[Lancito01]

What

agc_driver_mesh_barycentric_emulation fails on every run on a device with VK_KHR_fragment_shader_barycentric since #9193:

AGC graphics: the pixel shader did not emulate barycentrics

#9193 added SpirvTarget::fragmentBarycentricBuiltins and cleared it in PixelBarycentricEmulation's Draw, but MeshBarycentricEmulation's Draw still copies it from the device target and only clears fragmentShaderBarycentricEnabled. With the built-ins left on, HostInterpolationChecker::Lower reads the I/J VGPRs from BaryCoordKHR, so barycentricEmulation.active stays false and the test stops at its first Require. This sets target.fragmentBarycentricBuiltins = false there, as #9193 did for the pixel test. Test-only change.

Tested

Windows 10 22H2 (19045), WinLibs GCC 15.2.0 r7, Release, BUILD_TESTING=ON, all AGC execution tests built. ctest --test-dir build -R "^agc_driver_(mesh|pixel)_barycentric_emulation$" -V with ANYPS5_REQUIRE_VULKAN=1, ANYPS5_NO_SHADER_CACHE=1 and VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation, 5 fresh-process runs each:

tree	agc_driver_mesh_barycentric_emulation	agc_driver_pixel_barycentric_emulation
main 228cd64b538	fails 5/5 (the pixel shader did not emulate barycentrics)	passes 5/5
this change on 228cd64b538	passes 5/5 (mesh barycentric emulation tests passed: 932 perspective and 932 linear I/J pixels, 932 raw vertex value pixels)	passes 5/5
this PR, rebased on main 14332cd8d0	passes 3/3	passes 3/3

VK_LOADER_DEBUG=layer shows VK_LAYER_KHRONOS_validation inserted (Khronos nightly Windows build); no log contains VUID- or was not found but was requested by env var VK_INSTANCE_LAYERS. tools/check_conventions.py --base origin/main passes and tools/pr_overlap.py reports no overlap.

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
• Fixes: none
• Unblocks: none
• GPUs tested: NVIDIA GeForce RTX 3080 (10de:2216, Ampere), driver 610.62, Windows 10 22H2 (19045)
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

[MarcZX10]

Reviewed exact head 552d38908301cdbf6cb8a41ce7256de121f65a23. The one-line test setup change clears fragmentBarycentricBuiltins alongside fragmentShaderBarycentricEnabled, matching the existing pixel barycentric emulation setup and the described mesh-test failure. The PR includes repeated main-versus-change runs with validation enabled on an RTX 3080; this host has no compatible Vulkan device for independently reproducing those GPU results.\n\nValidation here: tools/check_conventions.py --base refs/remotes/upstream/main passed and git diff --check passed. No local GPU runtime result is claimed.

[milescoviello]

A Linux + NVIDIA data point for this head (552d389083), next to the Windows run in the description:

• Setup: Ubuntu 26.04 (Linux 7.0), GCC 15.2, Release, NVIDIA GeForce RTX 4070 Ti on 615.71.09, ANYPS5_REQUIRE_VULKAN=1, ctest -R '^agc_driver_mesh_barycentric_emulation$'.
• main 14332cd8d0: fails in each of 3 runs with the pixel shader did not emulate barycentrics.
• This head: passes 3 of 3 (932 perspective and 932 linear I/J pixels, 932 raw vertex value pixels).
• Validation layer: the Khronos layer 1.4.363 reports no VUID- on either side, with one Vulkan device created per side. That run used my local copy of the same line on the same base 14332cd8d0, before I found this PR.
• lavapipe on the same host (ANYPS5_GPU=llvmpipe): the test skips there (the device has no mesh shader compiler that handles these programs). That's why CI didn't catch the failure.

I came to the same one-line change from a validation-layer sweep of main, and dropped my branch once I found this PR.

AI-assisted: yes (Claude Code)

[Lancito01]

thanks for the review!

[maskym]

Windows + NVIDIA result for head 552d389083, merged into main f759e165d9 (the PR's GPUs tested line names no NVIDIA run on Windows 11 or on Turing).

This PR's own tests: agc_driver_mesh_barycentric_emulation passes (0.97 s); it fails on main on this device (Turing, driver 621.01, Windows 11), so this adds a Turing data point to the Ampere/Windows 10 one in the description, all passed.

