[gjonn]

What

When a shader uses one guest sampler with an image that needs point filtering (sint, depth-bits, or a converted format without in-shader filtering) and also with an ordinary image, ResourceMaterializer::Apply adds a point-filtered copy of that sampler after the original samplers. The ResourceSnapshot only holds the original samplers, so DescriptorBindingBuilder::Populate threw guest sampler index is out of range on the copy. It threw in both the sampler binding and ProveUnnormalized, which reads every sampler's descriptor.

• SamplerResource::copyOf records the sampler a copy was made from (NoCopy otherwise). Apply sets it on the copy, next to forcePointFiltering.
• GuestSamplersDescriptor takes the ShaderInfo and reads a copy's descriptor from its source sampler. The point-filter rewrite of word 2 still applies to the copy.
• The field is part of the shader disk cache encoding: the size assert and the per-element minimum are updated, and the cache format version goes from 7 to 8.
• New execution test ImageSampleSharedSampler: a compute shader samples a 32_SINT texture and an 8_8_8_8_UNORM texture through one bilinear sampler, at a coordinate between two texel columns. It checks that the SINT result is the exact point-sampled texel and that the UNORM result is the bilinear blend (0.25), so the copy is point-filtered and the original still filters.
• The disk cache round-trip test sets copyOf.

Tested

Windows 11, MinGW, Release, NVIDIA Vulkan device.

• The new test throws DescriptorBindingBuilder::Populate guest sampler index is out of range on unmodified main and passes with this change.
• All 171 agc_driver_*, agc_shader_memory, agc_masked_select, agc_shader_disk_cache and image_sample_gather tests pass. This includes every Vulkan sampler execution test (sample reduction, unnormalized, 8_SINT, emulated color compare, gathers, LOD clamp, derivatives).
• agc_driver_mesh was not built: tests/Mesh.cpp does not compile under MinGW (setenv, std::aligned_alloc), and this change does not touch it.
• tools/check_conventions.py --base origin/main passes.

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

🤖 Generated with Claude Code

[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
#1292	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp L175-176 core/shader/recompiler/Optimization/src/DescriptorBindingBuilder.cpp L155, L159, L205, L348 core/shader/recompiler/ShaderDiskCache.cpp L73-75, L522
#1487	core/libs/prx/libSceAgcDriver/tests/ShaderDiskCache.cpp L175-176 core/shader/recompiler/Optimization/src/DescriptorBindingBuilder.cpp L155, L159, L205, L348 core/shader/recompiler/ShaderDiskCache.cpp L73-75, L522

[adriadam10]

Checked the diff, built and ran the full ctest on top of current main (455/455), and confirmed the new test needs copyOf to compile and pass. CI is green on all jobs.

One fix added as our own commit on your branch:

• ShaderDiskCache.hpp: main is already at FormatVersion = 8, so this change (new copyOf field in the sampler encoding) bumps it to 9. Otherwise a cache written by main@8 would be decoded with the wrong layout. The description's "7 to 8" is stale because of this.

Note: #1292 touches the same lines in DescriptorBindingBuilder.cpp and ShaderDiskCache.cpp; whichever merges second will need a rebase.

Merging once CI is green on the pushed commit.