[S-TIR] Fix the ScheduleError from missing guard in CheckInline

[cchung100m]

Hi Commiters,

This PR is trying to fix issues #18380. Any suggestions would be appreciated if you are available.

Root Cause

The ScheduleError comes from a missing guard in CheckInline before calling GetScopRoot

Solution

Skip Inline if this is the root block

[gemini-code-assist]

Code Review

This pull request introduces a check in AutoInlineNode::CheckInline to prevent the inlining of root blocks by returning InlineType::kNoInline if the block's parent reference is null. This ensures that blocks without a parent scope are correctly identified as non-inlinable. I have no feedback to provide as no review comments were submitted.

[tlopex]

The current guard block_sref->parent == nullptr only handles the case where the block has no parent at all.

That is not sufficient here: if a root block is wrapped in a For loop, then parent is non-null, but there is still no enclosing SBlockNode. In that case, GetScopeRoot will still throw RootBlockError when it walks up the parent chain and fails to find an SBlockNode.

I think the guard should mirror the logic in GetScopeRoot, e.g. by traversing the parent chain and checking whether any ancestor is an SBlockNode. If none is found, return InlineType::kNoInline.

It would also be good to add a test where a root block is wrapped in a For loop, since the current check would miss that case.

  const StmtSRefNode* p = block_sref->parent;                   
  for (; p != nullptr; p = p->parent) {                                                   
    if (p->stmt->IsInstance<SBlockNode>()) break;
  }                                                                                       
  if (p == nullptr) {                                           
    return InlineType::kNoInline;
  }

[tlopex]

LGTM! Thanks

[cchung100m]

Thanks to @tlopex for the prompt reply and suggestions. 😄