feat(agc): depth bounds test and depth bias

[MichonGoddijn231849]

What

• Depth bounds: DB_DEPTH_CONTROL's DEPTH_BOUNDS_ENABLE runs the Vulkan depth bounds test with DB_DEPTH_BOUNDS_MIN/MAX (0x008/0x009). Without a depth surface it is still refused. Bounds outside [0, 1] are refused unless VK_EXT_depth_range_unrestricted is available, because clamping them would change the result at a stored depth of exactly 0 or 1.
• Depth bias: PA_SU_SC_MODE_CNTL's POLY_OFFSET_FRONT/BACK_ENABLE enable Vulkan depth bias:
  • constant factor = PA_SU_POLY_OFFSET_FRONT/BACK_OFFSET;
  • slope = _SCALE / 16 (the hardware scale is in 1/16 units);
  • clamp = PA_SU_POLY_OFFSET_CLAMP.
• Refused depth bias:
  • bias whose PA_SU_POLY_OFFSET_DB_FMT_CNTL units differ from the depth format's. A context that never writes the register is taken to use the format's units;
  • front and back bias that differ while both faces are drawn.
• Pipelines: both are dynamic state, so draws that differ only in bounds or bias values share a pipeline.
• Draw key: the registers join it.
• Still refused: depth-conditional color writes (DB_DEPTH_CONTROL bits 30/31).
• Device: depthBounds and depthBiasClamp are enabled when the device has them.

GTA V draws its deferred light volumes with the depth bounds test, and its shadow maps with depth bias. Both had been refused, so the scene got no point or spot lights and no shadows.

Tested

Windows 11 Pro, AMD Ryzen 5 7600, NVIDIA GeForce RTX 5080 (driver 32.0.16.1074), 32 GB DDR5-5600, MinGW-w64 GCC 15.2.0.

• GTA V (PPSA04264), with fix(agc): depth fast clears through the HTILE metadata #374: the light volumes and shadow-map draws are no longer rejected, and the Prologue's interior lights show.
• Dreaming Sarah (PPSA02929): the title menu, its text and the first scene render as on main.
• Demon's Souls (PPSA01342), with fix(libc): reserve the guest arena at its fixed range around what Windows already placed in it #231 and fix(libkernel): read, pread, AIO and APR write into guest memory as host writes #232: it reaches the main menu.
• ctest: all tests pass. New cases cover the decode of bounds and bias, and out-of-range bounds being refused without VK_EXT_depth_range_unrestricted and accepted with it.

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 (touches the same DrawKeyRegisters line as fix(agc): depth fast clears through the HTILE metadata #374; whichever lands second takes both ranges)
• AI-assisted: yes. Written with Claude (Anthropic) under my direction; I reviewed and tested it.

🤖 Generated with Claude Code

[oneandonlydean]

One exactness note: without VK_EXT_depth_range_unrestricted, out-of-range depth bounds are clamped to [0, 1]. That changes the result where the stored depth is exactly 0 or 1. For example, with DB_DEPTH_BOUNDS_MIN above 1 the hardware rejects every pixel, but after the clamp a pixel at depth 1 passes. Refusing bounds outside [0, 1] when the extension is missing would keep it exact; that's what my integration branch does.

[MichonGoddijn231849]

Agreed, thanks. Pushed: bounds outside [0, 1] are now refused when VK_EXT_depth_range_unrestricted is missing instead of clamped, with a test, and the branch is rebased onto main.

[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
#453	core/libs/prx/libSceAgcDriver/Graphics/src/State.cpp L705-707