[Fix][Kernel] Low-shared-memory fused prepare_wy_repr + compute_w_u k…

[guangxing83]

Low-shared-memory fused prepare_wy_repr + compute_w_u kernel for ungated DeltaNet.

[gemini-code-assist]

Code Review

This pull request introduces a new low-shared-memory fused kernel, fused_prepare_compute_w_u_tl_maca, for ungated DeltaNet in TileLang, replacing the previous implementation in deltanet_fwd.py. Feedback on the new kernel highlights critical race conditions in the repeated squaring loop due to missing thread synchronizations (T.sync_threads()). Additionally, it is recommended to enforce that the dynamically computed RHS tile size block_R is a multiple of 16 to ensure Tensor Core alignment and prevent performance degradation or compilation failures.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.