Open
Conversation
…include gfx1250 rmsnorm
Contributor
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
|
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
vgokhale
reviewed
Apr 26, 2026
| sharedLayoutWeights: gl.constexpr = gl.SwizzledSharedLayout(1, 1, 1, order=[0]) | ||
|
|
||
| # create a swizzled shared layout for the output | ||
| gl.SwizzledSharedLayout(1, 1, 1, order=[1, 0]) |
Contributor
There was a problem hiding this comment.
This isn't assigned to anything but is probably also not needed since the output is not TDM stored so you don't need a shared layout for it?
vgokhale
reviewed
Apr 26, 2026
|
|
||
| # Loop through the rows of the input tensor by NUM_PROG blocks | ||
| for row_idx in range(row_start, n_rows, NUM_PROG): | ||
| input_ptr + (row_idx * input_row_stride) |
Contributor
There was a problem hiding this comment.
This isn't assigned to anything?
vgokhale
reviewed
Apr 26, 2026
| rms_norm = a * norm_factor * weights | ||
| # store rms norm and the norm factor | ||
| gl.store( | ||
| rsigma_ptr + row_start, norm_factor.to(rsigma_ptr.dtype.element_ty) |
vgokhale
reviewed
Apr 26, 2026
| USE_BLOCK = COL > BLOCK_SIZE | ||
| NUM_PROG = min(ROW, get_num_sms()) | ||
|
|
||
| grid = (NUM_PROG,) |
Contributor
There was a problem hiding this comment.
I think you can put min(ROW, get_num_sms()) here.
vgokhale
reviewed
Apr 26, 2026
| output = torch.empty_like(input, device=input.device) | ||
| rsigma = torch.empty((ROW,), device=input.device, dtype=input.dtype) | ||
|
|
||
| MAX_FUSED_SIZE = 65536 // input.element_size() |
Contributor
There was a problem hiding this comment.
Comment for the magic number?
vgokhale
reviewed
Apr 26, 2026
| @@ -0,0 +1,39 @@ | |||
| # SPDX-License-Identifier: MIT | |||
Contributor
There was a problem hiding this comment.
I don't think we want two different files. We want a single API and the wrapper decides whether to call triton (gfx950 and earlier) or gluon (gfx1250, if a gluon kernel exists).
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.
Motivation
Create rmsnorm kernel in gluon for gfx1250
Technical Details
Translated existing triton implementation into a gluon equivalent.
Test Plan
Added a test reference in existing test_rmsnorm.py for gluon implementation.
Test Result
Passed all test condition