Skip to content

[CD][CUDA13][PTXAS] Remove ptxas bundle from PyTorch cu13 Binary#174716

Closed
nWEIdia wants to merge 2 commits intopytorch:mainfrom
nWEIdia:main-revert-ptxas-cu13-bundle
Closed

[CD][CUDA13][PTXAS] Remove ptxas bundle from PyTorch cu13 Binary#174716
nWEIdia wants to merge 2 commits intopytorch:mainfrom
nWEIdia:main-revert-ptxas-cu13-bundle

Conversation

@nWEIdia
Copy link
Copy Markdown
Collaborator

@nWEIdia nWEIdia commented Feb 10, 2026

The ptxas bundling was introduced in #163988 to workaround issues users may face due to #163801

Fortunately, on the triton upstream side, triton-lang/triton@884fdae finally landed, which is means #163801 is permanently fixed.

In addition, pytorch's triton commit pin has been updated via #178821

We can now roll back #163801 .

In between, we unified the arm sbsa build with x86, so revert won't work. Manually reverting the export.

Test plan: download and check the binary size to confirm 1) ptxas is gone from both x86 and sbsa (even though I only added to sbsa cu13 initially) 2) unit test that ran on #163988 should still pass.

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben @jataylo @eqy @ptrblck @tinglvv @malfet @atalman

@nWEIdia nWEIdia requested a review from a team as a code owner February 10, 2026 21:11
@pytorch-bot
Copy link
Copy Markdown

pytorch-bot Bot commented Feb 10, 2026

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/174716

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

❌ 7 Cancelled Jobs

As of commit 55f4526 with merge base 1aa5e36 (image):

CANCELLED JOBS - The following jobs were cancelled. Please retry:

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

@pytorch-bot
Copy link
Copy Markdown

pytorch-bot Bot commented Feb 10, 2026

This PR needs a release notes: label

If your changes are user facing and intended to be a part of release notes, please use a label starting with release notes:.

If not, please add the topic: not user facing label.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "topic: not user facing"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

@nWEIdia nWEIdia added ciflow/binaries Trigger all binary build and upload jobs on the PR release notes: build release notes category labels Feb 10, 2026
@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Feb 10, 2026

Two follow-ups needed:

  1. pytorch is switching to the triton commit that uses ptxas from cu13.1
  2. in between the triton introduced the TRITON_PTXAS_BLACKWELL_PATH, which this compiler.py is not aware of.

@bdhirsh bdhirsh requested a review from tinglvv February 10, 2026 23:20
@bdhirsh bdhirsh added the triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module label Feb 10, 2026
@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Mar 30, 2026

@pytorchbot rebase

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

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

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Successfully rebased main-revert-ptxas-cu13-bundle onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout main-revert-ptxas-cu13-bundle && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the main-revert-ptxas-cu13-bundle branch from 4028921 to 84e3531 Compare March 30, 2026 23:19
@pytorch-bot pytorch-bot Bot added the ciflow/torchtitan Run TorchTitan integration tests label Mar 30, 2026
@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Mar 31, 2026

@pytorchbot rebase

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

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

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Successfully rebased main-revert-ptxas-cu13-bundle onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout main-revert-ptxas-cu13-bundle && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the main-revert-ptxas-cu13-bundle branch from 84e3531 to 1935092 Compare March 31, 2026 18:04
@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 2, 2026

@pytorchbot rebase

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

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

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Successfully rebased main-revert-ptxas-cu13-bundle onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout main-revert-ptxas-cu13-bundle && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the main-revert-ptxas-cu13-bundle branch from 1935092 to 786e656 Compare April 2, 2026 20:12
@nWEIdia nWEIdia requested review from atalman and malfet April 5, 2026 02:55
Copy link
Copy Markdown
Contributor

@atalman atalman left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm

@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 21, 2026

@pytorchbot rebase

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

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

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Successfully rebased main-revert-ptxas-cu13-bundle onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout main-revert-ptxas-cu13-bundle && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the main-revert-ptxas-cu13-bundle branch from 786e656 to 8abcbd1 Compare April 21, 2026 23:36
@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 21, 2026

The follow up question is:
would any pytorch unit test on any platform (most likely dgx spark and THOR) require to invoke ptxas-blackwell binary ? As it currently stands (after merging this PR), they would only get the ptxas binary, which is

""ptxas": "12.9.86"

@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 22, 2026

Previously without bundled ptxas, dgx spark or Thor users do not have a convenient way of getting ptxas (since triton did not ship a cuda 13.1 ptxas as it current does), so the bundled ptxas -- for people who know, would be a last resort to solve the problem without installing CTK.

Now, even when the PyTorch code (compile_task.py) only uses ptxas (not ptxas-blackwell in triton terminology), advanced users can still manually rename ptxas-blackwell.

@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 22, 2026

@pytorchbot rebase

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

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

nWEIdia added 2 commits April 22, 2026 19:12
12.9), which fixes pytorch#163801
   permanently.
Reverting the bundling of ptxas, as even without ctk, users could get it
from simply pip installing triton.
@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Successfully rebased main-revert-ptxas-cu13-bundle onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout main-revert-ptxas-cu13-bundle && git pull --rebase)

@pytorchmergebot pytorchmergebot force-pushed the main-revert-ptxas-cu13-bundle branch from 8abcbd1 to 55f4526 Compare April 22, 2026 19:12
@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 23, 2026

@pytorchbot merge

@pytorch-bot pytorch-bot Bot added the ciflow/trunk Trigger trunk jobs on your pull request label Apr 23, 2026
@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 23, 2026

@pytorchbot merge

@pytorchmergebot
Copy link
Copy Markdown
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

@nWEIdia
Copy link
Copy Markdown
Collaborator Author

nWEIdia commented Apr 23, 2026

@pytorchbot merge -i

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

Labels

ciflow/binaries Trigger all binary build and upload jobs on the PR ciflow/inductor ciflow/torchtitan Run TorchTitan integration tests ciflow/trunk Trigger trunk jobs on your pull request Merged module: inductor open source release notes: build release notes category triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants