[AlxcBTW]

What

FPV Simulator (PPSA10698 v01.006.000, Unity 2021.3 URP) ends when it presents a frame:

[gpu] color target 0x231590000 stays non-resident: AGC graphics: storage image access to depth/stencil surface 0x231590000 is not implemented
[videoout] presentation failed: AGC graphics: sampling the stencil plane of depth surface 0x230cd0000 (1920x1080, vk format 130) as a 1920x1080 texture of guest format 130 (vk 43), tile mode 7, dimension 1, levels 0-0, slice 0 is not implemented (T# 02315900 c8200000 010dc1df 91b00fac 00000000 00700000 00000000 00000000)
terminate called after throwing an instance of 'std::runtime_error'

Unity reuses the memory of transient render targets. The T# base, 0x231590000, is the stencil plane of the D32S8 surface at 0x230cd0000, and the first line shows the title binding that address as a color target. The view is a 1920x1080 RGBA8 sRGB texture, which reads the color data, not an 8-bit stencil plane.

9d65003 (boykopovar#1495) already treats this as memory reuse for the depth plane: a view with another extent, or a depth-plane view in a format that cannot hold the plane, falls through to the ordinary texture path. Stencil-plane views of the same extent kept the old handling: they all went to the depth surface, which throws for any format other than R8 uint. This applies the same rule to the stencil plane: a view whose texels are not one byte wide cannot hold the 8-bit plane, so DepthSurfaceTexture returns nothing and the memory is read like any texture.

• An 8-bit view of the plane's own extent still samples the stencil plane.
• An 8-bit view in another format than R8 uint still throws, as before.

Related: boykopovar#454 (open) is broader: it retires a depth surface once another writer covers its depth or stencil plane. This PR extends only the format rule from boykopovar#1495 and doesn't track writers. Both change the same lines in DepthSurfaceTexture, so whichever lands second needs a small rebase.

Tested

Windows 11, WinLibs GCC 15.2.0 r7, Release, RTX 5070 Laptop (driver 616.92).

• New cases in DepthSurfaceReuse.cpp (agc_driver_graphics), on a 320x180 D32S8 surface: an R8 uint view of the stencil plane still samples the plane; RGBA8 and R32F views of the same extent are read as reused memory. Without the source change, agc_driver_graphics fails on the RGBA8 case with the original error: sampling the stencil plane of depth surface 0x60000000 (320x180, vk format 130) as a 320x180 texture of guest format 56 (vk 37) ... is not implemented.
• Related tests (agc_driver_graphics, agc_driver_depth_bias_per_face, agc_driver_pending_exact_surface, agc_driver_prepared_graphics, agc_driver_cmask_fast_clear, agc_pass_hazards): 6/6.
• Suites:
  • Non-GPU tests: 446/446.
  • agc_driver on NVIDIA RTX 5070 Laptop: 253/253.
  • check_conventions.py passes.
• In-game: not re-run on this branch. The stop above is from a local build of main with other patches.

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 in TechnicalDebt
• No notes, investigation .md files or images added to the repository
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

Upstream issue: boykopovar#2994

[AlxcBTW]

Merged upstream as boykopovar#3277 (carried unchanged by @oneandonlydean, authorship kept).