Unique cuda support #8899
Closed
Unique cuda support #8899
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
yueyericardo
requested review from
apaszke,
colesbury,
ezyang,
gchanan,
soumith and
zdevito
as code owners
Collaborator
|
I don't see a synchronization call ( |
Contributor
Author
|
@ngimel Thanks for your remind. I already changed the code and result. |
Contributor
|
ROCm build OOMed: |
ezyang
approved these changes
Jun 28, 2018
Contributor
ezyang
left a comment
ezyang
left a comment
There was a problem hiding this comment.
Nice! Inverse index algorithm is not "great" but it seems to be good enough.
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Contributor
|
@pytorchbot retest this please |
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Contributor
|
@yueyericardo If you want to unblock this for merge, I'd advise preprocessoring out the implementation when ROCm is building (so we'll have unique for CUDA but not ROCm). Use |
Contributor
Author
|
@ezyang I already implemented your advise.Thank you. |
Contributor
|
Great, thanks! |
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Jul 9, 2018
Summary:
Add cuda support for unique.
There is a simple test below for a tensor including 1M <int> data.
And the performance is faster.
```python
Performance
cpu: 0.05040597915649414 s
x: tensor([1, 3, 1, ..., 4, 9, 4])
x output: tensor([1, 2, 3, 4, 5, 6, 7, 8, 9])
x inverse: tensor([0, 2, 0, ..., 3, 8, 3])
gpu: 0.015192985534667969 s
y: tensor([1, 3, 1, ..., 4, 9, 4], device='cuda:0')
y output: tensor([1, 2, 3, 4, 5, 6, 7, 8, 9], device='cuda:0')
y inverse: tensor([0, 2, 0, ..., 3, 8, 3], device='cuda:0')
```
```python
Code
import torch
import time
x=torch.randint(1,10,(1000000,),dtype=torch.long)
device = torch.device("cuda")
y=x.to(device)
start = time.time();
output,inverse = x.unique(sorted=True,return_inverse=True)
stop = time.time();
print('cpu:',stop-start,'s')
print('x:',x)
print('x output:',output)
print('x inverse:',inverse)
start = time.time();
output1,inverse1 = y.unique(sorted=True,return_inverse=True)
torch.cuda.synchronize();
stop = time.time();
print('gpu:',stop-start,'s')
print('y:',y)
print('y output:',output1)
print('y inverse:',inverse1)
```
Closes pytorch/pytorch#8899
Reviewed By: SsnL
Differential Revision: D8677655
Pulled By: ezyang
fbshipit-source-id: 09df3f0602f235c5d36c7a6e7e1d89dbf82570bb
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Jul 13, 2018
Summary:
Add cuda support for unique.
There is a simple test below for a tensor including 1M <int> data.
And the performance is faster.
```python
Performance
cpu: 0.05040597915649414 s
x: tensor([1, 3, 1, ..., 4, 9, 4])
x output: tensor([1, 2, 3, 4, 5, 6, 7, 8, 9])
x inverse: tensor([0, 2, 0, ..., 3, 8, 3])
gpu: 0.015192985534667969 s
y: tensor([1, 3, 1, ..., 4, 9, 4], device='cuda:0')
y output: tensor([1, 2, 3, 4, 5, 6, 7, 8, 9], device='cuda:0')
y inverse: tensor([0, 2, 0, ..., 3, 8, 3], device='cuda:0')
```
```python
Code
import torch
import time
x=torch.randint(1,10,(1000000,),dtype=torch.long)
device = torch.device("cuda")
y=x.to(device)
start = time.time();
output,inverse = x.unique(sorted=True,return_inverse=True)
stop = time.time();
print('cpu:',stop-start,'s')
print('x:',x)
print('x output:',output)
print('x inverse:',inverse)
start = time.time();
output1,inverse1 = y.unique(sorted=True,return_inverse=True)
torch.cuda.synchronize();
stop = time.time();
print('gpu:',stop-start,'s')
print('y:',y)
print('y output:',output1)
print('y inverse:',inverse1)
```
Closes pytorch/pytorch#8899
Reviewed By: SsnL
Differential Revision: D8677655
Pulled By: ezyang
fbshipit-source-id: 09df3f0602f235c5d36c7a6e7e1d89dbf82570bb
goodlux
pushed a commit
to goodlux/pytorch
that referenced
this pull request
Aug 15, 2018
Summary:
Add cuda support for unique.
There is a simple test below for a tensor including 1M <int> data.
And the performance is faster.
