Skip to content

Support for Eigen thread pool #20157

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
Closed
ilia-cher wants to merge 15 commits into from
Closed

Support for Eigen thread pool #20157

ilia-cher wants to merge 15 commits into from

Conversation

@ilia-cher
Copy link
Contributor

@ilia-cher ilia-cher commented May 6, 2019 •
edited
Loading

Stack from ghstack:

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch
pytest -s -v test/test_jit.py

Differential Revision: D15248708

@pytorchbot pytorchbot added caffe2 module: build Build system issues module: internals Related to internal abstractions in c10 and ATen labels May 6, 2019
This was referenced May 6, 2019
Closed
Closed
Closed
Closed
Closed
Closed
Closed
Closed
Closed
Closed
@jamesr66a
Copy link
Collaborator

jamesr66a commented May 6, 2019

Any way to test this?

@ilia-cher
Copy link
Contributor Author

ilia-cher commented May 6, 2019 •
edited
Loading

@jamesr66a I did build PT with PARALLEL_BACKEND=NATIVE to use native thread pool and with USE_EIGEN_THREADPOOL=1 to use Eigen, then ran test_torch.py, a lot of the ops in these tests use parallelism and we also have dedicated parallelization tests there (e.g. test_cpu_parallel)
We don't have a dedicated thread pool test in PT itself, but there's one in Eigen, we could maybe also include it

@ilia-cher
Copy link
Contributor Author

ilia-cher commented May 6, 2019

@jamesr66a I also ran test_jit.py to test explicit fork/join parallelism

@ilia-cher ilia-cher requested a review from zheng-xq May 6, 2019 20:52
@jamesr66a
Copy link
Collaborator

jamesr66a commented May 6, 2019

Ok. But should we get this tested in CI somehow?

@ilia-cher
Copy link
Contributor Author

ilia-cher commented May 6, 2019

Makes sense, I guess we should add a few more builds, not only for Eigen but for native and TBB backends too; I think it's better to send a separate PR, it will require some changes to our CI builds and @kostmo review

@suo suo removed their request for review May 10, 2019 17:06
This was referenced May 13, 2019
Closed
Closed
@cpuhrsch
Copy link
Contributor

cpuhrsch commented May 20, 2019

Where will this be needed? For now we wanted to avoid adding Eigen as a dependency. cc @soumith

@ilia-cher
Copy link
Contributor Author

ilia-cher commented May 21, 2019

some benchmarking data

./build/bin/at_launch_benchmark
Launching 10000000 tasks using 48 threads
Time to run 10000000 iterations 19.216 s.

(48 - size of the thread pool)

vs. when built with USE_EIGEN_THREADPOOL=0:

./build/bin/at_launch_benchmark
Launching 10000000 tasks using 48 threads
Time to run 10000000 iterations 187.441 s.

when using 50 parallel threads launching tasks instead of a single serial loop

Launching 10000000 tasks using 48 threads
Time to run 10000000 iterations 9.916 s.
vs.
Launching 10000000 tasks using 48 threads
Time to run 10000000 iterations 95.326 s.

ilia-cher added 6 commits May 21, 2019 23:32
Update on "Support for Eigen thread pool"
578cf89 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
662adf3 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
d67a2ac 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
e03ec15 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
dd344ed 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
018796e 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
This was referenced May 23, 2019
Closed
Closed
ilia-cher added 9 commits May 23, 2019 00:11
Update on "Support for Eigen thread pool"
36668b9 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
a0ed864 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
6b55821 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
6ac7d9f 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
7f28866 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
144e229 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
658ca03 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
09f1cae 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
Update on "Support for Eigen thread pool"
c1d756a 
Support for Eigen thread pool

Summary:
Adding support for Eigen thread pool as a replacement
for the current native thread pool

Test Plan:
USE_EIGEN_THREADPOOL=1 USE_CUDA=0 PARALLEL_BACKEND=NATIVE BLAS=MKL
USE_MKLDNN=1 USE_OPENCV=1 USE_FFMPEG=1 python setup.py develop --cmake
pytest -s -v test/test_torch.py::TestTorch

gh-metadata: pytorch pytorch 20157 gh/ilia-cher/16/head
@ilia-cher ilia-cher closed this Jun 6, 2019
@facebook-github-bot facebook-github-bot deleted the gh/ilia-cher/16/head branch October 28, 2019 22:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@soumith soumith Awaiting requested review from soumith

@dzhulgakov dzhulgakov Awaiting requested review from dzhulgakov

@VitalyFedyunin VitalyFedyunin Awaiting requested review from VitalyFedyunin

@cpuhrsch cpuhrsch Awaiting requested review from cpuhrsch

@jamesr66a jamesr66a Awaiting requested review from jamesr66a

@zheng-xq zheng-xq Awaiting requested review from zheng-xq

Assignees

No one assigned

Labels

caffe2 module: build Build system issues module: internals Related to internal abstractions in c10 and ATen

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@ilia-cher @jamesr66a @cpuhrsch @pytorchbot