Skip to content

[CuTeDSL] Make the dynamic-layout guide use explicit singleton strides - #3757

Open
0z5a wants to merge 1 commit into
NVIDIA:mainfrom
0z5a:docs/dynamic-layout-guide-strides
Open

0z5a wants to merge 1 commit into
NVIDIA:mainfrom
0z5a:docs/dynamic-layout-guide-strides

Conversation

@0z5a

@0z5a 0z5a commented Oct 20, 2026 •

Copy link
Copy Markdown

The mark_layout_dynamic example says dimensions 0 and 2 have stride 1, but its permute produces stride 4 for both. mark_layout_dynamic(leading_dim=0) therefore raises before the example can print its layout.

Construct the intended layout with torch.empty_strided, and print t1 / t2 after defining those variables.

Validated the six positive cases t0–t5 on GH200 (main editable core, torch 2.11.0) and H100 (DSL 4.8.1, torch 2.13.0). The original and print-only repair both fail; explicit strides complete all six cases with the expected layouts. Deliberate error snippets elsewhere in the guide are outside this change.

Fixes #3760.

Signed-off-by: 0z5a <192209249+0z5a@users.noreply.github.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[DOC] Dynamic-layout guide example uses incompatible singleton strides

2 participants