Create a replayDomain utility to allow reuse

[Priya2698]

Separating out replaying domain to a function to allow reuse.
For Issue #5079, PR #5090 will also replay allocation.

[Priya2698]

!test

[github-actions]

Review updated until commit e086d0f

Description

• Created reusable replayDomain utility function

• Applied replay logic consistently across domains

• Propagated padding and parallelization selectively

• Fixed axis mapping with proper error checks

---

Changes walkthrough 📝

	Relevant files
Enhancement	transform_rfactor.cpp Introduced replayDomain utility and replaced inline replay logic csrc/transform_rfactor.cpp Extracted repeated replay logic into replayDomain function Added parameters to control padding and parallelization propagation Used replayDomain for producer and consumer domain replay Removed redundant loops and improved error messaging +61/-46

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 No relevant tests

⚡ Recommended focus areas for review Default Parameter Usage The function replayDomain uses default parameters for ignore_ids, propagate_padding, and propagate_parallelization. However, passing empty default containers by value could lead to performance overhead or unintended behavior. Consider using const references for the default container arguments to avoid unnecessary copying. std::vector<IterDomain*> replayDomain( const std::vector<IterDomain*>& replay_domain, const std::unordered_map<IterDomain*, IterDomain*>& replay_to_target_map, const std::unordered_set<IterDomain*>& ignore_ids = {}, bool propagate_padding = false, bool propagate_parallelization = false) { Redundant Condition Check The condition propagate_parallelization || replay_id->isDeviceDim() || replay_id->isStream() may lead to unintended parallelization behavior. Since device and stream parallel types are always preserved, the propagate_parallelization flag might be redundant in those cases. This could cause confusion or incorrect assumptions about when parallelization is applied. if (propagate_parallelization || replay_id->isDeviceDim() || replay_id->isStream()) { target_id->parallelize(replay_id->getParallelType()); } Loop Index Increment in Lambda In the original code, there was an incorrect use of i++ inside a lambda within std::transform, which has been removed. While the new code avoids this issue by using replayDomain, it's important to confirm that all uses of the utility function correctly handle index management and iteration without relying on side effects. std::vector<IterDomain*> new_producer_domain = replayDomain( original_td->loop(), original_to_producer_id_map, /*ignore_ids=*/{}, /*propagate_padding=*/true, /*propagate_parallelization=*/true);

[naoyam]

LGTM

[jjsjann123]

lgmt

[Priya2698]

!test --diff

[Priya2698]

@xwang233 I can't access the codediff artifact. Any suggestions?

[xwang233]

it's not pushed to web artifact yet, please check again later