fix(shader): preserve pixel helper lanes in EXEC copies

[santhoshkumar-dev]

What

A pixel shader that enables whole-quad mode, copies EXEC to an SGPR, and restores that copy loses its helper lanes. hostExecWord intersects the saved mask with a Vulkan ballot of active invocations, which excludes those helpers. Subsequent interpolation leaves incorrect coordinates in the helper lanes, so implicit texture sampling selects the wrong mips. OlliOlli World (PPSA02327, #5316) shows maze patterns on characters and terrain, including dashed triangle edges on the Radlandia map.

Pixel-stage EXEC reads now intersect against the whole-quad expansion of the host active-lane mask. This retains the helper bits already present in guest EXEC while still rejecting absent host quads. Other stages retain the existing host-lane mask. There are no title checks, fixed mip levels, sampler overrides, or diagnostic environment switches in this change.

The mask semantics follow the AMD RDNA 2 ISA, section 12.3, S_MOV_B64 and S_WQM_B32/B64, printed page 111. Vulkan permits helper invocations to be inactive for non-quad subgroup operations; see Helper Invocations. Console hardware was not measured.

Tested

Windows 11 Pro build 26200, GCC 15.2 Release, AMD Radeon RX 6700 XT, AMD Vulkan 26.8.1 / Windows display driver 32.0.21045.5002.

• Windows system-library build passed. Eight focused CTest cases passed with Vulkan required and no skips: shader memory, masked-select optimization, host EXEC lanes, LOD clamps, barycentric emulation, pixel helpers, scalar SOP1, and the new implicit-derivative regression.
• The regression draws 24 skewed triangles with individually marked mip levels. It saves/restores EXEC after WQM and checks mip selection at every covered pixel for wave32/wave64, native/emulated barycentrics, and expected mip levels 0/1/2. Each of the 12 cases checks 7,817 pixels with zero incorrect selections.
• Negative control on unmodified main d92a8da: the first case fails with 983 incorrectly sampled pixels. Restoring only this EXEC correction makes all 12 cases pass. The branch was then rebased onto db81992, whose intervening changes add optional draw-program tracing, and the library build and eight tests were rerun.
• OlliOlli World v01.005.000: the map and tutorial NPCs render cleanly with normal mip selection; the user confirmed both scenes. The game check used d92a8da plus only the EXEC correction, with the shader disk cache disabled and all rendering overrides removed. It ran for 215 seconds and exited normally with code 0. No FPS or audio improvement is claimed.
• Repository conventions and whitespace checks passed. Linux, other GPU vendors, and a full CTest-suite run were not tested locally.

Before and after screenshots

Original screenshots supplied by the tester. The after images show the EXEC correction with normal automatic mip selection.

Scene	Before EXEC correction	After EXEC correction
Characters
Radlandia map

Original PNGs and screenshot provenance. Camera positions and UI state differ; these images document rendering correctness.

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; no stubs added
• No notes, investigation 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: OlliOlli World PPSA02327 OlliOlli World (PPSA02327) #5316 (texture corruption)
• GPUs tested: AMD Radeon RX 6700 XT, AMD Vulkan 26.8.1 / display driver 32.0.21045.5002, Windows 11 Pro build 26200
• AI-assisted: yes (Codex)

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 6326	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1860
PR 6341	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1860
PR 6606	core/libs/prx/libSceAgcDriver/CMakeLists.txt L1860
PR 7410	docs/dev/debt/shader.md L80

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 1341	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 1368	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 1395	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 2434	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 4650	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 4891	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5021	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 5371	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 5426	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 5431	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5499	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5500	docs/dev/debt/shader.md
PR 5925	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 5941	docs/dev/debt/shader.md
PR 6009	docs/dev/debt/shader.md
PR 6044	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6185	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6186	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6188	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6192	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6218	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6234	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6235	docs/dev/debt/shader.md
PR 6280	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 6300	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6320	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6360	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6412	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 6454	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6470	docs/dev/debt/shader.md
PR 6481	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6486	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6488	docs/dev/debt/shader.md
PR 6501	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6503	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6535	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6613	docs/dev/debt/shader.md
PR 6683	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6701	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6747	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6763	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6788	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6808	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6856	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6868	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6937	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6941	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6946	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6947	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6963	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6973	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6977	docs/dev/debt/shader.md
PR 6987	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6989	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 6995	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7002	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7085	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 7201	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7223	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 7332	core/libs/prx/libSceAgcDriver/CMakeLists.txt docs/dev/debt/shader.md
PR 7388	docs/dev/debt/shader.md
PR 7429	docs/dev/debt/shader.md
PR 7615	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7642	core/libs/prx/libSceAgcDriver/CMakeLists.txt
PR 7644	core/libs/prx/libSceAgcDriver/CMakeLists.txt

[zaryabzip]

NVIDIA and Intel run, for the "Linux, other GPU vendors, and a full CTest-suite run were not tested locally." line. Windows 11 (26200), RTX 3060 Laptop GPU (NVIDIA 617.42) and Intel UHD Graphics (device 0x9a60), WinLibs GCC 15.2.0 r7 (UCRT), Release, BUILD_TESTING=ON. Head 27b196bee merged onto main 8856f180c (clean); the GPU was picked with ANYPS5_GPU, and the verbose test logs show the selected device.

• agc_driver_pixel_implicit_derivatives with the PR: passes on both GPUs, incorrect=0 in all 12 cases.
• With ExecMask.cpp from main and the PR's test:
  • Intel UHD: fails at the first case, emulate0 wave64 mip0 covered=7817 incorrect=983, then AGC graphics: implicit derivatives selected the wrong mip at 983 pixels.
  • RTX 3060: passes, 3 of 3 runs, incorrect=0 in all 12 cases. The test does not detect the old behaviour on this NVIDIA GPU.
• ctest -R "agc|shader|guest_memory|write_watch" -j4 on the RTX 3060: 592/592 pass with the PR, and 591/591 on main (the new test is the difference). Both skip the same 3 tests (agc_host_import_sparse, agc_driver_sparse_dma_buf_import, agc_driver_stencil_ref_export).

Not run: OlliOlli World, Linux, the suite on the Intel GPU.

AI-assisted: yes (Claude Code)