[ATen][Sparse] Use Third-Party Eigen for sparse addmm #101814

Aidyn-A · 2023-05-18T18:26:22Z

This pull request adds a functionality for addmm with the following layouts:

    add(a_csr, b_csr)

    addmm(c_csr, a_csr, b_csr)
    addmm(c_csr, a_csr, b_csc)
    addmm(c_csr, a_csc, b_csc)
    addmm(c_csr, a_csc, b_csr)

using Eigen for non-MKL builds.

Edit: thanks to the comment #101814 (comment) the following ops were also added:

    add(a_csc, b_csc)

    addmm(c_csc, a_csr, b_csr)
    addmm(c_csc, a_csr, b_csc)
    addmm(c_csc, a_csc, b_csc)
    addmm(c_csc, a_csc, b_csr)

pytorch-bot · 2023-05-18T18:26:25Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/101814

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit e92166b with merge base e872bf8 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

Aidyn-A · 2023-05-19T15:22:56Z

@pytorchbot label "ciflow/trunk"

cpuhrsch · 2023-05-23T18:23:53Z

Thank you for the pull request @Aidyn-A - which issue does this address? Do you have timings on the performance?

Aidyn-A · 2023-05-23T19:59:03Z

Hi @cpuhrsch,
The reason why this PR exist is that sparse operations add, addmm and matmul are not available on CPUs without MKL. Therefore, the following architectures are left out:

All machines with ARM based CPUs
All Windows OS machines
All modern Mac machines

Among existing issues I found this one:

Adding sparse addmv and triangular_solve support on CPU - Mac OS - Apple Silicon M2 #96972

I do not have performance numbers at the moment, but I can provide some.

test/test_sparse_csr.py

aten/src/ATen/native/eigen/SparseBlas.cpp

amjames

Couple of comments, however all of them assume that there is general agreement on introducing an eigen dependency.

aten/src/ATen/native/eigen/SparseBlas.cpp

aten/src/ATen/native/eigen/SparseBlas.h

aten/src/ATen/native/eigen/SparseBlas.cpp

facebook-github-bot · 2023-05-30T18:11:17Z

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

cpuhrsch · 2023-05-30T18:11:40Z

I'm going to import this to get a sense for the internal build setup and whether this needs to be co-landed.

facebook-github-bot · 2023-06-21T03:51:07Z

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

cpuhrsch · 2023-06-21T03:52:09Z

Various internal build system changes should make this GH1 landable, but I'm reimporting to verify. I'll be on PTO in a few days, but back July 10th. If it doesn't land before I go on PTO, I'll pick this back up then.

cpuhrsch · 2023-06-21T04:23:34Z

aten/src/ATen/native/eigen/SparseBlasImpl.cpp

@@ -0,0 +1,381 @@
+#include <ATen/native/eigen/SparseBlasImpl.h>


Can you move this file into
aten/src/ATen/native/*sparse*/eigen/SparseBlasImpl.cpp instead of aten/src/ATen/native/eigen/SparseBlasImpl.cpp?

Hope 6a91e00 is what you meant

facebook-github-bot · 2023-06-21T18:02:17Z

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

Aidyn-A · 2023-08-09T08:37:33Z

Hi @cpuhrsch,
Hope you were able to apply the system changes you mentioned. Shall I rebase this PR to be up-to-date with the latest PyTorch?

cpuhrsch · 2023-09-12T17:39:17Z

@Aidyn-A - Yes, could you please try to rebase and I'll confirm that it works internally? Please mention me once that's done.

Aidyn-A · 2023-09-12T17:40:40Z

@pytorchbot rebase

pytorchmergebot · 2023-09-12T17:43:15Z

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

pytorchmergebot · 2023-09-12T17:43:21Z

Successfully rebased eigen_for_sparse_addmm onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout eigen_for_sparse_addmm && git pull --rebase)

facebook-github-bot · 2023-09-12T17:54:15Z

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

pytorchmergebot · 2025-03-11T16:38:03Z

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

pytorchmergebot · 2025-03-11T16:38:08Z

Successfully rebased eigen_for_sparse_addmm onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout eigen_for_sparse_addmm && git pull --rebase)

Aidyn-A · 2025-04-30T16:12:34Z

@pytorchbot rebase

pytorchmergebot · 2025-04-30T16:14:13Z

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

pytorchmergebot · 2025-04-30T16:14:17Z

Successfully rebased eigen_for_sparse_addmm onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout eigen_for_sparse_addmm && git pull --rebase)

Aidyn-A · 2025-05-05T08:45:36Z

@cpuhrsch re-opening this PR. Is there anything I can do to resolve the internal failures?

