[S-TIR] Fix Segfault when applying Parallel during TIR schedule rewriting

[cchung100m]

Hi Commiters,

This PR is trying to fix issues #18424. Any suggestions would be appreciated if you are available.

Root Cause

Unsafe dynamic-shape dereferences in AdjustParallelVectorize The code assumed IntImm for buffer shape / loop extent and dereferenced directly. With dynamic shapes, as() can be null, which can segfault before any try/catch handles it.

Solution

Replaced unsafe IntImm assumptions with null checks and GetLoopIntExtent(...); if contiguous analysis is not possible, conservatively disables that path instead of dereferencing null.

[gemini-code-assist]

Code Review

This pull request wraps the logic within the Apply method in a try-catch block to handle runtime::Error. Feedback indicates that AdjustParallelVectorize contains unsafe dereferences that could lead to segmentation faults when encountering dynamic shapes, which would not be caught by this block. Additionally, the new try-catch block is considered redundant with the caller's error handling and should include logging to avoid suppressing error messages.

[cchung100m]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces safety checks for non-integer buffer shapes and loop extents within the AdjustParallelVectorize function, preventing potential crashes when dealing with dynamic shapes. The code now correctly identifies when contiguous memory access cannot be statically analyzed and resets the fusion count. A review comment suggests optimizing the buffer access loop by breaking early once a non-analyzable access is encountered, as the minimum fusion count is already reached.

[cchung100m]

Hi @tlopex @mshr-h ,

This PR is trying to fix issues #18424. Any suggestions would be appreciated if you are available. 😄