fix(agc): render a nonzero slice of an XOR color view with MIP0_DEPTH 0 into its array slice

[Tamalero]

What

A color view of a nonzero slice with CB_COLOR_ATTRIB3 MIP0_DEPTH = 0 renders into the wrong bytes under the 64 KiB XOR swizzle modes. High on Life (PPSA15388) draws every face of its sky-light cube this way, and the cube comes back scrambled. That is the stale-mip pattern in #4736.

The downsample pass draws face f with:

• CB_COLOR_VIEW = 0x1c00a005 for face 5 (slice start = slice end = f);
• CB_COLOR_ATTRIB3 = 0x4dc6c000 (MIP0_DEPTH 0, 2D, SW_64KB_R_X).

Cause: DecodeColorBuffer keeps the array slice only when MIP0_DEPTH is nonzero. With MIP0_DEPTH 0 it adds slice × slice size to the address and renders a standalone 2D surface there, so the slice term of the swizzle is lost. In the R_X equations in TextureSwizzleEquations.hpp (from addrlib's GFX10 non-RB+ pattern tables), at 8 bytes per element, byte-address bits 8–11 are XORed with slice bits 3–0. Slice 0 is unaffected. For slices 1–5 the bytes land in permuted 256-byte blocks, and the cube texture that later reads slice f sees them in the wrong order. The pass also downsamples into mip tails, which share a 64 KiB block, so the permutation mixes mips there.

Fix: a nonzero slice of an XOR color view with MIP0_DEPTH 0 becomes slice slice of an array of max(slice, last slice) + 1 layers at the surface base. That is the path MIP0_DEPTH ≠ 0 already takes. Linear and non-XOR views are unchanged.

The source is the slice term in the existing addrlib-derived equations. This has not been measured on hardware: I have no capture showing the CB applies the slice term when MIP0_DEPTH is 0. The game's own use is the evidence. It draws face f this way and samples the six faces as a cube, so the slice term must apply.

Tested

• Linux, GCC 16, Release, RTX 4090 (NVIDIA 615.78).

• agc_driver_array_layer_target gains a case that draws into slice 1 with MIP0_DEPTH 0 and reads slice 1 back through a 4-layer array texture. It checks the result both while pending and after a re-upload from guest memory.

  • Negative check: without the fix the new case fails ("the color view did not keep its array and slice"). The existing MIP0_DEPTH 3 cases pass with and without the fix.

• Full agc_ suite on this branch: 464 of 464 pass (agc_driver_stencil_ref_export skipped on this device).

• tools/check_conventions.py --base origin/main passes.

• High on Life (PPSA15388), on a local stack: main 3377986a plus feat(agc): render layered draws into 3D color targets #1077, feat(agc): depth planes sampled as 2D arrays, cubes and integer views #425, fix(agc): bind a buffer descriptor past the range limit by its committed bytes, or empty #3202 and this change. The before values come from earlier runs on the same stack without it.

  Sky-light cube	Before	After
  Faces 1–5 of the downsampled cube	mips 3–7: 25–81 % stale texels; face 2 mip 5 holds face 4's data	every face smooth from 128² down to 1×1
  Cube B (GGX-filtered)	G ≈ 234 on faces 1, 2, 3 and 5	~2–3 on every face
  1×1 capture brightness	2660	0.73
  Sky SH coefficients (9×1)	65504 / non-finite	2.09, −0.003, −1.16, …

  The scene is still dark, because indirect light is missing (see AGC: a cube's packed-tail mips rendered per face are not seen by the cube's sampled view; the reflection capture filters a clear value (High on Life PPSA15388, black lighting) #4736). Direct light now renders correctly.

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 under Silent stubs in their library's technical debt file
• 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
• Unblocks: PPSA15388 AGC: a cube's packed-tail mips rendered per face are not seen by the cube's sampled view; the reflection capture filters a clear value (High on Life PPSA15388, black lighting) #4736 (sky-light cube; not the whole black scene)
• AI-assisted: yes

🤖 Generated with Claude Code

[xXPBSXx]

Windows test (MinGW GCC 15.2, Release, Windows 10, RTX 4080 SUPER). Built on main 88107b7de together with thirteen other open GPU PRs (their own commits cherry-picked, no merge commits), so only the tests that belong to this PR say something about it:

• tests/execution/ArrayLayerTarget.cpp (agc_driver_array_layer_target): 3 of 3 runs passed.

Not tested: no game run. The batch was built once, not per PR; nothing failed, so no bisect was needed.

AI-assisted (Claude Code).

[Tamalero]

Closing as superseded by #4551, which fixes the same bug (a nonzero slice of an XOR color view with MIP0_DEPTH 0 loses its slice term) and merged today. #4551 keeps the shifted 2D surface and folds the slice's equation bits into the pipe/bank XOR; this PR rendered the view into the array slice instead. Both put the bytes where the hardware does. #4551 was opened first, from Lost Soul Aside's reflection cube (#4387), which is the same symptom as High on Life's in #4736.

Thanks @xXPBSXx for the Windows run. I'll recheck High on Life's cube on main with #4551 and report on #4736.

Tested and written by Claude (AI assistant) on a contributor's machine.