…155357) This pull request adds the following ops for sparse matrices using Eigen library: ```python add(a_csr, b_csr) add(a_csc, b_csc) addmm(c_csr, a_csr, b_csr) addmm(c_csr, a_csr, b_csc) addmm(c_csr, a_csc, b_csc) addmm(c_csr, a_csc, b_csr) addmm(c_csc, a_csr, b_csr) addmm(c_csc, a_csr, b_csc) addmm(c_csc, a_csc, b_csc) addmm(c_csc, a_csc, b_csr) ``` Currently, the operations for sparse matrices on CPU are available through MKL only. The non-existence of MKL on `aarch64` causes the unavailability of these ops on any machines with ARM based CPUs, including Apple Silicon, AWS Graviton and NVIDIA Grace. This PR addresses this issue by using Eigen as a backend for the above ops. This is a re-factored version of my previous PR #101814. The main difference with the old one, this does not enable Eigen by default. Pull Request resolved: #155357 Approved by: https://github.com/pearu, https://github.com/eqy

…155357) This pull request adds the following ops for sparse matrices using Eigen library: ```python add(a_csr, b_csr) add(a_csc, b_csc) addmm(c_csr, a_csr, b_csr) addmm(c_csr, a_csr, b_csc) addmm(c_csr, a_csc, b_csc) addmm(c_csr, a_csc, b_csr) addmm(c_csc, a_csr, b_csr) addmm(c_csc, a_csr, b_csc) addmm(c_csc, a_csc, b_csc) addmm(c_csc, a_csc, b_csr) ``` Currently, the operations for sparse matrices on CPU are available through MKL only. The non-existence of MKL on `aarch64` causes the unavailability of these ops on any machines with ARM based CPUs, including Apple Silicon, AWS Graviton and NVIDIA Grace. This PR addresses this issue by using Eigen as a backend for the above ops. This is a re-factored version of my previous PR #101814. The main difference with the old one, this does not enable Eigen by default. Pull Request resolved: #155357 Approved by: https://github.com/pearu, https://github.com/eqy Co-authored-by: Eli Uriegas <eliuriegas@meta.com>

…ytorch#155357) This pull request adds the following ops for sparse matrices using Eigen library: ```python add(a_csr, b_csr) add(a_csc, b_csc) addmm(c_csr, a_csr, b_csr) addmm(c_csr, a_csr, b_csc) addmm(c_csr, a_csc, b_csc) addmm(c_csr, a_csc, b_csr) addmm(c_csc, a_csr, b_csr) addmm(c_csc, a_csr, b_csc) addmm(c_csc, a_csc, b_csc) addmm(c_csc, a_csc, b_csr) ``` Currently, the operations for sparse matrices on CPU are available through MKL only. The non-existence of MKL on `aarch64` causes the unavailability of these ops on any machines with ARM based CPUs, including Apple Silicon, AWS Graviton and NVIDIA Grace. This PR addresses this issue by using Eigen as a backend for the above ops. This is a re-factored version of my previous PR pytorch#101814. The main difference with the old one, this does not enable Eigen by default. Pull Request resolved: pytorch#155357 Approved by: https://github.com/pearu, https://github.com/eqy

…ytorch#155357) This pull request adds the following ops for sparse matrices using Eigen library: ```python add(a_csr, b_csr) add(a_csc, b_csc) addmm(c_csr, a_csr, b_csr) addmm(c_csr, a_csr, b_csc) addmm(c_csr, a_csc, b_csc) addmm(c_csr, a_csc, b_csr) addmm(c_csc, a_csr, b_csr) addmm(c_csc, a_csr, b_csc) addmm(c_csc, a_csc, b_csc) addmm(c_csc, a_csc, b_csr) ``` Currently, the operations for sparse matrices on CPU are available through MKL only. The non-existence of MKL on `aarch64` causes the unavailability of these ops on any machines with ARM based CPUs, including Apple Silicon, AWS Graviton and NVIDIA Grace. This PR addresses this issue by using Eigen as a backend for the above ops. This is a re-factored version of my previous PR pytorch#101814. The main difference with the old one, this does not enable Eigen by default. Pull Request resolved: pytorch#155357 Approved by: https://github.com/pearu, https://github.com/eqy Co-authored-by: Eli Uriegas <eliuriegas@meta.com>

