Dead code

[wujingyue]

A follow-up to #5177 (comment)

This change doesn't break any tests. Is that expected?

Possibly because of

Fuser/csrc/runtime/fusion_kernel_runtime.cpp

Lines 550 to 553 in e2d2264

	// TODO Remove binding the original fusion inputs when creating
	// heuristics for fusion segment.
	evaluator_precomputed_values->bindValues(
	group_to_run->getCompleteFusionInputs(), args);

?

[wujingyue]

!test

[github-actions]

Review updated until commit 9c82318

Description

• Removed dead code checking for concrete tensor domains

• Simplified logical domain cloning in fusion segmenter

• Eliminated unused variables and conditionals

• Improved code clarity and maintainability

---

Changes walkthrough 📝

	Relevant files
Bug fix	fusion_segmenter.cpp Simplify IterDomain cloning logic                                                csrc/fusion_segmenter.cpp Removed unused tv_is_concrete flag and loop Eliminated conditional logic for RFactor domains Simplified cloning to always use cloneWithoutRFactor Cleaned up redundant extent handling +1/-22

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 No relevant tests

⚡ Recommended focus areas for review Possible Issue The removal of logic that checks whether the logical domain contains all concrete sized extents and handles rFactor domains with symbolic extents may lead to incorrect cloning behavior for tensors with non-constant extents. The new code unconditionally clones without considering symbolic extent replacement, which was previously handled conditionally based on tv_is_concrete and isRFactorProduct. for (const auto& id : logical) { new_logical_domain.push_back(id->cloneWithoutRFactor()); }

[wujingyue]

!test

[jacobhinkle]

It's possible that segmentation changed since #630 was merged. One thing that changed is introduction of the ExprEval scheduler which now evaluates some of the segments so it won't hit the issue. I think we should look at that test DynamicTransformIssue418 and sprinkle in some pointwise ops to trigger the segmentation between two codegen'd segments again.

As to your change, it removes only the rFactor product flag and does not mark the extents for replacement, which would open us up to #418 and #629 again I think.