```python
Performance
cpu: 0.05040597915649414 s
x: tensor([1, 3, 1, ..., 4, 9, 4])
x output: tensor([1, 2, 3, 4, 5, 6, 7, 8, 9])
x inverse: tensor([0, 2, 0, ..., 3, 8, 3])
gpu: 0.015192985534667969 s
y: tensor([1, 3, 1, ..., 4, 9, 4], device='cuda:0')
y output: tensor([1, 2, 3, 4, 5, 6, 7, 8, 9], device='cuda:0')
y inverse: tensor([0, 2, 0, ..., 3, 8, 3], device='cuda:0')
```
```python
Code
import torch
import time
x=torch.randint(1,10,(1000000,),dtype=torch.long)
device = torch.device("cuda")
y=x.to(device)
start = time.time();
output,inverse = x.unique(sorted=True,return_inverse=True)
stop = time.time();
print('cpu:',stop-start,'s')
print('x:',x)
print('x output:',output)
print('x inverse:',inverse)
start = time.time();
output1,inverse1 = y.unique(sorted=True,return_inverse=True)
torch.cuda.synchronize();
stop = time.time();
print('gpu:',stop-start,'s')
print('y:',y)
print('y output:',output1)
print('y inverse:',inverse1)
```
Closes pytorch#8899
Reviewed By: SsnL
Differential Revision: D8677655
Pulled By: ezyang
fbshipit-source-id: 09df3f0602f235c5d36c7a6e7e1d89dbf82570bb
theweiho
added a commit
to theweiho/translate
that referenced
this pull request
May 16, 2019
Summary: pytorch/pytorch#8899 had added CUDA support for `torch.unique()` pytorch/pytorch#16145 has some timing stats that could be relevant --- Experiment results: https://fb.quip.com/olQOA853j0mb Words per second (`gpu-unique_wps_avg_vs_base`): 1.046x Total train time (`gpu-unique_total_train_time_vs_base`; excl ar_AR-fr_XX): 0.987x Even though train time reduction is pretty minimal (probably overshadowed by random variance, scheduling delay, etc), WPS does seem to be ~5% faster - so might as well land this. Training time for ar_AR-fr_XX increased significantly - but that's b/c it trained for many more updates (`gpu-unique_num_updates_avg_vs_base`) - and also ended up w/ +1.43 BLEU. I think this is probably just an anomaly? Differential Revision: D15073468 fbshipit-source-id: a9710738c827013afb35a67bd3a9be259b0e2d5f
theweiho
added a commit
to theweiho/translate
that referenced
this pull request
May 16, 2019
…torch#537) Summary: Pull Request resolved: pytorch#537 pytorch/pytorch#8899 had added CUDA support for `torch.unique()` pytorch/pytorch#16145 has some timing stats that could be relevant --- Experiment results: https://fb.quip.com/olQOA853j0mb Words per second (`gpu-unique_wps_avg_vs_base`): 1.046x Total train time (`gpu-unique_total_train_time_vs_base`; excl ar_AR-fr_XX): 0.987x Even though train time reduction is pretty minimal (probably overshadowed by random variance, scheduling delay, etc), WPS does seem to be ~5% faster - so might as well land this. Training time for ar_AR-fr_XX increased significantly - but that's b/c it trained for many more updates (`gpu-unique_num_updates_avg_vs_base`) - and also ended up w/ +1.43 BLEU. I think this is probably just an anomaly? Differential Revision: D15073468 fbshipit-source-id: 713288fc7c77f582840f270dd2e343a3b63f8fe5
theweiho
added a commit
to theweiho/translate
that referenced
this pull request
May 16, 2019
…torch#537) Summary: Pull Request resolved: pytorch#537 pytorch/pytorch#8899 had added CUDA support for `torch.unique()` pytorch/pytorch#16145 has some timing stats that could be relevant --- Experiment results: https://fb.quip.com/olQOA853j0mb Words per second (`gpu-unique_wps_avg_vs_base`): 1.046x Total train time (`gpu-unique_total_train_time_vs_base`; excl ar_AR-fr_XX): 0.987x Even though train time reduction is pretty minimal (probably overshadowed by random variance, scheduling delay, etc), WPS does seem to be ~5% faster - so might as well land this. Training time for ar_AR-fr_XX increased significantly - but that's b/c it trained for many more updates (`gpu-unique_num_updates_avg_vs_base`) - and also ended up w/ +1.43 BLEU. I think this is probably just an anomaly? Differential Revision: D15073468 fbshipit-source-id: 29c7eaaddd63d629866c7314920fe27b22690603
facebook-github-bot
pushed a commit
to pytorch/translate
that referenced
this pull request
May 17, 2019
Summary: Pull Request resolved: #537 pytorch/pytorch#8899 had added CUDA support for `torch.unique()` pytorch/pytorch#16145 has some timing stats that could be relevant --- Experiment results: https://fb.quip.com/olQOA853j0mb Words per second (`gpu-unique_wps_avg_vs_base`): 1.046x Total train time (`gpu-unique_total_train_time_vs_base`; excl ar_AR-fr_XX): 0.987x Even though train time reduction is pretty minimal (probably overshadowed by random variance, scheduling delay, etc), WPS does seem to be ~5% faster - so might as well land this. Training time for ar_AR-fr_XX increased significantly - but that's b/c it trained for many more updates (`gpu-unique_num_updates_avg_vs_base`) - and also ended up w/ +1.43 BLEU. I think this is probably just an anomaly? Reviewed By: akinh, jmp84 Differential Revision: D15073468 fbshipit-source-id: c2dba562b6d4fb4d15d2a56d03ce6a6e3ddff07d
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add cuda support for unique.
There is a simple test below for a tensor including 1M data.
And the performance is faster.