[fix] allow exporting custom Metal kernels without a Metal backend#3650
Open
xthomaswang wants to merge 1 commit into
Open
[fix] allow exporting custom Metal kernels without a Metal backend#3650xthomaswang wants to merge 1 commit into
xthomaswang wants to merge 1 commit into
Conversation
Author
|
This now conflicts with #3638, which made the tracing state thread-local. My plan is to rebase on main and make the new |
Collaborator
|
This basically looks good to me, a rebase and changing the counter to threadlcoal would be enough. |
- Move metal_kernel graph construction to backend/common so it builds on all platforms - During export tracing without Metal, record the kernel on a placeholder GPU stream which the importing process remaps to a real stream - Outside of export tracing, calling the kernel without Metal still raises an error Fixes ml-explore#3240
xthomaswang
force-pushed
the
fix/no-metal-custom-kernel-export
branch
from
a431ede to
7cbaeac
Compare
Author
Done, rebase on main and change counter to threadlocal. |
zcbenz
reviewed
Jun 14, 2026
| if ((device && *device != Device::gpu) || | ||
| (stream && stream->device != Device::gpu) || | ||
| (tl_stream && tl_stream->device != Device::gpu)) { | ||
| throw std::invalid_argument("[metal_kernel] Only supports the GPU."); |
Collaborator
There was a problem hiding this comment.
Is it necessary to require a gpu stream here? Can we just return to_stream(s) for exporting?
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Proposed changes
Fixes #3240
mx.fast.metal_kernelthrows at construction on non-Metal builds, so functionsusing custom Metal kernels can't be exported from Linux.
metal_kernelgraph construction to backend/common so it builds on all platformsThe same mechanism could support exporting
mx.fast.cuda_kernelfrom non-CUDAplatforms; happy to do that as a follow-up.
Checklist
Put an
xin the boxes that apply.pre-commit run --all-filesto format my code / installed pre-commit prior to committing changes