NVIDIA / NVIDIA/cutile-python

[FEA]: Support a MetaTensor that does not hold data for debug purposes.

Open
#56 11 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

feature request status: waiting-for-feedback
Dominant language
Python
Stars
2.2k
Forks
155
PR merge metrics
No merged PRs in 30d

Description

Is this a new feature, an improvement, or a change to existing functionality?

New Feature

How would you describe the priority of this feature request?

High

Please provide a clear description of problem this feature solves

Is it possible to support a MetaTensor that does not hold data for debug purposes?

Feature Description

We can utilize this to run the frontmost part of cuTile (python->hir->ir->TileIR) and either dump IR from each level for inspection or build tools like transpiler without having to create real data on GPU and even calling tileiras.

This comes from my personal experiments; (1) I have to develop on machines without Blackwell GPUs; (2) I have tried to implement a LSP server and cuTile -> numpy transpiler for debug purposes, but found it hard to catch up to the rapid development of cuTile mainstream.

Describe your ideal solution

If possible, I would like to discuss what the interface looks like and/or implement this.

Describe any alternatives you have considered

No response

Additional context

No response

Contributing Guidelines
  • I agree to follow cuTile Python's contributing guidelines
  • I have searched the open feature requests and have found no duplicates for this feature request

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue names no files, tests, or concrete entry points. Start by tracing the frontmost cuTile Python-to-HIR-to-IR-to-TileIR flow and review how MetaTensor currently stores data. Done would require an agreed interface for a data-free MetaTensor and support for dumping or inspecting IR without real GPU data.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
compilers
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.