Upstream PR 2998 Uniform Copy Dirtiness

Upstream PR #2998 — Derive uniform dirtiness during copying

Upstream work · Actionable research · Research and retired work

Current disposition — November 1, 2026

Reviewed: November 1, 2026. Fork source: b7140848 (full SHA b71408480a3a979a9c584663e1d1f400904c8bb7). Latest stable checked: v0.8.136-0-g458730bf5373, September 30, source 458730bf5373f1f0d027450fef7443e3af7a3ae1. GitHub state, branch recommendation, source presence and release availability are separate facts. See current status and release versus main.

Current field	November 1, 2026 observation
GitHub source	Upstream #2998
GitHub state / draft	Open / No
Current head	651079ce4de102ba75e8ea768f47a440def9349b
Originally reviewed head	651079ce4de102ba75e8ea768f47a440def9349b
Last metadata update (UTC)	2026-10-18T20:59:11Z
Head versus original review	Same recorded head; metadata/state updates do not indicate a new proposal-code head.
Current action	Open proposal; refresh integration and focused tests before handoff.
Fork/release availability	This upstream proposal’s GitHub state does not establish full fork/release inclusion. The historical overlap below is scoped to its recorded fork baseline; use the current feature register for delivered fork behavior.

The current head matches the recorded September review head. Later updated_at metadata does not establish changed proposal code. The original fork overlap remains a dated source comparison, not a fresh integration or native qualification result.

This refresh checked GitHub identity/state and the specific source boundaries named above. It did not rerun native games, XISO timings or integration qualification. Historical performance claims remain attached to their original heads.

Historical review — September 21, 2026

The review below is preserved as written, including its original state, head, fork baseline and proposed handoff. Its present-tense statements apply to that dated review; the current disposition above governs today’s action.

← Upstream Work

Field	Value
Upstream PR	xemu-project/xemu#2998
State at 2026-09-21 audit	Open
Reviewed upstream head	651079ce4de102ba75e8ea768f47a440def9349b
Last upstream update seen	2026-09-06
Reviewed upstream base	f9b14039e5bb56ae2d8f028e31e7cc19f13f7e12
Fork baseline	82d04d84bdb3de956f20c3537e1da2cece0e8da2
Relationship	Core mechanism already present downstream; companion fixes available
Related fork work	Mainkill1/xemu#138, Mainkill1/xemu#148
Contribution priority	Highest
Upstream contact status	No comment posted by this audit.

Summary

The upstream PR replaces unconditional copies plus whole-block hashing with comparisons during reflected uniform copying. The fork already uses the same compare-and-copy shape and has two useful follow-ups.

Upstream scope

The proposed uniform_copy() determines dirtiness while copying reflected elements. Upstream also preserves the important binding-switch rule: values matching a binding’s own previous allocation do not imply that the currently bound GPU allocation already contains them.

Corresponding fork work

Fork #138 fixes deterministic construction of pixel-shader bump uniforms by initializing unused stage-zero entries that the renderer could still compare/copy. Fork #148 separately avoids repeated scans of clean vertex-uniform row flags and covers ordinary writes, transform/RDI paths, CPU vertex processing, and snapshot restoration.

Difference that must remain explicit

The stage-zero initialization defect predates this upstream optimization; it should not be blamed on #2998. The dirty-row summary is a separate optimization and should not be presented as required for removal of the full-block hash.

Transferable regression and review cases

• Construct identical fragment-uniform values into buffers prefilled with different sentinels; every compared byte must match.
• Run the producer under MemorySanitizer where supported.
• Switch between bindings with equal values and confirm the selected allocation is still reconciled.
• Verify reflected padded/strided arrays compare and copy only their defined elements.
• Measure actual UBO staging and descriptor writes separately from CPU reconciliation calls.

Suggested upstream handoff

Offer #138’s deterministic producer regression first. Acknowledge upstream’s existing binding-switch handling. Mention #148 only as a separate follow-up for row-dirty scanning.

Claims this page does not support

• It does not prove the upstream branch merges cleanly with current upstream.
• It does not prove a native correctness or performance result unless a linked exact-head run says so.
• It does not transfer favorable performance percentages between hardware, renderers, games, or source heads.
• It does not authorize posting a comment or review upstream.

Before sending anything upstream

• Refresh the PR state, head SHA, description, changed files, and newest review comments.
• Recheck the corresponding fork implementation on current main.
• Run or cite the narrow regression on an exact source state.
• Verify every public link and remove private paths, captures, game data, credentials, and internal logs.
• Phrase source concerns as source concerns and reproductions as reproductions.
• State the uncovered behavior and evidence limit.
• Add the posted URL and upstream response to this page.

Tracking log

Date	Upstream head	Action	Result
2026-09-21	651079ce4de1	Fork/upstream source comparison recorded	Handoff prepared; nothing posted

References

• Upstream PR #2998
• Fork repository
• Upstream Work
• Pull Request Lessons