Simplify selfAllocationReplay

[wujingyue]

No description provided.

[wujingyue]

!test

[github-actions]

Review updated until commit ff8c560

Description

• Simplify selfAllocationReplay function

• Remove unnecessary reduction filtering

• Update replay logic to include all allocation IDs

---

Changes walkthrough 📝

	Relevant files
Enhancement	transform_replay.cpp Simplify selfAllocationReplay logic                                            csrc/transform_replay.cpp Removed reduction filtering from self_logical Updated ReplaySelf to use all self_allocation IDs Simplified logic for pushing reduction IDs +3/-11

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 No relevant tests

⚡ Recommended focus areas for review Possible Issue The change in self_logical from TensorDomain::noReductions(self->logical()) to self->logical() might lead to incorrect behavior if reduction IDs are present and need to be excluded. auto self_logical = self->logical(); Possible Issue The removal of self_allocation_no_reduction and the change in ReplaySelf instantiation might cause issues if non-reduction IDs are necessary for correct replay behavior. ReplaySelf replay(self_allocation, axis_map); Code Clarity The comment // NOLINTNEXTLINE(modernize-use-emplace) should be removed or justified if emplace_back is not applicable. new_contiguity.push_back(std::nullopt);

[wujingyue]

!test --diff