[Dextroid17]

What

Three writes in agc_driver_recorder_tests change guest memory directly and never call GuestMemory::MarkWritten: the two memsets (0x22, 0x33) in resourceReadTests and guest[offset + 8] ^= 0xffu in misalignedRegionTests. Since 98ce8af (#5906, draw snapshot reuse) the draw snapshot trusts the stamp, so it hands back the old bytes and the checks second draw observed overwritten input and the shader's patched offset into the draw snapshot misses the view's bytes fail. This adds a MarkWritten after each write, 3 lines in tests/Recorder.cpp. No driver code changes.

This only shows on top of the (9) a scan under a pending write was kept fix, because that check stops the run first. That fix is #5883 (the same hunk as 6bd076c), so this PR is stacked on it.

Discussion and the bisect: #1532.

Tested

• Lavapipe (llvmpipe, Mesa 26.2.4, LLVM 23.1.1; CachyOS, Linux 7.2.9), default CMake build (CMAKE_BUILD_TYPE empty, so unoptimized), main ab9a476. Branch plus the test(agc): account for unknown pending DCC key writes #5883 hunk: agc_driver_recorder_tests, no arguments, exit 0, Recorder read tracking and label tests passed, 3 of 3, about 2 s each.
• Without the stamps (main plus the test(agc): account for unknown pending DCC key writes #5883 hunk only) the run stops at second draw observed overwritten input. With the two memsets stamped it stops next at the shader's patched offset into the draw snapshot misses the view's bytes (3 of 3, main eea428c). With the (9) hunk, the run passes at 11adb34 (parent of 98ce8af) and stops at second draw observed overwritten input at 98ce8af, 2 of 2 each.
• python tools/check_conventions.py --base upstream/main passes.
• fma-rc reported the same change (rebuilt from my description on agc_driver_recorder_tests: on Linux NVIDIA a reported import write does not invalidate a cache, and the test has never run in CI #1532) passing 3 of 3 on RTX 5080, RADV and lavapipe. I did not run those.
• Not tested: this branch alone without the test(agc): account for unknown pending DCC key writes #5883 hunk (it stops at the (9) check), Windows, any title. This does not show the driver is correct for a CPU write that is not stamped; it only changes the test.

Checklist

• Based on current main; no other open PR implements the same functions (test(agc): account for unknown pending DCC key writes #5883 is the (9) hunk only; it does not touch these writes)
• One topic per PR; follow-ups go in a new PR
• No comments in code
• Unimplemented paths throw: n/a
• No notes or images added
• No third-party code
• Generic behaviour, not specific to one title
• Depends on: test(agc): account for unknown pending DCC key writes #5883
• Unblocks: none
• GPUs tested: lavapipe (llvmpipe, Mesa 26.2.4), Linux
• AI-assisted: yes

[github-actions]

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 1368	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 1395	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 5021	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 5925	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 6188	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 6222	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 6373	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 6747	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 6946	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp
PR 6947	core/libs/prx/libSceAgcDriver/tests/Recorder.cpp