[KMCompiler][ttx]opt for rope

[YangLong114514]

Description

The rope operator has been optimized for the ascend platform.

Changes

1. Modified the block_size calculation method to dynamically adjust the block_size value based on the UB upper limit of 192KB.

Performance

Using Ascend 910B and Triton 3.2.x of FlagTree:

shape	before	after	speedup= before/after
（1，4，32，64）	32.8846	27.9724	1.18
（2，8，1，32）	3.916	3.0602	1.28
（1，32，1024，32）	70.8734	67.5576	1.05
（2，16，512，128）	86.5778	76.9256	1.13
（1，32，4096，128）	342.8748	320.2106	1.07
（2，16，4096，128）	350.215	319.8304	1.10

Test

Correctness verified against baseline implementation across all shapes and dtypes listed above.

[gemini-code-assist]

Code Review

This pull request introduces optimizations and safety checks for the RoPE kernel on the NPU backend, including dynamic token block size estimation to prevent compile-time Unified Buffer (UB) overflow, simplified masking, and improved Triton loading and casting. The review feedback suggests replacing an iterative loop with an O(1) mathematical calculation to optimize block size limiting, and removing commented-out code to maintain codebase cleanliness.

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.