[Celegans12]

What

8_SRGB color targets throw unsupported color format 1 number type 6 component swap 0. Decoding them isn't enough: R8_SRGB attachments are optional in Vulkan, NVIDIA 610.62 and lavapipe have none, and with the decode alone both drop the writes.

• DecodeColorFormat maps 8_SRGB (component swap 0) to R8_SRGB; other swaps and 8_8_SRGB still throw.
• Unless R8_SRGB supports both COLOR_ATTACHMENT and COLOR_ATTACHMENT_BLEND, AttachmentProxyFormat picks RGBA8_SRGB (APS5_SRGB_ATTACHMENT_PROXY=1 forces it). The resident R8 image is blitted into an RGBA8 copy before the pass and back after it, and the pass writes R only through an RGBA8_SRGB view: encode, decode and blending are the device's, as for 8_8_8_8_SRGB.
• Devices with R8_SRGB attachments render straight into an R8_SRGB view of the resident image, like other resident targets. Proxied draws leave no draw recipe.
• Still throws on a proxy device: linear, mipmapped and 3D 8_SRGB targets; two TechnicalDebt lines.
• PPSA26344 v1.008: the G-buffer pass (slot 5 of six 4K targets, first 0x5050bc0000) and a multiply-blend pass (0x503b270000) throw this.
• feat(agc): multisampled color and depth targets with CB resolves #419 and fix(agc): store and read DCC keys over the console's metadata extent #788 also edit Draw (feat(agc): multisampled color and depth targets with CB resolves #419 recordDraw too), feat(agc): host import windows, least-recent eviction under the budget #290 and feat(agc): implement sint storage images through uint views #944 StorageTexture::release; all merge cleanly. Draft GTA V: the prologue plays (testing branch of the PRs it needs) #426 does too, but already conflicts with main.

Tested

• Windows 11, MinGW-w64 GCC 15.2.0, based on main 9ab937b: configure -DBUILD_TESTING=ON, full build, --target libs, ctest:
• 380/380 tests pass (ctest --output-on-failure)
• Same build on lavapipe (Mesa 26.2.4 llvmpipe, 8-wide subgroups, VK_DRIVER_FILES): 379/379 pass; agc_driver_flat_store is left out there because it takes several minutes on Windows llvmpipe
• New agc_driver_srgb8_color_target: multiply and DST_ALPHA blends (two in a continued pass), fast clear and MRT within 1 sRGB code; ZERO/ONE and outside-scissor bytes exact; fails on main.
• agc_driver_graphics: decode, refusals, proxy choice; won't build on main.
• PPSA26344 v1.008 (with the cache budget change): no color format 1 number type 6 skips, 3,355 before; 0 of 7 frame dumps from 250 to 480 s black, 9 of 10 before.
• Not tested: R8_SRGB attachments, Linux (CI), the console.

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
• AI-assisted: yes

[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
#1122	docs/dev/TechnicalDebt.md L256
#1182	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp L8
#1190	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp L8
#1199	core/libs/prx/libSceAgcDriver/tests/Graphics.cpp L8

[adriadam10]

Checked on RADV (direct R8_SRGB path and the forced RGBA8_SRGB proxy via APS5_SRGB_ATTACHMENT_PROXY=1): agc_driver_graphics and agc_driver_srgb8_color_target pass, full ctest 416/416 on top of current main; the new tests don't build or fail without the change; Linux/Windows CI green. Remaining 8_SRGB cases still throw and are listed in TechnicalDebt; merging.