Host: Windows 11 26300, Intel i7-12700K, NVIDIA GeForce RTX 2070 SUPER (vendor 0x10de, device 0x1e84; Vulkan loader 1.4.357 from the NVIDIA driver 621.01), Khronos Validation Layers nightly 1.4.366 (b2b44c0d37) through VK_ADD_LAYER_PATH, WinLibs GCC 15.2.0 (x86_64-ucrt-posix-seh, r6), Release, BUILD_TESTING=ON, Ninja.

Same focused run on main f759e165d9 and on this tree, same machine and GPU: VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation, ANYPS5_NO_SHADER_CACHE=1, ANYPS5_REQUIRE_VULKAN=1, VK_LOADER_DEBUG=error,warn,layer, ctest --test-dir build -R '^(agc|guest_video_out|video_out)' --parallel 4 -V.

• main: 1127/1128 passed (9 skipped), the failure being agc_driver_mesh_barycentric_emulation (AGC: agc_driver_mesh_barycentric_emulation fails since #9193 on devices with mesh shaders and VK_KHR_fragment_shader_barycentric #12121).
• This tree: every pre-existing test passes, including that one (the tree carries test(agc): disable barycentric built-ins in the mesh barycentric emulation test #12062, its fix); the only tests not run are feat(libSceVideoOut): limit the guest display to the host resolution #11948's three, whose executable fails to link on Windows (reported on that PR). The loader inserted VK_LAYER_KHRONOS_validation in every test (998 and 999 insertions), neither log contains was not found but was requested by env var VK_INSTANCE_LAYERS, and both report exactly one VUID, VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605 18 times (pre-existing on main, AGC: pipelines use differing per-attachment blend states while independentBlend is never enabled (VUID-VkPipelineColorBlendStateCreateInfo-pAttachments-00605) #11990 / fix(agc): enable independentBlend for color targets with different blend states #6782): this PR adds no new VUID- error.

The tree also carried the open PRs #11948, #12132, #12070, #11680, #11679, #11678, #11677, #11675, #11674, #11605, #11604, #11600, all merged cleanly, so the run is a joint one. Not run on a PS5.

AI-assisted: yes (Claude Code), the runs and the numbers above are from this machine.

[CruddOCE]

AMD result for this PR's head 552d38908 merged into main 0cd07dc4a (merged locally without a commit).

AMD Radeon RX 9070 GRE (RDNA 4, 1002:7550), driver 32.0.32015.7029 (Adrenalin 26.10.1), Windows 11 Home 10.0.26200, Ryzen 5 7500X3D, WinLibs GCC 15.2.0 r7, Release, ANYPS5_GPU="RX 9070" ANYPS5_REQUIRE_VULKAN=1 ANYPS5_NO_SHADER_CACHE=1, serial CTest.

agc_driver_mesh_barycentric_emulation	main 0cd07dc4a	this PR
RX 9070 GRE	fails 3 of 3: the pixel shader did not emulate barycentrics	passes 3 of 3

The other five agc_driver_mesh* tests also pass with the PR on this device.

The iGPU in the same machine (Radeon(TM) Graphics, 1002:164e, 32.0.21045.16010) exposes no mesh shaders, so the test is skipped there on both. The Khronos validation layer is not installed here, so no VUID- comparison was made.

AI-assisted: yes

[HabibUrRehmanBhattii]

Verified on current main (0cd07dc4ab25d924cbcebee5e0468a1b785042f7) on a third GPU:

• Windows 11 26200, NVIDIA GeForce RTX 4070, driver 616.92, WinLibs GCC 15.2.0 r7, Release, BUILD_TESTING=ON.
• Baseline (without this change): agc_driver_mesh_barycentric_emulation fails 3 of 3 runs, exit 1, AGC graphics: the pixel shader did not emulate barycentrics.
• With only this PR's change (target.fragmentBarycentricBuiltins = false; in core/libs/prx/libSceAgcDriver/tests/execution/MeshBarycentricEmulation.cpp): 3 of 3 runs pass (ctest -R '^agc_driver_mesh_barycentric_emulation$' --repeat until-fail:3; 5.11 s cold, 0.32 s each after).
• The device exposes mesh shaders and VK_KHR_fragment_shader_barycentric (mesh=1 in the test log).
• No Khronos validation layer was loaded in these runs (not installed on this machine), so no VUID- comparison is available from them.

This reproduces #12121 on a third device (RTX 4070; #12121 reports RTX 2070 SUPER and RTX 4090) and confirms this one-line change fixes it at the commit above.

AI-assisted: yes