fix(agc): rasterize a single-sample scan into a multisampled target at the pixel center

[AchrafReyani]

What

A draw whose rasterizer is single-sample (PA_SC_AA_CONFIG.MSAA_NUM_SAMPLES = 0) with MSAA_ENABLE off, into a multisampled color target, throws a color target's sample count differs from the rasterizer's.

The hardware supports this combination. radeonsi programs exactly this when GL multisampling is disabled on a multisampled framebuffer: in si_emit_msaa_config (si_state.c, mesa-25.2.0), sc_aa_config stays 0 unless rs->multisample_enable, and MSAA_ENABLE follows state->multisample, while CB_COLORi_ATTRIB keeps the framebuffer's samples. GL defines the result: with MULTISAMPLE disabled, rasterization is single-sample at the fragment center "except that the fragment coverage value is set to full coverage", so every sample gets the pixel's color.

Such a draw now rasterizes the color target's sample count with every sample location at the pixel center (the custom sample locations from #2169), so all samples are covered exactly when the center is. It applies only with no depth target, no alpha-to-coverage and MSAA_ENABLE off, and only when the first written color target is multisampled. Any other mix still throws, and MSAA_ENABLE on keeps #4309's checks.

Title: ProsperoPuzzles (PPSA99006) stops here on main 9bc3623 at its first draw into its 4x target:

  what():  AGC graphics: a color target's sample count differs from the rasterizer's

Register values at that draw (posted on #419): CB_COLOR0_ATTRIB 0x00012000 (4 samples, 4 fragments), PA_SC_AA_CONFIG 0x00000000, PA_SC_MODE_CNTL_0 0x00000022 (MSAA_ENABLE off), no depth. The title renders through blackbearreloaded/ps5-opengl (Mesa gallium on AGC), the same driver family as radeonsi. With this change it gets past that draw and stops at the next one, which samples that 4x image: DescriptorBindingBuilder::Populate multisampled image resources have no descriptor image shape (#3036).

Tested

Windows 11 (26200), RTX 4070, WinLibs GCC 15.2.0 r7, Release, main 9bc3623:

• agc_driver_graphics (multisampleTests): AA config 0, MSAA_ENABLE off and a 4x/4f target decode as 4 samples with customSampleLocations and all four locations at (0.5, 0.5). With MSAA_ENABLE on, the same draw is still refused (by fix(agc): accept MSAA_ENABLE on a single-sample target sampled at the pixel center #4309's check). This fails with main's State.cpp and passes with this change.
• ctest -E "^agc_|video_out|^shader|recompiler": 497/497. GPU suite (VK_LOADER_DRIVERS_DISABLE=*dzn*, Windows: VideoOut crashes intermittently when the Dozen (Vulkan on D3D12) driver is installed, even though the native driver is selected #3255): 441/441.
• ProsperoPuzzles, before and after (above).
• tools/check_conventions.py and tools/pr_overlap.py: clean.

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 (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: PPSA99006 ProsperoPuzzles (PPSA99006) on Windows / NVIDIA: MSAA blockers #4427
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

[xXPBSXx]

Windows test (MinGW, local build): PR head merged onto main c567576 without conflicts; agc_driver_graphics (ctest) passes, including the new test from this PR. Not tested: no game run, no other test suites.

AI-assisted (Claude Code).

[xXPBSXx]

Correction to my test comment: the base I actually built and tested was main 0ef9dc9, not c567576 (my local upstream/main ref had moved). Result unchanged: clean merge, agc_driver_graphics passes.

AI-assisted (Claude Code).