyunin
Copy link
Contributor Author

Done in #22874 and #23814

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@VitalyFedyunin VitalyFedyunin

Labels
better-engineering Relatively self-contained tasks for better engineering contributors module: cpu CPU specific problem (e.g., perf, algorithm) module: cuda Related to torch.cuda, and CUDA support in general module: porting Issues related to porting TH/THNN legacy to ATen native triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@ezyang @VitalyFedyunin