[johnmaia]

What

UploadPrepare merges overlapping regions. At the lease, a registered range takes a host import of exactly its size and records it in direct. When a descriptor range runs past that range's end, and into the next registered range, all three merge into one region. The merged region keeps the first range's import in direct.

UploadPrepare and UploadFinish then reuse that import while the import epoch is unchanged. Descriptor binds the view only up to the import's end and throws guest buffer view exceeds its GPU owner.

Both reuse sites now keep the import only while it covers the whole region. Otherwise the region looks an import up again (findImport already requires coverage) or is copied, as before.

Disney Epic Mickey: Rebrushed (PPSA15657, UE4) stopped at its first presentation on this. A 0x50000-byte import was kept for a merged 0x80000-byte region, under a 0x7f800-byte storage view at its base.

Tested

Linux (Ubuntu 24.04, GCC), RX 7600 (RADV).

• agc_driver_recorder_tests gets crossingImportTests. It registers two adjacent 64 KiB ranges, imports both, and binds a 512-byte view across their boundary in an address-based build. On main it throws guest buffer view exceeds its GPU owner; with the change it passes.
• The full ctest passes, 497/497.
• tools/check_conventions.py and tools/check_test_registration.py pass.
• PPSA15657, on main with a local stack of open PRs: before, its first presentation fails with this error. With the change, the stale import is dropped once, and the title gets past it to the next stops (Draw with no pixel program fails: "prepared shader artifact is missing for the requested static ABI" (null pixel program) #1901 / fix(agc): prepare the null pixel program at draw when its variant was not prepared #1902, Rect-list draw aborts with "prepared rectangle artifacts are missing" when its pixel variant was prepared at draw #1927 / fix(agc): prepare a rect-list draw's missing rectangle at draw from its prepared stages #1928, then layered color targets).

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed in TechnicalDebt
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• AI-assisted: yes

🤖 Generated with Claude Code

https://claude.ai/code/session_0134RWCyiyydJBbfKz8g9Bwe

[rmz-24]

Windows run: Windows 11 Pro 25H2 (10.0.26200.9457), RTX 5060 Ti (NVIDIA 617.42), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release, -DANYPS5_ENABLE_SPIRV_TOOLS=ON, full build plus libs. b597c7f cherry-picked onto current main d70b899 (applies cleanly).

• agc_driver_recorder_tests (not in ctest, run directly): crossingImportTests runs and passes on this host; the suite then stops at the unrelated refresh-store assertion from agc_driver_recorder_tests fails on Windows since #2096: the refresh store check, plus a more frequent device loss #2155, as main does on Windows. With the test change proposed in agc_driver_recorder_tests fails on Windows since #2096: the refresh store check, plus a more frequent device loss #2155 applied on top, the whole recorder suite passes.
• Full ctest: 509/510; the one failure was windows_import_modules, the intermittent Windows Error Reporting timeout fixed by test(relinker): launch the GUI startup fixtures without Windows Error Reporting #2227, unrelated to this change.

AI-assisted: yes

[oneandonlydean]

Closing: #1489 (merged since) drops the import in UploadPrepare when a merge takes a region past its end, which is this case. Please reopen if PPSA15657 still stops at guest buffer view exceeds its GPU owner on current main.