[Celegans12]

What

A sampler with FORCE_DEGAMMA (S# word 0 bit 20) throws guest sampler descriptor forces sRGB decoding which is not implemented at decode. This is the direct-sampler part of the earlier version, now on main without #757; the image table check follows once #757 lands.

• DecodeSamplerResource records the bit as forceDegamma (host sampler unchanged) and accepts it only for callers that pair the sampler with its images (forceDegammaPaired, also on SamplerCache::Get).
• The direct sampler binding pairs: RequireDegammaFormat lets only sRGB textures (8, 8_8, 8_8_8_8, BC1/2/3/7 SRGB; IsSrgbTextureFormat) through such a sampler, sampled as before, so decoded once.
• Hardware: on RDNA2 (gfx1035), 8_SRGB, 8_8_SRGB and 8_8_8_8_SRGB sample bit for bit the same with and without FORCE_DEGAMMA, point and bilinear. Forcing UNORM 8, 8_8 and 8_8_8_8 decodes per texel before filtering, contrary to the older GPCS4/fpPS4 notes; this PR still throws for those formats. BC sRGB formats remain unmeasured (measurement).
• Still throws: other textures through such a sampler, unpaired decodes and cache lookups.
• PPSA26344 v1.008: compute shader 0x80003fca00 samples a 64x1 or 1024x1 BC1_SRGB T# (format 170) through a bilinear forcing S#; 202 dispatches were skipped in one run.

Tested

• Windows 11, MinGW-w64 GCC 15.2.0, based on main f145139: configure -DBUILD_TESTING=ON, full build, --target libs, ctest:
• 489/489 tests pass (ctest --output-on-failure)
• Linux (WSL Ubuntu 24.04, GCC 13.3, Mesa llvmpipe, upstream's job configuration), merged with main f145139: 498/498 pass; the sampler tests also pass on Windows lavapipe (Mesa 26.2.4)
• New agc_driver_image_sample_force_degamma: a forcing S# samples 8_8_8_8_SRGB bit for bit as a plain one (point, bilinear); UNORM throws. It fails on main (guest sampler descriptor forces sRGB decoding which is not implemented).
• agc_driver_graphics: decode, unpaired refusal on a cache miss and hit, IsSrgbTextureFormat.
• PPSA26344 v1.008, with the version stacked on feat(agc): implement image and sampler tables indexed at runtime offsets #757 before perf(agc): baseline shader recompilation on registration #639: the "forces sRGB decoding" skips of 0x80003fca00 were gone. Not run again.
• Not tested: PPSA21564, Dreaming Sarah, console behaviour.

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

[EdouardSence]

Tested on AMD under Linux, since this PR wasn't: RX 9070 XT (RDNA4, GFX1201), RADV / Mesa 26.2.4, Linux 7.2.9 (CachyOS), GCC Release build.

• ctest: this PR merged onto main 07c2405, all 455 tests pass with ctest -j8 (main alone: 452/452 on the same machine).
• Astro Bot (PPSA21567, retail): built as main + feat(agc): implement deferred faults for reads through null SRT pointers #759 (needed to boot this title on main) + this PR (with feat(agc): implement image and sampler tables indexed at runtime offsets #757, which it stacks on), then one scripted run headless from boot through the title screen. No error and no amdgpu fault came from this change.
  • Presents per 10 s on the title screen: 229, 224, 227; with main + feat(agc): implement deferred faults for reads through null SRT pointers #759 alone: 226, 229, 226.
  • The run then stops, with and without this PR, on storage image access to depth/stencil surface ... is not implemented (the gap feat(agc): depth planes sampled as 2D arrays, cubes and integer views #425 fills). So nothing past the title screen is covered.

AI-assisted: the runs and this write-up were done by Claude (Claude Code) on my machine.

[adriadam10]

Rebased onto current main (only a TechnicalDebt adjacency conflict) and added one commit that updates the FORCE_DEGAMMA TechnicalDebt line with a hardware measurement.

• Measured on RDNA2 (gfx1035) with image_sample_lz: 8_SRGB, 8_8_SRGB and 8_8_8_8_SRGB sample bit for bit the same with and without FORCE_DEGAMMA, point and bilinear, so passing them through unchanged is exact there.
• For the follow-up: on the same hardware, UNORM 8, 8_8 and 8_8_8_8 through a forcing S# sample bit for bit as the matching sRGB format, i.e. decoded per texel before filtering, not after as the GPCS4/fpPS4 notes say. This PR still throws for them, which is right; the TechnicalDebt line now says so.
• BC sRGB formats were not measured.
• Checked: the libSceAgcDriver build and ctest pass after the rebase; the new execution test fails without the change; shadPS4#3420 skips sRGB images as described.

Merging once CI is green.

[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
#1415	core/libs/prx/libSceAgcDriver/Graphics/include/GuestSamplerResource.hpp L32
#1733	core/libs/prx/libSceAgcDriver/Graphics/include/Sampler.hpp L62 core/libs/prx/libSceAgcDriver/Graphics/src/Sampler.cpp L94