[xXPBSXx]

What

GuestBufferMemory::ViewBytes gives the descriptor range of a guest buffer view that starts adjustment bytes after a storage-buffer offset boundary (added in 38e89df). No test calls it directly; BufferUnalignedBase reaches it only through a whole draw. Add guestBufferViewBytesTests to agc_driver_graphics_tests:

• an adjustment of a whole number of DWORDs (0, 4, 16, 256) adds exactly the adjustment to the byte count;
• an adjustment of 1, 2, 3, 5 or 7 bytes adds the DWORDs around the view (4 * ceil(adjustment / 4)), the case 38e89df describes as "the range covers the dwords around it".

Byte counts are multiples of four (4, 16, 4096). The expected values follow from that commit message and the function; the test does not change behaviour and passes on main.

Tested

Release build, MinGW-w64, Windows 10 (19045), main 5d9dcc42d plus this commit (edb74cb58; merge-tree against main is clean). ctest -R '^agc_driver_graphics$': passed (7.4 s). With the first expectation changed to bytes + adjustment + 1 the same test fails with "a view whose adjustment is a whole number of DWORDs did not add the adjustment", so the new cases run. tools/check_conventions.py --base upstream/main: all checks passed. No Vulkan device is needed by the new function; the executable itself ran against the NVIDIA GeForce RTX 4080 SUPER driver. Linux/lavapipe not tested.

Checklist

• Based on current main; no other open PR implements the same tests (open PRs that edit tests/Graphics.cpp, fix(agc): keep the guest's sample order where pixel (0,0)'s locations stand in for a 2x2 grid #6683, fix(agc): read a multisampled target's CMASK by its MSAA codes #6669 and fix(agc): bind image and sampler tables through update-after-bind limits #6647, change other 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); no stubs added
• No notes, investigation .md files or images added to the repository
• New third-party code is a submodule built from source (none added)
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: none (test coverage only)
• GPUs tested: NVIDIA GeForce RTX 4080 SUPER, Windows 10 (19045); Linux and lavapipe not tested
• AI-assisted: yes

AI-assisted (Claude Code).

🤖 Generated with Claude Code

[xXPBSXx]

@adriadam10 a short request for review of this test-only PR (guestBufferViewBytesTests in agc_driver_graphics_tests: the DWORD rounding of GuestBufferMemory::ViewBytes for whole-DWORD and for odd adjustments, byte counts that are multiples of four; +13/-0, no behavior change). Head edb74cb58 is MERGEABLE / CLEAN, the Conventions, Eligibility and Intake checks are green, ctest -R ^agc_driver_graphics$ passes (and fails with a deliberately wrong expectation, so the cases run).

AI-assisted (Claude Code).

[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 1395	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 4040	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 4650	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 5441	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 5499	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 5500	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 5925	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6188	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6215	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6222	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6235	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6297	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6320	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6457	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6479	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6529	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6535	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6548	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6647	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6730	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6808	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6973	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6987	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6989	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 6995	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp
PR 7002	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp

[jotitadev]

The new cases pass, but they only cover the case where both branches of ViewBytes are right. Every byte count in the test is a multiple of four (4, 16, 4096). Off a DWORD boundary, the function returns 4 * (adjustment / 4 + bytes / 4 + 1). That truncates bytes / 4 on its own, so when adjustment % 4 + bytes % 4 > 4 the range stops before the view's last bytes:

bytes	adjustment	ViewBytes	bytes the view spans from the aligned start
7	2	8	9
6	3	8	9
3	3	4	6
2	3	4	5
4098	3	4100	4101

Those byte counts can happen: the size comes from ShaderBufferResource::GetSize(), which is Stride() * NumRecords(), or NumRecords() when the stride is 0, with no rounding (ShaderResources.cpp, the guest-buffer path that pushes {address, size, true, ...}). So a byte buffer with an odd record count, or a stride of 2 or 6, gets a descriptor range one to three bytes short of the bytes the shader may read. The same value also sizes the snapshot and overlap checks at the two ViewBytes(size, item.adjustment) call sites.

For the unaligned branch, the covering range is 4 * ((adjustment + bytes + 3) / 4). It equals the test's bytes + 4 * ((adjustment + 3) / 4) whenever bytes is a multiple of four. With that expectation and 7ull added to the outer list, adjustments 2, 3 and 7 fail on current main. Whether the fix belongs in this PR or a separate one is up to you. As it stands, the test only checks the formula where it agrees with the intended one.

Checked with a standalone copy of ViewBytes from this head, built with clang++ -std=c++20 on macOS: the PR's 27 cases pass, and the rows above come out as shown.

Written with AI assistance (Claude Code). I ran the checks above on my machine.

[MarcZX10]

The #6999 report confirms a real under-coverage in the off-DWORD branch. I am preparing a separate follow-up with the production formula correction and regression sizes 2, 3, 6, and 7. Since it touches the same test file, I plan to base it on your test commit and list Depends on: #6695, preserving your commit author. Please let me know if you would prefer a different landing order.

[MarcZX10]

The follow-up is now open as #7001, based on your test commit unchanged and declaring Depends on: #6695. The production formula and off-DWORD regression cases are included; conventions, eligibility and intake checks pass. Please let me know if you would prefer a different landing order.