[ee6aba54]

What

CB_COLOR_VIEW can select a range of slices (SLICE_START to SLICE_MAX); a draw into a 3D color target then addresses the slice per primitive through the render target index export. The decoder rejected every view whose last slice differs from its first ("color views of several array slices are unsupported"), and the index export was accepted but never routed.

Two commits:

• A 3D target's view carries its slice count (clamped to the surface depth), the resident image's attachment view becomes a 2D array over those slices, and the framebuffer takes the layer count, which every color target of the draw must share. A depth target alongside a layered draw is still refused, and 2D array targets keep the single-slice rule.
• For such a draw, PA_CL_VS_OUT_CNTL reaches the vertex stage info (and the draw hash, the recompile cache key and the request serialization, now version 18), so the render target index export becomes a gl_Layer output: per vertex, or per primitive in a mesh shader (the emitter already produced that output; the validation refused it). SPIR-V targets below 1.5 take the vertex built-in from SPV_EXT_shader_viewport_index_layer; the device enables VK_EXT_shader_viewport_index_layer, the validation accepts the vertex built-in with that capability or ShaderLayer and the mesh built-in, and dynamic rendering begins with the framebuffer's layer count instead of 1. Single-slice views keep ignoring the export, which lands on their slice as the hardware clamp would; the viewport index export stays ignored and keeps its one-time log line.

Tested

• Ubuntu 24.04, Linux 7.0, GCC 13.3, Release, based on 33ae282e4. ctest of a tree combining this branch with fix(libSceAgcDriver): accept DCC on 3D color targets #4499 and feat(libSceAgcDriver): storage image access to depth surfaces #4500: 954 tests pass; video_out_flip_* and agc_driver_stencil_ref_export are skipped on this machine.
• New agc_vertex_layer_export (recompiler only) checks the SPIR-V 1.3 module takes gl_Layer from the extension without raising its version, the 1.5 module from ShaderLayer, and a misc vector export without the index declares no layer output. New agc_driver_layer_export_slices (RTX 5080) draws into slices 0-3 of a 128x128x4 R32 volume with a vertex shader exporting slice 2 and reads every slice back: only slice 2 changes, also after the volume is written back to guest memory and uploaded again. agc_driver_graphics covers the decode (slice range, clamping, inverted range) and the validation (the vertex Layer built-in needs the device flag). agc_shader_memory pins the new serialization version.
• PPSA15304 (engine: Unreal Engine 5): the draw into a 48-slice 48x48x48 3D target that stopped the title is a mesh shader draw exporting the slice per primitive; it now runs with the export routed. Together with fix(libSceAgcDriver): accept DCC on 3D color targets #4499, feat(libSceAgcDriver): storage image access to depth surfaces #4500 and a fix for the compute dispatch look-ahead submitted separately, the title reaches its main menu. The mesh built-in is covered only by this title run; no test builds a mesh module with a per-primitive Layer output.
• Windows not built or tested.

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
#1077			core/libs/prx/libSceAgcDriver/tests/VertexLayerExport.cpp
#1313	core/libs/prx/libSceAgcDriver/Execution/src/Driver/Draw/Draw.cpp L147
#3633	core/libs/prx/libSceAgcDriver/Graphics/src/State.cpp L996, L1009-1010
#3763	core/libs/prx/libSceAgcDriver/Execution/src/Driver/Draw/Draw.cpp L147 core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp L267 core/libs/prx/libSceAgcDriver/Graphics/include/Pipeline.hpp L18, L50, L70, L126 core/libs/prx/libSceAgcDriver/Graphics/include/State.hpp L57 core/libs/prx/libSceAgcDriver/Graphics/include/Texture.hpp L163, L536 core/libs/prx/libSceAgcDriver/Graphics/src/Draw.cpp L740, L1968, L2156, L2532 core/libs/prx/libSceAgcDriver/Graphics/src/Pipeline.cpp L72, L84, L567, L582, L592, L605, L632 core/libs/prx/libSceAgcDriver/Graphics/src/ShaderValidation.cpp L48, L130, L136-138, L145, L245, L566, L587 core/libs/prx/libSceAgcDriver/Graphics/src/State.cpp L100-102, L694-701, L972, L996, L1008-1014 core/libs/prx/libSceAgcDriver/Graphics/src/Texture.cpp L1329, L1335, L1342, L1345, L1348 core/libs/prx/libSceAgcDriver/tests/Graphics.cpp L1319, L3325-3329 core/libs/prx/libSceAgcDriver/tests/VertexLayerExport.cpp core/shader/recompiler/ControlFlow/src/RequestSerializer.cpp L404, L407, L423, L568 core/shader/recompiler/SpirvBackend/src/SpirvMemory/SpirvModuleSetup.cpp L109-111		core/libs/prx/libSceAgcDriver/tests/VertexLayerExport.cpp
#4451	core/libs/prx/libSceAgcDriver/tests/ShaderMemory.cpp L1640, L1645

[ee6aba54]

Closing as a duplicate of #1077 (layered draws into 3D color targets, the render target index export routed to gl_Layer, VK_EXT_shader_viewport_index_layer) and #3763 (the dynamic rendering layer count). I missed both when I checked for overlapping pull requests.

For those pull requests, in case it helps: PPSA15304 (engine: Unreal Engine 5) renders into a 48-slice 48x48x48 3D target with a mesh shader that exports the slice per primitive, so the per-primitive Layer output of mesh shaders matters there as well as the vertex one. With an equivalent change, #4500 and #4586, the title reaches its main menu.