[PT2E][X86] Extend reused-module fallback to nn.Conv2d (follow-up to #4480)

[waynehacking8]

Summary

Follow-up to #4480 (reused nn.Linear fix), extending the same fix to nn.Conv2d. As @claude noted in the #4480 review, the conv annotation paths share the single-call-site assumption, so a reused nn.Conv2d (e.g. called inside an unrolled loop) under non-strict export still crashes prepare_pt2e:

• plain path → ValueError: conv partition has more than one output node (_annotate_conv2d)
• fusion paths → ValueError: Input partition has more than one output node (_get_output_nodes_of_partitions)

Stacked on #4480. This branch is based on fix-4478-reused-linear-fusion because the new test relies on _test_quantizer(..., strict=False) introduced there. Until #4480 merges, the diff here also shows the linear commits; I'll rebase onto main (conv-only) once #4480 lands. Happy to instead fold this into #4480 if you'd prefer a single PR.

Modifications

Mirror the linear fix:

• _annotate_conv2d_binary_unary / _annotate_conv2d_binary / _annotate_conv2d_unary: skip fusion when any partition has multiple output nodes, so a reused conv falls back to plain annotation.
• _annotate_conv2d: iterate partition.output_nodes and annotate each call site instead of raising on multi-output partitions.

Test

Adds SharedConvModule + test_conv2d_reused_with_fusable_post_op, covering the unary (relu), binary (add) and binary+unary (add+relu) fusion guards plus the plain fallback.

Coverage is real, not just "passes": with the fix reverted the test crashes with Input partition has more than one output node; with it, each call site is annotated as plain conv and prepare_pt2e succeeds (node_occurrence 2×q/dq-per-tensor + 2× weight dq, conv annotated at both sites, post-op unfused). Validated on torch 2.12 (CPU); ruff check/format clean.

---

This PR was written with the assistance of Claude (AI); I have reviewed and validated all changes on real hardware.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4482

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.