pytorch-bot bot added the release notes: sparse release notes category label May 18, 2023

pytorchbot added the open source label May 18, 2023

Aidyn-A marked this pull request as ready for review May 19, 2023 01:23

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label May 19, 2023

jbschlosser requested review from amjames, cpuhrsch and pearu and removed request for amjames May 19, 2023 20:52

jbschlosser added the triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module label May 19, 2023

cpuhrsch reviewed May 23, 2023

View reviewed changes

test/test_sparse_csr.py Outdated Show resolved Hide resolved

ngimel reviewed May 24, 2023

View reviewed changes

aten/src/ATen/native/eigen/SparseBlas.cpp Outdated Show resolved Hide resolved

amjames reviewed May 24, 2023

View reviewed changes

aten/src/ATen/native/eigen/SparseBlas.cpp Outdated Show resolved Hide resolved

aten/src/ATen/native/eigen/SparseBlas.h Outdated Show resolved Hide resolved

aten/src/ATen/native/eigen/SparseBlas.cpp Outdated Show resolved Hide resolved

cpuhrsch reviewed Jun 21, 2023

View reviewed changes

pytorchmergebot force-pushed the eigen_for_sparse_addmm branch from 6e8cb4a to 2e428d2 Compare September 12, 2023 17:43

pytorchmergebot force-pushed the eigen_for_sparse_addmm branch from 5d21600 to cec2a79 Compare March 11, 2025 16:38

github-actions bot closed this Apr 10, 2025

Aidyn-A reopened this Apr 30, 2025

Aidyn-A and others added 14 commits April 30, 2025 16:14

use eigen for sparse addmm

deb6982

fix no-op build

9d2b232

update test_addmm_all_sparse

c1f14e4

fix win test and bazel build

56eb79f

add rule for lite interpreter

5167815

bazel: add eigen as dep

a6af29d

handle M=nan case

3cf3e55

remove duplicated code

c845bba

consistent namespaces

8422c4a

check contiguity

0da1639

sparse add(csc, csc, csc)

de363cd

mv files

ce80127

avoid mobile

3433c4a

fix typo

e92166b

pytorchmergebot force-pushed the eigen_for_sparse_addmm branch from cec2a79 to e92166b Compare April 30, 2025 16:14

github-actions bot closed this Jun 4, 2025

Aidyn-A mentioned this pull request Jun 6, 2025

[ATen][CPU][Sparse] Use Third-Party Eigen for sparse add and addmm #155357

Closed

		@@ -0,0 +1,381 @@
		#include <ATen/native/eigen/SparseBlasImpl.h>

[ATen][Sparse] Use Third-Party Eigen for sparse addmm #101814

[ATen][Sparse] Use Third-Party Eigen for sparse addmm #101814

Uh oh!

Conversation

Aidyn-A commented May 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented May 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/101814

✅ No Failures

Uh oh!

Aidyn-A commented May 19, 2023

Uh oh!

cpuhrsch commented May 23, 2023

Uh oh!

Aidyn-A commented May 23, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

amjames left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

facebook-github-bot commented May 30, 2023

Uh oh!

cpuhrsch commented May 30, 2023

Uh oh!

facebook-github-bot commented Jun 21, 2023

Uh oh!

cpuhrsch commented Jun 21, 2023

Uh oh!

cpuhrsch Jun 21, 2023

Choose a reason for hiding this comment

Uh oh!

Aidyn-A Jun 21, 2023

Choose a reason for hiding this comment

Uh oh!

facebook-github-bot commented Jun 21, 2023

Uh oh!

Aidyn-A commented Aug 9, 2023

Uh oh!

cpuhrsch commented Sep 12, 2023

Uh oh!

Aidyn-A commented Sep 12, 2023

Uh oh!

pytorchmergebot commented Sep 12, 2023

Uh oh!

pytorchmergebot commented Sep 12, 2023

Uh oh!

facebook-github-bot commented Sep 12, 2023

Uh oh!

pytorchmergebot commented Mar 11, 2025

Uh oh!

pytorchmergebot commented Mar 11, 2025

Uh oh!

Aidyn-A commented Apr 30, 2025

Uh oh!

pytorchmergebot commented Apr 30, 2025

Uh oh!

pytorchmergebot commented Apr 30, 2025

Uh oh!

Aidyn-A commented May 5, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

Aidyn-A commented May 18, 2023 •

edited

Loading

pytorch-bot bot commented May 18, 2023 •

edited

Loading

Aidyn-A commented May 23, 2023 •

edited

Loading